Abstract The biological functions of circular RNA (circRNAs) in cancers have garnered significant attention, particularly for their potential as biomarkers. However, the roles of circRNAs in ovarian cancer (OC) and their applicability for early detection of this malignancy remain underexplored. We performed RNA sequencing on ovarian cancer cell lines to identify circRNAs associated with OC. The functional mechanisms of the identified circRNAs were elucidated through bioinformatics analysis. The discriminating ability of biomarkers was assessed using receiver operating characteristic (ROC) analysis. RNA sequencing analysis revealed that 170 known circRNAs were correlated with ovarian cancer. Through the circRNA-miRNA-mRNA regulatory network, we identified 9 circRNAs that interact with 8 miRNAs, subsequently regulating the expression of 324 mRNAs. Functional enrichment analysis, protein-protein interaction (PPI) network analysis, and hub gene analysis indicated that these circRNAs and miRNAs may play a role in regulating MAPK, Wnt, and ErbB signaling pathways. We validated these circRNAs and miRNAs expression profiles in cell, tissue, and plasma samples, identifying four candidates—hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935, and hsa-miR-338-3p—that expression level positively correlate with ovarian cancer development. These markers were then combined into a circRNA and miRNA detection (CMD) panel for ovarian cancer detection. The area under the curve (AUC) values obtained from ROC analysis demonstrated that these individual candidates, as well as the CMD panel, exhibited superior discriminatory ability for OC compared to traditional biomarkers such as CA125, HE4, and the ROMA index in our sample set, which included 28 healthy controls and 22 ovarian cancer patients. Notably, the CMD panel showed exceptional potential for distinguishing early-stage OC samples from healthy controls, achieving an AUC of 1. In this study, we elucidated the functional mechanisms of a set of circRNAs associated with OC through multi-omics analysis and demonstrated that the combination of circRNAs and miRNAs into a biomarker panel holds significant potential for early detection of ovarian cancer. Supplementary Information The online version contains supplementary material available at 10.1038/s41598-025-11641-3. Keywords: Ovarian cancer, Biomarker, CircRNA-miRNA panel, Bioinformatics, Clinical early detection Subject terms: Diagnostic markers, Computational biology and bioinformatics Introduction Ovarian cancer (OC) is a malignant gynecological disease of the female reproductive system. It is the fifth leading cause of cancer death in women in the world^[34]1,[35]2. Due to the lack of typical signs and symptoms, most patients have already been diagnosed with stage III (5%) or stage IV (51%) ovarian cancer at initial presentation and are accompanied by extensive peritoneal metastases. This results in an annual survival rate of only about 5%^[36]3–[37]5. Improving early diagnosis is a research focus. At present, human epididymal protein 4 (HE4) and carbohydrate antigen 125 (CA125) and ROMA index are the most common biomarkers to predict OC in the clinic, but these biomarkers are all at the protein level and have low sensitivity and specificity^[38]6–[39]8. Therefore, it is of great significance to discover novel biomarkers to promote the early screening and detection of OC. Circular RNA (circRNA) is a class of non-coding RNA molecules that do not have 5’ terminal cap and 3’ terminal poly (A) tail and which form circular structures with covalent bonds. This transcript is generated by back splicing or exon skipping of the precursor mRNA^[40]9–[41]11. CircRNA is characterized by being structurally stable, conserved and high-abundance within cells, and displays complex tissue- and stage-specific expression patterns that are refractory to degradation by exonuclease RNaseR and is more stable than linear transcripts^[42]12,[43]13. As a stable endogenous circRNA, circRNA usually regulates the protein expression by binding to downstream miRNA. It can regulate the expression of parental genes, regulate alternative splicing events or mRNA translation, and act as a molecular sponge for miRNA or RNA-binding proteins^[44]14,[45]15. Currently, several studies have shown that circRNA can participate in tumor development in many ways, and it has certain research value in the diagnosis and treatment of tumors^[46]16,[47]17. Because of its high stability, conservation, abundance and specificity, it is considered a promising biomarker for cancer diagnosis and prognosis. Recent studies have suggested that circRNAs play important roles in the proliferation, migration, invasion of OC^[48]18. Wang et al. identified the novel circRNA hsa_circ_0000173 (circATP2B4) in OC tissue, promoted a new mechanism to advance OC progression by affecting the polarization direction of TAMs. It has the potential to serve as a future diagnosis and treatment target of OC^[49]19. Wu et al. found and identified a circRNA (circFBXO7) downregulated in OC, and the downregulation of circFBXO7 released its spongy miR-96-5p, which then induced the degradation of MTSS1. The circFBXO7/miR-96-5p/MTSS1 axis is an important modulator of the Wnt signaling pathway^[50]20. Although these studies have highlighted the specific roles of circRNAs in OC, research on the systematic profiling of circRNA expression remains limited. Multi-omics analysis of circRNAs could comprehensively reveal their functions in ovarian cancer and their potential value as biomarkers. In this study, we utilized high-throughput RNA sequencing (RNA-seq) to identify potential circRNA correlated with OC and evaluated their clinical utility, along with their potential targeted miRNAs, as biomarkers for OC detection. The study workflow is shown in Fig. [51]1. Firstly, at the cellular level, we screened and identified the OC related circRNAs (DECs). By constructing a comprehensive circRNA-miRNA-mRNA regulatory network. Protein-protein interaction (PPI) networks and Kyoto Encyclopedia of Genes and Genomes (KEGG) functional enrichment analysis, a comprehensive understanding of the potential roles of the circRNA in OC have been revealed. Secondly, at the cells, tissue and clinical blood sample levels, the co-expression of circRNA and their potential targeted miRNAs were validated. Finally, combining CircRNAs and MiRNAs Diagnostic panel (CMD panel) was constructed through binary logistic regression, ROC analysis reveal that this diagnostic panel demonstrated have higher discriminate ability in OC than established biomarkers. Our study offers new insights into the involvement of circRNAs in OC development through the ceRNA regulatory mechanism and provides a new diagnostic panel for accurate early OC diagnosis. This may contribute to the clinical development of biomarkers for early diagnosis of OC based on liquid biopsy. Fig. 1. [52]Fig. 1 [53]Open in a new tab Study design flowchart. An overview of experimental design aimed at identifying potential circRNA and miRNA biomarkers for the early diagnosis of ovarian cancer. Materials and methods Cell culture Human ovarian cancer cell lines NIHOVCAR3 and SKOV3, along with the HEK293T cell line, were procured from the Chinese Academy of Sciences Cell Bank in Beijing, China. NIHOVCAR3 cells were cultured in 1640 medium supplemented with insulin and 20% fetal bovine serum (VivaCell, Shanghai, China). SKOV3 cells were maintained in McCoy’s 5 A medium (Gibco, Billings, MT) containing 10% fetal bovine serum, while HEK293T cells were cultured in DMEM (Gibco, Billings, MT, USA). All cultures were incubated at 37 °C in a 5% CO[2] atmosphere. CircRNA isolation and quality control CircRNA was extracted from the cultured cells, including the ovarian cancer cell lines NIHOVCAR3 and SKOV3, as well as the control cell line HEK293T. Total RNA was isolated using the RNeasy Mini Plus Kit (Qiagen, Germany). For each sample, 2 µg of RNA was utilized as the input material for subsequent preparations. Ribosomal RNA was first removed, and the remaining rRNA-free residue was purified via ethanol precipitation. The linear RNA was then digested with 3 U of RNase R (Epicentre, USA) per microgram of RNA, preparing it for sequencing library construction. The quantity and quality of the RNA were assessed using a NanoDrop spectrophotometer (Thermo Fisher Scientific, MA, USA). CircRNA sequencing and preliminary analysis Total RNA was treated with DNase I and RNase R. The sequencing cDNA libraries for RNA-Seq were generated by NEBNext^® Ultra™ Directional RNA Library Prep Kit for Illumina^® (New England Biolabs, USA) following manufacturer’s recommendations. The circRNA sequencing was implemented using Illumina PE150 platform (Illumina, USA), and all cell samples sequencing included 3 replicates were operated by Liaoning Baihao biotech. (Shenyang, China). Reference genome and gene model annotation files were downloaded from the genome website directly. The index of the reference genome was built using Hisat2 (v2.0.5) and paired-end clean reads were aligned to the reference genome using Hisat2 (v2.0.5). CircRNA prediction was performed using CIRI2 (v2.0.6) and find_circ (v1.2) separately and intersected based on their location on the chromosome. We annotated the predicted circRNAs using the circBase database. Each count was normalized for the circRNA expression analysis using junction reads per billion (RPB) mapped reads. Quantification of gene expression level: the raw counts were first normalized using TPM Normalized expression level = (readCount * 1,000,000)/libsize (libsize is the sum of circRNA read counts). Screening of differentially expressed circRNAs associated with ovarian cancer Gene expression levels between NIHOVCAR3 and HEK293T, as well as SKOV3 and HEK293T, were compared using the “edgeR” R package. The criteria for screening differentially expressed circRNAs were set at p < 0.05 and |log2FC| > 1. To visualize the differential expression of circRNAs, we employed the “ggplot2” package to generate volcano plots and the “pheatmap” package to create heat maps. Furthermore, a Venn diagram was employed to filter and illustrate overlapping circRNAs, which were defined as circRNAs associated with ovarian cancer. Annotation and structure prediction of candidate circRNAs The relevant annotation data for the candidate circRNAs were obtained from “CircBase.” The potential open reading frame (ORF) structures, RNA-binding protein (RBP) sites, and microRNA response elements (MREs) for these candidate circRNAs were predicted and visualized using “CSCD” and “CircPrimer.” Construction of circRNA-miRNA-mRNA regulatory network In May 2024, the database tools “ENCORI” and “circInteractome” were accessed to predict circRNA-miRNA interactions. In June 2024, the web-based database tools “miRDB”, “TargetScan”, and “miRWalk11” were utilized to predict miRNA-mRNA interactions. Subsequently, a circRNA-miRNA-mRNA regulatory network was constructed based on upregulated and downregulated circRNAs. The network diagram was visualized using “Cytoscape” software. Gene enrichment analysis The mRNA in the circRNA-miRNA-mRNA regulatory network was enriched by KEGG pathway and annotated by GO. The GO functions and