The clinicopathological data of 223 consecutive patients with ccRCC from the Department of Urology, Zhongshan Hospital, Fudan University (Shanghai, China) between February 2005 and June 2007 were retrospectively analyzed in this study. All involved patients signed informed consent. Here were the inclusion criteria we followed: (1) undergone radical or partial nephrectomy at our institute; (2) pathologically diagnosed ccRCC; and (3) adults with age ≥18 years. The exclusion standards were: (1) accompanied with other malignant tumors; (2) preoperative systemic treatment; and (3) without complete/available follow-up/clinical/pathological data. Clinicopathological variables included age, gender, tumor size, TNM stage, Eastern Cooperative Oncology Group Performance Status and presence of tumor necrosis. Metastasis or recurrences were defined by images and histopathology information. Several urological pathologists reassessed the tumor histological type, differentiation and stage according to the 2004 WHO criteria31 and the American Joint Committee on Cancer 2010 TNM classification.32 Disease progression was identified via RECIST1.1 criteria.33 The time between curative surgery and recurrence or metastasis was defined as recurrence-free survival (RFS). Overall survival (OS) was defined as the time from the date of surgery to the date of death or last visit. The last collection of follow-up data was performed in June 2019. Archived tissue samples and surgical specimens of all patients were collected for the study. We randomly divided Zhongshan Cohort into two independent sets, that is, Discovery Cohort (n=112) and Internal Validation Cohort (n=111). We also enrolled patients from The Cancer Genome Atlas (TCGA) cohort as External Validation Cohort (n=532). Patient characteristics of TCGA data set were obtained from UCSC Xena (https://xenabrowser.net/datapages/) in January 2020. A total of 532 patients with ccRCC were feasible for the analysis with gene expression statistics, clinical data and follow-up information. Detailed patient characteristics of immunohistochemistry (IHC) analyses were presented in online supplemental table 1.

In addition, we collected resected tumor specimen from 42 fresh RCC samples undergone radical or partial nephrectomy in stage I and II from March 2019 to August 2019. Of these samples, eight of them were paired with peripheral blood and paratumor tissue. These samples were performed with flow cytometry (FCM) examinations. Detailed patient characteristics of FCM analyses were presented in online supplemental table 2. Flowchart of the patient enrollment for IHC and FCM analysis was listed in online supplemental figure 1.

Tissue microarray (TMA) construction was described previously.16 34 Single and double color immunohistochemical staining were carried out according to the protocols detailed previously.16 34 TMA slides were scanned on NanoZoomer-XR (Hamamatsu). Two urological pathologists independently counted immune cell number while blinded to the clinical data. Immune cells were evaluated as the average of positive cell number in three representative and independent fields. Representative pictures for IHC were shown in online supplemental figure 2. Details of IHC antibodies were shown in online supplemental table 3.

Fresh tissues were collected and then examined as soon as the tumors were resected during surgery. Peripheral blood mononuclear cells (PBMCs) were isolated from heparinized venous blood by lysing solution (BD Biosciences). Surgical tumor tissues and non-tumor tissues were minced and digested with collagenase IV (Sigma), and incubated by RBC lysis buffer (BD Biosciences). Then PBMCs and single cell suspensions were separately stained with membrane markers (monoclonal antibodies) for 40 min at 4℃. After disposed by Fixation/Permeabilization Solution Kit (BD Biosciences) according to manufacture protocol, intracellular protein was stained by corresponding antibodies. Cell suspensions and PBMC were stained with fluorochrome-labeled antibodies and preserved with cell staining buffer. FACS data were collected via BD FACSCelesta and analyzed via Flowjo V.10.0 (Tree Star). Details of FCM antibodies were provided in online supplemental table 4.

In the TCGA cohort, the CXCL13⁺CD8⁺T cell signature comprised CD2, CD3E, CD5, CD7, CD8A, CD8B, CD27, CD28, HIF-1α, TGF-β1, TGF-β2, TGF-β3, CTGF, SMAD2, SMAD3, CXCL13 and CXCR5.35–37 Signature score was defined as the mean of normalized expression of related genes [log₂ (FPKM +1)]. Gene signature was listed in online supplemental table 5. Absolute proportion of CD8⁺T cells in TCGA Cohort was obtained by CIBERSORT calculation.

