The GEO database (https://www.ncbi.nlm.nih.gov/geo/) supplied the ovarian cancer single-cell RNA sequencing dataset, obtained under accession number GSE181955. The samples comprised one normal ovary, two primary ovarian cancer specimens, and two omental tissues. Tissue slices for spatial transcriptomics were acquired from GSE211956. This study did not necessitate ethical approval as it utilized publicly available data.

The Seurat package (version 4.3.0) was used to process gene expression data in R (version 4.2.0) (Butler et al., 2018). 500 < nCount <100,000, 300 <nFeature <7,500, mitochondrial gene expression ≤25% of total genes, and erythrocyte gene expression ≤5% of total genes were the quality control criteria used to screen the cells. The data were first normalized using “NormalizeData” (Jiang et al., 2022), followed by “FindVariableFeatures” (Wu et al., 2021; Zhao et al., 2024a; Lin L. et al., 2024; Ni et al., 2025; Hou et al., 2025; Li et al., 2025; Wang et al., 2024) to select the first 2,000 variable genes, then followed by “ScaleData” to standardize the data. Next, we downscaled the obtained data using principal component analysis (PCA) and used the Harmony R package (version 0.1.1) to handle batch effects. Then, we used the “FindClusters” and “FindNeighbors” tools (Ge et al., 2024; Jin et al., 2024; Nie et al., 2024; Zhang Y. et al., 2024) to cluster the first 30 principal components (PCs) into cells. Lastly, based on the important PCs, gene expression was visualized using uniform manifold approximation and projection (UMAP) (Ge et al., 2024; Ding et al., 2023).

To identify marker genes for each cluster, Seurat’s FindAllMarkers tool was used to conduct a differential gene expression analysis across cell clusters (Zhang et al., 2019; Huang et al., 2024; Ding et al., 2023). After that, we used the singleR package to identify and label various cell clusters according to the patterns of marker gene composition. These were then manually confirmed and improved using the CellMarker database.

Odds ratios were computed using the methods outlined in the study to evaluate the preference of tumor cell subtypes for cancer (Zheng et al., 2021).

The authors used AUCell (Aibar et al., 2017), a technique for locating active genes in scRNA-seq data, to evaluate the stemness of TC subtypes. Furthermore, cell stemness was assessed using the CytoTRACE R package (version 0.3.3), which allowed for speculative inference about the chronological order of cell differentiation (Lin et al., 2024b; Sun et al., 2024; Li et al., 2024).

The Monocle 2 algorithm (Qiu et al., 2017; Zhao et al., 2023; Feng et al., 2025), which condensed high-dimensional gene expression data into a lower-dimensional space for display, was used to predict pseudotime trajectories of TCs subtypes. Trajectories were created from the cells, and each branch point was recognized. The Slingshot R package (version 2.6.0) was used to investigate lineage and pseudotime links in more detail (Zhou et al., 2024; Zhao et al., 2024b; Shao et al., 2024). This technique fitted branching trajectories with synchronized master curves and created lineage structures using clustering-based minimal spanning trees.

Using the ClusterProfiler R package (version 4.6.0), we enriched differentially expressed genes (DEGs) using Gene Ontology (GO) (Zhao et al., 2022a; Lin et al., 2022b; Zhao Z. J. et al., 2021), Kyoto Encyclopedia of Genes and Genomes (KEGG) (Wang et al., 2020; Zhao Z. et al., 2021), and Gene Set Enrichment Analysis (GSEA) (Zhang et al., 2023b; Chen Y. et al., 2022a; Sun et al., 2022a). We set significance of all terms at an adjusted p-value whose threshold was less than 0.05 and analyzed the data (Alexa et al., 2006). The filtering criteria for fold change were based on a log2 fold change threshold of 1, which means that only genes with a fold change greater than or equal to 1 (or less than or equal to −1) were considered as differentially expressed.

SCENIC identified active transcriptional regulators from single-cell data. The pySCENIC R software package (version 0.10.0) and Python (version 3.7) for single-cell regulatory network inference and clustering analysis were utilized to identify transcription factors (TFs) with notable expression differences in various TCs subtypes. To start, we used GRNBoost to find putative target genes linked to every TF. In order to find potential sites for direct binding, more DNA-motif analysis was carried out. AUCell has been used to evaluate the activity of the regulators in individual cells, and the top five TFs have been chosen based on the results. The website https://reorigins.aertslab.org/cistarget/ is where the human gene-motif rankings were first created. The Connection Specificity Index (CSI) methodology has been used to define regulatory modules in order to find certain association partners (Suo et al., 2018). Lastly, we classified different regulator modules using hierarchical clustering based on Euclidean distance. Studying the relationships between various regulators is made easier by building the regulator linkage network with a threshold value of 0.65.

