Spatial dynamics of CD39+CD8+ exhausted T cell reveal tertiary lymphoid structures-mediated response to PD-1 blockade in esophageal cancer

Abstract

Despite the success of immune checkpoint blockade (ICB) therapy for esophageal squamous cell cancer, the key immune cell populations that affect ICB efficacy remain unclear. Here, imaging mass cytometry of tumor tissues from ICB-treated patients identifies a distinct cell population of CD39⁺PD-1⁺CD8⁺ T cells, specifically the TCF1⁺ subset, precursor exhausted T (CD39⁺ Tpex) cells, which positively correlate with ICB benefit. CD39⁺ Tpex cells are predominantly in the stroma, while differentiated CD39⁺ exhausted T cells are abundantly and proximally within the parenchyma. Notably, CD39⁺ Tpex cells are concentrated within and around tertiary lymphoid structure (TLS). Accordingly, tumors harboring TLSs have more of these cells in tumor areas than tumors lacking TLSs, suggesting Tpex cell recruitment from TLSs to tumors. In addition, circulating CD39⁺ Tpex cells are also increased in responders following ICB therapy. Our findings show that this unique subpopulation of CD39⁺PD-1⁺CD8⁺ T cells is crucial for ICB benefit, and suggest a key role in TLS-mediated immune responses against tumors.

Fig. 1. Study design and clinical data for the 31 enrolled patients.

a Tumor, blood, and secondary lymphoid organ (SLO) collection from esophageal squamous cell carcinoma (ESCC) patients treated with nivolumab monotherapy for imaging mass cytometry (IMC), mass flow cytometry (MC), and multicolor flow cytometry (FC). Among the 31 enrolled patients, one patient (KU10) was ineligible for analysis due to no history of fluoropyrimidine-based and platinum-based chemotherapy. For more details, see the “Methods”. b Clinical data and samples analyzed for each of the 31 ESCC patients. TLS: tertiary lymphoid structure, R: responder, NR: non-responder, NE: not evaluable, NA: not applicable. c Clinical outcomes of 30 eligible ESCC patients: responders (complete response; CR + partial response; PR + stable disease; SD or non-CR/non-PD ≥ 6 months), non-responders (SD or non-CR/non-PD < 6 months + progressive disease; PD). d Representative PD-L1 immunostaining for an ESCC patient. Combined positive score (CPS) in Rs (n = 9 patients) vs. NRs (n = 13 patients); two-sided Mann-Whitney U-test (P = 0.95). Five NE cases are excluded. PD-L1 staining across 27 independent samples was validated by a pathologist to ensure accuracy. d Error bars indicate mean ± SEM. a was created in BioRender. Kenro, T. (2024) BioRender.com/s50j163. Source data are provided as a Source Data file.

Fig. 2. CD39⁺PD-1⁺CD8⁺ T cells exhibit a proliferative exhausted phenotype.

a Schematic illustration of IMC processing of tumor samples with or without tertiary lymphoid structures (TLSs). b IMC processing example for a tumor with TLSs (patient ID: KU37). The IMC images are pseudo-colored as follows: Pan-cytokeratin (PanCK) (white), CD20 (green), CD3 (blue), SMA (red), and Collagen (pink). c Spatial distribution images of PD-1⁻CD8⁺ (orange plots) and PD-1⁺CD8⁺ (blue plots) T cells, derived from three representative IMC images, pseudo-colored as follows: E-cadherin (green) and Nuclei (dark blue). Tumor parenchyma in distribution images is outlined in red. d Each fraction (%) of the annotated immune profile among total cells from the tumor-ROIs for 27 patients at pre-treatment. e Proportion (%) of PD-1⁺CD8⁺ T cells among total cells in tumor ROIs, compared between responders (R, n = 9) vs. non-responders (NR, n = 13); two-sided Mann-Whitney U-test (P = 0.0045). Five NE cases are excluded. f Two representative UMAP plots of CD8⁺ tumor-infiltrating lymphocytes (TILs) (n = 555 cells each) within the tumor-ROIs, colored by the expression of each marker. The selected UMAP plots were chosen for their clear depiction of exhaustion patterns, which were observed in a subset of cases among a total of 11 independent ROIs. g Proportion (%) of cells expressing CD39 among PD-1⁺CD8⁺ T cells in paired pre-treatment PBMCs vs. TILs; two-sided Wilcoxon matched-pairs signed-rank test (n = 27, P < 0.0001). h Median mass intensity (MMI) of Ki67 and scaled expression of TOX in CD39⁺PD-1⁺CD8⁺ TILs vs. CD39⁻PD-1⁻CD8⁺ TILs; two-sided Wilcoxon matched-pairs signed test (each n = 25 independent tumor-ROIs, P < 0.0001 for both comparisons). Two cases with cold tumors lacking CD39⁺PD-1⁺CD8⁺ T cells are excluded. e, g, h Error bars indicate mean ± SEM. e, g n refers to patients. a was created in BioRender. Kenro, T. (2024) BioRender.com/s50j163. Source data are provided as a Source Data file.

