Introduction

Gastric cancer is among the most prevalent malignancies worldwide, occupying the fifth position in terms of tumor incidence and the fourth in mortality worldwide.1 Over 70% of individuals with gastric cancer are diagnosed at locally advanced or metastatic stages in China. The overall prognosis of patients with gastric cancer is poor due to a high proportion of patients with advanced disease. Furthermore, effective treatment options for these patients remain lacking. Systemic chemotherapy as initial therapy is the mainstay of treatment for advanced gastric cancer. However, the median survival time is still limited to approximately 15 months when patients are treated with chemotherapy alone.2

Recently, several cohort studies have validated the effectiveness of programmed cell death protein 1 (PD-1) inhibitors plus chemotherapy in the first-line treatment of advanced gastric cancer.3–7 Additionally, conversion surgery has demonstrated the potential to prolong survival by over 20 months in patients with stage IV gastric cancer who have achieved a satisfactory clinical response to immunochemotherapy.8 Nonetheless, there is still a considerable unmet need for higher efficacy rates. Due to intratumor heterogeneity, only a few patients derived benefits from this treatment regimen. Therefore, it is absolutely imperative to screen the predominant beneficiary population and reverse the refractoriness in patients treated with immunochemotherapy.

Yoshida et al had categorized patients with stage IV gastric cancer into four groups based on anatomical relationships and surgical techniques. Although this classification aids in selecting appropriate candidates for conversion therapy, its precision is limited according to the follow-up results.9 Thus, precise molecular stratification of stage IV gastric cancer is necessary. The efficacy of immunochemotherapy is strongly linked to the response to immune checkpoint inhibitors, especially PD-1 inhibitors. A series of biomarkers represented by microsatellite instability-high (MSI-H) / mismatch repair deficiency (dMMR) status and programmed cell death-ligand 1 (PD-L1) expression have been frequently used to predict the response to PD-1 inhibitors in gastric cancer. Nevertheless, the incidence of MSI-H/dMMR gastric cancer is low and reported to be 5–15%.10 Additionally, PD-L1 expression is not an optimal biomarker for anti-PD-1 therapies. Even if the combined positive score exceeds 10, approximately 30% of patients with microsatellite stability (MSS) gastric cancer would not benefit from immunochemotherapy.11 Thus, it is crucial to discover novel biomarkers for immunotherapy. Additionally, identifying new molecular targets to reverse the refractoriness is also an urgent priority for non-responders.

Tertiary lymphoid structures (TLSs) are lymphoid-like structures formed in chronically inflamed non-lymphoid tissues, including tumors.12 Effector T cells and high-affinity antibodies within TLSs could directly target and eliminate tumor cells because TLSs are not encased in membranes. Previous studies have demonstrated that a high density of TLSs was associated with extended survival in gastric cancer.13 14 Apart from that, it has been reported that TLSs and B cells were implicated in the response to immunotherapy across various tumors, such as melanoma,15 renal cell carcinoma,16 head and neck squamous cell carcinoma17 and sarcoma.18 However, the relationship between TLSs and immunotherapy efficacy in gastric cancer remains largely unexplored. Moreover, inducing the formation or maturation of TLSs represents an effective anticancer strategy with promising clinical implications.19 Nonetheless, there are currently no safe and effective agents available for inducing TLS formation in cancer treatment. Of note, CXCL13 has been reported to be essential for TLS formation. Moreover, CXCL13 played a critical role in the humoral immune response and the maintenance of the germinal center.20 With the development of single-cell sequencing, new T-cell subpopulations, such as CXCL13+ T cells, were continually identified. Multiple studies have shown that intratumoral CXCL13+ T cells could recruit B lymphocytes to promote TLS formation and maturation.21 22 However, the characteristics and the cell-intrinsic regulatory mechanisms of these CXCL13+ T cells remain elusive. Moreover, investigating the mechanisms of CXCL13+ T cell-B-cell interactions could provide novel ideas about TLS-inducing agents.

