Reprogramming the Intrahepatic Cholangiocarcinoma Immune Microenvironment by Chemotherapy and CTLA-4 Blockade Enhances Anti–PD-1 Therapy

Intrahepatic cholangiocarcinoma (ICC) is an aggressive biliary tract cancer (BTC) characterized by late clinical presentation, frequent recurrence after local therapies, and resistance to systemic therapy (1, 2). The incidence of ICC has been increasing in the United States during the last two decades (1, 3). ICC arises in the liver, and its oncogenic drivers and microenvironment differ from extrahepatic biliary cancers, which is relevant for developing new therapeutic approaches. For patients diagnosed with BTC at advanced stages, systemic chemotherapy with gemcitabine and cisplatin (GC) has been the standard care over the last decade (4–6). This combination therapy provides a short delay in progression, but new therapeutic strategies are urgently needed because of the rapid development of resistance. Molecularly targeted therapies have been approved for specific ICC subsets recurring after GC, all of which have limited efficacy (4, 5). Overall, the 5-year survival for patients with ICC is a dismal 8%. Therefore, developing novel therapeutic strategies to improve ICC treatment efficacy remains a significant unmet need.

An attractive approach to impact ICC more broadly is using immune-checkpoint blockade (ICB) therapy to reactivate and enhance antitumor immunity. The ability of ICBs—such as antibodies against programmed cell death protein (PD)-1 or its ligand PD-L1, or cytotoxic T lymphocyte antigen (CTLA)-4—to induce durable responses in advanced disease has established immunotherapy as a new pillar of cancer treatment (7). In BTC, a clinical trial of pembrolizumab—an anti–PD-1—showed an overall response rate (ORR) of 17% (8, 9). Notably, a recent randomized phase III trial (TOPAZ-1) showed that GC chemotherapy with the PD-L1 antibody durvalumab demonstrated a hazard ratio of 0.80 for overall survival (OS) and an ORR of 27% versus 19% with GC alone as a first-line treatment for advanced BTC (10). The addition of the CTLA-4 antibody tremelimumab to GC/durvalumab showed a 70% response rate in a phase II study in BTC patients (11). Similarly, combining tremelimumab with durvalumab enhances the ORR in advanced BTC patients compared with durvalumab monotherapy (12).

These breakthroughs raise critical new questions about the mechanisms of the interaction between GC and ICB and their impact on the immunosuppressive microenvironment of ICC, a key mediator of treatment resistance. Previous studies demonstrated that chemotherapeutic drugs affect the viability of cancer cells and can also exert immunostimulatory effects. Some of these effects may be mediated by targeting immunosuppressive cells, whereas others may be mediated by increased immunogenicity secondary to cancer cell killing. However, a substantial unmet need in this area of research has been the availability of animal models that reproduce the hallmarks of human ICC oncogenesis and microenvironment. It was reported that gemcitabine could deplete myeloid-derived suppressor cells (MDSC) in several tumor-bearing animals and enhance antitumor immune activity (13). Gemcitabine can also polarize tumor-associated macrophages toward antitumor phenotypes in pancreatic cancer (14). In addition to these direct immunostimulatory effects, gemcitabine, and cisplatin have been reported to enhance the antigenicity and immunogenicity of different tumors by upregulating the expression of HLA class I in cancer cells (15, 16). Moreover, the expression of PD-L1 in cancer cells can be upregulated by cisplatin in human carcinomas (17). ICC tumor growth inhibition can be induced by adding ICB to GC chemotherapy in preclinical models (18). However, how GC chemotherapy affects antitumor immunity in ICC remains unclear.

Previous studies demonstrated that, although PD-L1 is expressed in ICC cells, PD-1-expressing lymphocytes infiltrate only the fibrous septa but not in tumor lobules (19). Other findings corroborate these observations: T regulatory cells (Treg) often accumulate in ICCs, most effector CD8⁺ CTLs, and T helper cells are sequestered at the tumor margins (20). These features may explain the limited efficacy of using ICB alone in clinical trials. In addition, ICCs are poorly perfused and hypoxic, contributing to their “immunologically cold” tumor microenvironments (21–23). These factors likely mediate resistance to ICB alone, as seen in most ICC patients (24–26). These limitations have shifted the focus to revealing and targeting the critical mechanisms of cancer immune evasion that are barriers to infiltration and activation of CTLs and ICB response. The infiltration of CD8⁺ CTLs is essential for effective cancer immunotherapy (27, 28). Chemokine receptor CXCR3 is expressed primarily on activated CD8⁺ CTLs that produce perforin, granzyme, and IFNγ (29, 30). The interaction of CXCR3 and its ligands, including interferon-inducible chemokines CXCL9, CXCL10, and CXCL11, is essential for the function of T cells. Although CXCR3 is not required for CD8⁺ T-cell migration, it mediates intratumoral CD8⁺ T-cell responses to anti–PD-1 therapy (30) and chemotherapy (31).

In the current study, we used two orthotopic ICC models that show characteristic oncogenic mutations; aggressive progression in the liver and at metastatic sites (lymph node and lung); development of ascites and pleural effusions; desmoplastic tumors with hypoperfused vasculature and hypoxic microenvironment; and ICB resistance. We generated orthotopic tumors using cells established from two distinct genetically engineered mouse models: SS49 cells from Idh2^R172K/Kras^G12D mice and 425 cells from TP53^KOKras^G12D mice (22, 32). Using these immunotherapy-resistant ICC models, we evaluated the outcome of GC chemotherapy with single or dual ICB therapy, the impact of these treatments on the tumor microenvironment and antitumor immunity, and the potential mechanism of benefit of these treatments.

