The tissue microarray was constructed by Outdo Biotech Company (Shanghai, China) from 190 CCA patients who underwent surgical procedures in 2006 to 2017 at The First Affiliated Hospital of Nanjing Medicine University. The expression of ROBO1 was evaluated by the grade semiquantitative scoring system. The intensity was classified as negative (0), weak (1), moderate (2), or strong (3), and the density of positive cells in the target region was scored as follows: 0–5% (0), 6–35% (1), 36–70% (2), and >70% (3). The overall score was calculated by multiplying intensity and density. Each microarray tissue point was scored by two independent pathologists, and the average score was taken as the final score. The score of tumor tissue/normal tissue >1 was defined as upregulated; otherwise, it was defined as downregulated.

The patients were followed up regularly until death or October 25, 2019. The use of clinical samples was approved by the Ethics Committee of The Affiliated Hospital of Nanjing Medical University. Written informed patient consent was obtained in accordance with regional regulations.

Human CCA cell lines RBE, HCCC9810, QBC939, and HuCCT1 and human umbilical vein endothelial cell line (HUVEC) were purchased from the Chinese National Human Genome Center (Shanghai, China). The cholangiocarcinoma cell lines were cultured in Dulbecco’s modified Eagle’s medium (Gibco, USA) supplemented with 10% fetal bovine serum (Gibco) and antibiotics (1% penicillin/streptomycin; Gibco). The HUVECs were cultured in F12-K medium (Gibco) containing 10% fetal bovine serum.

Six-week-old male BALB/c nude mice (GemPharmatech Co., Ltd., China) were purchased to perform animal studies. Suspensions containing 1.0 × 10⁷ RBE cells were injected subcutaneously into the groin area of nude mice. The tumors were measured with a caliper every week until these were harvested 6 weeks later. Tumor volume is calculated according to the following formula: volume = width² × length/2.

In total, 200 μl Matrigel (BD Bioscience, USA) was added to a 24-well plate and incubated at 37°C for 30 min. The HUVECs were resuspended with conditioned medium (CM) derived from tumor cells, and 1.0 × 10⁵ HUVECs were seeded into each well of the 24-well plate. Images were taken 8 h later.

The chicken chorioallantoic membrane (CAM) assay was performed using specific pathogen-free fertilized chicken eggs (Jinan SAIS Poultry Co., Ltd, China). The eggs were incubated at 37°C under 80% humidity for 6 days. A 1-cm-diameter window was opened on the eggshell. The surface of the air sac floor was removed to expose the CAM. A 0.5-cm-diameter filter paper was placed onto the CAM, and 100 μl conditioned medium was added onto the filter paper before the window was closed with a sterile adhesive tape. After 3 days of incubation, the CAM was fixed with stationary solution (methanol/acetone = 1:1) and harvested. The photos of the CAMs were taken using a digital camera, and second- and third-order vessels were counted to assess the effects of the conditioned medium on angiogenesis.

The ROBO1 wild type and c.838G>T (p.E280*) mutation overexpression lentiviruses were constructed by Genechem Co., Ltd. (Shanghai, China). The RBE and HCCC9810 cells were seeded in 6-well plates and then infected with lentivirus according to the manufacturer’s protocol. After 48 h, the medium was replaced with complete medium. The infected cells were treated with 10 μg/ml puromycin in order to select stable overexpressing cells. ROBO1 wild type and c.838G>T overexpression in RBE and HCCC9810, respectively, were confirmed by RT-qPCR and Sanger sequencing (Genewiz, China).

Total RNA was extracted from stable negative control, ROBO1^WT, and ROBO1^E280* groups of RBE cells using TRIzol reagent (Invitrogen, California, USA) according to the manufacturer’s protocol. The concentration and the purity of total RNA were checked by NanoDrop 2000 Nucleic Acid and Protein Analyzer (Thermo Scientific, MA, USA). RNA-Seq analysis was performed by Shanghai Personalbio Technology Co., Ltd. (Shanghai, China).

