CXCL13-expressing CD4+ T cells coordinate the lymphocytes triad to promote the anti-tumor immunity in NSCLC Authors: Danping Liu, Jiajun Luo, Wei Guo, Junrui Ma, Xiaobao Yang, Yihan Xia, Pu Li, Dakang Xu, Zhidong Gu Published: 2026 ABSTRACT ABSTRACTCD4+ T cells are indispensable for CD8+ T cells-mediated anti-tumor immunity, while little is known about how CD4+ T cells coordinate with other cells to promote CD8+ T cells activity. In this study, by coupling single-cell RNA sequencing (scRNA-seq) with multiplex immunohistochemistry staining (mIHC), CXCL13-expressing CD4+ T cells were explored to recruit CXCR5+ B cells to form CXCL13+CD4+ T cells:: CXCR5+ B cells:: CD8+ T cells triad, promoting the anti-tumor immunity and heralding a favorable prognosis. A CXCL13-expressing subset of CD4+ T cells was identified to be associated with better prognosis and tumor-reactive hallmarks. The further cell types interaction analysis revealed a specific interaction between CXCL13+CD4+ T cells and CXCR5+ B cells in comparison with other cell types. The mIHC verified that an elevated level of CXCR5+ B cells infiltration in the proximity of CXCL13+CD4+ T cells. The subsequent Entropy analysis, which characterizes the colocalization among more than two cell types, was employed to reveal the spatial distribution pattern. The results revealed that CXCL13+CD4+ T cells and CXCR5+ B cells co-localized with cytotoxic CD8+ T cells, forming a functional lymphocyte triad, whose existence heralded a better prognosis. Collectively, by integrating scRNA-seq with mIHC, we demonstrated that the recruitment of CXCR5+ B cells by neoantigen reacting to CXCL13+CD4+ T cells through CXCL13–CXCR5 signaling contributed to the lymphocyte triad, leading to an enhanced anti-tumor immunity and heralding a favorable prognosis in non-small-cell lung cancer (NSCLC).Keywords: CXCL13+CD4+ T cells, lymphocytes triad, spatial analysis, CXCL13–CXCR5, NSCLC Introduction The functionality of T cells orchestrates the anti-tumor immunity. For decades, the direct tumor killing has been ascribed to the cytotoxic CD8+ T cells, whereas the effector function of CD4+ T cells is conventionally considered to be focused on licensing antigen-presenting cells APCs for CD8+ T cell priming, facilitating and sustaining innate and adaptive anti-tumor reactions. 1-3 It has been demonstrated that the proper anti-tumor capacity of CD8+ T cells requires priming and activation in a CD4+ T cell-dependent manner. 3-7 With the deep profiling of intratumoral T cell compartments, recent studies have highlighted the importance of intratumoral CD4+ T cells in eliciting potent anti-tumor immune response against a wide spectrum of cancers, especially NSCLC, meanwhile characterized the molecular signatures of tumor reactive CD4+ T cells. 3 , 5 , 7-13 Hence, CD4+ T cells are responsible and indispensable for coordinating an effective adaptive immune response against tumor in the tumor microenvironment (TME). However, the mechanism by which these tumor-reactive CD4+ T cells unleash the anti-tumor activity of CD8+ T cells in NSCLC remains poorly understood. Following the cytokine milieu of TME and other stimulation signals, CD4+ T cells can differentiate into several subsets under the control of distinct transcriptional regulatory programs and thus exhibit a broad spectrum of action and complexity. 7 It has been recognized that an unconventional subset of CD4+ T cells carrying the cytotoxic features, such as the expression of IFN-γ, granzymes, and perforin, directly participates in the tumor eradication in a MHC-II dependent manner. 1 , 4 , 7 , 14-16 Aside from the direct cytotoxicity against tumors, cytokines released from intratumor CD4+ T cells coordinate the innate and adaptive anti-tumor immune functions in the TME through remodeling the cell types distribution in the TME and forming functional spatial architectures, such as the well-known tertiary lymphoid structures (TLS). 3 , 13 , 16-18 In a murine lung cancer model, CD4+ T cells are reported to reshape the distribution of CD8+ T cells and B cells to form “lymphonets” which supports the cytotoxic T cell response during immunotherapy through CXCL10-CXCR3 signaling. 19 In human melanoma and ovarian cancer, CXCL13+ CD4+ T cells are found to activate tumoricidal macrophages and CD8+ T cells, as well as co-localizing with B cells and promoting their maturation, which participate the formation of TLS to elicit antitumor response of cellular and humoral immunity. 18 , 20 These findings suggest that these tumor-reactive CD4+ T cells reshape the TME towards a tumor rejecting states through interacting with other cell type, while whether and how tumor-reactive CD4+ T cells regulate other anti-tumor components and thus promote the CD8+ T cells-dependent tumor eradication in the TME of NSCLC has yet to be deciphered. Despite the utility of imaging based and sequencing based diagnosis, these approaches failed to obtain the functional states of infiltrated immune cells. 21 , 22 With advance of scRNA-seq and multiplexed imaging technologies coupled with state-of-art analyzing algorithms, we are now capable of profiling cells with in-depth transcriptional features and locating the cells with specific phenotypes in the TME to reveal their interacting partners. 22-28 Here, through scRNA-seq profiling of intratumoral CD4+ T cells in human NSCLC, we identified a unique subset of CXCL13-expressing CD4+ T cells carrying the tumor-reactive features whose infiltration level was validated to be correlated with a favorable clinical outcome by multiplexed staining of patient resections. By analyzing the interacting cell types of these tumor-reactive CXCL13+CD4+ T cells and revealing ligand‒receptor signaling, CXCL13+CD4+ T cells were found to be associated with the infiltration of CD8+ T cells and B cells, which are recruited through CXCL13‒CXCR5 signaling. Based on a multiplex panel containing PanCK, CD8, CD4, CD20, CXCL13, and CXCR5, we localized and phenotyped the tumor cells, T cells, and B cells in the TME of NSCLC. Consistent with the results from scRNA-seq, CXCL13+CD4+ T cells were found to co-localize with CXCR5+ B cells and CD8+ T cells and accumulate in the TLS-like structure. An entropy algorithm was employed to reveal the distribution pattern of CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells. The TME was dissected into multiple units by Entropy algorithm to calculate the prevalence of these three cell types, and a high prevalence suggested a more aggregated distribution of these three cell types to form a CXCL13+CD4+ T cell:: CXCR5+ B cell:: CD8+ T cell triad instead of an irrelevant random distribution. The results suggested that patients with a high prevalence exhibit a prolonged overall survival, implying a more coordinated and potent anti-tumor response. Collectively, our data indicated that CXCL13+CD4+ T cells could coordinate with CD8+ T