Comparative impact of tertiary lymphoid structures and tumor-infiltrating lymphocytes in cholangiocarcinoma

The present study included patients who had received a histologically confirmed diagnosis of locally advanced or metastatic biliary tract cancer at any of the participating hospitals (assessed on the basis of the Response Evaluation Criteria in Solid Tumors (V.1.1)). Other clinical data were obtained from a relevant study.7 The patients received fixed-dose therapy (nivolumab, 200 mg; gemcitabine, 800 mg/m²; and S-1, 80/100/120 mg) on day 1 of a 2-week cycle. S-1 was continued until the occurrence of any adverse event or the need for dose reduction (as determined by the investigators). Treatment cycles were continued until the progression of the disease, the occurrence of intolerable toxic reactions, withdrawal of consent, or other reasons. Subsequent cycles were initiated only for patients satisfying specific criteria on day 1. Dose adjustment cannot be conducted for nivolumab. However, the doses of gemcitabine and S-1 can be reduced twice—to minimum doses of 400 mg/m² and 60 mg/day, respectively—without further escalation.

To study inflamed and non-inflamed tumors in patients with cholangiocarcinoma, we analyzed the RNA-Seq data of cholangiocarcinoma samples. The data were downloaded from The Cancer Genome Atlas (TCGA; https://portal.gdc.cancer.gov/↗) and the Gene Expression Omnibus (RRID:SCR_005012; https://www.ncbi.nlm.nih.gov/geo/↗) databases. A TCGA data set—TCGA-cholangiocarcinoma (TCGA-CHOL; n=36)—was used for discovery, whereas three GEO data sets, all of which consist of intrahepatic cholangiocarcinoma (ICC) cases, GSE119336↗ (n=15), GSE162396↗ (n=12), and GSE215997↗ (n=13), were used for validation. For the transcriptomic analysis of ICC, three GEO datasets (GSE119336↗, GSE162396↗, and GSE215997↗) were used. GSE119336↗ and GSE162396↗ provided RNA-Seq and gene expression data from ICC tumors and matched normal tissues. GSE215997↗ contained transcriptomic profiles of cholangiocarcinoma and adjacent normal tissues. Raw reads data (FASTQ files) were downloaded from GEO and processed using standard bioinformatic workflows, including quality control, alignment and normalization. We used microenvironmental cell population (MCP)-counter29 to quantify the abundance (MCP scores) of eight immune cells (T cells, CD8⁺ T cells, cytotoxic lymphocytes, natural killer (NK) cells, B-cell lineages, monocytic lineage, myeloid DCs, and neutrophils) and two stromal cells (endothelial cells and fibroblasts). Gene expression data (fragments per kilobase million values) were log2-transformed. MCP scores were calculated using 109 transcriptomic markers. Cholangiocarcinoma samples were stratified into inflamed and non-inflamed tumors based on MCP scores through hierarchical clustering (distance metric: Euclidean distance; linkage criterion: Ward method). On the basis of MCP score heatmaps and hierarchical clustering results, we defined samples with high and low MCP scores across cell types as inflamed and non-inflamed tumors, respectively. Furthermore, principal component analysis (PCA) was conducted on 109 transcriptomic marker genes expression to identify the predominant genes for classification. A pathologist (Y-CY) reviewed whole-slide images from the TCGA-CHOL data set to detect the presence of TLSs in inflamed and non-inflamed tumors.

The profiles and relevance of cholangiocarcinoma cell lines were assessed using data from the Cancer Cell Line Encyclopedia and Depmap portal website (https://depmap.org/portal/ccle/↗). Customized statistics for each site were evaluated to obtain significant data (p≤0.05).

To analyze the genetic differences in each sample, tumors were collected and performed the 440-gene panel ACTOnco from ACT Genomics, with sequencing carried out on the Ion Torrent platform by Thermo Fisher Scientific. The methodology followed was as previously outlined.7 All samples were sequenced with an average mean depth of ≥500×, ensuring that at least 75% of sequenced regions had coverage at ≥100×.

For immunohistochemical analysis, 4 μm thick formalin-fixed, paraffin-embedded cholangiocarcinoma tissue specimens were stained (overnight at 4°C) with primary antibodies against CD4, CD3, CD8, CD21, MS4A1 (CD20), and CD79A. Simultaneously, control specimens were incubated in diluent without primary antibody. After incubation, the slides were washed thrice (5 min/wash) with Tris-buffered saline containing 0.1% Tween 20 and then visualized using the Real Envision Detection System, Peroxidase/DAB+, Rabbit/Mouse (K500711; Dako). After washing, the slides were counterstained with hematoxylin and then analyzed through microscopy by a pathologist in a blinded manner.

