EZH2 Inhibition Restores Tumor Suppressor SFRP1 Activity by Reprogramming Extrachromosomal Circular DNA Dynamics in Ovarian Cancer

Simple Summary

Extrachromosomal circular DNA (eccDNA) contributes significantly to cancer progression through oncogene amplification and increased tumor heterogeneity. However, its regulatory interplay with epigenetic modulators such as EZH2 is not well characterized. Here, the effect of Tazemetostat, a specific EZH2 inhibitor, is examined on both the eccDNA repertoire and gene expression profiles in ovarian cancer. By combining Circle-seq with RNA-seq, we demonstrate that EZH2 suppression leads to substantial reprogramming of eccDNA. Integrated multi-omics revealed 67 genes with changes in eccDNA levels that correlated with altered transcription; among these genes, 11 were identified as potential drivers of drug response. Spatial single-cell RNA sequencing further highlighted that SFRP1 is the only tumor suppressor gene uniformly reactivated following Tazemetostat treatment. SFRP1 also showed pronounced expression heterogeneity in cancer-associated fibroblasts, suggesting that its function within the tumor niche is context-dependent. Our results provide initial evidence that targeting EZH2 restructures eccDNA architecture and reinstates SFRP1-mediated tumor suppression. These findings define a previously unrecognized epigenetics–eccDNA pathway that promotes malignant plasticity and treatment resistance, providing a new conceptual basis for addressing eccDNA-driven cancers.

Abstract

Extrachromosomal circular DNA (eccDNA) has emerged as a pivotal contributor to cancer progression, facilitating oncogene amplification, dysregulated gene expression, and tumor heterogeneity. Despite its significance in cancer, the interplay between eccDNA and key epigenetic regulators such as EZH2 remains largely unexplored. In this study, we systematically investigate the correlation between Tazemetostat, a highly selective EZH2 inhibitor, and alterations in the eccDNA landscape and transcriptional programs in ovarian cancer. Through integrated profiling using Circle-seq and RNA sequencing, we demonstrate that EZH2 inhibition is associated with markedly reprogrammed eccDNA dynamics. Furthermore, multi-omics integration identified that 67 genes exhibited concordant changes in both eccDNA abundance and transcript expression. Subsequent analyses also pinpointed 11 genes as putative effectors of drug response. Notably, spatial single-cell transcriptomics identified SFRP1 as the most consistently reactivated tumor suppressor across eccDNA, bulk expression, and spatial datasets, based on predefined statistical and biological criteria, by Tazemetostat. Moreover, SFRP1 was one of the genes that varied the most within cancer-associated fibroblast populations, exhibiting distinct spatial expression patterns. Taken together, this study establishes the first potential evidence that EZH2 inhibition may reprogram eccDNA dynamics to potentially restore SFRP1 tumor suppressor expression in ovarian cancer. By integrating multi-omics and spatial single-cell transcriptomics, we uncovered a novel epigenetic–eccDNA axis that may contribute to oncogenic plasticity and therapeutic resistance. This could result in a paradigm shift in targeting eccDNA-driven malignancies.

1. Introduction

Epigenetic modifications are recognized as central drivers of gene regulation and are increasingly implicated in the progression of cancer. Enhancer of zeste homolog 2 (EZH2), a catalytic component of the polycomb repressive complex 2, catalyzes the trimethylation of histone H3 at lysine 27 (H3K27me³), leading to gene silencing and affecting critical cellular pathways involved in proliferation, differentiation, and survival [1,2]. EZH2 is dysregulated across multiple cancer types, including ovarian cancer, and is associated with aggressive disease phenotypes, therapy resistance, and poor prognosis [3]. Targeting EZH2 with small-molecule inhibitors, such as Tazemetostat, has emerged as a promising therapeutic approach to mitigate EZH2 and its oncogenic effects [4]. Preclinical studies have demonstrated the antitumor activity of Tazemetostat in various solid tumor models, including ovarian cancer [5]. Although FDA-approved, Tazemetostat has shown inconsistent efficacy as a monotherapy in solid tumors [1]; innovative combination strategies are needed in patients with elevated EZH2 activity.

Extrachromosomal DNAs (ecDNAs) drive oncogene amplification and intratumoral heterogeneity in cancer [6,7]. As such, tumors containing EcDNA have elevated replication stress and can be selectively targeted by CHK1 inhibition [8].

The circular form of extrachromosomal DNA, known as extrachromosomal circular DNA (eccDNA), has gained attention as a distinctive molecular entity in cancer. It has the potential to drive oncogene amplification and increase tumor heterogeneity. Unlike linear chromosomal DNA, eccDNA exists independently in circular form and can carry genomic fragments, including entire genes, regulatory elements, and repetitive sequences, resulting in changes to dynamic gene expression [6,9]. Studies have shown that eccDNA can act as a reservoir for oncogenes and resistance factors, contributing to adaptive responses under therapeutic pressure [6,10]. For instance, by investigating the spatial–temporal evolution of eccDNA in glioblastoma, researchers found that ecDNA amplification, particularly of EGFR, is linked to treatment resistance and shorter survival. Thus, a potential therapeutic window is possible for early intervention. While ecDNA has shown significant effects on gene expression and can be influenced by drug treatments [11], the specific molecular impact of EZH2 inhibition on eccDNA profiles remains largely unexplored.

Several clinical trials of Tazemetostat have been conducted in ovarian cancer, yet its efficacy has been less than satisfactory. The mechanisms of action and resistance to Tazemetostat in ovarian cancer remain unclear. In this study, we utilized TOV-112D ovarian cancer cells to evaluate Tazemetostat-related changes in eccDNA profiles. This could yield valuable information regarding the influence of eccDNA formation and oncogene expression on EZH2 inhibition. Specifically, investigating alterations in eccDNA-associated gene expression and the potential of these changes as prognostic biomarkers may provide new avenues for understanding drug responses and patient stratification in ovarian cancer [12].

To explore these questions, we assessed how treatment with Tazemetostat modulates eccDNA profiles in TOV-112D ovarian cancer cells. By integrating eccDNA mapping with transcriptomic analysis, we aimed to identify eccDNA-associated genes that respond to EZH2 inhibition. Incorporating spatial single-cell transcriptomic data enabled further prioritization of genes with potential clinical relevance [13]. Our findings provide evidence that EZH2 inhibition is associated with changes in eccDNA architecture and expression of tumor suppressor-related genes, revealing a potential link between epigenetic regulation and eccDNA-mediated alterations to gene expression.

Importantly, unlike previous studies that primarily focused on eccDNA as carriers of oncogene amplification or drug resistance determinants, the present study introduces a conceptual framework in which epigenetic perturbation is associated with systematic remodeling of eccDNA landscapes at a genome-wide scale. By integrating Circle-seq, transcriptomics, and spatial single-cell data, this work extends the functional scope of eccDNA research from structural variation to epigenetically linked transcriptional coordination, thereby providing a novel perspective on eccDNA regulation under targeted therapy.