cells and CXCR5+ B cells to promote the anti-tumor immunity and the defined triad could serve as a biomarker to differentiate the outcomes of patients, yielding the future of precise medicine in oncology. Materials and methods Data acquisition scRNA-seq data were obtained from GSE144945, which contains CD45+ cells from ten patients diagnosed with NSCLC. GSE131907, including 11 primary LUAD tissues that profiled both immune cells and other non-immune components in the TME, such as endothelial cells and fibroblasts, was used for validation. The RNA-seq data of 526 LUAD patients and 501 LUSC patients from the TCGA database were downloaded from the UCSC website. H&E diagnostic slides of NSCLC patients were acquired from the TCGA database. Single-cell data acquiring and processing Cells were first subjected to stringent quality control based on the number of RNA features between 200–6000, the percentage of mitochondrial genes <10%, and the proportion of erythroid genes less than 3%. Cells that passed QC were normalized using the recommended settings of 'NormalizeData' and the ‘vst’ method, and 2000 highly variable genes were screened. The cell cycle score was calculated using the ‘CellCyclesScoring’ algorithm and normalized by the “ScaleData” function. PCA dimensionality reduction was then carried out using the 2000 highly variable genes obtained above. The batch effect between different patients was also removed by the ‘Harmony’ algorithm as recommended. Finally, using the UMAP algorithm for dimensionality reduction, different clusters were obtained using the ‘FindNeighbors’ and ‘FindClusters’ functions. According to the expression of gene markers, cell clusters were defined as CD4+ T cell (CD3D, CD3E, and CD4), Treg (FOXP3 and IL2RA), CD8+ T cell (CD8A and CD8B), NK (FCGR3A and KLRD1), and B cell (MS4A1 and CD79A), Macrophage (CD68and CD14), DC cell (CLEC4C), ILC2 (KIT and IL1RL1), and proliferating cell (MKI67 and TOP2A). Next, we further analyzed the CD4+ T cell cluster and obtained a total of four different CD4+ T cell clusters with unique transcriptional features. The cell clusters were annotated as CD4_CXCL13, CD4_CCL5, CD4_IL7R, and CD4_CCR7, according to the highly expressed genes generated by the ‘FindAllMarkers’ function. The R package (version 3.1.5) was applied in this section. To reaffirm the specificity of CXCL13 expression in CXCL13+CD4+ T cells, we analyzed an additional validation scRNA-seq dataset (GSE131907) including 11 primary LUAD tissues that profiled both immune cells and other non-immune components in the TME, such as endothelial cells and fibroblasts. Identification of tumor-reactive CD4+ T cells Gene signatures of tumor-specific CD4+ T cells were obtained from Frank J. Lowery et al. 12 The 40-gene version of NeoTCR4 (NeoTCR-40) was previously shown to have the highest sensitivity and specificity and was therefore used to estimate the tumor reactivity score. The tumor reactivity score of each CD4+ T cell cluster was obtained using the ‘AddMoudule’ algorithm. To further prove that CXCL13+CD4+ T cells are tumor-reactive CD4+ T cells, we collected the marker genes of tumor-reactive CD4+ T cells (CXCL13, GNG4, ETV7, and TOX) and visualized their expression in different CD4+ T cell clusters using the ‘FeaturePlot’ function. Extracting related genes and normalizing them to CD4. 13 Normalization of CXCL13 expression by CD4 was defined as the ratio of the expression levels of CXCL13 and CD4 (FPKM). Cytotoxic scores were calculated using the ssGSEA algorithm to evaluate the expression levels of cytotoxicity-related genes, which is consistent with established protocols. 29 Gene ontology enrichment analysis We performed Gene Ontology enrichment analysis using the ‘ClusterProfile’ package in R. 30 Upregulated genes in CXCL13+CD4+ T cells were generated by the ‘FindMarkers’ algorithm and then were analyzed using the ‘enrichGO’ function. The cutoff for significant enrichment was set at p < 0.01, q < 0.05. Gene sets in Gene Ontology-Biological Process (GO BP) were used for the enrichment analysis. Decoding the immune infiltration in the TME by RNA-seq data The “IOBR” package of R software was applied to decipher TME. 28 , 31 The ESTIMATE algorithm was used to calculate the StromaScore and ImmuneScore based on the RNA-seq data. Gene expression signatures are used to draw a reasonable inference about the fraction of immune and stromal cells in TME. 32 TLS signatures calculation The calculation process refers to research and TLS signature genes that include genes obtained from the quoting literature. 33 EPIC (estimating the proportions of immune and cancer cells) To explore the correlation of CXCL13+CD4+ T cells with other types of infiltrated immune cells, we apply the “EPIC” algorithm and detect the main cell types in the tumor. 34 The fraction of different cell type was generated and their correlation with CXCL13+CD4+ T cells was calculated by the Spearman test. Cell communication analysis of scRNA-seq data Cell communication analysis between CXCL13+CD4+ T cells and B cells was performed using CellChat version 1.6.1. The default ‘CellChatDB’ was applied. The ‘ComputeCommunProb’ function was run to infer cell-to-cell interactions using ‘truncatedMean’ algorithm. The ‘NetVisual_bubble’ function with signaling set to XCR, CCL, and CXCL was performed to demonstrate chemotaxis between CD4+ T cells and B cells. The expression of specific ligands and receptors in different cell clusters was shown using ‘DotPlot’. Patient cohort Human NSCLC specimens were collected from Ruijin Hospital, Shanghai Jiao Tong University School of Medicine with written informed consent was obtained prior to participation. An experienced pathologist used H&E staining to diagnose the pathological type of each patient. A total of 82 FFPE tumor tissues (LUAD) were constructed on tissue microarrays (TMA) for subsequent staining analysis. The basic information of the patients is shown in Table S1. The study was conducted in accordance with the Declaration of Helsinki, and all patients provided written informed, the Ethics Committee of Ruijin Hospital, Shanghai Jiao Tong University School of Medicine-approved consent for the use of their tissues for research purposes. Multiplex immunohistochemistry The FFEP slide was heated at 60 °C for 1 hour, followed by deparaffinization with xylene and rehydrated through an ethanol solution with a descending gradient (100%, 95%, 90%, 85%, 50%, ddH2O). Afterward, the sections were microwaved for antigen repair. Peroxidase blocking was performed with endogenous peroxidase blocking solution, and blocking solution was used to block non-specific antibody binding. The primary antibodies included anti-CD8a, anti-CD4, anti-CXCR5, anti-CXCL13, anti-CD20 and anti-pan-keratin antibodies. Then, labeling of cells with a polymer horseradish peroxidase (HRP)-labeled secondary antibody (Akoya Biosciences). Next, Opal-tyramide signal amplification (TSA)-conjugated fluorescent dyes were used to visualize the staining results. The cell nuclei were counterstained with DAPI, and the TMA slide was mounted for later detection using the Vectra Polaris System (Akoya Biosciences). Supplementary Table S2 shows the detailed information. Transwell migration assay Transwell migration assay was performed to verify that CXCL13+CD4+ T cells recruit B cells and that this can be blocked by CXCL13 neutralizing antibodies. Twenty-four-well transwell chambers (14347, LABSELECT) with 5 µm pore filters were used. First, plate-bound αCD3/αCD28 Ab (10 μg/ml each, 70976S, Cell Signaling Technology) was incubated at 4 °C in the dark for 24 hours. To minimize edge effects, the outer wells were sequestered with 500 μL of sterile PBS. PBMC were isolated and pooled from five healthy donors. 