Single-cell transcriptome analysis reveals TOX as a promoting factor for T cell exhaustion and a predictor for anti-PD-1 responses in human cancer

Abstract

Background: T cells exhibit heterogeneous functional states in the tumor microenvironment. Immune checkpoint inhibitors (ICIs) can reinvigorate only the stem cell-like progenitor exhausted T cells, which suggests that inhibiting the exhaustion progress will improve the efficacy of immunotherapy. Thus, regulatory factors promoting T cell exhaustion could serve as potential targets for delaying the process and improving ICI efficacy.

Methods: We analyzed the single-cell transcriptome data derived from human melanoma and non-small cell lung cancer (NSCLC) samples and classified the tumor-infiltrating (TI) CD8⁺ T cell population based on PDCD1 (PD-1) levels, i.e., PDCD1-high and PDCD1-low cells. Additionally, we identified differentially expressed genes as candidate factors regulating intra-tumoral T cell exhaustion. The co-expression of candidate genes with immune checkpoint (IC) molecules in the TI CD8⁺ T cells was confirmed by single-cell trajectory and flow cytometry analyses. The loss-of-function effect of the candidate regulator was examined by a cell-based knockdown assay. The clinical effect of the candidate regulator was evaluated based on the overall survival and anti-PD-1 responses.

Results: We retrieved many known factors for regulating T cell exhaustion among the differentially expressed genes between PDCD1-high and PDCD1-low subsets of the TI CD8⁺ T cells in human melanoma and NSCLC. TOX was the only transcription factor (TF) predicted in both tumor types. TOX levels tend to increase as CD8⁺ T cells become more exhausted. Flow cytometry analysis revealed a correlation between TOX expression and severity of intra-tumoral T cell exhaustion. TOX knockdown in the human TI CD8⁺ T cells resulted in downregulation of PD-1, TIM-3, TIGIT, and CTLA-4, which suggests that TOX promotes intra-tumoral T cell exhaustion by upregulating IC proteins in cancer. Finally, the TOX level in the TI T cells was found to be highly predictive of overall survival and anti-PD-1 efficacy in melanoma and NSCLC.

Conclusions: We predicted the regulatory factors involved in T cell exhaustion using single-cell transcriptome profiles of human TI lymphocytes. TOX promoted intra-tumoral CD8⁺ T cell exhaustion via upregulation of IC molecules. This suggested that TOX inhibition can potentially impede T cell exhaustion and improve ICI efficacy. Additionally, TOX expression in the TI T cells can be used for patient stratification during anti-tumor treatments, including anti-PD-1 immunotherapy.

Keywords: Anti-PD-1 immunotherapy; Immune checkpoint; Intra-tumoral T cell exhaustion; Single-cell RNA sequencing.

Fig. 1

Prediction of regulatory factors involved in mediating intra-tumoral T cell exhaustion by single-cell transcriptome analysis. a Overview of the strategy used for identifying the candidate genes associated with T cell exhaustion using single-cell transcriptome profiles of TI CD8⁺ T cells. b, c Correlation between the expression levels of immune checkpoint (IC) genes and TOX with those of PDCD1, which is a marker of the exhaustion state in b melanoma (derived from GSE72056) and c non-small cell lung cancer (NSCLC) (derived from GSE99254). Individual cells that express a gene of interest at values higher than the threshold value are indicated in red in the t-stochastic neighbor embedding (tSNE) plots. d Examples of differentially expressed transcription factors (TFs) between the PDCD1-high and PDCD1-low cells in melanoma or NSCLC. The distribution patterns of gene expression in the single-cell for PDCD1-low subset and PDCD1-high subset are summarized as violin plots. The difference was tested using the Wilcoxon rank-sum test (**** P < 0.0001)

