KKU-213A (10) and KKU-100(11) were obtained from the Japanese Collection of Research Bioresources Cell Bank, Osaka, Japan. Stable gemcitabine-resistant CCA cell lines (CCA-GemR) were developed from the parental cell lines KKU-213A and KKU-100 by continuous exposure to gradually increasing doses of gemcitabine, as previously described (12). Parental CCA cell lines were cultured in complete Dulbecco's modified Eagle's medium (DMEM) containing 5 mM glucose, supplemented with 10% fetal bovine serum, 100 U/ml penicillin, and 100 µg/ml streptomycin (all from Gibco/BRL; Thermo Fisher Scientific, Inc.) and maintained at 37˚C in a humidified incubator with 5% CO₂. Gemcitabine-resistant CCA cell lines were maintained under the same conditions but were additionally treated with the IC₁₀ concentration of gemcitabine. The present study was approved (approval no. HE661419) by the Khon Kaen University Ethics Committee for Human Research (Khon Kaen, Thailand).

To identify common molecules related to gemcitabine resistance in cancer cell lines, transcriptomic datasets from stable gemcitabine-resistant cancer cell lines were retrieved from the GEO database (https://www.ncbi.nlm.nih.gov/geo/), including GSE116118 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE116118) (13), GSE208659 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE208659) (14), and GSE140077 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE140077) (15). In addition, a transcriptomic dataset (GSE76297, https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE76297) (16) from 92 normal and 91 Thai tissues from patients with CCA was also used to determine the expression levels of candidate genes in patients with CCA. DEGs were investigated using GEO2R (www.ncbi.nlm.nih.gov/geo/geo2r) and Galaxy (https://usegalaxy.org/) (17). DEGs were determined based on the criteria of an adjusted P-value <0.05 and log2|FC| >0.5. Data visualizations, including volcano plots and Venn diagrams, were generated using RStudio version 2024.12 (https://posit.co/) and jvenn (https://jvenn.toulouse.inrae.fr/app/index.html) (18), respectively.

Genes correlated with RTKN2 expression in patients with CCA were identified using the cBioPortal platform (https://www.cbioportal.org) (19). The TCGA-CHOL dataset was selected, and Spearman correlation analysis was performed to determine the genes that were positively and negatively correlated with RTKN2 expression. To investigate the biological significance of RTKN2-correlated genes, pathway enrichment analysis was performed separately for positively and negatively correlated genes using Enrichr (https://maayanlab.cloud/Enrichr/). Enrichment results were ranked by adjusted P-values using the Benjamini-Hochberg correction, and the most significantly enriched pathways (adjusted P<0.05) were reported and visualized.

Total RNA was extracted from CCA cell lines using TRIzol reagent (Invitrogen; Thermo Fisher Scientific, Inc.). The RNA concentration was measured using NanoDrop^™ 2000 Spectrophotometer (Thermo Fisher Scientific, Inc.). The extracted RNA from CCA cell lines was converted to cDNA using High-capacity cDNA Reverse Transcription Kit (cat. no. 4368814, Applied Biosystem^™; Thermo Fisher Scientific, Inc.). Using real-time PCR, the mRNA expression levels of candidate genes associated with gemcitabine resistance were determined in CCA cell lines. Real-time PCR was performed using LightCycle 480 real-time PCR system (Roach Diagnostics). Each PCR condition contained 2X LightCycle 480 SYBR Green I Master (Roach Diagnostics), 5 mM of primers, and 40 ng of cDNA. The amplification was initiated by incubating at 95˚C for 5 min, followed by 40 cycles at 95˚C for 20 sec, 65˚C (RAB1B, JAG1, PACS1 and ANKS6) and 63˚C (SF3B4 and RTKN2) for 10 sec, and 72˚C for 20 sec. Mean and standard deviation values of the cycle of quantification threshold (Cq) and melting temperature were calculated. The expression of each gene was normalized with β-actin. The gene expression levels were determined using the 2^-ΔΔCq method (20). All PCR primers and expected PCR product sizes are shown in Table I.

To evaluate the effect of gemcitabine on RTKN2 expression, parental CCA cell lines (KKU-213A and KKU-100) were plated in 10-cm culture dishes and incubated for 24 h. The cells were then treated with gemcitabine at concentrations corresponding to IC₀ and IC₅₀ values. After 48 h of incubation, total RNA was extracted and RT-qPCR was performed as aforementioned.

The amino acid sequence of human RTKN2 was retrieved from the Uniprot website (UniProt ID: Q8IZC4; https://www.uniprot.org/), and utilized for constructing the 3D structure using AlphaFold3(21). The protonation states of ionizable amino acid residues of the constructed RTKN2 were assigned at pH 7.4 using the PDB2PQR web server version 3.6.1 (https://server.poissonboltzmann.org/pdb2pqr) (22,23). The protein structure was then minimized using molecular dynamics (MD) simulations for 100 ns using the AMBER24 software (https://ambermd.org/). The AMBER ff19SB force field was applied to the protein (24). The TIP3P water model (25) was used to solvate the system. MD simulations were performed under periodic boundary conditions in the isothermal-isobaric (NPT) ensemble at 310 K and 1 atm.

