Non-small cell lung cancer: Analysis using mass cytometry and next generation sequencing reveals new opportunities for the development of personalized therapies

• Published in: Journal of clinical oncology Vol. 38; no. 15_suppl; p. e21026
• Main Authors: Raju Paul, Susan, Bagaev, Alexander, Valiev, Ivan, Zyrin, Vladimir, Zaitsev, Aleksandr, Dyykanov, Daniyar, Nuzhdina, Katerina, Kotlov, Nikita, Frenkel, Felix, Korek, Skylar, Reeves, Patrick, Davies, Diane L., Wright, Cameron D, Ott, Harald, Muniappan, Ashok, Tsiper, Maria, Fowler, Nathan, Lanuti, Michael, Ataullakhanov, Ravshan, Poznansky, Mark C
• Format: Journal Article
• Language: English
• Published: 20.05.2020
• ISSN: 0732-183X; 1527-7755
• DOI: 10.1200/JCO.2020.38.15_suppl.e21026

Abstract only e21026 Background: Comprehensive molecular profiling and the use of biomarkers as companion diagnostics have transformed precision medicine for cancer patients. To identify patient-specific tumor microenvironment and biomarker profiles, we assessed the accuracy of our deconvolution algorithm in identifying cellular compositions from whole exome (WES) and whole transcriptome (RNA-seq) sequencing of solid tumors compared with cell populations identified by Mass Cytometry by Time of Flight (CyTOF) in surgically resected tissue from non-small cell lung cancer (NSCLC) patients. Methods: Resected NSCLC tissue was divided for RNA-seq and WES of whole tissue (n = 9) and for generating tissue single cell suspensions through mechanical dissociation and enzymatic digestion (n = 11). Bulk RNA-seq and CyTOF were performed on all cell suspensions. Cellular phenotypes were identified using clustering algorithms in CyTOF and predicted from bulk RNA-seq using our proprietary computational method. Results: Cellular composition reconstructed from RNA-seq correlated with the composition detected by CyTOF (R 2 = 0.922, n = 7) from cell suspensions. To recover the cell percentage from bulk RNA-seq, a machine learning framework was trained on the cell compendia comprising 7,117 unique cell type RNA-seq profiles. A two-stage hierarchical learning procedure generated a gradient boosting Light GBM model that included training on artificial RNA-seq mixtures of different cell types. With this method, we found that stromal and malignant cells were depleted during single cell suspension preparation, resulting in statistically significant differences in the tumor cell composition reconstructed from solid tissue and single cell suspensions. Immune cell types namely T cells and macrophages were similarly represented in both the bulk tumor tissue and matched single cell suspensions. Transcriptomics revealed a subgroup of patients whose tumors were B-cell-enriched, which was validated in other NSCLC cohorts and was associated with greater CD4+ and CD8+ T cell infiltration and improved clinical outcomes. Conclusions: Since preparation of single cell suspensions leads to the loss of several cellular components, RNA-seq of tumor bulk tissue better describes the molecular and cellular properties of the tumor microenvironment. The combination of RNA-seq and WES of tumor tissue provides a comprehensive profile of cellular composition, suggesting that this combination is ideal for precision medicine applications.