A comprehensive profile of TCF1+ progenitor and TCF1− terminally exhausted PD-1+CD8+ T cells in head and neck squamous cell carcinoma: implications for prognosis and immunotherapy

The heterogeneity of exhausted T cells (Tex) is a critical determinant of immune checkpoint blockade therapy efficacy. However, few studies have explored exhausted T cell subpopulations in human cancers. In the present study, we examined samples from two cohorts of 175 patients with head and neck squamous cell cancer (HNSCC) by multiplex immunohistochemistry (mIHC) to investigate two subsets of Tex, CD8+PD1+TCF1+ progenitor exhausted T cells (TCF1+Texprog) and CD8+PD1+TCF1− terminally exhausted T cells (TCF1−Texterm). Moreover, fresh tumor samples from 34 patients with HNSCC were examined by flow cytometry and immunohistochemistry to further investigate their properties and cytotoxic capabilities and their correlation with regulatory T cells (Tregs) in the tumor immune microenvironment (TIME). mIHC and flow cytometry analysis showed that TCF1−Texterm represented a greater proportion of CD8+PD1+Tex than TCF1+Texprog in most patients. TCF1+Texprog produced abundant TNFα, while TCF1−Texterm expressed higher levels of CD103, TIM-3, CTLA-4, and TIGIT. TCF1−Texterm exhibited a polyfunctional TNFα+GZMB+IFNγ+ phenotype; and were associated with better overall survival and recurrence-free survival. The results also indicated that larger proportions of TCF1−Texterm were accompanied by an increase in the proportion of Tregs. Therefore, it was concluded that TCF1−Texterm was the major CD8+PD1+Tex subset in the HNSCC TIME and that these cells favor patient survival. A high proportion of TCF1−Texterm was associated with greater Treg abundance.


INTRODUCTION
Therapeutic advances in immune checkpoint blockade (ICB) therapy have rapidly emerged in the past few years, yielding gratifying results related to the treatment of several kinds of tumors. The programmed death-1 (PD-1)-blocking antibody nivolumab was the first immune checkpoint inhibitor approved by the United States Food and Drug Administration (FDA) for metastatic head and neck squamous cell cancer (HNSCC), receiving approval in 2016 on the basis of the results of the CheckMate 141 trial. 1 However, due to the complex components of the tumor immune microenvironment (TIME), the overall response rates to ICB therapies for HNSCC range from just 19 to 33%. [1][2][3] The HNSCC TIME is populated by dysfunctional immune cells, inhibitory cytokines, and exosomes carrying suppressive ligands. 4,5 Among all immune cell phenotypes, cytotoxic T cell exhaustion is a hallmark of many cancers and strongly affects ICB therapy.
The term "T cell exhaustion", which was originally derived from a mouse chronic virus infection model, is now commonly used to define a dysfunctional status of T cells stimulated by continuous antigen burden. 6 The characteristics of exhausted T cells are a high expression of inhibitory receptors, reduced cytotoxic activity, and reduced proliferation potential. 6,7 Recently, researchers have made progress in understanding exhausted CD8 + T cells (Tex). Two subpopulations of Tex were found in the TIME, progenitor exhausted T cells (Tex prog ) and terminally exhausted T cells (Tex term ), and these subpopulations are characterized by distinct epigenetic and transcriptional states. [8][9][10][11] They are distinguished by the expression of TCF1, which is encoded by TCF7 and plays a fundamental role in T cell development, stemness, and memory formation. [12][13][14] TCF1 was highly expressed in Tex prog cells, but not in Tex term , thus TCF1 expression could unambiguously identify Tex prog cells. 13 CD8 + PD1 + TCF1 + Tex prog (TCF1 + Tex prog ) possessed stem-like properties and could be differentiated into Tex term but exhibited relatively lower cytotoxicity. CD8 + PD1 + TCF1 − Tex term (TCF1 − Tex term ) expressed more cytokines and exerted strong antitumor activity. In particular, TCF1 + Tex prog were discovered within the tumor region and lymph node and responded preferentially to checkpoint blockade in melanoma. 8,10,15,16 Previous studies on HNSCC focused mostly on whole tumorinfiltrating CD8 + T cells and suggested that the abundance of infiltrating CD8 + T cells was associated with better overall survival (OS). [17][18][19] Further, a subset of CD8 + CD103 + tissue-resident memory T cells (T RM ) are critical for protective immunity in HNSCC. 20,21 However, previous studies have shown that the prognostic value of Tex is controversial. In hepatocellular carcinoma, PD1 hi CD8 + Tex was associated with poor outcomes, whereas in triple-negative breast cancer, they were associated with a better prognosis. [22][23][24][25] A possible explanation might be that the TIME consists of distinct Tex subsets that exhibit distinct cytokine production and phenotypic characteristics. Therefore, to further understand the heterogeneity and prognostic value of PD1 + CD8 + Tex, we aimed to explore the profiles of TCF1 + Tex prog and TCF1 − Tex term in HNSCC.