All statistical analyses were performed with SPSS v24.0 (IBM, SPSS, Chicago, IL, USA) and Graphpad Prism 8 (GraphPad Software, La Jolla, CA, USA). Differences between the two groups were analyzed by Student’s t-test. Moreover, χ ² test was used to analyze correlations between ROBO1 expression and clinicopathological variables. Overall survival (OS) and disease-free survival were analyzed with Kaplan–Meier methods, and log-rank test was applied for comparison. A multivariate analysis was performed using Cox proportional hazard regression model. Differences were considered as statistically significant when *p <0.05, **p < 0.01, or ***p <0.001. Supplementary information about materials and methods is provided in Supplementary Material .

To demonstrate the genomic landscapes of CCA, we performed whole-exome sequencing of 67 CCAs (including 43 pCCAs and 24 iCCAs). We identified significant mutated genes, including MACF1, METTL14, ROBO1, and so on. Furthermore, we performed combined annotation–dependent depletion (CADD) to evaluate the potential deleteriousness of gene mutations and listed them in descending order ( Supplementary Table S1 ). ROBO1 p.E280* (c.838G>T, mutation rate 2.99%, found in 2 iCCAs of 67 CCAs) ranked first, with a CADD score of 38. The same mutation of ROBO1 also occurred in colorectal cancer cell line HT115 according to the COSMIC database. The ROBO1 p.E280* mutation induced a premature stop codon in the position of the 280th aa and may encode a truncated protein of ROBO1. The E280* mutant contains Ig1-2 domains and a part of Ig3 domain of the extracellular region, with the rest of the extracellular region and all the transmembrane and intracellular domains untranslated ( Figure 1A ). These observations indicated that the truncated mutations of ROBO1 may be highly deleterious in the progression of carcinogenesis, but the causality between this mutation and CCA has not yet been established.

Firstly, we explored the expression file of wild-type ROBO1 in CCA patients. The RT-qPCR of 99 pairs of CCA patients showed that ROBO1 mRNA expression was significantly downregulated in CCA tissues compared to their adjacent normal tissues ( Figure 1B ). Furthermore, the protein expression of ROBO1 was also significantly downregulated in CCA tissues ( Figure 1C ). To investigate the clinical significance, we analyzed the expression levels of ROBO1 by tissue microarray in 190 CCA patients. Representative immunohistochemical staining images of ROBO1 high/low expression patterns are shown in Figure 1D . The detailed immunohistochemistry (IHC) scores of the 190 pairs of tissues are shown in Supplementary Figure S1 . As shown in Figure 1E , ROBO1 was downregulated in 63.7% (121 of 190) of CCAs compared with normal tissues.

The association between ROBO1 protein expression and the clinicopathological characteristics of CCA was analyzed and is shown in Supplementary Table S2 . ROBO1 protein expression was significantly correlated with patient’s gender (p = 0.006) and marginally significantly correlated with serum CA199 (p = 0.087), tumor location (p = 0.068), T stage (p = 0.069), and clinical stage (p = 0.061). Besides these, the univariable analysis showed that ROBO1 expression, histological grade, tumor thrombus, nerve invasion, N stage, and clinical stage were significantly associated with the OS of CCA patients. Furthermore, the multivariate analysis showed that ROBO1 expression was an independent risk factor in patients with CCA ( Table 1 ). The Kaplan–Meier survival curves demonstrated that CCA patients with upregulated ROBO1 had longer postoperative overall survival and disease-free survival than those with ROBO1 downregulated expression ( Figures 1F, G ). These results collectively suggested that the downregulation of ROBO1 may have a stimulatory role in the progression of CCA and predict poor survival.

The mRNA and protein expression levels of ROBO1 in CCA cell lines were detected by RT-qPCR and Western blot ( Supplementary Figures S2A, B ). RBE and HCCC9810 cells were used to detect the functional role of ROBO1^WT and ROBO1^E280* due to the relatively low expression and no congenital E280* mutation ( Supplementary Figure S2C ). Stably overexpressing ROBO1^WT and ROBO1^E280* cells were constructed with lentivirus transfection, and the transfection efficiency was confirmed by RT-qPCR and WB ( Supplementary Figures S2D, E ). As shown in Supplementary Figure S2E , ROBO1^E280* encoded a truncated protein (predicted molecular weight, 31 kDa; observed molecular weight, 43 kDa).

