Genomic and immune profiling characterizes immune heterogeneity and a TIGIT-associated subtype in intrahepatic cholangiocarcinoma.
Oct 19, 2026DOI: 10.1097/hc9.0000000000001052Journal ArticleHepatology Communications
Written by: Anastasiia KorostelevaJosephine Sophie ReschkeDawei YiSilke MarhenkeJinbo ZhaoValery VolkTanja Reineke-PlaaßSimon PeterCharlotte HoffmannDoan Duy Hai TranChristoph GerdesMelanie BathonWalter KolchFriedrich FeuerhakeVadim ZhernovkovAnna SaborowskiArndt Vogel
INTRODUCTION
Intrahepatic cholangiocarcinoma (iCCA) is an aggressive and molecularly heterogeneous primary liver cancer with a rising global incidence. Over the past decade, advances in precision oncology have begun to reshape its therapeutic landscape. Selective inhibitors targeting recurrent genomic alterations (GAs) can induce durable responses in molecularly defined subgroups, and immune checkpoint inhibitors (ICIs) have demonstrated consistent, albeit modest, clinical activity in patients with biliary tract cancers (BTCs).1 However, overall survival (OS) gains remain limited for most patients. A major barrier to further progress is the lack of robust, widely validated biomarkers beyond actionable GAs to reliably identify patients most likely to benefit from systemic therapies, and particularly from the addition of ICIs.Large-scale genomic and transcriptomic profiling has defined recurrent molecular signatures in iCCA that inform biology, prognosis, and therapeutic opportunities.2 Canonical oncogenic drivers include FGFR2 gene fusions as well as IDH1/2 and BRAFV600E mutations, among others, which define therapeutic niches.345 Alterations in chromatin modifiers (eg, BAP1, ARID1A), tumor suppressors (TP53, CDKN2A/B), and genes involved in DNA repair and genome stability are also observed, contributing to heterogeneity in clinical behavior. At the pathway level, aggregated signatures point to frequent perturbation of RTK–RAS signaling, PI3K–AKT–mTOR pathways, WNT signaling, and cell-cycle regulatory networks.6 Integrative transcriptomic classifiers separate tumors into “inflamed” (immune-rich) and “non-inflamed” (immune-excluded or immune-desert) phenotypes, and more granular schemes capture stromal composition, macrophage polarization, and T-cell exhaustion signatures that carry prognostic and potential predictive value.7Clinically actionable features associated with sensitivity to ICIs are relatively uncommon in iCCA. Microsatellite instability (MSI) tumors and tumors with high tumor mutational burden (TMBhigh)—both biomarkers of ICI responsiveness across cancer types—are rare in iCCA.2 Similarly, defects in other genomic processes that generate abundant neoantigens occur infrequently, limiting the proportion of patients expected to mount robust neoantigen-driven antitumor immune responses. PD-L1 expression assessed by immunohistochemistry has been explored as a potential predictive marker in BTC, but its utility is inconsistent: PD-L1 is often expressed at low or heterogeneous levels in iCCA, and PD-L1 positivity does not predict response to PD-1/PD-L1 blockade in this disease.8910 Consequently, PD-L1 status alone is insufficient for patient selection, and response rates to ICI remain modest outside of MSI/TMBhigh subsets.
Recent spatial and single-cell studies underscore additional complexity: immune checkpoint (IC) molecules such as PD-1, TIGIT, LAG-3, and CTLA-4 are heterogeneously distributed within tumors and can be differentially enriched in tumor cores, stroma, or invasive margins. Co-expression patterns and the physical proximity of exhausted or regulatory lymphocytes to cancer cells may further modulate response to therapy. These spatial dimensions are not captured by bulk sequencing and may explain discordant biomarker–response relationships observed clinically.
Given the limited prevalence of classical ICI biomarkers in iCCA, there is an urgent need for integrated biomarker strategies that combine GAs, pathway activity, transcriptional immune signatures, and spatially resolved immune phenotyping. Such multimodal approaches could identify immune-responsive subgroups, reveal mechanisms of primary resistance, and prioritize rational combinations of ICIs. This study aims to comprehensively map IC expression and immune architecture in iCCA across molecularly defined subtypes, to inform biomarker development and guide next-generation immunotherapy strategies.
METHODS
The study cohort consisted of 144 patients with histologically confirmed intrahepatic cholangiocarcinoma (iCCA) who underwent curative-intent surgical resection at Hannover Medical School between 2000 and 2019. Patients were included if adequate formalin-fixed paraffin-embedded tumor tissue and corresponding molecular and clinical data were available for integrated genomic, transcriptomic, and spatial immune profiling analyses. Clinical and pathological data were collected retrospectively from institutional databases and medical records. Variables included demographic characteristics, tumor morphology, pathological stage, tumor grade, lymph node status, resection margin status, vascular invasion, multifocality, and survival outcomes. Detailed baseline characteristics of the study population are provided in Supplemental Table S1, https://links.lww.com/HC9/C498. The use of archived tissues was approved by the local institutional review board (IRB)/ethics commission of Hannover Medical School (approval number 3612-2017), and the study was conducted in accordance with the principles of the Declaration of Helsinki and the 2018 Declaration of Istanbul. All patients provided written informed consent to participate. Independent analyses from the same cohort were previously reported.10 We used multi-omics data, including: (i) multiplex immunohistochemistry (mIHC) data that quantifies immune-cell populations and their expression of IC markers within different tissue regions. Two panels characterized the spatial distribution of cytokeratin AE1/3+ tumor cells, CD34 as a marker for vascular endothelial cells, and key immune markers, including CD3 as a pan-T-cell marker, TIGIT+, PD-1+, LAG-3+, and CTLA-4+ (ii) bulk RNA-seq and (iii) panel DNA sequencing.