We used the SS49 and 425 murine ICC cell lines obtained from the Bardeesy Lab at Massachusetts General Hospital (MGH), established from spontaneous tumors in Idh2^R172K/Kras^G12D and p53^–/–Kras^G12D mice, respectively (22, 32). All cells used for experiments were passaged less than 5 times and were authenticated before in vivo use. Cells were cultured in Dulbecco's Modified Eagle Medium (DMEM; ThermoFisher) with 10% fetal bovine serum (FBS; Hyclone, SH30071.03) and 10% penicillin–streptomycin (Gibco #15070063) in 5% CO₂ at 37°C. Mycoplasma contamination was routinely performed before in vivo studies for all cell lines using MycoAlert Mycoplasma Detection Kit (Lonza #LT07-318). No genetic manipulations were performed for the cells used in this study.

Animal experiments were performed in the animal facility of MGH under specific pathogen-free conditions. All animal experiments were performed under the Institutional Animal Care and Use Committee (IACUC) at Massachusetts General Hospital–approved protocol (2014N000083). Studies complied with all guidelines outlined regarding animal research in the IACUC Policies and Guidance of MGH Research Institute.

For the therapeutic studies, orthotopic ICCs were reproducibly induced by grafting 425 cells intrahepatically in syngeneic C57Bl/6/FVB F1 mice or Rag1^–/–/C57Bl/6 mice (for the T-cell transfer studies), whereas SS49 cells were implanted in syngeneic C57Bl/6 mice. The mice were purchased from the MGH Center for Comparative Medicine (CCM). Six- to 8-week-old male and female mice were used for experiments. The suspensions of the murine ICC cells mixed with Matrigel (Corning #354234, 1:1 volume/volume with the cell suspension) were injected into the subcapsular region of the mesolimbic liver parenchyma using small 0.5 mL syringes with 28-gauge needles. To avoid leakage of tumor cells from the injection sites, which might lead to local spread and “seeding” of metastasis in the peritoneal cavity, we limited the injection volume to 20 μL (10⁶ cells in 20 μL per mouse). In addition, a steady and slow injection was performed to prevent leakage of the injected cell suspension and to minimize the damage to the surrounding liver tissues. After removing the needle, the liver surface at the site of the needle tract was covered with Gelfoam (Pfizer #9034201) for 5 minutes to reduce bleeding and potential backflow (33, 34). All treatments were initiated in mice with established tumors when the tumors reached 5 mm in diameter, measured by high-frequency ultrasound imaging. Tumor growth and treatment response were also monitored by ultrasound imaging. The p53-null (425-ICC) and Idh-mutant (SS49-ICC) murine models showed reproducible aggressive primary tumor growth and spontaneous lung metastases. Spontaneous metastatic burden was evaluated by considering the number and the size of the nodules, as previously described (33). For survival studies, moribund status was used as the endpoint per protocol, defined as symptoms of prolonged distress, >15% weight loss compared with the starting date, body condition score >2, and tumor size of >15 mm in diameter.

Tumor growth and treatment response were monitored by high-frequency ultrasound imaging. For this model's longitudinal evaluation of tumor growth, we used an ultrasound device (Vevo 2100, VisualSonics) equipped with specific probes for small-animal imaging at appropriate time points 6 days after implantation and then every 3 days (for the 425-ICC model) and 6 days (for the SS49-ICC model). Imaging to assess tumor growth noninvasively longitudinally was conducted under isoflurane anesthesia (33). The ultrasound measurement was discontinued upon the demise of over 50% of the mice in a treatment group, and the maximum tumor size is 15 mm in diameter allowed per protocol.

GC were purchased from Pfizer and Fresenius Kabi, respectively. Mouse anti–PD-1 (clone RMP-014), anti–CTLA-4 (clone 9D9), anti-CD4 (clone GK1.5), and anti-CD8b (clone 53-5.8) were purchased from Bio X Cell. GC were administered intraperitoneally (i.p., 60 mg/kg for gemcitabine and 0.3 mg/kg for cisplatin twice/week), anti–PD-1 and anti–CTLA-4 were administered by retrobulbar injection (10 mg/kg, every three days), and depleting anti-CD4 and anti-CD8b were administered by intraperitoneal injection (10 mg/kg, thrice a week).

For adoptive cell transfer studies, we enriched CD8⁺ T cells from Cxcr3^–/–/C57Bl/6 mice purchased from CCM at the MGH or nontransgenic C57Bl/6 mice to 80% purity using EasySep Mouse CD8⁺ T-Cell Isolation Kit (Stemcell Technologies #19853). Then, CD45⁺CD3⁺CD8⁺ cells, identified by flow antibodies (CD45: BioLegend, 30-F11; CD3e: BioLegend, 145-2C11; CD8a, BioLegend, 5H10-1), were sorted to 98% purity using AriaII FACS sorter in RPMI/with 10% FBS. Cells were washed with PBS twice and concentrated to 1×10⁶ cells per 100 μL. Cells were transferred via tail vein to Rag1^–/–/C57Bl/6 mice two days before the ICC implantation into the liver.

Six-μm-thick frozen sections of 425-ICC tissue were prepared for immunofluorescence (IF). We used an anti-CD31 to identify endothelial cells, an anti-α-SMA to identify perivascular cells and an anti-CD8 for staining T cells. All primary antibodies are listed in Supplementary Table S1. All secondary antibodies were purchased from Jackson ImmunoResearch. Frozen sections from O.C.T. Compound-embedded tissue blocks were washed with PBS and treated with normal donkey serum (Jackson ImmunoResearch #017000121) for blocking. Primary antibodies were applied overnight at 4°C, followed by the reaction with appropriate secondary antibodies for 2 hours at 24°C. Analysis was performed in five random fields in the tumor tissues under ×400 magnification using a laser-scanning confocal microscope (Olympus, FV-1000). Data were analyzed using ImageJ (US NIH, RRID: SCR_003070) and Photoshop (Adobe Systems Inc., RRID: SCR_014199) software.

