Gene expression profiles data from TCGA and GEO databases were retrieved for analysis. The GSE18520 and GSE26712 datasets (https://www.ncbi.nlm.nih.gov/geo) were obtained from GEO. GSE18520 included 10 normal tissues and 53 ovarian cancer samples, while GSE26712 comprised 10 normal ovarian tissues and 185 ovarian cancer samples. Clinical data from TCGA (https://www.cancer.gov/ccg/research/genome-sequencing/tcga/studied-cancers)—including patient age, clinical stage, pathological grade, lymph node and vein infiltration, vascular invasion, and survival outcomes—were extracted from the database and linked to 620 tumor samples and 10 normal samples. Due to the significant disparity in the number of tumor samples and normal samples in the TCGA database, calculations such as the odds ratio (OR) may introduce substantial bias. Therefore, TCGA data is primarily utilized in the field of tumor genomic analysis. Additionally, single-cell sequencing data from GSE184880, consisting of 7 ovarian tumor tissues and 5 adjacent non-tumor tissues, were incorporated into the integrative analysis.

We initially screened 14 ovarian cancer gene expression datasets from GEO, including GSE3149, GSE9891, GSE26712, GSE14764, GSE15622, GSE18520, GSE19829, GSE23554, GSE26193, GSE27651, GSE30161, GSE51373, GSE63885, and GSE65986. Among these, only GSE18520 and GSE26712, along with TCGA dataset, included matched normal and tumor tissue samples and were therefore selected for downstream analysis. For each of the three datasets, univariate logistic regression analyses were performed in R to evaluate the association between CLDN gene expression (dichotomized at the median) and disease status (tumor vs. normal). Odds ratios (ORs) and corresponding p-values were calculated, and genes showing statistically significant associations (p 0.05) in all three datasets were considered consistently dysregulated. Prognostic hazard ratio (HR) for claudin family members was extracted from the KM Plotter database (https://kmplot.com/analysis/index.php?p=service&cancer=ovar). A meta-analysis was conducted using a random-effects model, and forest plots were generated with Revman software (5.4, The Cochrane Collaboration, London, UK). Genes with hazard ratios significantly different from 1 (95% CI not including 1) were deemed prognostically relevant. Additionally, receiver operating characteristic (ROC) curves were constructed for each CLDN gene using the pROC package in R. Genes with an area under the curve (AUC) greater than 0.7 in all three datasets were identified as having strong diagnostic potential.

Using DEcenter, we identified significantly elevated expression of claudins in ovarian tumor tissues compared to normal tissues. To further examine the expression patterns of CLDN genes at single-cell resolution, we analyzed a publicly available scRNA-seq dataset (e.g., GSEXXXXXX) downloaded from GEO, preprocessing the data with the Seurat package in R [33]. Quality control was performed by excluding cells with fewer than 200 or more than 10,000 detected genes, as well as those with greater than 5% mitochondrial read content. Additionally, only genes expressed in at least three cells were retained for downstream analysis. The filtered data were normalized and corrected for batch effects before undergoing dimensional reduction. Principal component analysis (PCA) was conducted to capture major sources of variation, and the top principal components were used as input for Uniform Manifold Approximation and Projection (UMAP) to generate a two-dimensional embedding for visualization. Cell clusters were identified using the FindClusters algorithm, and their identities were annotated by comparing cluster-specific expression profiles to reference bulk RNA-seq datasets from Blueprint and ENCODE, using the SingleR package (Dvir Aran lab, Israel) [34]. Finally, the expression patterns of target genes—including CLDN6, CLDN9, and CLDN10—were examined across distinct cell populations to determine their cellular context and potential functional roles.