Next-generation sequencing was performed using the hybrid-capture FoundationOne CDx (F1CDx) assay (Foundation Medicine, Inc., Boston, MA, USA), as previously described, in a Clinical Laboratory Improvement Amendments (CLIA)–certified and College of American Pathologists (CAP)–accredited laboratory.11
All analyses were performed in R (v4.4.3) using RStudio (Posit, v2024.04.2+764.pro1) on Ubuntu 22.04.5 LTS. Gene Set Variation Analysis (GSVA) was used to assign patients to Stroma-Tumor-Immune Microenvironment (STIM) classes based on established classification.712 Kaplan–Meier survival curves and Cox proportional hazards regression were used to evaluate survival differences between STIM classes. MCP-counter, a transcriptomic-based method that estimates the absolute abundance of tissue-infiltrating immune and stromal cell populations using gene expression markers specific to each lineage, was used to deconvolute RNA sequencing data into immune and stromal cell populations.To visualize and identify frequent GAs, we used ComplexHeatmap, patterns of co-occurring and mutually exclusive alterations were assessed with the Maftools R package.1314 Differentially expressed genes (DEGs) were identified using DESeq2.15 Transcription factor (TF) activity inference was performed using the decoupleR R package and CollecTRI database, integrating prior knowledge of TF-target interactions.16 Finally, fgsea (Fast Gene Set Enrichment Analysis) was conducted to identify enriched biological pathways associated with differentially expressed genes in the TIGIThigh patients.17
All statistical analyses were performed using R and GraphPad Prism. Patients were stratified into High and Low groups using the median value as the predefined cutoff for Kaplan–Meier analyses. Survival analyses were performed using the survival package in R. Univariable and multivariable Cox proportional hazards regression models were used to assess associations with survival outcomes. Continuous and categorical variables were compared using the Wilcoxon rank-sum test and the Fisher exact test, respectively. p-values were adjusted using the Benjamini–Hochberg method where appropriate. Statistical analyses for Figures 2C, F were performed in GraphPad Prism. Detailed statistical methods are provided in the Supplemental Methods, https://links.lww.com/HC9/C497. All analysis scripts are publicly available at: https://github.com/AnastasiiaKorosteleva/TIGIT-Expression-in-iCCA-Is-Associated-with-Immune-Activation-and-Regional-Heterogeneity/.
Additional information on mIHC and statistical analyses is provided in the supplements.
RESULTS
A total of 144 patients with resected iCCA were included in the study. Baseline clinicopathological characteristics of the cohort are summarized in Supplemental Table 1, https://links.lww.com/HC9/C498. To characterize the tumor immune microenvironment, we integrated transcriptomic, genomic, and mIHC data. Tumors were stratified according to the STIM classification (Figure 1A), which stratifies iCCA into 5 biologically distinct subtypes based on tumor-intrinsic features, stromal composition, and immune microenvironment characteristics.7 In our cohort, ~47% of tumors fell into the “inflamed” category, comprising the Immune Classical subtype (~27.5%) and the Inflammatory Stroma subtype (~19.5%). The remaining 53% were classified as “non-inflamed,” including the Desert-like (~12%), Hepatic Stem-like (~27.5%), and Tumor Classical (~13.5%) subtypes. Consistent with the pivotal study, the Hepatic Stem-like subtype was associated with the most favorable OS (Figure 1B). In contrast, the Desert-like and Tumor Classical subtypes exhibited the poorest prognosis, reflecting more aggressive tumor biology.
STIM-based stratification of patients with intrahepatic cholangiocarcinoma. (A) Tumors were classified according to the stroma, tumor, and immune microenvironment (STIM) framework7 using single-sample gene set enrichment analysis (ssGSEA).12 The 30 most frequent genetic alterations are depicted. Pathways and genes were defined accordingly.6 (B) Kaplan–Meier curves comparing overall survival across STIM groups. (C) Summary of differential genetic alterations associated with each STIM class (each class compared with all remaining samples). p-values were calculated using the Fisher exact test.