CellChat inferred ligand-receptor-mediated intercellular communication. Depending on the ligand-receptor level, regulatory frameworks were constructed and interactions were analyzed using the CellChat R software program (version 1.6.1) (Jin et al., 2021; Lin et al., 2024c; Liu et al., 2024). Using the “netVisualDiffInteraction” and “IdentifyCommunicationPatterns” features, we estimated the number of communication patterns and visualized variations in intercellular communication strength, setting a significant level where the p-value was 0.05.

We utilized the integrated scRNA-seq dataset as a reference to execute cell type decomposition within the histological structures of the ST slide, employing the robust cell type decomposition (RCTD) approach with doublet_mode set to “full” (Cable et al., 2022).

Cell-cell interactions were assessed using stLearn, followed by a modified pseudo-time trajectory analysis in spatial contexts via stLearn (Pham et al., 2023), which utilized PAGA trajectory analysis based on tissue-wide SME normalized gene expression data to reveal relationships among subtypes. A pseudotime spatial trajectory algorithm was utilized to describe the malignant growth across the sections, identifying spatial and transcriptional connections among the subtypes.

We investigated the impact of ovarian cancer-associated TCs on patient survival prediction using key marker genes unique to crucial TCs subtypes. To find the most significant predictive genes, LASSO regression (Zheng R. et al., 2022; Zheng R. Z. et al., 2022; Li X. Y. et al., 2022) was performed after univariate Cox analysis. A risk score model, which is defined as Risk score = ∑ i n X i × Y i , was then established by doing multivariate Cox regression analysis to determine risk coefficients for each gene (Chen B. et al., 2022; Zou et al., 2022; Zhao J. et al., 2022). Using the “surv_cutpoint” tool to find an ideal threshold, patients were divided into low-risk and high-risk categories. The “Survival” R package (version 3.3.1) was used to analyze the survival of these cohorts (Lin et al., 2022c; Lin et al., 2022d; Lin et al., 2023), and the “ggsurvplot” function was used to show the survival curves. ROC curves were produced using the “timeROC” package (version 0.4.0) in order to evaluate the model’s prediction accuracy (Wang J. et al., 2022). AUC was then defined as the area under the ROC curve and responds to the accuracy of this predictive model (Zhao et al., 2022c). In order to purify transcriptomes from bulk data without distinguishing individual cells, CIBERSORT was also used to estimate the number of cell types in bulk RNA-seq data (Newman et al., 2019). Transcripts per million (TPM) values were used to pre-normalize the TCGA bulk RNA-seq data. Using 1,000 permutations without batch correction, a signature matrix was constructed using TPM-normalized datasets for particular cell types. To enable CIBERSORT to estimate the cell type fractions in both cohorts, ovarian cancer patients from the TCGA database were randomly assigned to training and testing cohorts in a 1:1 ratio according to survival status.

The CIBERSORT R program (version 0.1.0) was used to quantify immune cell infiltration and expression differences between various risk score groups. Correlations between immune cells and risk scores, modeled genes, and OS were then examined. Tumor purity, immunological score, EATIMATE score, and stromal score levels across different risk groups were assessed using the Xcell algorithm and the ESTIMATE R package (version 1.0.13).

We used the American Type Culture Collection’s Caov-3 and SK-OV-3 cell lines, which we cultivated at 37°C, 5% CO₂, and 95% humidity. The SK-OV-3 cell line was kept in MD10 medium with the same serum and antibiotic concentrations, whereas the Caov-3 cell line was cultivated in MD02 medium with 10% fetal bovine serum and 1% penicillin-streptomycin.

For every transfection, we utilized Lipofectamine 3000 RNAiMAX (Invitrogen, United States). Six-well plates were seeded with cells at 50% confluence, and the cells were transfected with the knockdown models (Si-FN1-1 and Si-FN1-2) and negative control (si-NC), respectively. GenePharma’s (Suzhou, China) short interfering RNA (siRNA) constructs were used to knock down FN1.

The cells were cultivated for 10 days after being seeded into six-well plates. The cells were then fixed for 15 minutes using 4% paraformaldehyde. The cells were then stained for 15 min using 0.5% crystal violet. Lastly, we used ImageJ software (National Institutes of Health, Bethesda, Maryland, United States) to take pictures of the colonies and determine their number.

Transfected Caov-3 and SK-OV-3 cell lines were seeded at 5 × 10³ per well in 96-well plates, and their cell activity was assessed using the CCK-8 test. The cells were then cultivated for the entire day. We next filled each well with 10 μL of CCK-8 reagent (Vazyme, A311-01) and incubated them for 2 hours at 37°C in the dark. Lastly, on days 1, 2, 3, and 4 after transfection, we measured absorbance at 450 nm using a microplate reader (Thermo, A33978) and plotted the average OD values.

