Human GBC cell lines GBC-SD and NOZ (sourced from The Chinese Academy of Sciences and The iCell Bioscience Inc., respectively) were cultured in Dulbecco’s Modified Eagle’s Medium (DMEM; California, USA) supplemented with 1% antibiotics and 10% fetal bovine serum; GIBCO, Grand Island, NY, USA. Nude mice (BALB/c, age: 6 – 8 weeks) were obtained from the Cavens Model Animal Research Company (Suzhou, China). A lentivirus vector was procured from the Shanghai Jikai Gene Company (Shanghai, China). The GPR64 monoclonal antibody (host: mouse; isotype: IgG1) was obtained from Wuhan Sanying Biotechnology Company (Wuhan, China).

All animal experiments were performed in adherence to the guidelines and protocols of the China Animal Protection Association. The ethics committee of Bengbu Medical University provided its ethical approval (Ethics Approval Letter (2024) 377). Relevant datasets (i.e., GSE238179 and GSE255497) were downloaded from the Gene Expression Omnibus (GEO) database and organized and visualized using Perl and R software. The data was obtained in the following order: Search keywords (Biliary Tract Cancer:1631), Top Organisms (Homo sapiens: 1625), Entry type (Series: 76), Study type (Expression profiling by array: 21). The first two items that best matched the search criteria were selected, that is GSE238179:8 and GSE255497:32. The corresponding clinical information is given in Supplementary File 1 and Supplementary File 2 . Exosome-related genes were sourced from the https://www.genecards.org/website (861 websites) and the related literature (18 sources).

To reduce the data bias between the two datasets (i.e., GSE238179 and GSE255497), principal component analysis (PCA) was performed on the expression matrix before and after eliminating the batch effects. The filtering condition was set to logFC absolute value >1 and P < 0.05, and the differential analysis and visualization were conducted using relevant R packages (limma, dplyr, pheatmap, and ggplot2).

To further explore the potential biological functions of DEGs in the pathogenesis of GBC, we analyzed the differential genes of GBC by Gene Ontology (GO), Kyoto Encyclopedia of Genes and Genomes (KEGG), and Gene Set Enrichment Analysis (GSEA) methods for analysis. KEGG and GSEA enrichment analysis revealed the enrichment of the related pathways and genes.

Least Absolute Shrinkage and Selection Operator (LASSO) regression analysis was performed on all the DEGs of GBC, and the results of LASSO regression analysis were regarded as model genes. Then, the highest accuracy and the lowest error rate were determined by the Support Vector Machine (SVM) algorithm and further applied to identify the DEGs. The key genes were identified in the LASSO regression model and SVM analysis. Finally, a receiver operating characteristic (ROC) curve was drawn for the key genes to evaluate the accuracy of the model.

To explore the function and role of DEGs in the immune microenvironment of GBC, the GSVA package and GSEA Base package of R software were used to immunologically score the DEGs of GBC. The key gene scores were extracted for analysis, and the correlation heat map was visualized with the R pheatmap package to analyze the correlation results between immune cells and the key differential genes.

A cell suspension (1 mL; density of 3 – 5 x 10⁴/mL) was mixed with 3 mL of DMEM and incubated in a 6-well plate for 16 – 24 h. A certain amount of virus and infection enhancer was then added in accordance with the MOI and virus titer of the cells specified in the instructions. The mixture was cultured for 12 – 16 h, and the medium was refreshed to continue culturing for the determination of the change time based on cell morphology. After approximately 3 days of incubation, the cells were transfected under a fluorescence microscope, puromycin was used to screen the uninfected cells, and the virus volume was calculated as follows: (MOI × cell count)/virus titer. (GPR64-OE: GPR64 Overexpression, GPR64-NC: GPR64 Negative Control). The RT-PCR primer sequence used in the study was GPR64 F:CAGGCGTCAAACCCCAGAG, GPR64 R:CCAGTTAAGGTGCCATTCGTTAT. GAPDH F: CAGGAGGCATTGCTGATGAT,GAPDH R: GAAGGCTGGGGCTCATTT.

Six different groups (namely, GBC-SD-GPR64-OE, GBC-SD-GPR64-NC, GBC-SD, NOZ-GPR64-OE, NOZ-GPR64-NC, and NOZ) of cell suspensions were prepared at the same density and injected into the upper-right abdomen of nude mice (n = 30, Density: 5 × 10⁷/mL). The tumor volume was measured once every 5 days (volume [mm³] = 0.5 × width² × length). On day 30, all mice were euthanized, and their tumor tissues were harvested for subsequent analyses.

