Stress-driven reprogramming of plasmacytoid dendritic cells in intrahepatic cholangiocarcinoma defines a reversible targetable immunosuppressive state Authors: Mei-Ru Chen, Xiao-Li Xie, Yan-Li Zhou, Jin-Mei Tian, Ying Zhao, Suping Hou, Yuan-Yuan Wang, Ji-Chao Lu, Dong-Qiang Zhao, Zhao-Bo Cui, Li-Xian Zhang Published: 2026 ABSTRACT BackgroundPlasmacytoid dendritic cells (pDCs) have been implicated in both restraining and promoting intrahepatic cholangiocarcinoma (iCCA), leaving their clinical relevance and therapeutic potential unresolved.MethodsMendelian randomization was used to assess the causal association between circulating pDC levels and iCCA risk. Bulk and single-cell transcriptomic analyses were performed to characterize pDC-related programs and tumor-conditioned states, and multiplex immunofluorescence was used to define spatial distribution and clinical associations in iCCA tissues. To assess reversibility of stress-associated pDC features, IRE1α RNase activity was pharmacologically inhibited with 4µ8C under tumor-conditioned stress in vitro.ResultsGenetically predicted higher circulating pDC levels were associated with lower iCCA risk, consistent with a systemic protective association. In bulk cohorts, higher expression of pDC markers (CLEC4C, NRP1, IL3RA) was associated with an immune-inflamed microenvironment and improved survival in early-stage disease. Single-cell analyses indicated that intratumoral pDCs acquired stress-associated transcriptional programs, including enrichment of endoplasmic reticulum stress and unfolded protein response pathways. In vitro, 4µ8C reduced IRE1α-dependent XBP1 splicing and partially restored type I interferon-linked activation and pDC immunogenic readouts under tumor-conditioned stress. Spatial profiling further showed that higher intratumoral CD303⁺IRF7⁺ pDC activation was associated with advanced stage and poorer overall survival, whereas higher activation in adjacent non-tumor tissues correlated with more favorable outcomes.ConclusionsTogether, these findings support a context-dependent, stress-associated pDC program in iCCA and provide a rationale for further evaluating the IRE1α-XBP1 stress axis as a potential approach to modulate pDC-associated immune states within the tumor microenvironment.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12957-026-04489-y. Introduction Intrahepatic cholangiocarcinoma (iCCA) is a malignant tumor originating from the biliary epithelium, with a globally increasing incidence [1]. Patients are often diagnosed at advanced stages, when surgical resection is frequently no longer feasible [2]. Moreover, the disease responds poorly to conventional radiotherapy and chemotherapy, leading to an exceptionally dismal prognosis and a five-year survival rate of less than 20% [3]. This grim prognosis highlights the urgent need to elucidate the immunological mechanisms underlying iCCA progression and to identify new therapeutic entry points within the tumor microenvironment (TME). In recent years, the role of the tumor immune microenvironment (TIME) in the pathogenesis and progression of CCA has garnered increasing interest. The TME constitutes a complex ecosystem wherein the composition, function, and spatial distribution of immune cells significantly influence tumor advancement and treatment response [4, 5]. Among these immune components, plasmacytoid dendritic cells (pDCs) have emerged as multifunctional innate immune players with dual roles in tumor immunity [6]. Recognized as professional type I interferon (IFN-I) producing cells, pDCs are crucial mediators of antiviral responses and have the capacity to initiate antitumor immunity [7]. However, accumulating evidence indicates that pDCs exhibit functional plasticity in cancer: in some contexts, they effectively activate T cell mediated antitumor immunity, while in others, they may promote immune tolerance and tumor growth [8]. pDCs demonstrate considerable functional heterogeneity across human cancers, with their prognostic impact varying by tumor type and immunological context. In malignancies such as breast [9], ovarian [10], and oral squamous cell carcinoma [11], pDC infiltration is often linked to immunosuppression and poor clinical outcomes. Conversely, in settings like follicular lymphoma [12] or Epstein-Barr virus associated nasopharyngeal carcinoma [13], pDC presence correlates with better survival, suggesting antitumor functions. This context-dependent behavior underscores the influence of local microenvironmental cues on pDC function. Notably, the profoundly fibrotic and metabolically stressed iCCA microenvironment may impose transcriptional reprogramming on infiltrating pDCs, potentially shifting them from an immune-activating to a tolerogenic state, a hopothesis that remains largely unexplored. In this study, we hypothesize that the iCCA TME reprograms pDCs from a systemic immune-protective state toward a locally immunosuppressive, tumor-associated phenotype. To test this, we integrated Mendelian randomization (MR) to assess causal associations between circulating pDC levels and iCCA risk, bulk and single-cell transcriptomic profiling to delineate molecular alterations, and multiplex immunofluorescence (mIF) to define spatial and clinical relevance. Furthermore, we examined whether pharmacologic inhibition of the IRE1α-XBP1 stress pathway could reverse this dysfunctional state and restore pDC immunogenicity, thereby providing mechanistic insight and a conceptual basis for therapeutic reprogramming of pDC function in iCCA. Methods Mendelian randomization assumptions and design MR leverages single-nucleotide polymorphisms (SNPs) as instrumental variables (IVs) to infer causal relationships, minimizing confounding [14]. This two-sample MR study evaluated the causal effects of immune cells on iCCA risk, adhering to three core assumptions: (1) genetic variants strongly associate with the exposure; (2) variants are independent of confounders; and (3) variants influence the outcome exclusively through the exposure [15]. These assumptions are illustrated in Fig. 1. Fig. 1Three main assumptions of Mendelian randomization. (By Figdraw) Three main assumptions of Mendelian randomization. (By Figdraw) Genome-wide association study (GWAS) data sources Summary statistics for iCCA were obtained from a GWAS of 466,348 European-ancestry individuals in the UK Biobank, using ICD10 code C22.1 [16]. For immune traits, publicly available summary