Across the cohort, we identified a total of 1507 GAs, comprising single-nucleotide variants (n=897), small insertions (n=74) and deletions (n=128), as well as complex structural events such as copy-number alterations (n=360), and gene fusions (n=48). The 5 most frequently GAs were CDKN2A loss (30%), CDKN2B loss (27%), TP53 mutations (22%), MTAP loss (20%), and BAP1 mutations (19%) (Figure 1A). Several of these alterations were significantly associated with unfavorable clinical outcomes. Specifically, CDKN2A loss (HR=2.67), CDKN2B loss (HR=2.42), MTAP loss (HR=2.57), and TP53 mutations (HR=1.92) were linked to shorter OS (Supplemental Figures S1A, B, https://links.lww.com/HC9/C499). The most prominent co-occurring GAs included MTAP loss with CDKN2A/B loss, concurrent KRAS/APC and TP53 alterations, DIS3 with APC alterations, and co-occurrence of MCL1 amplification with ATM alterations (Supplemental Figure S1C, https://links.lww.com/HC9/C499). In contrast, the strongest patterns of mutual exclusivity were observed between IDH1 mutations and MTAP loss, FGFR2 fusions and TP53, ARID1A, or IDH1 mutations, as well as between KRAS and PBRM1 mutations.
We next examined the distribution of recurrent GAs across STIM-defined TIME classes (Figures 1A, C). In the Desert-like class, BRCA2 mutations were significantly enriched, occurring in 30.8% of cases versus 6.0% in the other classes (p=0.0137). The Hepatic Stem-like group exhibited a distinct molecular profile that was overall in line with the original publication by Martin-Serrano et al:7 BAP1 mutations were significantly overrepresented (37.5% vs. 11.1% in other classes; p=0.0012). FGFR2 fusions were also significantly more frequent (22.5% vs. 7.8%; p=0.039) and IDH1 mutations showed a higher prevalence (27.5% vs. 15.6%), although this difference did not reach statistical significance (p=0.147). Conversely, several tumor suppressor alterations and focal deletions occurred less frequently in the Hepatic Stem-like class: CDKN2A deletions were observed in 17.5% of Hepatic Stem-like tumors compared with 35.6% in other classes (p=0.0407), CDKN2B deletions in 15.0% versus 33.3% (p=0.0348), MTAP deletions in 7.5% versus 25.6% (p=0.018), and TP53 mutations in 7.5% versus 28.9% (p=0.0062). On the contrary, in the Inflammatory Stroma class, TP53 mutations were significantly enriched (41.7% vs. 17.9% in other classes; p=0.0266), and KRAS mutations were numerically higher though not statistically significant (25.0% vs. 10.4%; p=0.087). Collectively, these results reveal class-specific enrichment and depletion of key oncogenic events, further supporting the assumption that STIM-defined microenvironment phenotypes are associated with distinct underlying molecular alterations.
At the pathway level (Figure 1A, lower panel), enrichment analysis using TCGA-derived oncogenic signaling signatures identified RTK–RAS (69%), PI3K (41%), and WNT (20%) as the most frequently altered cascades.6 Within STIM classes, mutations affecting the Cell Cycle pathway were significantly enriched in the Inflammatory Stroma group (21.9% vs. 7.6% in other classes; p=0.0257). The Tumor Classical class was significantly enriched for TGF-Beta pathway alterations (22.7% vs. 5.6%; p=0.0172). By contrast, the Immune Classical class exhibited a significantly lower prevalence of Cell Cycle pathway mutations (2.2% vs. 13.4% in the remainder of the cohort; p=0.0428).
To assess spatial heterogeneity of IC expression in iCCA, we used a compartmental sampling strategy with regions of interest (ROIs) annotated to capture tumor core, stroma, invasive margin, and adjacent liver as previously described (Figure 2A).10 This spatially resolved design allowed comparison of immune profiles across histologically distinct compartments rather than by bulk averaging. For immune phenotyping, we applied a targeted mIHC panel comprising canonical T cell and an endothelial marker, CD34, alongside key ICs (Figure 2B). CD3 served as a pan-T-cell marker to normalize checkpoint-positive fractions, and PD-1, TIGIT, LAG-3, and CTLA-4 were included to interrogate complementary axes of T-cell exhaustion and regulation.
Multiplex immunohistochemistry profiling of spatial immune landscapes in intrahepatic cholangiocarcinoma. (A) Regions of interest (ROIs) were placed into distinct compartments for subsequent comparative analyses: tumor core, tumor stroma, tumor margin, and adjacent liver. (B) Representative composite image illustrating expression of indicated immune checkpoints on CD3+ T cells. (C) Dot plots showing cell densities (cells/mm2) of tumor cells (AE1/3+), endothelial cells (CD34+), T cells (CD3+), and other cells across tumor core, tumor margin, and tumor stroma. Statistical comparisons were performed using the Kruskal–Wallis test followed by the Dunn post hoc test. # indicates significance across all regional comparisons (p<0.05). (D) Densities of CD3 cells (co-)expressing the indicated checkpoints according to tissue region, as well as (E) percentage thereof. (F) Region-specific distribution of checkpoint expression on CD3+ T cells. Statistical comparisons were performed using the Friedman test. Statistical significance is indicated by (p*<0.05). (G) Waterfall plots illustrating intertumoral heterogeneity between patients in immune checkpoint expression across regions. Cases were ordered according to the proportion of IC+ phenotypes of total CD3+ T cells across the cohort.