For CD8 staining of murine 425-ICC lung metastatic tissues, we used rabbit anti-mouse CD8α (CST, #98941). Briefly, 5-μm-thick sections of murine 425-ICC lung metastatic tissues were deparaffinized in xylene for 5 minutes × 2 times, rehydrated using a graded alcohol series, and placed in a citrate buffer at 97°C for 20 minutes for antigen retrieval. The slides were kept for 30 minutes at room temperature, washed with PBS, placed in 3% H₂O₂ for 10 minutes to block the endogenous peroxide, and placed in blocking buffer with 5% normal donkey serum for 1 hour. The primary antibody against CD8α (CST, #98941) was applied overnight at 4°C, and peroxidase-conjugated anti-rabbit IgG (Abcam #ab214880) for the secondary antibody was applied for 1 hour. Finally, the sections were developed with the 3,3′-diaminobenzidine color solution (Dako #K3467) for 3 minutes at room temperature. Then, hematoxylin (Thermo Scientific #72511) was used as a chromogen, and the slides were consecutively counterstained for 1 minute. After mounting, images of the slides were taken using a bright-field microscope (Olympus, BX40) under ×100 magnification by Canon camera. Data were analyzed using Photoshop (Adobe Systems Inc., RRID: SCR_014199) software.

The sectioning and deparaffinization were the same for H&E staining of murine 425-ICC lung metastatic tissues. After hydration, hematoxylin was applied for 2 minutes to stain cell nuclei, followed by a brief rinse in water. Eosin (Sigma-Aldrich #102439) was subsequently applied for 3 minutes to stain the cytoplasm. After mounting, slides were taken images using a bright-field microscope (Olympus, BX40) under ×40 magnification by a Canon camera. Data were analyzed using Photoshop software.

Total RNA from 425-ICC tumor tissues after 8 days of treatment was isolated using Rneasy Mini Kit (Qiagen Inc.) and analyzed by the NanoDrop system. One μg of template RNA was used for reverse transcriptase PCR, and then it was diluted 100-fold to serve as the template for each well, with triplicates being used. qPCR was carried out using iTaq Universal SYBR Green Supermix (Bio-Rad, Inc.) and was amplified at the annealing temperature of 60°C with the primers (Supplementary Table S2) on an Mx3000P qPCR System (Agilent). Cxcl9, Cxcl10, Cxcl11, and Cxcr3 genes were assessed. Gapdh was used as a housekeeping gene, and the relative amount of mRNA was calculated by the 2^−ΔΔCT method.

Total RNA was extracted from the 425-ICC tissues using Qiagen kits. RNA sequencing (RNA-seq) was performed on Illumina Novaseq at the Molecular Biology Core Facilities, Dana-Farber Cancer Institute (Boston, MA). The quality control of RNA-seq raw data was performed by FastQC (RRID: SCR_014583). After quality control, the low-quality bases and adaptors contamination was removed by Cutadapt (RRID: SCR_011841). The quality of clean yield data was examined again by FastQC software. Next, the clean data were aligned to mouse reference genome mm10 by Hisat2 (RRID: SCR_015530) with a maximum of 2 mismatches per read. After data mapping, samtools (RRID: SCR_002105) and HTSeq-count (RRID: SCR_011867) were used to count the number of reads aligned to the gene features. The differentially expressed genes were identified by edgeR (RRID: SCR_012802) with a cutoff of |log(fold change)| >1 and P < 0.01. Differentially expressed genes were annotated using Gene Ontology (GO), Biocarta, Reactome, PID, and Kyoto Encyclopedia of Genes and Genomes public databases.

We obtained dissociated single cells from 425-ICC tissues from mice treated with GC alone or GC combined with dual ICB and harvested on day 10. A magnetically activated cell sorting column was used to obtain CD45⁺ and CD45^– cells, and then the cells were mixed at a ratio of 1:1 and captured using the 10X Chromium controller. After cell barcoding, reverse transcription, and cDNA amplification, amplified cDNA was used for 3′ RNA-seq library generation following the manufacturer's instructions for the Chromium Next GEM Single Cell 3′ Reagent Kit v3.1 (10X Genomics #PN-1000269). The completed single-cell RNA library was sequenced on a NextSeq 2000 (Illumina) by the MGH NextGen Sequencing Core. Data quality control was processed by Cell Ranger v7.0.0 (RRID: SCR_017344). Cells were clustered by Seurat (RRID: SCR_016341) and were annotated by the representative markers. The differentially expressed genes were identified by the FindMarkers function of the Seurat R package, and the scores of hypoxia, cytotoxicity, and exhaustion were calculated by the AddModuleScore function of Seurat. Pathway enrichment analysis of differentially expressed genes was annotated using Geno Ontology (GO) and Hallmark databases.

Cells were washed with the buffer, fixed, and permeabilized with FoxP3/Transcription Factor Staining Buffer Set (eBioscience/Thermo Fisher Scientific) to stain the intracellular markers. Harvested cells were incubated in DMEM with Cell Activation Cocktail (with Brefeldin A; BioLegend # 423303) for 4 hours at 37°C. The cells were stained with the cell-surface antibodies and intracellular marker in the buffer with brefeldin A. Anti-mouse CD16/32 (clone 93, BioLegend) was added for FcR blockade and incubated for 5 minutes at room temperature. After another washing step, antibodies for cell phenotyping were added, and cells were incubated for 40 minutes at room temperature. The monoclonal antibodies used for flow cytometry analysis were specific for CD8a (BioLegend, 5H10-1), FoxP3 (Thermo Fisher Scientific, FJK-16s), Cxcr3 (Thermo Fisher Scientific, CXCR3-173), IFNγ (BioLegend, XMG1.2), CD45 (BioLegend, 30-F11), CD3e (BioLegend, 145-2C11), and CD4 (BD Pharmingen, RM4-5). The gating strategy of flow cytometry analysis is shown in Supplementary Fig. S1.