Prior to the experiment, we first starved the cells for the entire day in serum-free media. The cell suspension was then added to the superior chamber of Costar plates that had been pre-treated with Matrigel (BD Biosciences, United States), while serum-enriched media was present in the inferior chamber. Following 2 days of incubation, we treated the cells with 4% paraformaldehyde and evaluated their capacity for invasion using crystal violet staining.

Stable transfected cells were put into 6-well plates, and their growth was monitored until confluence. We made a scratch in each well using a sterile 200-μL pipette tip, cleaned the wells with PBS to get rid of any debris, and then incubated them in a medium free of serum. We took pictures of the scratch both then and 48 h later, and we measured its width using ImageJ software.

In 6-well plates, 5 × 10³ transfected Caov-3 and SK-OV-3 cells were added to each well, and the cells were cultivated for the entire night. In the meanwhile, we combined 10 mM EdU with serum-free medium to create a 2× EdU working solution. Following a 2-h incubation period at 37°C, we rinsed the cells with PBS, fixed them for 30 min with 4% paraformaldehyde, and permeabilized them for 15 min with 2 mg/mL glycine and 0.5% Triton X-100. After that, we dyed them for half an hour at room temperature using 1X Apollo and 1X Hoechst 33,342. Finally, cell proliferation was evaluated using fluorescence microscopy.

Initially, eight tissue samples—normal ovarian tissue, omental tissue, and ovarian cancer tissue—were obtained from GEO to ascertain the cell types implicated in the progression of ovarian cancer. We would conduct a series of multi-omics analyses and experimental validation of the obtained sample data (Figure 1). Genes failing to meet the minimal expression threshold were excluded following an assessment of the quality and completeness of the raw data. Following the removal of batch effects and quality control, 35,726 cells were retained and categorized into 28 clusters (Figure 2A). Through the analysis of cellular genetic profiles and prevalent markers, these 28 cell clusters were conclusively identified as six principal cell types: fibroblasts (DCN), myeloid cells (LYZ), proliferating cells (MKI67), epithelial cells (EPCs, WFDC2), T/NK cells (CCL5), B and plasma cells (IGKC) (Figure 2B). The fraction of EPCs was increasing, but the proportion of T/NK cells was decreasing in ovarian cancer tissues, omental tissues, and normal ovarian tissues. Specifically, myeloid cells exhibited a decline from omental tissue to ovarian cancer tissue and an increase correspondingly from normal ovarian tissue to omental tissue. EPCs predominantly resided in the G1 and S phases, whereas T/NK cells were more prevalent in the G1, G2/M, and S phases, with G2/M being the most frequent (Figures 2C–E). The main enriched genes in the bubble diagram for each cell type corresponded with the known marker genes (Figure 2F). Finally, the quantity of RNAs expressed in EPCs was illustrated by violin plots and UMAP plots, which further indicated that the characteristic RNAs were predominant across all cell types (Figure 2G). Ovarian cancer tissues demonstrated significantly elevated expression levels of both items in comparison to other tissues. We additionally associated EPCs with ovarian cancer, as prior studies indicated that ovarian cancer originates from three distinct tissue types: around 3 to 5 percent from germ cells, 5 to 8 percent from stromal cells, and 85 to 95 percent from epithelial cells.

We then utilized inferCNV to distinguish malignant TCs from EPCs due to the robust correlation between EPCs and ovarian cancer TCs (Supplementary Figure S1). We chose T cells and NK cells as reference cell types for the inferred CNV analysis based on their well-established role as non-malignant, genetically stable immune populations within the tumor microenvironment. These cells are not expected to harbor tumor-specific genomic alterations, making them appropriate baselines for distinguishing CNV signals that are characteristic of malignant cells. To distinguish various TCs subtypes, we down-clustered the 10,953 TCs identified according to the flag genes produced by the cells. The subtypes were subsequently assigned their corresponding genes: C0 TNFRSF18+ TCs, C1 DAPL1+ TCs, C2 SLC40A1+ TCs, C3 FN1+ TCs, C4 MEIS2+ TCs, and C5 CFAP126+ TCs (Figures 3A–C). C0 TNFRSF18+ TCs were found in all three tissues, C1 DAPL1+ TCs and C4 MEIS2+ TCs were more common in both normal and tumor tissues overall, and C2 SLC40A1+ TCs were most common in one of the tumor samples. The analysis found that the omental tissues were dominated by C3 FN1+ TCs (Figures 3D,E). The propensity of ovarian cancer cells to disseminate to the omentum and the incidence of intraperitoneal metastases are well recognized (Arie et al., 2013; Miyamoto et al., 2023). Furthermore, the omentum was among the initial sites where malignancies establish or disseminate into the peritoneal cavity, serving as a vital pre-metastatic environment for the progression of invasive disease (Etzerodt et al., 2020; Bella et al., 2022). A comparable study indicated that FN1, the gene associated with C3 FN1+ TCs, served as a marker correlated with unfavorable outcomes, while another investigation revealed its involvement in regulating the advancement of ovarian cancer (Wei et al., 2021; Meagher et al., 2022). The precise mechanism by which it led to ovarian cancer remained unidentified. The amplification of JUNB was associated with a worse outcome in ovarian cancer patients, while the gene MSLN, prevalent in C3 FN1+ TCs, was previously thought to have a role in the peritoneal dissemination of ovarian TCs (Hilliard, 2018; Perez-Benavente et al., 2022) (Figure 3F). These investigations further corroborated our previous hypotheses. Consequently, ovarian tumor cells were likely to establish the initial cluster of tumors in the omentum, which would then generate malignant cancer cells and ultimately result in ovarian cancer. Furthermore, we might deduce that C5 CFAP126+ TCs predominantly favored the G2/M phase, while C2 SLC40A1+ TCs exhibited a high preference for the S phase. We subsequently calculated the variances in nCount RNA, nFeature RNA, and CNV score expression levels among subtypes and tissue sources (Figure 3G). It would be beneficial for us to examine the pathways of the TCs from omental tissues to ovarian cancer tissues, as the three results exhibited notable discrepancies between ovarian cancer and omental tissues, highlighting significant heterogeneity between these two tissue types.