Analysis of cell-type distribution across histological compartments revealed distinct spatial patterns (Figure 2C). As expected, tumor cells showed a marked spatial gradient across tissue compartments with the highest densities in ROIs placed within the tumor core [median (IQR): 3639.3 (2712.2–4519.1) cells/mm2], with a pronounced reduction at the invasive margin [1090.2 (661.4–1633.3) cells/mm2] and in the stroma [616.3 (246.7–821.2) cells/mm2]. CD3+ T cells showed the lowest median density in the tumor core [91.2 (51.5–182.3) cells/mm2] and progressively higher densities in the stroma [656.4 (266.3–1328.7) cells/mm2] and invasive margin [921.9 (546.2–1500.1) cells/mm2], while adjacent liver tissue exhibited intermediate levels [572.5 (332.2–851.8) cells/mm2]. Endothelial cell (CD34) densities as a surrogate for microvessel density were reduced in the tumor core compared with stromal and margin regions. Median values and (IQRs): were 163.9 (107.8–276.6) cells/mm2 in the core, 187.2 (128.4–266.3) cells/mm2 in the stroma, and 205.1 (123.9–284.4) cells/mm2 in the margin, with the lowest densities observed within adjacent liver tissue [89.1 (63.9–154.6) cells/mm2].
Most infiltrating CD3+ immune cells were negative for the assayed checkpoints across all regions, including the tumor core (86.9%), tumor stroma (88.1%), tumor margin (91.6%), and adjacent liver (92.9%) (Figure 2E). Among CD3+ T cells, TIGIT was the most frequently observed inhibitory receptor across all the regions, both in terms of relative abundance and density. TIGIT+ T cells accounted for 4.6% of T cells in the tumor core, 7.6% in the stroma, 4.2% in the invasive margin, and 4.9% in the adjacent liver. Consistently, TIGIT+ T-cell densities were highest in the stromal compartment [27.9 (5.2–75.9) cells/mm2], followed by the margin [18.8 (7.3–62.8) cells/mm2], liver [18.1 (6.2–40.5) cells/mm2], with low densities in the tumor core [1.8 (0.5–5.1) cells/mm2] (Figure 2D). PD-1 was the second most prevalent immune checkpoint and showed a more even spatial distribution across regions. PD-1+ T cells represented 2.9%, 1.6%, 2.4%, and 1.3% of T cells in the tumor core, stroma, margin, and liver, respectively, with corresponding cell densities ranging from 0.6 (0.0–2.6) cells/mm2 in the tumor core to 5.0 (0.7–26.2) cells/mm2 in the margin, and intermediate densities in the stroma [2.1 (0.0–11.1) cells/mm2] and liver [2.1 (0.0–8.2) cells/mm2].
LAG-3 expression was comparatively sparse and largely restricted to tumor-associated compartments. LAG-3+ T cells comprised 2.7% of T cells in the tumor core, 0.9% in the stroma, and 0.6% in the margin, but were rare in the liver (0.1%). In line with this, LAG-3+ T-cell densities were detectable in the core [0.5 (0.0–2.4) cells/mm2], stroma [2.0 (0.6–6.3) cells/mm2], and margin [1.9 (0.0–5.9) cells/mm]. CTLA-4 was the least abundant immune checkpoint, representing ≤1% of T cells across all compartments (1% in the core, 0.3% in the stroma, 0.3% in the margin, and 0.2% in the liver). Correspondingly, CTLA-4+ T-cell densities were low across regions, with median values and (IQRs): 0.4 (0.0–1.1) cells/mm2 in the tumor core, 1.0 (0.0–3.7) cells/mm2 in the stroma, 1.6 (0.4–3.5) cells/mm2 in the margin, and 0.3 (0.0–1.2) cells/mm2 in the liver. These findings are in line with a previous study, describing the tumor core as “cold” centers with fewer T cells and reduced IC expression compared with the invasive margin.18
Co-expression of multiple ICs on the same T cell was uncommon (Figures 2D, E). TIGIT+/PD-1+ double-positive T cells accounted for 0.5% of T cells in the tumor core, 1.1% in the stroma, 0.7% in the margin, and 0.6% in the liver, with the highest density observed in the stroma [0.9 (0.0–8.3) cells/mm2]. TIGIT+/LAG-3+ T cells were rare across all regions (≤0.4%), while PD-1+/LAG-3+ double-positive T cells were most prevalent in the tumor core (1%).
An IC-centric analysis (Figure 2F) corroborated TIGIT as the dominant inhibitory receptor within the stromal compartment, where it showed the highest expression levels [39.5 (8.5–116.9) cells/mm2]. In contrast, PD-1 expression was particularly enriched at the invasive margin [5.6 (0.9–28.3) cells/mm2]. LAG-3 positivity was mainly observed in the stroma [2.1 (0.6–7.6) cells/mm2], while CTLA-4 showed low but detectable expression in both the stroma [1.0 (0.0–4.0) cells/mm2] and the margin [1.4 (0.4–3.6) cells/mm2], consistent with its overall rarity. Figure 2G illustrates the intratumoral/intertumoral heterogeneity of immune checkpoint expression in the TIME of iCCA.
We next evaluated the prognostic relevance of IC expression across spatial compartments (Figure 3). Cox proportional hazards modeling revealed no statistically significant associations for most IC-expressing cell subsets after correction for multiple testing (Figure 3A). Several cell type–specific variables, including endothelial cells and PD-1+/LAG-3+ double-positive T cells, showed suggestive associations with outcome; however, these associations did not remain significant after multivariable adjustment, including relevant clinical covariates (Figure 3A and Supplemental Figure S4, https://links.lww.com/HC9/C502).