2. Materials and Methods

2.1. Cell Culture and Treatment

The human ovarian carcinoma cell line TOV-112D was maintained in Dulbecco’s Modified Eagle Medium (DMEM) containing 10% fetal bovine serum (FBS; Yeasen, Shanghai, China), along with 100 U/mL penicillin and 100 μg/mL streptomycin (BasalMedia, Shanghai, China). All cell lines were authenticated using PCR analysis and verified as free from mycoplasma contamination. Cells were plated at optimal densities and allowed to adhere for 12–24 h prior to treatment. Subsequently, TOV-112D cells were exposed to 1 μM Tazemetostat (EPZ-6438)—an inhibitor targeting EZH2—for 24 h. This compound was obtained from Selleck Chemicals (catalog no. S7128, Houston, TX, USA). All experiments were performed with three independent biological replicates (n = 3), each representing cells from separate passages treated on different days.

2.2. Immunoblotting

Proteins were extracted in an ice-cold lysis buffer with 50 mM Tris/HCl (pH 7.5), 0.5% Nonidet P-40 (NP-40), 1 mM EDTA, 150 mM NaCl, 1 mM dithiothreitol (DTT), 0.2 mM phenylmethylsulfonyl fluoride (PMSF), 10 μM pepstatin A, 1 μg/mL leupeptin, and 10% protease inhibitor cocktail. Equal amounts of clear cell lysate (20–80 μg) were used for immunoblotting (IB) analysis. File S1 presents the original versions of the gel and blot images.

2.3. Circular DNA Isolation, Purification and Sequencing

EccDNA was performed using an optimized workflow adapted from the Circle-seq protocol [14]. Briefly, genomic DNA was extracted from TOV-112D cells utilizing the TIANprep Mini Plasmid Kit (TIANGEN, Beijing, China). The total DNA fraction was then subjected to an alkaline denaturation–renaturation process to separate chromosomal DNA, lipids, and proteins. Purification was achieved via ion-exchange chromatography on a Plasmid Mini AX column (A&A Biotechnology, Gdańsk, Poland).

To remove contaminating linear DNA, the preparation was treated with Plasmid-Safe ATP-dependent DNase (Epicentre, Madison, WI, USA) and supplemented with the MssI restriction enzyme, which selectively digests mitochondrial circular DNA (approximately 16 kb), thereby facilitating exonuclease access to any remaining linear DNA ends. The efficiency of linear DNA removal was verified via quantitative PCR (qPCR) targeting the COX5B locus.

EccDNA-enriched material was then amplified through phi29 polymerase-mediated rolling circle amplification using the REPLI-g Midi Kit (QIAGEN, Hilden, Germany), and the amplified products were purified with AMPure XP magnetic beads (catalog no. A63880; Beckman Coulter, Indianapolis, IN, USA) and fragmented to an average length of 200–300 bp via ultrasonic shearing (Bioruptor, Diagenode, Seraing, Belgium). DNA libraries were prepared following the NEBNext Ultra DNA Library Prep Kit protocol (New England Biolabs, Ipswich, MA, USA) and sequenced on an Illumina NovaSeq 6000 platform (Illumina, San Diego, CA, USA) using paired-end 150 bp (PE150) reads, yielding approximately 80 million read pairs per sample. The Circle-seq sequencing and library preparation were carried out by Jiayin Biotechnology Ltd. (Shanghai, China). Quality control was performed at multiple steps: (1) DNA concentration measured by NanoDrop (A260/280 ratio > 1.8, A260/230 ratio > 2.0); (2) DNA integrity verified by agarose gel electrophoresis; and (3) eccDNA enrichment confirmed by qPCR showing >95% reduction in COX5B linear DNA signal.

2.4. The Bioinformatics Analysis of eccDNA Abundance Data

To identify robust molecular targets driven by EZH2-regulated eccDNAs, we adopted a two-tier screening strategy. First, we defined a ‘Discovery Set’ of 67 genes based on our experimental data in TOV-112D cells, selecting candidates that showed concurrent upregulation in both Circle-seq (eccDNA abundance) and RNA-seq (mRNA expression) following Tazemetostat treatment. Second, to ensure clinical relevance and reduce cell-line artifacts, we further refined this list by cross-referencing it with external clinical datasets, resulting in a ‘Cross-validated Subset’ of 11 genes that maintained prognostic significance in patient cohorts. The following analyses proceed from this broad discovery set to the refined, validated subset. Sequencing data from the Circle-seq assay were analyzed to detect and quantify eccDNA elements. The Circle-Map software package (version 1.1.4) was employed to identify circular DNA structures by recognizing sequence breakpoints and patterns characteristic of eccDNA [15]. This approach leverages soft-clipped reads and structural variant signals within the sequencing data to define eccDNA junctions with high precision.

Following detection, Samtools (version 0.2) was used to extract raw counts of soft-clipped reads corresponding to these breakpoints [16], which provided an estimate of the relative abundance of each eccDNA molecule. To identify eccDNA species showing significant differences in expression between control and Tazemetostat-treated samples, the edgeR package (version 0.6.9) was applied [17]. Statistical significance was determined using the quasi-likelihood F-test with FDR correction (Benjamini–Hochberg method). Differential eccDNA elements were defined as those with |log2FC| > 1 and FDR < 0.05, and the analysis incorporated both statistical thresholds and biological relevance to ensure reliable results.

Genomic annotation of the detected eccDNA loci was carried out using Bedtools (version 2.27.1), allowing positional data to be integrated with known gene features. Annotated eccDNA coordinates were reformatted into BED files, which facilitated downstream genomic overlap analyses. These files were cross-referenced with entries from circBase [18], a public repository for circular RNA data, to determine overlapping or unique eccDNA elements across control and treatment groups. The final datasets categorized eccDNA elements as upregulated or downregulated, providing the foundation for subsequent integrative analyses.

2.5. Transcriptome Sequencing

Total RNA was isolated from cell samples using the TRIzol reagent (Cat. No. 15596018; Thermo Fisher Scientific, Waltham, MA, USA). RNA quality and purity were evaluated using several complementary approaches: 1% agarose gel electrophoresis was used to check integrity and rule out degradation or contamination; the NanoPhotometer^® spectrophotometer (IMPLEN, Westlake Village, CA, USA) assessed sample purity; and the Qubit^® RNA Assay Kit with the Qubit^® 2.0 Fluorometer (Life Technologies, Carlsbad, CA, USA) determined RNA concentration. RNA integrity was further verified using the Agilent 2100 Bioanalyzer and its RNA Nano 6000 Kit (Agilent Technologies, Santa Clara, CA, USA). For each sample, 3 µg of total RNA was used to construct the sequencing library, and libraries were prepared using the NEBNext^® Ultra™ RNA Library Prep Kit for Illumina^® (Cat. No. E7490, USA; New England Biolabs, Ipswich, MA, USA), with index barcodes added to label individual samples.