To further explore the functional roles of CLDN6, CLDN9, and CLDN10 in OC at the transcriptome level, we dichotomized tumor samples into high- and low-expression groups based on the median expression level of each gene. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were then performed using the clusterProfiler R package. To investigate intercellular communication patterns at single-cell resolution, we applied the “CellChat” R package and the CellchatDB human database (https://rdrr.io/github/sqjin/CellChat/man/CellChatDB.human.html) to infer ligand–receptor-mediated signaling networks. Overlapping enriched pathways identified across multiple CLDN-associated analyses were selected for further evaluation. Finally, Gene Set Enrichment Analysis (GSEA, v4.3.2, Broad Institute, Cambridge, MA, USA, https://www.gsea-msigdb.org/gsea) was conducted to assess the enrichment of these candidate pathways in high-versus low-expression groups, thereby evaluating the functional impact of CLDN6, CLDN9, and CLDN10 expression on key biological processes.

ROC curves and corresponding area under the curve (AUC) values were generated for a multi-gene signature combining CLDN6, CLDN9, and CLDN10, and compared with those of established clinical biomarkers—including carbohydrate antigen 125 (CA125), human epididymis 4 (HE4), carcinoembryonic antigen (CEA), and alpha-fetoprotein (AFP)—in three independent datasets.

All 8 tissue samples were collected from the Gynecology Department Specimen Repository of the First Hospital of Lanzhou University. The present study received ethical approval from the First Hospital of Lanzhou University. Written informed consent approving this study was obtained from each patient.

The human ovarian epithelial carcinoma cell lines OV-90 and SK-OV-3 were obtained from the Cell Bank of Type Culture Collection of the Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences (Shanghai, China). The luciferase-expressing SK-OV-3 (Luc1) cell line was purchased from Meisen Chinese Tissue Culture Collections (Zhejiang, China). All cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM; Invitrogen, Waltham, MA, USA), supplemented with 10% fetal bovine serum (FBS) and 1% penicillin-streptomycin solution (both from Invitrogen). Cells were maintained at 37 °C in a humidified atmosphere containing 5% CO₂ and were routinely passaged when reaching approximately 80% confluence. All cell lines were validated by STR profiling and tested negative for mycoplasma.

Specific small interfering RNAs (siRNAs) targeting CLDN6, CLDN9, and CLDN10, along with a non-targeting siRNA as a negative control (NC), were synthesized by GenePharma (Shanghai, China). For stable knockdown, a lentiviral vector encoding short hairpin RNA (shRNA) against CLDN6 was also obtained from GenePharma. Transient transfections were performed in OV-90 and SK-OV-3 cells using Lipofectamine 2000 (Invitrogen) according to the manufacturer’s instructions. Knockdown efficiency was validated 48 hours post-transfection by RT-qPCR, and cells were subsequently harvested for functional assays. To generate stable CLDN6-knockdown OV-90 (Luc1) cells, lentiviral particles were used to infect target cells, followed by puromycin selection. The resulting cell line was employed to establish both subcutaneous xenograft tumors and experimental lung metastasis models in immunodeficient mice.

Total RNA was extracted from cultured cells using TRIzol reagent (Invitrogen), according to the manufacturer’s instructions. One microgram of total RNA was reverse-transcribed into cDNA using the miScript II Reverse Transcription Kit (218161, Qiagen, Germantown, MD, USA). Quantitative real-time PCR (qPCR) was performed using the miScript SYBR Green PCR Master Mix (Qiagen) on an ABI Prism 7900HT Sequence Detection System (Applied Biosystems). Each reaction was carried out in a final volume of 20 µL and included 40 cycles of amplification under the following thermal cycling conditions: 95 °C for 3 min (initial activation), followed by 95 °C for 15 sec, 58 °C for 30 sec, and 72 °C for 7 min per cycle. The relative mRNA expression levels were calculated using the 2^{-Δ⁢Δ⁢Cq} method, with GAPDH or ACTB used as endogenous reference genes.