Prognostic relevance of immune checkpoint–expressing T-cell populations in intrahepatic cholangiocarcinoma. (A) Schematic overview of the Cox proportional hazards regression model used to evaluate associations between immune-cell populations and patient survival across tissue regions. (B) Box plots showing the distribution of immune checkpoint–negative T cells (CD3+/TIGIT−/PD-1−/LAG-3−/CTLA-4−) across tissue regions. Statistical comparisons were performed using the Wilcoxon test with Benjamini–Hochberg correction. Significance levels are indicated by (adjusted p*<0.05). (C) Kaplan–Meier survival analysis stratified by the abundance of immune checkpoint–negative T cells in the tumor core. Patients in the lowest quartile (25%) were compared with all remaining patients. Survival differences were assessed using the log-rank test. (D) Kaplan–Meier survival curves stratified by the proportion of CD3+/TIGIT−/PD-1−/LAG-3+/CTLA-4− cells among CD3+ T cells in the segmented tumor area of the tumor stroma region (see Supplemental Methods, https://links.lww.com/HC9/C497, and Supplemental Figure S2, https://links.lww.com/HC9/C500). The median value (0) was used as the cutoff, resulting in a comparison of patients with non-detectable versus detectable levels of this cell population. Survival differences were assessed using the log-rank test. (E) Kaplan–Meier survival curves stratified by the proportion of CD3+/TIGIT+/PD-1−/LAG-3−/CTLA-4− cells among CD3+ T cells in the tumor core (stromal subregion). The median value was used as the cutoff, and survival differences were assessed using the log-rank test.
The lowest fraction of IC-negative (IC−) CD3+ T cells, defined as CD3+/TIGIT−/PD-1−/LAG-3−/CTLA-4− cells among total CD3+ cells, was found in the tumor core, followed by stroma and invasive margin (Figure 3B). Using the 25th percentile of IC-T cell frequency as a cutoff, Kaplan–Meier analysis demonstrated that a higher proportion of IC-T cells in the tumor core was associated with significantly improved OS (Figure 3C); no comparable associations were observed in stroma or margin. Quartile-based analyses of the Kaplan–Meier plot in Figure 3C (Supplemental Figures S3A, B, https://links.lww.com/HC9/C501) indicate that the survival effect is most evident in patients with fewer than 25% IC− CD3+ T cells in the tumor core compared with the remainder of the cohort.
Spatially resolved analyses of individual ICs also revealed region-specific effects (Figures 3D, E and Supplemental Figure S2A, https://links.lww.com/HC9/C500): LAG-3+ T-cell density in the tumor area of tumor stroma ROIs correlated with worse survival (p=0.038; Figure 3D), consistent with an immunosuppressive function of LAG-3+ T cells. Median was used as a cutoff for Figures 3D–E. In contrast, higher densities of TIGIT+ T cells within the stromal area of the tumor core ROIs correlated with improved OS (p=0.045; Figure 3E), suggesting that in this context, TIGIT expression may either mark a partially active T-cell population or reflect compensatory immune responses at sites of high antigen exposure.19 These results indicate that the prognostic relevance of IC expression in iCCA is heterogeneous and context dependent, underscoring the need for region-specific immune profiling in biomarker development.
We next analyzed the spatial relationship between IC-expressing T cells and adjacent tumor cells (Figure 4A). Although the distance from IC+ T cells to the nearest cancer cell showed little variation between different ICs, overall T-cell proximity differed by tumor compartment. As expected, in the tumor core, T cells were typically located within a few tens of micrometers of the nearest cancer cell; in the stroma, distances were greater (~50–100 µm); and at the invasive margin, distances were intermediate (≈25–50 µm) (Figure 4B). When analyzing specific IC+/CD3+ T-cell subsets, CD3+/LAG-3+ cells consistently exhibited closer proximity to tumor cells across all regions compared with IC− CD3+ T cells (padj<0.05 for core, stroma, and margin). By contrast, CD3+/PD-1+ T cells were more distant from cancer cells in the tumor core (core: padj<0.05; stroma and margin: ns), while double-positive CD3+/PD-1+/LAG-3+ T cells were located closer to tumor cells in both the core and margin (core and margin: padj<0.05; stroma: ns). CD3+/TIGIT+ T cells were found at greater distances from tumor cells in the stroma and margin, with no significant spatial difference observed in the core (stroma and margin: padj<0.05; core: ns). The spatial positioning of IC+ T cells may reflect distinct functional roles within the TIME.