In this study, we applied multiplex immunohistochemistry (mIHC) and flow cytometry to investigate the comprehensive profile of TCF1 + Tex prog and TCF1 − Tex term in HNSCC, including their spatial distribution, prognostic value, cytokine secretion, and correlation with regulatory T cells (Tregs) in the TIME. We found that CD8 + PD1 + TCF1 + Tex prog and CD8 + PD1 + TCF1 − Tex term coexisted in the HNSCC TIME and that the latter was the major subset. The density of tumor-infiltrating TCF1 − Tex term could serve as an independent prognostic indicator in HNSCC. In addition, the level of infiltrating Tregs increased as the proportion of TCF1 − Tex term became higher. We propose that these findings could provide new strategies and indications for optimal immunotherapeutic regimens for HNSCC patients.

RESULTS
TCF1 − Tex term was the major PD1 + CD8 + Tex subset in the HNSCC TIME A five-color mIHC scheme that included anti-PanCK (brown), anti-CD8 (red), anti-PD-1 (green), anti-TCF1 (yellow), and DAPI (blue) was developed to comprehensively profile the spatial distribution of TCF1 + Tex prog and TCF1 − Tex term . In this scheme, CD8 + PD-1 + TCF1 + cells represented TCF1 + Tex prog , and CD8 + PD-1 + TCF1 − cells were TCF1 − Tex term . 8 Interestingly, these two subtypes were both observed in formalin-fixed, paraffin-embedded (FFPE) samples from HNSCC patients (Fig. 1a). Then, we quantitated the numbers of these two subtypes by StrataQuest software. 26 We found that TCF1 − Tex term were significantly more abundant than TCF1 + Tex prog in the total tumor region (Fig. 1b). To further analyze the spatial distributions of these subpopulations, we deeply classified the intratumoral and stromal regions by PanCK staining (Fig. S1a and Fig. 1a). The results showed that both TCF1 + Tex prog and TCF1 − Tex term were significantly enriched in the stroma rather than the intratumoral region (Fig. 1c). In addition, the number of stromal TCF1 − Tex term was significantly higher than that of stromal TCF1 + Tex prog , but there was no statistical significance in the intratumoral region (Fig. S1b, c).