Primer sequence:

CLDN6 F: 5^′-TGTTCGGCTTGCTGGTCTAC-3^′

R: 5^′-CGGGGATTAGCGTCAGGAC-3^′

CLDN9 F: 5^′-ATGCAGTGCAAGGTGTACGA-3^′

R: 5^′-ATCAGGCCAAGGTCGAAAGG-3^′

CLDN10 F: 5^′-CTGTGGAAGGCGTGCGTTA-3^′

R: 5^′-CAAAGAAGCCCAGGCTGACA-3^′

GAPDH F: 5^′-ACAACTTTGGTATCGTGGAAGG-3^′

R: 5^′-GCCATCACGCCACAGTTTC-3^′

Formalin-fixed, paraffin-embedded tissue sections (4 µm thickness) were deparaffinized in xylene and rehydrated through a graded ethanol series (100%, 95%, and 70%) for 5 minutes each. Antigen retrieval was performed via heat-induced epitope retrieval (HIER) using a pressure cooker at 120 °C for 15 minutes in Tris-EDTA buffer (10 mM Tris, 1 mM EDTA, pH 8.0). Endogenous peroxidase activity was blocked with 3% hydrogen peroxide for 10 minutes at room temperature. Sections were incubated overnight at 4 °C with the primary antibody (CLDN6 1:200, 75055; CLDN9 1:500, 192398; CLDN10 1:400, 52234), followed by incubation with a horseradish peroxidase (HRP)-conjugated secondary antibody (Goat Anti-Rat IgG (H+L), 1:4500, AS106, ABCLonal, Wuhan, Hubei, China). Immunoreactivity was visualized using 3,3^′-diaminobenzidine (DAB) as chromogen, and nuclei were counterstained with Gill’s Hematoxylin III (Carl Roth). The antibodies for CLDN6/CLDN9/CLDN10 are all provided by Abcam Company (Cambridge, UK). After brief bluing in distilled water, sections were dehydrated and mounted with EcoMount (Biocare Medical) for microscopic examination.

For the wound healing assay, cells were seeded at a density of 1 $\times$10⁵ cells per well in 6-well plates and cultured overnight to achieve confluent monolayers. Wounds were created by scratching the cell layer with a sterile 200-µL pipette tip. Loosely adherent cells were removed by gently washing three times with phosphate-buffered saline (PBS), and serum-free medium was added to minimize cell proliferation. Images were acquired at the time of scratching (0 h) and 24 hours later at identical locations. The extent of wound closure was analyzed using ImageJ software (version 1.53t, National Institutes of Health, Bethesda, MD, USA), and the relative migration rate was calculated as follows: [(A₀ – A₂₄)/A₀] $\times$ 100%, where A₀ and A₂₄ represent the wound area at 0 h and 24 h, respectively.

For the Transwell migration assay, a polycarbonate membrane insert with an 8 µm pore size (Corning) was used. Ovarian cancer cells (5 $\times$ 10⁴ per well) were seeded into the upper chamber in serum-free medium, while complete medium containing 10% fetal bovine serum (FBS) was placed in the lower chamber as a chemoattractant. After incubation for 24 hours, non-migrated cells on the upper surface were removed with a cotton swab, and migrated cells on the lower surface were fixed with methanol, stained with crystal violet, and imaged under a light microscope (IX-73, Olympus, Tokyo, Japan). The number of migrated cells was quantified from five random fields per chamber using ImageJ software. Experiments were performed using CLDN6-knockdown cells and negative control cells to assess the effect of CLDN6 silencing on migratory capacity.

Total proteins were extracted from ovarian cancer tumor tissues, matched adjacent non-tumor tissues, and OC cell lines. Equal amounts of protein lysates were separated by SDS-PAGE and transferred onto polyvinylidene fluoride (PVDF) membranes. After blocking with 5% non-fat milk in Tris-buffered saline containing 0.1% Tween-20 (TBST) for 1 hour at room temperature, membranes were incubated overnight at 4 °C with primary monoclonal rabbit antibodies against the target protein (e.g., anti-CLDN6, 1:1000, 75055). Following three washes with TBST, membranes were incubated with horseradish peroxidase (HRP)-conjugated goat anti-rabbit IgG secondary antibody (1:4500, AS106, ABCLonal) for 1 hour at room temperature. After additional TBST washes, protein bands were visualized using an enhanced chemiluminescence (ECL) detection reagent according to the manufacturer’s instructions.