In this research, we assessed tumor infiltration by diverse immune cell phenotypes, including TLSs and CXCL13+ T-cell populations, in tissue samples obtained before and after immunochemotherapy from patients with advanced gastric cancer. Additionally, we found that vitamin B6 supplementation or targeting pyridoxal kinase (PDXK) could boost the efficacy of immunotherapy primarily promoting the formation of TLSs through the upregulation of CXCL13 expression in CXCL13+ CD160+ CD8+ T cells. This study will provide new insights into the treatment of advanced gastric cancer.

MethodsData collection

The gastric cancer single-cell RNA sequencing (scRNA-seq) data sets used in this research were obtained from GSE206785, GSE167297, GSE150290, and GSE183904 which were from GEO database (https://www.ncbi.nlm.nih.gov/geo/), CRA002586 which was from the Genome Sequence Archive (https://ngdc.cncb.ac.cn/gsa/browse/CRA002586), and another scRNA data set was submitted by Anuja Sathe et al. We collected and collated the expression profile, along with the clinical information of the ARCG cohort (GSE66229), which contained 300 gastric cancer samples. We gathered the spatial transcriptome sequencing data from nine primary gastric cancer samples and one peritoneal metastatic lesion from GSE251950.

Spatial transcriptomics analysis

The gene-spot matrices generated after spatial transcriptomic (ST) data processing from ST and visium samples were analyzed with the Seurat package (V.4.1.1) in R. Spots were filtered for a minimum detected gene count of 200 genes. In contrast, genes with fewer than 10 read counts or expressed in fewer than 3 spots were removed. Normalization across spots was performed with the LogVMR function. Dimensionality reduction and clustering were performed with independent principal component analysis (PCA) with the first 30 principal components (PCs). We collected TLS-50 signatures from Wu et al in online supplemental table S123 and then applied the “AddModuleScore” function with default parameters in the Seurat package to calculate the gene scores of the TLSs. The cell type signature score derived from the scRNA-seq data set was added to “metadata” of the ST data set with the “AddModulScore” function with default parameters in Seurat. Spatial feature expression plots were generated with the “SpatialFeaturePlot” function in the Seurat package. We applied the SpaGene method to identify distinct clusters for each sample.

Single-cell transcriptome analysis

CellRanger (V.6.0.2) was used to read mapping and gene expression quantification. Seurat package (V.4.1.1) was applied for downstream analysis. Cells with less than 1,000 unique molecular identifiers (UMIs) or >15% mitochondria genes were excluded. Doublets were assessed using the DoubletFinder (V.2.0.3) algorithm for each sample. We used single-cell variational inference (scVI) tools to remove batch effects to ensure that our single-cell analysis was not unduly driven by the outlying characteristics of one or a few patients’ transcriptional profiles. We scaled data with the top 4,000 most variable genes by using the FindVariableFeatures function and used variable genes for PCA, used FindNeighbors to get nearest neighbors for graph clustering based on the top 50 PCs, and used FindCluster to obtain cell subtypes, and visualized cells with the Uniform Manifold Approximation and Projection (UMAP) algorithm. Then, we filtered T cells based on CD3D, CD3E, and CD3G expression and used FindCluster to obtain T-cell subtypes.

Clinical samples and TLSs assessment

In this study, 116 clinical advanced gastric cancer tissues were obtained from patients who underwent radical gastrectomy at the First Affiliated Hospital of Nanjing Medical University between 2016 and 2018 and none of these patients were treated with neoadjuvant chemotherapy or immunotherapy. Additionally, 15 gastric cancer tissues from patients who received immunotherapy combined with chemotherapy were obtained between 2019 and 2024, either from radical gastrectomy or preoperative gastroscopy with gastric biopsy at the same institution. The immunotherapy regimens included camrelizumab, nivolumab, tislelizumab, sintilimab. The chemotherapy regimens included oxaliplatin plus calcium levofolinate, 5-fluorouracil, S-1 plus oxaliplatin, oxaliplatin plus capecitabine, 5-fluorouracil plus leucovorin, oxaliplatin, and docetaxel. Trastuzumab was recommended for patients with HER2-positive cancers. The tumor regression grade (TRG) was evaluated by experienced pathologists according to the American Joint Committee on Cancer/College of American Pathologists system. TRG 0 means a complete response (CR), with no residual tumor cells and fibrosis extending through the different layers of the gastric wall. The details for these 15 patients are shown in table 1.