Distances between immune checkpoint–expressing T cells and tumor cells. (A) Representative visualization of tumor–T cell distance measurements. CD3+ cells are indicated by green dots, and cytokeratin AE1/3 expression (red) serves as a surrogate marker for tumor cells. (B) Box plots comparing the distances between CD3+ cells to the nearest tumor cell expressing either none (IC−) or the indicated checkpoints. Source (“from”) cell types are shown on the x-axis. Box plots comparing the distances of CD3+ cells expressing either no immune checkpoint (IC−) or the indicated checkpoints to the nearest tumor cell. Source cell types are shown on the x-axis. Statistical comparisons were performed using the Wilcoxon test with Benjamini–Hochberg correction; indicates adjusted p<0.05. CD3+/IC− cells are defined as CD3+/TIGIT−/PD-1−/LAG-3−/CTLA-4− T cells. (C) Cox proportional hazards regression analysis identifying distance-based immune-cell metrics associated with patient survival. Statistical significance was determined using Benjamini–Hochberg–adjusted p* values. (D, E, F, G) Kaplan–Meier survival curves stratified by median distance values. Analyses were performed using subsegmented tumor and stromal areas, as described in the Supplemental Methods, https://links.lww.com/HC9/C497, and Supplemental Figure S2, https://links.lww.com/HC9/C500.
To assess the prognostic impact of spatial immune-cell distribution, we performed Cox proportional hazards models using continuous distance metrics and supplemented these with Kaplan–Meier analyses (Figures 4C, D). Although Cox regression did not identify statistically significant associations, several non-significant trends suggested that immunity to tumor cell proximity may affect outcome. A shorter distance between IC−/CD3+ T cells and cancer cells in the tumor area of the tumor core was associated with worse OS (Figure 4D), as was the closer proximity of endothelial cells to tumor cells in the stroma area of the tumor core (Figure 4E). Conversely, greater distances between CD3+/TIGIT+ T cells and tumor cells in the stroma area within the tumor core ROIs correlated with poorer outcome, consistent with the cell density survival analyses (Figures 4F, G). These associations remained statistically significant in a multivariable analysis adjusted for age, tumor stage, resection margin, tumor diameter, tumor number, and duct morphology (Supplemental Figure S5, https://links.lww.com/HC9/C503).
Finally, we compared TIME composition across STIM classes using mIHC data and tested enrichment with Wilcoxon rank-sum tests versus all other classes (Figures 5A–C). Consistent with Martin-Serrano et al,7 the Inflammatory Stroma class in our cohort was enriched for features of T-cell exhaustion, exhibiting increased densities of CD3+/TIGIT+, CD3+/TIGIT+/PD-1+, and CD3+/TIGIT+/LAG-3+ subsets, and a reduced proportion of IC-T cells. The Hepatic Stem-like class showed the lowest densities of CD3+/LAG-3+, CD3+/TIGIT+/PD-1+, CD3+/TIGIT+/LAG-3+, and CD3+/PD-1+/LAG-3+ T cells, consistent with its previously reported “non-inflamed” immunosuppressive phenotype characterized by enrichment of M2-like macrophages. The Immune Classical class exhibited a more distinct pattern: within the tumor core, CD3+/TIGIT+ T cells were significantly reduced versus other classes, with no consistent differences for other IC or cell types. The Desert-like and Tumor Classical classes showed no significant deviations in exhausted or checkpoint-positive T-cell densities relative to other STIM categories. Overall, these data position the Inflammatory Stroma and Hepatic Stem-like classes at opposite ends of the IC activity spectrum in iCCA: one defined by abundant exhausted T-cell populations in a fibroblast-rich stroma, the other by marked depletion of these subsets.
Features of the microenvironment according to STIM class. (A–C) Tumor microenvironment features were compared for each STIM class versus all other classes using the Wilcoxon test with Bonferroni correction. Red boxes indicate cell populations with higher density relative to other classes, whereas green boxes indicate lower density. Statistical significance is indicated by *, , and , corresponding to Benjamini–Hochberg–adjusted p*-values <0.001, 0.001–0.01, and 0.01–0.05, respectively. No significant differences were observed for the Desert-like and Tumor Classical classes. Analyses were performed using multiplex immunohistochemistry (mIHC) data, with log2-transformed cell densities measured across tissue regions (tumor core, tumor stroma, tumor margin, and adjacent liver). (D) Cell-type deconvolution from transcriptomic data using the MCP-counter method to estimate the abundance of immune-cell populations.
To further characterize immune and stromal differences across STIM classes, we compared MCP-counter–derived20 cell-type enrichment scores (Figure 5D) between each class and all remaining tumors using two-sided Wilcoxon rank-sum tests, followed by Benjamini–Hochberg FDR correction. The Hepatic Stem-like class (n=41) demonstrated significantly lower enrichment of Neutrophils (padj=1.16×10−8), T cells (9.37×10−8), B lineage (4.02×10−5), Monocytic lineage (6.79×10−5), Myeloid dendritic cells (7.81×10−5), Cytotoxic lymphocytes (1.11×10−5), CD8 T cells (1.96×10−5), Endothelial cells (padj<0.05), and NK cells (padj<0.05), indicating a broadly immune-depleted phenotype. The Immune Classical group (n=41) was characterized by significantly higher Neutrophil enrichment (padj=2.31×10−8). In line with previous findings, increased neutrophil infiltration has been linked to worse clinical outcomes, and this association persists even in the presence of CD3+ T cells.21 The Inflammatory Stroma class (n=26) showed markedly higher levels of T cells (padj=9.86×10−9), Cytotoxic lymphocytes (2.50×10−8), B lineage (3.28×10−8), Endothelial cells (1.30×10−5), Myeloid dendritic cells (1.90×10−4), CD8 T cells (9.56×10−4), Monocytic lineage (padj<0.01), NK cells (padj<0.05), and Fibroblasts (padj<0.05), consistent with an inflamed and stromal-enriched TIME with prominent fibroblast infiltration, as previously reported.7 Finally, the Tumor Classical class (n=18) displayed significantly lower Endothelial cell enrichment (padj<0.01). Regarding clinical parameters, histopathologically defined large duct type was enriched in the Tumor Classical class, while small duct type was enriched in the Hepatic Stem-like class (padj<0.05 for both). No significant associations were found for gender distribution across classes.