Fig. 2

Gene expression dynamics along the pseudotime of T cell exhaustion. a Single-cell trajectories across three distinct states of CD8⁺ T cells derived from human melanoma (GSE72056). The cells were classified into different T cell types using Monocle 2 based on the following criteria: effector (CD62L⁻, CD127⁻, PDCD1⁻), exhausted (PDCD1⁺), memory (either CD62L⁺ or CD127⁺), ambiguous (classified into multiple cell types), and unknown (classified into none of the cell types). The ambiguous cells and unknown cells were not visualized in the t-stochastic neighbor embedding (tSNE) plot. Based on the enriched cell type, the cells were classified into three states (of CD8⁺ T cell): effector, exhausted, and memory states (P < 2.2e−16 for exhausted state and memory state, P = 7.07e−07 for effector state by the binomial test). b Distribution of CD8⁺ T cells from each patient of the dataset across three branches of single-cell trajectories. c, d The expression dynamics of immune checkpoint (IC) genes and TOX along the pseudotime of CD8⁺ T cells in two alternative trajectories from the effector state to the memory state or to the exhausted state were summarized using BEAM analysis (c) and scatter plots with regression curves (right column for the trajectory toward exhausted state and left column for the trajectory toward memory state) (d). The significance of upregulated expression in the exhausted T cells (or memory T cells) relative to the effector T cells was tested by one-tailed Mann-Whitney U test

Fig. 3

Correlation of TOX expression with the severity of CD8⁺ T cell exhaustion in human tumors. a–c Flow cytometric analysis of the tumor-infiltrating (TI) CD8⁺ T cells isolated from human non-small cell lung cancer (NSCLC) (n = 20) and head and neck squamous cell carcinoma (HNSCC) (n = 15). a Representative plots showing the co-expression of TOX and immune checkpoint (IC) molecules (PD-1, TIM-3, and TIGIT) in the TI CD8⁺ T cells. b Percentage of TOX⁺ cells in the two subpopulations of the TI CD8⁺ T cells (expressing or not expressing a specific IC molecule). Each line in the graph indicates the data derived from the same tumor tissue of each individual patient. c TOX protein levels in the three subsets of TI CD8⁺ T cells with different severities of exhaustion, i.e., PD-1⁻TIM-3⁻ (orange), PD-1⁺TIM-3⁻ (blue), and PD-1⁺TIM-3⁺ (red). Histogram represents TOX expression level in each subset of the TI CD8⁺ T cells. Percentage of TOX-expressing cells in each subset is described in the histogram, and mean fluorescence intensity (MFI) for TOX expression in each subset is indicated in parenthesis. A dashed line represents the boundary separating the TOX protein expression. Distribution of TOX-expressing subsets of TI CD8⁺ T cells across patients is summarized in grouped scattered plots. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001. All statistical analyses were performed using unpaired Student’s t test

Fig. 4

TOX-dependent regulation of the expression of immune checkpoint (IC) molecules and the production of effector cytokines in the tumor-infiltrating (TI) CD8⁺ T cells in human non-small cell lung cancer (NSCLC). a The expression level of IC molecules in the TI CD8⁺ T cells in human NSCLC when TOX has been knocked down. b Production of IFN-γ and TNF-α in the TI CD8⁺ T cells in human NSCLC when TOX has been knocked down. Each line in the graph indicates data derived from the same tumor tissue of each individual patient. ns, not significant; *P < 0.05; **P < 0.01; ***P < 0.001. All statistical analyses were performed using paired Student’s t test

Fig. 5

TOX expression level in the tumor-infiltrating (TI) T cells can predict prognosis and anti-PD-1 therapy response. a Violin plots to depict the distribution of TOX expression levels for three groups of cells derived from melanoma: T cells, other immune cells, and cancer cells. b Correlation between TOX expression level in the T cells and TOX expression level in the CD8⁺ T cells. c Overall survival analysis of The Cancer Genome Atlas (TCGA) cohorts of patients with subcutaneous melanoma (SKCM). d Overall survival analysis of TCGA cohorts of patients with non-small cell lung cancer (NSCLC) (with only the top 25% tumor mutation burden). The patients were classified into high-TOX for those with top 30% TOX-level and low-TOX for the rest. e–g Waterfall plot to depict anti-PD-1 immunotherapy response based on three independent cohorts of patients with melanoma from Hugo et al. (e), patients with NSCLC from Jung et al. (f), and patients with NSCLC recruited from Yonsei Cancer Center (YCC) (g). The baseline represents median level of TOX expression normalized to the level in the TI T cells. P values are calculated using the two-tailed Mann-Whitney U test (testing the association of responder status with level of TOX expression in TI T cells). h Area under the receiver operating characteristics curve (AUROC) for the retrieval of responders based on the TOX expression level in the TI T cells