The cut-off values were the median values in three cohorts. For CXCL13⁺CD8⁺T cells density in Discovery Cohort,≥40/ mm² was defined as high and <40/ mm² was defined as low. For holistic CD8⁺T cells density in Discovery Cohort, ≥86/ mm² was defined as high and <86/ mm² was defined as low. We directly used this threshold to test in Internal Validation Cohort. The cut-off value of CD8⁺T cell proportion and CXCL13⁺CD8⁺T cell signature level was 9.20% and 3.1493, respectively, in External Validation Cohort (TCGA Cohort). The cut-off value of TLS was defined as absence or presence. For FCM analyses, we also used the median proportion of CXCL13⁺CD8⁺T cells/ CD8⁺T cells (0.389) as the cut-off value.

Continuous variables were analyzed with Mann-Whitney test as appropriate. The relationship between CXCL13⁺CD8⁺T cell density and patient characteristics was evaluated by χ² test. The correlation analyses were evaluated with non-parametric (Mann-Whitney U test and Spearman’s test) tests. Wilcoxon matched-pairs signed rank test was used for non-parametric pairing t-tests. Kruskal-Wallis test was applied to detect the association between CXCL13⁺CD8⁺T cells infiltration level and tumor stage. Results were shown in mean±SD. Kaplan-Meier method and Log rank test were applied to demonstrate survival curves between different groups in the cohort. Multivariate analyses of cox regression model were applied to estimate HRs and 95% CIs. All statistical analyses were performed with SPSS V.19.0 (IBM), R software V.4.0.2 (R Foundation for Statistical Computing, http://www.r-project.org/), Graphpad V.8.0 and Medcalc V.15. The heatmap was performed with use of https://software.broadinstitute.org/morpheus/. All of these analyses were performed at the statistically significant level of 5% (p<0.05) and two tailed.

The large amount of tumor-infiltrated CD8⁺T cells in ccRCC could exhibit various immune functional phenotypes. First, we found that exhausted T cells could be defined via expression levels of inhibitory receptors PD-1 and Tim-3. More importantly, CXCL13 expression was upregulated in these exhausted T cells (online supplemental figure 3), so we put the emphasis on CXCL13⁺CD8⁺T cells.

The existence of CXCL13⁺CD8⁺T cells was confirmed by double-stained IHC. IHC staining images that represented different infiltration levels of CXCL13⁺CD8⁺T cells in intratumor tissue and nontumor tissue (black arrow: CD8⁺T cells, blue arrow: CXCL13⁺ cells, red arrow: CXCL13⁺CD8⁺T cells) through two enlargement factors (200×, 400×) were presented (figure 1A). We found that CXCL13⁺CD8⁺T cells were more abundant in intratumoral tissues compared with nontumor tissues (p<0.001, figure 1D). We also conducted the flow cytometry analyses of fresh ccRCC specimens (figure 1B,C). Similar results were obtained through comparison of cell proportion and median fluorescence intensity (MFI) level (figure 1E,F). IHC data also revealed the positive correlation between the infiltration level of CXCL13⁺CD8⁺T cells and tumor stage (p=0.046), UCLA Integrated Staging System (UISS) stage (p<0.001), and the stage size grade and necorsis score (SSIGN) grade (p=0.030) (figure 1G). More CXCL13⁺CD8⁺T cells were infiltrated in tumor tissue in later tumor stage (p=0.006), later UISS stage (p=0.017) and higher SSIGN grade (p=0.042) in ccRCC (figure 1H). Conclusively, these results suggested that tumor-infiltrating CXCL13⁺CD8⁺T cells were potentially involved in ccRCC tumor progression.