Messenger RNA was enriched from total RNA using oligo(dT)-attached magnetic beads. The isolated mRNA was fragmented under elevated temperature in the NEBNext First-Strand Synthesis Reaction Buffer (5×). First-strand cDNA was synthesized with random hexamer primers and M-MuLV Reverse Transcriptase (RNase H −), followed by second-strand cDNA synthesis using DNA Polymerase I and RNase H. Overhanging ends were blunted through enzymatic activity, and A-tailing of 3′ ends was performed prior to ligation with hairpin-loop adaptors. Size selection of 250–300 bp fragments was achieved using the AMPure XP system (Beckman Coulter, Indianapolis, IN, USA).

Adaptor-ligated fragments were treated with USER Enzyme (New England Biolabs, Ipswich, MA, USA) at 37 °C for 15 min, followed by a 5 min incubation at 95 °C. PCR amplification was performed using Phusion High-Fidelity DNA polymerase, universal and index-specific primers. Final libraries were purified using the AMPure XP system and assessed for quality on the Agilent 2100 Bioanalyzer. Cluster generation was performed using a TruSeq PE Cluster Kit v3-cBot-HS (Illumina, San Diego, CA, USA) on the Illumina cBot system, followed by sequencing on an Illumina NovaSeq 6000 platform to produce 150 bp paired-end reads. Sequencing services were conducted by Jiayin Biotechnology Ltd. (Shanghai, China).

2.6. The Bioinformatics Analysis of Transcriptome Data

Raw FASTQ files obtained from RNA sequencing were subjected to quality assessment and cleaning. Adapter sequences, reads containing poly-N stretches, and low-quality bases were removed to produce high-confidence clean reads. Key quality indicators—Q20, Q30, and GC content—were calculated to evaluate sequence integrity and ensure that only reliable reads were included for downstream analyses. These steps established a robust dataset for accurate transcriptome profiling.

Clean reads were aligned to the reference genome using STAR (v2.7.9a) (Spliced Transcripts Alignment to a Reference), which is an aligner known for its exceptional mapping speed and precision in detecting splice junctions. STAR not only identifies canonical and noncanonical splice sites but also captures chimeric or fusion transcripts, enabling comprehensive analysis of transcriptomes. Its ability to process full-length RNA molecules provides a deeper view of gene structure and alternative splicing events. STAR alignment parameters are listed as follows: –outFilterMultimapNmax 20, –alignSJoverhangMin 8, –alignSJDBoverhangMin 1, –outFilterMismatchNmax 999, –alignIntronMin 20, and –alignIntronMax 1000000.

Gene-level quantification was performed using HTSeq (v0.6.0) [19], which counted the number of reads mapped to each gene. Expression levels were normalized as FPKM (Fragments Per Kilobase of transcript per Million) mapped reads. By accounting for both sequencing depth and gene length, meaningful comparisons were drawn between genes and samples. Differential gene expressions between experimental groups were analyzed using the edgeR package (v3.36.0). Genes were considered significantly differentially expressed if they met the criteria of an absolute log₂ fold change > 1 and an adjusted p-value < 0.05 after false discovery rate (FDR) correction. This statistical framework ensured that only biologically meaningful and statistically reliable changes in gene expression were retained for interpretation.

2.7. Correlation Analysis of eccDNA and RNAseq

To integrate eccDNA sequencing data with corresponding RNA expression profiles from the same ovarian cancer cell samples, a multi-step bioinformatics strategy was employed. This pipeline enabled the systematic comparison of eccDNA occurrence and transcriptional activity to identify potential regulatory associations. The eccDNA dataset was first annotated with genomic coordinates, and each circular DNA element was linked to its chromosomal origin. This mapping step provided the foundation for cross-referencing eccDNA loci with genes located within or near those regions, allowing insights into their possible functional impact. The integrated workflow consisted of the following steps: (1) Genomic annotation: eccDNA coordinates were mapped to genes using Bedtools (v2.30.0) intersect with a 10 kb window. (2) Data merging: eccDNA read counts and gene expression values (FPKM) were matched by gene symbol. (3) Correlation analysis: Pearson correlation coefficients were calculated between eccDNA abundance and gene expression across all six samples. (4) Significance filtering: Genes with |r| > 0.8 and p < 0.05 were retained as eccDNA-correlated genes (n = 67).

Next, the eccDNA and RNAseq datasets were merged based on shared genomic coordinates. This integration made it clear whether specific eccDNA elements correlated with alterations in gene expression patterns. The relationships between eccDNA presence and transcript abundance were quantified through correlation analyses, identifying instances where circular DNA formation may coincide with the upregulation or downregulation of nearby genes.

By aligning eccDNA loci with corresponding expression changes, we were able to assess the potential cis-acting effects of eccDNA on gene transcription within the same genomic neighborhood. This integrative approach provided a framework to examine whether eccDNA formation influences transcriptional regulation, thereby revealing new layers of complexity in gene regulation and cancer genome dynamics.

2.8. Public Data Integration for Tazemetostat-Responsive Genes

The primary goal of integrating public data was to validate the differentially expressed genes identified in our dataset and establish a high-quality list of Tazemetostat-responsive genes. To accomplish this, we accessed and downloaded the transcriptomic dataset GSE262877 from the Gene Expression Omnibus (GEO) database [20].

For each of these groups, we performed a differential expression analysis to identify genes with altered expression due to Tazemetostat treatment by comparing the three treated samples against three untreated samples. The processed gene expression matrix was downloaded from GEO, and analysis was conducted using the ‘limma’ (Linear Models for Microarray Data) package in R (v4.4.2) [21]. We began by normalizing the expression data to ensure sample comparability. ‘Limma’ then fit linear models to each gene’s expression values, estimating gene-specific means and variability between groups. Through empirical Bayes moderation, ‘limma’ enhances statistical power, particularly for smaller datasets, by employing variance shrinkage toward a common mean. This process produces results that include p-values, adjusted for multiple testing and fold-change values, effectively highlighting genes that are significantly up- or downregulated in response to Tazemetostat treatment. This approach provides a reliable and well-annotated resource for future research on Tazemetostat-responsive genes. To observe shared differentially expressed genes (DEGs) among different treatments, an upper triangle heatmap was implemented using the Seaborn package (v0.11.2) in Python (v3.9.7) [22]. To ensure the robustness of the identified targets, we defined a ‘validated gene list’ by selecting genes that were consistently identified as differentially expressed (|log2FC| > 1 and p < 0.05) in at least three out of the five independent comparison groups (ne1262, ne154, ne155, prabl, and prlnc) within the GSE262877 dataset.

