Abstract A47: Characterization of immune cell composition across cancer types in pan-cancer genomic cohorts

Abstract

Abstract Introduction: Identifying immune cell signatures in individual tumors can help guide treatment selection for patients. Numerous deconvolution methods have been developed to estimate immune cell fractions from bulk gene expression data, but they have yet to be systematically applied and cross-validated in pan-cancer genomic cohorts. We undertook this study to cross-validate immune cell quantification methods across 25 cancer types spanning 11,011 samples and provide a public immuno-oncology resource. Methods: Using gene expression data from both The Cancer Genome Atlas (TCGA) and the International Cancer Genome Consortium (ICGC), we employed and compared six methods to estimate immune cell fractions in each cancer type (CIBERSORT, quanTIseq, EPIC, TIMER, MCP-counter, xCell; immunedeconv R package). We mapped these results to a common vocabulary of five broad cell categories for comparison: T cells, B cells, natural killer cells (NK cells), macrophages/monocytes, and myeloid dendritic cells (mDCs). In parallel, we computed immune cell proportions for seven cancer types using single-cell RNA-seq (scRNA-seq) data from a single-cell tumor immune atlas of 217 patients. We correlated the median immune cell fractions estimated from bulk deconvolution with scRNA-seq proportions for each cancer type. To demonstrate the application of this resource, we compared the immune cell fractions (1) in adjacent normal versus tumor tissues, and (2) in head and neck squamous cell carcinoma (HNSC) and colorectal adenocarcinoma (COAD) tumor subtypes using multivariable linear regression adjusted for age and sex. Results: Overall, 9,689 TCGA samples and 1,322 ICGC samples were analyzed, and correlations across six deconvolution tools were performed. Two well-performing cell estimation methods, EPIC and quanTIseq, demonstrated good correlation between median deconvoluted T cell enrichment score and single-cell cytotoxic CD8+ T cell populations (spearman coefficient; EPIC = 0.71, quanTIseq = 0.43). However, cross-cancer type correlations between scRNA-seq and bulk estimated fractions were not statistically significant for most cancer types. In multivariable regression comparing immune cell estimates across cancer subtypes, we found increased T cell (EPIC: OR=1.65, 95% CI = 1.23-2.21, quanTIseq: OR = 1.30, 95% CI = 1.20-1.40) and B cell (EPIC: OR = 2.92, 95% CI = 2.05-4.15, quanTIseq: OR = 1.48, 95% CI = 1.37-1.60) enrichment in HPV-positive compared to HPV-negative HNSC. In COAD, we found increased T cell enrichment in the microsatellite instability (MSI) subtype compared to the genome stable subtype (OR = 1.16, 95% CI = 1.01-1.33). Conclusion: Overall, our study provides a public resource of immune cell annotations for two widely used, pan-cancer genomics cohorts and can facilitate future studies that aim to characterize the tumor immune microenvironment. Furthermore, we validated the use of this resource by demonstrating enrichment of specific immune cell subsets in HNSC and COAD subtypes, consistent with prior reports of these well-characterized cancers. Citation Format: Makda Getachew Zewde, Daniel Fulop, Alexander Tsankov, Kuan-lin Huang. Characterization of immune cell composition across cancer types in pan-cancer genomic cohorts [abstract]. In: Proceedings of the AACR Special Conference: Tumor Immunology and Immunotherapy; 2022 Oct 21-24; Boston, MA. Philadelphia (PA): AACR; Cancer Immunol Res 2022;10(12 Suppl):Abstract nr A47.