Since the infiltration level of CXCL13⁺CD8⁺T cells might relate to tumor progression, the clinical merit of this specific CD8⁺T subtype required further exploration. Next, we employed Kaplan-Meier analyses to investigate the prognostic value of CXCL13⁺CD8⁺T cells in patients with ccRCC from three independent cohorts (Discovery cohort, Internal Validation Cohort and External Validation Cohort). CD8⁺T cells alone failed to define the prognosis of patients with ccRCC in these three cohorts (discovery OS p=0.164, RFS p=0.200, figure 2A; internal validation OS p=0.340, RFS p=0.190, figure 2C; external validation OS p=0.784, RFS p=0.770, figure 2E), which revealed high heterogeneity for CD8⁺T cells in ccRCC. However, we observed that high level of tumor-infiltrating CXCL13⁺CD8⁺T cells represented significantly shorter OS and RFS than low level of tumor-infiltrating CXCL13⁺CD8⁺T cells in patients with ccRCC (OS p=0.008; RFS p=0.029; figure 2B) in discovery cohort. Shorter OS in CXCL13⁺CD8⁺T cell high subgroup was observed in two validation cohorts as well (internal validation OS p=0.018, external validation OS p=0.010, figure 2D). Although the CXCL13⁺CD8⁺T cell signature failed to predict RFS in external validation cohort (RFS p=0.085, figure 2F), high infiltration of CXCL13⁺CD8⁺T cells indicated shorter RFS in internal validation cohort (RFS p=0.011, figure 2D). Cox regression multivariate analysis demonstrated the infiltration level of CXCL13⁺CD8⁺T cells as an independent prognostic factor in three cohorts (online supplemental table 6). Conclusively, these results above proved that infiltration level of CXCL13⁺CD8⁺T cells could serve as an independent prognosis predictor for poor outcomes in ccRCC.

In order to explain the prognostic merit of CXCL13⁺CD8⁺T cells, we felt the necessity to explore the status of intratumoral CXCL13⁺CD8⁺T cells. We conducted FCM analyses on 42 fresh ccRCC tumor tissues in order to detect the immune function of intratumoral CXCL13⁺CD8⁺T cells. CXCL13⁺CD8⁺T cells within ccRCC tissues exposed higher level of immune checkpoints (PD-1, Tim-3, T cell immunoreceptor with Ig and ITIM domains (TIGIT) and CTLA-4) compared with CXCL13⁻CD8⁺T cells (PD-1 p=0.002, Tim-3 p<0.001, TIGIT p=0.002; CTLA-4 p=0.007; figure 3A). Additionally, CXCL13⁺CD8⁺T cells had relatively better proliferation ability (Ki67 p=0.008, figure 3B). Besides, we found that intratumoral CXCL13⁺CD8⁺T cells produced fewer effective molecules (tumor necrosis factor (TNF)-α p=0.026, interferon (IFN)-γ p=0.028, figure 3C), which might explain the exhausted state of CXCL13⁺CD8⁺T cells. The expression of cytotoxic molecules (PRF-1, GZMB) and degranulation marker (CD107a), however, showed no significant difference between CXCL13⁺CD8⁺T cells and CXCL13⁻CD8⁺T cells (PRF-1 p=0.447, GZMB p=0.994, CD107a p=0.504, figure 3C). In addition, the proportion of CXCL13⁺CD8⁺T cells was correlated positively with the proportion of Tim-3⁺CD8⁺T cells and negatively with the proportion of IFN-γ⁺CD8⁺T cells (Tim-3 p=0.005, Spearman r=0.558; IFN-γ p=0.047, Spearman r=−0.489, figure 3D). The relationship of CXCL13⁺CD8⁺T cells proportion and other markers of CD8⁺T cells (PD-1, TIGIT, CTLA-4, Ki67 and TNF-α) were shown in online supplemental figure 4). Conclusively, these findings revealed that the specific subset, CXCL13⁺CD8⁺T cells, exhibited highly exhausted status, diminished antitumor function and better proliferation ability.