Apoptosis was assessed using an Annexin V-FITC/Propidium Iodide (PI) apoptosis detection kit (AP101, MultiSciences, Hangzhou, Zhejiang, China), according to the manufacturer’s instructions. Briefly, harvested cells were washed once with cold phosphate-buffered saline (PBS), then resuspended in Annexin V Binding Buffer. Aliquots of 100 µL cell suspension were incubated with 5 µL Annexin V-FITC and 5 µL PI for 15 minutes at room temperature (approx. 25 °C) in the dark. Following staining, 400 µL of Annexin V Binding Buffer was added directly to each sample without washing to maintain cell viability and prevent loss of early apoptotic cells. Samples were analyzed within 1 hour by flow cytometry using a Longcyte flow cytometer (C2060, Challenbio, Beijing, China). Fluorescence signals were detected in the FL1 (FITC) and FL2 (PI) channels. Data were processed using FlowJo v10 software (BD Life Sciences, Franklin Lakes, NJ, USA), with apoptotic cells defined as Annexin V-positive (early and late apoptosis). All experiments were performed in biological triplicate.

Cells were harvested, washed in PBS, and fixed in cold 70% ethanol for 30 minutes at 4 °C. Fixed cells were washed and stained with propidium iodide (MultiSciences) for 15 minutes in the dark. Samples were analyzed immediately using a flow cytometer. DNA content was measured ($\ge$10,000 events/sample), and cell cycle distribution (G0/G1, S, G2/M phases) was quantified using FlowJo software. Experiments were performed in triplicate.

Cells were harvested, washed in PBS, and resuspended in warm culture medium. JC-1 dye (final concentration: 5–10 µM; MultiSciences) was added, and cells were incubated for 15–30 minutes at 37 °C in the dark. Cells were washed with PBS, resuspended, and analyzed immediately by flow cytometry. Fluorescence was detected in the FL1 (green monomers, low $\mathrm{\Delta }$$\mathrm{\Psi }$m) and FL2 (red aggregates, high $\mathrm{\Delta }$$\mathrm{\Psi }$m) channels. The FL2/FL1 ratio quantified $\mathrm{\Delta }$$\mathrm{\Psi }$m. Carbonyl cyanide m-chlorophenylhydrazone (CCCP)-treated cells served as depolarized controls. At least 10,000 events per sample were acquired. Data were analyzed using FlowJo software. Experiments were performed in triplicate.

Cells were harvested, washed with PBS, and incubated with 10 µM DCFH-DA (2^′,7^′-dichlorodihydrofluorescein diacetate; Beyotime, Shanghai, China) for 30 minutes at 37 °C in the dark. After washing, cells were resuspended in fresh medium and incubated for 15 minutes at 37 °C to allow hydrolysis of DCFH-DA to DCFH. Reactive oxygen species (ROS) oxidized non-fluorescent DCFH to fluorescent DCF. Samples were analyzed immediately by flow cytometry. Data were analyzed using FlowJo software. Experiments were performed in triplicate.

Female nude mice (4–6 weeks old) were randomly assigned to subcutaneous xenograft tumor and lung metastasis model groups. CLDN6-knockdown OV-90 cells were injected subcutaneously or via the tail vein. Each mouse received a subcutaneous inoculation of 2 $\times$ 10⁶ cells in 200 µL of PBS. Tumor volume was calculated using the formula V = (a $\times$ b²)/2, where a represents the longest diameter and b the shortest. Metastatic dissemination was monitored using an in vivo small-animal imaging system (PerkinElmer, Waltham, MA, USA). The method of euthanasia for the animals after the experiment was carried out by using an excessive dose (In the gas anesthesia machine, after induction of anesthesia with 5% isoflurane, 1.5% isoflurane was used to maintain anesthesia. At the time of euthanasia, 8% isoflurane was inhaled for 10 minutes until death was confirmed.) of isoflurane through an inhalation anesthesia machine.

The DECenter platform was used to analyze transcriptomic differences and identify gene expression disparities between cancerous and adjacent non-tumor tissues. Additionally, differential gene expression analysis was conducted using the limma package in R, with results considered statistically significant at $|$log₂FC$|$ $>$1 and adjusted p-value 0.05.