The last MD snapshot of the RTKN2 structure (Fig. S1) was subsequently analyzed for the potential ligand-binding sites using the CavityPlus 2022 web server (http://pkumdl.cn:8000/cavityplus) (26,27). Sites 2, 5, and 9, which exhibited strong druggability (Table SI), were then selected as ligand-binding sites for molecular docking. The coordinates (x, y, z) of these binding sites were as follows: Site 2 (94.957, 76.997, 99.134), site 5 (113.028, 76.794, 108.707), and site 9 (45.114, 90.615, 54.979). For virtual screening, compounds (excluding those with a molecular weight >1,000 g/mol, glycans, peptides, and lipids) from the drug repurposing compound library plus (https://www.medchemexpress.com/screening-libraries.html, accessed on 1st March 2025) were docked into each of the three binding sites using AutoDock Vina version 1.2.6 (28,29). The top three compounds exhibiting the highest binding affinities (lowest binding energy) at each site were selected as potential hits.

All RT-qPCR data are presented from three independent biological experiments (triplicate in each experiment). Statistical analysis was performed using SPSS statistics 24.0 software (IBM Corp.) and GraphPad Prism 9.2 software (Dotmatics). To determine whether the observed expression showed a significant difference between the two groups (resistance vs. parental), data that demonstrated normality and homogeneity of variance were compared using Student's t-test, while data with non-normal distribution or unequal variance were analyzed using the Mann-Whitney U test. For visualization, box-and-whisker plots displaying the median and interquartile range (IQR) were used in nonparametric cases, whereas for parametric analyses (t-test), the mean value with 95% confidence intervals (CI) is presented. P-values <0.05 were considered to indicate statistically significant differences.

A total of three transcriptomic datasets of stable gemcitabine-resistant cell lines and their corresponding parental cell lines, including GSE116118, GSE208659, and GSE140077, were retrieved from the GEO database. These datasets included cell lines from CCA (MT-CHC01R1.5 gemcitabine resistant/MT-CHC01 parental), gallbladder (NOZ-GemR gemcitabine resistant/NOZ parental), and pancreatic cancers (BxPC-3-GR gemcitabine resistant/BxPC-3 parental). These cancers are commonly grouped as pancreatobiliary cancers because they arise from foregut-derived pancreatobiliary epithelium and share overlapping histopathological features, clinical behavior, and poor prognosis. By integrating these three datasets, the aim was to minimize bias arising from cell line specificity and to identify reproducible DEGs consistently associated with gemcitabine resistance.

The analysis of DEGs between gemcitabine-resistant cell lines and their corresponding parental cell lines identified a total of 7,751, 2,730, and 2,206 DEGs in GSE116118, GSE208659, and GSE140077, respectively. These results, including upregulated and downregulated genes in the three stable gemcitabine-resistant cancer cell lines, are shown in Table II. Volcano plots in Fig. 1A show the upregulated and downregulated DEGs. Venn diagrams were used to determine common DEGs across the three transcriptomic datasets from stable gemcitabine-resistant cancer cell lines. The analysis identified 14 DEGs shared among the three stable gemcitabine-resistant cancer cell lines (Fig. 1B). The 14 DEGs were RAB1B, JAG1, PACS1, ANKS6, SF3B4, RTKN2, PGM2, DNAH11, NT5DC1, CDK14, IGFBP6, RNFT1, RBBP9, and ZDHHC18. These commonly upregulated DEGs may be associated with gemcitabine resistance in cancer cells.

To prioritize genes potentially relevant to CCA, the expression levels of the 14 upregulated DEGs in patients with CCA were further validated using the transcriptomic dataset (GSE76297), which includes 92 normal and 91 CCA tissues from Thai patients. The results indicated that six of the 14 upregulated DEGs, including RAB1B, JAG1, PACS1, ANKS6, SF3B4, and RTKN2, were significantly upregulated in CCA tissues compared with adjacent normal tissues (Figs. 2A and S1).

The expression levels of the six upregulated DEGs associated with gemcitabine resistance were further validated in two stable gemcitabine-resistant CCA cell lines using RT-qPCR. As shown in Fig. 2B, RTKN2 was significantly elevated in stable gemcitabine-resistant CCA cell lines, namely KKU-213A-GemR and KKU-100-GemR, compared with their parental CCA cell lines (KKU-213A and KKU-100). However, RAB1B, JAG1, PACS1, ANKS6, and SF3B4 expression levels were not significantly altered in the two stable gemcitabine-resistant CCA cell lines.