2.9. Functional Enrichment and Mutational and Survival Analysis

Functional annotation of differentially expressed or eccDNA-associated genes was conducted using the g:Profiler web platform. Significance of enrichment was determined using all human protein-coding genes as the reference background, and terms were ranked based on adjusted p-values. Enriched Gene Ontology (GO) terms and pathway categories provided insights into biological processes, molecular functions, and cellular components significantly associated with the input gene sets.

For data visualization, we employed the DOSE package from Bioconductor (v3.20.0), which facilitates enrichment mapping and semantic similarity calculations among annotated terms [23]. Functionally related terms were connected in the resulting networks, and edge thickness represented the degree of similarity. This network-based approach highlights biologically coherent modules, simplifies interpretation by clustering redundant annotations, and reveals disease-related or functionally linked gene groups.

Survival analyses were performed using TCGA pan-cancer data (n = 2683 samples) accessed via cBioPortal [24]. For immunotherapy-specific survival analysis, we utilized already-published cohorts of anti-PD-1 treatment groups as annotated in the original spatial transcriptomics study [25]. These datasets were selected based on (1) availability of matched genomic and survival data; (2) adequate sample size for statistical power; and (3) clinical annotation quality.

2.10. Spatial Single-Cell Transcriptome Data Processing and Downstream Analysis

This study utilized publicly available spatial single-cell transcriptome data [25], which was processed following the methodology detailed in [25]. Briefly, raw sequencing data were preprocessed to remove low-quality reads, and the gene expression levels were quantified at the single-cell level. Spatial barcoding information was decoded to map individual cells to their respective spatial locations within the tissue, and doublet detection and removal were performed to ensure the purity of single-cell datasets. The resulting processed data were quality-controlled to filter out low-quality cells and genes based on metrics such as gene count, UMI count, and mitochondrial gene expression.

Following preprocessing, the spatial single-cell transcriptome data were analyzed using a pipeline consistent with the methods described in [25]. Dimensionality reduction was performed using principal component analysis (PCA), followed by clustering and visualization via uniform manifold approximation and projection (UMAP). Cell clusters were annotated based on marker gene expression profiles. Spatially resolved gene expression patterns were visualized using Seurat (v5) [26]. All analyses were conducted using R (version 4.4.2), and reproducibility was ensured through version-controlled workflows.

To characterize transcriptional heterogeneity and spatial gene expression patterns, we utilized the Seurat package (v5) to identify both highly variable genes (HVGs) and spatially variable genes (SVGs) from spatial single-cell transcriptome data. HVGs were first identified using the FindVariableFeatures function with default parameters, allowing us to capture genes that exhibit high variability in expression across individual cells, independent of their spatial context. To uncover genes exhibiting spatially structured expression patterns, we applied the FindSpatiallyVariableFeatures function, which integrates spatial coordinates and gene expression to detect SVGs enriched in specific tissue regions. This dual approach enabled the identification of both cell-intrinsic and spatially patterned gene expression programs, providing critical insights into spatial heterogeneity and potential microenvironmental interactions within the tumor landscape.

2.11. Statistical Analysis

All statistical tests were two-tailed unless otherwise specified. For differential expression analyses (RNA-seq and eccDNA-seq), p-values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) method. Statistical significance was defined as FDR < 0.05 and |log2FC| > 1. Sample sizes (n = 3 biological replicates) were determined based on standard practices in transcriptomic studies and pilot data variance estimates. Correlation analyses used Pearson correlation coefficients with significance defined as p < 0.05.

3. Results

3.1. Tazemetostat Reduces H3K27me³ Levels in TOV-112D Ovarian Cancer Cells

In this study, we developed a comprehensive computational workflow (Figure 1A) to systematically investigate the effects of Tazemetostat on eccDNA dynamics and associated gene expression. Firstly, TOV-112D ovarian cancer cells were treated with Tazemetostat in three independent replicates, while the samples treated with DMSO served as controls. IB analysis confirmed that Tazemetostat treatment reduced H3K27me³ levels and increased H3K27Ac levels, without affecting EZH2 expression (Figure 1B). These results confirm the on-target activity of the inhibitor in our experimental model. As expected, Tazemetostat reduces H3K27me³ without altering EZH2 protein abundance, which is consistent with its known mechanism as a catalytic inhibitor rather than a transcriptional or protein-stability modulator. Both treated and untreated groups underwent Circle-seq to profile eccDNA and RNA sequencing (RNAseq) and capture transcriptomic changes. This dual profiling generated high-resolution data on eccDNA abundance and gene expression in response to Tazemetostat. Secondly, RNAseq data were correlated with the Circle-seq results to identify genes with expression changes specifically linked to eccDNA. This correlation revealed 67 genes that exhibited consistent regulation of eccDNA abundance and expression due to EZH2 inhibition.

Comparative analyses using multiple published drug response datasets were undertaken to explore reliable Tazemetostat-response genes. Differential gene expressions between Tazemetostat-treated and untreated cell lines were cross-referenced with external data, including the GSE262877 dataset from the NCBI Gene Expression Omnibus. This external dataset provided additional validation, focusing on consistent drug response genes across different experimental conditions. Lastly, through the integration of experimental and public datasets, a final set of 11 eccDNA-associated genes was identified (e.g., ABCB1, KIT, and SFRP1). These genes were further characterized by analyzing their mutation patterns and association with patient survival outcomes using publicly available cancer genomics databases. Building on this clinical relevance, we also conducted an exploratory analysis of immunotherapy. Rather than serving as an experimental prediction, this analysis involved overlaying Tazemetostat-responsive genes with published survival data from a cohort not treated with a PD-1 to explore potential prognostic implications. Given the growing prominence of immunotherapy in ovarian cancer treatment, this supplementary evaluation was included in this study to contextualize our findings within a broader therapeutic landscape and to highlight avenues for future clinical application. In summary, this workflow highlights eccDNA’s regulatory role in mediating drug response under EZH2 inhibition, providing novel biomarkers and therapeutic targets for precision oncology. The stepwise integration of experimental and external datasets ensures robust and biologically relevant findings, providing a versatile framework applicable to any eccDNA and RNAseq data analysis.

3.2. Characterization of Tazemetostat-Related Changes in eccDNA Features

The impact of Tazemetostat treatment on eccDNA dynamics was investigated by analyzing multiple features of eccDNA in treated and control samples. Figure 2A illustrates the size distribution of eccDNA in Tazemetostat-treated samples (yellow) and control samples (red). We observed distinct variations in eccDNA size profiles between the two groups, indicating that Tazemetostat may be associated with altered eccDNA size dynamics in TOV-112D cells. Specifically, certain size ranges of eccDNA were more prevalent in treated samples, suggesting that EZH2 inhibition selectively modulates eccDNA formation. These differences in size distribution may reflect changes in gene regulatory circuits influenced by Tazemetostat, with eccDNA sizes potentially serving as markers for drug response.