enrichment analysis of genes regulated by circRNA-miRNA network were carried out with clusterProfiler. The classification of ontology covers three domains: Biological Process (BP), Cellular Component (CC) and Molecular Function (MF). The p-value produced by topGO denotes the significance of GO terms enrichment in genes. The lower the p-value, the more significant the GO Term (p-value ≤ 0.01 is recommended). The KEGG pathways enrichment analysis of genes regulated by ceRNA network was performed using clusterProfiler package to map genes in KEGG pathways. The p-value denotes the significance of the pathway correlated to conditions. The lower the p-value, the more significant the pathway (p-value ≤ 0.5). The data were shown in Supplementary Table [54]1. The R package clusterProfiler (v4.6.0) was used for GO and KEGG functional annotation and enrichment analysis. EnrichGO and enrichKEGG functions were used for gene enrichment analysis. The ggplot2 package (v3.4.2) was used for visualization of the results. Screening of hub genes by PPI network We utilized all potential target genes to identify hub genes associated with circRNAs linked to ovarian cancer (OC). Protein interaction networks were analyzed, and potential protein interaction links were predicted using the “STRING” database. By applying “Cytohubba” and the “MCODE” plug-in in Cytoscape software, we filtered for hub genes with strong connections in the PPI networks based on the maximum population centrality (MCC) scoring method. Blood sample collection Normal plasma samples (n = 28) and ovarian cancer plasma samples (n = 22) for this study were obtained from Liaoning Cancer Hospital & Institute (Shenyang, China) between May 23, 2023, and May 23, 2024. All participants were aged 20 to 75 years. The OC patients were diagnosed based on pathological examination and had not received any preoperative anticancer treatment. Clinical data for the patients are presented in Supplementary Table [55]2. Peripheral blood samples (5 mL) were collected from the veins of all participants and placed into 5 mL EDTA anticoagulation tubes, which were gently mixed by inversion for subsequent experiments. After isolation, the plasma was snap-frozen in dry ice and stored at −80 °C until needed. Informed consent was obtained from all participants. The study was approved by the Ethics Committee of Dalian University of Technology, with approval number DUTSBE240612-02. RNase R digestion assay Total RNA (5 µg) extracted from blood sample was incubated at 37℃ for 15 min, either with or without 4U/µg of RNase R (Beyotime, China, Shanghai). Following this incubation, RNase R was inactivated by heating at 70℃ for 15 min. Subsequently, reverse transcription and quantitative real-time PCR (qRT-PCR) were performed to detect the expression levels of the circRNAs and their associated miRNAs. Quantitative PCR analysis Total RNA was isolated from blood samples using TRIzol^® Plus RNA Purification Kit (Invitrogen, Waltham, MA, USA) in a high-speed freezing centrifuge (Cence) following the manufacturer’s instructions. We used the spectrophotometer (Implen, Germany, NanoPhotometer-NP80) to measure RNA concentration and stored samples at −80℃. Briefly, total RNAs were digested with RNase R to remove linear RNAs and enrich circRNAs. Then, the enriched circRNAs were amplified and transcribed into cRNA utilizing a random priming method. For the reverse transcription reaction of circRNA and miRNA, the GoScript™ Reverse Transcription (Promega, USA) was used to synthesize cDNA. The subsequent quantitative PCR reaction was performed following an GoTaq^® qPCR Master Mix (Promega, USA) protocol through an Applied Biosystems QuantStudio 5 system (Applied Biosystems, Massachusetts, USA). GAPDH and U6 were selected as the reference genes for normalizing the expression of circRNAs and miRNAs, respectively. The results were calculated and analyzed with the relative expression level Delta-Ct (ΔCt). All the specific primers in the study were synthesized by Genescript Biotech (Nanjing, China). The primer sequences are listed in Supplementary Table [56]S1. The assay was carried out in triplicates and repeated three times. Statistical analysis The statistical analysis was conducted using IBM SPSS software (SPSS, USA) and GraphPad Prism 9.0 (GraphPad software, USA). For data that followed a normal distribution, an independent samples t-test was employed to compare the measurement data. CircRNAs having |log[2]FC| ≥ 1 and p-value ≤ 0.05 were selected as the significantly DECs. The diagnostic performance of OC-associated circRNA, miRNA biomarkers, and circRNA-miRNA panels was assessed through ROC curve analysis. The AUC was calculated using GraphPad Prism 9.0, with a value exceeding 0.5 considered statistically significant. The SPSS binary logistic regression analysis model was employed to establish a circRNA and miRNA combined diagnostic panel by fitting the expression levels of hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935, and hsa-miR-338-3p. Results Characterizations of circRNA associated with OC CircRNAs are characterized by their unique back-spliced junction, which leads to chimeric reads alignment in the RNA-seq data, effectively distinguishing them from linear RNAs (Fig. [57]2A). To date, there have been few studies focused on circRNAs related to ovarian cancer. To address this gap, we selected two representative human ovarian cancer cell lines, NIHOVCAR3 and SKOV3, along with the HEK293T, which is a non-carcinoma cell line derived from human embryonic kidney cells, and is unrelated to gynecological tissues or organs. Our circRNA sequencing