We further performed flow cytometry to investigate these two types of PD1 + CD8 + Tex in 34 fresh HNSCC tumor samples and 10 paired draining lymph nodes (dLNs). The clinicopathologic characteristics of HNSCC patients are shown in Table S1. The flow cytometry gating strategy is displayed in Fig. S1d. We found significantly more CD8 + PD-1 + TCF1 − Tex term than CD8 + PD-1 + TCF1 + Tex prog in HNSCC tumor samples which were consistent with the mIHC results (Fig. 1d), whereas, in dLN, the proportion of TCF1 + Tex prog was significantly higher than that of TCF1 − Tex term (Fig. 1e). We tried to explore the correlation of Tex subsets between the tumor and dLN. The results showed that TCF1 − Tex term in dLNs was positively associated with TCF1 − Tex term in tumors, while there was no correlation for TCF1 + Tex prog (Fig. 1f). All these results indicated that TCF1 − Tex term was the major Tex subpopulation in the HNSCC TIME and the proportion of TCF1 − Tex term in TIME was correlated with that in dLN. TCF1 -Tex term conferred better prognosis in HNSCC As CD8 + PD-1 + TCF1 + Tex prog and CD8 + PD-1 + TCF1 − Tex term were detected in HNSCC, we investigated their prognostic value in the Sun Yan-Sen University (SYSU) cohort with 102 patients (patient clinicopathological information is presented in Table S2), based on the density of these two subsets in the mIHC results. We found that the abundance of TCF1 − Tex term rather than TCF1 + Tex prog was associated with better OS (Fig. 2a, b) and recurrence-free survival (RFS, Fig. S2a, b). Then, we performed univariate and multivariate OS analyses of multiple clinicopathological parameters in the SYSU cohort. The results showed that among these variables, nodal invasion, clinical-stage, radiotherapy, and TCF1 − Tex term density were favorable for OS in the univariate log-rank test (Table S3). Then, these variables were entered into multivariate Cox regression analyses. In multivariate analyses, the TCF1 − Tex term density was an independent prognostic factor for OS (Table 1).
To validate the prognostic value of the TCF1 − Tex term population, we further applied a validation cohort with 73 patients (Zhejiang University cohort, ZJU cohort, patient clinicopathological information is given in Table S2). Consistent with the results from the SYSU cohort, a high density of TCF1 − Tex term also predicted a better OS and RFS in the ZJU cohort, while there was no statistically significant prognostic difference between patients with high and low densities of TCF1 + Tex prog (Fig. 2c, d, Fig. S2c, d). Univariate and multivariate OS analyses of multiple clinicopathological parameters in the ZJU cohort also demonstrated that the TCF1 − Tex term density was indicative of prognosis (Table S3 and Table 1). These results indicated that TCF1 − Tex term density could serve as an independent prognostic factor for HNSCC. TCF1 + Tex prog and TCF1 − Tex term exhibited distinct cytokine production and phenotypic characteristics in HNSCC Based on the different prognostic values of TCF1 + Tex prog and TCF1 − Tex term , we investigated the cytokine production and phenotypic profiles of these two subsets in fresh tumors. The production capability of interferon-γ (IFNγ), tumor necrosis factorα (TNFα), and granzyme B (GZMB) was analyzed. TNF-α, IFN-γ, and GZMB are important cytokines that play a critical role in antitumor response. However, there are differences in their functions. TNF-α could promote T cell proliferation and survival and GZMB acts as exocytosis of granule components in the direction of the target cell, leading to a lethal hit of the target cell. We found that GZMB production was significantly higher in CD8 + PD-1 + TCF1 − Tex term than in CD8 + PD-1 + TCF1 + Tex prog (Fig. 3a, b). However, there was no significant difference in IFNγ generation between these two subsets ( Fig. 3c). In contrast to the case for GZMB, the production of TNFα was significantly higher in TCF1 + Tex prog than in TCF1 − Tex term (Fig. 3d). Moreover, TCF1 − Tex term showed a higher proportion of cells with the TNFα + GZMB + IFNγ + polyfunctional phenotype (Fig. 3e). It might indicate that highly TNFα-expressing TCF1 + Tex prog might associate with its property of proliferation and TCF1 − Tex term produced higher GZMB, showing a direct attack on the tumor cells.
