Pan-cancer mapping of single CD8+ T cell profiles reveals a TCF1:CXCR6 axis regulating CD28 co-stimulation and anti-tumor immunity

Abstract

CD8⁺ T cells must persist and function in diverse tumor microenvironments to exert their effects. Thus, understanding common underlying expression programs could better inform the next generation of immunotherapies. We apply a generalizable matrix factorization algorithm that recovers both shared and context-specific expression programs from diverse datasets to a single-cell RNA sequencing (scRNA-seq) compendium of 33,161 CD8⁺ T cells from 132 patients with seven human cancers. Our meta-single-cell analyses uncover a pan-cancer T cell dysfunction program that predicts clinical non-response to checkpoint blockade in melanoma and highlights CXCR6 as a pan-cancer marker of chronically activated T cells. Cxcr6 is trans-activated by AP-1 and repressed by TCF1. Using mouse models, we show that Cxcr6 deletion in CD8⁺ T cells increases apoptosis of PD1⁺TIM3⁺ cells, dampens CD28 signaling, and compromises tumor growth control. Our study uncovers a TCF1:CXCR6 axis that counterbalances PD1-mediated suppression of CD8⁺ cell responses and is essential for effective anti-tumor immunity.

Keywords: CD28; CXCR6; T cell dysfunction; T cell exhaustion; TCF1; human; meta-analysis; pan-cancer; single cell.

Graphical abstract

Figure 1

Pan-cancer approach reveals highly generalizable T cell expression programs (A) Analysis approach. scRNA-seq data of CD8⁺ TILs from seven human cancers (left) were analyzed with GMDF (center) revealing shared (pan-cancer) and context (tumor)-specific expression programs (right). (B) GMDF identified 6 pan-cancer T cell expression programs. Top 100 genes (rows) of each program (color bar on the left) and their weights (W matrix) across programs from the different GMDF solutions (columns). Right: representative genes from each program. (C) Chronic activation score (y axis) and activation score (x axis) for each CD8⁺ T cell (dot) from each study (panels) with cells classified as dysfunctional, effector, or naive/memory by their signatures, or as “balanced,” if their chronic activation score is at the expected level based on the LOWESS regression line (black). Pearson’s R and association p value are shown. (D) Average expression (row Z score, color bar) of genes (rows) from the decoupled pan-cancer dysfunction program in CD8⁺ T cell subsets (columns) from 9 scRNA-seq studies stratified by expression of canonical markers (naive/memory: CCR7, TCF7, LEF1, and SELL; effector: NKG7, CCL4, CST7, PRF1, GZMA, GZMB, IFNG, and CCL3; dysfunctional: PDCD1, TIGIT, HAVCR2, LAG3, and CTLA4). (E) Distribution of overall expression scores (y axis) of the decoupled pan-cancer dysfunction program (minus immune checkpoints: PDCD1, TIGIT, HAVCR2, LAG3, and CTLA4) in CD8⁺ T cells stratified as in (D). Middle line: median; box edges: 25^th and 75^th percentiles, whiskers: most extreme points that do not exceed $\pm$ IQR∗1.5; further outliers are marked individually. ∗∗∗p < 0.001, mixed-effects test.

Figure 2

The pan-cancer T cell dysfunction program predicts ICB responses (A) Uniform manifold approximation and projection (UMAP) embedding of CD8⁺ T cell profiles (dots) from 48 melanoma tumors, colored by the overall expression of the decoupled pan-cancer naive/memory (I), effector (II), and dysfunction (III), or pan-cancer cell cycle (IV) program, ICB response status (V), or pre- versus post-treatment biopsy status (VI). (B) Left: true positive (y axis) and false positive (x axis) rates when predicting the clinical response of the tumor based on different levels of the pan-cancer dysfunction program score at either the CD8⁺ T cell (top) or sample (bottom) level. Right: distribution of the overall expression (y axis) of the decoupled pan-cancer dysfunction program in responders or non-responders at the cell (top) and sample (bottom) level. Middle line: median; box edges: 25^th and 75^th percentiles, whiskers: most extreme points that do not exceed $\pm$ IQR∗1.5; further outliers are marked individually. (C) Distribution of overall expression of the decoupled pan-cancer dysfunction program in the responding and non-responding lesions of patients (x axis) with mixed responses. Statistical significance was determined by Student's t test (B, right) and Fisher combined test (C).