A tissue microarray (TMA) containing 63 cores, each 1.5 mm in diameter, was constructed; 54 of the cores were representative of the 18 mouse tumors of the 425-ICC model analyzed by H&E. Cores from a normal spleen of C57Bl/6/FVB F1 mouse were also included in the TMA as controls. To perform IHC staining with mass cytometry antibodies, the TMA slide was dewaxed in xylene and rehydrated in an alcohol gradient. Slides were incubated in Antigen Retrieval Agent pH 9 (Agilent S2367) at 96°C for 30 minutes; the slide was blocked with 3% BSA in PBS for 45 minutes at room temperature, followed by an overnight stain at 4°C with the antibody cocktail listed in Supplementary Table S3. Cell-ID Intercalator-Ir (Standard BioTools PN 201192A) was used for DNA labeling. Ruthenium tetroxide 0.5% Aqueous Solution (Electron Microscopy Sciences PN 20700-05) was used as a counterstain. Images were acquired using a Hyperion Imaging System (Standard BioTools, RRID: SCR_023195) at the Johns Hopkins Mass Cytometry Facility. Upon image acquisition, multilayered ome.tiff images were generated for each core and exported using MCD Viewer (Standard BioTools, RRID: SCR_023007). Using HALO 3.2, Area Quantification FL v2.1.2 was used to quantify the area density of individual imaging mass cytometry (IMC) markers: CD8a, CD31, and Ir for DNA/Nuclear labeling. Area Quantification FL v2.1.2 was also used to quantify the colocalized area density of IMC marker combinations CD3⁺TCF1⁺, CD3⁺CD8⁺TCF1⁺, and Ir for DNA/Nuclear labeling. All area density data were normalized by nuclear area density. The resulting data were analyzed by the Kruskal–Wallis test with pairwise comparisons using GraphPad Prism (v9.2.0, RRID: SCR_002798). Single-cell data obtained from segmented IMC images were analyzed following the workflow for IMC data (35), and plots were drawn in R-4.2.2.

The χ² or Fisher test was used to compare categorical variables, and the Mann–Whitney U test was utilized to compare two groups with quantitative variables. When the experimental cohort includes more than two groups, including quantitative variables, one-way ANOVA with Tukey multiple comparisons test was applied unless specified in the figure legends. The Kaplan–Meier method generated survival curves underlying the log-rank test and Cox proportional hazard model. The hazard ratio and 95% CI were calculated for statistical survival analyses for murine models. All analyses were performed using JMP Pro 11.2.0 (SAS Institute Inc.), and data were presented as mean values ± SEM. Significant difference between experimental groups was determined when P values were less than 0.05.

Data not found in the manuscript and its associated supplementary data files are available from the corresponding author (DGD) upon reasonable request. The IMC data set has been deposited in Zenodo (DOI: 10.5281/zenodo.7439428). The bulk RNA-seq and single-cell RNA-seq (scRNA-seq) data are available in the Gene Expression Omnibus database under accession numbers GSE253275 and GSE253627. Reagents are available via materials transfer agreements.

We first evaluated the in vivo efficacy of dual ICB (anti–PD-1 and CTLA-4) combined with GC chemotherapy versus each intervention type alone using p53-null (425-ICC) and Idh-mutant (SS49-ICC) murine models (Fig. 1A and D). Mice received up to 6 injections of ICB and 9 injections of GC.

We found that all combination treatments were feasible in these models. Treatments were administered until mice became moribund, tumors reached the maximum size allowed per protocol, or tumors became undetectable by imaging. These tumors were resistant to dual ICB therapy alone; however, combining GC with dual ICB (hereafter, GC/dual ICB) significantly delayed the growth of established ICCs in both models (Fig. 1B and E; Supplementary Fig. S2A and S2B). Moreover, GC/dual ICB substantially improved median OS compared with chemotherapy alone in mice bearing these aggressive ICCs (Fig. 1C and F). In addition, GC/dual ICB treatment significantly reduced the formation of lung metastases (Supplementary Fig. S2C–S2E), despite limited infiltration by CD8⁺ T cells in the metastatic lesions (Supplementary Fig. S2F), and reduced disease-related morbidity, i.e., the incidence of bloody ascites and pleural effusions (Supplementary Fig. S2G–S2J). GC/dual ICB therapy was associated with weight loss in some mice but less than the 15% specified per protocol (Supplementary Fig. S2K and S2L). These data showed that standard GC chemotherapy could convert p53-null and Idh-mutant ICCs to dual ICB therapy‒responsive tumors and that combination therapy was feasible.

In separate experiments, we examined the effect of GC/dual ICB combination therapy on the immune microenvironment of ICC. To this end, we conducted a time-matched study in 425-ICC-bearing mice. All mice were sacrificed after 8 days of treatment with dual ICB (anti–PD-1 and CTLA-4) combined with GC chemotherapy when tumor growth was already significantly delayed compared with each intervention alone. Immune cell infiltration was measured by IF in tumor sections and flow cytometry in dissociated tumor tissues. GC chemotherapy treatment increased CD8⁺ T-cell infiltration and proliferation/activation in ICC tissue (Fig. 1G and I). Because these murine ICCs are typically hypoperfused due to blood vessel abnormalities (33), we also used IF for CD31 (an endothelial marker) and α-SMA (a perivascular cell marker) to evaluate changes in vascular density. We found that GC increased the density of vessels covered by perivascular cells, consistent with the normalization of ICC vasculature (Fig. 1H, J and K; ref. 36).