statistics from the GWAS Catalog (accessions GCST90001391 to GCST90002121; https://www.ebi.ac.uk/gwas/ ) were used, encompassing 731 immunophenotypes: absolute cell counts (n = 118), median fluorescence intensities (n = 389), morphological parameters (n = 32), and relative cell counts (n = 192) [17]. The original GWAS involved 3,757 non-overlapping European individuals genotyped for approximately 22 million SNPs using the Sardinian sequence-based reference panel, with associations tested after covariate adjustment [18]. All data were publicly accessible, and informed consent was obtained in original studies; no additional ethical approval was required. Selection of instrumental variables SNPs associated with exposures were selected at a genome-wide significance threshold of P < 1 × 10⁻⁵ and minor allele frequency > 0.01 [19]. Linkage disequilibrium was pruned (r² < 0.01, window > 10,000 kb), and palindromic SNPs were excluded. Instrument strength was assessed via F-statistics (F = beta²/se²), excluding weak instruments (F < 10). For reverse MR, IVs associated with iCCA were selected at P < 1 × 10⁻⁵. Bulk transcriptomic data analysis RNA-sequencing and clinical data for cholangiocarcinoma were downloaded from The Cancer Genome Atlas (TCGA; https://portal.gdc.cancer.gov/ ), including 36 tumor and 9 adjacent normal samples. Validation used GSE107943 from the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/ ), comprising 30 iCCA tumors with annotations. Immune infiltration was further assessed via ssGSEA for signature gene set enrichment. Analyses were performed in R with ggplot2 for visualization. Scores were compared using the Wilcoxon rank-sum test (P < 0.05). Differential expression of pDC markers (CLEC4C, NRP1, IL3RA) employed edgeR (version 3.32.1), with log2 fold-change > 0.25 and adjusted P < 0.05. Spearman correlations with immune infiltration scores were visualized using corrplot. Clinical correlations stratified samples by stage, differentiation, and prognostic classes (class A: favorable; class B: poor) from GSE107943 [20], using Wilcoxon rank-sum or t-tests. Separately, to quantify the activated pDC program in bulk cohorts, an activated pDC ssGSEA score was calculated using a scRNA-seq–derived activated pDC signature (CLEC4C, IL3RA, GZMB, IRF7, OAS1, ISG20, MX1). Fibroblast abundance in TCGA-CHOL tumors was inferred from bulk RNA-seq using MCP-counter (Fibroblasts score) as a CAF surrogate. Activated CD8 T-cell scores were obtained by ssGSEA. Tumors were classified into four CD8–CAF states by median splits of activated CD8 and fibroblast scores, and Activated pDC ssGSEA scores were compared across groups using Kruskal-Wallis with BH-adjusted pairwise Wilcoxon tests. Tumor purity was estimated by ESTIMATE and used for purity-adjusted analyses (partial Spearman correlation and multivariable regression). Single-cell RNA sequencing data analysis Data from GSE138709 were processed using Seurat (version 4.0.5) in R. Normalization, scaling, and dimensionality reduction via principal component analysis were performed, followed by clustering with the Louvain algorithm (resolution 0.8), which yielded 23 clusters. These clusters were visualized using uniform manifold approximation and projection (UMAP). DC subsets were identified using markers: pDCs (GZMB, LILRA4, IL3RA), cDC1 (CLEC9A, FLT3, IDO1), and cDC2 (CD1C, FCER1A, HLA-DQA1). Extracted DCs were re-normalized, subjected to PCA, and subclustered (resolution 0.8) into 7 populations. Differential expression used FindAllMarkers (min.pct = 0.3, logfc.threshold = 0.25), with top 10 markers per subcluster visualized by heatmap. pDCs (cluster 0) were confirmed via specific markers (CLEC4C, NRP1, IL3RA for pDCs; THBD, XCR1, CLEC9A, CADM1, IRF8 for cDC1; CD1C, ITGAM, SIRPA, CD209, CD47 for cDC2). Tumor versus normal pDC differential expression employed FindMarkers (min.pct = 0.25, logfc.threshold = 0.25; adjusted P < 0.05, |log2FC| > 0.25). Upregulated genes underwent protein-protein interaction (PPI) network construction in STRING, imported to Cytoscape (version 3.10.1), and modular analysis via MCODE. Hub genes were enriched for biological processes using WEBGestalt ( https://www.webgestalt.org/option.php, adjusted P < 0.05). Construction and specificity assessment of an activated pDC signature: Canonical pDC markers (CLEC4C/CD303, IL3RA, GZMB, and IRF7) were combined with pDC-enriched interferon-stimulated genes (ISGs). A predefined ISG panel (IFI27, ISG20, RSAD2, IFIT3, MX1, OAS1, ISG15, IFIT1, IFITM3, XAF1, and IFI6) was ranked by mean expression in pDC cells, and the top candidates were visualized together with pDC markers. To evaluate specificity, average expression was calculated across annotated cell types (pDC, cDC1, cDC2, and others) using Seurat’s AverageExpression and displayed as a heatmap. The final activated pDC signature comprised seven genes: CLEC4C, IL3RA, GZMB, IRF7, OAS1, ISG20, and MX1. Patient cohort and tissue samples Surgical specimens were collected from a total of 48 iCCA patients, including 14 patients treated at Hengshui People’s Hospital and 24 patients who underwent radical surgery at the Second Hospital of Hebei Medical University between 2018 and 2024. All cases provided paired tumor and adjacent normal tissue samples, along with complete follow-up data. Inclusion: histologically confirmed iCCA; exclusion: prior neoadjuvant therapy, concurrent malignancies, insufficient tissue. Clinicopathological data encompassed TNM stage (AJCC 8th), lymph node metastasis, differentiation, and OS. The study received approval from the Research Ethics Committee of the Second Hospital of Hebei Medical University and Hengshui People’s Hospital for the research involving human samples. Written informed consent was obtained from all patients. Isolation and culture of human pDCs Peripheral blood was obtained from healthy adult volunteers recruited at the Second Hospital of Hebei Medical University. The study received approval from the Research Ethics Committee of the Second Hospital of Hebei Medical University (Approval No. 2024-R474). Written informed consent was obtained from all participants before enrollment. Peripheral blood mononuclear cells (PBMCs) were isolated by density-gradient centrifugation. Purified pDCs were obtained from PBMCs using the Plasmacytoid Dendritic Cell Isolation Kit II (Miltenyi Biotec, Auburn, CA) according to the manufacturer’s protocol, achieving > 95% purity. Cells were resuspended