The molecular impact of Tazemetostat was further assessed by examining the GC content distribution patterns of eccDNA regions and their flanking sequences. For each condition (treated and control), we randomly selected 1000 eccDNA sequences and analyzed three regions: the eccDNA sequences themselves, their upstream regions (1000 bp upstream from the eccDNA start), and their downstream regions (1000 bp downstream from the eccDNA end). We also generated 1000 random genomic sequences of equivalent length as the eccDNA for comparison. As shown in Figure 2B,C, we observed distinct GC content distribution patterns between Tazemetostat-treated and control samples. In both conditions, the GC content of eccDNA and its adjacent regions differed from the random genomic background, with treated samples showing a notably different distribution pattern compared to controls. These differences in GC content patterns suggest that Tazemetostat treatment may correlate with the selection or formation of eccDNA from regions with specific GC compositions, potentially affecting the regulatory potential of these sequences.

Next, we examined the chromosomal distribution of eccDNA to determine whether Tazemetostat treatment affects eccDNA frequencies across the genome. Figure 2D presents the eccDNA counts per megabase (Mb) across chromosomes in treated and control samples. Our analysis revealed a non-uniform distribution, with notable reductions in eccDNA counts for several chromosomes, such as chromosomes 1, 2, and 3, in Tazemetostat-treated samples. This suggests that Tazemetostat may be linked to reduced eccDNA formation or stability in a chromosome-specific manner. These findings indicate that changes in eccDNA distribution are non-uniform across chromosomes following EZH2 inhibition.

Finally, to investigate the functional significance of eccDNA, we analyzed its distribution across genomic regions adjacent to genes and CpG islands. As shown in Figure 2E, Tazemetostat-treated samples exhibited differential eccDNA coverage in these regions compared to controls. Specifically, we observed increased eccDNA coverage near CpG islands, particularly in regions 2 kb upstream (CpG2kbU) and downstream (CpG2kbD) or adjacent to genes (Gene2kbU and Gene2kbD). These results suggest that Tazemetostat influences the formation or maintenance of eccDNA in gene-regulatory regions. Additionally, shifts in eccDNA coverage across functional genomic elements, such as exons, introns, and untranslated regions (UTRs), highlight the potential role of eccDNA in mediating gene expression changes. These findings provide insights into how EZH2 inhibition selectively impacts eccDNA distribution, offering suggestions about its regulatory roles in ovarian cancer cells.

This comprehensive analysis of eccDNA characteristics underscores the multifaceted impact of Tazemetostat on eccDNA size, composition, chromosomal distribution, and functional relevance, shedding light on its mechanism of action in modulating gene regulatory networks.

3.3. The Transcriptome Analysis Identified Significant Pathways Linked to Up- and Downregulated Genes in Response to Tazemetostat Treatment

Differential expression and enrichment analyses based on RNAseq data were conducted to investigate the effects of EZH2 inhibition when treated with Tazemetostat in TOV-112D ovarian cancer cells. Figure 3A presents an overview heatmap comparing gene expression patterns between Tazemetostat-treated and control samples. The visualized heatmap reveals distinct expression patterns across the three treated and three control samples. Orange indicates upregulated genes and purple indicates downregulated genes. Hierarchical clustering was applied to group genes with similar expression profiles, potentially identifying functionally related genes or shared pathway involvement. In brief, our analysis identified 573 upregulated and 22 downregulated genes (|log2FC| > 1, p-value < 0.05) in response to Tazemetostat treatment. These 595 differentially expressed genes (DEGs) provide insights into the molecular mechanisms of EZH2 inhibition and its impact on eccDNA dynamics in TOV-112D cells, potentially revealing new prognostic markers for ovarian cancer.

The function of the 595 DEGs was explored through Gene Ontology (GO) enrichment map analysis. This involved annotating the genes with GO terms and connecting the enriched terms based on semantic similarity. Figure 3B illustrates the network analysis of these DEGs, revealing various biological processes influenced by Tazemetostat treatment. The key positive regulatory processes identified included the regulation of cell communication, positive regulation of multicellular organismal processes, and positive regulation of developmental processes. These processes are essential for cell signaling, development, and differentiation, suggesting that Tazemetostat may enhance these functions in ovarian cancer cells. Conversely, notable negative regulatory processes include the negative regulation of biological processes, signaling, and developmental processes.

To further distinguish between up- and downregulated genes, an additional KEGG pathway enrichment analysis was performed. Figure 3C displays the top 20 significantly enriched pathways associated with 573 upregulated genes. The most significantly enriched pathways include “Pathways in cancer,” “Transcriptional misregulation in cancer,” “PI3K-Akt signaling pathway,” and “Wnt signaling pathway.” These pathways are known to play critical roles in oncogenesis. For instance, “Pathways in cancer” highlights key signaling cascades frequently altered in malignancies, serving as central hubs for tumor progression and therapeutic targeting. The “PI3K-Akt signaling pathway” is a crucial regulator of cell growth, survival, and metabolism, and its dysregulation is a hallmark of many cancers, contributing to resistance to apoptosis and uncontrolled proliferation. The “Transcriptional misregulation in cancer” pathway highlights the impact of genetic and epigenetic changes in altering gene expression, driving tumorigenesis. These enriched pathways provide valuable insights into the molecular mechanisms underlying cancer in this context and highlight potential targets for therapeutic intervention. By identifying specific pathways with significant impact, this analysis offers a foundation for further investigation into the regulatory networks driving oncogenesis and their potential clinical applications.

As shown in Figure 3D, the 22 downregulated differentially expressed genes are significantly enriched in pathways such as “Pathways in cancer,” “Herpes simplex virus 1 infection,” “Neuroactive ligand–receptor interaction,” and “Calcium signaling pathway.” Among these, “Pathways in cancer” showed the highest level of enrichment based on its −log10(p-value), indicating its critical involvement in the biological processes affected by the experimental conditions. This pathway encompasses multiple signaling networks and molecular mechanisms central to tumorigenesis, including cell proliferation, apoptosis, and immune response. The significant enrichment of this pathway underscores the potential of the identified genes to influence cancer-related processes, making them targets for therapeutic intervention.

The “Calcium signaling pathway,” another highly enriched pathway, plays a primary role in regulating intracellular signaling processes, such as cell growth, apoptosis, and cellular metabolism. Disruption in calcium signaling is commonly implicated in cancer progression, particularly in processes like metastasis. The presence of this pathway among the most enriched pathways highlights its contribution to the changes in gene expression observed. Overall, the significant enrichment of pathways involved in cancer and cellular signaling underscores the intricate interplay between these downregulated genes and their potential functional roles in oncogenesis and cellular homeostasis.