To further understand the impact of adding dual ICB to GC chemotherapy on endothelial cells, we examined transcriptomic changes by scRNA-seq of dissociated murine ICC. The analysis showed that endothelial cells from the GC/dual ICB treatment group had higher expression levels of genes related to angiogenesis, including Ets1, Kdr, and Nrp1, compared with those in the GC chemotherapy alone group (Fig. 2A and B). Overall, there was an enrichment in pathways associated with increased endothelial-cell migration and vascular endothelial growth factor receptor signaling (Fig. 2C). We previously showed that chemotherapy improved vascular function in this ICC model (33). To understand how GC alters the proportion and gene expression of other cell fractions in the ICC tumor microenvironment, we performed scRNA-seq analysis and used our published data set (ICC cells and cancer-associated fibroblasts, selected in normoxic ex vivo culture; ref. 31) as a control. We observed a marked elevation in hypoxia-related pathways in the tumors from the GC group relative to the control (Supplementary Fig. S3A–S3F). Tumors from the GC/dual ICB group had a lower hypoxia score of tumor cells compared with the GC group (Supplementary Fig. S3G). This is consistent with functional vascular normalization. Among immune cells, the combination of GC and dual ICB reduced the proportion of Apoe⁺ macrophages, which have been reported to promote immune suppression in cancer (37), whereas the proportion of neutrophils tended to increase (Supplementary Fig. S3H and S3I). Meanwhile, T cells from the GC/dual ICB group were found to have higher cytotoxicity and exhaustion scores compared with the GC alone group (Supplementary Fig. S3J–S3L), which indicated that the combination treatment could increase both the cytotoxicity and exhaustion of T cells. Thus, these data indicate that adding dual ICB to chemotherapy promoted normal vessel function and antitumor immunity in ICC.

To directly examine changes in the tumor vasculature, ICC tissues collected after 20 days of treatment with GC alone, GC/anti–CTLA-4, or GC/dual ICB were evaluated by IMC. Analyses confirmed the significant increase in CD31⁺ vessel area and the intratumoral CD8⁺ T cells in ICC tissues from the GC/dual ICB group (Fig. 2D–G). Moreover, CD8⁺CD3⁺TCF1⁺, as well as CD8⁺CD3⁺GZMB⁺ infiltration, was significantly increased in ICCs from the GC/dual ICB treated mice (Fig. 2H and I). These findings were consistent across image-based marker density analysis (Fig. 2D–I) and unbiased clustering analysis of single-cell data derived from image segmentation of IMC (Supplementary Fig. S4A–S4C).

Prior studies have demonstrated that an effective ICB (anti–CTLA-4) immunotherapy induces vascular normalization and increases the recruitment of effector T cells (38). In addition, others have shown that effective antitumor immunity (including that induced by PD-1 blockade) and vascular normalization are reciprocally regulated (39). Thus, our data show that GC/dual ICB therapy can increase CTL infiltration and proliferation and normalize the vasculature in “immunologically cold” and hypoperfused ICCs.

Next, we tested the specific roles of CTLA-4 and PD-1 blockade in the efficacy of combined GC/dual ICB therapy in ICC. To this end, we evaluated in mice with established orthotopic murine 425-ICC the efficacy of treatment with (i) GC and dual ICB (anti–PD-1 and anti–CTLA-4), (ii) GC with anti–PD-1, (iii) GC with anti–CTLA-4, (iv) GC chemotherapy alone, (v) dual ICB alone, (vi) anti–PD-1 alone, (vii) anti–CTLA-4 alone, or (viii) isotype-matched IgG control. All agents were administered until mice became moribund, tumors reached the maximum size allowed per protocol, or tumors became undetectable by imaging. We found that GC treatment combined with anti–PD-1 therapy induced a growth delay and survival advantage that was not superior to GC alone in this model. In contrast, the combination of GC and anti–CTLA-4 treatment caused a significant delay in tumor growth and an increase in median OS compared with GC alone treatment, although its efficacy was inferior to GC with dual ICB in this ICC model (Fig. 3A and B; Supplementary Fig. S5A).

Prolonged, continuous administration of GC/dual ICB combination therapy induced toxicity in this experiment, including weight loss leading to treatment breaks in some mice (Supplementary Fig. S5B). The effects of treatment on the inhibition of lung metastases were significant only after combination therapy of GC with anti–CTLA-4 or GC with dual ICB (Supplementary Fig. S5C). Similarly, disease-related morbidity was reduced in all GC-treated groups and, to the greatest extent, in the GC/dual ICB group (Supplementary Fig. S5D and S5E). These data demonstrate that CTLA-4 blockade is essential for the efficacy of combination GC/dual ICB treatment in this ICC model, which is resistant to immunotherapy alone and can render these tumors responsive to anti–PD-1 therapy.

Next, to decipher the mechanism of benefit of GC/dual ICB therapy, we first repeated the time-matched experiments to evaluate the effects of combination therapy by immune profiling of responding murine 425-ICC tumors at day 20, when the differences in growth delay were significant between groups (Supplementary Fig. S6A). In this experiment, ICC-bearing C57Bl/6/FVB F1 mice were treated with (i) GC alone, (ii) dual ICB alone, (iii) GC plus anti–CTLA-4, (iv) GC combined with dual ICB, or (v) IgG control. Mice from the control group and dual ICB alone had to be sacrificed on day 10 due to rapid tumor progression; these tumor tissues were collected and used as a reference.

To evaluate the overall frequency of tumor-infiltrating lymphocytes (TIL), we counted the T-cell receptor (TCR)⁺ cells by flow cytometry in enzymatically digested ICC tissues. We found significantly higher frequencies of TILs in all GC-treated groups; moreover, there was a significant increase in the fraction of TILs after GC with anti–CTLA-4 and GC with dual ICB versus GC alone (Fig. 3C; Supplementary Fig. S6B). Among TILs, approximately 60% were CD8⁺ T cells in all groups (Fig. 3D; Supplementary Fig. S6C). Moreover, the fraction of CD8⁺ IFNγ⁺ T cells was higher in the tumors from GC-treated groups and was significantly increased after GC/dual ICB (P<0.05 vs. GC alone; Fig. 3E; Supplementary Fig. S6D). Notably, the fraction and the number of Ki67⁺CD8⁺ T cells were significantly higher in tumors after GC/anti–CTLA-4 and GC/dual ICB treatment (Fig. 3F; Supplementary Fig. S6E). Ki67⁺CD4⁺ conventional T cells also increased after GC/anti–CTLA-4 and GC/dual ICB treatment. However, the frequency of CD4⁺ cells was substantially lower than that of CD8⁺ T cells (Supplementary Fig. S6F).