Fig. 3. CD39⁺PD-1⁺CD8⁺ T (CD39⁺PD-1⁺Tex) cells, forming a TCF-1⁺ subset linked to clinical outcomes of anti–PD-1 therapy, exhibit distinct spatial tumor organization reflective of their hierarchical patterns.

a IMC image of representative CD39⁺ precursor exhausted T (CD39⁺ Tpex, solid arrowheads: TCF1⁺CD39⁺PD-1⁺CD8⁺) and CD39⁺ differentiated exhausted T (CD39⁺ dTex, open arrowheads: TCF1⁻CD39⁺PD-1⁺CD8⁺) cells. The IMC images are pseudo-colored: Nuclei (dark blue), CD8 (blue), and CD39, PD-1, TCF-1 (yellow). Scale bar = 50 μm. b Density (cells/mm²) of CD39⁺PD-1⁺ Tex cells relative to CD39⁺ Tpex cells in the tumor (parenchyma) and stroma areas (n = 27); Pearson’s correlations using a two-sided test (r = 0.795, P < 0.0001) (no adjustments for multiple comparisons). R: circles, NR: squares, NE: triangles. c Density (cells/mm²) of CD39⁺ Tpex cells in Rs (n = 9) vs. NRs (n = 13); two-sided Mann-Whitney U-test (P = 0.0082). Five NE cases are excluded from the analysis. (d) Spatial distribution of CD39⁺ Tpex (TCF1⁺CD39⁺PD-1⁺CD8⁺, orange) and CD39⁺ dTex (TCF1⁻CD39⁺PD-1⁺CD8⁺, blue) cells, derived from the IMC image. The red line outlines the tumor parenchyma. The IMC images are pseudo-colored as follows: PanCK (white), Vimentin (dark blue), and Collagen (red). e Proportions (%) of annotated immune cells in the tumor or stroma determined from 27 patients; two-sided Wilcoxon matched-pairs signed test (n = 21, CD39⁺ Tpex vs. CD39⁺ dTex cells: P = 0.0097). Six cold tumors without CD39⁺ Tpex or CD39⁺ dTex cells are excluded from the paired analysis. f Density (cells/mm²) of CD39⁺ Tpex (TCF-1 + ) and CD39⁺ dTex (TCF-1-) cells in the tumor vs. stroma; two-sided Wilcoxon matched-pairs signed test (each n = 27, CD39⁺ Tpex: P = 0.0067, CD39⁺ dTex: P = 0.236). Stroma: circles, Tumor: squares. g Proportion (%) of TCF-1⁺ cells among CD39⁺PD-1⁺ Tex cells in the tumor vs. stroma; two-sided Wilcoxon matched-pairs signed test (n = 23, P = 0.0237). Four cold tumor cases lacking CD39⁺PD-1⁺ Tex cells in either area are excluded. c, e, f, g Error bars indicate mean ± SEM. b, c, e, f, g n refers to independent tumor-ROIs. a, d Similar staining and distribution patterns were confirmed among 27 independent tumor-ROIs. a was created in BioRender. Kenro, T. (2024) BioRender.com/s50j163. Source data are provided as a Source Data file.

