Multiomics prediction and immunogenic validation of personalized neoantigens in cholangiocarcinoma patients Authors: Baoluhe Zhang, Bao Jin, Xueshuai Wan, Xiao Liu, Xiang’an Wu, Jiali Xing, Yuxin Wang, Haifeng Xu, Yiyao Xu, Mei Guan, Yilei Mao, Yating Bai, Junqi Zhu, Qian Chen, Shunda Du Published: 2026 ABSTRACT BackgroundCholangiocarcinoma (CCA) is an aggressive biliary tract cancer with limited treatment options, underscoring the need for breakthrough precision therapies. Neoantigen-based therapy is a promising novel strategy for cancer treatment. This study investigates the potential of neoantigen peptides in CCA, emphasizing their role in personalized immunotherapy and providing insights for vaccine design.MethodsPaired samples from 33 CCA patients underwent whole-exome sequencing and RNA sequencing to profile the mutational and neoantigen landscape. Besides, personalized neoantigen peptides were synthesized and immunogenicity was validated through enzyme-linked immunospot assay. Meanwhile, corresponding T cell receptor (TCR) sequences were predicted by bioinformatics.ResultsTP53 was the most frequently mutated gene (57.58%) and a significant positive correlation between tumor mutational burden and neoantigen load was observed (r = 0.6511, p < 0.0001). A total of 798 neoantigens were identified in 22 samples, predominantly characterized by human leukocyte antigen (HLA) A02:07 and A02:01. Additionally, 16 neoantigen peptides from three CCA patients were successfully synthesized and these peptides induced robust T-cell responses, with the most potent epitopes derived from UBN1, C2orf47 and ITPR2. Moreover, bioinformatic analysis further predicted in silico putative TCR sequences potentially associated with the candidate neoantigens. In Pt01, the Vβ5-6Jβ1–2 sequence had the highest predicted affinity; in Pt02, the Vβ2Jβ1–4 sequence had the highest predicted affinity; and in Pt03, the Vβ29-1Jβ2–7 sequence had the highest predicted affinity. The sequence motifs of the high-frequency TCR clusters were also demonstrated.ConclusionsOur findings show that candidate immunogenic neoantigen peptides can be identified in CCA and are supported by proof-of-concept functional validation. Together, these results provide a framework for the downstream validation of neoantigen targets and the development of personalized immunotherapy approaches.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12967-026-08471-w. Background Cholangiocarcinoma (CCA) is a formidable and perilous malignancy, originating from the intricate network of epithelial cells in both the intrahepatic and extrahepatic biliary duct system [1]. The manifestation of CCA depends on the complex interplay of underlying diseases and diverse anatomical locations, emerging from the epithelial cells lining the bile ducts, peribiliary glands, or even hepatocytes [2]. The tumor microenvironment (TME) of CCAs is marked by a regrettable lack of immunogenicity, fostering the proliferation of immunosuppressive cell types at an elevated level [3, 4]. Furthermore, the abundant presence of cancer-associated fibroblasts significantly contributes to the desmoplastic nature of CCA [5]. Within the complex immunosuppressive milieu of TME, CCAs exhibit pronounced resistance to conventional chemotherapy and demonstrate limited efficacy for immune checkpoint inhibitor (ICI) monotherapy. Hence, the urgency of formulating innovative and alternative therapeutic strategies for cholangiocarcinoma cannot be overstated [6]. The intricate landscape of tumor neoantigens, arising from mutated genes within tumor cells, gives rise to abnormal proteins. These proteins undergo degradation into peptides, a cascade that culminates in their recognition by the cellular antigen presentation system (major histocompatibility complex, MHC). This process transforms them into distinctive surface markers specific to tumor cells [7]. The evolution of next generation sequencing (NGS) technology has ushered in an era where increasingly refined mutation analysis algorithms facilitate the identification of diverse mutations in NGS data [8]. This, in turn, empowers the prediction of neoantigens, yielding more precise and granular insights into the realm of tumor neoantigens [9]. In previous reports, the realm of neoantigen-specific vaccines has witnessed notable success in treating various cancers, including melanoma [10–12], gastric cancer [13], lung cancer [11, 14], rectal cancer [15] and hepatocellular carcinoma [16, 17]. However, the efficacy and safety of neoantigen-based immunotherapy in CCA patients—particularly within the Chinese population—remain largely unexplored, highlighting a promising area for future research. In this study, we performed whole-exome sequencing (WES) and RNA sequencing on paired tumor and blood samples from Chinese CCA patients, systematically delineated their somatic mutation and neoantigen landscape, and demonstrated a strong association between tumor mutational burden (TMB) and neoantigen load. We then synthesized a subset of high-priority neoantigenic peptides and confirmed their capacity to elicit autologous T-cell responses by interferon-gamma (IFN-γ) enzyme-linked immunospot (ELISPOT) assay. Furthermore, we performed motif-based analysis of T cell receptor (TCR) repertoires to explore putative sequence patterns associated with high-affinity neoantigen candidates, generating hypotheses for future studies of neoantigen-reactive T-cell responses. Together, this work aims to establish an integrative framework for neoantigen identification in CCA and to provide candidate targets and exploratory TCR features for subsequent experimental validation in the context of precision immunotherapy. Methods WES and somatic mutation calling Tumour