Table 1

Basic information about 15 patients with stage IV gastric cancer who received immunochemotherapy as the initial treatment option

The pathological assessment of TLSs was primarily based on H&E-stained slides. TLSs are defined as dense lymphocyte aggregates (>100 cells) located among tumor cells or tumor fibrosis area. Furthermore, TLSs are characterized as a dense aggregation of unencapsulated B cells (CD20), accompanied by an adjacent T-cell zone (CD4 and CD8). Therefore, we performed immunohistochemistry (IHC) staining in some cases to adequately reveal the morphology and spatial distribution of the constituent cells of TLSs. These samples were incubated with anti-CD20, anti-CD8 or anti-CD4. The TLS histological scoring system was applied to gastric cancer tissues based on the results of IHC and H&E staining. The scoring system includes two aspects: (1) the number of TLS (N); (2) the ratio of TLS area versus tumor area (TLS %). The formula for TLS score was as follows: Score=N×TLS %.

Multiplex immunofluorescence

The prepared formalin-fixed and paraffin-embedded (FFPE) sections (4 µm) were first incubated in a dry oven and then immersed in xylene for deparaffinization. Rehydration was also performed. The sections were then placed with Tris-EDTA buffer for antigen retrieval. After cooled to room temperature (RT), slides were immersed in peroxidase-blocking solution to block endogenous peroxidase activity. Subsequently, multiplex immunofluorescence (mIF) kit (BioMed World, WAS15021) was used for staining. The slides were incubated with the primary antibodies at 4°C overnight. Then, slides were incubated with horseradish peroxidase (HRP)-conjugated secondary antibodies at room temperature. A quick wash in 1× phosphate buffered solution (PBS) buffer was followed by incubation with an appropriate fluorophore-conjugated tyramide signal amplification (TSA) at RT. Again, the slides were exposed to microware treatment to strip the tissue-bound primary/secondary antibody complexes and ready for labeling of the next marker. The primary antibodies (CD8, Abcam, 1:1000; CD20, Proteintech, 1:2000; CXCL13, Abcam, 1:2000; CD160, Abcam, 1:500; CD4, CST, 1:2000; PDCD1, Proteintech, 1:5000; CXCR5, Abcam, 1:1000), HRP-conjugated secondary antibodies (BioMed World, WAS12011), and fluorophore-conjugated TSA (TSA570, WAS10031; TSA520, WAS10021; TSA620, WAS10041; TSA690, WAS10061) were repeated until all markers were labeled. Finally, the slides were added with 4’, 6-diamidino-2-phenylindole (DAPI) in the dark at RT and mounted with anti-fade mounting medium. The PANNORAMIC MIDI II (3DHISTECH) was used to take images of tissue samples.

Metabolic activity analysis

The R package scMetabolism (V.0.2.1) assesses sets of genes from metabolic pathways to quantify the metabolic activity of single cells based on the scRNA-seq expression matrix. The scMetabolism analysis was performed with Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolism pathways and the VISION quantifying method.

Mice

Female and male mice (6–8 weeks) were used for all experiments. The 615 and C57BL/6 mouse strains were procured from the Shanghai Laboratory Animal Research Center. On acquisition, a 1-week acclimatization period was provided in a pathogen-free housing environment. For subcutaneous xenograft experiments, either 1×106/100 µL mouse forestomach carcinoma (MFC) cells (MeisenCTCC, CTCC-400–0334) or MFC-Luciferase cells (MeisenCTCC, CTCC-0497-Luc2) were digested and resuspended in PBS before being subcutaneously injected into 615 mice (with six mice per treatment group). The calculation of tumor volume was performed using the following formula: volume (mm3)=width2×length/2. In orthotopic xenograft experiments, tissue blocks (1 mm3) obtained from subcutaneous tumors were surgically transplanted into the stomach wall of 615 mice (with six mice per treatment group). The spontaneous gastric cancer model was created by exposing C57BL/6 mice to 240 ppm N-methyl-N-nitrosourea (MNU) in their drinking water.

