Epithelial characteristics of ovarian clear cell carcinoma at single-cell resolution

Abstract

Ovarian clear cell carcinoma (OCCC) represents a rare and aggressive subtype of epithelial ovarian cancer with distinctive clinical and molecular characteristics. However, the identification, origin, and molecular features of the malignant epithelial cells in OCCC remain poorly studied. We establish an OCCC-associated transcriptional landscape using single-cell RNA sequencing and investigated the properties of epithelial cells in tissues from normal ovaries, ovarian endometriosis, primary OCCC and recurrent OCCC to assess the status of malignant epithelial cells. We identify a specific subcluster of malignant epithelial cells and further analyze them to discover 173 candidate factors associated with OCCC. Regulon and pseudotime trajectory analyses reveal six transcription factors (TFs) and their corresponding targets among these candidate factors, highlighting their roles in OCCC onset and reoccurrence. Through experimental validation, we confirm the crucial involvement of STAT3, KLF5, and TRIM28 in the proliferation and migration of OVISE cells. Silencing these three TFs also results in the down-regulation of their associated TF targets linked to OCCC. Overall, we characterize complex malignant-like cell populations at single-cell resolution and highlighted several TFs and their targets, providing essential resources for understanding the regulatory mechanisms underlying OCCC initiation and recurrence.

This study conducted a comprehensive scRNA-seq of ovarian clear cell carcinoma (OCCC), identifying a malignant epithelial subcluster and 173 associated factors. Analysis highlighted crucial roles of STAT3, KLF5, and TRIM28, as a valuable resource for understanding OCCC.

Introduction

Ovarian clear cell carcinoma (OCCC) is a rare, aggressive, and chemoresistant tumor that constitutes approximately 13% of cases of epithelial ovarian cancer (EOC)^1,2. It is considered an understudied subtype of ovarian cancer with an extremely poor prognosis. The median overall survival (OS) for recurrent OCCC (rOCCC) patients is a mere 25.3 months, and the 5-year post-recurrence survival rate for rOCCC is a mere 13.2%, with over two-thirds of patients succumbing within 12 months and 93.1% 24 months^3,4.

OCCC exhibits distinct clinical and molecular features compared to other gynecological malignancies². Although “clear cell” is a pathology term denoting malignant epithelial cells in OCCC, and certain factors, such as KRT7 and PAX8, have been identified as playing crucial roles in OCCC diagnosis^4,5, the unique attributes of these cell populations in comparison to other epithelial cells, as well as their origin, have not been fully elucidated. Recently, an association between OCCC and endometriosis has been proposed, suggesting that endometrial-type epithelial cells could serve as precursors⁶. Notably, endometriosis is associated with an increased risk of OCCC^7,8. The constituents of ovarian endometriotic (endometrioma) cysts may influence the endometriotic environment and contribute to the transformation of endometriotic lesions into OCCC⁹. Recent single-cell transcriptomic analysis revealed a correlation between OCCC and endometriosis, suggesting that endometrial-type epithelial cells may serve as precursors for these tumors¹⁰.

The difficulties in procuring primary OCCC (pOCCC) and rOCCC neoplasms present a considerable practical impediment to obtaining ample tissue samples, thereby leaving the in vivo origins of malignant-like cell populations and their molecular attributes largely ambiguous. To bridge these significant knowledge gaps related to the onset and reoccurrence of OCCC, here, we initially established an OCCC-associated lineage via single-cell RNA sequencing (scRNA-seq) and subsequently examined the status of malignant epithelial cells to explore the transcriptional landscape of the epithelium. Furthermore, we identified distinct malignant-like cell populations and evaluated them to identify additional biomarkers associated with OCCC. Through this comprehensive approach, we successfully characterized the complex malignant-like cell population at single-cell resolution, thus elucidating several pivotal factors that underlie the initiation and recurrence of OCCC.

Results

Generation of single-cell transcriptomic landscapes for OCCC

In this study, we conducted a comprehensive analysis of the pathological and molecular features of the following human ovarian tissues: normal ovary, endometrioma, pOCCC, and rOCCC. First, hematoxylin and eosin (H&E) staining was performed to clarify the general characteristics of the healthy ovaries, endometriomas, pOCCC and rOCCC (Fig. 1A). Meanwhile, no tumor condition was detected in the normal ovary and endometrioma groups.

Fig. 1 Construction of the OCCC transcriptional landscape. [Images not available. See PDF.]

A H&E staining of human ovarian tissues from the normal ovary, endometrioma, pOCCC and rOCCC. B An illustrated flow chart of the scRNA-seq procedure. C UMAP visualization of human ovarian cell populations. D UMAP visualizations of representative marker genes. E Dot plot visualization of marker genes. The diameter of the dot corresponds to the fraction of cells expressing each gene in each cluster. The color intensity represents the average normalized expression level. F Cellular components of different sample types for each subcluster of human ovarian cell populations. Scale bar: 50 μm.

To comprehensively establish the transcriptomic landscapes associated with OCCC, we curated scRNA-seq data from multiple sources. This compilation included our recently generated datasets comprising one pOCCC sample and four rOCCC samples, as well as four additional published datasets consisting of one normal ovary sample and three endometrioma samples (Fig. 1B). After applying the filtering criteria, a total of 83,209 high-quality cells were retained for subsequent bioinformatics analysis, with an average of approximately 3524 unique molecular identifiers (UMIs) per cell and approximately 1200 genes detected per cell.

Cellular heterogeneity was investigated by uniform manifold approximation and projection (UMAP), and cells were grouped according to the similarity of their unique gene expression (Fig. 1C and Figure S1A, B). Unsupervised classification of individual cell transcriptomes identified eight distinct cellular clusters, namely, epithelial cells, B cells, plasma cells, T/NK cells, myeloid cells, fibroblasts, endothelial cells, and pericytes, based on the expression patterns of marker genes (Fig. 1C). To investigate potential markers for OCCC, we visualized the top 50 genes with the highest expression levels in each cell cluster using a heatmap (Figure S1C). Additionally, dot plots and UMAP visualizations were generated to visualize representative marker genes (Fig. 1D, E).

Among these prominent cell populations (Fig. 1F and Figure S1D), the proportion of the fibroblast population was significantly lower in pOCCC and rOCCC than in normal ovaries. Conversely, the proportion of the myeloid cell population gradually increased with increasing malignancy. Additionally, the T/NK cell population exhibited high enrichment in endometriomas, while the epithelial cell population demonstrated a marked increase in the pOCCC and rOCCC samples.

Identification of malignant epithelial cells via copy number variations (CNVs)

To elucidate the potential origin of OCCC from endometrioma in the ovaries, we initially investigated inferred CNVs within the epithelial cell populations of the normal ovary, endometrioma, pOCCC, and rOCCC samples across chromosomal intervals, to assess the resemblance of endometrioma to malignant epithelial cells. The epithelial cells in the normal ovary exhibited minimal CNV signals and served as a reference for calibration. Notably, the epithelial cells within endometriomas demonstrated a heightened median CNV score, showing a significant elevation in OCCC epithelial tissues (Fig. 2A, B). Subsequently, we reclustered the epithelial cell populations using the K-means method, which led to the identification of ten distinct subclusters. Notably, subcluster 10 (malignant-like cells) emerged as the subpopulation of malignant epithelial cells with the highest CNV scores (Fig. 2C). By scrutinizing the cellular constituents across various samples, subcluster 2, 3, and 9 exhibited the highest abundance in OCCC (Fig. 2D, E). Furthermore, a progressive rise in the ratio of malignant tumor cells (specifically cells in subcluster 1, 4, 5, 6, 7, 8, and 10) was noted from endometrioma to pOCCC and rOCCC (Fig. 2D, E). These findings suggest that epithelial cells in endometriomas may contribute to malignancy, with rOCCC demonstrating a greater degree of malignancy than OCCC.

Fig. 2 Analysis of CNVs in epithelial cells. [Images not available. See PDF.]

