MicroRNA-Mediated Rewiring of the PI3K–AKT–mTOR Axis Reveals Divergent Regulatory Networks and Metabolic Vulnerabilities in Endometriosis-Associated Ovarian Cancers (EAOCs)

1. Introduction

Endometriosis-associated ovarian cancers (EAOCs) represent a distinct subgroup of epithelial ovarian malignancies that arise in the context of endometriosis. The term includes ovarian clear cell carcinoma (OCCC) and endometrioid ovarian carcinoma (EnOC) [1,2]. Together, they represent the second most common subtype of ovarian carcinomas after the high-grade serous ovarian cancer (HGSOC) group and account for approximately 10–20% of all epithelial ovarian cancers. The prevalence differs geographically, with the highest incidence observed in East Asian populations, where it can reach 25–30% [3].

Clinically, EAOC patients tend to present at an earlier stage and are generally younger compared to HGSOC patients. The prognosis is overall better; however, advanced-stage and recurrent disease are often chemo-resistant and carry a worse prognosis than HGSOC [4]. Such differences in biological behaviour dictate a subtype-specific approach when investigating the molecular mechanisms underlying the disease.

The mechanistic target of rapamycin (mTOR) is a serine/threonine protein kinase belonging to the phosphatidylinositol 3-kinase–related kinase (PIKK) family. It exists in two functionally distinct complexes, the mechanistic target of rapamycin complex 1 (mTORC1) and the mechanistic target of rapamycin complex 2 (mTORC2) and is regarded as the central regulator of cellular homeostasis, balancing anabolic and catabolic processes in response to external and internal cues [2,5].

Previously, we investigated the mTOR pathway dysregulation in chemotherapy-naive high-grade serous primary ovarian cancers using transcriptomic and microRNA (miRNA) sequencing data from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) consortia projects [5].

To model the mTOR pathway dysregulation in ovarian clear cell (OCCC) and endometrioid ovarian carcinoma (EnOC), we adopted an integrative approach, examining multiple dependent datasets to achieve sufficient sample size. Gene expression meta-analysis approach was adopted, considering that these two histological subtypes are not represented in TCGA ovarian cancer cohort, and given their relatively low prevalence, the integrative approach facilitated robust evaluation of the transcriptional alterations.

2. Materials and Methods

2.1. Dataset Selection

The Gene Expression Omnibus (GEO) repository ( GEO Accession viewer) was searched using the following MeSH terms and free-text keywords: ((((“ovarian neoplasms”[All Fields] OR “ovarian cancer”[All Fields] OR “ovarian carcinoma”[All Fields]))) OR “ovarian neoplasms”[MeSH Terms]) AND ((“ovarian clear cell carcinoma”[All Fields] OR “clear cell ovarian carcinoma”[All Fields] OR “OCCC”[All Fields] OR “endometrioid carcinoma”[All Fields] OR “endometrioid adenocarcinoma”[All Fields] OR “ENOC”[All Fields] OR “endometriosis associated ovarian carcinoma”[All Fields] OR “EAOC”[All Fields])). We then applied the filters “Homo sapiens” and “Expression profiling by high-throughput sequencing.” To identify mRNA-containing data sets.

The initial search yielded 48 GEO datasets. Each dataset was manually screened, and selection was limited to datasets that included ovarian clear cell and/or endometrioid ovarian adenocarcinoma samples from patients. Cell lines, xenografted models, single-cell RNA sequencing, and datasets based on blood-, serum-, or urine-derived biomarkers (e.g., platelet RNA profiles) were also excluded. Additionally, datasets where the EAOC that were not clearly stratified into endometrioid or clear cell were also excluded (Supplementary Table 1).

Each dataset was further examined to include only samples with clear annotations as ovarian clear cell or endometrioid adenocarcinoma across all stages and grades. Where neoadjuvant chemotherapy, interval debulking surgery, refractory or resistant were explicitly mentioned, or sample-level data could not be obtained with certainty, datasets were excluded. Datasets with a single sample or mixed histology were also removed. The cohort included six datasets. Following batch correction and data normalisation steps, a further dataset was excluded, and a total of five datasets were included in the final analysis (Table 1). (The sample-level data are included in the supplementary Table 2).

For normal ovarian tissues, transcript read counts generated using the RSEM v1.3.3 pipeline were downloaded from the Genotype-Tissue Expression (GTEx) (GTEx Portal) project, and a total of 150 normal ovarian tissue samples were included (Sample IDs are in Supplementary Table 3).