and matched normal DNA were extracted using the TIANamp Genomic DNA Kit (DP304, TIANGEN, Beijing, China) from fresh tumour tissue and paired blood sample. Libraries were constructed by the Agilent SureSelectXT Human All Exon V6 Kit (5190–8864, Agilent Technologies, Santa Clara, USA) and sequenced with next-generation sequencing. Genomic DNA was fragmented, end-repaired, adenylated at the 3’ ends, end-connected, amplified, purified, and size-selected in the process of library construction, then was sequenced on the Illumina X10 platform (Illumina Inc., San Diego, CA, USA). For the WES data, sequencing reads containing too many Ns (> 5%), low-quality bases (> 15% bases with quality ≤ 19) and adaptor sequences were discarded. Then high-quality paired-end reads were aligned to the human reference genome (hg19) by Burrows-Wheeler Aligner (BWA, version 0.7.15, BWA-MEM algorithm). For each paired sample, somatic single-nucleotide variant (SNV) and insertion/deletion (InDel) mutations were simultaneously analyzed using Mutect2 from the Genome Analysis Toolkit (GATK, version 4.1) in tumor–normal paired analysis mode. And the candidate somatic mutations were filtered using FilterMutectCalls to remove sequencing artifacts and germline polymorphisms. Then all detected mutations were annotated using ANNOVAR (version 2016-02-01). The following criteria were utilized to further filter somatic variants based on the ANNOVAR annotation files: (a) retaining “PASS” items in the “Filter” column; (b) retaining mutations that mutated in exonic or splicing region; (c) deleting “synonymous SNV” and “unknown” items in the “ExonicFunc.refGene” column; (d) retaining mutations that not included in public 1000 Genomes databases or the frequency of the alternative allele was < 0.01. The remaining mutations were identified as somatic mutations and were subsequently used for further analysis. RNA-seq and gene expression Tumour RNA was extracted using the TRIzol® Reagent (15596018, Invitrogen, Burlington, USA) from fresh tissue. Libraries were constructed using a NEBNext® UltraTM II RNA Library Prep Kit (#E7775, NEB, MA, USA) and sequenced with NGS. Total RNA was fragmented, reverse transcribed into complementary DNA, base ‘A’ added in the 3’ ends, adapter connected, amplified and purified, and then sequenced on Illumina X10 platform (Illumina Inc., San Diego, CA, USA). Clean reads were mapped to the hg19 human genome using Bowtie 2 (version 2.2.4) software from Tophat 2 (version 2.0.10) with default parameters. The program Cufflinks (version 2.2.1) was used to calculate the expression levels of genes in terms of fragments per kilobase million (FPKM). Prediction of neoantigens and corresponding T cell receptor (TCR) sequences HLA class I typing was performed using OptiType (version 1.3.1) based on WES data from normal tissue, using default parameters. Neoantigen prediction was conducted using the pVACseq pipeline (version 4.0.10). Peptide–MHC binding affinity was assessed using both NetMHC (version 2.22.1) and PickPocket (version 2.22.1), with binding strength quantified by half-maximal inhibitory concentration (IC50) using default parameters. In general, peptides with IC50 < 500 nM were considered potential binders, and peptides with IC50 ≤ 150 nM were classified as strong binders. Only peptides derived from genes with an expression level of ≥ 1 FPKM were considered for downstream analysis. TCR–neoantigen binding prediction was performed using the ImRex platform (https://github.com/pmoris/ImRex), a deep learning–based tool trained on experimentally validated TCR–peptide–MHC interaction datasets.Validated neoantigen peptide sequences were input into ImRex, and the software predicted TCR β-chain sequences with high binding affinity based on structural and sequence features. Candidate TCRs were ranked by predicted interaction probability scores, and the top-ranking TCR sequences for each neoantigen were selected for downstream analysis. The output included complementarity-determining region 3 (CDR3) β sequences and their corresponding V–J gene segments. These results were used to identify potential neoantigen-specific TCR clonotypes and assess their therapeutic relevance. Exploratory identification of tumor-associated antigen (TAA) candidates To explore potential TAA candidates in CCA, tumor RNA-seq data from our CCA patients were compared with 262 normal liver RNA-seq data obtained from the GTEx V11 database (gene_reads_v11_liver.gct.gz). Differentially expressed genes (DEGs) analysis was performed using DESeq2 (version 1.36.0). Genes met with the following criteria were deemed significant DEGs: the fold change of expression ≥ 1.5 and P-value < 0.05. Then the representative overexpressed DEGs in CCA tumors with potential clinical implementability were focused, including 13 antibody–drug conjugate (ADC) targets and 224 cancer-testis antigen (CTA) genes (Detailed gene list are provided in Additional file 1: Table S1). Immunogenicity, hydrophilicity and toxicity prediction of neoantigens peptides To better understand potential immunogenic features and practical feasibility of selected candidate neoantigen peptides, the immunogenicity-related and physicochemical properties were further characterized prior to synthesis, including antigenicity, IFN-γ–induction potential, hydrophobicity and hydrophilicity, and toxicity. The antigenicity of peptides was predicted by VaxiJen (version 2.0), and peptides with a score ≥ 0.4 were considered antigenic. The potential to stimulate IFN-γ production was evaluated using the IFNepitope server (https://webs.iiitd.edu.in/raghava/ifnepitope/predict.php), and peptides with a prediction score > 0 were classified as IFN-γ positive. The hydrophobicity and hydrophilicity of