The diet (from Jiangsu Xietong Pharmaceutical Bio-engineering) was strictly defined as non-VB6 diet, normal-VB6 diet, and excessive-VB6 diet, which contain 0, 6 mg, and 60 mg of vitamin B6 per kg diets, respectively. Mice were treated with anti-PD-1 antibody intraperitoneally (200 µg/mouse, every 3 days, Bio X Cell) or MS023 (2 mg/mouse, every day, Selleck) 1 week after subcutaneous injection of tumor cells or orthotopic transplantation. PBS was used as a negative control.

Patient-derived tumor fragments

Gastric cancer samples were collected from patients undergoing gastrectomy. The samples were diced into approximately 1 mm diameter fragments and then incorporated into an artificial extracellular matrix within a 96-well plate (Corning). The composition of the artificial extracellular matrix was elaborated below: 10% fetal bovine serum (FBS) - Dulbecco’s Modified Eagle’s Medium (DMEM) containing 1× Eagle’s minimal essential medium (MEM) non-essential amino acids (Sigma-Aldrich), 1 mg/mL collagen I (Corning) and 4 mg/mL phenol red-free ice-cold Matrigel (BD Biosciences). Patient-derived tumor fragments (PDTFs) cultures were stimulated with 10 µg/mL anti-PD-1 alone (Selleck, Sintilimab), 0.5 mM pyridoxal alone (Selleck), or the combination of anti-PD-1 and pyridoxal where indicated.

Analysis of secreted mediators

The supernatants of PDTFs cultures were collected after 48 hours of culture. The indicated cytokines and chemokines within the supernatants were detected using human IL-2 ELISA Kit (Solarbio, SEKH-0008), human IFN-γ ELISA Kit (Solarbio, SEKH-0046), human CXCL13 ELISA Kit (Solarbio, SEKH-0072) and human TNF-α ELISA Kit (Solarbio, SEKH-0047) according to the manufacturers’ instructions.

Statistical analysis

The data analysis was performed using GraphPad Prism V.10.0.2 software and SPSS V.27.0.1. Statistical significance was assessed employing two-tailed unpaired t-tests for comparing two groups, while one-way analysis of variance followed by Bonferroni’s multiple comparisons test was used for comparisons involving more than two groups. The results of in vitro functional assays were reported based on a minimum of three independent experiments and were expressed as mean±SD.

Additional experimental procedures are provided in the online supplemental information.

ResultsThe abundance and maturation of intratumoral TLSs were associated with improved prognosis and the efficacy of immunochemotherapy

Previous studies have demonstrated that a high enrichment of TLSs may prolong the survival of patients with gastric cancer.13 14 Initially, we evaluated the abundance of TLSs in tumor tissues from 116 patients with gastric cancer (table 2) and observed that the TLSs-high group had a superior outcome compared with the TLSs-low group (online supplemental figure S1A). Therefore, TLSs could serve as a prognostic biomarker for gastric cancer. Then, we began to investigate whether TLSs could be used as biomarkers to predict the efficacy of immunotherapy in advanced gastric cancer. Accordingly, we conducted a retrospective analysis using gastroscopic biopsies obtained from patients with stage IV gastric cancer prior to initiating immunotherapy (table 1). The results revealed a positive correlation between the response to immunotherapy and the density of intratumoral TLSs before treatment initiation (online supplemental figure S1B). Therefore, TLSs could be used as an index in the evaluation of immunotherapeutic effect in advanced gastric cancer.

Table 2

Association between TLS scores and clinicopathological features in patients with gastric cancer