In addition, we evaluated the expression of CD103, T cell immunoglobulin and mucin-domain containing-3 (TIM-3), cytotoxic T lymphocyte antigen 4 (CTLA-4), lymphocyte activation gene-3 (LAG-3), and T cell immunoglobulin and ITIM domain (TIGIT) in these two subsets by flow cytometry. CD103 is a marker of T RM that binds to E-cadherin on epithelial cells, and TIM-3, CTLA-4, LAG-3, and TIGIT function as coinhibitors. 5,6 The flow cytometry results revealed that TCF1 − Tex term expressed significantly higher levels of CD103 than that of TCF1 + Tex prog (Fig. 3f). We also found that the mean fluorescence intensity (MFI) of TIM-3, CTLA-4, and TIGIT was stronger in TCF1 − Tex term , but there was no significant difference in the expression of LAG-3 (Fig. 3g, h). These data suggested that TCF1 + Tex prog produced a higher level of TNFα, and although TCF1 − Tex term expressed high levels of immune  Fig. 1 CD8 + PD-1 + TCF1 + progenitor exhausted T cells (TCF1 + Tex prog ) and CD8 + PD-1 + TCF1 − terminally exhausted T cells (TCF1 − Tex term ) in the head and neck squamous cell carcinoma (HNSCC) microenvironment. a Representative images of HNSCC multiplex immunohistochemistry (mIHC) slides showing CD8 + PD-1 + TCF1 + Tex prog (yellow arrow) and CD8 + PD-1 + TCF1 − Tex term (white arrow) cells. b Quantitative analysis revealed that the density of TCF1 − Tex term cells was higher than that of TCF1 + Tex prog in the whole tumoral region (n = 102). c Quantitative analysis revealed that progenitor and terminally exhausted T cells were more abundant in the stromal region (n = 102). d Representative contour plots of TCF1 + Tex prog and TCF1 − Tex term in tumor by flow cytometry and quantitative analysis showed that TCF1 − Tex term were more abundant than TCF1 + Tex prog in tumors (n = 34). e Representative contour plots of TCF1 + Tex prog and TCF1 − Tex term in draining lymph nodes (dLNs) by flow cytometry and quantitative analysis showed that TCF1 + Tex prog were more abundant than TCF1 − Tex term in dLNs (n = 10). f Flow cytometry results showed that TCF1 − Tex term in dLNs was positively associated with TCF1 − Tex term in tumors, while there was no correlation for TCF1 + Tex prog (n = 10). Paired t tests were performed for the data in (b-e). Spearman's correlation test for (f). *P < 0.05, **P < 0.01. Scale bars: 200 μm for left (a) and 50 μm for right (a) checkpoint receptors, it also expressed higher levels of GZMB, exerting a stronger antitumor response.
A high proportion of TCF1 − Tex term was associated with more Treg infiltration Tregs are increased within tumor sites of various tumor types in humans and mice, limiting the antitumor immune response. 27 Tregs promote intratumoral T cell exhaustion by modulating the expression of several coinhibitors and the exhaustion-associated transcriptomic signature of CD8 + T cells. 28 We, therefore, used flow cytometry to investigate the association between the percentages of Tregs and the ratios of TCF1 + Tex prog to TCF1 − Tex term which represented the status of PD1 + CD8 + Tex proportion. We found that, although not statistically significant, the proportion of CD4 + Foxp3 + Tregs increased as the ratio of TCF1 + Tex prog /TCF1 − Tex term decreased (Fig. 4a). Then, we further divided these patients into two groups according to the TCF1 + Tex prog /TCF1 − Tex term ratio in tumors, with the median value as a cutoff. Tumors with a TCF1 + Tex prog /TCF1 − Tex term ratio ≥ 0.6 were defined as the high ratio group and those with a TCF1 + Tex prog /TCF1 − Tex term ratio < 0.6 were defined as the low ratio group. We found that the levels of tumor-infiltrating Tregs were significantly higher in the low ratio group than in the high ratio group (Fig. 4b, c). To confirm the flow cytometry results, we further stained Foxp3 in FFPE samples from these patients. In accordance with our findings, Foxp3 was significantly expressed at higher levels in low ratio patients (Fig. 4d, e).
Together, these results indicated that a high proportion of TCF1 − Tex term was associated with more tumor-infiltrating Tregs, which might constrain the persistent antitumor activity of TCF1 − Tex term .