Among all markers of T-cell activation tested (IL2, IL12, IFNγ, and Cxcr3), we found that the frequency and number of CD8⁺Cxcr3⁺ and CD8⁺Cxcr3⁺IFNγ⁺ T cells were significantly increased in tumors treated with GC/dual ICB compared with GC alone or GC combined with anti–CTLA-4 (Fig. 3G and H; Supplementary Fig. S6G and S6H). We also found significant increases in CD4⁺FoxP3⁺ Tregs, resulting in comparable CD8⁺ T cell to Treg ratios among GC-treated groups, but the absolute counts of Tregs were low (Supplementary Fig. S7A–S7D). Further phenotyping of CD8⁺ T cells showed that GC/dual ICB therapy reduced the markers of immunosuppression (markers associated with CD8⁺ T-cell exhaustion), including PD-1, TIGIT, Tim3, GITR, Vista, and Lag3, and increased markers that indicate high immunity such as CD44, marking effector function, and CD69, an early activation marker (Supplementary Fig. S7E).

To determine the impact of anti–PD-1 therapy, we analyzed the immune microenvironment of 425-ICC tumors treated with control, anti–PD-1, GC, GC/anti–PD-1, and GC/dual ICB. Given the difference in tumor growth, we collected tumors of control and anti–PD-1 on day 10 and for GC-containing groups on day 20, consistent with the prior experiment. Including a GC/anti–PD-1 group enabled us to discern whether combining GC with dual ICB treatment markedly elevated the prevalence of Cxcr3⁺CD8⁺ T cells compared with GC/anti–PD-1 alone (Supplementary Fig. S7F). Anti–PD-1 therapy significantly increased the frequency of CTLA-4⁺PD-1^– cells among conventional CD4⁺ T cells compared with IgG or GC alone (Fig. 3I). Addition of CTLA-4 blockade to the GC and anti–PD-1 regimen resulted in a nonsignificant reduction in the fraction of CTLA-4⁺PD-1^–CD4⁺ T cells compared with GC/anti–PD-1, but a significant decrease in the proportion of CTLA-4⁺PD-1⁺ conventional CD4⁺ T cells and Tregs compared with GC alone (Fig. 3J and K; Supplementary Fig. S7G and S7H). In addition, we found that GC/dual ICB treatment significantly reduced the frequency of Ly6C^loCD11b⁺ monocytic myeloid-derived suppressor cells (M-MDSC) compared with control or anti–PD-1 alone (Supplementary Fig. S7I). These data suggest a potential alleviation of immune suppression with the combined treatment.

We further validated our findings in the SS49-ICC model, where we observed that the combination of GC with both anti–CTLA-4 and anti–PD-1 therapies significantly increased the proportion of TCF1⁺CD8⁺ T cells compared with the combination of GC with anti–PD-1 alone. This emphasizes the therapeutic advantage of adding anti–CTLA-4 to the regimen in enhancing the effectiveness of the treatment (Supplementary Fig. S7J).

These results indicate that adding CTLA-4 blockade to GC increases the number of activated CTLs and that adding PD-1 blockade further increases the fraction of CD8⁺Cxcr3⁺IFNγ⁺ activated T cells in ICC tissue. They also raise the question of whether the expression of Cxcr3 ligands changes after GC/dual ICB treatment.

Using tissues collected in a time-matched manner, we used RNA-seq to evaluate the transcriptional changes in murine 425-ICC tissues after 8 days of treatment with GC alone, dual ICB alone, and GC/dual ICB versus IgG control. Bulk RNA-seq analysis showed significant increases in the expression of over 1,200 genes in the GC/dual ICB group compared with the other groups (Supplementary Fig. S8A and Supplementary Data set S1). Gene enrichment analyses showed significant activation of pathways related to cellular immunity (Supplementary Fig. S8B). Of these, the transcription of Cxcl9, Cxcl10, and Cxcl11, as well as their receptor Cxcr3, increased after GC/dual ICB combination therapy versus control (n = 3; Supplementary Fig. S9A–S9D). Activating this chemokine pathway is critical for CTL function and response to anti–PD-1 therapy (30, 40, 41). We used qPCR analysis to validate the increase in Cxcl10 and Cxcl11 expression and found a trend for increased Cxcl9 and Cxcr3 expression at this early time point (n = 5; Supplementary Fig. S9E–S9H).

Next, we performed depletion experiments to determine whether CD4⁺ or CD8⁺ T cells mediate the benefit of GC/dual ICB combination therapy (Supplementary Fig. S10A). CD4⁺ cell depletion did not affect median OS after either GC/anti–CTLA-4 or GC/dual ICB therapy, whereas CD8⁺ cell depletion completely compromised the efficacy of both treatments (Fig. 3L and M). Median OS was 42 days in the GC/anti–CTLA-4 group versus 31 days in the survival of GC/anti–CTLA-4/anti-CD8β group (P = 0.0046) and 45 days in the GC/dual ICB group versus 30 days in the GC/dual ICB/anti-CD8β group (P = 0.0048).

Next, we used the adoptive T-cell transfer model to establish whether Cxcr3⁺CD8⁺ T cells mediated the efficacy of GC/dual ICB therapy in the murine 425-ICC model. We first transferred one million CD8⁺ T cells from Cxcr3^+/+/C57Bl/6 or Cxcr3^–/–/C57Bl/6 to Rag1^–/–/C57Bl/6 mice, which lack functional T cells. Two days after the T-cell transfer, we orthotopically implanted murine 425-ICC cells in the Rag1^–/–/C57Bl/6 mice. When tumors were established, the mice received GC/dual ICB therapy or GC alone, and tumor growth was monitored by ultrasound imaging (Supplementary Fig. S10B).