the peptides were analyzed using the Expasy Protpram online server tool (https://web.expasy.org/protpram), and a score ≥ 0 was interpreted as hydrophobic, whereas a score < 0 was interpreted as hydrophilic. Finally, potential toxicity was predicted using the ToxinPred server tool (https://webs.iiitd.edu.in/raghava/toxinpred/multi_submit.php), and peptides with a score ≤ 0 were classified as non-toxic, while those with a score > 0 were classified as toxic. Peptide synthesis The candidate neoantigen peptides and their corresponding wild-type peptides were synthesized as crude peptides (28–30 amino acids in length) by Bankpeptide biological technology Co., Ltd. (Hefei, China), with a purity greater than 95% and endotoxin levels below 0.1 EU/µg. For peptides derived from frameshift mutations, longer sequences were synthesized to ensure accurate epitope presentation. Each mutant amino acid was flanked by approximately 13–14 amino acids from the wild-type sequence. Peptides were dissolved in dimethyl sulfoxide at a concentration of 1 mg/mL and used at a final concentration of 2 µg/mL in the ELISPOT assay. ELISPOT assay Three patients were selected for functional validation with longitudinal follow-up blood samples available, providing sufficient PBMCs for immunogenicity testing. Prior to the ELISPOT assay, patient-derived PBMCs were expanded by in vitro stimulation with synthetic peptides. PBMCs were seeded at 5 × 106 cells/well in 96-well plates in a total volume of 1 mL AIM-V medium supplemented with 2% human AB serum (100–512, Gemini Bio, Woodland, USA) and 100 U/mL penicillin–streptomycin, and cultured at 37 °C with 5% CO₂. Synthetic peptides were added at a final concentration of 2 µg/mL. Recombinant human interleukin-2 (20 U/mL) and interleukin-7 (20 ng/mL) were added starting on day 1 and replenished every 3 days. Cultures were maintained with half-medium replacement every 3 days, with cytokines restored to the indicated final concentrations after each medium change. On day 10, the specific T cell responses to each peptide were evaluated by ELISPOT analysis. Subsequently, the Human IFN-γ ELISpotBASIC kit (HRP) (3420–2 H, Mabtech, Nacka Strand, Sweden), containing capture antibody 1-D1K, biotinylated detection antibody 7-B6-1 and Streptavidin-HRP were used for IFN-γ testing according to the manufacturer’s instructions. AEC Staining Kit (AEC101-1KT, Sigma-Aldrich, New Jersey, USA) was used for color development. Specifically, nitrocellulose-backed multiscreen ELISPOT plates (MSIPS4510, Millipore, Boston, USA) were pre-wet with 15 µL 35% ethanol for 1 min. Plates were washed 5 times with sterile water 200 µL/well. IFN-γ capture antibody (1-D1K, 15 µg/mL, Mabtech 3420–2 H) was plated at 100 µL/well. Plates were incubated at 4 ℃ overnight. On the second day, plates were washed again with PBS (5 times with 200 µL/well) and then plated with 200 µL/well medium containing 10% Human serum AB (100–512, Gemini Bio, Woodland, USA). Plates were allowed to incubate for 2 h at room temperature. Medium were discarded, and expanded PBMCs were added at 2.5 ~ 5 × 104 cells/well with a total volume of 100 µL/well (peptides were added at a final concentration of 2 µg/mL). Then plates were then incubated for 24 h at 37℃ at 5% CO2. On day 3, cells were removed from ELISPOT plates. Then plates were washed with PBS for 5 times and 100 µL/well of biotinylated anti-human IFN-γ detection antibody (7-B6-1-biotin, final concentration 1 µg/mL) was added into plates. Plates were incubated for 2 h at room temperature, then washed with PBS buffer for 5 times in 200 µL/well. Streptavidin-HRP was diluted (1:100) in PBS with 0.5% Human serum AB (100–512, Gemini Bio, Woodland, USA), then added to each well (100 µL/well). Plates were incubated for 1 h at room temperature and then washed 5 times with PBS. 100 µL/well substrate solution was added into plates and plates were incubated in the dark at room temperature for 10 min [18]. Development was stopped by washing the plate extensively with sterile water. All operations are conducted in a sterile environment. Water was discarded vigorously off, and the plates were left to dry. Spots were inspected using a dissection microscope (GP-300 C, KSGAOPIN, Kunshan, China). Immunohistochemistry (IHC) IHC was performed to assess the expression of four immune checkpoint molecules in tumor tissues, including cytotoxic T lymphocyte antigen 4 (CTLA4), programmed cell death protein 1 (PD1), T cell immunoglobulin and mucin-domain containing protein 3 (TIM3), and T cell immunoreceptor with Ig and ITIM domains (TIGIT). Firstly, 5 μm sections were deparaffinized and rehydrated, after which heat-induced antigen retrieval was carried out in 1 mM EDTA buffer (pH 9.0) using a pressure cooker. Slides were heated until boiling and retrieval was continued for 3 min once maximum pressure was reached. Subsequently, the cooker was allowed to depressurize, and the slides were removed and cooled to room temperature. After cooling, sections were rinsed in deionized water for 5 min and then endogenous peroxidase activity was blocked using 3% hydrogen peroxide for 15 min at room temperature in the dark, followed by three washes with PBS (5 min each). Next, slides were blocked with 10% goat serum for 1 h at room temperature; after removal of the blocking solution, sections were incubated overnight at 4 °C with the following primary antibodies against CTLA4 (Biolynx, BX50210, 1:800), PD1 (Abcam, ab216352, 1:200), TIM3 (Biolynx, BX50205, 1:800), and TIGIT (Abcam, ab243903, 1:500). On the following day, slides were equilibrated at room temperature for 15 min, washed three times with PBS (5 min each), and then incubated with the secondary antibody working solution for 30 min at room temperature. After