6 × 105 PBMCs in 600 µL of the culture medium were seeded in the bottom chamber and stimulated with plate-bound αCD3/αCD28 Ab. To disrupt CXCL13/CXCR5 interactions during chemotaxis, a neutralizing antibody against CXCL13 (2 µg/mL, MAB801, R&D Systems) was added to the culture medium. Then, a total of 2 × 105 PBMCs labeled with CellTrace Violet dye (CTV, Cat# C34557, Invitrogen) in 200 µL of culture medium were seeded in the upper chamber, neutralizing antibody against CXCR5 (10 µg/mL, MAB190, R&D Systems) was added to the culture medium. The cells were incubated at 37 °C to allow migration. After 36 h of migration, the cells in the lower chambers were collected, stained with Fixable Viability Dye eFluor™ 780 and CD19-PE, and analyzed with a BD LSRFortessa X20. Spatial analysis APTIME software was used for tissue segmentation and definition of the phenotypes of the cells. The cells were segmented based on DAPI staining, and markers, including CD20, CD4, CD8, CXCL13, CXCR5 and PanCK, were used to distinguish different cell types. The positive thresholds of different markers were further determined according to their staining intensity and specificity. The cell densities, counts and proportions were automatically acquired and then analyzed using the APTIME software. The spatial localization (x, y) of different cells in each core was also generated via APTIME software for subsequent spatial analysis. The tumor boundary was manually selected based on the expressions of DAPI and PanCK. Target cells distributed 100 µm away from tumor boundary were calculated separately. The number of target cells within 10 µm of each reference cell is defined as the ratio of the total number of target cells (within 10 µm of reference cells) to reference cells. A 100 µm area around each CXCL13+CD4+ T cell was delineated, and the proportion of CXCR5– B cells versus CXCR5+ B cells distributed 0–100 µm around CXCL13+CD4+ T cells (number of bands = 10) was calculated separately in patients. Entropy was introduced from a spatial-platform agnostic toolkit named spatial image analysis of tissues (SPIAT). 23 Briefly, each tissue was divided into 225 (15 × 15) grids, and in each grid, Entropy is calculated based on the co-localization of CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells. A high value of entropy indicates that the cells are co-localized. Next, the proportion of grids with entropy values higher than 0.75 was calculated, which was termed prevalence. A higher prevalence indicates that the CXCL13+CD4+ T cells:: CXCR5+ B cells:: CD8+ T cells triad is present in greater quantities in the patients. Histological evaluation of TLS score and TLS counts Diagnostic H&E staining samples were downloaded from TCGA. Slides with poor staining quality were excluded, and a total of 94 LUAD and 106 LUSC patient slides were included for subsequent analysis. TLS evaluation was analyzed by professional pathologists. The area of each slide was calculated, excluding areas of necrosis and debris, by the pathologists. The score is defined according to this research standard. 35 In brief, the density of TLSs was divided into four classes: no TLSs (score 0), one or two TLSs (score 1), more than three TLSs (score 2), and massive TLSs (score 3). TLS counts in the TMA were assessed by pathologists to examine their association with prevalence. Statistical analysis The data were analyzed using R (version 4.1.0). The quantitative data were compared by two-tailed Student's t-test for unpaired data. Overall survival was evaluated via Kaplan‒Meier curves and log-rank tests. The correlation between CXCL13+CD4+ T cell density and B cell density and TLS-like structure were determined using Spearman analysis. Differences in the mean cell number of CXCR5– B cells and CXCR5+ B cells around CXCL13+CD4+ T cells in the same patient were analyzed using paired t-tests. Multivariate Cox regression analysis was constructed to reveal the correlations between CXCL13+CD4+ T cell level and NSCLC risk. p values of less than 0.05 were considered significant. Results Infiltration of tumor reactive CXCL13+CD4+ T cells heralds a favorable outcome To address how tumor-reactive CD4+ T cells interact with other immune cells and thus elicit anti-tumor immunity, here we first employed scRNA-seq and mIHC assays to identify tumor-reactive CD4+ T cells (Figure 1A). Based on canonical markers of immune cells, unsupervised clustering by UMAP algorithm was employed to characterize different cells in NSCLC (Figure S1A,B). Obtaining a total of 9 clusters of immune cells, we further targeted on conventional CD4+ T cells (CD4+ Tconv) to reveal heterogeneity, and subsequently four different CD4+ T cell subsets with distinct transcriptional features were characterized, including CD4_CXCL13, CD4_CCL5, CD4_IL7R, and CD4_CCR7 (Figures 1B,C, S1B,C). To further test the anti-tumor capacity of the four subsets of CD4+ Tconv cells, the tumor reactivity score of each CD4+ Tconv cell cluster was obtained, and the results proved that the CD4_CXCL13 cluster was tumor-specific CD4+ T cells (Figure 1C). Further, other hallmark genes, such as CXCL13, TIGIT, BATF, and TOX, were also found to be highly expressed in CD4_CXCL13 cluster 12 , 13 , 36 (Figures 1D, S1D). The transcriptional data suggested that CXCL13 is a unique marker of the CD4_CXCL13 cell cluster that distinguishes it from other subgroups and immune cell types (Figure 1E). To reaffirm the specificity of CXCL13 expression in CXCL13+CD4+ T cells, we analyzed an additional validation scRNA-seq dataset (GSE131907) that profiled both immune cells and other non-immune components in the TME, such as endothelial cells and cancer-associated fibroblasts (Figure S1E). The results suggested that CXCL13 was exclusively expressed in CXCL13+CD4+ T cells rather than other cell types. Notably, both the tumor-reactive score and the expression of all 40 genes in the signature ranked highest in CXCL13+CD4+ T cells among all identified subsets and other subsets of CD4+ T cells, including regulatory T cells (Treg) (Figure