Subsequently, to assess the malignancy level of the TCs in each subtype, we utilized CytoTRACE to gain an early insight into the differentiation extent of each subtype (Figures 4A–C). C2 SLC40A1+ TCs and C5 CFAP126+ TCs, characterized by a significant prevalence in ovarian cancer tissues, demonstrated a low degree of differentiation, suggesting a higher malignancy in these two subtypes of TCs. This was consistent with the tissue percentage and tissue preference of each subtype found in Figures 3D,E. We focused on C3 FN1+ TCs, which exhibited a relatively lower degree of malignancy, suggesting that they might still be in the growth and development phase rather than fully mature. This corroborated our previous hypothesis that malignant TCs linked to ovarian cancer may develop in the omentum. Bubble plots, which were displayed as violin plots and UMAP plots, showed that the stemness genes enriched to C3 FN1+ TCs were primarily BMI1, CTNNB1, KDM5B, and MYC (Figures 4D,E). Significant differences in expression levels were observed for these four genes between ovarian cancer and omental tissues. Expression levels were markedly elevated in ovarian cancer tissues compared to omental tissues, suggesting a potential role in facilitating the malignant proliferation of TCs within omental tissues. Previous studies have demonstrated that BMI1 promotes TCs growth and metastasis in ovarian cancer by altering TCs angiogenesis and extracellular matrix structure, primarily through the regulation of adhesion plaques and the PI3K/AKT signaling pathway (Zhao et al., 2018). The Wnt/β-catenin pathway, established by CTNNB1 encoding β-catenin, was a crucial signaling pathway implicated in the epithelial-mesenchymal transition (EMT) and has been shown to play a significant role in the carcinogenesis of ovarian cancer. (Arend et al., 2013). Ultimately, we evaluated the stemness of each subtype’s cells using AUC scoring, revealing that C3 FN1+ TCs obtained a high score, indicating their potential to promote the proliferation of malignant TCs (Figure 4F).

We employed the Monocle and SlingShot methodologies for trajectory inference to examine the sequence of developmental trajectories of TCs subtypes. We initially constructed a graph depicting the differentiation trajectory of pseudotime utilizing Monocle (Figure 5A). C0 TNFRSF18+ TCs, C4 MEIS2+ TCs, and C5 CFAP126+ TCs were distributed at all developmental stages, whereas C1 DAPL1+ TCs was primarily disseminated in the latter stages of development, according to a comparison of the subtypes’ differentiation trajectories using Figure 5B. On the other hand, C3 FN1+ TCs essentially did not go beyond node 1, while C2 SLC40A1+ TCs and C3 FN1+ TCs were primarily seen in the early phases of tumor development. Figure 5C illustrated the differentiation trajectories from various tissue origins. The differentiation trajectory of TCs was divided into seven phases, with findings indicating that the highest proportion of C3 FN1+ TCs, the earliest subtype in development, was observed in stage 1 (Figure 5D). The violin plots illustrating the developmental sequence indicated that omental tissues developed prior to ovarian cancer tissues, exhibiting a statistically significant difference, with C3 FN1+ TCs identified as the earliest subtype of TCs (Figures 5E,F).