three PBS washes (5 min each), DAB substrate was applied for 5 min and the reaction was stopped using deionized water. Sections were then counterstained with hematoxylin for 3 min and rinsed under running water, differentiated in acid alcohol for 5 s, rinsed again, and blued in ammonia water for 5 s followed by a final water rinse. Finally, slides were dehydrated through graded ethanol (75% for 1 min, 95% for 1 min, and absolute ethanol for 2 min twice), cleared in xylene for 2 min twice, and mounted with neutral resin for microscopic evaluation. Then IHC staining was quantitatively evaluated using the H-score method. For each case, 4–5 non-overlapping high-power fields (200×) were randomly captured and staining intensity was categorized as 0 (no staining), 1 (weak staining), 2 (intermediate staining), or 3 (strong staining). The H-score was calculated as H-score = 1 × (weakly stained cells%) + 2 × (moderately stained cells%) + 3 × (strongly stained cells%), yielding a total score ranging from 0 to 300. H-scores were independently assessed by two pathologists blinded to clinical and molecular data and discrepancies were resolved by consensus after discussion. Due to limited tissue availability, IHC assay and H-score quantification was performed in 30 of 33 cases. Statistical analysis All correlation analyses were performed using the Spearman correlation coefficient (two-sided). Comparisons between two groups were conducted using the Mann–Whitney U test and data were presented as mean ± SEM. All statistical analyses were performed using GraphPad Prism 8.0.1 (San Diego, USA). Summary of somatic mutations for all samples was generated by R package ‘maftools’ (version 2.12.0). Gene mutation landscape was plotted with the R package ‘ComplexHeatmap’ (version 2.12.1). The scatter diagrams and bar charts were drawn using GraphPad Prism 8.0.1 software (San Diego, USA). Motifs of TCR sequences were generated by online website of the MEME Suite (https://meme-suite.org/meme/index.html). In all analyses, a two-sided P value < 0.05 was considered statistically significant. Results Gene mutation landscape of Chinese CCA patients Among 33 CCA patients, we identified 3338 mutations, including 11.23% (375/3338) insertion/deletion (InDel) and 88.77% (2963/3338) single-nucleotide variant (SNV) (Detailed data are provided in Additional file 1: Table S2). The predominant SNV mutation patterns were C > T (51.74%), C > A (14.68%), and T > C (10.12%) (Fig. 1A). All patients exhibited a median of 63 somatic mutations, with a range from 12 to 803. Subsequent analysis focused on genes displaying mutations in at least two patients. TP53 emerged as the most frequently mutated gene, with alterations observed in 57.58% (19/33) cases. Following TP53, other frequently mutated genes included ARID1A (18.18%, 6/33), ZNF469 (15.15%, 5/33), KRAS (15.15%, 5/33), MUC16 (15.15%, 5/33) and CRB1 (12.12%, 4/33) (Fig. 1B). Additionally, we identified 52 shared mutations across 50 genes that were present in at least two patients (Detailed data are provided in Additional file 1: Table S3). Besides, three CCA patients who underwent neoantigen peptide synthesis exhibited 178 somatic mutation sites affecting 175 genes, with SNV being the predominant mutation type (Fig. 1C). Fig. 1Gene mutation landscape in CCA patients. (A) Genomic landscape of somatic alterations in tumor tissues. (B) Top 6 frequently somatic mutated genes. (C) Summary of somatic mutation counts and types in three cases. CCA, cholangiocarcinoma Gene mutation landscape in CCA patients. (A) Genomic landscape of somatic alterations in tumor tissues. (B) Top 6 frequently somatic mutated genes. (C) Summary of somatic mutation counts and types in three cases. CCA, cholangiocarcinoma Correlation between tumor mutational burden (TMB) and neoantigen load and immune checkpoint molecule expression in CCA patients To better understand the relationship between TMB and neoantigen generation in CCA patients, we performed a comprehensive analysis of mutation data from 33 cases. TMB values ranged from 0.20 to 13.23, with a median of 1.04. A strong positive correlation was observed between TMB levels and the number of predicted neoantigens (r = 0.6511, p < 0.0001) (Fig. 2A). To further explore the correlation between TMB and the tumor immune microenvironment, we performed IHC staining for four immune checkpoint molecules (CTLA4, PD1, TIM3, and TIGIT) in tumor samples (30 of 33 cases; three cases were not evaluable due to insufficient tissue sample) and quantified staining using H-scores derived from 4 to 5 fields per case. Patients were stratified into high TMB (H-TMB) (n = 14) and low TMB (L-TMB) (n = 16) groups based on the median TMB value across the cohort. Compared with the L-TMB group, the H-TMB group showed higher H-scores for these immune checkpoint molecules (Fig. 2B, C). The detailed H-score data are provided in Additional file 1: Table S4. In addition, correlation analyses showed that the H-scores of CTLA4, PD1, TIM3, and TIGIT were positively associated with TMB (Fig. 2D). Given the presence of one extreme TMB value (13.23), we performed a sensitivity analysis by excluding this case to assess the robustness of the observed associations. After removing this outlier, the correlation between TMB and predicted neoantigen load remained significant in the remaining 32 patients (r = 0.6167, p = 0.0002) (Additional file 2: Figure S1A). Similarly, in the subset of 29 patients with available H-score data, TMB remained significantly correlated with the H-scores of all four immune checkpoint molecules (CTLA4: r = 0.3992, p = 0.0319; PD1: r = 0.5480, p = 0.0021; TIM3: r = 0.5661, p = 0.0014; TIGIT: r = 0.4034, p = 0.0300) (Additional file 2: Figure S1B). Fig. 