analysis identified a total of 28,956 circRNAs using the find_circ software and 43,146 circRNAs using CIRI2. Following cross-analysis, we found 9,949 circRNAs that were consistently detected (Supplementary Figure [58]S1A). Fig. 2. [59]Fig. 2 [60]Open in a new tab CircRNA generation, size and distribution in ovarian cancer cells. Diagram presenting the forward-splicing of a linear RNA and the back-splicing of a circRNA. (B) The proportion of circRNAs from different sources. (C ) Circos map of the density distribution of all circRNAs on chromosomes. (D) Hierarchical clustering and heat map analysis of differentially expressed circRNAs between HEK293T and NIHOVCAR3, (E) SKOV3. (F) The Venn diagram displayed consistent expression changes across the comparisons of NIHOVCAR3 vs. HEK293T and SKOV3 vs. HEK293T. A small proportion of 9,949 circRNAs originated from introns and intergenic regions, while the majority were derived from protein-coding exons (Fig. [61]2B). We marked these 9,949 circRNA genes on the genomic circle diagram to display the position of the circRNA genes on the chromosome in detail. Their transcripts were mainly distributed on chromosome chr1, followed by chr3, chr2 and chr5 (Fig. [62]2C). To identify the circRNA expression profiles in ovarian cancer (OC), we employed screening criteria (|log2FC| > 1, p < 0.05) to compare the circRNAs from the OC cell lines NIHOVCAR3 and SKOV3 with those from the normal HEK293T cell line. Our analysis revealed that 637 circRNAs were differentially expressed in NIHOVCAR3 compared to HEK293T, with 331 circRNAs up-regulated and 306 down-regulated (Fig. [63]2D, Supplementary Figure [64]S1B). In the comparison between SKOV3 and HEK293T, we found 616 differentially expressed circRNAs, of which 308 were up-regulated and 308 down-regulated (Fig. [65]2E, Supplementary Figure [66]S1C). Notably, 244 circRNAs exhibited consistent expression changes across both comparisons. We consider these 244 circRNAs as potential candidates related to ovarian cancer. Based on the “CircBase” database, 244 candidate circRNAs were annotated to 170 known circRNAs (Fig. [67]2F). We will conduct further research based on these findings. Construction of circRNA-miRNA-mRNA regulatory network To reveal the role of circRNA in the occurrence and development of OC, we constructed a circRNA-miRNA-mRNA interaction network using 170 differentially expressed circRNAs. Initially, we predicted miRNAs associated with circRNAs in two circRNA-miRNA prediction databases. In the ENCORI database, we predicted 70 circRNAs associated with 184 miRNAs, and in the circInteractome database, we predicted 71 circRNAs associated with 293 miRNAs. There were 22 circRNAs and their potential targeted 21 miRNAs overlapped in both databases. Subsequently, we predicted miRNAs-related mRNAs from two miRNA-mRNA prediction databases. In the miRWalk database, we predicted 19 miRNAs associated with 11,834 mRNAs, in the miRDB database, we predicted 19 miRNAs associated with 7,603 mRNAs. Finally, we obtained 8 miRNAs and 324 predicted target mRNAs that existed in two databases. Based on these 8 miRNAs, we identified 9 related circRNAs (Fig. [68]3A, Supplementary Table 3). This led to the establishment of a circRNA-miRNA-mRNA regulatory network that comprises 9 circRNAs, 8 miRNAs, and 324 mRNAs, based on the identified circRNA-miRNA and miRNA-mRNA interactions. Fig. 3. [69]Fig. 3 [70]Open in a new tab CircRNA-miRNA-mRNA network construction. Flowchart for circRNA, miRNA, and mRNA interaction network. (B) The interaction network includes a total of 4 upregulated circRNAs, along with their potential targeted 4 miRNAs and 142 mRNAs. Dark blue nodes represent upregulated circRNAs, lilac nodes represent miRNAs, gray nodes represent mRNAs. (C) The interaction network includes a total of 5 downregulated circRNAs, along with their potential targeted 7 miRNAs and 292 mRNAs. Green nodes represent downregulated circRNAs and gray represent miRNAs. As illustrated in Fig. [71]3B and C, this regulatory network features 4 up-regulated circRNAs (hsa_circ_0000284, hsa_circ_0007440, hsa_circ_0049101, and hsa_circ_0006935) and 5 down-regulated circRNAs (hsa_circ_0004926, hsa_circ_0000638, hsa_circ_0005466, hsa_circ_0007292, and hsa_circ_0006737). The two networks all include 8 miRNAs interacting with circRNAs: hsa-miR-338-3p, hsa-miR-330-5p, hsa-miR-508-3p, hsa-miR-543, hsa-miR-326, hsa-miR-485-3p, hsa-miR-346, and hsa-miR-1197. Notably, hsa-miR-330-5p, hsa-miR-543, and hsa-miR-508-3p are targeted by both up-regulated and down-regulated circRNAs (Table [72]1). Table 1. 9 candidate circrnas information in circRNA-miRNA-mRNA network. circRNA ID Gene NIH vs. T293 Log2FC SK vs. T293 Log2FC miRNA mRNA hsa_circ_0000638 ETFA −2.9884 −1.5751 hsa-miR-1197 EIF5 hsa_circ_0005466 PDPK1 −1.6415 −1.2330 hsa-miR-326 ERBB4 hsa-miR-330-5p GAB1 hsa_circ_0007440 CTNNA1 2.6815 1.7542 hsa-miR-338-3p RIMS1 hsa_circ_0006935 PAPSS1 2.0831 1.5433 hsa-miR-338-3p CBL hsa_circ_0049101 MUC16 2.1266 1.0532 hsa-miR-330-5p MUC16 hsa_circ_0006737 NOP14 −1.6799 −1.4488 hsa-miR-346 FGF7 hsa_circ_0007292 ATP5C1 −1.9431 −1.1322 hsa-miR-485-3p EIF3J hsa-miR-508-3p hsa_circ_0000284 HIPK3 1.2306 1.2346 hsa-miR-508-3p ZNF692 hsa_circ_0004926 TCFL5 −1.2481 −2.3005 hsa-miR-543 FAM10A [73]Open in a new tab To further investigate the molecular mechanism underlying circRNA involvement in transcriptional regulation, we constructed a protein-protein interaction (PPI) network from all mRNAs from the circRNA-miRNA-mRNA network. A PPI network was established to determine the interactions among 324 mRNAs of interest. After excluding disconnected