A The relative expression levels of InferCNV signals on each chromosome in different human ovarian specimens. Normal ovary-derived epithelial cells were used as the control. B CNV scores of the normal ovary, endometrioma, pOCCC and rOCCC samples. The statistical differences were analyzed using the Wilcoxon rank-sum test. C K-means clustering for similar patterns of CNV signals in epithelial cells. D Cellular component analysis of K-means clusters in different sample types. E Cellular component analysis of the normal ovary, endometrioma, pOCCC and rOCCC samples in different K-means clusters.

Identification of malignant-like cell populations in OCCC

The primary objective of this study was to ascertain the distinct cell populations of malignant-like cells in OCCC. Therefore, we further annotated the subclusters of epithelial cells as (Epithelial Population 1) EP1 – (Epithelial Population 5) EP5 and performed t-distributed stochastic neighbor embedding (tSNE) to visualize the subclusters and corresponding groups (Fig. 3A, B). It is widely acknowledged that malignant-like cells are predominantly found in pOCCC and rOCCC. Surprisingly, we observed a significant increase in both the number and ratio of the EP1 subcluster in the pOCCC and rOCCC samples compared to that in the normal ovary or endometrioma sample (Fig. 3C, D). t-SNE also revealed that a small proportion of EP1 epithelial cells were present in normal ovaries and endometriomas, but the proportions of these cells increased dramatically in the pOCCC and rOCCC samples (Figure S2A).

Fig. 3 Subcluster analysis of epithelial cells in different human ovarian specimens. [Images not available. See PDF.]

A t-SNE visualization of subclusters (EP1-EP5) of epithelial cells. B t-SNE visualization of epithelial cells in different tissue types. C Cellular fraction of subclusters in different tissue types. D The number of cells in different subclusters in each sample type. E Dot plot visualization of PAX8 and KRT7 in each subcluster. The diameter of the dot corresponds to the fraction of cells expressing each gene in each subcluster. The color intensity represents the average normalized expression level. F tSNE visualization of PAX8/KRT7 double-negative, PAX8-positive, KRT7-positive, and PAX8/KRT7 double-positive epithelial cells. G IHC staining of PAX8 in normal ovary, endometrioma, OCCC, and rOCCC ovarian tissues. H IHC staining of KRT7 in normal ovary, endometrioma, OCCC, and rOCCC ovarian tissues. I The 5 genes with the most abundant expression in the EP1–EP5 subclusters of epithelial cells. Scale bar: 50 μm.

Due to the mixture of normal epithelial cells and malignant like cells in epithelial cell populations, we subsequently investigated the expression patterns of two well-known markers, PAX8 and KRT7, in each subcluster of epithelial cells in OCCC. Surprisingly, we observed high enrichment of both PAX8 and KRT7 in the EP1 subcluster (Fig. 3E). The tSNE plot clearly showed that PAX8- and KRT7-positive cells constituted the majority of the EP1 population, particularly in the pOCCC and rOCCC samples (Fig. 3F and Figure S2B). Therefore, we inferred that the EP1 subcluster represents the malignant-like cell population in human ovaries.

Furthermore, we demonstrated that PAX8/KRT7 double-positive epithelial cells accounted for a significant proportion of the malignant-like cell population in pOCCC, with the vast majority being present in rOCCC (Figure S2C–F), revealing the heterogeneity of malignant epithelial cells in pOCCC and rOCCC. To validate the expression patterns of PAX8 and KRT7, we performed immunohistochemistry (IHC) staining on normal ovary, endometrioma, pOCCC, and rOCCC ovarian tissues, which yielded results consistent with those of the scRNA-seq analysis (Fig. 3G, H). Additionally, Fig. 3I displays the five most abundantly expressed genes in the EP1–EP5 subclusters of epithelial cells, which may serve as potential targets for malignant-like cell populations. Our data represent a pioneering application of scRNA-seq technology to elucidate the molecular characteristics of malignant-like cell populations in human ovaries.

Identification of genes associated with pathogenicity in OCCC

Given the limited knowledge surrounding pOCCC and rOCCC, our subsequent investigation aimed to identify potential targets involved in the regulation of these conditions by analyzing transcriptomic profiles within malignant-like cell populations. To uncover pathogenic genes associated with OCCC, we conducted an expression trend analysis within malignant-like cell populations (EP1) of the normal ovary, endometrioma, pOCCC, and rOCCC samples. Intriguingly, we identified a set of genes exhibiting distinct expression trends across the four sample types (Fig. 4A). Specifically, our emphasis was on profile 12, identified with the lowest P-value, and identified 933 genes that tended to exhibit low expression in normal ovary/endometrioma but high expression in pOCCC/rOCCC (Fig. 4B, C).

Fig. 4 Trend analysis of malignant-like cell populations in human ovarian specimens. [Images not available. See PDF.]

A Expression trend analysis within malignant-like cell populations (EP1) of normal ovary, endometrioma, pOCCC, and rOCCC. The x-axis represents the order of the four groups. B The number of genes in each expression trend profile. C Expression trend curves of profile 12 genes in different samples. D DO analysis of profile 12 genes. E GO analysis of profile 12 genes. F Volcano plots of DEGs for the comparisons of normal ovary vs. pOCCC and normal ovary vs. rOCCC. G Venn diagram for the number of common and unique DEGs among normal ovary, pOCCC, and rOCCC. H Venn diagram for the number of profile 12 genes and common DEGs identified in (G).

The Disease Ontology (DO) serves as a standardized ontology for human diseases, aiming to furnish the biomedical community with consistent, reusable, and enduring descriptions of human disease terms, phenotype characteristics, and associated medical vocabulary concepts. DO enrichment analysis revealed notably enriched human disease DO terms in DEGs compared to the entire genome background. Notably, DO analysis revealed that genes within profile 12 were significantly enriched in ovarian carcinoma and other clear cell-related carcinomas, suggesting their potential as reliable candidate targets for clear cell-originating ovarian carcinoma (Fig. 4D). Additionally, Gene Ontology (GO) enrichment analysis indicated that genes within profile 12 primarily participated in gene expression processes, such as transcription and protein synthesis-related events (Fig. 4E). Furthermore, Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis demonstrated the involvement of 12 genes in various metabolic pathways, tight junctions, focal adhesions, and other pathways (Figure S3A).

To further refine our list of candidate targets, we performed a differential expression analysis (Fig. 4F) and identified 453 genes that were differentially expressed in both pOCCC and rOCCC compared with normal ovarian cells (Fig. 4G and Figure S3B). Notably, 173 differentially expressed genes (DEGs) exhibited expression trends consistent with those in profile 12 (Fig. 4H and Figure S3C). The DO analysis revealed that these 173 genes were enriched in multiple cancers, particularly malignant tumors of ovarian origin (Figure S3D). Collectively, our findings suggest that these 173 identified genes may play a crucial role in the pathogenicity of OCCC.

Transcription factor (TF) analysis in malignant-like cell populations

Subsequently, we constructed a TF regulatory network module, referred to as a regulon, using SCENIC software to predict TFs and their corresponding target genes. Within this context, we identified a total of 34 regulons, 10 of which were associated with 6 TFs, thereby reflecting transcriptional regulatory activity specific to malignant-like cell populations in the ovaries. Notably, these 10 relevant regulons exhibited heightened activation within both the pOCCC and the rOCCC malignant-like cell populations (Fig. 5A). To gain further insight into the activity distribution within the malignant-like cell populations, we employed tSNE as a visualization technique (Fig. 5B-C and Figure S4A). Crucially, we also explored the expression levels of the aforementioned 6 TFs and their targets in the malignant-like cell populations, revealing significant up-regulation in both pOCCC and rOCCC (Fig. 5D–F and Figure S4B). Furthermore, through the examination of interaction networks and Sankey diagrams, it became evident that several targets were subject to regulation by multiple TFs, thus revealing a complex web of interactions among these factors (Fig. 5E, G).

Fig. 5 Regulon analysis of OCCC-associated TFs and targets in malignant-like cell populations. [Images not available. See PDF.]

A Heatmap of 10 regulons for OCCC-associated candidate TFs. B t-SNE visualization of the activity distribution of the different groups. C t-SNE visualization of the activity distributions of different regulons. D Dot plot view of 6 OCCC-associated TFs. E Sankey diagram for OCCC-associated TFs and their targets identified in malignant-like cell populations. F Dot plot view of OCCC-associated TF targets. G Interaction network between TFs and their targets identified in malignant-like cell populations.