To further elucidate trajectories among all TC subtypes and tissues, we subsequently employed SlingShot analysis. As the diverse tissues advanced in the sequence of omental-ovarian cancer, with minimal involvement from normal tissues in the developmental process, the distinct subtypes evolved and differentiated in the subsequent order: C3 FN1+ TCs, C2 SLC40A1+ TCs, C1 DAPL1+ TCs, C0 TNFRSF18+ TCs, C4 MEIS2+ TCs, and C5 CFAP126+ TCs (Figures 5G,H). The results of SlingShot’s analysis aligned with Monocle’s findings, indicating that omental tissue and C3 FN1+ TCs subtypes were likely origins of ovarian cancer formation. Examining the role of C3 FN1+ TCs in the malignant proliferation of TCs within omental tissue, leading to carcinogenesis, might provide new insights for targeted therapies in ovarian cancer. The functional processes associated with the lineage trajectory of TCs subtypes were subsequently determined using GO-BP enrichment analyses (Figure 5I). Lastly, the dynamic trend diagrams were used to illustrate the distribution and variations in expression of specific genes across TCs subtypes in pseudotime (Figure 5J). The initial high expression of FN1, the gene that wass named for C3 FN1+ TCs, and the subsequent low level of fluctuation state following an abrupt decline piqued our interest. This illustrated the importance of C3 FN1+ TCs being highly active during the initial stages.

We performed enrichment pathway analyses to examine the role of C3 FN1+ TCs in tumor regeneration and to understand the specific functional activities of these cells during early developmental stages. It was intriguing to observe that the pathway identified by GO-BP enrichment analysis of C3 FN1+ TCs was cytoplasmic translation, a mechanism that was closely associated with TCs proliferation (Figures 6A,D). Furthermore, prior research has shown that protein translation promoted the growth of ovarian cancer tumors, a mechanism that was similar to the one we have identified (Guo et al., 2019). Moreover, the KEGG enrichment analysis was used to identify the critical pathways for C3 FN1+ TCs, including ribosome, allograft rejection, and type I diabetes mellitus. It was intriguing to observe that we collected research that established type I diabetes mellitus as a risk factor for cancer and that individuals with this condition were more likely to develop ovarian cancer (Swerdlow et al., 2023; Suh and Kim, 2019). Utilizing the word cloud, we identified morphogenesis as the most relevant pathway associated with C3 FN1+ TCs, indicating its involvement in the early development of ovarian cancer (Figure 6B). The hypothesis was validated through a review of relevant literature, which revealed that multiple genes associated with tissue and cellular morphogenesis play a role in the development of ovarian cancer (Shi-Peng et al., 2017; Chen et al., 2021). GSEA has identified enriched activities including the development of epithelium, cell differentiation, positive regulation of developmental processes, cell adhesion, morphogenesis of animal organs, and growth factor response (Figure 6C). The cancer cells from the omentum spread and adhered to the ovary, leading to the development of ovarian carcinoma and morphological changes in the ovary. Additionally, the epithelial cells in C3 FN1+ TCs experienced mutations during development to form TCs, which subsequently responded to growth factors, proliferated, and differentiated, ultimately resulting in the formation of malignant proliferating cells. Finally, we identified upregulated and downregulated pathways in C3 FN1+ TCs through GSEA once again (Figure 6E). The pathway was enhanced in response to antigen processing, peptide antigen presentation, and epidermal growth factor. Epidermal growth factor has been demonstrated to enhance ovarian epithelial cell proliferation, facilitate EMT, elevate cancer cell invasiveness and drug resistance, and negatively impact patient differentiation and prognosis (Rodriguez et al., 1991; Grassi et al., 2017; Oh et al., 2014).

We employed pyscenic to examine the gene regulatory architecture of C3 FN1+ TCs to understand the influence of TFs on cellular activities. A preliminary categorization of all TCs based on regulatory activity was conducted initially (Figure 7A). The UMAP plot derived from the activities of the TFs exhibited reduced discretization, as indicated by the clustering results, thereby excluding confounding variables. The C3 FN1+ TCs exhibited limited discretization and were predominantly located on the right side of the figure. Based on the CSI, we divided the TF subtypes into two regulatory factor modules, M1 and M2, in order to see the correlation between them (Figure 7B). Each of these two regulatory factor modules gathered TFs that might potentially work together to regulate genes. In the M2 regulatory module, we could observe that C3 FN1+ TCs had the greatest regulon activity score (Figure 7C). As a result, the primary regulators of the transcription carried out by C3 FN1+ TCs were also the regulatory factors grouped in M2. ATF3, CEBPB, NR2F1, JUND, and YY1 were the five main regulators in C3 FN1+ TCs, as shown in Figures 7D–F. We demonstrated the distribution of these five regulators on UMAP and the expression of each in the various TCs subtypes (Figures 7G–L). Almost all five regulators exhibited low expression in other TCs subtypes while demonstrating high expression in C3 FN1+ TCs, suggesting a specialized expression pattern. Research indicated that cells from recurrent solid tumors demonstrated increased expression of ATF3, implying its role in promoting ovarian cancer aggressiveness, treatment resistance, and recurrence (Bopple et al., 2024). Furthermore, elevated JUNB levels have been shown to enhance tumor growth and metastasis in mice by altering the TGF-β2-stimulated response from an antiproliferative to a pro-invasive one. Additionally, tumor genomic data suggest that JUNB amplification correlates with poor prognosis in ovarian cancer patients (Perez-Benavente et al., 2022). The most notable finding was that NR2F1, absent in nearly all other subtypes, regulated the TGF-β1-driven EMT, influencing immunological response and platinum sensitivity (Liang et al., 2022). In a nutshell, TF analyses suggested that C3 FN1+ TCs played a promoting role in ovarian carcinogenesis and were associated with poor prognosis.