Because TIGIT was the most frequently expressed IC in our cohort, we performed transcriptomic profiling to define the molecular context of TIGIThigh tumors. To stratify tumors, we used a percentile-based approach based on immune-cell densities. Specifically, TIGIThigh tumors were defined as those falling within the top 25% (≥75th percentile) of CD3+/TIGIT+/PD-1−/LAG-3−/CTLA-4− T-cell density in the stroma. Using this cutoff, 42 tumors were classified as TIGIThigh, while the remaining tumors (n=99) were classified as TIGITlow. The stromal compartment was selected for stratification because it showed the highest overall TIGIT+ T-cell densities and the greatest dynamic range across patients.
Differential expression analysis revealed a distinct molecular signature for TIGIThigh tumors (Figure 6A). Among the top upregulated genes were chemokines and receptors implicated in immune recruitment (eg, CCL4, CCR4), along with SLAMF7 and CD200R1. Pathway enrichment indicated coordinated activation of immune reaction and antigen-presentation programs (Figure 6B), including T-cell receptor (TCR) signaling, CD28 co-stimulation, B-cell receptor (BCR) signaling, interferon-γ/α/β responses, and IL-3/IL-5/GM-CSF signaling alongside parallel enrichment of immunoregulatory pathways such as IL-10 signaling, PD-1 signaling, and negative regulation of DDX58/IFIH1-mediated RNA sensing.
Molecular characterization of TIGIThigh tumors. (A) Heat map showing the top 50 significantly upregulated differentially expressed genes in TIGIThigh tumors, ranked by adjusted p-value based on DESeq2 analysis.15 (B) Functional pathway enrichment analysis identifying upregulated and downregulated gene sets associated with TIGIThigh tumors was performed using Fast Gene Set Enrichment Analysis (fgsea).17 Reactome gene sets were obtained using the msigdbr R package, and significantly enriched pathways (p-adjusted <0.05) were identified in TIGIThigh tumors. (C) Transcription factor (TF) activity analysis identifying upregulated and downregulated TFs driving regulatory programs of TIGIThigh tumors. Analysis was performed using the decoupleR R package, with TF activity scores calculated using the univariate linear model (ULM) method.16 (D) Genetic alterations enriched in TIGIThigh tumors were identified by comparing TIGIThigh samples with all other tumors using the Fisher exact test.
Transcription factor activity inferred by a univariate linear model with the CollecTRI database16 showed increased activity of TFs linked to T-cell differentiation and antigen presentation (FOXP3, RUNX1, NF-κB, IRF1, AP-1, RFX) and reduced activity of TFs associated with developmental plasticity and chromatin regulation (HDAC1, SOX2, GLI2) (Figure 6C).
Genomic profiling identified multiple alterations enriched in TIGIThigh tumors (Figure 6D). Significantly more frequent somatic mutations included APC (20.7% vs. 4.9%, p=0.0148), CDKN2A (13.8% vs. 2.9%, p=0.0425), DIS3 (20.7% vs. 3.9%, p=0.0080), FLT1 (17.2% vs. 4.9%, p=0.0423), and GNAS (10.3% vs. 1.0%, p=0.0338). Mutations in MLH1, PDGFRB, and QKI were detected in 10.3% of TIGIThigh patients and were absent in the comparator group (all p=0.0100). Additional recurrent mutations included RNF43 (13.8% vs. 2.0%, p=0.0216), RUNX1 (6.9% vs. 0%, p=0.0477), and SMAD4 (13.8% vs. 2.9%, p=0.0425). Copy-number gains enriched in TIGIT-high tumors included AKT2 (10.3% vs. 1.0%, p=0.0338), FGF10, PPP2R1A, and RICTOR (each 6.9% vs. 0%, p=0.0477), while TEK loss occurred uniquely in TIGIThigh cases (6.9% vs. 0%, p=0.0477). TIGIT expression did not associate significantly with any STIM class (p=0.2646).
Collectively, these data indicate that TIGIT expression delineates a molecular subtype marked by concurrent immune activation and regulatory checkpoint signaling, underpinned by a characteristic transcriptional program and a distinct spectrum of genetic alterations.
DISCUSSION
In this integrated study of 144 resected iCCA combining STIM-based transcriptomics, next-generation sequencing, and compartmental mIHC, we show that microenvironment class, genotype, and spatial immune architecture are interdependent and define clinically relevant iCCA subsets. These findings refine prior classifiers and have practical implications for biomarker development and therapeutic selection in a disease with limited immunotherapy benefit.