S1F,G). Infiltration of tumor reactive CXCL13+CD4+ T cells heralds a favorable outcome. (A) Schematic overview revealing the workflow. (B) UMAP dimensionality reduction analysis of CD4+ T cell clusters. (C) Violin plots showing the tumor reactivity score in CD4+ T cell clusters. ***p < 0.001. (D) Volcano plot comparing differentially expressed genes of CD4_CXCL13 cell cluster with other CD4+ T cell clusters. (E) A dot diagram revealing the expression of CXCL13 in each cell cluster. (F) Representative images of expression of Panck, CXCL13, and CD4 in adjacent normal tissue and tumor tissue of NSCLC. Yellow arrows depict CXCL13+CD4+ T cells. (G) Comparison of the density and frequency of CXCL13+CD4+ T cells in adjacent normal tissue and tumor tissue. n (adjacent normal) = 82, n (tumor) = 82. **p < 0.01. (H) Survival plot of CXCL13+CD4+ T cell density estimated by mIHC in NSCLC. n (Low) = 41, n (High) = 41. (I and J) Survival plots of normalization of CXCL13 expression by CD4 in TCGA LUAD cohort and TCGA LUSC cohort. In TCGA LUAD cohort, n (Low) = 159, n (High) = 354. In TCGA LUSC cohort, n (low) = 104, n (high) = 389. We next sought to characterize the infiltration of CXCL13+CD4+ T cells in the TME of NSCLC. A cohort containing 82 patients with NSCLC carried the weight of further mIHC assay labeling the CD4, CXCL13, and PanCK to reveal the localization and infiltration level of CXCL13+CD4+ T cells. As expected, the infiltration of CXCL13+CD4+ T cells only exists in the tumor tissue, with few cells in the adjacent normal tissue, suggesting that the induction of CXCL13+CD4+ T cells required the tumor contexture (Figure 1F,G). By comparing the overall survival time with different infiltration levels of CXCL13+CD4+ T cells, we found that high level of CXCL13+CD4+ T cells also indicated favorable outcomes (Figure 1H), while CXCL13–CD4+ T cells showed little correlation (Figure S2). To further validate these results, we tested the prognostic value of CXCL13+CD4+ T cells in two TCGA cohorts containing 526 LUAD patients and 501 LUSC patients (Figure 1I,J). The expression of CXCL13 was normalized to that of CD4 to represent the infiltration level of CXCL13+CD4+ T cells in the TME based on bulk RNA-seq data as described in a previous study. 13 CXCL13 was basically expressed only in CD4_CXCL13 cluster, further suggesting that the CXCL13/CD4 ratio can be regarded as the infiltration level of CXCL13+CD4+ T cells in the total CD4+ T cells (Figure 1E). In line with previous data, the high infiltration of CXCL13+CD4+ T cells was associated with a better prognosis (Figure 1I,J). We further performed a multivariate Cox regression including the level of CXCL13+CD4+ T cells, gender, age, tumor stage, and the American Joint Committee on Cancer (AJCC) stage to test the prognostic value of the infiltration of CXCL13+CD4+ T cells and we found that a high level of CXCL13+CD4+ T cells may serve as an independent indicator of favorable overall survival in patients with NSCLC (Figure S3). CXCL13+CD4+ T cells promote the anti-tumor immunity indirectly With the correlation between CXCL13+CD4+ T cells and tumor prognosis, we proceeded to explore how CXCL13+CD4+ T cells exert anti-tumor function. We conducted gene enrichment analysis based on the up-regulated genes in the CXCL13+CD4+ T cells (Figure 2A). Consistent with the highest tumor reactivity score, pathways associated with T cell activation and leukocyte cell‒cell adhesion were enriched in CXCL13+CD4+ T cells, implying their response to the tumor antigen in the TME. Recent works have confirmed that tumor-specific CD4+ T cells can exert anti-tumor function directly by killing tumor cells through perforin and granzymes or indirectly by recruiting other immune cells. To explore the function of CXCL13+CD4+ T cells, we compared the expression of T cell function-associated genes among each cell subset 29 (Figure 2B). In line with the results from the signaling pathway analysis, these cells exhibited a markedly up-regulated expression of activation marker genes. The expression of cell activation-related genes, including TNFRSF18, CD82, and TNFRSF4, was significantly up-regulated compared with other clusters (Figure 2C). We further characterized CXCL13+CD4+ T cells based on the scRNA profiling and compared the identity of these cells with those of canonical helper T cell signatures (Tfh, Th1, Th2, Th17, and naive T cells) (Figures 2B, S1H). The results suggested that CXCL13+CD4+ T cells shared features with both Th2 and Tfh cells, which are well known for their regulation on humoral immunity, particularly the function of B cells. 37 , 38 Further, despite a lack of direct effector factors, cytokines and co-stimulatory molecules that are known to drive B cell maturation and activation, such as IL-21 and CD40LG, were found to be highly expressed in CXCL13+CD4+ T cells. Notably, IL-21 from tumor reactive CD4+ T cells has been reported to enhance the CD8+ T cells mediated anti-tumor immunity and CD40-CD40L signaling was reported to be required for induction of germinal center B cells in the murine lung adenocarcinoma. 39 However, CXCL13+CD4+ T cells expressed relatively low level of effector and cytotoxic molecules, such as granzyme B and perforin (encoded by GZMB and PRF1, respectively). These data indicated a low cytotoxicity of CXCL13+CD4+ T cells and a non-direct tumor eradication. Further, we compared the distribution of CXCL13+CD4+ T cells within the TME to explore whether CXCL13+CD4+ T cells directly contact with tumor cells in comparison with CD8+ T cells. The contact with tumor cells requires proximity to tumors; thus, we defined the juxtatumoral region by the tumor boundary to test whether the cells were distributed in proximity of tumor cells (Figure 2D). The cytotoxic CD8+ T cells were found to infiltrate into the proximity of the tumor cells, suggesting their direct contact with tumor cells, while the vast majority of the CXCL13+CD4+ T cells were distributed outside the boundary, which implied that the CXCL13+CD4+ T cells were not responsible for direct tumor killing through direct interaction (Figure 2D). To test the association between the cytotoxicity mediated by CD8+ T cells, by using ssGSEA algorithm, we comprehensively assessed the expression level of cytotoxicity-related genes and output a cytotoxic score (Figure 2E). A high cytotoxic score represents a high expression of cytotoxicity-related genes and implies a more effective anti-tumor response. We tested the relationship between CXCL13+CD4+ T cells infiltration level and cytotoxic score in both TCGA LUAD and LUSC cohort which turned out that the patients with a high level of CXCL13+CD4+ T cells infiltration showed a significantly increased cytotoxic scores, suggesting a boosted anti-tumor immunity. Collectively, these data implied that CXCL13+CD4+ T cells may promote anti-tumor immunity through facilitating other tumor-killing