For the purpose of defining and comprehending complex biological interactions, we conducted research on communication webs between ligands and receptors as well as intercellular linkages. Using CellChat analysis, we initially reported the intercellular communication webs that were present in ovarian cancer, which included a large number of cell types and a variety of TCs subtypes (Figure 8A). Concurrently, utilizing the C3 FN1+ TCs subtype of interest as the originating cell, we established a communication network for it (Figure 8B). We observed a notable cellular association between proliferating cells and myeloid cells, as well as C3 FN1+ TCs cells. To understand the communication patterns in ovarian cancer, we analyzed the relationship between the populations of cells that released and received signaling molecules and the fundamental communication dynamics among the cells (Figure 8C). This discovery enabled us to identify three distinct patterns of incoming and outgoing signaling, together with the signaling molecules associated with each pattern. Subsequently, we employed heatmaps to illustrate the expression of numerous signaling molecules in both incoming and outgoing signaling pathways across different cell types and TCs subtype, respectively (Figure 8D). The efferent signaling pathway in C3 FN1+ TCs exhibited substantial expression of MIF, APP, FN1, AGRN, WNT, and NECTIN. Finally, we measured the ligand-receptor network to ascertain the exact outgoing and incoming cellular communication patterns relevant to the six TCs subtypes (Figure 8E). C3 FN1+ TCs predominantly accepted THBS, GRN, and EGF signaling molecules in incoming communication, while they primarily emitted COLLAGEN, FN1, and SEMA3 signaling molecules in outgoing communication. The focus of our investigation was the FN1 dominant signaling pathway. Additional research was required to comprehensively elucidate the impact of FN1 signaling molecules on ovarian cancer development.

A circular diagram illustrated the intercellular communication framework, with FN1 serving as the signaling molecule. A hierarchical diagram was utilized to refine the specific intercellular connections between cells utilizing FN1 as the signaling molecule (Figures 9A,B). The findings indicated that C3 FN1+ TCs, fibroblasts, T/NK cells and myeloid cells exhibited significant signaling cross-talk, with FN1 serving as the secreted ligand. Subsequently, we elucidated the intercellular communication network of the FN1-CD44 signaling pathway and further identified the receptor CD44 (Figures 9C,D). Myeloid cells and T/NK cells were two examples of ligand-receiving target cells associated with both targeted therapy and ovarian cancer. Myeloid cells, encompassing monocytes, granulocytes, dendritic cells, and macrophages, constituted a significant proportion of the TME in cancer and were crucial in regulating tumor spread (van Vlerken-Ysla et al., 2023). The molecular pathways facilitating interaction between macrophages and disseminated cancer cells might serve as innovative targets for the prevention of metastasis and disease recurrence. It was exhilarating to observe that tissue-resident macrophages in the omentum have been empirically demonstrated to facilitate the metastatic dissemination of ovarian cancer in prior studies (Etzerodt et al., 2020; An and Yang, 2021; Jazwinska et al., 2023). Furthermore, immunotherapy aimed at T cells in ovarian cancer suggested that the advancement of ovarian cancer was affected by the interactions between T cells and cancer cells (Blanc-Durand et al., 2023; Nasiri et al., 2023). Using the centrality measure method, which assessed the relative importance of each cell type in this process, the cell types mediating and influencing FN1 signal-mediated intercellular communication were identified. Myeloid cells acted as receivers, mediators, and influencers within the FN1 signaling pathway, whereas C3 FN1+ TCs were significantly expressed as senders, as illustrated in Figure 9E. The heatmap again demonstrated the significant signaling interactions generated by fibroblasts and C3 FN1+ TCs functioning as secretory cells on myeloid cells (Figure 9F). The findings demonstrated that C3 FN1+ TCs and fibroblasts primarily utilized FN1 as a ligand to interact with myeloid cells, with CD44 serving as the receptor. Bubble and violin plots depicted cell-cell interactions among several ligands, with FN1 serving as the ligand (Figures 9G,H). These findings increased our interest in the study of FN1 as well as C3 TCs subtype even more, so we next analyzed them further in the context of the clinic through the ST.