Fig. 4. The hierarchical exhaustion patterns of CD39⁺PD-1⁺ Tex cells lead to differences in their proximity to the tumor parenchyma.

a Proportion (%) of stromal immune cell subsets residing within 40 μm of the parenchyma among those within 400 μm of the parenchyma (tumor) from 27 independent tumor-ROIs. b Proportion (%) of stromal immune subsets within 400μm of the tumor parenchyma from 27 patients, compared across four distance ranges (0–10 μm, 10–20 μm, 20–30 μm, 30–40 μm) using a repeated measures analysis of variance (ANOVA), followed by post-hoc tests with Bonferroni correction. The following immune subsets were analyzed: TCF-1⁺CD39⁺PD-1⁺CD8 (n = 19), TCF-1⁻CD39⁺PD-1⁺CD8 (n = 22), PD-1⁺Treg (n = 26), PD-1⁻Treg (n = 27), CD8⁺ Tnaive (n = 22), CD8⁺ Tmemory (n = 27), CD4⁺ Tnaive (n = 25), CD4⁺ Tmemory (n = 27). Cases lacking the respective stromal immune subset within 400μm of the tumor parenchyma are excluded from paired analysis. c Violin plots of median distance to the tumor of stromal immune cells from 27 patients; two-sided Wilcoxon matched-pairs signed test (n = 18, CD39⁺ Tpex vs. CD39⁺ dTex cells: P = 0.0304). Nine tumor-ROIs without stromal CD39⁺ Tpex or CD39⁺ dTex cells are excluded from the paired analysis. d IMC image of representative CD39⁺ Tpex and CD39⁺ dTex cells in the stroma. The cells marked with a square in each image are CD39⁺PD-1⁺ Tex cells within the red-colored stroma. The left cells are TCF-1⁺, and the right cells are TCF-1⁻. Similar distribution patterns were confirmed among 27 independent tumor-ROIs. *P < 0.05. b Error bars indicate mean ± SEM. b, c n refers to independent tumor-ROIs. Source data are provided as a Source Data file.

Fig. 5. The CD8⁺ T cell composition within TLSs associated with ICB effect is distinct from that in conventional tumor regions.

a Rate (%) of clinical response for TLS-detectable (n = 13 patients) vs. TLS-undetectable tumors (n = 14 patients); two-sided Fisher’s exact test (P = 0.0225). b Fractions (%) of annotated immune cells residing in TLSs per each of 12 ROIs harboring TLSs. c Comparison of the density (cells/mm²) of each immune cell between TLSs (circles) and the tumor and stroma areas (squares) (n = 12); two-sided Wilcoxon matched-pairs signed test (each n = 12, naive CD8: P = 0.001, memory CD8: P = 0.569, naive CD4: P = 0.1099, memory CD4: P = 0.3804, PD-1⁺ CD8: P = 0.0024). d The representative full TLS IMC image pseudo-colored: E-cadherin (green), CD20 (blue), CD3 (red), and Collagen (pink). The magnified images show CD39⁺ Tpex cells (open arrowheads: TCF1⁺CD39⁺PD-1⁺CD8⁺ T cells) and CD39⁺ dTex cells (solid arrowheads: TCF1⁻CD39⁺PD-1⁺CD8⁺ T cells), pseudo-colored: Nuclei (dark blue), CD8 (blue), CD39, PD-1, TCF-1 (yellow). White scale bar = 100 μm, yellow scale bar = 50 μm. *P < 0.05. c Error bars indicate mean ± SEM. c n refers to independent tumor-ROIs with TLSs. d Similar staining and distribution patterns were confirmed among the 13 independent tumor-ROIs harboring TLSs. d was created in BioRender. Kenro, T. (2024) BioRender.com/s50j163. Source data are provided as a Source Data file.