at RPMI-1640 supplemented with 10% FBS and gentamicin (10 µg/mL), seeded into 96-well plates, and cultured at 37 °C in a humidified 5% CO₂ incubator for 24–48 h. To simulate the iCCA microenvironment, pDCs were cultured with 30% tumor-conditioned medium (supernatant from HuCC-T1 cultures, filtered through 0.22 μm), collectively referred to as the “TME condition”. Pharmacologic modulation of the IRE1α-XBP1 pathway To assess the reversibility of ER-stress-induced dysfunction, pDCs were treated with the IRE1α RNase inhibitor 4µ8C (20 µM; DMSO < 0.1%) alone or in combination with the TLR9 agonist CpG-ODN 2216 3 µg/mL (InvivoGen). Control cells received equal volumes of vehicle. After 4 h of treatment, cells were harvested for XBP1 splicing analysis; after 24 h, supernatants and cell lysates were collected for cytokine and gene expression assays. Measurement of IFN-α secretion Supernatants from pDC cultures (24 h post-treatment) were analyzed for human IFN-α using a commercial ELISA kit (Thermo Fisher Scientific) according to the manufacturer’s protocol. Absorbance was read at 450 nm with a 620 nm reference. Concentrations were derived from a standard curve and normalized to viable cell counts when necessary. Quantification of mRNA Total RNA was extracted from pDCs using the RNeasy Mini Kit (Qiagen) and reverse-transcribed with a cDNA synthesis kit (Takara). Expression of genes was measured by SYBR Green–based qPCR (Bio-Rad CFX Connect) with GAPDH as the internal control. Relative expression levels were calculated using the 2⁻ΔΔCt method compared with the control. Co-culture assay for T-cell activation To evaluate the functional recovery of pDC immunogenicity, pretreated pDCs were co-cultured with autologous CD8⁺ T cells (purified by negative selection; Miltenyi Biotec) at a 1:10 ratio in the presence of 10 ng/mL OKT3 (Bio-Express). Supernatants collected at 48 h were assayed for IFN-γ and Granzyme B using human ELISA kits (Thermo Fisher Scientific) following the manufacturer’s instructions. Cytokine concentrations were expressed as pg/mL or fold change versus TME. Cytotoxicity assay Tumor cell killing was quantified using a lactate dehydrogenase (LDH) release assay (Nanjing Jiancheng Bioengineering Institute) according to the manufacturer’s instructions. HuCC-T1 target cells were seeded in 96-well plates and co-incubated with mixed effector cells composed of pretreated pDCs and autologous CD8⁺ T cells at a fixed internal ratio of 1:10 (pDC: T cell). The overall effector-to-target (E: T) ratio was 5:1, and co-culture was maintained for 6 h before LDH quantification. LDH release was measured at 450 nm. Multiple immunofluorescence Tissue sections were first subjected to endogenous peroxidase inactivation, antigen retrieval, and blocking with BSA. Slides were incubated at 4 °C overnight with primary antibodies against CD303/BDCA2 (Abcam; ab239078; 1:1000) and IRF7 (Abways; CY5645; 1:1000). After washing, HRP-linked secondary antibodies were applied. Signal amplification was performed using a TSA-based system (TSA Kit, SB-Y6112, ShareBio), followed by antibody stripping with elution buffer at 37 °C. Sections were coverslipped with anti-fade mounting medium. Images were acquired using a laser scanning confocal microscope, and quantification was performed in ImageJ. To characterize the spatial distribution of pDCs, regions of interest (ROIs) were predefined as (i) tumor core (intratumoral area distant from the tumor–stroma interface), (ii) invasive margin (the tumor–stroma interface and adjacent infiltrative front), and (iii) peritumoral/adjacent normal tissue. For each specimen, three non-overlapping representative ROIs were analyzed per region under identical acquisition settings. pDC density was quantified as the number of CD303⁺ cells per mm², and the activation ratio was calculated as (CD303⁺IRF7⁺ cells / total CD303⁺ cells) × 100%. All evaluations were performed in a blinded manner. Statistical analysis MR analyses used TwoSampleMR in R, including inverse-variance weighted (IVW; primary method), weighted median, weighted mode, simple mode, and MR-Egger [21]. Horizontal pleiotropy was assessed via MR-PRESSO (global, outlier, distortion tests), with outliers removed and re-analysis. Heterogeneity and pleiotropy were evaluated using Cochran’s Q and MR-Egger intercept tests; MR-Egger was prioritized if pleiotropy was detected. Scatter and funnel plots visualized the results. Suggestive SNPs (P < 10⁻⁵) were identified via Phenoscanner V2 ( http://www.phenoscanner.medschl.cam.ac.uk/ ) and excluded. Data normality used the Shapiro-Wilk; comparisons employed t-tests or Wilcoxon tests; correlations via Spearman or Kruskal-Wallis; survival via Kaplan-Meier and log-rank. Benjamini-Hochberg adjusted multiple comparisons. For spatial analyses, CD303⁺ density and activation ratio were compared across normal, invasive margin, and tumor core using paired Friedman tests with BH-adjusted post hoc Wilcoxon tests. P < 0.05 was significant. Results Causal estimates between the 731 immune cell traits and iCCA Following rigorous IV selection and harmonization, 731 SNPs associated with immune cell traits were included in the MR analysis. All selected SNPs possessed F-statistics exceeding 10, confirming their strength as IVs. Using the IVW method as the primary approach, two-sample MR was performed to assess the causal effects of immunophenotypes on iCCA risk. Initial analysis identified 593 SNPs significantly linked to iCCA across 27 immune cell phenotypes. Subsequent screening using Phenoscanner revealed that 166 of these SNPs were associated with 320 non-immunological traits (Supplementary Table S2). To minimize confounding, 26 SNPs associated with 27 known risk factors for iCCA [22, 23], including primary sclerosing cholangitis, hepatitis B/C, metabolic conditions, and others, were excluded (Supplementary Table S3). After this adjustment, re-analysis yielded 501 SNPs significantly associated with iCCA across 23 immune cell phenotypes, encompassing dendritic cell, regulatory T cell, myeloid cell, TBNK, T cell maturation, monocyte, and B cell panels. A significance threshold of P < 0.05 was applied. Notably, 14 immune cell phenotypes demonstrated a protective effect against iCCA (Fig. 2; Supplementary Table 4). These included plasmacytoid dendritic cell (% dendritic cells), effector memory CD8⁺ T cells, CD8⁺ NKT cells, and several surface markers across T, B, and myeloid lineages. Conversely, 