Tissue samples from each patient with cholangiocarcinoma were sectioned and fixed in 10% formalin; this was followed by dehydration in graded alcohols and embedding in paraffin wax. Subsequently, 0.2 μm thick sections were sliced from the paraffinized blocks; the sections were deparaffinized through immersion in xylene and rehydrated. H&E was added to each slide, which was then rinsed with water. Next, each slide was dehydrated through immersion in graded alcohols and then in xylene (twice). Photomicrographs were obtained and interpreted by the pathologist (Y-CY). TLSs and TIL evaluation was performed in H&E-stained slides. TLS was defined as dense cellular lymphoid aggregates resembling germinal centers found in secondary lymphoid structures.30 TIL was defined as lymphocytes infiltrating tumor nests or located dispersed in the stroma between the carcinoma cells.31

Spatial transcriptomic data were generated using the GeoMx Digital Spatial Profiling (DSP) platform (NanoString Technologies), which enables the collection of spatially resolved gene expression data from formalin-fixed paraffin-embedded tissue sections. Tissue sections were stained with up to four visualization markers, and regions of interest (ROIs) were selected based on cell type-specific markers, allowing the profiling of tumor and microenvironment regions. To perform spatial analyses in the present study, we analyzed six cholangiocarcinoma samples derived from our study: three TLS-positive and three TLS-negative tumor tissue sections. Within each section, 12 ROIs were selected on the basis of visualization markers: CD19 (yellow) for B cells, CD3E (red) for T cells, KRT18 (green) for tumor cells, and DNA (blue) for nuclei. Each ROI only contained a single area of illumination (AOI) or segmentation, ensuring focused analysis of specific regions.

To define reference cell-type profiles, single-cell RNA sequencing (scRNA-seq) data from publicly available datasets were used. These datasets provided detailed profiles of immune and stromal cell populations, which were then curated to create a cell profile matrix. The matrix was constructed from a combination of flow-sorted peripheral blood mononuclear cells, scRNA-seq data from tumors, and RNA-seq data from flow-sorted stromal cells, collectively forming the SafeTME matrix. Deconvolution was performed using the SpatialDecon algorithm, which applies log-normal regression to spatial transcriptomic data to estimate the abundance of immune cell populations. The algorithm incorporates reference cell-type profiles from the SafeTME matrix to quantify the presence of 18 immune and stromal cell types within each ROI. The log-normal regression model corrects for variability in gene expression data, providing more accurate cell type estimates compared with classical least-squares methods. To ensure consistent and comparable gene expression measurements across samples, Q3 normalization was applied to the GeoMx DSP data. In this method, the 75th percentile (third quartile, Q3) of the signal from each ROI was calculated to account for technical variations across regions. This approach normalizes the data by adjusting each ROI’s gene expression values relative to its 75th percentile, thereby minimizing the impact of outliers or highly expressed genes. The Q3 normalized data were then used for downstream analyses, including differential gene expression and cell type deconvolution.

Tissue samples were profiled using the GeoMx Cancer Transcriptome Atlas (CTA) from NanoString Technologies, which provides a comprehensive spatial transcriptomic analysis across a variety of cancer types. The CTA enables the measurement of over 1800 genes specifically associated with key biological processes such as the TME, immune response, stromal interactions, and cancer progression. These genes cover a wide spectrum of pathways involved in tumor development, immune modulation, and therapeutic resistance.

Data are presented as mean±SD. Between-group differences were determined using the Mann-Whitney U test. The rates of progression-free survival and overall survival were calculated using the Kaplan-Meier method. Several clinicopathological factors were considered in the initial univariate analysis, which was performed using a log-rank test. Statistical analyses were performed using SPSS (V.17.0) for Windows (IBM SPSS).