Fig. 6. Spatial features of CD39⁺ Tpex cells characterized by abundant localization in TLSs and near the stroma are distinct from those of CD39⁺ dTex cells.

a Spatial distribution of indicated CD8⁺ T cell subsets in representative IMC images with TLSs The IMC images are pseudo-colored: PanCK (white), CD20 (green), CD3 (red), Collagen (pink), and Nuclei (dark blue). b Density (cells/mm²) of CD39⁺PD-1⁺CD8⁺ T cells in TLSs vs. combined tumor and stroma areas; two-sided Wilcoxon matched-pairs signed test (n = 12). c The density (cells/mm²) of CD39⁺ Tpex cells in the tumor, stroma, and TLSs was analyzed using a repeated measures ANOVA followed by post-hoc tests with Bonferroni correction (each n = 12, TLS vs. stroma, TLS vs. tumor). d Fraction (%) of TCF-1⁺ cells among CD39⁺PD-1⁺CD8⁺ T cells in the tumor, stroma, and TLSs from 12 independent tumor-ROIs with TLSs; repeated measures ANOVA using a mixed-effect model followed by post-hoc tests with Bonferroni correction (TLS: n = 11 vs. stroma: n = 10, TLS: n = 11 vs. tumor: n = 11). Cases with missing data for CD39⁺PD-1⁺CD8⁺ T cells in any area (TLS, stroma, or tumor) were treated as missing data, not excluded. e Density (cells/mm²) of CD39⁺ Tpex and CD39⁺ dTex cells within TLSs and stromal areas at varying distances from TLSs; repeated measures ANOVA followed by post-hoc tests with Bonferroni correction (n = 12 for each). Comparisons were made between TLS and 0–50 μm, and TLS and 150-200μm for both CD39⁺ Tpex and CD39⁺ dTex cells. f Proportions (%) of stromal immune cell subsets at varying distances from the TLS, relative to total cells within 400 μm of the parenchyma in 13 tumor-ROIs with TLSs. Comparisons were made among three distance ranges (0–100 μm: circles, 100–200 μm: squares, 200–300 μm: triangles) using a two-sided Wilcoxon matched-pairs signed test (0–100 μm vs. 200–300 μm). Analyzed subsets included TCF-1⁺PD-1⁺CD8 (n = 12), TCF-1⁺CD39⁺PD-1⁺CD8 (n = 11), TCF-1⁻CD39⁺PD-1⁺CD8 (n = 11), PD-1⁺Treg (n = 12), PD-1⁻Treg (n = 13), CD39⁻PD-1⁻CD8 (naive) (n = 12), CD39⁻PD-1⁻CD8 (memory) (n = 12), naive-convCD4 (n = 13), and memory-convCD4 (n = 13). Cases lacking the respective stromal immune subset within 400μm of the TLS were excluded. g Fraction (%) of TCF-1⁺ cells among CD39⁺PD-1⁺ Tex cells in the combined tumor and stroma areas from 25 patients; two-sided Mann-Whitney U-test (13 TLS-detectable vs. 12 TLS-undetectable tumors). Two cases with cold tumors lacking CD39⁺PD-1⁺CD8⁺ T cells are excluded. *P < 0.05. b, c, d, e, f, g Error bars indicate mean ± SEM. b, c, d, e, f n refers to independent tumor-ROIs with TLSs. a Similar staining and distribution patterns were confirmed among the 13 independent tumor-ROIs harboring TLSs. Source data are provided as a Source Data file.

Fig. 7. Circulating CD39⁺Ki67⁺ Tex (CD39⁺Ki67⁺PD-1⁺CD8⁺ T) cells, partly consisting of CD39⁺ Tpex (TCF-1⁺CD39⁺Ki67⁺PD-1⁺CD8⁺ T) cells, preferentially increase in frequency 2 weeks after ICB therapy.