By integrating transcriptomic data from TCGA and two GEO datasets (GSE18520 and GSE26712), we identified multiple differentially expressed CLDN genes across cohorts. In the TCGA cohort, 9 CLDN genes were significantly upregulated and 13 were downregulated; in GSE18520, 12 were upregulated and 5 were downregulated; and in GSE26712, 8 were upregulated and 3 were downregulated (Fig. 1). Cross-dataset integration revealed five CLDN genes—namely CLDN4, CLDN6, CLDN9, CLDN10, and CLDN16—that were consistently upregulated and significantly associated with ovarian cancer status (odds ratio $>$1, p 0.05) across all three datasets (Table 1).

Fig. 1.

The differential expression of Claudin genes. (A) The volcano plot displays differential expression in GSE18520. (B) The heatmap displays differential expression in GSE18520. (C) The volcano plot displays differential expression in GSE26712. (D) The heatmap displays differential expression in GSE26712. (E) The volcano plot displays differential expression in TCGA. (F) The heatmap displays differential expression in TCGA. TCGA, The Cancer Genome Atlas.

Survival analysis using the Kaplan-Meier Plotter database, combined with meta-analysis, revealed that high expression of CLDN1, CLDN2, CLDN6, CLDN9, and CLDN10 showed high expression in tumors. The genes CLDN1, CLDN2, CLDN6, CLDN9, and CLDN10 all demonstrate high expression in tumors. However, findings from different studies and researchers vary: some indicate these genes are risk factors associated with shorter patient survival, while others suggest the opposite. Therefore, a meta-analysis is required to integrate and evaluate data from different studies and datasets, in order to validate which genes are more likely to be risk genes. Fig. 2 shows a Meta-analysis of different CLDN gene families. The results showed that CLDN2 and CLDN10 showed HR 1 (Fig. 2B,E), and others showed HR $>$1 (Fig. 2A,C,D). Heterogeneity was assessed using meta-analysis, followed by further validation with data from the GSE dataset to analyze whether the CLDN family genes and hazard ratio (HR) showed statistical significance.

Fig. 2.

Statistically significant hazard ratio (HR) values for CLDN genes. (A) CLDN1. (B) CLDN2. (C) CLDN6. (D) CLDN9. (E) CLDN10. CLDN, Claudins.

We calculated the HR values and p-values of CLDN1, CLDN2, CLDN6, CLDN9, and CLDN10 in different datasets within the GSE database, respectively. Then, Cox regression was employed to analyze the significance of different CLDN genes of HR in OC (Table 2). The results revealed marked contradictions in CLDN2, which was therefore subsequently excluded from further analysis. CLDN6, 9, and 10 demonstrated consistent associations across different study analyses, indicating enhanced reliability of these genes in subsequent investigations.

Through the TCGA and GSE dataset, we analyzed the differential expression of CLDN family genes in tumor and normal tissues, identified significantly highly expressed genes, and calculated the odds ratios (OR) of CLDN genes with statistical significance. Additionally, we collected literature and performed a meta-analysis to assess the heterogeneity of HR for the CLDN family across different studies, identifying CLDN family genes with pooled HR values greater than 1 or less than 1. Furthermore, using the GSE dataset, we conducted Cox regression analysis to further examine CLDN family genes with HR values greater than 1 or less than 1 and significant associations (p 0.05). Based on the above analyses, the intersection of these conditions was taken, ultimately identifying CLDN6, CLDN9, and CLDN10 for subsequent ROC validation and experimental verification analysis (Fig. 3).

Fig. 3.

Venn diagram illustrating the intersection of TCGA, GSE18520, and GSE26712 datasets.