9 phenotypes were identified as risk factors, predominantly within Treg and TBNK panels. Fig. 2Forest plot of the association between immune cells and intrahepatic cholangiocarcinoma Forest plot of the association between immune cells and intrahepatic cholangiocarcinoma After false discovery rate (FDR) correction, one phenotype remained significantly protective: plasmacytoid dendritic cell (% dendritic cell) (OR = 0.59, 95% CI: 0.45–0.79, P = 0.0003, P-FDR = 0.039). Sensitivity analyses, including scatter plots, funnel plots, MR-Egger intercept, and Cochran’s Q tests, supported the robustness and consistency of these findings, with no evidence of pleiotropy or heterogeneity (Supplementary Figures S1-S6; Supplementary Table S3). Reverse MR analysis was conducted to evaluate potential reverse causation between iCCA and the 23 immune phenotypes. A nominally significant association was detected only for CD33⁺ HLA DR⁺ CD14dim (% CD33⁺ HLA DR⁺) using the MR-Egger method (OR = 1.06, 95% CI: 1.01–1.11, P = 0.024). No evidence of multicollinearity was observed (Supplementary Table S5). Together, these results strongly support a protective causal role of plasmacytoid dendritic cells in iCCA development, providing a genetic foundation for further investigation into their function in the iCCA microenvironment. Analysis of bulk transcriptomic data (TCGA and GEO) Analysis of the TCGA dataset revealed a significantly elevated infiltration level of pDCs in tumor tissues compared to normal adjacent tissues (P < 0.05, Fig. 3A). Furthermore, Spearman correlation analysis demonstrated significant positive correlations between key pDC-related markers (CLEC4C, NRP1, and IL3RA) and the infiltration scores of multiple immune cell types, including CD4⁺ T cells, CD8⁺ T cells, B cells, and dendritic cells (Fig. 3B), suggesting a broad coordination between pDC abundance and overall immune infiltration in the TME. Fig. 3Bulk transcriptomic analyses reveal the association of pDC abundance and key markers with immune context, tumor stage, and prognosis in iCCA. A pDC infiltration scores in tumor tissues compared to adjacent normal tissues from the TCGA dataset. B Spearman correlation analysis between expression of pDC-related markers (CLEC4C, NRP1, IL3RA) and infiltration scores of various immune cell types in the TCGA cohort. C, D Association of pDC infiltration score (C) and NRP1 expression (D) with tumor stage in the TCGA cohort. E, F Validation in the GEO dataset (GSE107943): Association of CLEC4C expression with tumor stage (E) and IL3RA expression with tumor differentiation grade (F). G NRP1 expression levels in patient subgroups with pre-established favorable (Class A) or poor (Class B) prognosis. Statistical significance is indicated as *P < 0.05, **P < 0.01, ***P < 0.001. TCGA, The Cancer Genome Atlas; GEO, Gene Expression Omnibus Bulk transcriptomic analyses reveal the association of pDC abundance and key markers with immune context, tumor stage, and prognosis in iCCA. A pDC infiltration scores in tumor tissues compared to adjacent normal tissues from the TCGA dataset. B Spearman correlation analysis between expression of pDC-related markers (CLEC4C, NRP1, IL3RA) and infiltration scores of various immune cell types in the TCGA cohort. C, D Association of pDC infiltration score (C) and NRP1 expression (D) with tumor stage in the TCGA cohort. E, F Validation in the GEO dataset (GSE107943): Association of CLEC4C expression with tumor stage (E) and IL3RA expression with tumor differentiation grade (F). G NRP1 expression levels in patient subgroups with pre-established favorable (Class A) or poor (Class B) prognosis. Statistical significance is indicated as *P < 0.05, **P < 0.01, ***P < 0.001. TCGA, The Cancer Genome Atlas; GEO, Gene Expression Omnibus We further evaluated the clinical relevance of pDC-related metrics. In the TCGA cohort, pDC infiltration scores were inversely correlated with tumor stage, with higher stages associated with reduced pDC abundance (P = 0.023, Fig. 3C). Similarly, expression of NRP1 was significantly lower in advanced-stage (III-IV) compared to early-stage (I-II) tumors (P = 0.012, Fig. 3D). These trends were corroborated in the independent GSE107943 dataset, where lower expression of CLEC4C was associated with advanced tumor stage (P = 0.041, Fig. 3E), and reduced IL3RA expression correlated with poorer tumor differentiation (P = 0.034, Fig. 3F). Notably, NRP1 expression was also linked to clinical outcome. Based on pre-established prognostic classifications (class A: favorable prognosis; class B: poor prognosis) [20], we found that NRP1 expression was significantly elevated in the favorable prognosis group (P = 0.00044, Fig. 3G), supporting its potential role as a positive prognostic marker. Collectively, these results underscore the significance of pDCs and their molecular signatures in iCCA, indicating their involvement in immune regulation and clinical progression, and highlighting their potential as therapeutic targets. Single-cell transcriptomic profiling reveals stress-adapted reprogramming of pDCs in iCCA Unsupervised clustering of scRNA-seq data from the GSE138709 dataset categorized all cells into 23 distinct clusters. Dendritic cell (DC) populations were identified based on established marker genes: plasmacytoid DCs (pDCs; GZMB, LILRA4, IL3RA), cDC1s (CLEC9A, FLT3, IDO1), and cDC2s (CD1C, FCER1A, HLA-DQA1), which formed clearly segregated clusters on UMAP visualization (Fig. 4A). The DC compartment was subsequently re-clustered at a resolution of 0.8, yielding seven transcriptionally distinct subpopulations (Fig. 4B). Among them, cluster 0 was confidently annotated as pDCs due to high expression of canonical markers (CLEC4C, NRP1, and IL3RA) and its discrete localization within UMAP space (Fig. 4C). Fig. 4Single-cell transcriptomic profiling identifies a transcriptionally reprogrammed pDC population in iCCA. A UMAP visualization of major dendritic cell (DC) subsets (pDCs, cDC1s, cDC2s) identified from the GSE138709 dataset. B Heatmap displaying the top 10 marker genes for each of the 7 subclusters identified after re-clustering DCs. C UMAP plot highlighting the distinct localization of cluster 0, annotated as pDCs based on high expression of canonical markers (CLEC4C, NRP1, IL3RA). D Bar plot comparing the proportion of pDCs (cluster 0) in tumor versus normal tissue samples. E Volcano plot showing differentially expressed genes in tumor-associated