For miRNA data, the Gene Expression Omnibus (GEO) repository (GEO Accession viewer) was queried using the following free-text keywords: (ovary) AND cancer. We then applied the filters “Homo sapiens” and “Non-coding RNA profiling by high-throughput sequencing.” The initial search identified 43 datasets. The same inclusion and exclusion criteria used for the mRNA datasets were applied to the miRNA dataset collection. Normal ovarian tissue samples from the GEO datasets were used as controls. After screening, a total of four datasets were included in the final cohort (see Supplementary Table 4 for details). These datasets comprised high-throughput, open-access miRNA expression profiles for ovarian clear cell carcinoma; however, no comparable datasets were available for EnOC. We proceeded to search ArrayExpress using the keywords “ovarian adenocarcinoma,” “ovarian cancer,” “ovarian clear cell adenocarcinoma,” and “endometrioid adenocarcinoma of the ovary”. The following filters were applied: Homo sapiens, high-throughput sequencing, RNA-seq of coding RNA, RNA-seq of non-coding RNA, and technology: sequencing array. The search was further limited to studies in which raw datasets were available. After applying these criteria, no additional studies were identified that met the predefined inclusion and exclusion criteria. We then searched the European Genome-Phenome Archive using the keyword “miRNA”; a total of 156 studies were screened, and access was requested and granted for the dataset EGAD00001009636, which contains miRNA data of 82 endometroid ovarian cancer samples and 30 clear cell ovarian cancer samples [6].

Table 2. Characteristics of the miRNA data sets included in the analysis.

Accession No.	Platform	Sample type	Total number of samples	No. of selected samples
				OCCC	EnOC	Normal
GSE200852	Illumina MiSeq	Ovary	52	20	-----	10
GSE230956	Illumina NextSeq 500	Ovary	16	4	------	--------
GSE134289	HiSeq X Ten	Ovary	6	-------	-------	3
GSE261800	Ion Torrent Proton	Ovary	40	-------	-------	20
EGAD00001009636	llumina NextSeq 500			30	82	----
Total Selected				54	82	33

Table 2. Characteristics of the miRNA data sets included in the analysis.

Accession No.	Platform	Sample type	Total number of samples	No. of selected samples
				OCCC	EnOC	Normal
GSE200852	Illumina MiSeq	Ovary	52	20	-----	10
GSE230956	Illumina NextSeq 500	Ovary	16	4	------	--------
GSE134289	HiSeq X Ten	Ovary	6	-------	-------	3
GSE261800	Ion Torrent Proton	Ovary	40	-------	-------	20
EGAD00001009636	llumina NextSeq 500			30	82	----
Total Selected				54	82	33

2.2. Data Processing

mRNA data was downloaded using prefetch function of sratoolkit v3.0.7 , then we used fasterq-dump (2.11.3) to extract the fastq files. Raw Data of all datasets were quality assessed using fastqc (v 0.11.9); two datasets were identified to contain poor-quality bases and adaptor sequences, which were trimmed using fastp (v 0.24.0). The fastp parameters used for the dataset GSE160692 were (-i -I -o -O -w -h -j -- low_complexity_filter – complexity threshold 30). In addition, fastp parameters were used for the dataset GSE 295399 ( --cut_front --cut_front_mean_quality 20 --cut_front_window_size 4 --cut_tail --cut_tail_mean_quality 20 --cut_tail_window_size 4 --qualified_quality_phred 20 --unqualified_percent_limit 40 --trim_front1 3 --trim_front2 3).

We used the human reference genome GRCh38 (GENCODE release v39) to generate the genome index for read alignment using STAR (v2.7.10a). Gene annotation from GENCODE v39 was used as the reference annotation. Reads were aligned to the reference genome using STAR (v2.7.10a), and gene-level expected counts were quantified using RSEM (v1.3.3), consistent with the GTEx RNA-seq processing pipeline [7,8]. Samples with a total alignment percentage <60% were excluded, and the total number of samples included at this point was 135 cancer samples.

Gene-level read counts from RSEM v.1.3.3 were imported into R (v 4.4.1) using the tximport package (v1.34.0). The GTEx Transcript read counts from RSEMv.1.3.3. (v 10), of 150 normal ovarian samples were downloaded from GTEx portal and collapsed to gene level reads.