We found that ICC growth after GC therapy was not different between mice receiving CD8⁺ T-cell transfer from Cxcr3^+/+/C57Bl/6 versus Cxcr3^–/–/C57Bl/6 donor mice. In contrast, the efficacy of GC/dual ICB therapy was significantly diminished in mice that received CD8⁺ T cells from Cxcr3^–/–/C57Bl/6 versus those received CD8⁺ T cells from Cxcr3^+/+/C57Bl/6 mice, with inferior tumor growth delay and survival benefits (Fig. 4A, B; Supplementary Fig. S11A). Treatment was not associated with significant weight loss in any groups (Supplementary Fig. S11B). Moreover, the metastatic lung burden was lowest in mice treated with GC/dual ICB and received CD8⁺ T-cell transfer from Cxcr3^+/+/C57Bl/6 mice (Supplementary Fig. S11C). Finally, unlike the other treatment groups, these mice were also free of bloody ascites and pleural effusions (Supplementary Fig. S11D and S11E). These data show that the benefits of GC combined with dual ICB partly depended on activated CD8⁺Cxcr3⁺ T cells.

The GC/dual ICB combination showed efficacy but was associated with adverse effects in continuous administration. Thus, we used the murine 425-ICC orthotopic model to test the effect of treatment deescalation based on scheduling the agents to leverage their mechanisms of action on efficacy. To this end, mice with established tumors were randomized into one of the following treatment groups: (i) GC until endpoint (maximum chemodrugs exposure); (ii) GC with anti–CTLA-4 and anti–PD-1 until endpoint (maximum chemo- and ICB exposure); (iii) GC until endpoint with 1 week of anti–CTLA-4 and maintenance anti–PD-1 from day 8 to endpoint (chemo/CTLA-4 blockade “priming”/reduced ICB drug exposure); and (iv) GC until endpoint with anti–PD-1 for the first week and anti–CTLA-4 from day 8 to endpoint (chemo/PD-1 blockade “priming”/reduced ICB drug exposure; Supplementary Fig. S12A). “Priming” with anti–CTLA-4 followed by GC/anti–PD-1 (group III) achieved the same OS benefit compared with maximum chemotherapy and ICB drug exposure (group II). Efficacy was achieved despite a lesser primary tumor growth delay because of equivalent antimetastatic effects and reduced ascites and pleural effusions (Fig. 4C and D; Supplementary Fig. S12B–S12F).

Over the last decade, ICBs have revolutionized systemic therapy for cancer after inducing durable responses in some malignancies. Unfortunately, these cases represent a minority (typically 20%–30%) for most tumor types (42). Using orthotopic ICC models, we show that GC chemotherapy converts ICCs from “immunologically cold” tumors into “hot” tumors by increasing effector CD8⁺ T-cell recruitment facilitated by induction of normalization of ICC vasculature. These effects enhanced the efficacy of immunotherapy with ICB via recruitment of activated CD8⁺Cxcr3⁺ T cells, leading to increased survival and reduced morbidity in two aggressive orthotopic models of ICC. Our study reveals that standard chemotherapy for ICC facilitates anti–CTLA-4‒based ICB therapies and the mechanisms of benefit for these treatment interactions (Fig. 4E).

Vascular normalization by GC chemotherapy combined with ICB in ICC may be indirectly mediated by killing cancer cells, the key source of proangiogenic factors. This concept was first demonstrated by hormone withdrawal in a hormone-dependent tumor model (43). Th1 cells were identified to play a crucial role in the reciprocal regulation of vessel normalization and antitumor immunity (39). GC treatment also increases the accumulation of CD8⁺ T cells and, when combined with dual CTLA-4 and PD-1 blockade, showed durable responses mediated partly by activation of CD8⁺Cxcr3⁺ CTLs. IMC analyses confirmed the increased vascularization and T-cell infiltration, associated with a reduction in the number of cancer cells in the murine ICC after GC and dual ICB combination treatment, consistent with IF and flow-cytometric studies.

CTLA-4 blockade played a critical role when combined with GC in tumor responses and induced an increase in activated CD8⁺ T cells despite the persistence of Tregs, which were comparatively less frequent. Given the recent data showing the feasibility and efficacy of GC chemotherapy with anti–PD-L1 therapy in BTC, the approach tested here is immediately applicable as combination therapy in ICC. The concept of using combination therapy to address resistance to anti–PD-1/PD-L1 therapy is widely accepted and is actively pursued in clinical studies (44–46). However, its implementation is limited by an incomplete understanding of treatment interactions in specific tumor contexts, serious toxicity concerns, and high economic costs. Our study demonstrates the efficacy of mechanism-based multimodality therapy for ICC, which minimizes drug exposure for the reagents used in combined strategies while maintaining optimal effectiveness.

Although early studies of Cxcr3 focused on its role in driving the recruitment of activated T cells into inflamed tissues, Cxcr3 was identified as an activation marker on T cells (30, 47–50). For CD8⁺ T cells, Cxcr3 might be involved in the directed migration of these cells and their activation and proliferation in tumors (51). Our study demonstrates that the reprogramming of the ICC immune microenvironment by chemotherapy and the enhancement of CTLA-4 blockade of anti–PD-1 treatment efficacy depends partly on Cxcr3⁺CD8⁺ T cells. Our immune profiling results show that, whereas CTLA-4 blockade alone did not significantly enhance the T-cell infiltration induced by GC chemotherapy, it increased the fraction of proliferating CD8⁺ T cells and treatment efficacy. Moreover, when added to GC/anti–PD-1 therapy, CTLA-4 blockade significantly increased the infiltration and activation of CD8⁺ T cells in ICC. Although ICC infiltration by CD4⁺ conventional T cells also increased after GC/dual ICB treatment, a similar trend was seen for Tregs. Depletion experiments demonstrated that treatment efficacy depended on CD8⁺ and not CD4⁺ T cells. These results are consistent with a comparatively lower frequency of CD4⁺ versus CD8⁺ T cells in the ICC tissues.