Pseudotime trajectory analysis of epithelial cell populations

To further elucidate the mechanism underlying pOCCC and rOCCC pathogenesis, we conducted pseudotime trajectory analysis of epithelial cell populations derived from normal ovary, endometrioma, pOCCC, and rOCCC (Fig. 6A). This analysis revealed the presence of seven distinct states that effectively distinguished epithelial cells within each sample (Fig. 6B, C). Notably, in normal ovaries, epithelial cells predominantly exhibited a propensity towards States 1-2, while in endometriomas, they leaned towards States 1-3. In contrast, epithelial cells from pOCCC were predominantly aligned with States 4-7, whereas those from rOCCC were inclined towards States 4-6 (Figure S5). Additionally, when we examined the pseudotime trajectory based on EP subclusters, we observed that malignant-like cell populations (EP1) tended to develop toward OCCC and rOCCC states (Fig. 6D–F).

Fig. 6 Pseudotime trajectory analysis of epithelial cell populations in human ovarian specimens. [Images not available. See PDF.]

A Pseudotime trajectory plot view of epithelial cell populations in normal ovary, endometrioma, pOCCC, and rOCCC. B Pseudotime trajectory analysis of epithelial cell populations in different states. C Ridge plot view of the pseudotime trajectory of epithelial cell populations in different sample types. D Pseudotime trajectory analysis of cells in different subclusters of epithelial cells. E Pseudotime trajectory analysis colored according to malignant-like cell population (EP1). F Pseudotime trajectory analysis of cells in different groups. G Heatmap for Branch 1 (States 1, 2 vs States 1, 3, 7, 5, 6) manner in epithelial cell populations. H Spline plots of representative gene expression dynamics for Branch 1 (States 1, 2 vs States 1, 3, 7, 5, 6) manner in epithelial cell populations. I Heatmap for Branch 2 (States 1, 3, 5, 6 vs States 1, 3, 7) manner in epithelial cell populations. J Spline plots of representative gene expression dynamics for Branch 2 (States 1, 3, 5, 6 vs States 1, 3, 7) manner in epithelial cell populations.

To distinguish state-biased genes to trace the source of malignant epithelial cells, we employed heatmap and spline plot views to illustrate the expression dynamics within branch 1 (States 1, 2 vs States 1, 3, 7, 5, 6) manner (Fig. 6G, H). Surprisingly, we discovered a set of genes crucial for the development of OCCC. Notably, the expression of TFs and their targets, such as KLF5, TMBIM1, SLC2A1, LAMC1, and PTHLH, was upregulated in both pOCCC and rOCCC (Fig. 6H). Furthermore, to further determine the source of the primary and recurrent malignant epithelial cells, we identified a significant number of state-biased genes expressed within Branch 2 (States 1, 3, 5, 6 vs States 1, 3, 7) manner (Fig. 6I, J). Notably, ANXA2, TMBIM1, SLC2A1, LAMC1, and PTHLH exhibited expression patterns biased toward rOCCC, while genes such as SMARCC1, VCL, CS, CTNNA1, MAFK, STAT3, PAX8, and ACTN4 displayed expression patterns biased toward OCCC (Fig. 6J). Collectively, these and the aforementioned findings suggest that these state-biased genes are likely to play a pivotal role in the pathogenicity of OCCC.

Identification of three TFs linked to OCCC pathogenicity

Our scRNA-seq analysis unveiled six potential TFs within malignant-like cell populations, hinting at their plausible involvement in OCCC development. Nevertheless, the functional implications of these TFs remained ambiguous. To tackle this, we executed functional confirmations in the context of two OCCC-associated cell lines (OVISE and TOV-21G) to evaluate impacts of TFs on OCCC and the consistent regulation of downstream targets by these TFs. Noteworthy discoveries include the pivotal functions of STAT3, KLF5, and TRIM28 in OCCC. Specifically, we silenced the STAT3, KLF5, and TRIM28 genes in OCCC cell lines and confirmed the efficacy of RNA interference (RNAi) through qRT‒PCR (Fig. 7A and Figure S6A). Cell growth was assessed using the CCK-8 assay, which revealed that the knockdown of the STAT3, KLF5, and TRIM28 genes potently inhibited cell proliferation (Fig. 7B and Figure S6B). Additionally, upon silencing the STAT3, KLF5, and TRIM28 genes, we observed a substantial decrease in the expression of representative targets associated with TFs and their respective targets identified through regulon analysis in this investigation (Fig. 7C–E and Figure S6C–E). Interestingly, via transwell assay, we also discovered that the knockdown of the STAT3, KLF5, and TRIM28 genes dramatically reduced the migration abilities of OCCC cell lines (Fig. 7F and Figure S6F). Consistent conclusions derived from both OCCC cell lines demonstrate that they effectively mirror the malignant-like epithelial cell populations identified in our scRNA-seq analysis. Taken together, our results underscore the crucial significance of STAT3, KLF5, TRIM28, and their downstream targets in OCCC, bolstering the concept that they coordinate an intricate regulatory framework propelling OCCC advancement.

Fig. 7 Functional validation of OCCC-associated TFs in OVISE cells. [Images not available. See PDF.]

A Knockdown efficiency of STAT3, KLF5, and TRIM28 in OVISE cells. B Detection of cell proliferation ability after knocking down STAT3, KLF5, and TRIM28. C Relative mRNA levels of PAX8, MAFK and TRIM28 after STAT3 knockdown. D Relative mRNA levels of PTHLH, SLC2A1, TMB1M1, CS, ACTN4, and CTNNA1 after knocking down TRIM28. E Relative mRNA levels of VCL and ANXA2 after KLF5 knockdown. F Transwell assays of the NC, STAT3 siRNA, KLF5 siRNA and TRIM28 siRNA cells. *p < 0.05, **p < 0.01, ***p < 0.001; scale bar: 100 μm.

Discussion

Ovarian carcinoma is the third most prevalent gynecological malignancy worldwide, posing a significant threat to women’s health^{11, 12, 13–14}. Given its asymptomatic nature during the early stages, EOC is the most lethal of all gynecological cancers, with more than 80% of cases being diagnosed after the tumor has already disseminated within the abdominal cavity^{15, 16–17}. One contributing factor to the lack of substantial progress in this field is the tendency to treat ovarian cancer subtypes as a single entity, even in large-scale clinical trials, disregarding distinct histological subtypes, molecular drivers, and rare pathological types, such as OCCC⁸. OCCC is a unique histologic subtype associated with a poorer prognosis and relative resistance to platinum-based chemotherapy^18,19. Moreover, rOCCC exhibits particularly high chemoresistance, with systemic therapy yielding response rates of less than 10%, even in patients with platinum-sensitive recurrence²⁰. Consequently, the development of novel therapeutic strategies tailored to the distinctive biology of pOCCC and rOCCC is imperative to improve outcomes for patients afflicted by this disease.