2Correlation analysis between TMB and neoantigen load and immune checkpoint molecule expression in CCA patients. (A) Correlation analysis between TMB and neoantigen load. (B) Quantitative comparison of immune checkpoint molecule expression (CTLA4, PD1, TIM3, and TIGIT) between H-TMB (n = 14) and L-TMB (n = 16) groups. The expression of immune checkpoint molecule was quantified with an H-score (ranging between 0 and 300) derived by multiplying a three-value staining intensity score (1, weak; 2, moderate; 3, strong) with the percentage of cells showing positivity (0–100%). The bar graphs are represented as mean ± SEM. The P values were calculated using a two-tailed Mann-Whitney U test. (C) Representative IHC images of CTLA4, PD1, TIM3, and TIGIT staining in H-TMB and L-TMB patients. Scale bars, 100 μm. (D) Correlation analysis between TMB and H-scores. CCA, cholangiocarcinoma. TMB, tumor mutational burden, IHC, immunohistochemistry Correlation analysis between TMB and neoantigen load and immune checkpoint molecule expression in CCA patients. (A) Correlation analysis between TMB and neoantigen load. (B) Quantitative comparison of immune checkpoint molecule expression (CTLA4, PD1, TIM3, and TIGIT) between H-TMB (n = 14) and L-TMB (n = 16) groups. The expression of immune checkpoint molecule was quantified with an H-score (ranging between 0 and 300) derived by multiplying a three-value staining intensity score (1, weak; 2, moderate; 3, strong) with the percentage of cells showing positivity (0–100%). The bar graphs are represented as mean ± SEM. The P values were calculated using a two-tailed Mann-Whitney U test. (C) Representative IHC images of CTLA4, PD1, TIM3, and TIGIT staining in H-TMB and L-TMB patients. Scale bars, 100 μm. (D) Correlation analysis between TMB and H-scores. CCA, cholangiocarcinoma. TMB, tumor mutational burden, IHC, immunohistochemistry Neoantigen profiling in CCA patients Across the 33 patients, 53 human leukocyte antigen (HLA) alleles were identified, including 13 HLA-A, 25 HLA-B, and 15 HLA-C alleles (Additional file 1: Table S5). The estimated allele frequencies in the global and Chinese populations from the public database (https://www.allelefrequencies.net/), together with the detection within our cohort, are summarized in Additional file 1: Table S5. We successfully identified tumor neoantigens in 22 out of the 33 cases, totaling 798 neoantigens (Fig. 3A-B). To characterize which HLA alleles are responsible for presenting the largest diversity of neoantigens within our cohort, we analyzed the total number of neoantigens (out of 798) that predicted to bind to each unique HLA allele. Notably, the predominant top 10 HLA alleles that could bind to most neoantigens were HLA-A02:07, HLA-A02:01, HLA-B40:01, HLA-A11:01, HLA-C03:03, HLA-A03:01, HLA-A30:01, HLA-A24:02, HLA-B35:01 and HLA-B07:02 (Fig. 3C). Detailed neoantigen data are provided in Additional file 1: Table S6. Fig. 3Comprehensive analysis of neoantigen profile and HLA binding across samples. (A) Predicted neoantigen number per individual sample. (B) Summary of neoantigen number under different IC50 binding affinity cutoffs. (C) Neoantigen counts binding with the top 10 HLA alleles that responsible for presenting the largest diversity of neoantigens in our patients. HLA, human leukocyte antigen. IC50, half-maximal inhibitory concentration Comprehensive analysis of neoantigen profile and HLA binding across samples. (A) Predicted neoantigen number per individual sample. (B) Summary of neoantigen number under different IC50 binding affinity cutoffs. (C) Neoantigen counts binding with the top 10 HLA alleles that responsible for presenting the largest diversity of neoantigens in our patients. HLA, human leukocyte antigen. IC50, half-maximal inhibitory concentration A total of 91 neoantigens were identified in the three cases, spanning 11 HLA genotypes. Specifically, for the Pt01 sample, which harbored 11 mutated genes (MIB2, SHC1, UBN1, BIRC2, OLFML2B, CEP170B, SYK, CAMTA2, TOM1L2, HCFC1, and ITGA4), the HLA genotype was heterozygous for A (A24:02, A11:01), B (B56:01, B13:01), and C (C03:04, C01:02). Two mutation types were observed, including frameshift (FS-deletions/insertions) and missense mutations. The predicted neoantigen-HLA binding affinities were as follows: 29 peptides with IC50 ≤ 500 nM, 9 with IC50 ≤ 150 nM, 3 with IC50 ≤ 100 nM, and 2 with IC50 ≤ 50 nM. For the Pt02 sample, 13 mutated genes were identified (ACVR1, AHNAK2, ERBB2, SNRNP40, C2orf47, NUBP2, SREBF1, PLSCR4, SMARCA4, RTN4, TP53, DAZAP1, and EPS8L2), with HLA genotypes being heterozygous for A (A24:02, A02:06), B (B56:01, B35:01), and C (C03:03, C07:02). Similar to Pt01, two types of mutations were included (frameshift and missense), and the affinity distribution was: 42 peptides with IC50 ≤ 500 nM, 12 with IC50 ≤ 150 nM, 4 with IC50 ≤ 100 nM, and none with IC50 ≤ 50 nM. For the Pt03 sample, 12 mutated genes were found (LDOC1, CCDC14, CIR1, ITPR2, TRAPPC10, ARHGAP17, AUTS2, GPBAR1, KLF13, LCMT2, MAP3K4 and MYL12B). The HLA genotypes included homozygous A (A30:01), heterozygous B (B07:02, B13:01), and heterozygous C (C07:02, C*03:04). Mutation types were frameshift and missense. The distribution of peptide-HLA binding affinities was as follows: 20 peptides with IC50 ≤ 500 nM, 6 with IC50 ≤ 150 nM, 4 with IC50 ≤ 100 nM, and 1 with IC50 ≤ 50 nM. Detailed data are provided in Additional file 1: Table S7. Exploratory analysis of upregulated TAAs in CCA Compared to normal tissues, 2127 significant DEGs were identified, among which 545 genes were significantly upregulated in CCA tumors (Additional file 1: Table S8). We then compared the upregulated gene list with curated gene sets of ADC targets and CTAs. No ADC target genes