3.4. The Correlation Between eccDNA and Gene Expression

In our study, we matched sequencing data for eccDNA and gene expression across six samples, providing an excellent opportunity to identify consistent changes in eccDNA-correlated gene expression. By correlating the eccDNA and RNAseq data, we identified 67 genes. Figure 4A offers a comprehensive functional overview of these 67 genes, the expression of which was consistently modulated by eccDNA dynamics in response to EZH2 inhibition in TOV-112D cells (Table S1). These genes were identified through an integrative analysis of eccDNA mapping and RNAseq data, linking the dynamics of extrachromosomal DNA with transcriptional changes in the cellular response to Tazemetostat treatment. The hallmark enrichment plot illustrates how these genes correspond to various established cancer hallmark categories by comparing them to reference gene sets. Significance is visually represented by colored slices, with the red dotted line marking the adjusted p-value threshold of 0.05 (Table S2). Notably, the cancer hallmarks “Tissue Invasion and Metastasis”, “Sustaining Proliferative Signaling”, and “Evading Immune Destruction” exhibit significant enrichment, highlighting the pivotal roles of these genes in key oncogenic processes.

“Tissue Invasion and Metastasis” was identified as a hallmark significantly enriched among these 67 genes, suggesting a possible association with processes involved in cancer cell dissemination and secondary growth. This hallmark is critical in the progression of cancer from localized disease to a systemic condition, often associated with poor prognosis [27]. This hallmark is critical in cancer progression and often correlates with poor prognosis. The enrichment of genes in this category implies that EZH2 inhibition may indirectly influence regulatory networks involved in extracellular matrix remodeling and epithelial-to-mesenchymal transition (EMT), although functional assays are required to confirm their direct role in metastatic potential. The enrichment of genes in this category suggests that eccDNA-associated genes are enriched in metastasis-related pathways, indicating a potential link between EZH2 inhibition and metastatic gene programs, which provides potential novel targets to limit metastasis.

“Sustaining Proliferative Signaling” represents another hallmark significantly enriched in this analysis. This hallmark reflects the ability of cancer cells to bypass normal regulatory mechanisms and sustain chronic cell division. Genes within this category are often associated with oncogenic signaling pathways, such as the PI3K/AKT, MAPK, or MYC pathways, which drive proliferation and survival. The enrichment of these genes implies that eccDNA-driven transcriptional changes may play a role in maintaining proliferative signaling even under therapeutic pressure from EZH2 inhibition. These findings underscore the potential for eccDNA to support key cancer traits, even in the context of targeted therapy, by modulating gene expression at the transcriptional level.

Furthermore, the enrichment of the “Evading Immune Destruction” pathway points to a possible link between these eccDNA-associated genes and immune modulation. This observation aligns with the concept that tumor cells utilize specific mechanisms to evade immune surveillance. “Evading Immune Destruction” likely encompasses genes and pathways employed by cancer cells in various immune evasion strategies. This could include upregulation of immune checkpoint proteins that inhibit T cell activation, secretion of immunosuppressive factors, recruitment of immunosuppressive cell types such as regulatory T cells or myeloid-derived suppressor cells, and downregulation of tumor antigen presentation. By uncovering this enrichment, the analysis highlights that these molecular alterations in cancers are closely linked to the tumor’s ability to circumvent immune surveillance and elimination. Therefore, targeting these immune evasion mechanisms could be a promising therapeutic approach to improve anti-tumor immunity and clinical outcomes for cancer patients. Further investigation into the specific genes and pathways involved could uncover potential vulnerabilities to exploit with immunotherapies or combination treatment strategies.

Beyond these three hallmarks, other categories such as “Genome Instability” were also noted in the analysis, though these were not as significantly enriched. Genome instability, for example, is a hallmark that promotes the accumulation of mutations and chromosomal alterations, providing a substrate for cancer evolution and adaptation. This is consistent with our previous findings showing that Cdh1 regulates mammary tumorigenesis by influencing genomic stability [28]. Meanwhile, evasion of immune destruction reflects cancer cells’ ability to suppress or escape immune surveillance, enabling their survival in a hostile microenvironment. While these categories were not the most significantly enriched, the interplay between eccDNA-driven gene expression and these hallmarks could provide additional insights into the broader impact of eccDNA dynamics on cancer biology. Collectively, this analysis highlights the diverse and critical roles of the 67 genes in shaping cancer progression and response to EZH2 inhibition, offering a rich resource for further functional and therapeutic exploration.

3.5. Validation of eccDNA-Correlated Tazemetostat Response Genes Using Public Dataset

Validation of the 67 eccDNA-correlated Tazemetostat response genes was carried out using the publicly available GSE262877 dataset to examine gene expression changes under various conditions. This dataset was selected for its well-characterized subgroups and comprehensive transcriptomic data, aiming to analyze and identify key differentially expressed genes between Tazemetostat-treated and control subgroups and to explore key molecular features associated with Tazemetostat.

The dataset includes 30 prostate tissue and cell line samples, which are grouped to analyze the effects of Tazemetostat. Specifically, it represents five expression classes: each group contains three untreated samples and another three samples treated with 5 µM Tazemetostat, allowing robust comparisons of drug response. These samples include 18 patient-derived organoid neuroendocrine neoplasm samples (across three groups: ne1262, ne154, and ne155) and 12 prostate adenocarcinoma (PRAD) cell line-based samples (across two groups: prabl and prlnc).

As shown in Figure 4B, the number of differentially expressed genes in our dataset that overlap with the five public datasets (ne1262, ne154, ne155, prabl, and prlnc) varies. For example, there are 123 overlapping genes between our eccDNA-correlated Tazemetostat result and the ne1262 dataset, indicating a significant overlap in the genes affected by Tazemetostat treatment in both datasets. This overlap varies across different datasets, with the highest overlap observed between prabl (676 genes) and prlnc (730 genes). This suggests that the response to Tazemetostat treatment may be more consistent within certain types of samples, such as prostate adenocarcinoma cell lines (prabl and prlnc). Additionally, the overlaps between our RNAseq results and other datasets (e.g., 47 genes with ne154, 40 genes with ne155) validate the differentially expressed genes identified in our eccDNA-correlated Tazemetostat DEG result, supporting the reliability of these findings for further research on Tazemetostat-responsive genes.

A total of 68 genes (Table S3) were collected to prepare a reliable gene list for our eccDNA-associated Tazemetostat response study. The consistency and validation of the differentially expressed genes identified in our eccDNA-correlated Tazemetostat study and two or more treatment groups from GSE262877 provide a strong foundation for further investigation into the effects of Tazemetostat on gene expression and eccDNA dynamics in cancers.

Based on the rigorous cross-validation criteria described in the Methods (detection in ≥3 independent datasets), we identified a robust set of 68 Tazemetostat-responsive genes (Table S3). These genes included both upregulated and downregulated responses, with fold-change directionality reported in Supplementary Table S3. This overlap between our eccDNA-associated Tazemetostat analysis and multiple independent treatment groups from GSE262877 provides strong support for the robustness of these gene signatures. Together, these concordant findings offer a solid foundation for further investigation into how Tazemetostat influences gene expression and eccDNA dynamics in cancer. The 68 genes included consist of both upregulated and downregulated Tazemetostat-responsive types identified across at least three public datasets; directionality is reported in Supplementary Table S3.

