The normalized level-3 data of the KIRC cohort in the TCGA database was acquired using the UCSC Xena (http://xena.ucsc.edu/). Clinicopathological, RNA-seq, DNA methylation (quantified by Illumina HumanMethylation450 BeadChip), and survival data were extracted for further analysis. In brief, gender, AJCC pathological stages, and histological grade of tumors were extracted. RNA-seq data were represented as log₂ (TPM + 0.001). Log₂ (fold change, FC) was calculated between two groups. The methylation level of CpG sites was calculated and represented by the β value. Survival data included PFS, DFS, DSS, and OS. PFS refers to the period from the date of diagnosis until the date of the first occurrence of a new tumor event (NTE), which includes a progression of the disease, locoregional recurrence, distant metastasis, new primary tumor, or death with the tumor. Patients who were alive without these event types or died without tumor were censored. DSS event is defined as death from the disease (Liu et al., 2018).

A total of 530 primary tumor cases and 72 adjacent normal (adj. N) tissues with RNA-seq data were included. All 530 tumor cases had OS data, among which 528, 116 and 519 cases had PFS, DFS, and DSS data, respectively. Besides, 309 tumor cases and 23 adj. N cases have RNA-seq and methylation data at the same time.

ScRNA-seq data of previous publication datasets were examined using scRNASeqDB (https://bioinfo.uth.edu/scrnaseqdb/index.php?csrt=18377095680907440044) (Cao et al., 2017) and PanglaoDB (https://panglaodb.se/index.html) (Franzen et al., 2019). The former one currently collected 38 scRNA-seq datasets, while the latter one collected 305 human scRNA-seq datasets.

Immunohistochemistry (IHC) staining of CXCL13 (antibody: HPA052613) in renal tumor tissues were retrieved from the HPA (https://www.proteinatlas.org/) (Uhlen et al., 2015; Uhlen et al., 2017). In this database, 11 renal cancer tissues were stained for CXCL13 expression, among which four were scored as low, while seven were scored as negative.

Human embryonic kidney cell line 293T, KIRC cell lines ORSC2 and 7860, renal cell carcinoma cell line ACHN, ovarian cancer cell line SKOV3, leukemia cell line K562, U266 were purchased from the ATCC. All cell lines were cultured at 37°C, 5% CO₂ in a humidified incubator in standard cell culture media as indicated by the provider.

In brief, BALB/C mice were immunized using CXCL13 recombinant protein. To Monoclonal antibody was generated by hybridoma technology. Firstly, ELISA was used to select antibodies that bind to CXCL13 recombinant protein. Cell ELISA was performed to screen competitive antibodies. Surface Plasmon Resonance (SPR) was utilized to screen high-affinity antibodies. The hybridoma cells were sub-cloned at least two times by limiting dilution method. Finally, CXCL13 antibodies with high purity were obtained using protein G affinity chromatography. The purity of the antibodies was higher than 95%, as determined by SDS-PAGE. A monoclonal CXCL13 antibody (18-C1-B3) was used in the current study.

Whole-cell lysates of 293T, ORSC2, 7860, ACHN, SKOV3, K562, and U266 cells were prepared using RIPA buffer (Beyotime, Wuhan, China) supplemented with protease inhibitor PMSF (MilliporeSigma, St. Louis, MO, USA). Protein concentrations were determined using a BCA protein assay kit (Thermo, Waltham, MA, USA). Protein extracts and CXCL13 recombinant Protein (PeproTech, Rocky Hil, NJ, USA) were separated by 15% SDS-PAGE and transferred onto a PVDF membrane (MilliporeSigma). PVDF membrane was blocked in TBS containing 5% nonfat dry milk and 0.1% Tween 20 with gentle shaking for 1 h. Membranes were incubated with a primary antibody generated in our lab (18-C1-B3, 1:2000) or a commercially available antibody (AF801, 1:2000, R&D Systems, Minneapolis, MN, USA) diluted in the blocking buffer at 4°C overnight. Then, they were incubated for 1 h with the corresponding horseradish peroxidase-linked secondary antibodies (Zhong Shan Golden Bridge Biotechnology, China) diluted 1:5000 in blocking buffer. Images were captured by a ChmiScope 6000 Touch (Clinx, Shanghai, China).