Among the CD8⁺ T cells subsets infiltrating ICC tissue after GC/dual ICB treatment, our IMC analyses showed an increase in CD8⁺TCF1⁺ and CD8⁺GZMB⁺ T cells, known to be associated with the activation and recruitment of CD8⁺Cxcr3⁺ T cells (50). Moreover, we found an increased frequency of CD44⁺ and CD69⁺ CD8⁺ T cells in the ICC microenvironment, consistent with early activation of effector T cells. CD8⁺ T-cell expansion seen after the addition of CTLA-4 blockade alone to GC was also associated with the expression of cell exhaustion markers, including PD-1, TIGIT, Tim3, and Lag3. Importantly, when GC was combined with dual CTLA-4/PD-1 blockade, expression of these immunosuppressive markers decreased on the infiltrating CD8⁺ T cells. These results demonstrate the role of CTLA-4 blockade and the importance of PD-1 blockade in efficacious ICB therapy with GC in ICC. Although PD-1 can partially enhance cytotoxicity, there is a subsequent increase in the exhaustion of T cells and other immune checkpoints such as CTLA-4, potential targets to overcome adaptive resistance. Moreover, our results indicate that the rational scheduling of the ICBs (anti–CTLA-4 “priming”) with GC chemotherapy and anti–PD-1 therapy may achieve equivalent efficacy with continuous dosing while reducing overall drug exposure in ICC patients. Although CTLA-4 and PD-1 serve as immune-checkpoint molecules, they act at different phases of the immune response (52). CTLA-4 blockade may be effective by priming antitumor immunity and expanding the infiltrated CD8⁺ T-cell pool. Conversely, anti–PD-1 priming may alleviate the inhibition of T cells already activated within the tumor microenvironment, which may be insufficient in ICC.

Although our study has demonstrated the pivotal role of CD8⁺ T cells in the context of GC/dual ICB combination therapy, further studies are needed to explore the function of myeloid cells. We detected a reduction in M-MDSC frequency, and scRNA-seq and IMC data pointed to a potential role of macrophages in ICC. Another limitation of our study is that the two murine ICC models carry the KRAS^G12D mutation in addition to Idh2 and p53. Although this suggests that GC/ICB may be effective against ICCs with this mutation, further validation across disease subtypes is needed in clinical studies.

In summary, we demonstrate that GC chemotherapy alone can normalize the vessels and increase activated T-cell infiltration in ICC. Combining GC with anti–PD-1/CTLA-4 immunotherapy was feasible and showed efficacy in murine ICCs resistant to ICB or GC/anti–PD-1 therapy. These new insights into chemotherapy/ICB treatment interactions may directly inform clinical trials for ICC designed to increase efficacy while reducing drug exposure.

Z. Amoozgar reports that she is currently an employee of Sanofi while maintaining a nonconflict relationship with MGH as a nonpaid employee and collaborator. This work is neither funded nor related to any work at Sanofi. S.M. Shin reports nonfinancial support from Standard BioTools outside the submitted work. L. Shi reports grants from NCI and the Cholangiocarcinoma Foundation during the conduct of the study. L.L. Munn reports grants from NIH during the conduct of the study and other support from SimBiosys outside the submitted work. D. Fukumura reports grants from MGH during the conduct of the study. N. Bardeesy reports grants from Servier Laboratories outside the submitted work. W.J. Ho reports other support from Rodeo/Amgen, Exelixis, and Standard BioTools, and grants from Sanofi and NeoTX outside the submitted work. R.K. Jain reports other support from Enlight Biosciences and Accurius Therapeutics, personal fees from SynDevRx, SPARC, Elpis Biopharmaceuticals, Tekla Life Sciences Investors, Tekla Healthcare Investors, Tekla World Healthcare Fund, Tekla Healthcare Opportunities Fund, Innocoll Pharmaceuticals, 28-seven Incorporated, Bristol Meyers Squibb, Cur Therapeutics, and Merck, grants from Niles Albert Research Foundation, Boehringer-Ingelheim, Sanofi, Jane's Trust Foundation, Ludwig Center at Harvard, and grants from U.S. National Cancer Institute outside the submitted work. D.G. Duda reports grants from Bayer, Exelixis, BMS, and Surface Oncology, and personal fees from Innocoll outside the submitted work. No disclosures were reported by the other authors.

J. Chen: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. Z. Amoozgar: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. X. Liu: Conceptualization, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. S. Aoki: Conceptualization, investigation, methodology, writing–review and editing. Z. Liu: Investigation, writing–review and editing. S.M. Shin: Investigation, methodology, writing–review and editing. A. Matsui: Investigation, methodology, writing–review and editing. A. Hernandez: Investigation, methodology, writing–review and editing. Z. Pu: Investigation, writing–review and editing. S. Halvorsen: Investigation, methodology, writing–review and editing. P.-J. Lei: Investigation, methodology, writing–review and editing. M. Datta: Investigation, methodology, writing–review and editing. L. Zhu: Investigation, writing–review and editing. Z. Ruan: Investigation, writing–review and editing. L. Shi: Methodology, writing–review and editing. D. Staiculescu: Investigation, writing–review and editing. K. Inoue: Investigation, writing–review and editing. L.L. Munn: Methodology, writing–review and editing. D. Fukumura: Methodology, writing–review and editing. P. Huang: Methodology, writing–review and editing. S. Sassi: Investigation, methodology, writing–review and editing. N. Bardeesy: Investigation, methodology, writing–review and editing. W.J. Ho: Investigation, methodology, writing–review and editing. R.K. Jain: Conceptualization, investigation, methodology, writing–review and editing. D.G. Duda: Conceptualization, resources, supervision, funding acquisition, methodology, writing–original draft, project administration, writing–review and editing.

This study was supported by NIH grants R01CA260857, R01CA254351, R03CA256764, and P01CA261669 (D.G. Duda), NIH grant R01CA247441 (L.L. Munn and D.G. Duda), Department of Defense PRCRP grants W81XWH-19-1-0284 and W81XWH-21-1-0738 (D.G. Duda), NIH grant K22CA258410 (M. Datta), NIH grant P30CA006973 (W. Ho), NIH grant U01CA224348, R01CA259253, R01CA208205, R01NS118929, U01CA261842 (R.K. Jain), and Cholangiocarcinoma Research Foundation postdoctoral fellowship (S. Aoki).