To determine the role of wild-type ROBO1 and its E280* mutation in the proliferation of CCA cell lines, a clone formation assay showed that, compared to the NC group, the number of cell colonies decreased significantly in ROBO1-overexpressing cells, which was reversed by E280* mutation ( Figure 2A ). Consistently, in CCK8 assays, ROBO1 overexpression significantly decreased the absorbance at OD450, while cells with E280* mutations showed a significantly higher absorbance than wild-type ROBO1 ( Figure 2B ). In addition, flow cytometry assays showed that the percentage of cells in the G0–G1 phase was significantly higher in the ROBO1^WT overexpression group, while the percentage of cells in the S and G2–M phase decreased significantly. The E280* mutation reverses the phenotype of wild-type ROBO1 in the flow cytometry assays ( Figure 2C , Supplementary Figure S3 ). The EdU assays also revealed that ROBO1 could inhibit the proliferation of CCA cells, while the mutant protein lost the function ( Figure 2D ). Furthermore, proliferation-related proteins cyclin D1 and PCNA were increased in ROBO1^E280* compared with ROBO1^WT-overexpressing cells ( Figure 3C ). These data indicated that ROBO1^E280* represses the tumor-suppressing effects of ROBO1^WT on CCA cell proliferation.

Furthermore, we detected the effects of ROBO1^WT and ROBO1^E280* on tumor angiogenesis using HUVEC tube formation and CAM assays. As shown in Figure 3A , ROBO1^WT overexpression in CCA cells suppressed tube formation of HUVECs, while E280* lost the function of anti-angiogenesis. Similarly, the conditioned medium derived from ROBO1^WT cells inhibited angiogenesis in the CAM assays, while CM derived from ROBO1^E280* partially lost the anti-angiogenic effects ( Figure 3B ). In addition, the downregulation of angiogenesis-related proteins VEGF and MMP2 in ROBO1^WT cells was reversed by ROBO1^E280* ( Figure 3C ).

Subsequently, experiments in vivo were performed to evaluate the effects of ROBO1^WT and ROBO1^E280* on angiogenesis and tumorigenesis. The RBE cells stably transfected with LV-ROBO1^WT, LV-ROBO1^E280*, and LV-NC were injected subcutaneously into nude male mice. The results showed that ROBO1 overexpression significantly inhibited tumor growth in vivo, which was partially reversed by ROBO1^E280* ( Figures 3D – F ). IHC staining of xenografts showed that the expression of Ki67 and VEGF was decreased in the ROBO1^WT group, while it was recovered in the ROBO1^E280* group ( Figure 3G ). Besides this, higher microvascular density was observed in ROBO1^E280* compared with the ROBO1^WT group, using CD31 staining ( Figure 3G ). Taken together, these results indicated that ROBO1^E280* loses the inhibitory effects of ROBO1^WT on angiogenesis and tumorigenesis of CCA.

To explore the mechanism under the loss-of-function phenotype of ROBO1^E280*, we performed subcellular fractionation analysis to identify the subcellular localization of the truncated protein. Western blotting analysis revealed that ROBO1^WT was detected on the cytomembrane, while ROBO1^E280* was found in the cytoplasm ( Figure 4A ). Immunofluorescence staining further confirmed the translocation of ROBO1^E280* ( Figure 4B ). Given that SLIT2 ligands interact with ROBO1 receptors on the cytomembrane, we thus performed co-immunoprecipitation assays to examine whether the SLIT2/ROBO1 interaction was affected by E280* mutation. The anti-SLIT2 antibody specifically co-immunoprecipitated ROBO1^WT, while ROBO^E280* was not detected in the immunoprecipitates of cells overexpressing ROBO1^E280* ( Figure 4C ), which revealed that ROBO1^E280* failed to interact with SLIT2. According to these results, the SLIT2/ROBO1 transmembrane signal transduction is interrupted by the E280* mutation of ROBO1.