cells, such as CD8+ T cells, rather than directly conducting tumor eradication. CXCL13+CD4+ T cells promote the anti-tumor immunity indirectly. (A) Gene enrichment analysis showing enriched pathways up-regulated in CD4_CXCL13 cell cluster compared with the other CD4 clusters. (B) Heatmap revealing the enrichment analysis of hallmark gene sets in CD4+ T cell clusters. (C) Violin plots showing the expression of TNFRSF18, CD82, and TNFRSF4 in CD4+ T cell clusters. (D) Representative images of the cell distribution of CXCL13+CD4+ T cells and CD8+ T cells around the tumor boundary. Comparison of cell density of CD8+ T cells and CXCL13+CD4+ T cells within ±100 µm of the tumor boundary. n (Inside) = 12, n (Outside) = 12. *p < 0.05, ns: not significant. (E) The analysis shows a significant difference in the cytotoxic ssGSEA score between low and high normalization of CXCL13 expression by CD4 groups in TCGA LUAD cohort and TCGA LUSC cohort. In TCGA LUAD cohort, n (High) = 263, and n (Low) = 263. In TCGA LUSC cohort, n (High) = 251, n (Low) = 250. ***p < 0.001. CXCL13+CD4+ T cells are associated with the infiltration of B cells and CD8+ T cells To explore the partner of CXCL13+CD4+ T cells in anti-tumor immunity, we conducted a correlation analysis between CXCL13+CD4+ T cells and other immune cells. Our previous data suggested that CXCL13 exclusively expressed in CD4+ T cells; thus, normalization of CXCL13 expression by CD4 was calculate to represent the infiltration level of CXCL13+CD4+ T cells based on bulk RNA-seq cohort. 13 First, the positive relationship of immune score estimate and normalization of CXCL13 expression by CD4, which was utilized to represent the infiltration of CXCL13+CD4+ T cells based on bulk-RNA seq data, was validated in LUAD and LUSC cohort from TCGA, suggesting a high level of normalization of CXCL13 expression by CD4 was in favor of immune cell infiltration (Figure 3A). There was no correlation between the stroma score analysis and the normalization of CXCL13 expression by CD4 (Figure S4). Thus, the partner of CXCL13+CD4+ T cells would be immune cells instead of stromal cells. Tumor-infiltrating lymphocytes (TILs) are the major contributors to immune defense against malignancy and have been recognized to be a biomarker for patient outcome stratification and reflected the immune infiltration status of TME. 40-42 Subsequently, the analysis of the relationship between different cell types and the level of normalization of CXCL13 expression by CD4 was conducted. The correlation between the expression of CXCL13 normalized to CD4 and B cells was the highest in comparison with the association with other cell types (Figure 3B). Notably, a high correlation between the expression of CXCL13 normalized to CD4 and CD8+ T cells was also found, which implied that CXCL13+CD4+ T cells coordinated an effective anti-tumor response. CXCR5 is the receptor of CXCL13 was found in previous studies. 43 , 44 CXCL13+CD4+ T cells are associated with the infiltration of B cells and CD8+ T cells. (A) Correlation scatter plots of ImmuneScore Estimate and normalization of CXCL13 expression by CD4 in TCGA LUAD cohort and TCGA LUSC cohort. In TCGA LUAD cohort, n = 526. In TCGA LUSC cohort, n = 501. (B) Lollipop plots depicting the relationship between different cell types and levels of normalization of CXCL13 expression by CD4. (C) Representative images of mIHC staining with seven markers. (D) Representative images of different cell types according to the identified markers. (E) Correlation scatter plot of CXCL13+CD4+ T cell density and B cell density. (F) Correlation scatter plot of CXCL13+CD4+ T cell density and CD8+ T cell density. (G) Correlation scatter plot of CXCL13+CD4+ T cell density and CD4+ T cell density. To verify the partnership between CXCL13+CD4+ T cells, B cells, and CD8+ T cells, we established a seven biomarkers panel staining for further analysis, including DAPI, tutor epitopes (PanCK), three different lymphocyte epitopes (CD4, CD8, and CD20), and a ligand–receptor pair (CXCL13, CXCR5) (Figure 3C,D). We assessed TILs, as the sum of T cells (CD8+ T cells and CD4+ T cells) and B cells (CD20+ B cells), to evaluate the immune infiltration status of TME in our data. Further, to normalize the variation of TILs in different individuals, we analyzed the correlation between the proportion of CXCL13+CD4+ T cells and B cells as well as CD8+ T cells in TILs. Based on the mIHC analysis results, consistent with previous data, CXCL13+CD4+ T cells shown a positively correlated with B cells, CD8+ T cells, and CD4+ T cells (Figures 3E–G, S5A–C). These data indicated that CXCL13+CD4+ T cells coordinated with lymphocytes in the TME to elicit the anti-tumor immunity. CXCL13+CD4+ T cells recruit B cells through CXCL13–CXCR5 axis It has been reported by others that CXCL13 functions through CXCR5 receptor signaling. 43 , 44 The colocalization of CXCL13 and CXCR5 was identified in NSCLC TME (Figure 4A). To further identify the potential partner of CXCL13+CD4+ T cells in the TME, we quantified the location of CXCR5, and the results suggested that CXCR5 was predominantly expressed in B cells (Figure 4B). Previously, CXCR5+ B cells were reported to show follicular-like features and participate in mature TLS function. 10 , 45 In our data, we performed further sub-clustering analysis on B cells and found that CXCR5 was exclusively expressed in memory B cell clusters, which showed expression of MS4A1 and TNFRSF13B, implying that these CXCR5+ B cells have potentially been exposed to tumor antigens (Figure S6A). Furthermore, cell communication analysis suggested that the CXCL13–CXCR5 axis regulates the cell‒cell interactions between CXCL13+CD4+ T cells and B cells, which was undetectable in other CD4+ T cell subsets (Figure 4C). The expression of specific ligands and receptors in different cell clusters also confirmed that CXCL13+CD4+ T cells had an essential connection with CXCR5+ B cells (Figure 4D). We then compared the distribution of B cells around CXCL13+CD4+ T cells by mIHC (Figure 4E). CXCR5+ B cells were more abundant than CXCR5– B cells proximal (within 10 μm) to CXCL13+CD4+ T cells (Figure 4F). Next, a comparison of distribution of CXCR5+ B cells and CXCR5– B cells from the CXCL13+CD4+ T cells within a radius of 0–100 μm was conducted, and the results turned out that the CXCR5+ B cells increased significantly within 10 μm from the CXCL13+CD4+ T cells (Figure S6B). These results indicated that CXCL13+CD4+ T cells coordinate with B cells and recruit B cells through the CXCL13–CXCR5 axis. CXCL13+CD4+ T cells recruit B cells through the CXCL13–CXCR5 axis. The representative images showing the colocalization of CXCL13 and CXCR5. (B) The representative images and localization of CXCR5+ cells evaluated by mIHC. (C) Expression of CXCR5-CXCL13 and CCL20-CCR6 in each CD4+ T cell cluster. (D) Expression of CXCR5 and CXCL13 in each cell