Figure 1

Single-cell and spatial transcriptome profiling of T cells and TLSs in the tumor microenvironment of gastric cancer. (A) The spatial feature plot showing the scores of TLSs, T cells, B cells and macrophage in different patients’ spatial transcriptomic data sets. P3, patient 3; P10, patient 10. (B) Comparison of the scores of different immune cells between TLS-high and TLS-low spots in different patients’ spatial transcriptomic data sets. (C) UMAP plot of reclustered T cells showing 10 clusters annotated in different colors. (D) Dot plot showing the average expression levels and cell expression proportions of the differential genes in these 10 T-cell clusters. Dot size encodes the percentage of cells expressing the gene, color encodes the average per cell gene expression level. (E) CD4-C8-CXCL13 and CD8-C6-CXCL13 exhibiting dysfunctional and inhibitory characteristics. (F) The heatmap showing the characteristic gene expression in each T-cell cluster. (G) UMAP plot showing the expression of CD4, CD8B, PDCD1, CXCL13, ITGAE and CD160 in T cells. (H) The plot showing the abundance of immune cells as the score of TLSs increased using the ACRG data set. Analysis includes 300 processed samples. (I) Correlation of lineage-normalized cell-type frequencies in the ACRG cohort. (J) Scatterplot demonstrating the correlation between the CD4-C8-CXCL13 and the B cells (left panel). Scatterplot demonstrating the correlation between the CD8-C6-CXCL13 and the TLS (right panel). (K) Survival curves showing the association between factors and overall survival in patients from the ACRG cohort (log-rank test). Survival curves for TLS (upper panel), CD4-C8-CXCL13 (middle panel) and CD8-C6-CXCL13 (lower panel) and p values are shown. ACRG, Asian Cancer Research Group; TLSs, tertiary lymphoid structures; UMAP, Uniform Manifold Approximation and Projection.

Subsequently, we set out to interrogate the effect of immunochemotherapy on the formation and maturation TLSs. We found that the number of intratumoral TLSs increased after treatment (online supplemental figure S1C). Then, we classified the TLSs into three maturity levels based on expression of the markers CD2321: (1) lymphoid aggregates with no evidence of CD23+ mature follicular dendritic cells (mFDC); (2) primary follicles containing scattered CD23+ mFDC; and (3) mature TLSs, which are secondary follicles with a meshwork of CD23+ mFDC (online supplemental figure S1D). We found that patients with a partial response (PR) had a higher proportion of intratumoral mature TLSs after treatment compared with patients with a poor response (online supplemental figure S1E). However, these phenomena were not obvious in patients with a CR. We speculated that this might be attributed to the absence of sustained antigen stimulation for a long period before detection.

To further validate the impact of TLSs on gastric cancer development, we calculated the gene scores of TLSs in the spatial transcriptome data of gastric cancer and observed the presence of apparent TLS regions. Consistent with previous reports, TLSs-high spots were enriched with T cells, B cells, dendritic cells (DCs), and macrophages in contrast to the TLSs-low spots (figure 1A,B). In addition, we found high expression of CD3D (T-cell marker), CD79A (B cell marker), HLA-DQA1 (DC marker), and CD68 (macrophage marker) in TLSs-high regions (online supplemental figure S1F,G).

Collectively, these results reveal that intratumoral TLSs are associated with improved prognosis and the efficacy of immunochemotherapy. Besides, immunochemotherapy could induce TLS formation and maturation. However, what are the key elements that contribute to TLS formation and maturation in the immune microenvironment?

CXCL13+ CD160+ CD8+ T cells and CXCL13+ PDCD1+ CD4+ T cells were critical factors to promote TLS formation or maturation

We performed an integrative analysis of large-scale single-cell transcriptomes across six data sets using the scVI algorithm and re-clustered T cells and identified ten distinct subsets according to their differential genes (figure 1C,D). These T-cell subsets included C0-KLRC1 CD8+ T subsets, C1-CCR6 CD4+ T subsets, C2-GZMK CD8+ T subsets, C3-FOXP3 CD4+ T subsets, C4-GNLY CD8+ T subsets, C5-CCR7 CD4+ T subsets, C6-CXCL13 CD8+ T subsets, C7-TNF CD8+ T subsets, C8-CXCL13 CD4+ T subsets, and C9-IL17A CD4+ T subsets. Among these subsets, CD4-C8-CXCL13 and CD8-C6-CXCL13 were characterized by upregulated expression of CXCL13.