To further verify the effect of gemcitabine treatment on RTKN2 expression in CCA cells, parental CCA cell lines, KKU-213A and KKU-100, were treated with gemcitabine at IC₅₀ concentrations for 48 h. Specifically, KKU-213A cells were treated with 0 and 0.72 µM gemcitabine, while KKU-100 cells were treated with 0 and 0.102 µM gemcitabine. As shown in Fig. 2C, RTKN2 mRNA expression was significantly upregulated in cells treated with gemcitabine in both cell lines. These findings revealed that RTKN2 is inducible by gemcitabine exposure and may contribute to the adaptive resistance mechanism in CCA cells.

Genes correlated with RTKN2 expression were identified using the TCGA Cholangiocarcinoma dataset via the cBioPortal platform. A total of 210 genes were found to be significantly correlated with RTKN2 expression (q-value <0.05) (Table SI). Pathway enrichment analysis demonstrated that genes negatively associated with RTKN2 showed statistically significant enrichment in several metabolic and mitochondrial pathways, such as ‘aerobic respiration and respiratory electron transport’, ‘respiratory electron transport’, ‘TP53 regulates metabolic genes’, and ‘complex IV assembly’. These findings suggest that high RTKN2 expression may be associated with the downregulation of mitochondrial oxidative phosphorylation and TP53-mediated metabolic control, implicating RTKN2 in metabolic reprogramming in CCA. However, no statistically significant pathways were identified in genes positively associated with RTKN2 (Table III).

The 3D structure of RTKN2 was constructed using AlphaFold3 (Fig. S2). Due to the lack of reported ligand-binding sites for RTKN2, the CavityPlus web server was utilized to predict potential ligand-binding pockets on the RTKN2 protein. The results showed that there were 31 possible ligand-binding sites on RTKN2 (Table SII). Among these, the binding pockets 2, 5, and 9 of RTKN2 exhibited strong druggability, with a drugscore of 4,037, 3,544, and 1,062, respectively (Fig. 3). Therefore, these three binding pockets of RTKN2 were selected for further study.

To identify potential small-molecule inhibitors of RTKN2, docking-based virtual screening from 5,396 compounds available in the drug repurposing compound library plus was performed on three possible binding sites of RTKN2 using AutoDock Vina. As shown in Fig. 3 and Table IV, the potential screened inhibitors for the binding site 2 included MK-3207 (-10.49 kcal/mol), phthalocyanine (-11.40 kcal/mol), and orforglipron (-11.47 kcal/mol). The best-scoring compounds for the binding pocket 5 included NKP608 (-10.94 kcal/mol), tegatrabetan (-11.10 kcal/mol), and MK-3207 (-11.31 kcal/mol). The best-scoring compounds for the binding pocket 9 included umbralisib (-10.62 kcal/mol), MK-3207 (-10.66 kcal/mol), and vepdegestrant (-10.91 kcal/mol). Notably, MK-3207 was found to be a potential inhibitor at all three binding sites of RTKN2. Taken together, the docking findings revealed the top three candidate compounds at each ligand-binding site, with binding affinities ranging from -10.49 to -11.47 kcal/mol, suggesting them as potential RTKN2 inhibitors.

The top-ranking compounds exhibited multiple favorable interactions with RTKN2, including hydrogen bonds, halogen interactions, π interactions, electrostatic attractions, and, in particular, van der Waals forces. Comprehensive 2D interaction profiles for each ligand-binding site (Fig. 4) revealed specific interactions between the functional groups of the ligand and key amino acid residues of RTKN2. The Asp154 and Lys415 residues at the binding site 2 of RTKN2 were found to electrostatically form π-anion and π-cation interactions with the aromatic and alkyl moieties of MK-3207, phthalocyanine, and orforglipron. In addition, the fluorine atoms of MK-3207 and orforglipron formed halogen interactions with i) Leu439 and Glu440 and ii) Asp154 and Thr425 residues, respectively. For ligand-binding site 5, instead of forming halogen interactions, the fluorine atoms of MK-3207 potentially formed hydrogen bonds with Gln479, Ser491, and Asp496 residues. By contrast, the fluorine atoms of NKP608 formed halogen forces with Glu452 and Gln479 residues. The Ile446, Lys449, Phe457, Pro469, Leu473, and Ile482 residues were found to be hotspots for the binding of NKP608, tegatrabetan, and MK-3207 via several π interactions. In the case of binding site 9, the aromatic moieties of all screened ligands were stabilized by alkyl/π-alkyl interactions and van der Waals forces. Notably, the fluorine atoms of MK-3207 formed halogen interactions with Lys585 and Pro586 residues. Moreover, this compound potentially formed a hydrogen bond with Pro9 residue. The amino group and the fluorine atom of umbralisib potentially formed hydrogen bonds with Thr580, Lys585, Pro586, and Ile595 residues. Similarly, vepdegestrant formed two hydrogen bonds with Arg7 and Thr83 residues. These results suggest that the identified compounds hold promise as lead candidates for the development of RTKN2-targeted inhibitors. By potentially disrupting RTKN2 function, these compounds may offer a novel therapeutic approach to overcome gemcitabine resistance in CCA.