cluster. (E) Representative images depicting B cells in proximity to CXCL13+CD4+ T cells. (F) Numbers of B cells proximal (within 10 μm) to CXCL13+CD4+ T cells from 34 patients. The data are the means ± s.e.m. (G) Representative flow plots of the transwell migration assay. (H) B cell migration counts in the presence or absence of anti-CXCL13/anti-CXCR5 neutralizing antibody (n = 4). ***p < 0.001. To mimic the recruitment in the TME and prove the specificity of the recruitment, peripheral blood mononuclear cells (PBMCs) from patients were used to carry the weight of experiments rather than purified T cells and B cells. We pre-coated the lower chamber with αhCD3/CD28 to mimic the antigen stimulation and induced antigen-reactive CD4+ T cells (CXCL13+CD4+ T cells), and those PBMCs seeded in the upper well, set to represent the immune cells in the circulation of patients, were labeled with cell trace violet (CTV) dye to be distinguished their counterparts in the lower well. The cells in the lower well were collected for the flow cytometry analysis to detect the recruited cells that were positive for CTV. The results revealed that the count of CTV+CD19+ B cells in the lower well was significantly higher after 36 h of stimulation in comparison with the unstimulated control, which represented chemokine-independent basal level migration. Further, we used the CXCL13 and CXCR5 blockade to disrupt this signaling that considered to be pivotal for the recruitment of B cells by CXCL13+CD4+ T cells. The results suggested that both CXCL13 and CXCR5 blockade abolished the recruitment of CTV+CD19+ B cells, either treated individually or in combination (Figures 4G,H, S7). These results suggested that the recruitment of B cells was dependent on T cells derived CXCL13 and the CXCR5 on B cells, and thus verified that CXCL13+CD4+ T cells recruit B cells. Triad among CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells promotes the NSCLC patients prognosis Previously, CXCL13 was reported to govern the recruitment of CD8+ T cells into the TME by interacting with CXCR3. 46 , 47 We performed a cell-cell interaction analysis by the CellChat algorithm to verify this pattern in lung cancer, and the results turned out that CXCL13+CD4+ T cells showed interaction with CD8+ T cells through CXCL13–CXCR3 signaling, implying that CXCL13+CD4+ T cells also recruited the CD8+ T cells. Notably, antigen presentation and activation-related signalings, such as CD80-CD28 and HLA-related interactions, were also found between CXCL13+CD4+ T cells and CD8+ T cells (Figuer S8). Given the strong chemoattractant effect of CXCL13+CD4+ T cells, we hypothesize that CXCL13+CD4+ T cells may serve as a coordinator that bridge the CXCR5+ B cells and CD8+ T cells to collaborate in exerting anti-tumor immunity. Following this notion, after confirming the quantitative associations between CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells, we sought to identify the spatial association among these three cell types. The mIHC images suggested that the proximity of CXCL13+CD4+ T, CXCR5+ B, and CD8+ T cells coexisted in situ in the NSCLC TME (Figure 5A). Subsequently, entropy was introduced from a spatial platform named SPIAT to investigate the spatial heterogeneity of the co-occurrence of different cell types (Figure 5B). In brief, the TME was separated into multiple units by the Entropy algorithm to calculate the prevalence of these three cell types, and a high prevalence suggested a more aggregated distribution of these three cell types instead of an irrelevant random distribution. The entropy algorithm was employed to reveal the distribution pattern of CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells. Spots with high prevalence identified in the TME suggested the existence of CXCL13+CD4+ T cells:: CXCR5+ B cells:: CD8+ T cells triad (Figure 5C). Further, patients with a high frequency of CXCL13+CD4+ T cells, namely, CXCR5+ B CD8+ T cells triad, heralded a better prognosis, indicating the anti-tumoral synergy between CXCL13+CD4+ T cells, CXCR5+ B cells, and cytotoxic CD8+ T cells (Figure 5D). To adjust the contribution of the infiltration level and address how the composition of lymphocytes was symphonized to exert anti-tumor immunity, we calculated the proportion of different cell types in tumor-infiltrating lymphocytes and conducted multivariate Cox analysis (Figure 5E). The results revealed that only the triad itself could serve as a protective factor, and the individual cell densities were not sufficient to herald an improved outcome, which suggested the significance of spatial coordination in anti-tumor immunity rather than the additive effect of all cell types. Collectively, the triad among CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells in TME promoted the prognosis of NSCLC patients. Triad among CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells promotes the NSCLC patients prognosis. (A) The representative images showing the proximity of CXCL13+CD4+ T cell, CXCR5+ B cell, and CD8+ T cell. (B) The calculation process of a pattern of spatial heterogeneity. Briefly, each tissue was divided into 225 (15 × 15) grids, and in each grid, entropy was calculated based on the co-localization of interested cell types. A high value of entropy indicates that the cells are co-localized. Next, the proportion of grids with entropy higher than the positive value was calculated, which was termed prevalence. A higher prevalence indicates that the interested cell type triad is present in greater quantities in the patients. Different colors were used to distinguish different cell types. (C) The representative images showing the calculation results for prevalence. (D) Survival plot of the prevalence level in NSCLC. n (Low prevalence) = 42, n (High prevalence) = 40. (E) The multivariable cox analysis of different cell types proportion in tumor-infiltrating lymphocytes. CXCL13+CD4+ T cells are associated with TLS-like structure Our results showed that the triad among CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells in TME predicted a favorable outcome. We wondered whether such lymphocyte aggerates would correlate with TLS. We combined our mIHC data with HE imaging to assess the relationship between the triad and TLS. The patients with high prevalence showed a higher number of TLS detected in the TME, suggesting that the triad identified in our data was closely associated with TLS (Figures 6A, S9A). The mIHC images also showed that CXCL13+CD4+ T cells existed in TLS-like structure in NSCLC tissues (Figure 6B). Further, normalization of CXCL13 expression by CD4 was tested to evaluate the immune infiltration characteristics in two levels of tertiary lymphoid structure signatures groups in LUAD and LUSC cohorts from TCGA, and the results showed that normalization of CXCL13 expression by CD4 in high level of tertiary lymphoid structure signatures group