Single-cell transcriptomic sequencing has become a relatively mature technology and has been widely used in multiple aspects of research^{21, 22, 23, 24–25}. Previously, transcriptomic atlases of ovarian cancers have also been demonstrated to reveal transcription landscapes at single-cell resolutions, revealing differential transcription factors related to intratumor heterogeneity and chromatin accessibility, identifying immune cell types with possible roles in the cancer niche, and constructing a prognostic model for ovarian cancer metastasis^{26, 27, 28–29}. Recent scRNA-seq studies have revealed specific cell subtypes and tumor heterogeneity in patients with high-grade serous ovarian carcinoma, preliminarily establishing the association between the molecular subtype and disease prognosis^30,31. However, to date, comprehensive studies of malignant-like cell populations derived from epithelial cells in rare pathological types of EOC, such as OCCC, are lacking. In the present study, we not only acquired tissues from four patients with rOCCC who underwent chemotherapy following primary surgical treatment but also obtained a rare fresh pOCCC tissue sample from a patient without prior chemotherapy. Additionally, we combined these samples with four additional published datasets (one normal ovary sample and three endometrioma samples) to explore the origins of OCCC in humans. Within the three endometrioma cases sourced from previously published datasets utilized in this study, instances of superficial peritoneal endometriosis or deep infiltrating endometriosis were observed. Among the three patients, one presented gynecological complications associated with adenomyosis, while another was undergoing exogenous hormone therapy at the time of the surgical procedure. Upon pathological examination, all endometriomas displayed endometrial-type epithelium and stroma, along with the presence of hemosiderin. It is noteworthy that no clinical signs of neoplastic proliferation were identified in these ovarian tissues. Comprehensive patient characteristics for normal ovary and endometrioma specimens utilized in this investigation are detailed in Table S1. Consequently, we successfully identified a substantial proportion of malignant-like cell populations originating from malignant epithelial cells within human ovarian tissues. It is important to acknowledge that in normal ovaries and endometriomas, epithelial cells are present solely on the surface of the ovarian tissue. Our immunostaining analyses reveal distinct presence of PAX8 and KRT7 within the ectopic glandular epithelium of the endometrial glands located within the endometriotic lesion. Specifically, positive staining for both PAX8 and KRT7 was identified in the dilated glandular structures of the endometrioma. These glands, instead of retaining their usual rounded morphology, displayed significant dilation in the histological sections. The expanded state of these glands gives the visual impression that the immunoreactivity forms a “border-like” pattern at the peripheries. Our samples from pOCCC and rOCCC were obtained specifically from the interior of the carcinoma, excluding any normal epithelium within the tumor tissue inside the ovaries. The expressions of PAX8 and KRT7 have been confirmed in OCCC^{32, 33, 34, 35–36}, and our research corroborates these observations. Therefore, our analytical strategy employed a dual-validation concept. Given that our captured epithelial cells represented a heterogeneous mixture (potentially including normal and malignant cells), our first step was to objectively define all epithelial cell populations within the dataset. Subsequently, within these objectively defined epithelial clusters, we rigorously assessed the expression of established markers, including PAX8 and KRT7. Hence, PAX8 and KRT7 positivity in the subcluster of epithelial cells of pOCCC and rOCCC should be regarded as indicative of malignant epithelial cell populations. This finding sheds light on the key factors underlying the development of malignant epithelial cells in both pOCCC and rOCCC.

As noted, normal ovarian tissue harbors a limited number of epithelial cells, whereas endometriomas demonstrate a slight increase in epithelial cell proliferation, and OCCCs exhibit pronounced expansion driven by pathological transformations. Previously, Atiya et al. identified a subset of endometriosis-derived mesenchymal stem cells (enMSCs) characterized by loss of CD10 expression that specifically supports OCCC growth³⁷. Although the cellular origin of OCCC remains controversial, mounting evidence—including histopathological, epidemiological, and molecular data—supports its frequent association with endometriosis and endometriotic ovarian cysts^{38, 39–40}, which may be one of the potential core therory. One of the important developments in our work is that we used PAX8 and KRT7, two classic markers of OCCC, to annotate malignant-like cell populations in epithelial cells. Notably, PAX8- and KRT7-positive cells accounted for a large proportion of the malignant-like cell population in the OCCC and rOCCC samples, while only a few PAX8- or KRT7-positive cells were found in endometriomas. We utilized CNV scores to assess intratumoral heterogeneity and malignant progression in the epithelium of endometriomas, pOCCC and rOCCC. While only a few malignant-like cells were observed in endometriomas, CNV signals increased slightly in the endometrioma epithelium and sharply in the pOCCC and rOCCC epithelia, thereby supporting the theory of an endometrial origin for OCCC. Although there are certain similarities between endometrioma and OCCC, alterations in the expression of pathogenic genes may be pivotal events in the qualitative transformation from endometrioma to OCCC. Recent study also indicated that integrated genomic and epigenomic analyses could provide further insights into the cellular origin of OCCC, though its etiology is likely multifactorial⁴¹. Consequently, additional biomarkers capable of predicting primary and recurrent OCCC must be identified to aid in the elucidation of the origin of malignant-like cell populations in ovarian carcinoma.

Mutations in genes such as adenine thymine-rich interactive domain 1 A (ARID1A) and phosphatidylinositol 4,5-bisphosphate 3-kinase catalytic subunit alpha (PIK3CA) have been identified in approximately 50% to 60% of OCCCs⁴² Moreover, ARID1A and PIK3CA mutations were found to frequently coexist^43,44. ARID1A encodes a member of the switch/sucrose nonfermentable (SWI/SNF) family of proteins and is dependent on glutamine metabolism, and the inhibition of glutaminase alone or in combination with immune checkpoint blockade has synergistic effects on suppressing ARID1A-inactivated tumors⁴⁵. Eutopic endometrial glands in ovarian cancer and endometriosis exhibit a high prevalence of PIK3CA mutations that deviate from the patterns observed in tumors. Studies have demonstrated that oxidative stress significantly down-regulates ARID1A expression in both endometriosis and endometriosis-associated ovarian cancer⁴⁶. Moreover, accumulating evidence indicates that diminished ARID1A expression at both transcriptional and translational levels plays crucial roles in the initiation and progression of endometriosis-related ovarian neoplasms⁴⁶. This association is particularly pronounced in OCCC, where ARID1A deficiency has been mechanistically linked to tumorigenesis^46,47. The consistent observation of ARID1A loss in these malignancies suggests its potential as both a diagnostic marker and therapeutic target. However, it is important to note that the mutations identified in the eutopic endometrium may not represent the sole driver mutations in these cancers⁴⁸.

In the present study, three TFs (STAT3, KLF5 and TRIM28) and their OCCC-relevant targets were found to play key roles in OCCC via our scRNA-seq analysis and functional validation. High levels of hepatocyte nuclear factor (HNF) 1β, which is known to activate STAT3 and NF-κB signaling and lead to immune suppression⁴⁹, were found in OCCC. Previous studies have shown that constitutive activation of STAT3 and key roles of IL6-STAT3-HIF signaling occur in OCCC^50,51. For SMARCC1, only a small number of OCCC cases have SMARCC1 mutations⁴³, and it is still unclear whether these mutations affect the clinical behavior of OCCC. KLF5, the other TF identified in our study, is known to form a transcriptional complex with EHF and ELF3 that drives expression from superenhancer regulatory elements and aberrantly activated during OC progression and PARPi resistance⁵². Targeting the DNA repair pathway represents a fundamental strategy when combined with PARPi to combat resistance⁵³. The down-regulation or mutation of ARID1A hinders DNA homologous recombination repair through the activation of the PI3K/Akt1 pathway, thereby heightening tumor cell susceptibility to PARPi and the combination therapy of PARP inhibitors with ionizing radiation, underscoring the pivotal role of PI3K/Akt1 signaling in sensitizing ARID1A-deficient OCCC to PARPi treatment⁵⁴. Specifically, inhibiting KLF5 also amplifies the sensitivity of ovarian cancer cells to PARPi and fosters DNA damage, positioning KLF5 as a crucial therapeutic target in PARPi resistance⁵². Our investigation unveils the broad regulatory impact of KLF5 on downstream targets, which could potentially contribute to the mechanisms of PARPi resistance in OCCC. Further study is imperative to elucidate the precise regulatory mechanisms through which KLF5 modulates PARPi resistance in OCCC. On the other hand, some studies have confirmed the involvement of TRIM28 in tumor formation^{55, 56–57}, but no strong evidence of an association between TRIM28 and OCCC has been found. Our data provide additional evidence that TFs such as STAT3, KLF5 and TRIM28 and their targets participate in the formation of ovarian malignant-like cell populations and OCCC pathogenicity.

The scRNA-seq data in this study were derived using 3’-end capture technology, which offers limited read depth and short fragment lengths, rendering accurate mutation detection unreliable due to insufficient coverage and alignment precision. Given these limitations, we concentrated on employing complementary methodologies—such as CNV analysis, variance in gene expression, and trajectory modeling—to delineate malignant attributes within epithelial subpopulations. Future investigations incorporating long-read sequencing are imperative to validate mutational drivers within these subpopulations.