Based on the results on the immune function of intratumoral CXCL13⁺CD8⁺T cells and the ‘paradoxical’ role of CD8⁺T cells in antitumor immunity, we assumed that the immune function of overall CD8⁺T cells might get restrained in ccRCC. The global characterization of CD8⁺ T cells was subsequently investigated based on CXCL13⁺ CD8⁺ T cell infiltration level. CD8⁺T cells in CXCL13⁺CD8⁺T cells high infiltration group possessed elevated immune checkpoints including PD-1, Tim-3 and TIGIT (p<0.001, p=0.023, p=0.024, respectively, figure 4B). Additionally, effective molecules like IFN-γ and TNF-α expressed by CD8⁺T cells were fewer in CXCL13⁺CD8⁺T cells high infiltration group (p=0.029, p=0.010, respectively, figure 4C). Though these CD8⁺T cells possessed increased exhausted markers and decreased activated molecules, we found the proportion, cytotoxic molecules (GZMB and PRF-1), degranulation capability (CD107a) and proliferation ability (Ki-67) of CD8⁺T cells showed no significant difference in CXCL13⁺CD8⁺T cells high/low infiltration group (figure 4A,C,D). As we mentioned above in figure 2, infiltration level of CD8⁺T cells alone could not define the prognosis of ccRCC patients. Merged survival curves, divided via CXCL13⁺CD8⁺T cells infiltration level and CD8⁺T cells infiltration level, indicated that combined division by CXCL13⁺CD8⁺T cells level and CD8⁺T cells level could serve as a prognosticator for patients with ccRCC as exhibited in figure 4E (p<0.001). We found that CD8⁺T cells abundance possessed the ability to independently define poor prognosis in CXCL13⁺CD8⁺T cells high infiltration group (p=0.009, online supplemental figure 5A). Nevertheless, CD8⁺T cells alone still could not define the prognosis in CXCL13⁺CD8⁺T cells low infiltration group (p=0.141, (online supplemental figure 5B). Conclusively, these results revealed that intratumoral CXCL13⁺CD8⁺T cells abundance might correlate with dysfunctional CD8⁺T cells infiltration, which led to poor outcomes in patients with ccRCC.

We found that tumor-infiltrating CXCL13⁺CD8⁺T cells represented a highly exhausted subtype. Therefore, the specific TME that induced the switch of exhausted T cells provoked our interest. Heatmap was used for intuitionistically exhibiting the relationship between the level of CXCL13⁺CD8⁺T cells and tumor stage, immune-related cells (CD8⁺T cells, CD4⁺T cells, Th1 cells, Th2 cells, T_reg cells, T_FH cells, TAMs, M1 macrophages, M2 marcophages, neutrophils, dendritic cells [DC], mast cells, natural killer [NK] cells and B cells) and immune-activated markers (GZMB, IFN-γ) (figure 5A). The number of total CD8⁺T cells was not related with CXCL13⁺CD8⁺T cells infiltration level (figure 5B), which was corresponded with results above in figure 4. Intriguingly, we found that microenvironment with high level of intratumoral CXCL13⁺CD8⁺T cells contained more CD4⁺T cells, Th2 cells (Gata-3⁺CD4⁺T cells), T_reg cells (Foxp3⁺CD4⁺T cells) and TAMs (CD68⁺ cells) (figure 5C,D). The level of intratumoral CXCL13⁺CD8⁺T cells showed negative connection with Th1/Th2 ratio (figure 5C). Additionally, microenvironment with high level of intratumoral CXCL13⁺CD8⁺T cells possessed decreased NK cells and GZMB⁺ cells (figure 5E,F). The density of CXCL13⁺CD8⁺T cells was correlated positively with the amount of TLS (p=0.026, figure 5G). Though TLS was recognized as a trigger for antitumor response,38 we found that TLS amount could not indicate clinical outcomes for patients with ccRCC (online supplemental figure 6). Conclusively, tumor-infiltrating CXCL13⁺CD8⁺T cells were assumed to interrelate with immunoevasive contexture in ccRCC TME.