Figure 3

CXCR6 and CXCL16 expression in human and murine tumors tracks with dysfunctional T cells and myeloid cells, respectively (A) Mean expression (color bar) and fraction of expressing cells (dot size) for CXCR6 and CXCL16 (columns) across cell types (rows) in different human tumor studies (panels). (B) GSEA plots obtained for the KEGG_ANTIGEN_PROCESSING_AND_PRESENTATION gene set when compared against the ranking of genes based on their co-expression with CXCL16 in macrophages in the indicated tumors. (C) Frequency of CXCR6⁺ cells (y axis, mean ± SEM) in the indicated subsets (x axis) of total (top) or Ova-Dextramer⁺ (bottom) CD8⁺ TILs harvested from B16F10, B16Ova, or Mc38Ova^hi tumors (top: B16F10, n = 8, 2 experiments combined. B16Ova, n = 10, 2 experiments combined. Mc38Ova^hi, n = 9, 2 experiments combined. Bottom: B16Ova, n = 3, 1 experiment. Mc38Ova^hi, n = 15, 2 experiments combined). (D) Representative distributions of expression levels (x axis, fluorescence intensity) in CXCR6⁺ and CXCR6⁻ CD8⁺ TILs from B16Ova tumors (n = 4, 1 experiment). (E) Representative distributions of CXCL16 surface (top) or intracellular (bottom) expression (x axis, fluorescence intensity) in myeloid cells from Mc38Ova^hi, B16Ova, or B16F10 tumors (n = 4 per tumor, 1 experiment). (F) Left: B16Ova tumor area (y axis, mean ± SEM) of early-stage or late-stage tumors. Middle: frequency (y axis, mean ± SEM) of PD1- and TIM3-expressing CD8⁺ TILs (x axis) in early- or late-stage tumors. Right: frequency of CXCR6⁺ cells (y axis, mean ± SEM) in the indicated subsets (x axis) of CD8⁺ TILs from early- or late-stage B16Ova tumors (n = 12, 2 experiments combined). Statistical significance was determined by Student’s unpaired t test (C, F). ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

Figure 4

CXCR6 expression increases upon ICB and is repressed by TCF1 (A and B) Left: tumor area (y axis, mean ± SEM) over time (x axis) of Mc38Ova^hi (A) or B16Ova (B) implanted in wild-type (WT) mice and treated with anti-PD1 (A), anti-PD-L1 + anti-TIM3 (B), or isotype control. Right: frequency of CXCR6⁺ cells (y axis, mean ± SEM) in the indicated subsets (x axis) of CD8⁺ TILs harvested from tumors treated as above (A: n = 4–6 per group, 1 experiment. B: n = 3–4 per group, 1 experiment). (C) Expression (row-normalized TPM, color bar) of top differentially expressed genes (rows) between WT OTI (E8i-Cre^-, Tcf7^FL/FL) and Tcf7 cKO OTI (E8i-Cre⁺, Tcf7^FL/FL) cells at different time points after T cell activation (columns) (n = 3, 1 experiment). (D) Mean expression (color bar) and fraction of expressing cells (dot size) of key genes (rows) in different CD8⁺ T cell clusters (columns) as determined by scRNA-seq of cells from B16Ova tumors implanted in WT or Tcf7 cKO mice (n = 3 per group combined, 1 experiment). (E) Frequency of CXCR6⁺ cells (y axis, mean ± SEM) in the indicated subsets (x axis) of CD8⁺ TILs from B16Ova tumors from WT (black) or Tcf7 cKO (green) mice (n = 3–7 per group, representative of 2 experiments). (F) Percent input (y axis, mean ± SEM) following chromatin immunoprecipitation (ChIP) PCR of the Cxcr6 locus with anti-TCF1 or rabbit IgG control antibodies (x axis) in WT or Tcf7 cKO CD8⁺ T cells (n = 8, 5 experiments combined). (G–I) Luciferase activity (RLU, relative light unit, y axis, mean ± SEM) in HEK293T cells transfected with Cxcr6 locus-containing pGL4.10 luciferase reporters together with either empty vector (control) or vectors encoding the indicated transcription factors (x axis). Firefly luciferase activity is presented relative to constitutive Renilla luciferase activity (n = 3, representative of 2 experiments). Statistical significance was determined by linear mixed model (A left, B left), Student’s unpaired t test (A right, B right, E, F comparing WT to Tcf7 cKO anti-TCF1 samples), Student’s paired t test (F comparing WT samples), or one-way ANOVA with Tukey’s multiple comparisons test (G, H, I). NS = not significant, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001, ∗∗∗∗p < 0.0001.