DISCUSSION
Recent studies have revealed that there are at least two distinct subsets of Tex in melanoma: CD8 + PD-1 + TCF1 + stem-like Tex prog and CD8 + PD-1 + TCF1 − Tex term . 8,10,15 However, these Tex subsets have not been explored in the context of HNSCC. In this study, we first explored the comprehensive profile of TCF1 + Tex prog and TCF1 − Tex term in HNSCC, paving the way to maximize the function of Tex reinvigoration by ICB therapy. We identified that TCF1 + Tex prog and TCF1 − Tex term were both detected in HNSCC by mIHC and flow cytometry. In most HNSCC patients, TCF1 − Tex term were more abundant than TCF1 + Tex prog in tumors, consistent with the results in melanoma. 29 This is partially due to the higher expression of CD103 on TCF1 − Tex term . However, TCF1 + Tex prog was the main Tex subtype in dLNs. It has been reported that tumor dLNs could act as key orchestrators of antitumor responses induced by ICB therapy and generate the TCF1 + Tex prog that seed the tumor. 16 We also found that TCF1 -Tex term expressed more GZMB and displayed a phenotype of greater polyfunctional TNFα + GZMB + IFNγ + production, whereas TCF1 + Tex prog could produce more TNFα which plays an important role in T cell proliferation and survival. 30 Fig. 2 The density of CD8 + PD-1 + TCF1 − terminally exhausted T cells (TCF1 − Tex term ) indicated a favorable prognosis in head and neck squamous cell carcinoma (HNSCC) based on the density from mIHC results. a There was no significant overall survival (OS) analysis according to high and low densities of CD8 + PD-1 + TCF1 + progenitor exhausted T cells (TCF1 + Tex prog ) in the SYSU cohort. b High density of TCF1 − Tex term predicted a better OS in the SYSU cohort. c There was no significant OS analysis according to high and low densities of TCF1 + Tex prog in the ZJU cohort. d High density of TCF1 − Tex term indicated a better OS in the ZJU cohort. The log-rank test was used for (a-d) Survival analysis revealed that a high abundance of TCF1 − Tex term rather than TCF1 + Tex prog predicted better OS and RFS in two clinical cohorts. However, the TCF1 + Tex prog levels were especially associated with better tumor regression and better overall response in ICB-treated patients. 8,10,32 Patients with melanoma who had a higher percentage of TCF1 + Tex prog respond to ICB therapy for a longer duration. Strategies aiming to increase the levels of progenitor and terminally exhausted T cells were proposed by some researchers. Baharom et al. applied a selfassembling nanoparticle vaccine that linked neoantigen peptides to a Toll-like receptor 7/8 agonist (SNP-7/8a) to induce a higher proportion of CD8 + PD-1 + TCF1 + Tex prog , resulting in a superior antitumor response in combination with a PD-L1 blockade. 33 In addition, Ptpn2 deletion in CD8 + T cells enhanced the generation and cytotoxicity of TCF1 − Tex term , promoted tumor control, and improved PD-1 checkpoint blockade responses to B16 tumors. 34 Even though progenitor and terminally exhausted T cells played important roles in the antitumor response, we noticed that CD4 + Foxp3 + Tregs were elevated as TCF1 + Tex prog /TCF1 − Tex term ratio decreased. Previous studies reported that high levels of tumorinfiltrating Foxp3 + Tregs were associated with better prognosis and locoregional control in HNSCC. [35][36][37] However, Foxp3 is a crucial regulator of Tregs that is responsible for a substantial amount of their suppressive potential. 38 This result indicates that the good prognosticator, Foxp3 + Tregs, might be accompanied by high levels of TCF1 -Tex term . Foxp3 expression correlated with decreased TCF1 and reduced accessibility of TCF1-bound chromatin regions. 39 Tregderived IL-10 and IL-35 can also directly induce the expression of multiple coinhibitors and drive T cell dysfunction. 28 This result suggested that Tregs might affect the formation of TCF1 + Tex prog through transcriptional regulation, resulting in an imbalance in Tex subtypes. In addition, we found that TCF1 − Tex term expressed higher levels of TIM-3, CTLA-4, and TIGIT, which are markers of a dysfunctional subset of tumor-infiltrating CD8 + PD1 + T cells in cancer. In a transgenic HNSCC mouse model, anti-TIM-3 monoclonal antibody injection induced a reduction in Tregs and increased IFN-γ production on CD8 + T cells, suggesting that the antitumor immune response was enhanced by TIM-3 blockade through relieved suppression of negative immune factors. 40 Therefore, increased Tregs and upregulated inhibitory receptors on TCF1 − Tex term might construct an immunosuppressive microenvironment in the TIME, resulting in a barrier to the efficacy of immunotherapy. These findings indicated that ICB therapy in combination with Treg deletion might serve as a new strategy to achieve a stronger T cell immune response in the TIME when designing immunotherapy regimens for patients with HNSCC, especially for those with TCF1 + Tex prog /TCF1 − Tex term ratios lower than 0.6.
