Cancer immunotherapies that target tumor-specific T cells have benefited many cancer patients, but the clinical efficacy varies greatly among different cancer types. Tumor-infiltrating T cells often enter a dysfunctional state, widely known as T cell exhaustion, and the antitumor functions of effector T cells are regulated by multiple factors, including the presence of regulatory T cells (T_reg cells). The states and abundances of T cells vary across tumor microenvironments (TMEs) of different cancer types, which may fundamentally influence different clinical parameters such as drug response to immunotherapies.

To build a high-resolution pan-cancer T cell atlas, we performed single-cell RNA sequencing (scRNA-seq) on tumors, paracancerous tissues, and blood samples from patients of various cancer types and collected additional published scRNA-seq datasets. The diverse data were integrated after correcting confounding factors and batch effects. This atlas was composed of scRNA-seq data from 316 patients across 21 cancer types. T cell receptor (TCR) sequences of individual T cells with gene expression profiles were assembled to characterize the expansion and dynamics of T cells. Various computational methods were applied to investigate the features and abundance of T cells across cancer types.

We identified multiple potentially tumor-reactive T cell (pTRT) populations in cancer patients. The states of the pTRTs varied dramatically in the tumor microenvironment of different cancer types. For CD8⁺ T cells, the major pTRTs were exhausted T cells and exhibited high heterogeneity. We computationally inferred two major developmental paths to T cell exhaustion, through effector memory T cells and tissue-resident memory T cells, respectively, and both were prevalent among cancer types. We also noted the state transitions between terminal exhausted T cells and cells such as natural killer (NK)–like T cells, Type 17 CD8⁺ T cells (Tc17 cells) cells, and CD8⁺ T_reg cells, but such transitions tend to occur in specific cancer types. For CD4⁺ T cells, follicular helper T cell (T_FH)/T helper 1 (T_H1) dual-functional T cells, which appeared to originate from T_FH cells, were also notable pTRTs and correlated with the tumor mutation burden. We also found that the transcriptional programs of pTRTs could be affected by transforming growth factor–β (TGF-β) and interferons in the TMEs. The abundances of T cell states vary dramatically depending on cancer types. On the basis of tumor-infiltrating T cell compositions, cancer patients could be immune-typed as a group with high frequencies of terminal exhausted CD8⁺ T cells and another group with high frequencies of tissue-resident memory CD8⁺ T cells, and the immune types were associated with clinical traits such as patient survival and responses to immune checkpoint blockade.

We depicted the pan-cancer landscape of T cell heterogeneity and dynamics in the TME and established a baseline reference for future temporal or spatial studies associated with cancer treatments. The systematic comparison across cancer types revealed the commonalities and differences of T cell states in different TMEs. Our detailed signature, dynamics, and regulations of tumor-infiltrating T cells will facilitate the development of immunotherapies, and our proposed immune-typing can aid the therapeutic and diagnostic strategies that target T cells.

We analyzed approximately 390,000 T cells from 316 patients of 21 cancer types by means of scRNA-seq. Combining gene expression profiles and T cell receptor sequences, we …