To investigate whether TLSs can predict immunotherapy responses in patients with cholangiocarcinoma, the tumor samples derived from our clinical study—a single-arm, phase II trial investigating the potential of nivolumab plus modified gemcitabine and S-1 therapy as a first-line treatment modality. Given that TLS is organized with distinct T and B cell zones and follicular DCs,12 a total of 32 biopsy and 16 resection/surgery samples were analyzed by H&E staining to identify TLS. The numbers of TLS-positive and TLS-negative samples were 7 and 9, respectively (figure 1A). Compared with TLS-negative patients, TLS-positive patients exhibited significantly improved immunotherapy responses, which were characterized by substantial increases in both the objective response rate (complete response+partial response; 0% vs 71%, respectively) and disease control rate (complete response+partial response+stable disease; 67% vs 100%, respectively; figure 1B). Furthermore, the rates of progression-free survival and overall survival were higher in TLS-positive patients than in TLS-negative patients (p=0.03 for both; figure 1C,D). Interestingly, to investigate whether the formation of TLS is related to different oncogene mutations, we sorted out the most common frequency mutation genes in CCA (Cholangiocarcinoma) to compare in TLS-positive and TLS-negative patients. Our results observed that the formation of TLS is not strongly related to any different oncogene mutations (figure 1E). Collectively, our results support the potential of TLSs for predicting the outcomes of combined immunotherapy in patients with cholangiocarcinoma.

Cholangiocarcinoma tumors have traditionally been considered to exhibit low immune infiltration and limited responses to immune checkpoint inhibitors (ICIs).32 33 To study the complex TME, we analyzed the RNA-Seq data of 36 cholangiocarcinoma samples (derived from the TCGA-CHOL data set) by using MCP-counter.29 On the basis of MCP score heatmaps and hierarchical clustering results, the samples were classified as inflamed or non-inflamed tumors. Substantial immune infiltration was observed in inflamed tumors (figure 2A). Similar findings were obtained for the samples derived from the GSE119336↗, GSE162396↗, and GSE215997↗ data sets, all of which consist of ICC cases. These datasets include 15 samples from China and 25 samples from South Korea (12 and 13 samples, respectively). These findings indicate the presence of inflamed tumors in patients with cholangiocarcinoma (figure 2C).

To identify pivotal genes and immune cells within inflamed and non-inflamed tumors, we conducted a PCA using 109 transcriptomic markers derived from the MCP score calculations, enabling the comprehensive characterization of molecular and immune landscape differences between the two tumor subtypes. The analysis confirmed that the expression patterns of these 109 genes, represented by principal component 1 (PC1), effectively distinguished between inflamed and non-inflamed tumors (figure 2B), underscoring the potential of these gene signatures as reliable biomarkers for tumor classification. Thus, we ranked these genes on the basis of their weighting coefficients. The list of top 10 of 109 genes indicated the substantial upregulation of genes related to T cells and B-cell lineages, particularly a subset of B cells (table 1). Collectively, the results validated the presence of inflamed tumors in patients with cholangiocarcinoma. In various cancers, the tumor mutation burden serves as a key determinant of the tumor response to immunotherapy.34 35 The tumor mutation burden can also indicate the status of immune infiltration across tumor regions.34 Thus, using data from the TCGA-CHOL data set, we investigated the tumor mutation burden in inflamed and non-inflamed tumors (online supplemental figure 1). Surprisingly, no significant difference was noted in the tumor mutation burden between inflamed and non-inflamed tumors. Therefore, compared with other cancers, distinct approaches can be used for differentiating inflamed tumors from non-inflamed tumors in cholangiocarcinoma.

Cytotoxic T cells play crucial roles in immunotherapy by activating and regulating the immune system to fight pathogens or foreign antigens.36 In the context of immune responses, the effects of TLSs on the TME remain poorly understood, which prompted us to conduct the present study. We identified well-known TLS signatures from the literature.27 Analysis of the TCGA-CHOL and GEO datasets demonstrated a marked overexpression of TLS-associated gene signatures in inflamed tumors (figure 2C). To further explore this association, we reviewed whole-slide images from the TCGA-CHOL cohort and identified a significant correlation between the presence of TLSs and inflamed tumor phenotypes (figure 2D). Additionally, immunohistochemical analysis of cholangiocarcinoma samples from our clinical study confirmed TLS positivity, characterized by MS4A1+follicular B cells surrounded by CD3+T cells (figure 2E). In parallel, multiplex IHC conducted on the same patient’s specimens which coexistence of TLS and TIL, demonstrating that TLSs exhibited a higher density of B cells (CD20) and an increased expression of T cells (CD3, CD4, CD8) compared with TIL (online supplemental figure 2). Our findings suggest that TLSs exhibit features commonly associated with inflamed tumors in cholangiocarcinoma.