Furthermore, we performed functional experiments to verify the interruption of SLIT2/ROBO1 signal by E280* mutation of ROBO1. As shown in Figure 5A , SLIT2 treatment significantly decreased the number of colonies in ROBO1^WT cells, while no significant decrease was observed in ROBO1^E280* cells. Consistent with the clone formation assays, the absorbance at OD450 was significantly decreased with SLIT2 treatment in ROBO1^WT cells, while SLIT2 treatment in ROBO1^E280* cells showed no significant differences ( Figure 5B ). The flow cytometry cell cycle analyses and EdU assays demonstrated similar results ( Figure 5C and Supplementary Figures S4A, B ). Furthermore, SLIT2-treated ROBO1^WT CCA cells significantly decreased HUVEC tube formation compared with untreated cells, which is not observed in ROBO1^E280* cells ( Figure 5D ). Consistent with the functional experiments, western blotting analysis showed that cyclin D1, PCNA, VEGF, and MMP2 were significantly downregulated in SLIT2-treated ROBO1^WT cells, which was not observed in SLIT2-treated ROBO1^E280* cells ( Figure 5E ). These results revealed that ROBO1^E280* disrupts the SLIT2-induced tumor-suppressing effects on proliferation and angiogenesis in CCA.

To explore the potential target of ROBO1^WT and ROBO1^E280*, we conducted RNA-seq in negative control, ROBO1^WT, and ROBO1^E280* cells. We screened a set of target genes whose expression was reversed in ROBO1^E280* cells compared with ROBO1^WT cells ( Figure 6A ). Among these genes, RT-qPCR and western blotting analysis verified that OLFML3, which plays a role in tumor-related angiogenesis (21, 22), was decreased in the ROBO1^WT group, while it was recovered in the ROBO1^E280* group ( Figures 6B, C ). Subsequently, we detected OLFML3 mRNA expression in clinical samples. The RT-qPCR of 99 pairs of CCA tissues demonstrated that OLFML3 mRNA expression was significantly upregulated in CCA tissues compared with normal tissues ( Figure 6D ). Figure 6E showed the typical OLFML3 protein expression in CCA tissues and matched normal tissues. Moreover, linear regression analysis revealed that OLFML3 mRNA expression was significantly negatively correlated with ROBO1 expression ( Figure 6F ).

Next, we performed a series of functional experiments to elucidate the role of OLFML3 in the proliferation and angiogenesis of CCA. The knockdown efficiency of OLFML3 siRNA was verified using RT-qPCR and western blotting analysis ( Figures 6G, M ). As shown in Figure 6H , the knockdown of OLFML3 significantly decreased the colony numbers compared with the control cells. In CCK8 assays, the absorbance of OD450 was also decreased in OLFML3-knockdown cells ( Figure 6I ). In addition, the percentage of G0–G1 phase cells increased, while that of S phase cells decreased in OLFML3-knockdown cells compared with control cells. In HCCC9810 OLFML3-knockdown cells, the percentage of G2–M cells also showed a significant decrease ( Figure 6J and Supplementary Figure S5A ). Furthermore, the EdU assays showed that the percentage of EdU-positive cells was reduced in OLFML3-knockdown cells compared with the control cells ( Figure 6K ). Subsequently, tube formation assays demonstrated that the depletion of OLFML3 reduced the angiogenic capacity of HUVECs ( Figure 6L ). Accordingly, western blotting analysis revealed that the expression of cyclin D1, PCNA, VEGF, and MMP2 was downregulated by the knockdown of OLFML3 ( Figure 6M ). These results indicated that OLFML3 may be a target gene of ROBO1 and act as a potential oncogene of CCA.

Subsequently, we knocked down OLFML3 in ROBO1^E280* cells and examined whether the knockdown of OLFML3 could restore the lost tumor-suppressing effects of ROBO1^E280*. In clone formation assays, the knockdown of OLFML3 decreased the increased colony number in ROBO1^E280* cells ( Figure 7A ). Similar results were observed in CCK8 assays ( Figure 7B ). In cell cycle assays, the depletion of OLFML3 in ROBO1^E280* elevated the proportion of cells in G0–G1 phase and decreased that of cells in S phase ( Figure 7C and Supplementary Figure S6A ). Besides this, the EdU assays showed that the depletion of OLFML3 partly decreased the EdU-positive proportion in ROBO1^E280* cells ( Figure 7D ). Moreover, the enhanced angiogenic capacity of ROBO1^E280* was reversed by the depletion of OLFML3 ( Figure 7E ). Western blotting analysis showed that the increased expressions of cyclin D1, PCNA, VEGF, and MMP2 in ROBO1^E280* cells were reversed by the knockdown of OLFML3 ( Figure 7F ). Taken together, the above-mentioned results revealed that OLFML3 is essential for ROBO1^E280*-induced proliferation and angiogenesis in CCA.