In summary, this comprehensive single-cell transcriptional landscape of OCCC serves as a valuable and timely resource, offering insights into the endogenous targets responsible for the emergence of malignant-like cell populations within ovarian epithelial cells. This further enhances our understanding of this rare subtype of EOC. Additional large-scale multiomics studies are undoubtedly warranted to confirm the roles of potential targets and facilitate further advancements in clinical diagnosis and treatment strategies for OCCC.

Materials and Methods

Patients for scRNA-seq

Fresh tumor tissues were collected from pOCCC (n = 1) and rOCCC (n = 4) patients at Fudan University Shanghai Cancer Center. This study was approved by the Institutional Ethics Committee of Fudan University Shanghai Cancer Center (2108241-25). Written informed consent was obtained from all OCCC participants. All ethical regulations relevant to human research participants were followed. OCCC tumor specimens were collected at the time of surgical resection from the untreated primary tumor. Patients with recurrent tumors had undergone cycles of chemo/adjuvant therapy before recurrence. Datasets for normal ovary (n = 1) and endometriomas (n = 3) were obtained from a previously published study¹⁰. In total, nine human ovarian tissues were used for the further analysis of scRNA-seq.

Single-cell solution preparation

The specimens were cut into approximately 1–2 mm³ pieces and digested with a Solo^TM Tumor Dissociation Kit (Sinotech Genomics, JZ-SC-58201) at 37 °C for 30–60 min. Then, enzymatic digestion was stopped by the addition of excess RPMI-1640 medium, and the cell were filtered through a 40 μm cell strainer. The single-cell suspensions were kept on ice before being loaded onto a BD Rhapsody cartridge for single-cell transcriptome capture.

Single-cell transcriptome capture, library construction and sequencing

The cells were first stained with two fluorescent dyes, calcein AM and Draq7, for precise determination of cell concentration and viability via a BD Rhapsody™ Scanner (BD Biosciences). The cells were loaded in one BD Rhapsody microwell cartridge as previously described⁵⁸. The cell capture beads were then loaded in excess to ensure that nearly every microwell contained one bead, and the excess beads were washed away from the cartridge. After the cells were lysed with lysis buffer, the cell capture beads were retrieved, washed, and submitted to reverse transcription. Microbead-captured single-cell transcriptomes were converted into cDNA libraries containing cell label and unique molecular identifier (UMI) information. All procedures were performed with a BD Rhapsody cDNA Kit (#633773, BD Biosciences) and a BD Rhapsody Targeted mRNA & AbSeq Amplification Kit (#633801, BD Biosciences) strictly following the manufacturers’ protocols. All the libraries were sequenced in PE150 mode (paired-end for 150 bp reads) on the NovaSeq platform (Illumina).

Sequencing data processing

The raw sequencing reads were processed using the BD Rhapsody Whole Transcriptome Assay Analysis Pipeline (v1.8) following the official manuals. The paired-end reads containing invalid cell barcodes were filtered out according to the whitelist provided by BD. The sequencing reads were then mapped to the Human Ensembl release 93 reference genome. The transcript expression of each cell was quantified using unique molecular identifiers (UMIs).

Public data usage

The expression matrices of normal ovary and endometrioma were obtained from Gene Expression Omnibus (GEO) database (GSE213216). Along with our sequenced datasets, all expression matrices were pooled together and applied to preprocessing and analysis.

Doublets removal

DoubletFinder (v2.0.3 https://github.com/chris-mcginnis-ucsf/DoubletFinder) was applied on each sample separately to remove potential doublets. The doublet detection procedure briefly comprised the following steps: (1) generating N artificial doublets from the existing scRNA-seq data where N is 25% of total cell number; (2) merging and pre-processing the real and artificial datasets; (3) performing principal component analysis (PCA) with “dims = 50” and computing the proportion of artificial k nearest neighbors (pANN) for each cell within the PC space; and (4) ranking and thresholding the pANN values based on the expected number of doublets. The expected doublet number was estimated as 7.6 * 10^-6 * N + 5.27 * 10^-4 where N is the total cell number in each sample.

Single-cell data analysis

All expression matrices were imported into Seurat (v4.1.1) for downstream analysis. Cells with >43,000 UMIs, >20% mitochondrial gene content, <200 or >6000 detected genes were removed. Cells were retained for downstream processing, including (1) normalization of total UMI counts to 10,000 and log-transformation; (2) selecting the top 3000 highly variable genes using “vst” method in Seurat; (3) regressing out the effects of UMI count and percentage of mitochondrial with linear model using ScaleData function.

The processed data were then used to calculate PCA with 50 components. Batch effects were corrected in PC space with Harmony, an iterative soft clustering method. The first 50 corrected principal components (PCs) were used as input for Uniform manifold approximation and projection (UMAP) with the RunUMAP function (n.neighbors = 30, min.dist = 0.3) and t-distributed stochastic neighbor embedding (t-SNE) with the RunTSNE function (perplexity = 30). Both non-linear dimensional reductions were calculated for visualization.

Unsupervised clustering was performed using the shared nearest neighbor (SNN) modularity optimization algorithm with the Louvain method, based on Euclidean distances in the first 50 Harmony corrected PCA space. The clustering resolution was set to 0.5.

The log-normalized matrices were subsequently loaded into the SingleR R package for cell type annotation, which based on correlating gene expression of reference cell types with single-cell expression. In order to conduct the differential expression analysis for each cluster or cell type, we employed the default “FindMarkers” approach from Seurat, specifically utilizing the non-parametric Wilcoxon rank sum test. Up-regulation analysis for each cell cluster was determined at thresholds of ln(fold change) > 0.25 and P-value < 0.01. The differential expression analyses between two groups for each cell type were conducted using “MAST” as the testing methodology, which accounted for the varying cellular detection rates as a covariate. DEGs with a fold change >2 and a P-value < 0.05 were considered significant. Cell types were identified by comparing the DEGs of each cluster to the canonical marker genes, with assistance by SingleR software.

Expression trend analysis

Expression trend analysis is a methodology that categorizes gene expression patterns based on the shapes of their expression curves across various stages or conditions. This approach facilitates the identification of gene clusters exhibiting similar expression dynamics. In our investigation, we utilized the Short Time-series Expression Miner (STEM) 1.3.11 (https://www.cs.cmu.edu/~jernst/stem/) for conducting the trend analysis. We provided a file containing the expression levels of all DEGs, organized in a biologically logical order. The parameters were configured as follows: -pro 20 -ratio 1.0, where -ratio corresponds to the log₂ value of the fold change threshold for differential expression. STEM subsequently executed the trend analysis, grouping the genes into distinct modules.

Disease Ontology analysis

We leveraged the Disease Ontology (DO), a rigorously standardized framework for human disease classification, to elucidate the potential biomedical signatures associated with the target gene set. DO enrichment analysis was executed by contrasting the target gene set against the entire genomic background. Significance of enrichment for each DO term was assessed using a hypergeometric distribution-based statistical test implemented via the phyper() function in R. Terms yielding P-values < 0.05 were deemed statistically enriched.

CNV estimation analysis

Initial CNVs for samples or groups were estimated by the inferCNV 1.10.0 R package (https://github.com/broadinstitute/infercnv). Normal ovaries that mainly contained nonmalignant epithelial cells were used as controls. InferCNV calculates the expression baseline based on the expression level of genes in the normal ovary, after which the baseline value is subtracted from the expression level of each gene in the cell to obtain the relative expression level. Subsequently, windows of 100 genes were established on the chromosomes to infer CNV events in single cells based on the relative expression levels of the genes in that window. To reduce the possibility of false-positive CNV calls, the default Bayesian latent mixture model was used to determine the posterior probabilities of alterations in each cell. Low-probability CNVs were filtered using the default threshold value of 0.5. Cells with identical patterns of CNV signals were partitioned into groups using K-means clustering. For K-means clustering, we utilized the kmeans function from the stats package in R with the default parameters: iter.max = 10 and nstart = 1. The determination of K, the quantity of clusters, was conducted utilizing the elbow technique. We computed the within-cluster sum of squares (WSS) for various K values and graphed the outcomes. The optimal K value was selected at the “elbow” juncture in the plot, denoting the point at which the rate of WSS reduction undergoes a pronounced shift. This approach facilitates the identification of the juncture where further cluster additions do not notably enhance clustering efficacy. For each cell, the CNV score was assigned using the mean value of the squares of the CNV signals.