was generally higher than that in the low-level group (Figure 6C,D). Furthermore, the relationship between normalization of CXCL13 expression by CD4 and tertiary lymphoid structure signatures was confirmed, indicating that CXCL13+CD4+ T cells were positively associated with TLS (Figure 6E,F). Next, we found that the TLS structure could be identified in HE image of the NSCLC tissues in the TCGA cohort (Figure 6G). To further investigate whether the CXCL13+CD4+ T cells are associated with TLS-like structure, paired diagnostic HE-stained slides were downloaded from TCGA database, and TLS scores were calculated for subsequent analysis. Sankey diagrams were created for visual presentation between the normalization of CXCL13 expression by CD4 and the TLS scores in TCGA LUAD cohort and TCGA LUSC cohort (Figure 6H,I). The results suggested that the vast majority of patients with high normalization of CXCL13 expression by CD4 showed high TLS score, indicating that CXCL13+CD4+ T cells are associated with TLS-like structure. Functionally, we further analyzed the correlation between CXCL13+CD4+ T cells infiltration and established markers of TLS maturity, including CD21 (encoded by CR2), CD23 (encoded by FCER2), and PNAd (which is composed of CD34, DACAM1, and GLYCAM1) in TCGA LUAD and LUSC cohort (Figure S9A,B). The results revealed a high positive correlation between CXCL13+CD4+ T cell infiltration and the follicle dendritic cell markers CR2 and FCER2 while a negative or weak positive correlation with peripheral node addressin marker (CD34, DACAM1, and GLYCAM1). Further, we characterized CXCR5+ B cells and found that these B cells showed both memory- and follicular-like features (Figure S6A). Given that the positive correlation with follicle dendritic cells markers and interaction with CXCR5+ follicular like B cells, the identified triad may be a stage of primary follicle-like TLS or a more detailed functional unit of mature TLS which has been recognized for mediating anti-tumor T cell immunity. 45 CXCL13+CD4+ T cells are associated with TLS-like structure. Correlation scatter plot of prevalence and TLS-like structure counts in mIHC cohort. (B) The multiplex staining images showing CXCL13+CD4+ T cells existing in TLS-like structure in NSCLC tissues. Yellow arrows depict CXCL13+CD4+ T cells. (C, D) The analysis showing a significant difference in the levels of normalization of CXCL13 expression by CD4 between the low- and high-tertiary lymphoid structure signature groups in TCGA LUAD cohort and TCGA LUSC cohort. In TCGA LUAD cohort, n (High) = 263, and n (Low) = 263. In the TCGA LUSC cohort, n (High) = 251, n (Low) = 250. The data are means ± s.e.m. ***p < 0.001. (E, F) Correlation scatter plots of normalization of CXCL13 expression by CD4 levels and tertiary lymphoid structure signatures in TCGA LUAD cohort and TCGA LUSC cohort. In TCGA LUAD cohort, n = 526. In TCGA LUSC cohort, n = 501. (G) The representative HE image showing TLS-like structure existing in NSCLC tissue in the TCGA LUAD cohort. (H, I) Sankey diagram showing the associations between the levels of normalization of CXCL13 expression by CD4 and the TLS scores in TCGA LUAD cohort and TCGA LUSC cohort. In TCGA LUAD cohort, n = 94. In TCGA LUSC cohort, n = 106. (J) A cartoon depicting a proposed model of CXCL13+CD4+ T cells recruiting B cells through CXCL13–CXCR5 axis, then CXCL13+CD4+ T cells and CXCR5+ B cells co-localized with cytotoxic CD8+ T cells and thus forming a functional niche against tumor cells. Discussion The TME is a spatially organized unity of different cell types, and the multicellular structures that form within the TME represent the functionality of immune cells and are of pathological and prognostic value. 22 , 48-51 In addition to the heterogeneity of cell types revealed by scRNA-seq, the spatial analyses coupled with multiplexed imaging techniques provide deep insight into how diverse cell types were symphonized in the TME. 51 Here, through state-of-art spatial analyzing algorithm, our data identified that CXCL13-secreting CD4+ T cells recruit B cells through CXCL13–CXCR5 axis to form the CXCL13+CD4+ T cells:: CXCR5+ B cells:: CD8+ T cells triad to promote the anti-tumor immunity and predict the outcome of patients with NSCLC (Figure 6J). CD4+ T cells have been recognized to be the key coordinator in the TME to orchestrate anti-tumor immunity. 1 , 3 , 5 , 7-11 , 13 The infiltration of either total of subsets of CD4+ T cells has been reported by us and others to be of prognostic value. 3-13 , 17 , 52 Previous studies suggested the prognostic value of cytotoxic subsets of CD4+ T cells, like TFH-like cytotoxic CD4+ T cells, while the association between cytokine-producing compartment of CD4+ T cells and the overall survival of patients with NSCLC was little known. In our cohort, we demonstrated that high infiltration of CXCL13+CD4+ T cells was correlated with prolonged survival. Intriguingly, despite the association between favorable prognosis and CXCL13+CD4+ T cells, our scRNA-seq profiling of CXCL13+CD4+ T cells showed that these cells were activated by tumor antigen but lacked effector molecular that participates in the direct tumor killing. The further spatial analysis revealed that the majority of CXCL13+CD4+ T cells were distributed distantly from the tumor cells, further supporting its non-direct interaction with the tumor cells. Thus, we hypothesized that these tumor-reactive CXCL13+CD4+ T cells regulated other cellular components in the TME to promote the anti-tumor immunity. Aside from the direct killing, the cytokine-producing properties of CD4+ T cells are essential for coordinating the well function of other cell types, yet little is known about how CD4+ T cells regulate the distribution of other cells in the TME. Previous studies reported that CXCL13+CD4+ T cells participate in the formation of TLS and other spatial architectures in human melanoma and ovarian cancer. 3 , 10 , 13 , 16-18 In addition to recognize the functions of CD4+ T cells, our work defined the spatial remodeling capacity of CD4+ T cells in NSCLC. First, to identify the interacting partner of CXCL13+CD4+ T cells, our bulk RNA-seq data revealed a correlation between CXCL13+CD4+ T cells and lymphocytes, including CD8+ T cells and B cells, but not the stromal cells, indicating a specialized effect on the infiltration of immune cells. Further ligand‒receptor analysis suggested an exclusive expression pattern of the CXCL13‒CXCR5 axis between CXCL13+CD4+ T cells and CXCR5+ B cells. The subsequent multiplexed staining revealed the colocalization of CXCL13 and CXCR5 in the TME of NSCLC, suggesting that CXCL13+CD4+ T cells recruit CXCR5+ B cells through CXCL13‒CXCR5 axis. The transwell migration assay further validated the specificity and recruitment of B cells by CXCL13+CD4+ T cells, which rely on the CXCL13‒CXCR5 axis. In addition, we also identified the CXCL13–CXCR3 signaling between CXCL13+CD4+ T cells and CD8+ T cells. CXCR3 is known to be expressed on the activated CD8+ T cells and is responsible for the chemotaxis of CD8+ T cells. 