According to the literature reports, we reclassified T cells in patients with gastric cancer into naïve T cells, cytotoxic T cells, dysfunctional T cells, inhibitory T cells, proliferative T cells, effective T cells, and resident T cells based on their molecular features. CD4-C8-CXCL13 and CD8-C6-CXCL13 exhibited dysfunctional and inhibitory characteristics, as they expressed inhibitory signature genes (PDCD1, TIGIT, and TOX) and co-stimulatory signature genes (CD28, ICOS, and TNFRSF18) (figure 1E,F). Notably, gene expression density analysis confirmed the specific expression distribution of CD160 and ITGAE in CD8-C6-CXCL13 and the specific expression distribution of PDCD1 in CD4-C8-CXCL13 (figure 1G, online supplemental figure S2). Analysis of the bulk RNA-seq of gastric cancer tissues from the Asian Cancer Research Group (ACRG) further confirmed that CXCL13+ CD160+ CD8+ T cells were highly correlated with TLSs (R=0.83, p<0.0001) and CXCL13+ PDCD1+ CD4+ T cells were highly correlated with B cells (R=0.84, p<0.0001), which are crucial components of TLSs (figure 1H–J). Besides, we found that the factions of TLSs, CXCL13+ CD160+ CD8+ T cells, and CXCL13+ PDCD1+ CD4+ T cells were all associated with better survival in gastric cancer through analyzing data from the ACRG database (figure 1K).

Figure 2

CXCL13+ CD160+ CD8+ T cells and CXCL13+ PDCD1+ CD4+ T cell around TLSs at different stages of maturation were assessed by multiplex immunofluorescence. (A) Resected gastric tissues stained with H&E or subjected to immunohistochemistry (IHC) detection for CD8, CXCL13, CD160 and CD20. The framed areas are shown adjacently at a higher magnification. Images were captured under a light microscope. Scale bar, 500 µm. (B) Multiplex immunofluorescence staining of human gastric cancer tissues for detection of CXCL13 (pink), CD160 (orange), CD8 (red), CD20 (green) and DAPI in gastric cancer tissues. Scale bar, 200 µm. (C) Tumor sections stained with H&E and IHC, including CD4, CXCL13, PDCD1 and CD20. The framed areas are shown adjacently at a higher magnification. Scale bar, 500 µm. (D) Multiplex immunofluorescence experiments were performed to detect the degree of CXCL13+ PDCD1+ CD4+ T-cell infiltration around TLSs at different stages of maturation. Antibody panel: CXCL13 (pink), PDCD1 (orange), CD4 (red), CD20 (green). Scale bar, 200 µm. (E) The numbers of CXCL13+ CD160+ CD8+ T cells in patients with low (n=17) versus high (n=17) levels of TLSs were compared. (F) The density of CXCL13+ CD160+ CD8+ T cells were higher around mature TLSs. (G) Percentage of CXCL13+ CD160+ CD8+ T cells among CD160+ CD8+ T cells around TLSs at different stages of maturation. Data are presented as the mean±SD. ns, not significant. *p<0.05, ***p<0.001, two-tailed Student’s t-test. DAPI, 4’, 6-diamidino-2-phenylindole; TLSs, tertiary lymphoid structures.

Subsequently, we performed multiplex immunofluorescence experiments and found that TLSs colocalized with CXCL13+ CD160+ CD8+ T cells (figure 2A,B) and CXCL13+ PDCD1+ CD4+ T cells (figure 2C,D). Additionally, the numbers of CXCL13+ CD160+ CD8+ T cells were significantly higher in tissues with abundant TLSs (figure 2E). Furthermore, we observed a higher density of CXCL13+ CD160+ CD8+ T cells around mature TLSs (figure 2F). Aside from that, the ratios of CXCL13+ CD160+ CD8+ T cells to total CD160+ CD8+ T cells reached 70.6% around mature TLSs (figure 2G).