STIM classification reproducibly separated tumors into inflamed (47%) and non-inflamed (53%) groups with distinct outcomes: Hepatic Stem-like tumors had the best OS, while Desert-like and Tumor Classical had the worst. Importantly, STIM classes correlated with specific genomic patterns. Hepatic Stem-like tumors were enriched for BAP1 mutations and FGFR2 fusions and depleted of CDKN2A/B, MTAP, and TP53 alterations; Desert-like tumors showed BRCA2 enrichment; and Inflammatory Stroma tumors harbored more TP53 alterations, a trend for KRAS, and increased cell-cycle pathway involvement. These associations support a model in which oncogenic programs and the stromal/immune milieu reciprocally shape tumor behavior.
Spatial profiling exposed pronounced compartmental heterogeneity in IC expression and immune-cell localization that bulk assays mask. TIGIT predominated in the stroma, PD-1 was relatively enriched at invasive margins, and LAG-3 and CTLA-4 were focal and core-restricted. The Inflammatory Stroma class exhibited high densities of exhausted/checkpoint-positive T cells (TIGIT+, TIGIT+/PD-1+, TIGIT+/LAG-3+), whereas Hepatic Stem-like tumors had the lowest densities of these subsets. These STIM-specific exhaustion signatures imply differing mechanisms of immune engagement: Inflammatory Stroma tumors contain antigen-experienced but suppressed T cells and may be amenable to combination checkpoint strategies, while Hepatic Stem-like tumors appear immunologically “colder” and may better suit targeted therapies or immune-priming approaches.
Stromal LAG-3+ T cells correlated with poorer OS, whereas stromal TIGIT+ T cells and higher fractions of IC-negative T cells in the tumor core associated with improved outcomes. Proximity metrics showed LAG-3+ cells closer to cancer cells and PD-1+ cells relatively more distant in the core. These findings may indicate that the prognostic meaning of checkpoint expression depends on marker identity and anatomical compartment, arguing for region-aware biomarker assays.
TIGIT emerged as the most frequently expressed checkpoint in our cohort and warrants focused discussion. Transcriptomic and genomic profiling of TIGIThigh tumors revealed coordinated upregulation of TCR/BCR signaling, interferon responses, and antigen presentation alongside immunoregulatory pathways, and a distinctive genomic repertoire (enrichment of APC, CDKN2A, DIS3 mutations, and AKT2 gains among others). This molecular profile suggests that TIGIT expression reflects active and sustained T-cell engagement, where inhibitory signaling plays a compensatory role, rather than irreversible terminal T-cell exhaustion. In our data, TIGIT expression and certain spatial features correlated with modestly improved OS, consistent with a compensatory marker of antigen-engaged, partially active immunity. The literature shows context dependence for TIGIT’s prognostic role: meta-analyses across cancers often link high TIGIT expression to worse outcomes,22 yet disease-specific studies report protective associations in entities such as small cell lung cancer, lung adenocarcinoma, and chronic lymphocytic leukemia.1923 Mechanistically, combined TIGIT and PD-1 blockade can expand non-exhausted T cells and restore effector function. In line with this rationale, rilvegostomig, a bispecific antibody targeting the PD-1 and TIGIT, is currently under active clinical development in cholangiocarcinoma.24 Our observation that TIGIThigh iCCA tumors display immune activation signatures and favorable spatial T-cell positioning suggests that anti-TIGIT combinations, particularly with PD-1/PD-L1 inhibitors, merit investigation in BTC and may be especially relevant for Inflammatory Stroma tumors.
Limitations of our study include the single-center resected cohort, targeted rather than whole-TIME spatial panels, and subgroup sizes limiting power for some comparisons. In addition, immune profiling was restricted to the primary tumor and adjacent tissue compartments, including the tumor core, stroma, invasive margin, and adjacent liver tissue. While this approach provides detailed insights into the spatial organization of the local tumor immune microenvironment, it does not capture immune responses occurring in tumor-draining lymph nodes or the peripheral circulation. These compartments may provide complementary information regarding T-cell priming, systemic immune activation, trafficking, and immune exhaustion, and could further refine biomarker development for immunotherapy. Future studies integrating spatial tumor profiling with analyses of lymph node and circulating immune-cell populations will be important to better understand the interplay between local and systemic antitumor immunity in iCCA. Nonetheless, the clear spatial patterns and genotype–phenotype coupling support the robustness of our findings. Nonetheless, the clear spatial patterns and genotype–phenotype coupling support robustness. Clinically, our data argue that STIM classification plus targeted genomic profiling can segment iCCA into groups with different vulnerabilities; that spatially resolved checkpoint mapping could be incorporated into biomarker development; and that exploratory post hoc analyses of ongoing BTC clinical trials in TIGIT combination therapies should incorporate transcriptome-based stratification, such as the STIM classification.
In summary, integrating STIM, genomics, and spatial immune phenotyping reveals layered heterogeneity in iCCA that shapes prognosis and therapeutic opportunities. Moving beyond single-region bulk readouts to transcriptome-informed, spatially resolved, multimodal assays will be critical to better predict immunotherapy benefit and design rational combination strategies in iCCA.