pDCs compared to normal pDCs. F Protein-protein interaction network of a core module of hub significantly genes in tumor pDCs. G Functional enrichment analysis of the core gene module from (F), showing significantly associated biological processes Single-cell transcriptomic profiling identifies a transcriptionally reprogrammed pDC population in iCCA. A UMAP visualization of major dendritic cell (DC) subsets (pDCs, cDC1s, cDC2s) identified from the GSE138709 dataset. B Heatmap displaying the top 10 marker genes for each of the 7 subclusters identified after re-clustering DCs. C UMAP plot highlighting the distinct localization of cluster 0, annotated as pDCs based on high expression of canonical markers (CLEC4C, NRP1, IL3RA). D Bar plot comparing the proportion of pDCs (cluster 0) in tumor versus normal tissue samples. E Volcano plot showing differentially expressed genes in tumor-associated pDCs compared to normal pDCs. F Protein-protein interaction network of a core module of hub significantly genes in tumor pDCs. G Functional enrichment analysis of the core gene module from (F), showing significantly associated biological processes Proportional analysis revealed a significant expansion of pDCs (cluster 0) in tumor samples compared with normal tissues (Fig. 4D). Differential gene expression analysis identified 30 upregulated genes in tumor-derived pDCs (Fig. 4E), and protein-protein interaction network mapping highlighted a tightly connected core module comprising 16 hub genes, including immediate-early transcription factors such as FOS, JUN, JUNB, FOSB, EGR1, EGR3, ATF3, IER2, NR4A1, and NR4A3, together with ZFP36, GADD45B, DUSP1, BTG2, CSRNP1, and CXCL8 (Fig. 4F). This transcriptional program is consistent with a stress-adaptive response of pDCs within the TME. Functional enrichment analysis of this core gene module revealed predominant activation of integrated stress response signaling, together with pathways related to RNA metabolism and responses to fibroblast and steroid hormones (Fig. 4G). Notably, several hub genes in this module, including ATF3, DUSP1, GADD45B, FOS, JUN, and NR4A family members, are well-recognized components of the endoplasmic reticulum (ER) stress-responsive transcriptional network, regulated downstream of the IRE1α-XBP1 branches of the unfolded protein response (UPR) [24–27]. Collectively, these findings suggest that tumor-associated pDCs in iCCA experience chronic ER stress and activate the IRE1α-XBP1 arm of the UPR, adopting a stress-adapted transcriptional state that may underlie their functional reprogramming within the TME. Because several genes in this program are interferon-stimulated and could be expressed across immune lineages, we further verified cell-type specificity in the scRNA-seq data. An average-expression heatmap of the top 5 ranked ISG candidates, together with canonical pDC markers, across pDCs, cDC1, cDC2, and other cells showed a collective enrichment in pDCs (Fig. S7A). Pharmacologic inhibition of the IRE1α-XBP1 pathway restores the IFN-I response and reactivates pDC immunogenicity To evaluate whether the stress-adapted state of tumor-conditioned pDCs is reversible, we pharmacologically inhibited IRE1α RNase activity using 4µ8C. Within 4 h of treatment under iCCA-like stress conditions, the ratio of spliced to total XBP1 (sXBP1/total XBP1) significantly decreased (P < 0.001), confirming effective blockade of IRE1α-dependent XBP1 splicing (Fig. 5A). Similarly, DNAJB9, a downstream target of XBP1s, was significantly upregulated in the TME group and reduced upon 4µ8C treatment, indicating the reversal of stress-induced ER-stress activation (Fig. 5B). Fig. 5Pharmacologic inhibition of the IRE1α–XBP1 pathway restores type I IFN signaling and pDC immunogenicity. A Inhibition of IRE1α RNase activity by 4µ8C reduced XBP1 splicing, as indicated by the decreased sXBP1/total XBP1 ratio. ***P < 0.001. B DNAJB9 expression (an XBP1s target) was increased under TME conditions and reduced by 4µ8C treatment. C-D pDCs cultured under tumor-conditioned medium (TCM)/TME conditions were treated for 24 h with vehicle, 4µ8C, CpG-ODN 2216, or both. IFN-α secretion (C) and MX1 expression (D) were enhanced by either treatment, with the combination producing the strongest response. E Co-culture of pretreated pDCs with autologous CD8⁺ T cells showed increased IFN-γ and Granzyme B production after 4µ8C or CpG-ODN 2216 stimulation, with the combination yielding the highest response. F Cytotoxic activity against HuCC-T1 cells was assessed by LDH release assay. Data represent mean ± SEM of three independent donors. Groups sharing the same letter are not significantly different (P > 0.05) Pharmacologic inhibition of the IRE1α–XBP1 pathway restores type I IFN signaling and pDC immunogenicity. A Inhibition of IRE1α RNase activity by 4µ8C reduced XBP1 splicing, as indicated by the decreased sXBP1/total XBP1 ratio. ***P < 0.001. B DNAJB9 expression (an XBP1s target) was increased under TME conditions and reduced by 4µ8C treatment. C-D pDCs cultured under tumor-conditioned medium (TCM)/TME conditions were treated for 24 h with vehicle, 4µ8C, CpG-ODN 2216, or both. IFN-α secretion (C) and MX1 expression (D) were enhanced by either treatment, with the combination producing the strongest response. E Co-culture of pretreated pDCs with autologous CD8⁺ T cells showed increased IFN-γ and Granzyme B production after 4µ8C or CpG-ODN 2216 stimulation, with the combination yielding the highest response. F Cytotoxic activity against HuCC-T1 cells was assessed by LDH release assay. Data represent mean ± SEM of three independent donors. Groups sharing the same letter are not significantly different (P > 0.05) After 24 hours, 4µ8C markedly increased IFN-α secretion (P < 0.05) and upregulated MX1 expression (P < 0.05), indicating restoration of type-I interferon signaling (Fig. 5C-D). To evaluate whether pharmacologic stress relief could cooperate with innate activation, pDCs were stimulated with the TLR9 agonist CpG-ODN 2216, which triggers IFN-I production through endosomal sensing. CpG stimulation alone produced a comparable enhancement, whereas the combination of 4µ8C and CpG further amplified both IFN-α and MX1 responses, significantly exceeding either single treatment (P < 0.05). These data suggest that pharmacologic inhibition of the IRE1α-XBP1 branch relieves stress-mediated repression and reactivates the IFN-I responsive program in pDCs. To assess whether this