For miRNA datasets, raw sequencing data were downloaded using prefetch function of sratoolkit v3.0.7, then we used fasterq-dump (2.11.3) to extract the fastq files. For the EGA dataset, we downloaded the raw sequence data as fastq files using the pyega3 client downloader (v5.2.0). Raw sequencing reads were assessed for quality using fastqc (v 0.11.9). Adaptor sequences were removed using fastp (v 0.24.0) for the GEO datasets and cutadapt (v 1.18) for the EGA dataset. The human reference genome GRCh38 (GENCODE release 49) was used, and a genome index was generated using Bowtie2 (v2.4.2). Small RNA sequencing reads in fastq format were aligned to the GRCh38 reference genome using Bowtie, applying the following parameters (-n 2 -l 15 -k 1). Alignment outputs were converted to sorted BAM files using samtools (v1.12). miRNA read quantification was performed using featureCounts of subread (v2.0.2) with the miRBase (release 22.1) GFF3 annotation file (hsa. gff3). The resulting count tables were subsequently combined into a miRNA count matrix for downstream analysis (Supplementary Table 5).

2.3. Data Integration and Normalisation

Low-abundance transcripts were filtered by retaining only those with a minimum of 5 read counts in at least 10% of samples. Batch correction was performed using ComBat-seq of the sva package of R (3.54.0). For the GEO cancer cohort, the study identifier was used as the batch variable. For the GTEx dataset, the sequencing centre identifier was used as the batch variable. Lowly expressed genes were filtered. The two matrices were then merged by Ensembl gene identifiers and log-transformed. Quantile normalisation was performed on the combined expression matrix using the normalise.quantiles function from the preprocessCore R package (v 1.68.0) to ensure a uniform distribution across samples and reduce the impact of heteroscedasticity [9]. Following normalisation, diagnostic checks were performed to confirm the comparability across the GTEx and GEO datasets (for more details, please see Supplementary Material 6).

2.4. Differential Expression Analysis

Differential Gene expression analysis between normal and EAOCs samples combined was conducted using the LIMMA package (v3.62.2) in Bioconductor (https://www.bioconductor.org/). Significantly dysregulated genes were defined as having an adjusted p-value < 0.05 and a log2FC (fold change) >1 or <−1 [10]. Differential expression analysis of miRNA counts in EAOCs samples versus normal ovarian tissues was performed in R using the DESeq2 package (v1.46.0) [11]. Statistical significance was defined using a threshold of adjusted p-value < 0.05 and an absolute log₂ fold change > 1. Variance stabilizing transformation was applied to count data using DESeq2, and principal component analysis was performed to assess sample clustering based on condition (Cancer versus Normal) and sample type (Normal, EnOC, and OCCC).

2.5. Comparative Transcriptomic Analysis Between Endometriosis-Associated Ovarian Cancer Subtypes

To assess shared and subtype-specific transcriptional changes between endometrioid ovarian carcinoma and ovarian clear cell carcinoma, overlap analysis was performed using the sets of significantly differentially expressed genes from each group. Venn diagrams were generated to visualise the number of unique and overlapping genes using ggVennDiagram (1.5.7) Package of R. Gene variance across endometrioid ovarian carcinoma and ovarian clear cell carcinoma samples was calculated, and the top 100 most variable genes were selected, mapped to gene symbols, and visualised across normal, endometrioid, and clear cell samples using heatmaps generated with the pheatmap (v1.0.13) package in R.

2.6. Uniform Manifold Approximation and Projection Analysis (UMAP)

Cancer samples were subsetted from the combined expression matrix. Gene variance was calculated across cancer samples, and the 2,000 most variable genes were selected for dimensionality reduction. Uniform Manifold Approximation and Projection was then performed in R using the umap package (v 0.2.10.0) and visualised using plotly (v 4.12.0) package for three-dimensional plots.

2.7. Functional Enrichment Analysis

Gene annotation was performed using the biomaRt package (2.62.1) in R, querying the Ensembl database (release 115). Ensembl gene identifiers were mapped to their corresponding gene symbols using the hsapiens_gene_ensembl dataset. Functional enrichment analysis was performed using the gprofiler2 (v 0.2.4) package in R. The top enriched gene ontology and Kyoto Encyclopedia of Genes and Genomes pathways terms were visualised using dot plots generated with ggplot2 (v 4.0.2).