Figure 5

CXCR6 knockout reduces anti-tumor immunity (A) Experimental design for adoptive transfer experiments. (B–F) B16Ova tumor-bearing mice were adoptively transferred with CRISPR CXCR6 KO OTI T cells, CRISPR control OTI T cells, or no transfer control as indicated in (A). (B) Tumor area (y axis, mean ± SEM) over time (x axis) (n = 5 per group, representative of 2 experiments). (C) Number of transferred, transduced OTI cells per mg of tumor tissue (y axis, mean ± SEM) (n = 5 per group, representative of 2 experiments). (D) Left: frequency of IFNγ⁺, TNF-α⁺, or Granzyme B⁺CD107a⁺ cells (y axis, mean ± SEM) in transferred, transduced OTI cells after ex vivo activation with Ova 257–264. Right: frequency of Ki-67⁺ cells (y axis, mean ± SEM) in transferred, transduced OTI cells (n = 4–5 per group, representative of 2 experiments). (E) Frequency of Tox⁺ cells (y axis, mean ± SEM) in the indicated subsets (x axis) of transferred, transduced OTI cells (n = 4 per group, 1 experiment). (F) Frequency of CX3CR1⁺ cells (left, y axis, mean ± SEM) and CX3CR1 expression level in CX3CR1⁺ cells (right, y axis, geometric mean fluorescence (MFI), mean ± SEM) in the indicated subsets (x axis) of transferred, transduced OTI cells (n = 4–5 per group, representative of 2 experiments). (G–I) B16Ova-tumor bearing mice were adoptively transferred with CXCR6 knockout (KO) OTI cells, WT OTI cells, or no T cells. (G) Tumor area (y axis, mean ± SEM) over time (x axis) (n = 4–5 per group, 1 experiment). (H) Frequency of Bcl-2⁺ cells (left, y axis, mean ± SEM) and level of Bcl-2 in Bcl-2⁺ cells (right, y axis, geometric mean fluorescence (MFI), mean ± SEM) in the indicated subsets (x axis) of transferred OTI cells (n = 4 per group, 1 experiment). (I) Frequency of Annexin V⁺, 7AAD⁻ cells (left, y axis, mean ± SEM) and of Annexin V⁺, 7AAD⁺ cells (right, mean ± SEM) in the indicated subsets (x axis) of transferred OTI cells (n = 4 per group, 1 experiment). Statistical significance was determined by linear mixed model (B, G) and Student’s unpaired t test (C–F, H, I). NS = not significant, ∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001.

Figure 6

CXCR6 KO cells have reduced CD28 co-stimulation (A) Heatmap showing differentially expressed genes (rows, Z score) between WT and CXCR6 KO OTI T cells (columns) sorted from T cell:DC co-cultures. Genes representative of the top 3 enriched pathways in WT and CXCR6 KO cells are labeled (n = 3, 1 experiment). (B) The top 3 significantly enriched pathways based on genes that were significantly higher in WT compared to CXCR6 KO OTI T cells from T cell:DC co-cultures. (C) The top 3 significantly enriched pathways based on genes that were significantly higher in CXCR6 KO compared to WT OTI T cells from T cell:DC co-cultures. (D) GSEA plot for the “costimulation by the CD28 family” (R-MMU-388841) gene set when compared against the ranking of genes based on their differential expression between WT and CXCR6 KO T OTI cells from T cell:DC co-cultures. (E) Experimental design for co-adoptive transfer experiments. (F–I) WT and CXCR6 KO OTI T cells were co-adoptively transferred into B16Ova tumor-bearing mice as indicated in (E) (n = 5–6 per time point, 1 experiment). (F) Frequency (y axis, mean ± SEM) of transferred OTI cells on day 3 (D3) and day 8 (D8) post-transfer (x axis). (G) Expression level (left, x axis, fluorescence intensity) of CD28 in WT and CXCR6 KO OTI CD8⁺ T cells versus fluorescence minus one (FMO, gray) prior to co-adoptive transfer. Frequency of CD28⁺ cells (middle, y axis, mean ± SEM) and CD28 expression level in CD28⁺ cells (right, y axis, geometric mean fluorescence [MFI]) in transferred OTI cells at day 3 (D3) and day 8 (D8) post-transfer. (H) Expression level (left, x axis, fluorescence intensity) of Bcl-xL in WT and CXCR6 KO OTI CD8⁺ T cells versus FMO (gray) prior to co-adoptive transfer. Frequency of Bcl-xL⁺ cells (middle, y axis, mean ± SEM) and Bcl-xL expression level in Bcl-xL⁺ cells (right, y axis, geometric mean fluorescence [MFI]) in transferred OTI cells at day 3 (D3) and day 8 (D8) post-transfer. (I) Expression level (left, x axis, fluorescence intensity) of CX3CR1 in WT and CXCR6 KO OTI CD8⁺ T cells versus FMO (gray) prior to co-adoptive transfer. Frequency of CX3CR1⁺ cells (middle, y axis, mean ± SEM) and CX3CR1 expression level in CX3CR1⁺ cells (right, y axis, geometric mean fluorescence [MFI]) in transferred OTI cells at day 3 (D3) and day 8 (D8) post-transfer. Statistical significance was determined by Student’s paired t test (F–I). NS = not significant, ∗p < 0.05, ∗∗p < 0.01.