Regulon analysis

To carry out transcription factor network inference, regulon analysis was performed on the SCENIC 1.12.0 R package (https://github.com/aertslab/SCENIC)⁵⁹. Briefly, the log-normalized expression matrix generated using Seurat was used as the input. A gene coexpression network was constructed via GENIE3. Each module was pruned based on a regulatory motif near a transcription start site via RcisTarget. Networks were retained if the TF-binding motif was enriched among its targets, while target genes without direct TF-binding motifs were removed. The activity of each regulon for each single cell was scored via the area under the curve (AUC) using the AUCell R package. For each regulon, we use AUCell to calculate the frequency histogram of gene expression levels within cells and represent the activity of regulons in cells by computing the area under the curve. Quantifying regulon activity through the AUCell value reflects whether the gene distribution within regulons in cells is concentrated in regions of high gene abundance. A higher AUC value for the same regulon indicates that the genes within the regulon are more concentrated in regions of high gene abundance within cells, with the transcription factor and associated genes expressing relatively higher levels compared to genes outside the regulon, indicating a stronger regulatory activity in activating target gene expression by the transcription factor.

Single-cell trajectory analysis

Single-cell trajectory analysis was performed using Monocle2 (v2.22.0; https://github.com/cole-trapnell-lab/monocle-release). Cells were ordered along a pseudo-temporal axis based on gene expression patterns, simulating dynamic processes such as differentiation. Dimensionality reduction was conducted using “DDRTree” algorithm (sigma = 0.001, param.gamma = 10, and tol = 0.001). Cells were then arranged into a branched trajectory, with the root defined as the population with the lowest pseudo-time value. Branch points represented cell fate decisions leading to distinct developmental paths. To identify branch-dependent genes, negative binomial generalized linear models were fitted, and significant genes were selected based on an FDR threshold of <1e-7.

Hematoxylin and eosin (H&E) staining

Ovarian tissues were fixed with formalin, dehydrated in an ethanol gradient, and embedded in paraffin after clearing with xylene. Tissue sections were cut to 5 μm, deparaffinized with xylene and rehydrated through an ethanol gradient. Sections were stained with hematoxylin for 10 minutes, differentiated with 1% ethanol hydrochloric acid for 20 seconds, treated with 1% ammonia water for 30 seconds, and then stained with eosin for 3 minutes. The tissue sections were then dehydrated with graded ethanol, cleared with xylene, sealed with neutral resin, and observed under a microscope. Tissue morphology was assessed in images acquired with a NanoZoomer tissue scanner (Hamamatsu).

Immunohistochemistry

The paraffin-embedded ovarian tissues were subjected to deparaffinization in xylene and rehydration through a series of alcohol solutions with decreasing concentrations. The sections were incubated in 3% hydrogen peroxide to block endogenous peroxidase activity and then microwaved in citrate buffer for antigen retrieval. The tissue slides were blocked by incubation with 5% BSA (bovine serum albumin) in PBS for 30 min and incubated with primary antibodies (rabbit anti-PAX8, 1:1000; Abcam, ab191870; rabbit anti-KRT7, 1:1000; Abcam, ab68459) at 4 °C overnight, followed by PBS washes and further incubation with horseradish peroxidase (HRP)-labeled anti-mouse immunoglobulin G (IgG) (H  +  L) (Zsbio, Beijing, China) at room temperature for 2 h. Thereafter, the sections were stained with 3,3’-diaminobenzidine (DAB) using a DAB chromogenic substrate kit (Zsbio, Beijing, China). After counterstaining with Mayer’s hematoxylin, the sections were dehydrated, cleared, and mounted. Tissue morphology was assessed in images acquired with a NanoZoomer tissue scanner (Hamamatsu).

Cell culture and transfection

The OVISE and TOV-21G cell lines were obtained from JCRB Cell Bank and ATCC, respectively, and maintained in DMEM supplemented with 10% fetal bovine serum (FBS) and penicillin/streptomycin (penicillin: 100 units/mL, streptomycin: 0.1 mg/mL) and was cultured at 37 °C in a 5% CO₂ humidified incubator. Prior to experimentation, these cells were tested for mycoplasma contamination.

To obtain stably transfected cells, cells were seeded into 6-well plates (6 × 10⁵ cells/well) and transfected at a final concentration of 100 nM using Lipofectamine 2000 (Lipo2000; 11668019, Invitrogen, USA). The supernatants were collected after 48 h for the following assays. qRT‒PCR was performed to determine the transfection efficiency. The siRNAs (Table S2) were designed and synthesized by RiboBio (Guangzhou, China).

RNA isolation, reverse transcription and qRT‒PCR

Total RNA was extracted by using TRIzol Reagent (15596026, Invitrogen, Waltham, MA, USA) according to the manufacturer’s protocol. First-strand cDNA was synthesized using a PrimeScript™ Reverse Transcriptase Kit (6210 A, Takara). Relative RNA levels were measured by qRT‒PCR with a LightCycler 480 PCR System (Roche, USA) using the TB Green Premix Ex TaqII (RR820, Takara) method. GAPDH was used as the reference gene. The sequences of the primers used are listed in Table S3.

Cell Counting Kit (CCK)-8 proliferation assay

The CCK − 8 assay (Beyotime, Shanghai, China, Cat# C0039) was used to determine OC cell viability. Transfected OVISE cells were digested with trypsin and resuspended in serum-containing medium. The cells in the suspension were counted using a cell counting board to calculate the number of cells per microliter of liquid. The total number of cells was divided by the number of cells per microliter to determine the volume of suspended cell solution that needed to be added. Cells were seeded in 96−well plates at a density of 2000 cells per well. After culture for 0 h, 24 h, 48 h, 72 h and 96 h, 100 µl of medium containing 10 µl of CCK − 8 reagent was added, and the cells were incubated at 5% CO₂ and 37 °C for 2 h. The absorbance was measured at a wavelength of 450 nm using a plate reader.

Cell migration assays

Transwell™ (Merck Millipore, 8 µm pore size) assays were also conducted to test cell migration ability as described previously⁶⁰. After transfection for 48 h, 2 × 10⁴ cells in 300 μl of serum-free medium were inoculated into the upper culture chamber. Medium (700 μl) supplemented with 10% FBS was added to the lower wells of a 24-well plate. After 24 h of incubation, the cells on the upper membrane surface were removed using cotton wool, and the cells on the lower membrane surface were fixed and stained with 0.1% (w/v) crystal violet (Beyotime Institute of Biotechnology). Cell images were captured under a microscope (Olympus, Japan).

Statistics and reproducibility

The quantitative results are presented as the mean ± standard error of the mean (SEM) and were analyzed by using GraphPad Prism 8.0 (GraphPad Software, CA, USA). Two-tailed Student’s t tests were used to determine significant differences between two groups, and one-way analysis of variance (ANOVA) was used to compare more than two groups; Dunnett’s multiple comparison test was used. *p < 0.05; **p < 0.01; ***p < 0.001.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Acknowledgements

We are grateful to Sinotech Genomics Co for scRNA-seq and Gene Denovo Biotechnology Co for assisting bioinformatics analysis in this study. The authors also wish to thank Yu Tang (Guangzhou OmicsMaster Biotechnology Co., Ltd.) and Di Shen (Guangzhou OmicsMaster Biotechnology Co., Ltd.) for their valuable contributions to this work. This study was funded by The National Natural Science Foundation of China (82272898, 82203723, 82271633), A three-year action plan to promote clinical skills and clinical innovation capacity of municipal hospitals by Shanghai Shenkang Hospital Development Center (SHDC2020CR5003-001), the Science and Technology Commission of Shanghai Municipality, China (21ZR1415000).