2.8. Functional mTOR Pathway Visualisation

The KEGGREST (v1.46.0) R package was used to retrieve the list of genes associated with the mTOR signalling pathway (hsa04150) from the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (release 117.0) [12,13]. mTOR DEGs were identified and mapped onto the KEGG mTOR signalling pathway using the Pathview (v1.46.0) R package [14].

2.9. Prediction and Integration of miRNA–mRNA Regulatory Interactions

Predicted gene targets of differentially expressed miRNAs (DEMs) were obtained using four prediction databases, TargetScan, miRDB, miRanda, and DIANA-microT through multiMiR (v 1.28.0) package in R [15]. Predicted targets were intersected with the mTOR pathway (DEGs) in EAOC, and only those experimentally validated were retained. The identified pairs were intersected with the miRNA from KEGG MicroRNAs in Cancer pathway (hsa05206) to identify cancer-relevant regulatory interactions.

2.10. Network Construction and Topological Analysis of mTOR Pathway miRNA–mRNA Interactions

Network analysis of miRNA–mRNA interactions was performed using Cytoscape software (version 3.10.3), developed by the Cytoscape Consortium (Cytoscape Consortium, San Diego, CA, USA) [16]. Hub genes and miRNAs were identified using degree centrality, defined as the number of regulatory edges incident to a node. Nodes with high degree centrality were considered candidate hubs.

3. Results

3.1. Transcriptomic Dysregulation of the mTOR Pathway in Endometriosis-Associated Ovarian Cancers (EAOC) and Ovarian Clear Cell

We initially evaluated six GEO datasets comprising OCCC and EnOC samples. Following batch correction and normalisation, GSE160692 was excluded due to integration incompatibility and so is one sample from GTEx. The final downstream analysis included 273 samples: 149 normal ovary samples (GTEx), 72 ovarian clear cell adenocarcinoma (OCCC) samples, and 52 endometrioid ovarian adenocarcinoma (EnOC) samples (GEO). After filtering low-abundance genes, 30,766 genes were expressed in the OCCC and EnOC samples. Differential gene expression analysis using the limma package in R identified 15,326 genes as significantly differentially expressed between endometriosis-associated ovarian cancer (EAOC) and normal ovarian tissue (adjusted p-value < 0.05 and absolute log2 fold change > 1) (Figure 1A).

Transcriptional similarity between endometrioid ovarian carcinoma (EnOC) and ovarian clear cell carcinoma (OCCC) was depicted in (Figure 1B), the sets of significantly differentially expressed genes relative to normal ovarian tissue were compared. A substantial proportion of genes (65%) were shared between the two subtypes, indicating a common transcriptional programme consistent with their classification as endometriosis-associated ovarian cancers (Figure 1B). However, a subset of genes were uniquely dysregulated in each subtype, highlighting potential subtype-specific biological processes.

To further investigate these subtype-specific differences, a direct differential expression comparison between EnOC and OCCC was performed. The top 100 differentially expressed genes were selected and visualised in a heatmap across normal, EnOC, and OCCC samples (Figure 1C). Despite the large overlap observed in the Venn analysis, the heatmap revealed clear separation between EnOC and OCCC based on these genes, with distinct expression patterns characterising each subtype.

The Uniform Manifold Approximation and Projection (UMAP) analysis further emphasised that EnOC and OCCC form distinct transcriptional clusters (Figure 2). (Supplementary Table 6 -7).

Subsequently, functional enrichment analysis was conducted for each subtype independently to identify significantly enriched biological pathways. Among the top enriched biological processes in ovarian clear cell carcinoma were regulation of cell proliferation, cellular response to stimulus, and signal transduction. At the molecular function level, protein binding, ion binding, and kinase-related activities were prominently enriched. Among the enriched Kyoto Encyclopedia of Genes and Genomes pathways were pathways in cancer, phosphoinositide 3-kinase–protein kinase B signalling, and metabolic pathways (Figure 3).

Among the enriched biological processes in endometrioid ovarian carcinoma were regulation of cell communication, cellular response to stimulus, and developmental processes. At the molecular function level, protein binding, ion binding, and transporter activity were prominently enriched. Among the enriched Kyoto Encyclopedia of Genes and Genomes pathways were cAMP signalling pathway and MAPK signalling pathway. (Figure 4).

