Vadim Kurbatov1, Agshin Balayev2, Areo Saffarzadeh3, Danielle R Heller4, Daniel J Boffa3, Justin D Blasberg3, Jun Lu5, Sajid A Khan6. 1. Department of Surgery, Yale School of Medicine, New Haven, Connecticut. Electronic address: vadim.kurbatov@yale.edu. 2. University of Zurich, Zurich, Switzerland. 3. Division of Thoracic Surgery, Department of Surgery, Yale School of Medicine, New Haven, Connecticut. 4. Department of Surgery, Yale School of Medicine, New Haven, Connecticut. 5. Department of Genetics, Yale School of Medicine, New Haven, Connecticut. 6. Section of Surgical Oncology, Department of Surgery, Yale School of Medicine, New Haven, Connecticut.
Abstract
BACKGROUND: Classification of lung adenocarcinoma (LUAD) currently relies on the TNM pathological staging system, which cannot fully account for the variability in postsurgery overall survival (OS). Despite the advances in immunotherapy and increased appreciation of the involvement of cancer immune microenvironment (IME) in cancer progression, the contribution of IME to postsurgery LUAD prognosis is not well understood. METHODS: We digitally inferred the contribution of 22 immune cell types or activation states to the tumor IME using CIBERSORT (Celltype Identification By Estimating Relative Subsets Of RNA Transcripts) analysis in an exploratory metadataset of 581 patients with early-stage LUAD. Patients were arranged based on similarity in IME using k-means clustering. Relationship to postsurgical OS was tested in univariable and multivariable models using Kaplan-Meier analysis and Cox proportional hazards modeling, respectively. To confirm survival relationships, a support vector machine classifier was constructed from a comparison of low-risk and high-risk IME groups. The classifier was applied to a the Cancer Genome Atlas LUAD validation dataset of 394 patients. RESULTS: Patients with an inferred IME enriched in resting mast cells and depleted of macrophages represented a low-clinical-risk group in both exploratory and validation cohorts. CONCLUSIONS: Variability in the digitally inferred composition of the tumor IME contributes to heterogeneity in postsurgical OS. Our data suggest that low inferred macrophage content and inferred resting activation state of intratumor mast cells are associated with improved clinical outcome. Computational inference can be used to define LUAD risk groups and help guide clinical decision making.
BACKGROUND: Classification of lung adenocarcinoma (LUAD) currently relies on the TNM pathological staging system, which cannot fully account for the variability in postsurgery overall survival (OS). Despite the advances in immunotherapy and increased appreciation of the involvement of cancer immune microenvironment (IME) in cancer progression, the contribution of IME to postsurgery LUAD prognosis is not well understood. METHODS: We digitally inferred the contribution of 22 immune cell types or activation states to the tumor IME using CIBERSORT (Celltype Identification By Estimating Relative Subsets Of RNA Transcripts) analysis in an exploratory metadataset of 581 patients with early-stage LUAD. Patients were arranged based on similarity in IME using k-means clustering. Relationship to postsurgical OS was tested in univariable and multivariable models using Kaplan-Meier analysis and Cox proportional hazards modeling, respectively. To confirm survival relationships, a support vector machine classifier was constructed from a comparison of low-risk and high-risk IME groups. The classifier was applied to a the Cancer Genome Atlas LUAD validation dataset of 394 patients. RESULTS:Patients with an inferred IME enriched in resting mast cells and depleted of macrophages represented a low-clinical-risk group in both exploratory and validation cohorts. CONCLUSIONS: Variability in the digitally inferred composition of the tumor IME contributes to heterogeneity in postsurgical OS. Our data suggest that low inferred macrophage content and inferred resting activation state of intratumor mast cells are associated with improved clinical outcome. Computational inference can be used to define LUAD risk groups and help guide clinical decision making.