To further characterize the features of TILs and TLSs within the TME, we analyzed six cholangiocarcinoma samples from our study. Base on H&E staining, the samples were stratified into TLS-positive (n=3) and TLS-negative samples (n=3). Following this stratification, the samples were subjected to analysis using GeoMx DSP technology for a detailed molecular and spatial assessment. ROIs were defined using four fluorescent markers to precisely distinguish key components of the TME. Each ROI contained a single AOI, ensuring focused analysis of specific regions. KRT18 antibodies (green) were used to differentiate between tumorous and stromal compartments, whereas CD3E (red) and CD19 (yellow) antibodies were used to detect T- and B-cell markers, respectively. In addition, 4′,6-diamidino-2-phenylindole (DAPI) staining (blue) was performed to identify cell nuclei. The organization of tumors, TILs, and TLSs is depicted in figure 3A–C. In tumors with fewer CD19⁺ and CD3⁺ cells, diffuse CD19 and CD3 expressions were observed around tumor cells, indicating the presence of TILs. By contrast, TLSs exhibited the concentrated aggregation of CD19⁺ and CD3⁺ cells, which facilitated the identification of TLSs. Based on these immunological markers, we classified all ROIs (n=72) into tumors (n=30), TILs (n=30), and TLSs (n=12).

We analyzed the expression of 1834 genes, derived from the GeoMx CTA, a panel specifically designed for comprehensive profiling of tumor biology, the TME, and the immune response, to study differential gene signatures between tumors, TILs, and TLSs. The resulting heatmap revealed distinct gene expression profiles between these regions, with notable differences observed between tumors and both TILs and TLSs (figure 3D). To compare TIL and TLS signatures from the 1834 genes, we conducted an analysis of differentially expressed genes (DEGs), focusing on those with significant upregulation or downregulation (log-2 fold change >1; p<0.05; figure 3E). The top 10 differentially expressed (upregulated) genes between TLSs and TILs were related to B cells. Among these genes, MS4A1, CD79A, TNFRSF13C, and POU2AF1 are essential for the activation, maturation, and survival of B cells. Additionally, CR2, also known as CD21, is a receptor that is expressed on B cells and follicular DCs. CCL19 and CXCL13 are key chemokines that recruit lymphocytes into TLSs and facilitate the formation of TLS.37 38 Notably, TLSs exhibited the overexpression of MS4A1, CD79A, and CR2, as revealed by the analysis of the TCGA data (table 1). Comparative pathway analysis between TILs and TLSs revealed elevated levels of B-cell receptor signaling, complement activation, and lymphocyte regulation and trafficking, while cytotoxic activity was reduced in TLSs compared with TILs (figure 3F). Additionally, we were interested in comparing TILs and TLSs from the same samples in a paired analysis. We selected three targets comprizing TIL (n=4) and TLS (n=4) sites. The samples were pooled and normalized, resulting in a total of 12 TILs and 12 TLSs for comparison. After conducting statistical analysis, we identified 364 DEGs, which were used for hierarchical clustering and bioinformatic predictions. The hierarchical clustering heat map revealed unique expression patterns, from which we extracted the DEG signature (online supplemental figure 3). This signature predicted diverse signaling pathways, including those related to immune responses and oncogenesis. Our analysis identified that cytotoxic T lymphocytes and several cytokines were more activated in the TIL group compared with the TLS group. Conversely, B lymphocytes and B cell receptor signaling were suppressed in the TIL group, suggesting that the TLS region harbored a higher proportion of B cells, consistent with our previous hypothesis (online supplemental table 1). Additionally, we identified several transcription factors involved in the TME. Based on previous studies, TBX21 (T-bet), PRDM1, and TOX have been reported to be transcriptionally regulated in neoantigen-specific TILs in cancer.39 Conversely, transcription factors such as PAX5 and IKZF1, which are implicated in promoting TLS formation or reducing TILs, were also consistent with our predictions40 41 (online supplemental table 2). Among these findings, we highlight several aspects that merit further investigation.