Proteins were extracted and quantified in accordance with the instructions provided with the RIPA lysis solution (Beyotime, Jiangsu, China) and BCA assay kit (Jiangsu, China), respectively, followed by loading, electrophoresis, membrane transfer, blocking with milk, treatment with primary antibody, washing, treatment with secondary antibody, and washing. Finally, ECL was used to expose the images in a gel imaging system (Bio-Rad, USA). Other antibodies used included B-cell lymphoma-2 (Bcl-2) antibody/neural cadherin (N-cadherin) antibody/Bcl-2-associated X protein (Bax) antibody/vimentin antibody/cysteine-dependent aspartate-specific protease-3 (caspase-3) antibody/goat anti-rabbit IgG (primary antibody dilution ratio of 1:1000, secondary antibody dilution ratio of 1:10000; Jiangsu Qinke Biological Research Center Co., Ltd., Jiangsu, China).

Paraffin sections of subcutaneous tumors obtained from the experimental mice were dewaxed and hydrated, followed by blocking with goat serum (Beyotime) for 30 min and incubation with the primary antibody in a wet box at 4 °C overnight. Next, the sections were washed and incubated with the secondary antibody. Color was developed using DAB (Beyotime) color developing solution after washing; brown and yellow staining indicated a positive result. After washing, the nucleus was stained and then gently dehydrated, meticulously sealed, and observed.

The paraffin sections of the murine subcutaneous tumors were dewaxed and hydrated. Hematoxylin staining of the cell nucleus and differentiation with 1% hydrochloride alcohol for 10 s were then performed, after which the slices were treated with 0.6% ammonia water and rinsed with clean water. These sections were then stained with eosin for 5 min after gradient ethanol dehydration and sealed using neutral gum gel. The stained sections were then imaged under a microscope, and the observations were recorded.

The experiment was conducted in accordance with the detailed steps described in the Transwell assay manual (23). Briefly, the cells were seeded into a 6-well plate chamber without and with matrix gel and incubated in serum-free culture medium. To this, 10% fetal bovine serum was added into the lower chamber medium and allowed to culture for 1 day in the cell incubator, followed by gentle washing with phosphate-buffered saline (PBS), fixing with polyformaldehyde, staining, and, finally, photography.

The cell suspension was inoculated into a 6-well plate after grouping. When the cells matured, a 200-µL pipette was used to scratch the cells and observe and photograph them at the same position under a microscope at 0, 24, and 48 h timepoints.

Cells showing an adherence rate of >90% were digested, centrifuged, resuspended, and counted. Then, 1 mL of the cell suspension was added to each well of a 6-well plate at a density of 1000 cells/mL and placed in a CO₂ incubator. The medium was changed according to predefined conditions. After 2 weeks of culture period, by when the colonies were formed, the cells were fixed and stained.

After the cells reached approximately 90% confluence, they were subjected to trypsin digestion and collection. The cells were washed in pre-cooled PBS twice, and a cell suspension was prepared. For cell staining, 5 µL of Annexin V-APC was added to the cell suspension and mixed well, to which 10 µL of PI was added and mixed. After incubation at room temperature for approximately 15 min, flow cytometry was performed, and the results were recorded.

All experiments were repeated more than three times. The data were analyzed using GraphPad Prism and expressed as the mean ± standard deviation. Statistical analyses were performed using the t-test, one-way analysis of variance (ANOVA), Mann–Whitney U-test, and Ruscall–Wallis test. Statistical significance was set to P < 0.05.

Relevant datasets (GSE238179, GSE255497) were downloaded from the GEO database. Using Perl and R, the data were processed, corrected, and visualized to identify intersecting genes ( Figures 1A, B ). Further differential analysis revealed 16 upregulated and 27 downregulated genes ( Supplementary File 3 ). The top 10 genes in each category are displayed in the heatmap ( Figure 1C ).

To explore the biological functions of DEGs in GBC pathogenesis, GO, KEGG, and GSEA enrichment analyses were performed ( Supplementarys Files 4 – 6 ). GO analysis showed enrichment in the following: Biological process: Wnt-signaling pathway, cell–cell signaling by Wnt, and potassium ion transmembrane transport; Cellular component: neuronal cell body, transmembrane transporter complex, and voltage-gated potassium channel complex; and Molecular function: potassium ion transmembrane transporter activity, voltage-gated potassium channel activity, and potassium channel activity ( Figure 1D ). KEGG analysis revealed the top three pathways as cAMP signaling, thyroid hormone signaling, and growth hormone synthesis, secretion, and action ( Figure 1E ). GSEA identified the top five pathways as Cell Cycle, Drug Metabolism – Cytochrome P450, Retinol Metabolism, Chemical Carcinogenesis – DNA Adducts, and Metabolism of Xenobiotics by Cytochrome P450 ( Figure 2A ).