We integrated scRNA-seq and spatial transcriptomics to elucidate the gene regulatory programs and cell-cell interactions that contribute to ovarian cancer development. Tissue sections were collected from an ovarian cancer patient, and ST data were processed (Figure 10A). We executed deconvolution with the RCTD approach to transfer cellular type labels from scRNA-seq data to ST data. Figure 10B presented the two examined outcomes. The C3 FN1+ TCs subtypes and EPCs were situated in the upper right quadrant of the slice, corresponding to the area identified as the tumor location in the tissue (Denisenko et al., 2024). Simultaneously, the locations of the C3 FN1+ TCs subtype and the C5 CFAP126+ TCs subtype corresponded to areas containing tumor tissue. Consequently, the data derived from ST indicated that the C3 FN1+ TCs subtype was indeed correlated with tumor formation in spatial locations. Subsequently, we illustrated the enrichment of spatial distributions for the C3 FN1+ TCs subtype, FN1, and myeloid cells. The findings indicated that the spatial locations of the C3 FN1+ TCs subtype and myeloid cells corresponded with the results derived from the RCTD, and that FN1 was also present in the vicinity of the C3 FN1+ TCs subtypes (Figure 10C). We utilized stLearn for extrapolation to further examine the cellular interactions at specific spatial areas. Our prior research of FN1-CD44 cell contacts in ovarian cancer demonstrated the expression of the receptor, ligand, and ligand-receptor in the corresponding sections (Figure 10D). The ligand FN1 was mostly expressed by the C3 FN1+ TCs subtype and fibroblasts, with the area of elevated expression aligning with the spatial distribution of these two cell types. The expression region of the receptor CD44 was situated between FN1, and there were areas within the tumor where both ligand-receptors were expressed. It was initially established that FN1 and CD44 interact through paracrine and proximal secretion mechanisms. Figure 10E illustrated a substantial quantity of cellular contacted in the area inhabited by fibroblasts, maybe linked to the preformation communication of the tumor. By evaluating the intensity of contacts across all areas in the tissue sections, we concluded that the intensity of cellular interactions was highest and statistically significant in the tumor region (Figure 10F). Our examination of the molecular and cytoarchitectural aspects of ovarian cancer corroborated the finding that the C3 FN1+ TCs subtype facilitated ovarian cancer via the FN1-CD44 signaling pathway, and that they were linked through a paracrine cell-cell communication network inside the ovarian cancer milieu.

The prognosis of patients classified according to strong or weak FN1 expression was initially validated. The survival curves (Guan et al., 2023; Sun et al., 2022b) for the two groups exhibited significant differences, suggesting that FN1 could serve as a valuable prognostic indicator for ovarian cancer (Figure 11A). Subsequently, univariate Cox regression analysis identified 17 mRNAs as potential predictive characteristics derived from the top 100 marker mRNAs of C3 FN1+ TCs (Figure 11B). Of these, it was discovered that HMGN3 and CXCR4 were protective factors (HR <1), whereas the other mRNAs were risk factors (HR >1). LASSO and multivariate regression analysis were employed to address multicollinearity among the mRNAs, ultimately identifying 10 genes associated with prognosis (Figures 11C,D). Analysis of the coefficient values of these prognostic mRNAs revealed that all were classified as risk mRNAs, except for HMGN3 and CXCR4 (Figure 11E). The study involved two groups: high and low FTRS (FN1+ TCs Risk Score) groups, aimed at investigating the influence of FN1 high-expressing TCs on ovarian cancer patients, utilizing the 10 prognostic mRNAs identified from the TCGA cohort (Figures 11F–H). Furthermore, we confirmed that the model effectively predicted the OS C-index (Figure 11I). The model’s predictive accuracy was evidenced by the ROC curve (Figure 11J). The survival outcomes were notably poorer in the FTRS high-expression group, and there was significant variation in survival between groups categorized by FTRS (Figure 11K). We employed the ROC curve to evaluate the rigor of this prediction, and the findings demonstrated enhanced accuracy (Figure 11L). Furthermore, we illustrated the upregulation and downregulation of the initial 30 DEGs via volcano plots (Figure 11M). Subsequently, we examined the pertinent functional pathways with KEGG enrichment methodologies. The signaling pathways regulating stem cell pluripotency and receptor activation in chemical carcinogenesis were highlighted in KEGG (Figure 11N). Both pathways were associated with the initiation of ovarian carcinogenesis, corroborating our previous theories regarding the function of TCs that express FN1.