nodes, the final PPI network comprised 169 nodes and 269 edges, visualized using Cytoscape. To identify hub genes within the PPI network, we calculated the connection degree of each mRNA using the cytoHubba plugin in Cytoscape. Additionally, we utilized the MCODE plugin to predict meaningful modules, leading to the identification of two core modules consisting of ten genes (Fig. [74]4A; Table [75]2). In cluster 1, we identified five hub genes: FGF7 (Fibroblast Growth Factor 7), ERBB4 (Erb-B2 Receptor Tyrosine Kinase 4), CBL (Cbl Proto-Oncogene), FGF1 (Fibroblast Growth Factor 1), and GAB1 (GRB2 Associated Binding Protein 1). Cluster 2 contained five hub genes: RIMS1 (Regulating Synaptic Membrane Exocytosis 1), RIMBP2 (RIMS Binding Protein 2), UNC13B (Unc-13 Homolog B), ERC1 (CAST Family Member 1), and UNC13A (Unc-13 Homolog A). Fig. 4. [76]Fig. 4 [77]Open in a new tab PPI network construction. (A) Hub gene screening for a PPI network constructed from 324 mRNAs. The dots represent genes, and the lines indicate the interaction relationship between the genes. The circles represent two clusters of hub genes. The upper diagram corresponds to cluster 1, and the lower diagram corresponds to cluster 2. The color shade represents the degree of correlation. (B) ErbB core pathway and the interrelationships between KEGG pathway and pathway. (C) The mutual relationship between core FGF genes and KEGG pathways. (D) The circRNA-miRNA-mRNA regulatory relationship based on the hub mRNA. The left column shows circRNA, the middle is miRNA and the right column is mRNA. Lines represent the interaction relationships between the genes. Table 2. Candidate hub gene and MCC score. Genes MCODE_score Degree Betweenness centrality Closeness centrality FGF7 4 9 0.05756899 0.30193906 ERBB4 4 9 0.03578392 0.28238342 RIMS1 3.733333333 7 0.01260872 0.24744608 CBL 4 12 0.05027006 0.27148194 FGF1 4 6 0.00747812 0.28056628 GAB1 4 5 0.000946 0.24942792 UNC13A 3.142857143 9 0.014956 0.25231481 NRXN1 3.428571429 19 0.14002235 0.30319889 SYT2 4 6 0.00542817 0.2454955 ERC1 3.428571429 6 0.00156303 0.2411504 UNC13B 3.428571429 7 0.0054633 0.24660633 RIMBP2 4 4 0 0.20566038 [78]Open in a new tab To further explore the specific mechanisms of these key genes, we performed KEGG pathway analysis and identified 26 pathways containing at least one hub gene. The KEGG pathway enrichment analysis revealed significant involvement of the hub genes in pathways such as the ErbB signaling pathway, Ras signaling pathway, MAPK signaling pathway, and PI3K-Akt signaling pathway (p < 0.001) (Fig. [79]4B). Furthermore, we examined the interrelationships among these key pathways and noted that the ERBB signaling pathway is central to many of the identified pathways, particularly in the enrichment analysis of predicted mRNAs (Fig. [80]4C). Subsequently, we reconstructed and visualized a circRNA-miRNA-hub gene regulatory network. In summary, we identified 11 ceRNA regulatory axes comprising six circRNAs (hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935, hsa_circ_0006737, hsa_circ_0005466, and hsa_circ_0000638), 5 miRNAs (hsa-miR-338-3p, hsa-miR-346, hsa-miR-330-5p, hsa-miR-326 and hsa-miR-1197), and seven hub genes (UNC13A, RIMS1, FGF7, FGF1, ERBB4, GAB1, and CBL) (Fig. [81]4D). Validation of the expression of interested circRNAs and miRNA in ovarian cancer cells To confirm the presence of the circRNAs and miRNAs of interest in cells, we extracted total RNA from cultured cells and subjected it to reverse transcription followed by PCR analysis. As shown in Supplementary Figure S2A, B, both circRNAs and miRNAs were detectable in RNA samples that had not undergone RNase R treatment. However, after treatment with RNase R—which digests linear RNAs but spares circular RNAs—we successfully detected circRNAs, whereas miRNAs were undetectable. This outcome confirms the circular nature of the circRNAs and the linear nature of the miRNAs. Subsequently, we performed sequencing analysis on the PCR products, revealing that these circRNAs contained the expected back-splicing junction sites characteristic of circular RNAs (Fig. [82]5A). Additionally, quantitative PCR results demonstrated that three circRNAs exhibited higher expression levels in ovarian cancer cells compared to HEK293T cells, while hsa_circ_0000638, hsa_circ_0005466, and hsa_circ_0006737 showed low expression levels (Fig. [83]5B), which are consistent with their expression characteristics in circRNA sequencing. These findings strongly indicate that the circRNA of interest is indeed present. Fig. 5. [84]Fig. 5 [85]Open in a new tab Identification of the expression of candidate circRNAs and miRNAs in cultured cells. (A) Sanger sequencing of qRT-PCR products for six differentially expressed circRNAs (DECs), including detailed information on the circRNAs and the back-splicing junction sites of their PCR products. (B) comparison of relative expression levels of six circRNAs between control cell HEK293T, and ovarian cancer cell SKOV3, NIHOVCAR3, ***P-value < 0.001, **P-value < 0.01, *P-value < 0.05, ns (not significant). Validation of the expression of interested circRNA and miRNA in ovarian tissues using GEO database To validate the expression of the circRNAs of interest in ovarian tissues, we reanalyzed the circRNA dataset [86]GSE192410 from the Gene Expression Omnibus (GEO). This dataset comprises three paired samples of ovarian tumor tissues and their corresponding normal ovarian tissues. Out of the six interested circRNAs, five were successfully detected, with hsa_circ_0049101 being undetectable, possibly due to its low expression levels (Fig. [87]6A). Specifically, hsa_circ_0007440 and hsa_circ_0006935 were significantly upregulated in OC samples compared to normal samples, while hsa_circ_0000638, hsa_circ_0005466, and hsa_circ_0006737 were significantly downregulated in OC samples (Fig. [88]6A). These expression patterns are consistent with our previously reported cellular-level findings, reinforcing their potential as reliable biomarkers. Fig. 6. [89]Fig. 6 [90]Open in a new tab Identification and validation of the expression of selected circRNAs and miRNAs using GEO datasets. (A) Box plots showing the relative expression levels of five circRNAs in the GEO external dataset [91]GSE192410. (B) Box plots showing the relative expression levels of five miRNAs in the GEO external dataset [92]GSE131790. **p-value < 0.01, *p-value < 0.05, ns (not significant). Additionally, we investigated the expression levels of miRNA—hsa-miR-338-3p, hsa-miR-346, hsa-miR-330-5p, hsa-miR-326, and hsa-miR-1197—in ovarian cancer tissues using the miRNA dataset [93]GSE131790 from GEO. This dataset includes miRNA expression data from 19 ovarian tumor samples and 6 normal epithelial cells samples obtained from resected fallopian tube tissues (Supplementary Figure S3). Among the investigated miRNAs, only hsa-miR-338-3p showed significant and elevated expression in OC samples compared to normal controls (Fig. [94]6B). These findings provide further evidence that the screened circRNA-miRNA pairs are reliable and relevant. Expression levels of circRNA and miRNA in clinical blood samples To validate the presence of ovarian cancer-related circRNAs and miRNAs in plasma, we performed RNase R assays on total circulating RNA samples. As shown in Supplementary Figure S4, six candidate OC-related circRNAs and five miRNAs were successfully detected, with all six circRNAs demonstrating resistance to RNase R treatment (Supplementary Figure S4A, B). These results confirm their presence in the blood samples. To further investigate the expression levels of these interested circRNAs and miRNAs correlated with ovarian cancer, we analyzed plasma samples from 22 OC patients (comprising 5 early stage (stage I and II), 9 stage III, 8 stage IV) and 28 healthy controls using qRT-PCR. Detailed patient demographics and cancer staging information are provided in Supplementary Table S2. As shown in Fig. [95]7A and Supplementary Figure S5, hsa_circ_0049101, hsa_circ_0006935, and hsa_circ_0007440 were significantly upregulated in OC patients compared to healthy controls. Furthermore, the expression levels of these three circRNAs were positively correlated with cancer stage, with their relative expression gradually increasing from early to advanced stages of disease progression. In contrast, hsa_circ_0000638, hsa_circ_0005466, and hsa_circ_0006737 were downregulated in OC patients; however, the differences in these circRNAs compared to the control group did not reach statistical significance. Fig. 7. [96]Fig. 7 [97]Open in a new tab Investigation of expression levels of circRNAs and miRNAs in the plasma at different stages using qRT-PCR experiment. The relative expression levels of the 6 circRNAs in plasma from different stage OC patients and controls. (B) The relative expression levels of 5 miRNAs in plasma from different stage OC patients and controls (Healthy control: 28, stage early: 5, stage III: 9, stage IV: 8). ****p-value < 0.0001, ***p-value < 0.001, **p-value < 0.01, *p-value < 0.05, ns (not significant). In terms of miRNAs, hsa-miR-338-3p, hsa-miR-330-5p, and hsa-miR-346 were significantly upregulated in advanced-stage cancers compared to the control group (Fig. [98]7B), whereas hsa-miR-1197 showed no significant difference in early-stage cancers. ROC analysis yielded consistent results (Supplementary Figure S[99]6). Combined with validation results of miRNAs in tissue samples, only hsa-miR-338-3p exhibited significant differential expression in both tissues and serum samples across clinical stages. Based on the above validation and analysis at the cellular level, tissue level, and in blood samples stratified by clinical stage, three circRNAs and one miRNA were ultimately selected as candidate potential biomarkers for ovarian cancer. Discriminate ability of CMD panel for OC early detection To enhance the clinical utility and feasibility, we developed a CMD panel comprising three highly expressed circRNAs (hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935) and one miRNA (hsa-miR-338-3p) for the detection of ovarian cancer (OC). Utilizing machine learning techniques, specifically binary logistic regression, we established the CMD panel as a diagnostic model (Fig. [100]8). Fig. 8. [101]Fig. 8 [102]Open in a new tab Discriminatory ability of CMD panel for OC early detection. ROC curves for CMD panel, individual circRNA and miRNA markers, and (B) ROC curves for CMD panel, CA125, HE4 and ROMA index for OC detection, discriminating 22 OC samples from the 28 controls. (C) ROC curves for CMD panel, individual circRNA and miRNA markers, and (D) ROC curves from CMD panel, CA125, HE4 and ROMA index for OC early detection, discriminating the 5 OC samples at early stage from randomly selected 5 control samples. The diagnostic performance of both the individual markers and the CMD panel was quantitatively evaluated through ROC curve analysis. Across all OC stage, the single-marker achieved AUC values: 0.9756 (95% confidence interval [CI], 0.7771-1.000) for hsa_circ_0049101, 0.9594 (95% CI, 0.6525–0.9904) for hsa_circ_0007440, 0.9554 (95% CI, 0.6540–0.9562) for hsa_circ_0006935, and 0.9050 (95% CI, 0.7001–0.9999) for hsa-miR-338-3p. The CMD panel demonstrated