3.6. Eleven Key Genes Associated with eccDNA Regulation and Tazemetostat Treatment

Key genes modulated by Tazemetostat treatment were identified by integrating eccDNA mapping with transcriptomic data from TOV-112D cells. By correlating eccDNA profiles with the RNAseq results, we pinpointed 67 genes that had consistently altered eccDNA dynamics in response to EZH2 inhibition. This approach highlighted eccDNA-driven changes in gene expression, ensuring robust and reproducible findings. We validated these differentially expressed genes (DEGs) using the public dataset GSE262877 from the NCBI GEO database [29], which includes various cancer cell lines treated with Tazemetostat. This cross-validation established a high-quality list of 68 Tazemetostat-responsive genes, ensuring consistency across multiple datasets. Ultimately, we narrowed this down to 11 genes in Figure 4C (ABCB1, APCDD1, ARMH4, GABRQ, KIT, LBH, LZTS1, NCAN, NMNAT2, SFRP1, and UST). These 11 overlapping genes represent high-confidence cross-dataset markers, while the remaining 56 eccDNA-regulated genes may reflect novel or TOV-112D-specific mechanisms. They are therefore analyzed separately. The 67-gene eccDNA–RNA set constitutes a primary and novel contribution of this work, representing the integrative eccDNA–transcriptional signature uniquely identified through our multi-omics framework. The smaller 11-gene subset reflects a conservative, cross-validated panel derived solely to highlight genes that show reproducibility across external Tazemetostat datasets. Importantly, this refinement does not reduce the significance of the full 67-gene discovery set, which may contain additional novel mechanisms that are not captured in publicly available datasets.

To identify the potential functional cluster for the 11 genes, we conducted a network analysis. The network visualization in Figure 4D of the 11 shared genes (ABCB1, APCDD1, ARMH4, GABRQ, KIT, LBH, LZTS1, NCAN, NMNAT2, SFRP1, and UST) highlights their interactions and functional relationships. Pink solid lines represent functional connections derived from the STRING database [30], indicating known or predicted functional associations between these genes. In our functional enrichment analysis, we also found that the 11 genes are enriched in 19 co-expression atlases from the Progenitor Cell Biology Consortium (PCBC) [31]. Therefore, we also added green dotted lines to indicate co-expression links based on data from the PCBC. This highlights how these genes are often expressed together in similar biological contexts. Notably, ARMH4 and LBH are isolated, showing no direct relationships with other genes in the network, which may suggest they have unique or less understood roles. The 11 selected genes are primarily involved in chemical synaptic transmission and developmental processes, underscoring their potential importance in neural communication and growth.

The potential prognostic relevance of these 11 genes in cancer was explored through comprehensive mutational and survival analyses across 2683 TCGA pan-cancer samples. As shown in Figure 5A, the oncoprint results reveal that NMNAT2, ABCB1, and GABRQ have the highest alteration frequencies at 11%, 9%, and 8%, respectively. The most common genetic alterations include missense mutations (both putative drivers and those of unknown significance) and amplifications. These findings underscore the significant genetic variability and potential functional impact that these genes have in the context of Tazemetostat treatment and cancer biology.

As shown in Figure 5B, our analysis of the TCGA pan-cancer dataset indicates that Non-Small Cell Lung Cancer (NSCLC) exhibits the highest frequency of genetic alterations, with approximately 60% of samples showing mutations, amplifications, or deep deletions in the analyzed genes. Melanoma also has a high frequency of alterations. Around 50–60% of samples were affected, indicating the crucial role these genes play in melanoma development and progression. Breast cancer displays moderate alteration frequency, with about 40% of samples exhibiting genetic changes, highlighting the potential involvement of these genes in breast cancer biology. These observations suggest that the 11 genes identified in this study are frequently altered in several major cancer types (Table S4), underscoring their importance in cancer research and therapy.

To assess clinical significance, an overall survival analysis was performed (Figure 5C), revealing alterations in the 11 genes in 893 patients (35%) and 915 samples (34%). Alterations were identified in 893 patients (35%) and 915 samples (34%) for the 11 genes. The Log-rank Test p-value of 1.769 × 10⁻³ indicates a statistically significant difference in survival between patients with and without these gene alterations. Specifically, the altered group showed a lower overall probability of survival, with a median overall survival of approximately 36.97 months (95% CI: 29.14–53.91) compared to 56.81 months (95% CI: 46.74) in the unaltered group. These findings suggest that alterations in these genes are associated with differences in survival outcomes among the analyzed pan-cancer samples. Overall, the TCGA pan-cancer analysis revealed significant genetic variability among the 11 key genes, with NMNAT2, ABCB1, and GABRQ showing the highest frequencies of alteration. These genetic alterations, including missense mutations and amplifications, were most prevalent in NSCLC and melanoma, indicating their crucial roles in these cancers. Survival analysis demonstrated that patients with alterations in these genes had a significantly lower overall survival probability, highlighting their potential impact on cancer prognosis and the importance of these genes in therapeutic strategies.

3.7. SFRP1 Correlates with Anti-Fibrotic and Immunomodulatory Effects in the Ovarian Cancer Microenvironment

Single-cell spatial transcriptomic analysis was performed on eight ovarian cancer samples [25] to explore potential immunotherapeutic targets at the intersection of epigenetic regulation, spatial gene expression heterogeneity, and tumor immune response. By intersecting the 11 Tazemetostat-responsive genes with spatially variable genes (SVGs) and highly variable genes (HVGs) identified in these datasets (Figure 6A), SFRP1 emerged as a uniquely responsive gene that not only reacts to EZH2 inhibition but also demonstrates dynamic, context-dependent expression across distinct regions of the tumor microenvironment. Another notable gene, LBH, was identified as an SVG but did not meet the HVG criteria.

Kaplan–Meier survival analyses further support the clinical relevance of SFRP1 (Figure 6B,C), demonstrating a significant association between high expression and improved patient outcomes in the context of anti-PD-1 immunotherapy. Importantly, the co-analysis with LBH (Figure 6B,C), another gene implicated in treatment response, strengthens the notion that epigenetically regulated factors can serve as biomarkers for immunotherapy efficacy.