The extent of immune infiltration in each group was assessed to comprehensively map immune cells in ovarian cancer. Figure 12A presented a thorough depiction of the distribution of 10 prognostically significant mRNAs and immune infiltration levels assessed by the three methodologies: ESTIMATE, CIBERSORT, and Xcell, illustrating the diverse immune infiltration statuses across different risk groups. The relative abundance of stromal and immune components within the tumor samples, enabling us to assess overall immune infiltration and stromal content differences between risk groups. The allocation of 22 immune cells across different risk categories was subsequently illustrated through a box line plot and heatmap utilizing the CIBERSORT tool (Figures 12B,C). Although M0 macrophages, M2 macrophages, and resting CD4 memory T cells constituted the predominant components of the immunological environment in ovarian cancer, the precise distinctions in distribution between the two groups could not be discerned. Subsequently, we illustrated the correlation between risk scores and 22 immune cell types within the tumor microenvironment of ovarian cancer (Figure 12D). The two primary risk factors were resting dendritic cells and active mast cells, whereas the two most effective protective variables were M1 macrophages and B cell memory. Subsequently, we employed a heatmap (Figure 12E) to analyze the correlations among diverse immune cells, eight risk genes, two protective genes (HMGN3 and CXCR4), risk scores, and prognosis. The risk scores corresponded with the findings of Figure 12D; however, we were focused on the contrasting results about the connection of M1 and M2 macrophages with overall survival (OS). Prognosis exhibited a positive correlation with M1 macrophages, whereas M2 macrophages demonstrated an inverse effect. M2 macrophages were associated with the risk factor TIMP3, while M1 macrophages were associated with the protective component CXCR4. Previous studies have also used TIMP3 to construct prognostic characteristics and risk stratification, and patients in the high-risk group had a poorer prognosis (Feng et al., 2022). Subsequently, to more effectively analyze the disparities in immune cell expression between the two risk groups, we conducted additional screening for immune cells that were more representative of the ovarian cancer tumor microenvironment (Figures 12F,G). A selection of six representative immune cells was made. It was noteworthy that immune cells positively correlated with risk scores were significantly expressed in the high group, while the low group demonstrated elevated expression of immune cells negatively correlated. Our primary emphasis was macrophages, revealing that M1 macrophages, favorably correlated with prognosis and negatively correlated with risk score, had substantially high expression in the elevated group. This indicated that M1 macrophages served as a protective immune cell type within the ovarian cancer TME. The literature also documented its cancer-inhibiting properties (Gunassekaran et al., 2021).

We calculated the stromal score, immunological score, and ESTIMATE score for both the low and high FTRS groups utilizing ESTIMATE. All values in the low FTRS group were significantly lower than those in the high FTRS group, indicating a greater degree of immune cell infiltration in the high FTRS group (Figure 12H). The group’s response to immunotherapy was then predicted using the Tumor Immune Dysfunction and Exclusion (TIDE), with elevated scores correlating to diminished efficacy of immune checkpoint inhibition medications (Jiang et al., 2018). The TIDE score was significantly higher in the high-risk group compared with the low-risk group (Figure 12I), suggesting that high FN1 expression significantly impeded the effectiveness of immune checkpoint blockade therapy. Immunophenotypic scores (IPS) of ovarian cancer patients were subsequently used to predict their responsiveness to anti-CTLA4 therapy (Figure 12J). The data showed that the response rate to anti-CTLA4 therapy was higher in the high-risk group. Figure 12K depicted six immune checkpoints highly expressed in the high-risk group, including CD276, CD86, HAVCR2, LAIR1, TNFSF9, and TNFRSF14, which collectively impeded antitumor immune responses. Figure 12L depicted the six drugs with low IC50 values in the high-risk group, suggesting that the high-risk group was more sensitive to these drugs. Overall, the findings suggested that TCs with elevated levels of FN1 might play a key role in immunotherapy resistance and could serve as a compelling biomarker in predicting survival time in patients with ovarian tumors.

The Caov-3 and SK-OV-3 cell lines were utilized in in vitro research to target FN1 mRNA degradation using siRNA to inhibit FN1 expression. Initially, mRNA levels were evaluated prior to and after to FN1 knockdown. Both cell lines exhibited a significant decrease in mRNA expression levels relative to the control group (Figure 13A). Following FN1 knockdown, the CCK-8 assay demonstrated a significant reduction in TCs viability (Figure 13B). Furthermore, the colony formation assay indicated that FN1 knockdown inhibited the aggregation of TCs (Figures 13C,D). Furthermore, transwell and scratch assays were employed to evaluate the migratory and invasive capacities of TCs, revealing a significant reduction in both following FN1 knockdown (Figures 13E–H). This result was also observed by the EdU staining assay (Figure 13I). The aforementioned findings indicated that FN1 knockdown inhibited ovarian cancer tumor growth by obstructing TCs aggregation, activity, invasion, and migration. This result has provided us with a novel approach for targeted therapy in ovarian cancer inside the clinical setting.