an AUC of 0.9789 (95% CI, 0.9484-1.0000) (Fig. [103]8A). Notably, compared to the established biomarkers CA125, HE4, and ROMA index, the AUC values for the serum biomarkers CA125 and HE4 were 0.7732 (95% CI, 0.6301–0.9152) and 0.6742 (95% CI, 0.5136–0.8351), respectively. The AUC values for the Premenopausal ROMA (%) and Postmenopausal ROMA (%) were 0.7023 (95% CI, 0.5446–0.8597) and 0.7715 (95% CI, 0.6286–0.9145), respectively. Both individual markers and the CMD panel demonstrated higher AUC values (Fig. [104]8B, Supplementary Figure S[105]7). These results suggest that the CMD panel holds significant potential as an effective biomarker for OC detection. Given the limited number of early-stage samples (n = 5) and control samples (n = 28), we performed ROC analysis on a subset of 5 early-stage OC samples and 5 randomly selected control samples to ensure analytical accuracy. As illustrated in Fig. [106]8C, both individual biomarker and CMD panel have relatively good detection performance for early stage OC. The CMD panel combining hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935, and hsa-miR-338-3p greatly improves the early diagnostic performance. Specifically, the CMD panel also maintained relatively high AUC values, achieving 100% accuracy and sensitivity in certain instances. Furthermore, compared to conventional clinical biomarkers, the CMD panel demonstrated higher sensitivity and specificity (Fig. [107]8D). These results indicate that the developed CMD panel demonstrates significant diagnostic capability, even in the early stages of ovarian cancer. Discussion Ovarian cancer (OC) is associated with the highest mortality among female genital malignancies, with an insidious early onset. The 5-year survival rate for early-stage OC exceeded 85%^[108]21. Therefore, effective early diagnosis is crucial for significantly reducing the mortality associated with this disease. There is increasing evidence that circRNAs and miRNAs play an important regulatory role in OC by directly or indirectly regulating the expression of cancer-related genes^[109]22,[110]23. Analyzing circRNA and miRNA biomarkers across various sample types (cells, tissues, and blood) holds promise for identifying potential diagnostic markers for OC. In our study, we successfully established a diagnostic panel containing 3 circRNAs and 1 miRNA through multi-omics analysis, capable of accurately detecting ovarian cancer. These findings suggest that circRNAs and miRNAs may play significant roles in the molecular pathways involved in OC, potentially serving as biomarkers for diagnosis or targets for therapeutic intervention. An increasing number of circRNAs have been reported in the context of tumorigenesis and shown to be relevant to cancer diagnosis, including circMUC16, and hsa_circ_0007440^[111]24,[112]25. However, a comprehensive understanding of circRNA expression patterns and their specific roles in OC remains limited. In this study, we constructed a circRNA-miRNA-mRNA regulatory network alongside a protein-protein interaction (PPI) network to elucidate the relationships among these molecules in OC. Through this analysis, we identified six circRNAs—hsa_circ_0049101, hsa_circ_0000638, hsa_circ_0005466, hsa_circ_0007440, hsa_circ_0006935, and hsa_circ_0006737—and five miRNAs—hsa-miR-338-3p, hsa-miR-346, hsa-miR-330-5p, hsa-miR-326, and hsa-miR-1197—as potential critical RNAs associated with OC. Among these circRNAs, some of them have been reported to correlate with cancers. hsa_circ_0049101 is derived from the MUC16 gene. Previous studies have reported that circMUC16 is among the up-regulated circRNAs in OC tissue and exhibits a close association with cancer staging and grading^[113]22. Gan et al. demonstrated that circMUC16 (hsa_circ_0049116) is up-regulated in OC cells and promotes invasion and metastasis of epithelial OC cells^[114]26. Additionally, circNOP14 (hsa_circ_0006737) has been demonstrated to be downregulated in various cancers, including colorectal cancer and liver cancer^[115]27,[116]28. Among them, hsa_circ_0007440 has been reported to play an important role in the malignant progression of osteosarcoma by binding to hsa-miR-338-3p. Our results indicate these circRNA may also correlated with OC^[117]29. MiRNAs have been demonstrated to play a crucial role as regulatory RNAs in OC. Specifically, hsa-miR-330-5p and hsa-miR-338-3p have been implicated in the pathogenesis of OC^[118]30. Overexpression of hsa-miR-330-5p was found to be correlated with increased cancer aggressiveness and stage in OC tissues and cells^[119]31. Hsa-miR-338-3p is down-regulated in OC tissues and cells, associated with reduced recurrence-free survival (RFS) and overall survival (OS), making it a potential prognostic biomarker for epithelial OC^[120]32. Additionally, studies have shown that miRNA molecules such as hsa-miR-326 can inhibit proliferation and invasion while promoting apoptosis in OC cells^[121]33. In this study, we demonstrate these OC-correlated miRNAs may interact with circRNAs. We employed the KEGG databases to gain insights into the potential function of circRNA-miRNA-mRNA interactions. KEGG pathway analysis revealed that most pathways were mainly associated with cancer. The most significant signalling pathways included the MAPK (hsa04010), Wnt (hsa04310), and ErbB (hsa04012) signalling pathways. Among them, hsa04012 is an important signaling pathway, which is closely related to cell proliferation and differentiation. The binding of ligands to ErbB receptors triggers downstream RAS/MAPK signaling pathways, thereby regulating the biological effects of cells^[122]34. Five genes CBL, ERBB4, GAB1, MAP2K1 and MAPK1 were enriched by the ErbB pathway. hsa04010 is also known as the RAS-RAF-MEK-ERK signaling cascade, and each component of this pathway is closely related to the occurrence and development of cancer^[123]35. 