CXCL13 expression in human KIRC tissues was evaluated using a human tissue microarray. Paraffin-embedded TMA was deparaffinized in xylene and rehydrated in gradients concentration of ethanol. TMA was treated with retrieval solution for 15 min at 99°C and incubated with anti-CXCL13 antibody in our lab overnight after blocking with goat serum for 1 h at room temperature. Following a wash, slides were incubated with DakoReal^TMEnVision^TM horseradish peroxidase-conjugated anti-mouse antibody for 30 min at room temperature and then visualized using diaminobenzidine (Dako, Carpinteria, CA, USA). Staining was analyzed via Imagscope Viewer (Leica Biosystems, Buffalo Grove, IL, USA).

The fractions of TIICs in TCGA-KIRC were acquired from the Cancer Immunome Atlas (https://tcia.at/home) (Charoentong et al., 2017). In this dataset, RNAs-seq data from TCGA was used to make an estimation using a strategy called CIBERSORT (Newman et al., 2019). This method characterizes the cell composition of complex tissues from their gene expression profiles. Briefly, it used 547 genes to distinguish 22 human hematopoietic cell phenotypes, including seven T cell types, naïve and memory B cells, plasma cells, NK cells, and myeloid subsets.

Statistical analysis was conducted using SPSS 25.0 and GraphPad Prism 8.1.2. Data were reported as the mean ± standard deviation (SD). Unequal variances t-test (Welch’s t-test) was performed for group-group comparison. Kaplan-Meier survival curves were generated. Survival difference was assessed by the Log-rank test. Correlation analysis was performed by calculating Pearson’s correlation coefficients. p < 0.05 was considered statistically significant.

Several previous studies indicated that CXCL13 expression might serve as a valuable biomarker in terms of therapeutic responses and prognosis in multiple cancers, including gastric cancer (Wei et al., 2018), breast cancer (Razis et al., 2020), and renal cancer (Zheng et al., 2018). Using RNA-seq data from TCGA-KIRC, we confirmed that CXCL13 expression was significantly higher in KIRC tumor tissues (N = 530) than in adj. N tissues (N = 72, log₂ FC = 5.387 ± 0.375, p < 0.001) (Figs. 1A–1B). No significant difference was observed between male and female patients (p = 0.26, Fig. 1C). However, CXCL13 expression was significantly higher in advanced stage (Stage III/IV vs. I/II, log₂ FC = 1.529 ± 0.248, p < 0.001) and high grade (grade 3/4 vs. grade 1/2, log₂ FC = 1.434 ± 0.239, p < 0.001) tumors (Figs. 1D–1E). Survival analysis confirmed that the groups with the highest quartile of CXCL13 expression had significantly worse PFS, DSS, and OS compared to the groups with the lowest quartile of CXCL13 expression (Figs. 1F–1H). In terms of DFS, we failed to identify a significant difference between the high and low CXCL13 expression groups by either quartile or median gene separation (Suppl. Figs. 1A–1B).

Although we confirmed the association between CXCL13 expression and unfavorable survival outcomes using data from TCGA-KIRC, the source of CXCL13 in the TME has not been well-characterized since RNA-seq data in TCGA-KIRC are from on bulk tissue. Two recent studies reported that renal cancer cell lines had significantly upregulated CXCL13 expression (Xu et al., 2019; Zheng et al., 2018). However, these findings are highly suspicious since CXCL13 mRNA expression was very limited in most of the tumor cell lines, according to the results in the HPA (Fig. 2A). Only several myeloid and bone marrow-derived cell lines, such as U-266, HL-60, and RH-30, have positive CXCL13 expression (Fig. 2A). HEK 293 cells (a human embryonic kidney cell line) has negative CXCL13 expression (Fig. 2A, black arrow).

To verify the RNA expression in KIRC cells, we also checked one previous scRNA-seq dataset (N = 116, GSE73121) based on primary and metastatic tumor cells (Kim et al., 2016), using scRNASeqDB. In this dataset, 34 parental metastatic KIRC (mKIRC) tumor cells, 36 patient-derived xenograft (PDX)-mKIRC tumor cells, and 46 PDX-primary KIRC (pKIRC) tumor cells were subjected to scRNA-seq. Results showed that among these cells, only one cell in the mKIRC group showed positive CXCL13 expression, while the rest 115 cells all had zero CXCL13 expression (Fig. 2B).

To explore the real source of CXCL13, we checked data in PanglaoDB, which including 305 human scRNA-seq datasets covering nearly all human tissues. Results identified three datasets with positive CXCL13 expression (https://panglaodb.se/search.html?query=%22CXCL13%22&species=3&tumor=1&nonadult=1), among which two datasets were T cells (SRA767245 and SRA767245), and another one was fibroblasts from embryonic stem cell lines (SRA676027).