To investigate mTOR pathway dysregulation in each subtype, we overlaid the mTOR DEGs in OCCC and EnOC onto the KEGG mTOR pathway (hsa04150) (Figure 5, and Figure 6).

The canonical mTOR signalling pathway in both subtypes showed upregulation of PIK3CB (encoding the p110β catalytic subunit of class IA PI3K) and MTOR, indicating transcriptional activation of the canonical PI3K/AKT/mTOR axis in EAOCs [17,18]. Contrary to our previous findings of RICTOR downregulation in high-grade serous ovarian cancer, there was no transcriptional downregulation of RICTOR detected in EAOCs [19,20]. The observed SLC7A5 upregulation also suggested enhanced amino acid uptake, highlighting possible pathway rewiring and the need for functional validation. We also observed a statistically significant difference in Raptor expression between EnOC and OCCC, although it did not meet the log2 fold change threshold of> 1 in either group (Figure 7).

3.2. The Regulatory miRNA Network of mTOR Pathway in Endometriosis-Associated Ovarian Cancers (EAOC).

miRNA datasets included four GEO datasets (GSE200852, GSE230956, GSE134289, and GSE261800) and The European Genome-phenome Archive dataset (EGAD00001009636). A total of 82 EnOC samples, 33 normal ovarian tissues, and 54 OCCC samples were included. Low-abundance miRNAs were subsequently removed, and a total of 650 miRNAs were included in downstream analysis. (Supplementary Table 8). The principal component analysis (PCA) demonstrated distinct clustering of EnOC, OCCC, and normal tissue samples. A subset of OCCC samples exhibited a transcriptional profile similar to EnOC. This observation may reflect heterogeneity within OCCC [21] (Figure 8).

Given the overlap in miRNA expression between EnOC and OCCC and the heterogeneity within the OCCC group. We explored the regulation of the mTOR pathway collectively in EAOCs rather than subtype specific.

Differential expression analysis identified 188 miRNAs (DEMs) that were significantly differentially expressed between EAOCs and normal ovarian tissue samples (adjusted p-value < 0.05 and log2FC > 1). Gene targets of significantly differentially expressed microRNAs were predicted and integrated with the mechanistic target of rapamycin pathway DEGs. Predicted interactions were retained if they appeared in at least two databases and were experimentally validated in the Multimir package database. Pairs with inverse regulatory directions (e.g., miRNA upregulated and mRNA downregulated) were considered biologically relevant. A total of 66 regulatory miRNA- mRNA pairs were identified, comprising 28 mTOR DEGs and 39 DEMs in EAOC and network of interactions was constructed (Figure 9).

To identify hub nodes within the network, degree centrality was computed, and scores of 3 or higher were used to define hub nodes. Among the miRNAs, hsa-miR-30a-5p, hsa-miR-30d-5p, hsa-miR-26a-5p, and hsa-miR-7-5p had the highest degree centrality. Among the mTOR pathway genes, mTOR gene exhibited the highest degree centrality. indicating its role as a central hub within the regulatory network. Other highly connected genes included neuroblastoma RAS viral oncogene homolog (NRAS), glycogen synthase kinase 3 beta (GSK3B), insulin receptor substrate 1 (IRS1), folliculin interacting protein 2 (FNIP2), frizzled class receptor 5 (FZD5), protein kinase C beta (PRKCB), DNA damage inducible transcript 4 (DDIT4), phosphoinositide-3-kinase catalytic subunit delta (PIK3CD), and AKT serine/threonine kinase 3 (AKT3) (Table 3).

4. Discussion

PI3K/AKT/mTOR dysregulation is frequently reported in endometriosis-associated ovarian cancers and is often attributed to genomic aberrations, particularly involving PIK3CA, PTEN, and ARID1A genes [22,23]. Our integrative in silico analysis provides a transcriptomic and microRNA-based characterisation of the mechanistic target of rapamycin signaling in endometriosis-associated ovarian cancers (EAOCs), highlighting shared and subtype-specific regulatory features between its two subtypes.