showed significant upregulation in the tumor samples of this cohort. However, three CTA-associated genes—OIP5, SPACA3, and CCDC110—were identified as significantly upregulated in CCA tumors compared to normal tissues, offering broader opportunities for therapeutic targeting. Detailed information were summarized in Additional file 1: Table S9. Immunogenicity, hydrophilicity and toxicity prediction of neoantigen candidates In three patients with sufficient PBMC sample for ELISPOT assay, a total of 19 candidate neoantigen sequences were meticulously selected for peptide synthesis based on an integrated consideration of tumor gene expression, predicted peptide-MHC binding characteristics, the relative binding advantage of mutant versus wild-type peptides, and practical manufacturability (peptide manufacturability, solubility considerations and synthesis constraints), with additional input from the peptide manufacturer regarding synthesis feasibility. Detailed information of these 19 neoantigens were shown in Additional file 1: Table S10. This comprised 6 peptides for Pt01 (involving 6 mutant genes and 2 HLA genotypes), 7 peptides for Pt02 (involving 7 mutant genes and 5 HLA genotypes), and 6 peptides for Pt03 (involving 6 mutant genes and 2 HLA genotypes). During the synthesis of neoantigen polypeptides, amino acid sequences were added to both sides of the predicted neoantigen sequence, and the total length of the sequence was generally controlled within 28–30 amino acids. Before synthesizing the neoantigen peptides, the physicochemical properties were evaluated through computational tools, including the antigenicity, IFN-γ production, hydropathy, and toxicity of T cell epitopes. The antigenicity scores of all 19 peptides were ≥ 0.4, thereby validating the antigenic characteristics of the selected peptides. All neoantigen peptides exhibited IFN-γ production scores greater than 0, indicating their potential to induce IFN-γ production. Moreover, T-cell epitope toxicity prediction indicated that all 19 peptides were classified as non-toxic. With respect to hydrophobicity, neoantigen peptides derived from HCFC1, SNRNP40, ACVR1, and LDOC1 were predicted to be hydrophobic (GRAVY scores ≥ 0), whereas the remaining peptides had GRAVY scores < 0, indicating hydrophilic properties. (Fig. 4) Fig. 4Summary of neoantigen peptides characteristics including physicochemical and other properties. HLA, human leukocyte antigen. IC50, half-maximal inhibitory concentration. MT Epitope Seq, mutated amino acid sequence. IFN-γ, interferon-gamma Summary of neoantigen peptides characteristics including physicochemical and other properties. HLA, human leukocyte antigen. IC50, half-maximal inhibitory concentration. MT Epitope Seq, mutated amino acid sequence. IFN-γ, interferon-gamma Validation of immunogenicity of neoantigen peptides and corresponding TCR prediction in three CCA patients Functional immunogenicity testing by ELISPOT assays was performed in three CCA patients (Pt01–Pt03) with available longitudinal follow-up peripheral blood mononuclear cells (PBMCs) samples, providing subset-based, proof-of-concept validation. A total of 16 peptides were successfully synthesized. As shown in Fig. 5A, D, and G, PBMCs stimulated with phytohemagglutinin (PHA, positive control) exhibited strong IFN-γ responses, whereas PBMCs stimulated with wild-type peptides (negative control) showed minimal IFN-γ secretion, indicating no significant nonspecific activation. Among the tested neoantigen peptides, UBN1 (Pt01; Fig. 5A–C), C2orf47 (Pt02; Fig. 5D–F), and ITPR2 (Pt03; Fig. 5G–I) induced the strongest T cell responses, as evidenced by the highest spot counts in IFN-γ ELISPOT assays (Fig. 5B, E, and H; highlighted in red). These peptides outperformed other candidate peptides in their respective patients, indicating their superior immunogenic potential. Furthermore, TCR sequence motifs corresponding to the top neoantigens were predicted using the ImRex platform and visualized as sequence logos (Fig. 5C, F, and I). For Pt01, a high-affinity motif including the Vβ5-6Jβ1–2 segment was identified. Pt02 showed a dominant TCR motif corresponding to Vβ2Jβ1–4, while Pt03 demonstrated a predominant motif involving Vβ29-1Jβ2–7 (Additional file 1: Table S11). These motifs represent the most enriched clonotypes predicted in silico and provide exploratory candidates for future validation of potential TCR–peptide–MHC interactions. Fig. 5Validation of neoantigen peptides in three patients. (A, D, G) ELISPOT assay results for Pt01, Pt02, and Pt03, respectively. (B, E, H) Quantitative analysis of ELISPOT spot counts for Pt01, Pt02, and Pt03, respectively. Black bars, positive control. Red bars, the most effective neoantigens (UBN1 for Pt01; C2orf47 for Pt02; ITPR2 for Pt03). Grey bars, other effective neoantigens. (C, F, I) CDR3β amino-acid sequence logo motif analysis of high-frequency TCR clones matched to peptide epitopes derived from the UBN1, C2orf47, and ITPR2 genes (Pt01: prediction score ≥ 0.995; Pt02: prediction score ≥ 0.94; Pt03: prediction score ≥ 0.99). The sequence logo displays amino-acid residue frequency and degree of conservation at each position. Letter height reflects degree of conservation. Letters for different amino acids at the same position are scaled according to their frequency. A prominent logo motif is considered as conserved amino-acids, and other logos mean highly variable amino-acids. Different colors distinguish residue classes (polar, charged, hydrophobic, etc.). CDR3, complementarity-determining region 3. ELISPOT, enzyme-linked immunospot. IFN-γ, interferon-gamma Validation of neoantigen peptides in three patients. (A, D, G) ELISPOT assay results for Pt01, Pt02, and Pt03, respectively. (B, E, H) Quantitative