In the diagnosis of ovarian cancer, there are four classic serum biomarkers: CA125, HE4, CEA, and AFP. This study plans to compare three CLDN family genes screened from public datasets with these classic serum biomarkers. The expression data of the three CLDN genes will be integrated to form a composite diagnostic and prognostic indicator for comparison and analysis. To prioritize candidate biomarkers, we implemented a three-dimensional screening strategy integrating odds ratio (OR), hazard ratio (HR), meta-analysis and area under the curve (AUC) values. We have established the following criteria for selecting the CLDN family genes: (1) OR $>$1 and p 0.05; (2) HR $>$1 or HR 1 and p 0.05; (3) AUC value of HR $>$0.7; (4) demonstration of validity in the meta-analysis results. The results showed that CLDN6, CLDN9, and CLDN10 were highly expressed in tumors, with OR $>$1 and p 0.05. Among them, CLDN10 had HR 1 and p 0.05, while CLDN6 and CLDN9 had HR $>$1 and p 0.05. Furthermore, the AUC values of all three genes consistently exceeded 0.7 across different datasets. These three genes met our screening criteria and were selected as candidate genes to proceed with model construction analysis and experimental validation.

First, we performed ROC curve analysis for CLDN6, 9, and 10 in different datasets separately, and the results showed that their AUC $>$0.75 (Fig. 4). And these three CLDN genes were integrated with diagnosis index as following.

Fig. 4.

The area greater than 0.7 under ROC curve of CLDN family in GSE18520, GSE26712 and TCGA dataset. (A) GSE18520 dataset. (B) GSE26712 dataset. (C) TCGA dataset. ROC, receiver operating characteristic.

diagnosis index (DI) =${\sum }_i{\beta }_i\times G_i$

In this formula, ${\beta }_{\text{i}}$ is the Cox regression coefficient of the ith CLDN genes, $G_{\text{i}}$ is the relative expression level of the ith gene. Subsequently, we compared the DI indicator (which integrates CLDN6, 9, and 10 genes) with individual classic serum biomarkers for analysis, with the results shown in Fig. 5. The results indicate that the combined DI index demonstrates superior performance compared to other traditional indicators. In Fig. 5A,B, the AUC of DI exceeds 0.99, demonstrating exceptionally high diagnostic efficacy. In Fig. 5C using the TCGA database, the AUC remains above 0.948. The results also reveal that among traditional diagnostic biomarkers, only HE4 maintains relatively stable diagnostic performance across multiple datasets, with results approaching those of the combined DI index.

Fig. 5.

The areas under ROC curves of our three combined genes and existing gynecological tumor markers in GEO database and TCGA. CA125, CEA, AFP, HE4 are currently available gynecological tumor markers. DI represents a combination of CLDN genes. (A) GSE18520 dataset. (B) GSE26712 dataset. (C) TCGA dataset. GEO, Gene Expression Omnibus; CA125, carbohydrate antigen 125; CEA, carcinoembryonic antigen; AFP, alpha-fetoprotein; HE4, human epididymis 4; DI, diagnosis Index.

Single-cell sequencing reveals cell type-specific mechanisms. Spatial expression localization: CLDN6/9/10 are highly expressed in the malignant epithelial cell subpopulation and are highly co-localized with the tumor core region (Fig. 6). Inter-cellular ligand-receptor analysis shows that CLDN6/9 activate the Wnt/$\beta$-catenin pathway (promoting metastasis). CLDN10 inhibits the Wnt signal (tumor suppressor effect) (Figs. 7,8).

Fig. 6.

The function and expression of Claudin proteins were validated at the single-cell level. (A) The expression of CLDN6 differs between normal tissues and tumors. (B) The expression of CLDN9 differs between normal tissues and tumors. (C) The expression of CLDN10 differs between normal tissues and tumors.

Fig. 7.

Expression levels of CLDN6, CLDN9 and CLDN10 in different cells in normal and tumor tissues. (A) UMAP visualizes the composition in normal ovarian tissue. (B) UMAP visualizes the composition in ovarian tumor tissue. (C) The expression of Claudin proteins in normal cells. (D) The expression of Claudin proteins in ovarian cancer cells. UMAP, Uniform Manifold Approximation and Projection.

Fig. 8.

