Context-specific functions of genes and cells

Division Of Basic Sciences - Nci

Investigators

Linked publications & trials

Abstract

(1) Intercellular communications in the tumor microenvironment (TME) is likely to influence, and influenced by, the transcriptional and phenotypic states of various cell types, ultimately affecting the clinical outcome. Details of such clinically relevant interactions between cellular states in IDH-mutant gliomas remains obscure. Here we develop a data-driven and "biomarker independent" computational pipeline to identify robust interactions among distinct cell states in the IDH-mutant glioma TME, termed cell state interactions (CSIs). The CSIs are associated with tumor progression and immunotherapy response, and are likely mediated through the interaction between cell surface molecules among the interacting cell types. The distinct cell states of malignant cells involved in CSIs resembled known neuronal lineages such as astrocytic and oligodendrocyte progenitor cells, and captured a unique multicellular network involving IDH1-mediated stemness of the cancer cells, resembling primitive embryonic development of brain, and immune cells. Overall, our work provides a novel approach to inferring clinically relevant cellular interactions in the TME with implications on patients stratification for therapeutic interventions. A manuscript reporting this work was reviewed and rejected by Sc Advances. A revised manuscript with additional analysis of spatial transcriptomic data is under review at MSB. (2) Gene signature refers to a set of genes whose expression characterizes the activity of a specific biological process or a cell state, e.g., EMT, proliferation, stemness, etc. Gene signatures are widely used to infer the state of a biological system based on transcriptomic data. However, the available gene signatures lack tissue or cellular context. For instance, a stemness signature derived from one tissue context may not be perfectly applicable to another tissue context. Across tissues, overlapping yet distinct sets of genes are likely to mediate a specific biological process. Here we derive cancer type-specific gene signatures, for 14 oncogenic hallmark processes across 23 cancer types, by integrating the available context-agnostic gene signatures with the protein interaction network and cancer-specific transcriptomic data. For several hallmarks, most notably EMT, DNA damage, and proliferation, the derived tissue-specific gene signatures are significantly more prognostic of survival than the corresponding reference signatures and better distinguish responders from non-responders in multiple clinical trial datasets. The context-specific genesets also reveal the tissue-specific genes and processes that interface with the core hallmark genes. We provide the projections for 14 cancer hallmarks across 23 human cancer types, along with the tool - Cancer Specific Transcriptomic Signature (CATS), as a Bioconductor package and GitHub repository. A manuscript reporting this work is being prepared. (3) Cytones are critical mediators of homestatic tissue function and cancer. However, the downstream response of cytokines are experimentally profiled only for a small minority of all cytokines, and being able to estimate cytokine activity in a tissue has significant clinical implications. Following up on our single cell miRNA prediction tool miRSCAPE, we are now developing a machine learning tool to predict cytokine activity in a tissue microenvironment based on the the transcriptome of the individual cell types in the microenvironment.