molecular restoration translated into improved effector activity, pretreated pDCs were co-cultured with autologous CD8⁺ T cells. Consistent with IFN-I pathway reactivation, 4µ8C treatment significantly increased IFN-γ and Granzyme B production compared to the control (pDCs alone), and the combination of 4µ8C and CpG yielded the highest levels, surpassing either treatment alone (Fig. 5E). Functionally, these molecular and cytokine changes culminated in enhanced tumor-cell lysis, as evidenced by elevated LDH release in the combination group (Fig. 5F). Together, these results demonstrate that pharmacologic blockade of the IRE1α-XBP1 pathway reverses stress-imposed suppression in pDCs, restores IFN-I production, and re-establishes their ability to activate cytotoxic T-cell responses in vitro. Relationship between intratumoral CD303⁺IRF7⁺ pDC infiltration and the overall tumor immune microenvironment To quantify activated pDC programs in bulk cohorts, we computed an Activated pDC ssGSEA score using a conservative scRNA-seq–derived 7-gene signature (CLEC4C, IL3RA, GZMB, IRF7, OAS1, ISG20, and MX1), whose preferential enrichment in pDCs was confirmed in our scRNA-seq data (Fig. S7A). Using this Activated pDC ssGSEA score, we evaluated its relationship with the overall TIME in TCGA and GEO cohorts. Across cohorts, the Activated pDC score tracked with an immune-inflamed microenvironment. In TCGA, it correlated with CD8⁺ T-cell, Treg, and macrophage signatures (Fig. 6A-C), and remained associated with CD8⁺ T cells and Tregs after stage adjustment, with myeloid effects further assessed using an M2 macrophage signature (Fig. 6D-F). Consistently, the high Activated pDC score group showed higher cytolytic activity and exhaustion scores than the low-score group (Fig. 6G). In GEO, the Activated pDC score showed similar immune associations, remaining positively correlated with CD8⁺ T cells and Treg ssGSEA scores, whereas correlations with macrophage and M2 macrophage signatures were not significant (Fig. S7B-G). Consistent with TCGA, cytolytic activity was higher in the high Activated pDC score group, while exhaustion showed no significant difference (Fig. S7H). Fig. 6Activated pDC programs associate with immune-stromal context in bulk iCCA cohorts and are validated by multiplex immunofluorescence. A-C In TCGA-CHOL, correlations between Activated pDC score and ssGSEA scores for (A) macrophages, (B) CD8⁺ T cells, and (C) Tregs. Lines indicate linear fit with 95% CI; R and P values are shown. D-F Partial Spearman correlations (adjusted for stage group: early vs. late) between Activated pDC score and ssGSEA scores for (D) M2 macrophages, E CD8⁺ T cells, and (F) Tregs; dashed lines indicate trends. G Cytolytic activity and exhaustion scores in tumors with low vs. high Activated pDC score (median split). H Association between Activated pDC score and inferred fibroblast abundance (MCP-counter fibroblasts score) after adjusting for tumor purity (partial Spearman and multivariable regression shown). I Activated pDC scores across four immune–stromal states stratified by median splits of activated CD8⁺ T-cell score (CD8high/low) and fibroblast score (CAFhigh/low); Kruskal-Wallis and BH-adjusted pairwise P values are indicated. J Representative mIF images of iCCA tissue compartments (adjacent normal, invasive margin, tumor core) stained for CD303 (pDC marker, yellow) and IRF7 (activation marker, red); nuclei counterstained with DAPI (blue). Scale bar, 50 μm. K Quantification of CD303⁺ pDC density (cells/mm²) across compartments. L pDC activation ratio (CD303⁺IRF7⁺/CD303⁺, %) across compartments. M Intratumoral pDC activation ratio in early-stage (I–II) vs. advanced-stage (III–IV) disease. N-O Kaplan-Meier overall survival stratified by high vs. low CD303⁺IRF7⁺ pDC levels in (N) tumor tissue and (O) adjacent normal tissue (median cutoff; log-rank test). Data are shown as mean ± SEM where applicable. Letters denote statistical groupings; groups sharing the same letter are not significantly different (BH-adjusted P > 0.05). **P < 0.01; ***P < 0.001 Activated pDC programs associate with immune-stromal context in bulk iCCA cohorts and are validated by multiplex immunofluorescence. A-C In TCGA-CHOL, correlations between Activated pDC score and ssGSEA scores for (A) macrophages, (B) CD8⁺ T cells, and (C) Tregs. Lines indicate linear fit with 95% CI; R and P values are shown. D-F Partial Spearman correlations (adjusted for stage group: early vs. late) between Activated pDC score and ssGSEA scores for (D) M2 macrophages, E CD8⁺ T cells, and (F) Tregs; dashed lines indicate trends. G Cytolytic activity and exhaustion scores in tumors with low vs. high Activated pDC score (median split). H Association between Activated pDC score and inferred fibroblast abundance (MCP-counter fibroblasts score) after adjusting for tumor purity (partial Spearman and multivariable regression shown). I Activated pDC scores across four immune–stromal states stratified by median splits of activated CD8⁺ T-cell score (CD8high/low) and fibroblast score (CAFhigh/low); Kruskal-Wallis and BH-adjusted pairwise P values are indicated. J Representative mIF images of iCCA tissue compartments (adjacent normal, invasive margin, tumor core) stained for CD303 (pDC marker, yellow) and IRF7 (activation marker, red); nuclei counterstained with DAPI (blue). Scale bar, 50 μm. K Quantification of CD303⁺ pDC density (cells/mm²) across compartments. L pDC activation ratio (CD303⁺IRF7⁺/CD303⁺, %) across compartments. M Intratumoral pDC activation ratio in early-stage (I–II) vs. advanced-stage (III–IV) disease. N-O Kaplan-Meier overall survival stratified by high vs. low CD303⁺IRF7⁺ pDC levels in (N) tumor tissue and (O) adjacent normal tissue (median cutoff; log-rank test). Data are shown as mean ± SEM where applicable. Letters denote statistical groupings; groups sharing the same letter are not significantly different (BH-adjusted P > 0.05). **P < 0.01; ***P < 0.001 To incorporate stromal features into our bulk immune infiltration analysis, we inferred fibroblast abundance using MCP-counter (Fibroblasts score) as a CAF-related surrogate. The Activated pDC score was not correlated with fibroblast abundance in univariate analysis (Spearman ρ = −0.003, P = 0.985), consistent with the strong influence of tumor purity on stromal signals in bulk transcriptomes. After adjusting for tumor purity, Activated pDC score showed an inverse association with fibroblast