To characterize the signatures of immune cell populations in TILs and TLSs, we estimated the properties of 18 types of immune cells, including T cells, B cells, NK cells, DCs, macrophages, neutrophils, mast cells, fibroblasts, and endothelial cells in each ROI (figure 4B) using SpatialDecon, which combines gene expression deconvolution with scRNA-seq data to estimate the spatial arrangement and abundance of immune cells in tissue samples and further validated against marker proteins to ensure accuracy.42 Our results revealed that TLSs exhibited higher properties of naive T cells, including CD4⁺ and CD8⁺ T cells, compared with TILs. In contrast, TILs had abundant CD8⁺ memory T cells and regulatory T cells. Notably, the predominant signature of T cells in TLSs were memory CD4⁺ T cells (figure 4C), which aligns with previous findings showing elevated proportions of CD4⁺ memory and naïve T cells in TLSs compared with the surrounding tumor tissue.43 We also observed elevated properties of B-cell lineages, including naive, memory, and plasma cells, with TLSs showing particularly high levels of memory B cells (figure 4D). Given that the primary function of DCs is to facilitate T-cell infiltration through the activation of naive T cells,21 44 it is noteworthy that myeloid DCs were more abundant in TLSs than in TILs (figure 4E). However, the properties of various innate immune cells, such as macrophages, NK cells, mast cells, and neutrophils, were lower in TLSs than in TILs (figure 4F–H). Collectively, our findings suggest that compared with TILs, TLSs exhibit elevated properties of naive T cells, B cells, and myeloid DCs but reduced properties of innate immune cells.

Due to the limited number of clinical samples, it was challenging to validate all proteins comprehensively. Therefore, we used RNA sequencing data from GeoDSP and applied statistical scoring based on the gene signatures of specific T-cell subtypes to assess the differences between TLSs and TILs (figure 5A, B,D), as well as between TILs in TLS-positive tumors and those in TLS-negative tumors (figure 5C). This gene expression-based scoring method effectively captures the characteristics and activity of different T-cell subtypes, compensating for the limitations of protein validation and providing reference data to further elucidate the functional roles of T cells in various TMEs. The signature of CTL, precursor-exhausted, and exhaustion are calculated by relative genes expression. The signature of CTL includes the z-score expression of GZMA, GZMB, TBX21, CX3CR1, GNLY, and PRF1.45,49 The signature of precursor-exhausted consists of the z-score expression of IL7R, LTB, CCR7 and TCF7.27 50 51 The signature of exhausted includes the z-score expression of EOMES, GZMK and CD27.52,56 Our results revealed that TLSs exhibited decreased expression of CTL-related genes and elevated expression of precursor-exhausted-related genes signature, implying that the T cells surrounding TLS exhibit high potential of mouldability instead of cytotoxicity (figure 5A,B). In the meantime, we also analyzed immune checkpoint molecules to examine the immunity function difference between T cells in TIL and TLS. Obviously, TILs exhibited significantly elevated levels of the following immune checkpoint molecules such as CTLA4, LAG3, TIM-3, OX40, and GITR (figure 5C). Additionally, we dissected the expression profiles of exhaustion-related genes that differed between TILs in TLS-positive and TLS-negative tumors. Interestingly, the signature of exhausted is higher in TIL in TLS-positive tumors than TIL in TLS-negative tumors, indicating that those precursor-exhausted T cells displayed in TLS have a high chance of being educated to prefer T cells with exhaustion characteristics (figure 5D). This evidence suggests that the T-cell subtypes within TILs of TLS-positive patients exhibit characteristics of exhausted T cells, facilitating a favorable immune environment that could enhance cancer immunity and be more responsive to immunotherapy.

To further explore the pathways and regulatory elements potentially implicated in TME activation, we performed a detailed comparative analysis of tumors and TILs, both in the presence and absence of TLSs. DEGs were analyzed using Ingenuity Pathway Analysis to identify relevant molecular pathways and regulators. Given our focus on understanding the mechanisms by which TLSs mediate enhanced responses to immunotherapy, we specifically examined the regulators of PD-L1 expressions (). Our analysis revealed activation of the HIF transcription factor family in TILs and tumors associated with TLSs. We hypothesize that HIF-2 may regulate PD-L1 expression, potentially facilitating immune evasion mediated by TLSs in the TME. online supplemental table 3

To validate the findings from our transcriptome analyses, we analyzed real-world clinical data from the T1219 clinical trial (figure 5E). Tumor proportion scores (TPS), which were calculated in terms of PD-L1 expression levels in tumors, did not differ between TLS-positive patients and TLS-negative patients. However, the proportion of patients with a combined positive score (CPS) of >1 was significantly higher in the TLS-positive group than in the TLS-negative group (43% vs 11%, respectively; p=0.007). Similar trends were obtained through immunohistochemical staining for LAG-3. These results suggest that the TME in TLS-positive patients is conducive to favorable immunotherapy responses, with PD-L1 being expressed primarily in the TME rather than on the tumor cells themselves.

Data are available on reasonable request.