Ensemble Analysis of Mixture and Single-Cell Transcriptomes in Tumor–Immune Microenvironment

Abstract

A variety of cell types form complex cellular networks and regulate each other in tissue microenvironment. RNA-sequencing (RNA-seq) of bulk tissue samples averages signals from heterogeneous cells, however, individual components can be inferred using in silico deconvolution with high scalability but limited resolution. Recently, single-cell RNA-sequencing (scRNA-seq) profiles the transcriptomes of individual cells, providing in-depth gene expression features of thousands of cells in a small number of samples. Co-analysis of bulk and scRNA-seq would complement the limitations in each platform. Here, I investigated tumor–immune microenvironment by joint analysis of bulk and scRNA-seq to determine cell type-specific transcriptomic signatures and clinical relevance in large-scale cohorts. Bulk RNA-seq from hepatocellular carcinoma cohorts were deconvolved to infer tumor-infiltrating immune cells. Infiltration of regulatory T cells prior to liver transplantation stratified patients at high risk of recurrence, however, the sensitivity of inference of rare cells in mixture samples was limited. To overcome low confidence of computational decomposition, I directly profiled transcriptomes of individual cells using scRNA-seq. First, 29,490 blood cells from Drosophila lymph glands were annotated and aligned to developmental trajectories to identify novel cell types and dynamic changes in differentiation under immune challenges. Next, the colorectal tumor microenvironment (TME) was profiled with both bulk and scRNA-seq to characterize molecular features of malignant cells and determine clinical relevance. The malignant cell-specific lncRNA SNHG16 was transcribed into two isoforms encoding micropeptides which were associated with clinical outcomes to different extents. Moreover, malignant signature programs were defined by normalizing gene expression variance in cancer cells. Proliferation and mesenchymal-like signatures were frequently found across genetic subclones, and the latter stratified colon cancer patients with worst outcomes. These results highlight that joint analysis of bulk and scRNA-seq not only identifies transcriptomic features with high specificity, but also generalizes the results to the cohort scale.