abundance (partial Spearman ρ = −0.359, P = 0.034; multivariable regression P = 0.0288; Fig. 6H). We then stratified TCGA-CHOL tumors into four CD8 and CAF immune stromal states using median splits of activated CD8+ T cell score (CD8high vs. CD8low) and fibroblast score (CAFhigh vs. CAFlow). Activated pDC scores differed significantly across these states (Kruskal-Wallis P = 0.00129; Fig. 6I), peaking in the CD8high/CAFlow group (median 1.365) compared with CD8low/CAFhigh (median 1.039; BH-adjusted P = 0.0036) and CD8low/CAFlow tumors (median 1.012; BH-adjusted P = 0.0036). Within CD8high tumors, CAF enrichment was associated with a lower Activated pDC score (CD8high/CAFhigh median 1.174 vs. CD8high/CAFlow median 1.365; BH-adjusted P = 0.0228), suggesting that activated pDC programs preferentially coincide with an immune-inflamed context and may be attenuated in CAF-enriched states. We next validated these observations in iCCA tissues using mIF to quantify total pDCs (CD303⁺) and activated pDCs (CD303⁺IRF7⁺) (Fig. 6J). Across spatial compartments, CD303⁺ pDC density was highest at the invasive margin, intermediate in the tumor core, and lowest in matched adjacent normal regions (BH-adjusted P < 0.05; Fig. 6K). Similarly, pDC activation, expressed as the activation ratio (CD303⁺IRF7⁺/CD303⁺, %), peaked at the invasive margin and remained elevated in the tumor core compared with adjacent normal tissue (BH-adjusted P < 0.05; Fig. 6L). Moreover, the intratumoral activation ratio was higher in advanced-stage (III-IV) than early-stage (I-II) tumors (P < 0.001; Fig. 6M), supporting an association between pDC activation and disease progression. Univariable logistic regression showed no association between the intratumoral pDC activation ratio (CD303⁺IRF7⁺/CD303⁺) and lymph node metastasis (β = −0.006, P = 0.923; OR = 0.994, 95% CI 0.872–1.138), suggesting limited predictive value of pDC activation for nodal spread in this cohort. Because our mIF quantification was performed at the patient level rather than on individual tumor nodules, we did not obtain nodulespecific measurements for multifocal cases; therefore, we cannot directly determine whether pDC reprogramming occurs uniformly across all tumor foci within the same patient. As an indirect assessment, we compared the intratumoral pDC activation ratio between multifocal and unifocal iCCA and observed no significant difference. Kaplan-Meier analysis further showed that higher levels of activated pDCs in tumor tissues were associated with worse overall survival (P < 0.05; Fig. 6E), whereas higher pDC activation in adjacent normal tissues correlated with better survival (P < 0.05; Fig. 6F). Collectively, these findings link activated pDC programs to distinct immune-stromal contexts in iCCA; however, these analyses are correlative, and further functional studies will be required to establish mechanistic causality. Discussion pDCs have long been recognized as pivotal sentinels of antiviral and antitumor immunity through their capacity to produce large amounts of IFN-I upon activation [28]. However, their role in solid tumors remains paradoxical, with reports describing both antitumor and pro-tumor activities across cancer types [29, 30]. In this study, we combined genetic, transcriptomic, and functional analyses to elucidate the context-dependent functions of pDCs in iCCA. Using MR, we first demonstrated a causal and protective association between genetically predicted circulating pDC levels and reduced iCCA risk. This genetic evidence implies that pDCs are not mere bystanders but active components of immune surveillance against biliary malignancy. Consistently, bulk transcriptomic analyses across the TCGA and GEO cohorts revealed that high pDC infiltration correlates with enhanced overall immune cell abundance and earlier tumor stage, indicating a favorable immune contexture when pDC presence is maintained. These bulk-level observations imply that in early-stage or less immunosuppressive contexts, pDCs retain their antitumor functionality, potentially reinforcing adaptive immunity through TLR9-mediated signaling [31]. This finding aligns with reports in colorectal cancer, where pDC-derived IFN-α is associated with improved survival [30], yet contrasts with hepatocellular carcinoma, where chronic inflammation may lead to premature pDC exhaustion [29], underscoring the unique fibrotic and immunomodulatory niche of iCCA. Single-cell transcriptomic analysis revealed that tumor-associated pDCs display a transcriptional program enriched for immediate-early and stress-responsive transcription factors, including FOS, JUN, ATF3, DUSP1, GADD45B, and members of the NR4A family. Such genes are rapidly induced by environmental or metabolic stress and constitute a conserved adaptive response network that integrates signals from the UPR and the integrated stress response [24–27]. Activation of this network is closely linked to the IRE1α-XBP1 branch of the UPR, which coordinates transcriptional adaptation to endoplasmic-reticulum stress [32, 33]. In our analysis, the downstream signature of this pathway was prominently enriched in tumor-associated pDCs, further corroborated by functional restoration upon pharmacological inhibition, supporting a key role for IRE1α-XBP1-mediated stress reprogramming in the iCCA microenvironment. Persistent stress signals in the TME, such as hypoxia, acidosis, and lipid overload, are known to impose endoplasmic-reticulum stress and activate the IRE1α-XBP1 pathway in immune cells [8, 34]. In the present study, we identify this pathway as a critical regulator of pDC dysfunction in iCCA. Chronic activation of IRE1α-XBP1 appears to restrain type-I interferon production, thereby weakening the immunogenic potential of pDCs and contributing to an immunosuppressive tumor milieu. Pharmacologic inhibition of this pathway by 4µ8C effectively reversed these stress-associated features, leading to the restoration of IFN-I secretion and recovery of T cell stimulatory activity. These findings indicate that the tolerogenic phenotype of tumor-associated pDCs is not fixed but rather imposed by microenvironmental stress, and that it can be reprogrammed through relief of ER stress. Combined treatment with 4µ8C and the TLR9 agonist CpG augmented IFN-I production, indicating that relieving ER stress can potentiate TLR9-mediated activation of pDCs and enhance their immunostimulatory function. Notably, while CD8+ T cells and Treg