analysis of ELISPOT spot counts for Pt01, Pt02, and Pt03, respectively. Black bars, positive control. Red bars, the most effective neoantigens (UBN1 for Pt01; C2orf47 for Pt02; ITPR2 for Pt03). Grey bars, other effective neoantigens. (C, F, I) CDR3β amino-acid sequence logo motif analysis of high-frequency TCR clones matched to peptide epitopes derived from the UBN1, C2orf47, and ITPR2 genes (Pt01: prediction score ≥ 0.995; Pt02: prediction score ≥ 0.94; Pt03: prediction score ≥ 0.99). The sequence logo displays amino-acid residue frequency and degree of conservation at each position. Letter height reflects degree of conservation. Letters for different amino acids at the same position are scaled according to their frequency. A prominent logo motif is considered as conserved amino-acids, and other logos mean highly variable amino-acids. Different colors distinguish residue classes (polar, charged, hydrophobic, etc.). CDR3, complementarity-determining region 3. ELISPOT, enzyme-linked immunospot. IFN-γ, interferon-gamma Discussion In this study, we performed WES and RNA-seq on tumor samples from 33 patients with CCA, identifying a total of 3293 somatic mutations, including 343 Indels mutations and 2,719SNVs. TP53 was the most frequently mutated gene (54.55%), followed by KRAS, ARID1A, and CRB1 (each at 15.15%). This mutational profile is highly consistent with previous studies in Chinese CCA cohorts [19–21]. As a key tumor suppressor, TP53 plays a central role in regulating the cell cycle, DNA repair, apoptosis, and differentiation [22]. Mutations in KRAS, on the other hand, drive aberrant activation of oncogenic signaling pathways, promoting tumor initiation and progression [23]. Prior research has suggested that CCA patients harboring concurrent KRAS and TP53 mutations tend to exhibit more aggressive biological behavior and may be more responsive to immunotherapy, whereas isolated KRAS mutations are associated with poorer prognosis and lower immunotherapeutic efficacy [24]. Further analysis revealed a significant positive correlation between TMB and the number of predicted neoantigens, indicating that TMB may influence the pool of potentially candidate neoepitopes in CCA patients. Besides, IHC analysis showed a higher expression tendency of immune checkpoint molecules in the H-TMB group, which may reflect differences in immune characteristics between H-TMB and L-TMB. TMB has been associated with favorable responses to ICIs therapy in previous studies [25, 26]. However, the predictive accuracy of TMB remains limited. Clinical responses to ICIs might be shaped by multiple factors, including the expression of immune checkpoint molecules, the presence of tumor-infiltrating lymphocytes, HLA class I expression, and immunosuppressive mediators. Our exploratory analyses raise the hypothesis that a subset of CCA patients may be more likely to benefit from ICIs therapy. However, it is necessary to further validate in prospective independent CCA cohorts with clinical endpoints because of the modest median TMB in our cohort. To assess the immunogenicity of selected neoantigens, we performed ELISPOT assays using patient-derived PBMCs. The results demonstrated that neoantigen peptides derived from UNB1 (Pt01), C2orf47 and EPS8L2 (Pt02), and ITPR2 (Pt03) could significantly induce IFN-γ secretion, indicating robust immune activation. The designed 28–30 amino acid long peptides spanned the mutant epitopes and included flanking wild-type sequences, which may enhance antigen processing and presentation via both MHC class I and II pathways. This design theoretically allows activation of both CD8⁺ cytotoxic and CD4⁺ helper T cell responses, providing a more comprehensive anti-tumor immune profile [27, 28]. However, functional validation was conducted in only three patients; therefore, these immunogenicity results are presented as proof-of-concept evidence and should be validated in larger independent cohorts before broader generalization. To further explore potential antigen-specific T cell recognition, we used the deep learning-based platform ImRex to predict TCR affinity for neoantigen-MHC complexes. Putative high-affinity TCR sequences were identified in Pt01, Pt02, and Pt03, including Vβ5-6Jβ1–2, Vβ2Jβ1–4, and Vβ29-1Jβ2–7, respectively. These sequences may reflect expanded T-cell clonotypes potentially associated with candidate neoantigen, offering an exploratory basis for downstream validation and for the future development of neoantigen-directed T-cell immunotherapy [29, 30]. However, further functional researches will be required to validate their neoantigen specificity and to determine whether these clonotypes are suitable for downstream translational applications. In our study, tumor-specific neoantigens derived from somatic mutations were predicted for most CCA cases. However, not all CCA patients could yield detectable candidate neoantigens, indicating a potential limitation of neoantigen-based strategies for all patients. As a complementary analysis, we therefore performed a transcriptome-based comparison between CCA tumor samples and normal liver tissues from the GTEx database to explore potential TAAs. Considering the translational feasibility of potential targets, we focused specifically on two categories of clinically relevant antigens: ADC targets and CTAs. Although no significant differences were detected in the ADC targets, three CTA-related genes were found to be significantly upregulated in CCA tumors. CTAs are characterized by restricted expression in germline tissues and aberrant activation in cancers, which makes them attractive candidates for immunotherapeutic targeting [31]. Collectively, these findings suggest that a number of TAAs may also exist