53 , 54 Meanwhile, CXCL13 has been reported to govern the recruitment of CD8+ T cells into the TME by interacting with CXCR3. 46 , 47 Here, we provided a clue that CXCL13+CD4+ T cells may simultaneously recruit both CD8+ T cells and B cells. These findings revealed the spatial remodeling capacity of CXCL13+CD4+ T cells and heightened the role of CXCL13+CD4+ T cells as a coordinator that bridges the CXCR5+ B cells and CD8+ T cells to collaborate in exerting anti-tumor immunity. Previously, CD4+ T cells were reported to participate in the triad of CD4+ T cell:CD8+ T cell:APC in solid tumors that is required for immunotherapy. 5 , 55 Progenitor CD8+ T cells were found to interact with CXCL13+CD4+ T cells within cellular triads around dendritic cells. 55 With the aim to quantify the cell-cell interaction and confirm whether such interaction would affect the prognosis, we employed Entropy algorithm to detect the colocalization among CXCL13+CD4+ T cells, CXCR5+ B cells, and CD8+ T cells. Our data revealed a CXCL13+CD4+ T cells:: CXCR5+ B cells:: CD8+ T cells triad that correlates with a favorable prognosis. Functionally, antigen presentation- and activation-related signaling, such as CD80-CD28 and HLA-related interaction, were also identified between CXCL13+CD4+ T cells and CD8+ T cells. Notably, CXCL13+CD4+ T cell infiltration was found to be positively correlated with the follicle dendritic cell markers CR2 and FCER2, while a negative or weak positive correlation with peripheral node addressin marker (CD34, DACAM1, and GLYCAM1). These data suggested that the identified triad may be a stage of primary follicle-like TLS, which has been recognized for mediating anti-tumor T cell immunity. 45 Distinct from the previous CD4+ T cell::CD8+ T cell::APC triad reported in the TME, we found that B cells could also play a role in the CD8+ T cells promoting triad, suggesting that there is crosstalk between cellular immunity and humoral immunity. The CXCR5+ B cells involved in the triad exhibited memory- and follicular-like features, indicating that these CXCR5+ B cells are potentially exposed to tumor antigens. We found that high CXCL13+CD4+ T cell infiltration can herald a better prognosis, while the origin of CXCL13+CD4+ T cells and their differentiation remained to be deciphered. Despite the prognostic value of the CXCL13+CD4+ T cells:: CXCR5+ B cells:: CD8+ T cells triad, how the function of CD8+ T cells was modulated requires further investigation. Conclusions In conclusion, upon mining scRNA-seq data from a public database, a specific CD4+ T cells subset expressing CXCL13 showed a potent neoantigen-reacting function and recruited B cells through CXCL13–CXCR5 signaling. mIHC further verified that patients with high CXCL13+CD4+ T cells infiltration showed a better prognosis. The infiltration of CXCL13+CD4+ T cells correlates with CD8+ T cells and B cells infiltration in TME, and it can recruit B cells through the CXCL13–CXCR5 axis. Spatial analysis revealed the formation of CXCL13+CD4+ T cells:: CXCR5+ B cells:: CD8+ T cells triad and its correlation with favorable outcomes. Moreover, the infiltration of CXCL13+CD4+ T cells was also associated with the formation of TLS, suggesting that this triad may represent the early, immature form of TLS. Overall, tumor-specific CXCL13+CD4+ T cells can form a triad with CD8+ T cells and B cells to exert durable anti-tumor immune responses. The identification of this triad shed new insight into the spatial regulatory mechanism of anti-tumor immunity and facilitated to predict the outcome of patients with NSCLC. Supplementary Material Table S3 List of abbreviations.xlsx Table S3 List of abbreviations.xlsx Table S2 Panel information.xlsx Table S2 Panel information.xlsx Table S1 Characteristics of patients.docx Table S1 Characteristics of patients.docx Figure S6.tif Figure S6.tif Figure S5.tif Figure S5.tif Figure S3.tif Figure S3.tif Figure S4.tif Figure S4.tif Figure S8.tif Figure S8.tif Figure S9.tif Figure S9.tif Figure S2.tif Figure S2.tif Figure S1.tif Figure S1.tif Figure S7.tif Figure S7.tif Supplementary Material Supplementary Material Funding Statement This work was supported by grants from the National Natural Science Foundation of China (Grant Nos. 82471820 and 82071811) and supported by Hainan Provincial Natural Science Foundation of China (Grant Nos. 824MS169 and 825MS201). Disclosure of potential conflicts of interest The authors declare that they have no potential conflict of interest​​​​​. Acknowledgements We are grateful to the authors of R packages from SPIAT. DX, WG, and ZG designed this study. DL and XY performed the mIHC experiment. DL and JL. performed the data analysis. DL and JM wrote the manuscript. JM provided help for the organization and revision of the manuscript. XY and YX helped with data interpretation. All the authors read and revised the manuscript critically and approved the final manuscript. Data availability statement The authors declare that the main data supporting the findings of this study are available within the article and its Supplementary Information. The data presented in this study and codes are available upon reasonable request from the corresponding author. The abbreviation used in this article was provided in the Supplementary Table S3. Ethics approval statement This study was conducted in accordance with the Declaration of Helsinki and was approved by the Ethics Committee of Ruijin Hospital, Shanghai Jiao Tong University School of Medicine (Protocol:KY 2018-104). All patients gave written informed consent for participation. Consent for publication Written informed consent has been obtained from the patients to publish this paper. Supplementary material Supplemental data for this article can be accessed at https://doi.org/10.1080/2162402X.2026.2684116. References Associated Data Supplementary Materials Table S3 List of abbreviations.xlsx Table S3 List of abbreviations.xlsx Table S2 Panel information.xlsx Table S2 Panel information.xlsx Table S1 Characteristics of patients.docx Table S1 Characteristics of patients.docx Figure S6.tif Figure S6.tif Figure S5.tif Figure S5.tif Figure S3.tif Figure S3.tif Figure S4.tif Figure S4.tif Figure S8.tif Figure S8.tif Figure S9.tif Figure S9.tif Figure S2.tif Figure S2.tif Figure S1.tif Figure S1.tif Figure S7.tif Figure S7.tif Supplementary Material Supplementary Material Data Availability Statement The authors declare that the main data supporting the findings of this study are available within the article and its Supplementary Information. The data presented in this study and codes are available upon reasonable request from the corresponding author. The abbreviation used in this article was provided in the Supplementary Table S3.