To identify key genes involved in GBC pathogenesis, LASSO and SVM were used for dual screening ( Figures 2B–E ), yielding three crucial DEGs: TMEM163, GPR64, and RNF112 ( Figure 2F ). ROC curve analysis was conducted to assess diagnostic performance: RNF112 (AUC = 0.853), GPR64 (AUC = 0.838), and TMEM163 (AUC = 0.8858) ( Figure 3A ). The combined AUC for all three genes was 0.978 ( Figure 3B ), indicating high diagnostic accuracy for GBC.

A nomogram was constructed to evaluate the correlation between gene expression and GBC risk. The results showed that lower gene expression corresponded to higher risk scores ( Figure 3D ). The calibration curve confirmed the predictive accuracy of the three-gene model ( Figure 3C ).

Expression levels of key genes were visualized in control and treatment groups, clearly showing that all three function as tumor suppressor genes ( Figure 3E ). Additionally, their interrelationships were analyzed ( Figure 3F ).

To further investigate the impact of key genes on the immune microenvironment in GBC, immune infiltration was assessed using ssGSEA. The results showed that these genes were associated with monocytes ( Figure 4A ). A heat map generated in R revealed correlations between key genes and immune cell infiltration. GPR64 was linked to activated B cells, activated CD8 T cells, effector memory CD8 T cells, immature B cells, and type 2 T helper cells. RNF112 was associated with activated B cells, activated CD8 T cells, central memory CD8 T cells, effector memory CD4 T cells, eosinophils, macrophages, mast cells, and type 1 and type 17 T helper cells. TMEM163 was associated only with monocytes ( Figure 4B ).

We combined the key genes with exosome-related genes (exosome genes:https://www.genecards.org/) to investigate their possible roles in GBC pathogenesis. GPR64 was identified as an exosome-associated gene ( Figure 5A ). Lentiviral transfection was used to overexpress GPR64 in GBC cell lines (GBC-SD, NOZ), and transfection efficiency was confirmed by RT-PCR ( Figure 5B ). GPR64 expression in the overexpression group was significantly higher than in the negative control group.

Colony formation assays showed that the number of colonies in the GPR64-OE group was significantly lower than in the GPR64-NC and Control groups ( Figure 5C ). WB analysis indicated reduced levels of the proliferation-related protein PCNA in the GPR64-OE group compared to the GPR64-NC and Control groups ( Figure 5D ).

Additionally, the wound healing rate at 0 – 24 h and 24 – 48 h was significantly lower in the GPR64-OE group than in the GPR64-NC and Control groups ( Figures 6A, B ). Transwell assays showed reduced cell migration and invasion in the GPR64-OE group compared to the corresponding Control group ( Figures 6C, D ). WB results further demonstrated that the expression of migration-related proteins Vimentin and N-cadherin was significantly lower in the GPR64-OE group than in the GPR64-NC and Control groups ( Figure 6E ) (all P < 0.05). These findings suggest that GPR64 overexpression inhibits the migration, invasion, and proliferation of GBC-SD and NOZ cells.

Flow cytometry analysis revealed that the apoptosis rate in the GBC-SD-GPR64-OE and NOZ-GPR64-OE groups was higher than that in the GBC-SD-GPR64-NC, NOZ-GPR64-NC, and Control groups ( Figure 7A ). WB analysis showed that pro-apoptotic proteins Bax and Caspase-3 were significantly higher in the GBC-SD-GPR64-OE and NOZ-GPR64-OE groups than in the GBC-SD-GPR64-NC, NOZ-GPR64-NC, and Control groups, while the anti-apoptotic protein Bcl-2 was significantly lower in the GBC-SD-GPR64-OE and NOZ-GPR64-OE groups than in the GBC-SD-GPR64-NC, NOZ-GPR64-NC, and Control groups ( Figure 7B ). These results indicate that GPR64 overexpression significantly enhances apoptosis in GBC cell lines (GBC-SD and NOZ).

Subcutaneous tumors were collected from three groups of nude mice ( Figure 8A ). On day 30, tumors in the GBC-SD-GPR64-NC, NOZ-GPR64-NC, and Control groups weighed significantly more than those in the GBC-SD-GPR64-OE and NOZ-GPR64-OE groups ( Figure 8B ), suggesting that GPR64 overexpression effectively suppresses tumorigenicity in GBC cells. Immunohistochemical staining of tumors from both cell lines (GBC-SD, NOZ) showed that the average optical density of GPR64-positive areas was higher in the GPR64-OE group than in the GPR64-NC and Control groups ( Figure 8C ).

Moreover, neovascularization in tumors in the GPR64-OE group was significantly lower than that in the GPR64-NC and Control groups ( Figures 8D, E ), suggesting that GPR64 exerts its tumor-suppressive effect by modulating tumor angiogenesis.

Overall, GPR64 significantly inhibits GBC progression by affecting the migration, invasion, proliferation, apoptosis, and angiogenesis of GBC cells.