associations were reproducible across cohorts, macrophage and M2 signatures showed cohort-dependent variability, underscoring the heterogeneity of myeloid and stromal signals in bulk transcriptomes and the importance of purity and stromal adjustment. The spatial and clinical analyses further supported the functional implications of this stress reprogramming. Multiplex immunofluorescence confirmed increased infiltration and activation (CD303+IRF7+) of pDCs at the invasive margin and within tumor regions compared with matched adjacent normal tissues. Notably, a higher intratumoral pDC activation ratio (CD303+IRF7+/CD303+) was associated with advanced TNM stage and poorer overall survival, whereas higher activation in adjacent non-tumor tissue correlated with a favorable prognosis. These opposite survival associations across tumor and adjacent tissues support spatially context-dependent pDC biology; however, given the correlative nature of bulk and spatial analyses, mechanistic causality will require in vivo perturbation models. Extending the bulk analysis to stromal features, purity-adjusted models revealed an inverse association between the Activated pDC program and inferred fibroblast abundance, and stratification by CD8-CAF states showed peak Activated pDC scores in CD8high/CAFlow tumors. Together, these data suggest that activated pDC programs preferentially coincide with an immune-inflamed context yet may be dampened in CAF-enriched states, pointing to an immune-stromal constraint on pDC activation in iCCA. This context dependence is consistent with our transcriptional data, indicating that iCCA pDCs are reprogrammed toward a stress-adapted state that restrains immunogenicity and may contribute to divergent clinical effects across tissue compartments. Mechanistically, sustained activation of the IRE1α-XBP1 axis may contribute to the functional polarization of pDCs through several non-exclusive and context-dependent mechanisms. First, XBP1s have been implicated in restraining type-I interferon production by interfering with IRF7 phosphorylation and nuclear translocation in chronically stressed dendritic cells [35]. Second, IRE1α signaling can engage MAPK-AP-1 and NF-κB pathways to induce immunoregulatory molecules such as PD-L1, thereby reinforcing a tolerogenic phenotype in tumor-associated antigen-presenting cells [36, 37]. Finally, prolonged ER stress has been shown to impair antigen cross-presentation and promote lipid accumulation in dendritic cells, collectively fostering an immunosuppressive microenvironment [38, 39]. In our study, pharmacologic inhibition of this pathway with 4µ8C reversed stress-associated signatures and restored type I IFN-linked activation in vitro, supporting the idea that alleviating ER-stress signaling can shift tumor-conditioned pDCs toward a more immunogenic phenotype and warranting further evaluation of this strategy in in vivo models. Conceptually, these findings integrate metabolic and proteostatic stress within the TME into the framework of immune regulation. By identifying the IRE1α-XBP1 pathway as a potentially reversible regulatory node for pDC activation, our study expands the traditional view of tumor-associated immune suppression and introduces a model of stress-mediated immunomodulation in iCCA. This framework helps reconcile the context-dependent roles of pDCs reported across cancer types and positions ER-stress signaling as a potentially actionable axis for immune reprogramming. However, one important limitation of our study is that the precise factors governing pDC recruitment and reprogramming within iCCA tumors remain unclear. While our data indicate that pharmacological modulation of the IRE1α-XBP1 pathway can revert pDCs to a more immunogenic state, we have not definitively shown whether this reprogramming is a consequence of tumor progression or a driver of it. Additionally, the mechanisms governing pDC recruitment and reprogramming within iCCA tumors require further clarification, particularly with respect to the relative contributions of tumor-intrinsic genetic features and the host immune context. Future studies using in vivo models, including immunocompetent or humanized systems with pDC manipulation, will be crucial to determine whether stress-associated pDC reprogramming drives tumor progression or reflects the evolving iCCA microenvironment. Moreover, these findings provide a rationale for future studies to test whether pharmacologic modulation of the IRE1α-XBP1 pathway can reshape pDC-associated immune states in vivo. Whether such modulation could be rationally integrated with existing immunotherapeutic strategies remains to be determined. Although our evidence is currently limited to correlative patient datasets and in vitro perturbation, these results support further evaluation of IRE1α-XBP1 inhibition as a potential strategy to restore type I IFN-driven pDC immunogenicity and strengthen antitumor T-cell responses in appropriate in vivo models. From a translational perspective, our findings raise the possibility that the spatial distribution and activation status of pDCs may have exploratory value in characterizing the immune contexture and prognosis of iCCA. In our spatial analyses, higher activated pDCs in the tumor core were linked to advanced TNM stage and poorer overall survival, consistent with prior evidence linking elevated peritumoral BDCA2⁺ (CD303⁺) pDC infiltration to adverse clinicopathological features and poor prognosis in iCCA [40]. However, the biological basis of these clinical associations is likely multifactorial, reflecting a complex interplay of stromal remodeling, metabolic stress, hypoxia, myeloid inflammation, and other immune-regulatory processes rather than a single immunosuppressive mechanism. Consequently, any direct relationship between pDC reprogramming, specific PD-1/PD-L1 checkpoint biology, and responsiveness to immuno-oncology therapies remains hypothetical and is not directly addressed by our current data. Prospective studies in iCCA cohorts treated with immuno-oncology therapies, integrating multiplex tissue profiling, single-cell analyses, and established biomarkers such as PD-L1 expression, MSI status, and tumor mutational burden, will be required to determine whether these pDC features have predictive, rather than purely prognostic, value or could inform future therapeutic strategies [41, 42]. Supplementary Information Supplementary Material 1. Supplementary Material 1.