To characterize the expression of CXCL13 at the protein level, we generated a CXCL13 monoclonal antibody (18-C1-B3). Western blot data indicated that KIRC cell lines (ORSC2 and 7860), as well as several other cancer cell lines, were negative for CXCL13 expression (Fig. 2C). Similar results were observed using a commercial CXCL13 antibody from an R&D system (AF801) (Fig. 2C). However, both 18-C1-B3 and AF801 could recognize the recombined CXCL13 protein (Fig. 2C).

Therefore, we infer that high CXCL13 expression detected by bulk-seq data in TCGA-KIRC was not from tumor cells but from other cells in TME. To validate this hypothesis, we checked IHC staining of 11 renal cancer tissues in the HPA, which used a commercial primary antibody against CXCL13 from Millipore Sigma (HPA052613). According to the CXCL13 staining score, 4 cases were low (Fig. 2D, up), while the rest 7 cases were not detected. Based on the images, we clearly saw the four cases with low CXCL13 staining all had abundant TIICs, which were generally CXCL13 positive (Fig. 2D up, purple arrows). The tumor cells around the TIICs also presented positive CXCL13 staining. As a secreted chemokine that can bind to typical receptors (such as CXCR5) on the surface of tumor cells, we speculated that the tumor cells-associated CXCL13 staining was a reflection of the CXCL13-receptor binding. In comparison, TIICs were rare in the cases scored as CXCL13 negative (Fig. 2D, down). Meanwhile, tumor cells were CXCL13 negative if no TIICs were nearby (Fig. 2D, down). To validate these findings, we also performed IHC using KIRC tissue array with our own anti-CXCL13 (18-C1-B3). Results confirmed that the issues lacking TIICs were generally CXCL13 negative (Fig. 2E, left). In comparison, TIICs and the surrounding KIRC tumor cells were CXCL13 positive (Fig. 2E, middle and right). These findings collectively suggested that CXCL13 might be mainly expressed by TIICs but no KIRC tumor cells.

Since we confirmed that CXCL13 was expressed by TIICs, we then explored its association with the faction of TIICs in TCGA-KIRC. The fraction scores calculated by the CIBERSORT method was used. A total of 22 TIIC types were estimated using RNA-seq data from TCGA-KIRC (Fig. 3A). Then, we assessed the correlation between CXCL13 expression and the TIICs (Fig. 3A). By setting moderate correlation (|Pearson’s r| ≥ 0.4) as the cut-off, we found that CXCL13 expression was negatively correlated with activated dendritic cells and resting memory CD4⁺ T cells (Figs. 3A–3C) but was positively correlated with gamma delta (γδ) T cells (Figs. 3A and 3D). Then, we assessed the survival difference between patients with the highest and lowest quartile of these TIICs. No significant survival difference was observed between groups stratified by activated dendritic cells (Figs. 3E and 3H). In comparison, a higher fraction score of resting memory CD4⁺ T cells was associated with better PFS (p = 0.006) and DSS (p < 0.001) (Figs. 3F and 3I). In contrast, a higher γδ T cell fraction was associated with significantly worse PFS (p < 0.001) and DSS (p = 0.014) (Figs. 3G and 3J).

Although CXCL13 dysregulation was confirmed in KIRC, the underlying mechanisms have not been explored yet. One recent study reported that CXCL13 might be transcriptionally silenced by CpG site methylation within its promoter (Ma et al., 2020). Therefore, we hypothesized that CpG site methylation might also influence CXCL13 expression in KIRC TME. Since we confirmed that CXCL13 was mainly expressed in TIICs in KIRC tissues, we infer that methylation 450k data in TCGA-KIRC could reflect CXCL13 DNA methylation profiles in TIICs. Therefore, we checked CXCL13 DNA methylation data in 309 KIRC tumors and 23 adj. N tissues with both RNA-seq and methylation data (Fig. 4A). Five CpG sites within the CXCL13 gene locus were included in the array, among which three sites were promoter-associated (cg16361705, cg17001652, and cg12020230) and two were gene-body associated (cg01134794 and cg06662476). Group comparison found four CpG sites (cg16361705, cg17001652, cg12020230, and cg01134794) were significantly hypomethylated in tumor tissues (Fig. 4B). Then, we generated plot charts to assess the correlation between CXCL13 expression and their methylation status (Figs. 4C–F). The methylation level of cg16361705 showed a nearly moderate negative correlation with CXCL13 expression (Pearson’s r = −0.38, p < 0.001, Fig. 4C).