In summary, the present study first identified TCF1 − Tex term as the major subtype in the TIME and then showed that this subtype is indicative of a favorable prognosis in HNSCC. The TCF1 − Tex term in the tumor environment was correlated with that in dLNs, which showed a higher proportion of Tex prog . Moreover, we found that a high proportion of TCF1 − Tex term was associated with more tumorinfiltrating Tregs. Overall, our study might provide a theoretical basis for optimal ICB immunotherapy in HNSCC. Immunohistochemistry (IHC) IHC staining was performed on FFPE samples as previously described. 41,42 Four-micrometer slides were undergone the following procedures: deparaffinized in xylene, rehydrated in graded alcohol, blocked in 3% hydrogen peroxide, antigen retrieval in Tris-EDTA buffer (pH 8.0). Next, the slides were incubated with Foxp3 (#98377S, Cell Signaling Technology, 1:100) primary antibodies at 4°C overnight. After being treated with HRP goat anti-rabbit secondary antibody, the slides were stained 3,3diaminobenzidine (DAB). Then Aperio AT2 scanner (Leica) was applied to scan the stained slides. At least three random interested areas (contained both intratumoral and stromal areas) of every section were chosen for quantification. Mean Foxp3 Hscores were quantified with background subtraction using Aperio eSlide Manager Quantification software. 43 mIHC Opal 7-color fluorescent IHC kit (NEL811001KT, PerkinElmer) was used to stain the multi-parametric immunofluorescence panel. First, an initial deparaffinization procedure was performed on all slides, followed by fixation in 10% formalin fixation and Tris-EDTA Tex prog CD8 + PD-1 + TCF1 + progenitor exhausted T cells, Tex term CD8 + PD-1 + TCF1 − terminally exhausted T cells, SYSU cohort patients with HNSCC who were treated with surgery at Sun Yat-Sen University, ZJU cohort patients with HNSCC who were treated with surgery at Zhejiang University, HR hazard ratio, 95% CI 95% confidence interval,not included, * P < 0.05

Patients and study cohorts
Two subsets of exhauted T cells in HNSCC Wang et al.   Fig. 3 Differences in cytotoxicity activation molecules and cell surface marker expression between CD8 + PD-1 + TCF1 + progenitor exhausted T cells (TCF1 + Texprog) and CD8 + PD-1 + TCF1 − terminally exhausted T cells (TCF1 − Tex term ) in head and neck squamous cell carcinoma (HNSCC) tumor tissues were detected using flow cytometry. a Representative contour plots of selected cytokine markers (TNFα, GZMB, and IFNγ) gated on TCF1 + Tex prog or TCF1 − Tex term . b TCF1 − Tex term showed higher GZMB production (n = 20). c There was no significant difference in IFNγ expression between these two subsets of exhausted T cells (n = 20). d TCF1 + Tex prog showed a higher TNFα production (n = 20). e TCF1 − Tex term displayed a higher proportion of cells with the polyfunctional phenotype of TNFα + GZMB + IFNγ + (n = 20). f Representative contour plots of CD103 and comparison of CD103 expression among TCF1 + Tex prog and TCF1 − Tex term (n = 24). g Mean fluorescence intensity (MFI) of TIM-3, CTLA-4, TIGIT, and LAG-3 gated on TCF1 + Tex prog or TCF1 − Tex term . h Quantitative analysis of TIM-3 (n = 24), CTLA-4 (n = 10), TIGIT (n = 10), and LAG-3 (n = 10) MFI between TCF1 + Tex prog and TCF1 − Tex term . Paired t tests were performed for (b-f and h). *P < 0.05, **P < 0.01, ****P < 0.000 1, ns P > 0.05 (pH = 9.0) antigen retrieval. Afterward, the slides were incubated with primary antibody, secondary-HRP antibody, and Opal TSA dyes. Subsequently