Author contributions

(I) Conceptualization, Data curation, Formal analysis: Qinhao Guo, Jun Yu, Xiaohua Wu, Bo Zheng and Hao Wen. (II) Funding acquisition, Investigation: Qinhao Guo, Xiaohua Wu, Bo Zheng and Hao Wen. (III) Methodology, Project administration: Qinhao Guo, Xia Chen, Rui Bi, Haiming Li, Meng Liu, Xingzhu Ju, Zheng Feng, Jun Zhu, Yizhen Li, Xin Wang, Qiuru Huang, Jiaxin Li, Xiaonan Zhou, Ying Zheng and Jun Yu. (IV) Writing - original draft: Qinhao Guo, Xia Chen and Jun Yu. (V) Writing - review and editing: Rui Bi, Haiming Li, Xingzhu Ju, Xiaohua Wu, Bo Zheng and Hao Wen. (VI) Final approval of manuscript: All authors.

Peer review

Peer review information

Communications Biology thanks Shannon Hawkins, Natalie Davidson and Amy L. Wilson for their contribution to the peer review of this work. Primary Handling Editors: Kuangyu Yen and Aylin Bircan, Johannes Stortz.

Data availability

The raw sequence data generated during this paper have been deposited in the Genome Sequence Archive (Genomics, Proteomics & Bioinformatics 2021) in National Genomics Data Center (Nucleic Acids Res 2022), China National Center for Bioinformation / Beijing Institute of Genomics, Chinese Academy of Sciences (HRA006442) that are publicly accessible at https://ngdc.cncb.ac.cn/gsa-human/browse/HRA006442. The scRNA-seq data used in the published dataset¹⁰ are available at NCBI GEO under accession number GSE213216. Source data were provided as Supplementary Data 1.

Code availability

Codes used to perform the analysis are available on GitHub: https://github.com/ycli1995/scRNAseq20250428. The specific computational tools and resources utilized in our single-cell analysis include: Seurat¹⁰ version 4.1.1 (https://satijalab.org/seurat/); DoubletFinder¹⁰ version 2.0.3 (https://github.com/chris-mcginnis-ucsf/DoubletFinder); inferCNV version 1.10.0 (https://github.com/broadinstitute/infercnv) (BMC Cancer 2025); SCENIC⁵⁹ version 1.12.0 (https://github.com/aertslab/SCENIC); Monocle2²² version 2.22.0 (https://github.com/cole-trapnell-lab/monocle-release); STEM version 1.3.11 (https://www.cs.cmu.edu/~jernst/stem/) (Aging Cell 2024).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-025-08617-4.

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

References

1. Matsumoto, T et al. Cytoplasmic EBP50 and elevated PARP1 are unfavorable prognostic factor s in ovarian clear cell carcinoma. Carcinogenesis; 2021; 42, pp. 1162-1170. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34323956]

2. Li, R et al. Sintilimab combined with bevacizumab in relapsed/persistent ovarian clear cell carcinoma (INOVA): an investigator-initiated, multicentre clinical trial-a study protocol of clinical trial. BMJ Open; 2022; 12, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/35613822][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9131073]e058132.

3. Kajiyama, H et al. Postrecurrent oncologic outcome of patients with ovarian clear cell carcinoma. Int J. Gynecol. Cancer; 2012; 22, pp. 801-806. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/22617480]

4. McCluggage, WG. Morphological subtypes of ovarian carcinoma: a review with emphasis on new developments and pathogenesis. Pathology; 2011; 43, pp. 420-432. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21716157]

5. Iwamoto, M; Nakatani, Y; Fugo, K; Kishimoto, T; Kiyokawa, T. Napsin A is frequently expressed in clear cell carcinoma of the ovary and endometrium. Hum. Pathol.; 2015; 46, pp. 957-962. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25971546]

6. Rodrigues, S., Braga, M., Félix, A. & Cunha, T. M. Clear cell carcinoma of the ovary: Clues for radiologists to perform a correct diagnosis. Curr. Probl. Diagn. Radiol.53, 271–279 (2024).

7. Barnard, ME et al. Endometriosis Typology and Ovarian Cancer Risk. JAMA; 2024; 332, pp. 482-489. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/39018030][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11255975]

8. Khalique, S; Lord, CJ; Banerjee, S; Natrajan, R. Translational genomics of ovarian clear cell carcinoma. Semin Cancer Biol.; 2020; 61, pp. 121-131. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31698086]

9. Yamaguchi, K et al. Contents of endometriotic cysts, especially the high concentration of free iron, are a possible cause of carcinogenesis in the cysts through the iron-induced persistent oxidative stress. Clin. Cancer Res; 2008; 14, pp. 32-40. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/18172249]

10. Fonseca, MAS et al. Single-cell transcriptomic analysis of endometriosis. Nat. Genet; 2023; 55, pp. 255-267. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36624343][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10950360]

11. Torre, LA et al. Ovarian cancer statistics, 2018. CA Cancer J. Clin.; 2018; 68, pp. 284-296. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29809280][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6621554]

12. Lheureux, S; Braunstein, M; Oza, AM. Epithelial ovarian cancer: Evolution of management in the era of precision medicine. CA Cancer J. Clin.; 2019; 69, pp. 280-304. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/31099893]

13. Yuan, X; Huang, Y; Guo, M; Hu, X; Li, P. Long non-coding RNA FAM83H-AS1 acts as a potential oncogenic driver in human ovarian cancer. J. Ovarian Res; 2021; 14, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/33413565][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7792319]6.

14. Siegel, RL; Miller, KD; Wagle, NS; Jemal, A. Cancer statistics, 2023. CA Cancer J. Clin.; 2023; 73, pp. 17-48. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36633525]

15. Lengyel, E. Ovarian cancer development and metastasis. Am. J. Pathol.; 2010; 177, pp. 1053-1064. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/20651229][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2928939]

16. Kipps, E; Tan, DSP; Kaye, SB. Meeting the challenge of ascites in ovarian cancer: new avenues for therapy and research. Nat. Rev. Cancer; 2013; 13, pp. 273-282. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/23426401][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4673904]

17. Siegel, RL; Miller, KD; Jemal, A. Cancer statistics, 2019. CA Cancer J. Clin.; 2019; 69, pp. 7-34. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30620402]

18. Lheureux, S et al. A Clinical and Molecular Phase II Trial of Oral ENMD-2076 in Ovarian Clear Cell Carcinoma (OCCC): A Study of the Princess Margaret Phase II Consortium. Clin. Cancer Res; 2018; 24, pp. 6168-6174. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30108107]

19. Ye, S; Liu, S; Xiang, L; Wu, X; Yang, H. 18F-FDG PET/CT-based metabolic metrics in recurrent tumors of ovarian clear cell carcinoma and their prognostic implications. BMC Cancer; 2019; 19, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30866858][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6417247]226.

20. Crotzer, DR et al. Lack of effective systemic therapy for recurrent clear cell carcinoma of the ovary. Gynecol. Oncol.; 2007; 105, pp. 404-408. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/17292461]

21. Shi, J et al. Spatio-temporal landscape of mouse epididymal cells and specific mitochondria-rich segments defined by large-scale single-cell RNA-seq. Cell Discov.; 2021; 7, 34. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34001862][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8129088]

22. Yu, J et al. Single-cell RNA sequencing reveals cell landscape following antimony exposure during spermatogenesis in Drosophila testes. Cell Death Discov.; 2023; 9, 86. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36894529][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9998446]

23. Yu, J et al. Single-cell RNA-sequencing reveals the transcriptional landscape of ND-42 mediated spermatid elongation via mitochondrial derivative maintenance in Drosophila testes. Redox Biol.; 2023; 62, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36933391][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10036812]102671.

24. Ji, C et al. Single-cell RNA sequencing reveals the lineage of malignant epithelial cells and upregulation of TAGLN2 promotes peritoneal metastasis in gastric cancer. Clin. Transl. Oncol.; 2023; 25, pp. 3405-3419. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37247132]

25. Wang, Y et al. Identification of genes related to glucose metabolism and analysis of the immune characteristics in Alzheimer’s disease. Brain Res; 2023; 1819, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37619853]148545.