Gene Set Enrichment Analysis (GSEA) enrichment analysis of CLDN6, CLDN9, and CLDN10. (A) CLDN6. (B) CLDN9. (C) CLDN10.

IHC and PCR results confirmed unanimously: The expression levels of CLDN6, CLDN10 proteins and their mRNAs in ovarian cancer tissues were significantly higher than those in normal tissues (Fig. 9). Western Blot results showed that ovarian cancer tissues had a higher CLDN6 expression in comparison to normal tissues (Fig. 10).

Fig. 9.

The expression of CLDN6, CLDN9, CLDN10 in clinical tissue samples. (A) Gene expression between OC and normal tissues using RT-PCR, ***p 0.001, ns, indicates no significant difference, n = 4. (B) Gene expression between OC and normal tissues using immunohistochemistry. Scale bar = 100 µm. OC, ovarian cancer; RT-PCR, reverse transcription-polymerase chain reaction.

Fig. 10.

The protein expression of CLDN6 in clinical tissue samples. (A) Gene expression between normal and OC tissues using Western-blot. (B) Quantitative expression of CLDN6 in normal ovarian tissue and OC tissue, ** p 0.01, n = 4.

To verify the functional driving role of CLDN6, CLDN9 and CLDN10, a gene knockdown experiment was conducted on the OV-90 and SK-OV-3 cell lines. si-CLDN10-2/3, si-CLDN9-2/3, si-CLDN6-1/3, were selected as the best silencing siRAN and used in subsequent experiments (Fig. 11A). The CCK8 and transwell experiments further demonstrated that reducing CLDN6 expression, rather than CLDN9 and CLDN10, could notably suppress cell proliferation and migration (Fig. 11B–D). Therefore, we decided to focus on CLDN6.

Fig. 11.

The effect of CLDNs knockdown on OC cells proliferation and migration. (A) Determination of knockdown efficiency of CLDN9, CLDN6 and CLDN10 in OV-90 and SK-OV-3 cell lines by qPCR. (B) The effect of knocking down CLDNs on the proliferation of the OC cells. (C) The effect of knocking down CLDNs on the transwell chamber migration of the OC cells. Scale bar = 50 µm. (D) Statistics of transwell chamber migration experiment results. (E) The effect of knocking down CLDN6 on the scratch migration of OV-90 and SK-OV-3 cells. Scale bar = 100 µm. (F) Statistical scratch test results. *p 0.05, **p 0.01, ***p 0.001; n = 3.

Next, a series of in vitro experiments showed that diminishing CLDN6 expression significantly reduced the wound-healing ability (Fig. 11E–F), promoted mitochondrial depolarization and cell apoptosis, induced ROS accumulation and partial G1 arrest in OC cell lines (Fig. 12A–D). The Western blot results showed that interfering with the expression of CLDN6 affected the protein expression of Wnt/$\beta$-catenin signaling pathway (Fig. 13). The results of the animal experiments showed that interfering with the expression of CLDN6 affected the size of the subcutaneous tumors and significantly reduced distant metastasis (Fig. 14).

Fig. 12.

Analysis of the effect of CLDN6 on OC cells by flow cytometry. (A) The mitochondrial membrane potential of OV-90 and SK-OV-03 cells were detected by flow cytometry and Statistics of experiment results. (B) Flow detection of cell apoptosis rate and data analysis. (C) Accumulation of ROS in cells and statistics of ROS relative fluorescence intensity experiment results. (D) Detection of cell cycle arrest was performed by flow cytometry. *p 0.05, **p 0.01, ***p 0.001; n = 3.

Fig. 13.

The protein expression of the Wnt/$\beta$-catenin pathway after the CLDN6 was knocked down. (A) Western blot analysis of Wnt/$\beta$-catenin signalling pathway-related proteins. (B) Statistical scratch test results. **p 0.01, ***p 0.001; n = 3.

Fig. 14.

In vivo experiments. (A) Subcutaneous tumor. (B) Tumor growth curve. ***p 0.001; n = 6. (C) Live imaging observation of lung metastasis. (D) Live imaging fluorescence intensity analysis. ***p 0.001; n = 3.