Lastly, single-cell and spatial transcriptomic analyses revealed that SFRP1 expression is highly variable among different cell types and across tumor regions, particularly within fibroblast-rich zones (Figure 6D–L). Violin plots and spatial maps confirmed that SFRP1 is enriched in specific cellular compartments, highlighting its potential to modulate stromal components of the tumor microenvironment. This spatial heterogeneity suggests that SFRP1 may play context-dependent roles in modulating the tumor niche, particularly within fibrotic regions driven by cancer-associated fibroblasts (CAFs) [32,33]. Since fibrosis is a hallmark of immunosuppressive tumor environments, the reactivation of SFRP1 may help counteract fibrotic remodeling and restore immunological responsiveness. Taken together, these findings suggest that SFRP1 is a dual-function gene capable of mitigating fibrosis and enhancing immune response. Therefore, it is positioned as a compelling candidate for combinatorial therapies involving epigenetic modulation and immune checkpoint blockade.

4. Discussion

The observed alterations in eccDNA dynamics following EZH2 inhibition reveal a previously unrecognized correlation between epigenetic regulation and extrachromosomal DNA biology. Conceptually, while the majority of existing literature characterizes eccDNAs primarily as structural vehicles for oncogene amplification and therapeutic resistance, our findings introduce a paradigm shift by positioning eccDNAs as highly dynamic elements that respond globally to epigenetic targeted therapies. These findings extend our understanding of how targeted epigenetic therapies may be associated with changes in genome architecture beyond their canonical effects on histone modifications. EZH2 overexpression is linked to a higher incidence of metastasis and a poorer prognosis [34]. Our data indicate that Tazemetostat significantly alters eccDNA dynamics in ovarian cancer cells, reflecting global epigenetic deregulation [35]. By integrating Circle-seq, transcriptomics, and spatial single-cell analyses, we identified 67 eccDNA-associated genes that respond to Tazemetostat treatment, with SFRP1 emerging as a spatially variable tumor suppressor consistently restored following EZH2 inhibition.

Previous studies have reported the diverse mechanisms by which eccDNA regulates gene expression. Firstly, eccDNA leads to an enhancement of the gene dosage effect [36]. The presence of eccDNA significantly amplifies the copy number of certain oncogenes, resulting in a marked increase in their expression levels. Secondly, studies have demonstrated that eccDNA interacts with chromosomal DNA more frequently. EccDNA carries potent enhancer signals, which, in turn, significantly impact the regulation of gene expression.

Recent studies have highlighted the complexity of EZH2 inhibitor responses across multiple cancers, including mechanisms involving lineage plasticity, modulation of the tumor microenvironment, and immune sensitivity [37,38,39]. These findings support the translational value of combining EZH2 inhibition with approaches targeting eccDNA- or microenvironment-driven resistance. EZH2 inhibition decreases H3K27me³. This leads to chromatin relaxation, increased DNA accessibility, and susceptibility to replication-associated DNA breaks [40]. eccDNA frequently originates from fragile chromosomal loci [41]. Although we did not perform ChIP-seq, publicly available EZH2 ChIP-seq datasets show enrichment near several eccDNA-associated loci, supporting a plausible link between EZH2-mediated repression, chromatin fragility, and eccDNA generation.

Loss of H3K27me³ upon EZH2 inhibition results in chromatin relaxation and increased accessibility, fostering DNA breakage, rearrangement, and eccDNA generation [40]. This link extends the function of EZH2 beyond transcriptional repression, positioning it as a safeguard against genome instability. Notably, changes in eccDNA abundance are not uniform but show sequence- and chromosome-specific biases. Such patterns may arise from selective chromatin remodeling at GC-rich and regulatory regions, which are particularly prone to eccDNA formation following epigenetic modulation [41].

SFRP1 antagonizes the Wnt/β-catenin signaling pathway, which is frequently dysregulated in ovarian cancer and is known to drive tumor cell proliferation and invasion. Restoration of SFRP1 expression can therefore inhibit these oncogenic processes, limiting tumor progression. Notably, the epigenetic silencing of SFRP1 via promoter hypermethylation has been widely reported in ovarian and other cancers, underscoring the significance of its reactivation; the upregulation of SFRP1 suggests a therapeutic reactivation of anti-tumor mechanisms within the tumor microenvironment [42]. Beyond SFRP1, our integrative analysis identified other candidates such as NRP1, SULF2, TNC, and PALLD. While NRP1 and SULF2 have established roles in microenvironment remodeling and metastasis, their specific regulation via eccDNA dynamics remains a novel observation that warrants further investigation.

However, several limitations of this study must be acknowledged. First, the present work is primarily based on correlative and computational analyses, integrating eccDNA profiling, bulk RNA-seq, and spatial single-cell transcriptomic data. No direct functional validation experiments (e.g., eccDNA perturbation or gene knockdown/overexpression assays) were performed, and therefore, causal relationships between eccDNA dynamics and gene regulation cannot be definitively established. Second, the pan-cancer survival analyses and immunotherapy-related survival associations were exploratory in nature, relying on publicly available datasets with heterogeneous clinical backgrounds. These analyses were not intended to provide predictive clinical conclusions but rather to offer hypothesis-generating insights into the potential relevance of eccDNA-associated genes.

Third, all eccDNA profiling experiments were conducted in a single ovarian cancer cell line (TOV-112D). Cell line–specific effects cannot be excluded, and future studies involving additional cell models and patient-derived samples will be required to assess the generalizability of these findings. Looking forward, our findings suggest potential combinatorial strategies to overcome epigenetic plasticity. Therapeutically targeting the lifecycle of eccDNAs—such as inhibiting DNA repair pathways involved in their genesis (e.g., MMEJ), disrupting chromatin regulators that maintain circular stability, or exploiting the transcriptional dependency of eccDNA templates—could synergize with EZH2 inhibitors. However, we emphasize that these approaches remain speculative. Extensive functional validation in preclinical models is required to determine whether blocking eccDNA dynamics can effectively prevent adaptive resistance to epigenetic therapies.

In summary, our work uncovers a novel regulatory axis wherein EZH2 inhibition reshapes eccDNA architecture, restoring tumor suppressor function and providing a foundation for innovative therapeutic strategies targeting epigenetic-eccDNA crosstalk in ovarian and potentially other cancers [43]. By integrating Circle-seq, transcriptomics, and spatial single-cell analyses, we present the first evidence that targeting EZH2 not only reshapes chromatin landscapes but also modulates eccDNA-mediated oncogenic plasticity. These results pave the way for advanced data integration and mining approaches to harness eccDNA dynamics in ovarian cancer, and potentially, other types of cancer.

5. Conclusions

In this study, we demonstrate that EZH2 inhibition with Tazemetostat is associated with reprogrammed eccDNA dynamics and increased SFRP1 expression in ovarian cancer cells. By integrating eccDNA profiling, transcriptomics, and clinical association analyses, we show that eccDNA-linked gene regulation reflects broader epigenetic reactivation of tumor suppressor pathways. These findings provide a mechanistic framework suggesting that eccDNA may serve as a dynamic indicator of epigenetic therapy response. Future studies will be needed to functionally validate these findings across additional ovarian cancer models and determine the translational relevance of eccDNA remodeling in patient tumors.