by the other cycles of staining consisted of antigen retrieval, blocking, primary antibody and secondary-HRP antibody incubation, and Opal TSA dye staining. Finally, the slides were mounted with Antifade Reagent (AR1109, BOSTER). DAPI was used for the nuclear counterstain. Images of the whole tissue specimens were acquired by the TissueFAXS platform (TissueGnostics). TCF1 + Tex prog and TCF1 − Tex term phenotypes were identified and quantitated using the tissue cytometry of StrataQuest software (V7.0.1.140, TissueGnostics). 44 Threshold cut-offs depicted by horizontal and/or vertical lines were determined manually by backgating to the greyscale and multiplexed tissue images for each biomarker (CD8, PD-1, TCF1). 45 The Opal detection fluorophores for exhausted T cells analysis Tissue dissociation Tissue dissociation was performed as the previous study. 23 Tumor tissues were cut into several pieces. One-piece was fixed in 10% formalin for IHC or mIHC verification and the other pieces were incubated with 1 mg·mL −1 Collagenase IV, 0.5 mg/mL Collagenase I (Gibco) and 0.1 mg·mL −1 DNase I (Biofroxx) in complete RPMI 1640 (Gibco) medium with 10% FBS for 30 min at 37°C. Digested tissues were filtered using a 70 μm cell strainer. Lymph nodes   Fig. 4 Regulatory T cells (Tregs) increased with a higher proportion of CD8 + PD-1 + TCF1 − terminally exhausted T cells (TCF1 − Tex term ). a Tregs negatively correlated with the ratios of progenitor exhausted T cells (TCF1 + Tex prog ) to TCF1 − Tex term (n = 21). b Flow cytometry result showed that CD4 + Foxp3 + Tregs were significantly increased in the low ratio group (high ratio group: n = 10, low ratio group: n = 11). c Representative contour plots of Foxp3 + Tregs among CD4 + T cells in the high/low TCF1 + Tex prog /TCF1 − Tex term ratio groups. d, e Representative IHC images of Foxp3 staining in FFPE samples from patients in the high/low ratio groups and quantitative Foxp3 H-score analysis showed that Foxp3 expressed higher in the low ratio group. Data are shown as the mean ± SEM. Spearman's correlation test was performed for a, and unpaired t tests were performed for (b, e). *P < 0.05, **P < 0.01. Scale bars: 200 μm for 100× and 50 μm for 400× were ground and filtered in the complete medium by a 40 μm cell strainer. Cells were then washed and treated with 1× red blood cell lysis buffer (BD Biosciences) for 2 min at room temperature. After washing with PBS (Gibco), the cell suspensions were then used for flow cytometry analysis.

Statistical analysis
All results were summarized as mean ± SEM, and statistical analysis was performed with Prism 7.0 (Graphpad Software). Differences between groups were evaluated by paired or unpaired two-tailed Student's t test. To determine the optimal cutoff point for the high-and low-density groups in our cohort and validation cohort, we used the maximally selected rank statistics from the "maxstat" R package. 46,47 The cutoff point for RFS was set to the same as the value for OS analysis. Survival analysis was performed and presented by using the Kaplan-Meier method. The difference in survival curves was tested by the log-rank test. The correlation analysis between groups was determined by Spearman's correlation test. P < 0.05 was considered statistically significant (*P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.000 1, ns P > 0.05).

DATA AVAILABILITY
Data are available upon reasonable request. All data relevant to the study are included in the article or uploaded as supplementary information.