a Study design and treatment. b CD39⁺Ki67⁺ Tex cells among PD-1⁺CD8⁺ T cells pre- and post-treatment from 28 patients; two-sided Wilcoxon matched-pairs signed test (n = 28, pre vs. post: P = 0.0001). c Schematic illustration of MC analysis. Six patient samples were randomly analyzed. d Unsupervised clusters visualized in the UMAP plots for concatenated PD-1⁺CD8⁺ T cells (each 17,616 cells) from 6 patient PBMCs pre- and post-treatment. C1: CD39⁺ dTex, C2: central memory T (Tcm), effector memory T (Tem), C3: stem cell memory T (Tscm), C4: CD39⁺ Tpex, C5: CCR6⁺ effector memory T, C6: effector memory T, C7: terminally differentiated effector memory T (Temra), C8: central memory T, C9: Tnaive, C10: central memory T, effector memory T cells. e Comparison of the frequency (%) of 10 clusters within circulating PD-1⁺CD8⁺ T cells from six patients, analyzed by MC at pre- vs. post-treatment; two-sided Wilcoxon matched-pairs signed-rank test for each cluster (each n = 6). f Mean mass intensity of CD39, Ki67, and TCF-1 for each cluster from 12 independent PBMCs; two-sided Wilcoxon matched-pairs signed test (TCF-1, n = 6, C1 vs. C4: P = 0.0049). g Representative plots of CD39⁺ Tpex (TCF-1⁺CD39⁺Ki67⁺PD-1⁺CD8⁺) and CD39⁺ dTex (TCF-1⁻CD39⁺Ki67⁺PD-1⁺CD8⁺) cells determined by manual gating. Similar staining was observed across 12 independent PBMCs. (h) Median mass intensity marker expression of manual-gated T-naive (PD-1⁻CD27⁺CD45RA⁺), CD39⁺ Tpex, and CD39⁺ dTex cells from 12 independent PBMCs determined by heatmap analysis. *P < 0.05. b, e Error bars indicate mean ± SEM. e, f n refers to independent PBMC samples. a, c were created in BioRender. Kenro, T. (2024) BioRender.com/s50j163. Source data are provided as a Source Data file.

Fig. 8. A preferential increase in proliferative CD39⁺ Tpex cells in the blood 2 weeks after ICB therapy is correlated with clinical benefit, and intratumoral CD39⁺ Tpex cells consistently upregulate Ki67 expression following ICB therapy.

a IFN-γ and TNF production by indicated subsets from 18 patients prior to ICB therapy, determined by FC analysis. % IFN-γ and % TNF were compared using a repeated measures ANOVA followed by post-hoc tests with Bonferroni correction (each n = 18, Ki67⁺CD39⁺PD-1⁺ vs. Ki67⁺CD39⁻PD-1⁺ vs. CD39⁻PD-1⁺ T cells). Samples were randomly analyzed. b Representative SLO image of CD39⁺ Tpex cells (open arrowheads). The whole SLO ROI is pseudo-colored as follows: CD3 (green), CD20 (blue), and SMA (red). In the magnified images, the colors are as follows: CD8 (red), PD-1 (green), CD39, TCF-1, TOX (blue), and nuclei (dark blue). This staining pattern is confirmed across three independent SLO samples. White scale bar = 100 μm. c Frequency of CD39⁺ Tpex and CD39⁺ dTex cells among PD-1⁺CD8⁺ T cells pre- vs. post-treatment from 24 patients; two-sided Wilcoxon matched-pairs signed test. CD39⁺ Tpex (Rs: n = 10, NRs: n = 14), CD39⁺ dTex (Rs: n = 10, NRs: n = 14). d % CD39⁺ Tpex cells among CD39⁺Ki67⁺ Tex (total Tex) cells from 24 patients pre-treatment and post-treatment; two-sided Mann-Whitney U-test. Pre-treatment (Rs, n = 10 vs. NRs, n = 14), Post-treatment (Rs, n = 10 vs. NRs, n = 14). e Kaplan-Meier PFS for post-% CD39⁺ Tpex cells among total Tex cells from 28 patients; log-rank test (% CD39⁺ Tpex cells in total Tex cells > 10%, n = 15 vs. < 10%, n = 13). Two cases with missing post-treatment PBMC samples are excluded from the analysis. f ROIs from biopsied tumors of three patients at pre-treatment and around the time of PD on ICB therapy, pseudo-colored as follows: PanCK (white), CD20 (green), CD3 (red), E-cadherin (blue), and nuclei (dark blue). g Density (cells/mm²) of each T cell type in the tumor and stroma pre-treatment (circles) and post-treatment (squares) from three paired tumor-ROIs. h Median nuclear expression of Ki67 in each intratumoral T cell type pre-treatment (circles) and post-treatment (squares) from three paired tumor-ROIs. *P < 0.05, **P < 0.0001. a, d Error bars indicate mean ± SEM. (a, c, d, e) n refers to independent PBMC samples. c, d NE cases and those with missing PBMC samples are excluded from the analysis. h Box plots indicate median values. b was created in BioRender. Kenro, T. (2024) BioRender.com/s50j163. Source data are provided as a Source Data file.