CXCL13+ CD160+ CD8+ T cells were strongly associated with clinical response to checkpoint immunotherapy

We then analyzed transcriptome sequencing data from 45 metastatic gastric cancer tissues obtained prior to the start of PD-1 immunotherapy.11 Our analysis unveiled a significant association between the extent of infiltration by CXCL13+ CD160+ CD8+ T cells and improved immunotherapy response (figure 3A). To further explore the correlation between CXCL13+ CD160+ CD8+ T cells and the response to immunotherapy, we performed multiplex immunofluorescence to quantify the intratumoral abundance of these cells in patients with different pathological responses. We found that responders (n=9) exhibited higher levels of CXCL13+ CD160+ CD8+ T cells in biopsy samples obtained prior to immunotherapy compared with non-responders (n=6) (figure 3B,C,G). Therefore, the number of CXCL13+ CD160+ CD8+ T cells may serve as an indicator of immunotherapy responsiveness. Moreover, there was a general increase in the number of intratumoral CXCL13+ CD160+ CD8+ T cells following immunotherapy, particularly in those with a PR to therapy (figure 3D,E,G,H). However, the number of intratumoral CXCL13+ CD160+ CD8+ T cells was decreased in patients with a CR (figure 3F, online supplemental figure S3A). This finding was consistent with the alterations observed in TLSs, suggesting a possible link to the absence of sustained antigen stimulation before detection.

Figure 3

CXCL13+ CD160+ CD8+ T cells were strongly associated with clinical response to checkpoint immunotherapy. (A) Transcriptome analysis revealed that the number of TLSs and CXCL13+ CD160+ CD8+ T cells were significantly higher in response (CR+PR) groups (n=12) compared with non-response (SD+PD) groups (n=33). (B) and (C) Enhanced CT scan of the abdomen before and after treatment with immunotherapy. The red arrows in the CT image indicate the tumor location. Multiplex immunofluorescence experiments were performed to assess the number of intratumoral CXCL13+ CD160+ CD8+ T cells in patients with a PR or SD response. Antibody panel: CXCL13 (pink), CD160 (orange), CD8 (red), CD20 (green), scale bar, 400 µm. (D–F) The Halo software was used to calculate the quantity of intratumoral CXCL13+ CD160+ CD8+ T cells both pre-immunotherapy and post-immunotherapy across patients exhibiting varied treatment responses. (G) Comparison of the number of CXCL13+ CD160+ CD8+ T cells between patients with CR or PR responses (n=9) and patients with SD or PD responses (n=6) prior to immunotherapy based on multiplex immunofluorescence staining. (H) Comparison of the number of infiltrating CXCL13+ CD160+ CD8+ T cells in patients with different responses following immunotherapy. Data are presented as the mean±SD. ns, not significant. *p<0.05, **p<0.01, ***p<0.001, two-tailed Student’s t-test. CR, complete response; PD, progressive disease; PR, partial response; SD, stable disease; ssGSEA, single-sample gene set enrichment analysis; TLSs, tertiary lymphoid structures.

Several studies have shown that CXCL13 could promote the recruitment of CXCR5+ B cells into tumor site, initiating the formation or maturation of TLSs. Analysis of the bulk RNA-seq from patients with gastric cancer in The Cancer Genome Atlas revealed a strong association between CXCL13+ CD160+ CD8+ T cells and B cells (figure 4A). Based on the integrated scRNA-seq data sets, we found CXCL13+ CD160+ CD8+ T cells were highly correlated to B cells and might communicate with B cells through CXCL13-CXCR5 signaling (figure 4B,C). Moreover, UMAP plots showed that CXCR5 was predominantly expressed in B cells (figure 4D). Interestingly, we found that there were more CXCR5+ B cells around TLSs in responders compared with non-responders after immunotherapy (figure 4E). This might be explained in part by the differential expression of CXCL13 in tissues. Additionally, we detected a higher abundance of CD38+ CD138+ CD27+ B cells (plasma cells) in responsive tissues (figure 4F). Furthermore, our results indicated that the amounts of IgG, IgM and IgA were quite different between patients with and without responsiveness (online supplemental figure S3B). Taken together, these findings imply that CXCL13+ CD160+ CD8+ T cells play a crucial role in influencing the formation and maturation of TLSs and are closely related to the efficacy of immunotherapy. However, what is the fine-tuning mechanism that affects CXCL13+ CD160+ CD8+ T-cell populations?