2 competitive circRNAs govern 14 genes to function in OC via MAPK signaling pathway. hsa_circ_0049101 (FGF1) and hsa_circ_0006737 (FGF7) have closely association to this pathway through influencing gene expression. Many studies have shown that dysregulation of the Wnt pathway can promote tumor cell proliferation and deterioration^[124]36. The findings tip that circRNAs and miRNAs may play significant role in the regulation of gene expression. However, the regulatory role of the screened circRNA and its ceRNA network in OC needs to be further validated by molecular and cellular experiments. In addition to analyzing cell samples, we further confirmed the expression of six circRNA and five miRNA in tissues and clinical blood samples at different FIGO stages. Among the six circRNAs examined, hsa_circ_0049101, hsa_circ_0007440, and hsa_circ_0006935 were significantly upregulated in the patient group compared to the control group, with their expression levels showing a significant positive correlation with advancing OC stages. Of the five miRNAs assessed, only hsa-miR-338-3p was upregulated, which aligns with the expression patterns observed in OC tissue within the GEO database. Consequently, hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935, and hsa-miR-338-3p were selected as potential ovarian cancer markers, as they exhibited higher expression levels in ovarian cancer cells, OC tumor tissues, and plasma samples from OC patients compared to controls. At present, the commonly used OC diagnostic biomarkers are CA125, HE4 and ROMA index. Molina et al. via binary logistic regression and ROC curve analysis, demonstrated that the sensitivity, specificity, and AUC values of HE4 in diagnosing OC were 79.3%, 98.9%, and 0.936 respectively, while corresponding values for CA125 were 82.9%, 70.9%, and 0.853 respectively. The corresponding values for ROMA were found to be 90.1%, 87.7%, and 0.952^[125]37. However, all of the above are protein-level biomarkers, and both sensitivity and specificity are insufficient, so there is a need to discover new RNA biomarkers. CircRNA has unique expression patterns, molecular stability, specificity, and widespread distribution in the human body, thus enabling its detection and quantification through liquid biopsy in body fluids^[126]38. As one of the most promising biomarkers, circRNA has been extensively studied. In a recent study, Wang et al. discovered that serum circSETDB1 promotes tumor development and is up-regulated in serous ovarian cancer (SOC), indicating its potential as a biomarker. ROC analysis demonstrated that serum circSETDB1 expression effectively differentiated SOC patients from healthy controls with an AUC of 0.8031, sensitivity of 78.33%, and specificity of 73.33%^[127]39. Ge et al. utilized microarray technology to detect the expression profile of circRNA in the plasma of OC patients. They established a combination (hsa_circ_0003972 and hsa_circ_0007288) known as circCOMBO through binary logistic regression analysis. The AUC for hsa_circ_0003972 and hsa_circ_0007288 were determined as 0.724 and 0.790, respectively. circCOMBO exhibited superior diagnostic value^[128]40. In this study, we combined hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935, and hsa-miR-338-3p as CMD panel. ROC analysis demonstrated that CMD panel had higher discriminate ability than the established biomarkers. These findings suggest that the cirRNA and miRNA hold great potential in OC detection, even in early stage of OC. There are certain limitations to our study, the analysis of regulatory networks or mechanisms in this study only identified DECs and miRNAs based on bioinformatics predictions, thus lacking in vivo experiments for result verification. Simultaneously, considering that the current method has a relatively small sample size for early-stage ovarian cancer. This limited cohort size may reduce the sensitivity in detecting differences, increasing the risk of false negatives, particularly for ovarian cancer subtypes with high heterogeneity. In future studies, we will validate this method by expanding the cohort size to systematically enhance the clinical applicability of the CMD panel, with the ultimate goal of integrating it into early ovarian cancer screening guidelines. Conclusions In conclusion, this study identified a set of DECs and their potential targeted miRNAs as detection biomarkers through multi-omics analysis, assessed at three levels: in ovarian cancer cells, tissues, and clinical blood samples. Additionally, it revealed important genes, pathways, and competing endogenous RNA (ceRNA) regulatory networks involved in ovarian cancer. Our findings suggest that hsa_circ_0049101, hsa_circ_0007440, hsa_circ_0006935, and hsa-miR-338-3p could serve as novel biomarkers for detection of ovarian cancer. The constructed CMD panel, which includes these three circRNAs and one miRNA, demonstrated superior diagnostic performance for early-stage ovarian cancer compared to single biomarkers and traditional clinical markers. We believe that this work can provide insights into the role of circRNAs in ovarian cancer, while also offering new strategies for the discovering of biomarkers for the early detection of ovarian cancer. Electronic supplementary material Below is the link to the electronic supplementary material. [129]Supplementary Material 1^ (3.8MB, docx) Acknowledgements