in CCA, particularly within the CTA family. Such targets may provide additional therapeutic opportunities for patients who lack suitable neoantigen candidates, thereby complementing personalized neoantigen-based approaches. This study has several limitations. First, the neoantigens reported here are computationally predicted candidates, and they have not been directly confirmed as naturally processed and presented peptides on tumor cells. Second, in the ELISPOT assays, cells were stimulated with synthetic long peptides, and the minimal epitopes responsible for the response and their cell-surface presentation by HLA molecules were not directly verified. Moreover, functional validation was limited to three patients with available follow-up PBMCs, so these results are proof-of-concept and may not be generalizable to the full cohort. Finally, the in silico TCR motif analyses are exploratory due to the lack of experimental validation. Therefore, it is necessary to conduct further studies in the future to validate these findings in larger independent cohorts and to strengthen mechanistic evidence by performing HLA-restriction analyses, applying immunopeptidomics to directly identify peptides presented on the tumor cell surface, and experimentally confirming neoantigen-reactive TCR sequences. Conclusions This study analyzed the mutational landscape and neoantigen profile of Chinese CCA, and we demonstrated that neoantigens can elicit T cell response. TCR motif analysis further identified exploratory clonotype patterns that may be associated with neoantigen stimulation. In summary, these results provide a resource of neoantigen candidates in CCA patients and exploratory TCR features for future experimental validation. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1: Additional file 1: Table S1. List of 13 ADC targets and 224 CTA genes. Note: ADC, antibody–drug conjugate; CTA, cancer-testis antigen. Table S2. Somatic mutations across the 33 CCA patients. Note: CCA, cholangiocarcinoma. Table S3. 52 shared somatic mutations among 33 CCA patients. Note: CCA, cholangiocarcinoma. Table S4. H-score quantification of immunohistochemistry in the CCA cohort (n = 30). Note: CCA, cholangiocarcinoma. Table S5. Global and Chinese allele frequencies and cohort detection of HLA alleles identified in the 33 CCA patients. Note: HLA, human leukocyte antigen. CCA, cholangiocarcinoma. Table S6. Summary of predicted neoantigens in 22 CCA patients. Note: HLA, human leukocyte antigen. MT, mutation. WT, wild type. Table S7. Summary of neoantigen, HLA isoforms and mutation types in three patients. Note: IC50, half-maximal inhibitory concentration. HLA, human leukocyte antigen. Table S8. Significantly DEGs identified in CCA compared with normal tissues of GTEx database. Note: DEGs, differentially expressed genes. CCA, cholangiocarcinoma. Table S9. Significantly upregulated CTA genes identified in CCA compared with normal tissues of GTEx database. Note: CTA, cancer-testis antigen. CCA, cholangiocarcinoma. Table S10. Predicted features of the 19 neoantigen peptides selected for synthesis. Note: HLA, human leukocyte antigen. MT, mutation. WT, wild type. Table S11. Predicted high-likelihood binding interactions between neoantigen peptides and TCR sequences in three patients. Note: CDR3, complementarity-determining region 3. TCR, T cell receptor. Supplementary Material 1: Additional file 1: Table S1. List of 13 ADC targets and 224 CTA genes. Note: ADC, antibody–drug conjugate; CTA, cancer-testis antigen. Table S2. Somatic mutations across the 33 CCA patients. Note: CCA, cholangiocarcinoma. Table S3. 52 shared somatic mutations among 33 CCA patients. Note: CCA, cholangiocarcinoma. Table S4. H-score quantification of immunohistochemistry in the CCA cohort (n = 30). Note: CCA, cholangiocarcinoma. Table S5. Global and Chinese allele frequencies and cohort detection of HLA alleles identified in the 33 CCA patients. Note: HLA, human leukocyte antigen. CCA, cholangiocarcinoma. Table S6. Summary of predicted neoantigens in 22 CCA patients. Note: HLA, human leukocyte antigen. MT, mutation. WT, wild type. Table S7. Summary of neoantigen, HLA isoforms and mutation types in three patients. Note: IC50, half-maximal inhibitory concentration. HLA, human leukocyte antigen. Table S8. Significantly DEGs identified in CCA compared with normal tissues of GTEx database. Note: DEGs, differentially expressed genes. CCA, cholangiocarcinoma. Table S9. Significantly upregulated CTA genes identified in CCA compared with normal tissues of GTEx database. Note: CTA, cancer-testis antigen. CCA, cholangiocarcinoma. Table S10. Predicted features of the 19 neoantigen peptides selected for synthesis. Note: HLA, human leukocyte antigen. MT, mutation. WT, wild type. Table S11. Predicted high-likelihood binding interactions between neoantigen peptides and TCR sequences in three patients. Note: CDR3, complementarity-determining region 3. TCR, T cell receptor. Supplementary Material 2: Additional file 2: Figure S1. Sensitivity analyses of correlations after excluding an extreme TMB value. (A) Spearman correlation between TMB and the number of predicted neoantigens in the remaining 32 patients. (B) Spearman correlations between TMB and IHC H-scores in the 29 patients with available H-score data. TMB, Tumor mutational burden. IHC, immunohistochemistry. Supplementary Material 2: Additional file 2: Figure S1. Sensitivity analyses of correlations after excluding an extreme TMB value. (A) Spearman correlation between TMB and the number of predicted neoantigens in the remaining 32 patients. (B) Spearman correlations between TMB and IHC H-scores in the 29 patients with available H-score data. TMB, Tumor mutational burden. IHC, immunohistochemistry.