Understanding the transcriptional consequences of oncogenic mutations is an important goal that may reveal new therapeutic targets for diverse cancers. Although single-cell methods hold promise for this task, it remains non-trivial to isolate and sequence DNA and RNA from the same cell at scale. Here we present a statistically motivated strategy that utilizes multiscale and multiomic analysis of individual human tumor specimens to deconstruct intra-tumoral heterogeneity by clarifying clonal populations of malignant cells and their transcriptional profiles. By combining deep, multiscale sampling of IDH-mutant astrocytomas with integrative, multiomic analysis, we reconstruct and validate the phylogenies, spatial distributions, and transcriptional profiles of distinct malignant clones. We identify a core set of genes that is consistently expressed by the truncal clone, including AKR1C3, whose expression is associated with poor outcomes in several types of cancer. Some derived clones exhibit significant enrichment with gene sets representing glioblastoma subtypes and nonmalignant cell types, including ependymal cells. Importantly, by genotyping nuclei for truncal mutations, we show that existing strategies for inferring malignancy from gene expression profiles of single cells may be inaccurate. Furthermore, we find that transcriptional phenotypes of malignancy persist despite loss of the mutant IDH1 protein following chr2q deletion in a subset of malignant cells. In summary, our study provides a generalizable strategy for precisely deconstructing intra-tumoral heterogeneity and clarifying the molecular profiles of malignant clones in any kind of solid tumor.We extend this approach to a metaanalysis of the cell-type specific dysregulation in the glioma microenvironment. Using the same statistically motivated approach to take advantage of inherent patterns of cell-type specific coexpression, we characterize the differentially expressed cell-type specific transcriptome. We perform this process on thousands of samples and hundreds of datasets from both glioma and normal samples. Finally by taking the difference in the in silico derived differential expression metrics for each cell type in glioma and normal contexts, we identify ideal markers of each cell type specifically in glioma but not normal and validate and filter them using orthogonal datasets. In summary, our study provides a generalizable strategy for precise identification of cell-type specific dysregulated genes using abundant bulk transcriptome data for any disease state involving solid tissues.