26. Olalekan, S; Xie, B; Back, R; Eckart, H; Basu, A. Characterizing the tumor microenvironment of metastatic ovarian cancer by single-cell transcriptomics. Cell Rep.; 2021; 35, 109165. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34038734]

27. Regner, MJ et al. A multi-omic single-cell landscape of human gynecologic malignancies. Mol. Cell; 2021; 81, pp. 4924-4941.e4910. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34739872]

28. Kan, T et al. Single-cell RNA-seq recognized the initiator of epithelial ovarian cancer recurrence. Oncogene; 2022; 41, pp. 895-906. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34992217]

29. Zhang, D et al. Establishment of an ovarian cancer omentum metastasis-related prognostic model by integrated analysis of scRNA-seq and bulk RNA-seq. J. Ovarian Res; 2022; 15, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36424614][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9686070]123.

30. Olbrecht, S et al. High-grade serous tubo-ovarian cancer refined with single-cell RNA seq uencing: specific cell subtypes influence survival and determine molecular subtype classification. Genome Med; 2021; 13, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34238352][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8268616]111.

31. Hao, Q et al. Single-cell transcriptomes reveal heterogeneity of high-grade serous ovarian carcinoma. Clin. Transl. Med; 2021; 11, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34459128][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8335963]e500.

32. Nonaka, D; Chiriboga, L; Soslow, RA. Expression of pax8 as a useful marker in distinguishing ovarian carcinomas from mammary carcinomas. Am. J. Surg. Pathol.; 2008; 32, pp. 1566-1571. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/18724243]

33. Wong, OGW; Li, J; Cheung, ANY. Targeting DNA Damage Response Pathway in Ovarian Clear Cell Carcinoma. Front Oncol.; 2021; 11, 666815. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34737943][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8560708]

34. Rhodes, A; Vallikkannu, N; Jayalakshmi, P. Expression of WT1 and PAX8 in the epithelial tumours of Malaysian women with ovarian cancer. Br. J. Biomed. Sci.; 2017; 74, pp. 65-70. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28367736]

35. Cameron, RI; Ashe, P; O’Rourke, DM; Foster, H; McCluggage, WG. A panel of immunohistochemical stains assists in the distinction between ovarian and renal clear cell carcinoma. Int J. Gynecol. Pathol.; 2003; 22, pp. 272-276. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/12819395]

36. Ohta, Y; Suzuki, T; Shiokawa, A; Mitsuya, T; Ota, H. Expression of CD10 and cytokeratins in ovarian and renal clear cell carcinoma. Int J. Gynecol. Pathol.; 2005; 24, pp. 239-245. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/15968199]

37. Atiya, HI et al. Endometriosis-Associated Mesenchymal Stem Cells Support Ovarian Clear Cell Carcinoma through Iron Regulation. Cancer Res; 2022; 82, pp. 4680-4693. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36219681][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9755968]

38. Anglesio, MS et al. Clear cell carcinoma of the ovary: a report from the first Ovarian Clear Cell Symposium, June 24th, 2010. Gynecol. Oncol.; 2011; 121, pp. 407-415. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21276610]

39. Pearce, CL et al. Association between endometriosis and risk of histological subtypes of ovarian cancer: a pooled analysis of case-control studies. Lancet Oncol.; 2012; 13, pp. 385-394. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/22361336][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3664011]

40. Anglesio, MS et al. Cancer-Associated Mutations in Endometriosis without Cancer. N. Engl. J. Med; 2017; 376, pp. 1835-1848. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28489996][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5555376]

41. Nishijima, A et al. Integrated genomic/epigenomic analysis stratifies subtypes of clear cell ovarian carcinoma, highlighting their cellular origin. Sci. Rep.; 2024; 14, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/39138354][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11322660]18797.

42. Shoji, T et al. Novel Therapeutic Strategies for Refractory Ovarian Cancers: Clear Cell and Mucinous Carcinomas. Cancers (Basel); 2021; 13, 6120. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34885229]

43. Itamochi, H et al. Whole-genome sequencing revealed novel prognostic biomarkers and promising targets for therapy of ovarian clear cell carcinoma. Br. J. Cancer; 2017; 117, pp. 717-724. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28728166][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5572180]

44. Murakami, R et al. Exome Sequencing Landscape Analysis in Ovarian Clear Cell Carcinoma Shed Light on Key Chromosomal Regions and Mutation Gene Networks. Am. J. Pathol.; 2017; 187, pp. 2246-2258. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28888422]

45. Wu, S et al. Targeting glutamine dependence through GLS1 inhibition suppresses ARID 1A-inactivated clear cell ovarian carcinoma. Nat. Cancer; 2021; 2, pp. 189-200. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34085048][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8168620]

46. Wang, Q; Wang, L; L, Y; Ding, X; Liu, A. Expression and significance of ARID1A mRNA in endometriosis- associated ovarian cancer. J. BUON; 2017; 22, pp. 1314-1321. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/29135119]

47. Ma, Y et al. Aberrant SWI/SNF Complex Members Are Predominant in Rare Ovarian Malignancies-Therapeutic Vulnerabilities in Treatment-Resistant Subtypes. Cancers (Basel); 2024; 16, 3068. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/39272926]

48. Murakami, K et al. Frequent PIK3CA mutations in eutopic endometrium of patients with ovarian clear cell carcinoma. Mod. Pathol.; 2021; 34, pp. 2071-2079. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/34172890][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8514336]

49. Oda, K; Hamanishi, J; Matsuo, K; Hasegawa, K. Genomics to immunotherapy of ovarian clear cell carcinoma: Unique opportunities for management. Gynecol. Oncol.; 2018; 151, pp. 381-389. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30217369][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7526052]

50. Anglesio, MS et al. IL6-STAT3-HIF signaling and therapeutic response to the angiogenesis inhibitor sunitinib in ovarian clear cell cancer. Clin. Cancer Res; 2011; 17, pp. 2538-2548. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/21343371]

51. Bixel, K et al. Targeting STAT3 by HO3867 induces apoptosis in ovarian clear cell carcinoma. Int J. Cancer; 2017; 141, pp. 1856-1866. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28646535][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5812255]

52. Wu, Y et al. KLF5 Promotes Tumor Progression and Parp Inhibitor Resistance in Ovarian Cancer. Adv. Sci. (Weinh.); 2023; 10, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37702443]e2304638.

53. Kulkarni, S; Seneviratne, N; Tosun, Ç; Madhusudan, S. PARP inhibitors in ovarian cancer: Mechanisms of resistance and implications to therapy. DNA Repair (Amst.); 2025; 149, 103830. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/40203475]

54. Yakovlev, VA; Sullivan, SA; Fields, EC; Temkin, SM. PARP inhibitors in the treatment of ARID1A mutant ovarian clear cell cancer: PI3K/Akt1-dependent mechanism of synthetic lethality. Front Oncol.; 2023; 13, 1124147. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36910637][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9992988]

55. Song, T et al. TRIM28 represses renal cell carcinoma cell proliferation by inhibiting TFE3/KDM6A-regulated autophagy. J. Biol. Chem.; 2023; 299, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/36935008][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10141522]104621.

56. Han, W-Y et al. WDR4/TRIM28 is a novel molecular target linked to lenvatinib resistance that helps retain the stem characteristics in hepatocellular carcino mas. Cancer Lett.; 2023; 568, [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37279851]216259.

57. Liang, M et al. E3 ligase TRIM28 promotes anti-PD-1 resistance in non-small cell lung cancer by enhancing the recruitment of myeloid-derived suppressor cells. J. Exp. Clin. Cancer Res; 2023; 42, 275. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/37865804][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10589970]

58. Fan, HC; Fu, GK; Fodor, SPA. Expression profiling. Combinatorial labeling of single cells for gene expression cytometry. Science; 2015; 347, 1258367. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/25657253]

59. Aibar, S et al. SCENIC: single-cell regulatory network inference and clustering. Nat. Methods; 2017; 14, pp. 1083-1086. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/28991892][PubMedCentral: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5937676]

60. Xie, M et al. The long intergenic non-protein coding RNA 707 promotes proliferation and metastasis of gastric cancer by interacting with mRNA stabilizing protein HuR. Cancer Lett.; 2019; 443, pp. 67-79. [PubMed: https://www.ncbi.nlm.nih.gov/pubmed/30502359]