Figure 4

CXCL13+ CD160+ CD8+ T cells communicated with B cells through CXCL13-CXCR5 interactions. (A) Correlation heatmap between various immune subset enrichments using the bulk RNA-seq of gastric cancer samples from TCGA cohort. (B) Bubble heatmap showing cell–cell communication inference between T-cell clusters and B cells through ligand and receptor binding. (C) Analyzing the correlation between different types of immune cells based on the integrated scRNA-seq data sets. (D) UMAP plot displaying the expression of CXCL13 in T cells and the expression of CXCR5 among all cell types. (E) Multiplex immunofluorescence experiments were performed to detect the degree of CXCR5+ B-cell infiltration around TLSs in patients with different responses following immunotherapy. Antibody panel: CXCL13 (orange), CXCR5 (red), CD20 (green), Scale bar, 100 µm. The framed areas are shown below at a higher magnification. White triangles indicated CXCR5+ B cells. The number of infiltrating CXCR5+ B cells in patients with different responses was statistically analyzed. (F) Multiplex immunofluorescence images showing the markers for CD38 (green), CD138 (red), CD27 (white) and DAPI. Scale bar, 100 µm. The framed areas are shown below at a higher magnification. White triangle indicated CD38+ CD138+ CD27+ B cells; yellow triangle indicated CD38+ CD138− CD27+ B cells. The number of infiltrating CD38+ CD138+ CD27+ B cells and CD38+ CD138− CD27+ B cells in patients with different responses was statistically analyzed. Data are presented as the mean±SD. ns, not significant. *p<0.05, ***p<0.001, two-tailed Student’s t-test. DAPI, 4’, 6-diamidino-2-phenylindole; PR, partial response; SD, stable disease; scRNA-seq, single-cell RNA sequencing; TCGA, The Cancer Genome Atlas; UMAP, Uniform Manifold Approximation and Projection.

The activation of vitamin B6 metabolic pathway could promote the expression and secretion of CXCL13 in CD160+ CD8+ T cells

The diverse environments encountered by T cells influence their phenotype.24 CXCL13+ CD160+ CD8+ T cells might adjust their needs to the limited tumor microenvironments. Consequently, we performed multiple metabolic feature analysis using scMetabolism, an R package for single-cell analysis. Interestingly, the results showed that vitamin B6 metabolism, glutathione metabolism and oxidative phosphorylation might be crucial for CXCL13+ CD160+ CD8+ T cells (figure 5A). PDXK, GPX4 and NDUFA4 might be putative targets, respectively (figure 5B, online supplemental figure S4A,B). We subsequently investigated the impacts of these three metabolic pathways on the immune microenvironment by ex vivo experiments using PDTFs,25 which could significantly minimize immune cell loss during short-term cultures. We found that there was a significant decrease in the amount of CXCL13 released in the supernatants on the stimulation of PDXK inhibitors (ginkgotoxin) (figure 5C). Moreover, IHC results showed that high levels of PDXK expression in responsive gastric cancer tissues following immunotherapy (figure 5D). Then, the MFC murine gastric cancer cell line was used to establish subcutaneous xenograft tumor models in 615 mice. These mice were randomly assigned to four groups based on distinct treatment conditions (n=6, each group) (figure 5E). After 3 weeks, tumors were excised and measured. The results showed that the PDXK agonist (MS023) plus PD-1 inhibitors group had the smallest tumor volume and had improved survival rates, although the PDXK agonist group and the PDXK inhibitor group exhibited equivalent tumor volumes than the PBS group (figure 5F,G, online supplemental figure S4C–F). Meanwhile, the number of lymphocyte aggregates within tumors was significantly increased in PDXK agonist-treated mice (figure 5H,I).

Figure 5

Targeting PDXK could promote the formation of TLSs and enhance the efficacy of immunotherapy in gastric cancer. (A) Heatmap displaying the metabolic feature for T-cell clusters using scMetabolism package. (B) Dot plot showing the expression of the genes encoding rate-limiting enzymes of vitamin B6 metabolism in different types of T cells. Dot size encodes the percentage of cells expressing the gene, color encodes the average per cell gene expression level. (C) Quantification of CXCL13 in PDTFs in the presence of different enzyme inhibitors measured by ELISA. (D) Representative H&E staining and PDXK immunohistochemistry of gastric cancer tissues with different responses following immunotherapy. Scale bar, 500 µm. (E) The schematic diagram of the animal experiments. (F–G) Images of tumors and tumor volume c