We demonstrated distinct transcriptional signatures of ovarian clear cell (OCC) and endometrioid ovarian adenocarcinoma in line with the previously published literature [24,25,26]. Despite the global differences, we observed upregulation of PIK3CB, mTOR and the amino acid transporter SLC7A5 genes in both OCCC and EnOC subtypes, coupled with preservation of RICTOR, a core mTORC2 component. Collectively, these findings suggest rewiring of the amino acids sensing mechanisms that feed into mTORC1 activation and conservation of mTORC2 activity in EAOCs [20,27]. These findings are consistent with previous in vitro studies using ovarian cancer cell line MDAH-2774 [28,29]

However, our functional enrichment analysis showed PI3K-Akt dominance in OCCC and MAPK/cAMP signalling in EnOC, suggesting that while the core mTOR machinery is conserved, the upstream parallel pathways driving its activation remain uniquely subtype-specific [30,31].

MicroRNAs are well-known post-transcriptional regulators of the PI3K/AKT/mTOR signalling network that have the ability to simultaneously regulate different components of the pathway to ensure permanent activation of the oncogenic signalling cascade [32,33] Although substantial overlap in the miRNA expression was observed between OCCC and EnOC, the PCA analysis revealed heterogeneity within the OCCC cohort. This observation is consistent with previous reports describing molecular heterogeneity within OCCC and may reflect biologically distinct subgroups with differing regulatory mechanisms and clinical behaviour [34,35,36]. However, investigation of these potential OCCC subtypes was beyond the scope of the present study; therefore, miRNA dysregulation was examined collectively within endometriosis-associated ovarian cancers (EAOCs).

The miRNA regulatory network analysis in our EAOCs cohort revealed the upregulation of several key hubs, notably hsa-miR-30a-5p, hsa-miR-30d-5p, hsa-miR-7-5p, and hsa-miR-26a-5p. Despite variability across profiling studies, miR-30a dysregulation is consistently reported and may be linked to oxidative stress, a known contributor to the malignant transformation of endometriosis. By identifying DDIT4 and PIK3CD as target genes of hsa-miR-30a-5p, we highlight a mechanistic link between oxidative stress and PI3K/AKT/mTOR activation [37,38].

We further demonstrated upregulation of hsa-miR-30d-5p in EAOC relative to normal ovarian tissue, in agreement with previous findings by Lee et al. Interestingly, their study showed that elevated expression of miR-181d, miR-30c, miR-30d, and miR-30e-3p was associated with significantly improved disease-free and overall survival outcomes [39]

We also observed upregulation of hsa-miR-7-5p in the EAOC cohort. Although hsa-miR-7-5p has not been extensively characterised in ovarian clear cell or endometrioid subtypes, previous studies have reported increased miR-7 expression in high-grade ovarian cancer. Notably, miR-7 has been suggested to function either as a tumour suppressor or an oncogene, depending on the cancer type and experimental context [40,41,42,43].

Beyond mTOR-specific targets, our network identified major highly connected genes, including GSK3B (Wnt signalling) and NRAS (MAPK signalling). This highlights how these hub miRNAs do not act in isolation but rather orchestrate critical cross-talk between multiple oncogenic pathways to drive an aggressive tumor phenotype [44] .

The concurrent downregulation of IRS1, GRB10, DDIT4, and PIK3CD under the same miRNA regulatory context hsa-miR-30a-5p, hsa-miR-30d-5p, and hsa-miR-7-5p suggests coordinated transcriptional reprogramming of the PI3K/AKT/mTOR network. As these genes include both upstream activators (IRS1, PIK3CD) and regulatory inhibitors or feedback components (GRB10, DDIT4), this pattern is more consistent with altered pathway control than with simple activation or suppression. In particular, downregulation of DDIT4 may relieve inhibitory pressure on mTORC1, whereas reduced IRS1 and PIK3CD expression may reflect diminished reliance on receptor-mediated signalling. Collectively, these findings suggest rewiring of the PI3K/AKT/mTOR axis with potentially increased intrinsic pathway activity in these histotypes [45,46,47].

Among the miRNAs identified in our cohort as targeting the mTOR gene, miR-199a has been previously shown to directly regulate mTOR expression and reverse cisplatin resistance in ovarian cancer and is further implicated in the KEGG microRNA epithelial ovarian cancer pathway [48].

A key strength of this study lies in the integrative analysis of transcriptomic and miRNA datasets, enabling the identification of coordinated regulatory networks and providing novel insight into mTOR pathway modulation in EAOC. By combining gene expression and miRNA profiling, this approach offers a more comprehensive systems-level understanding of pathway dysregulation . However, several limitations should be acknowledged. The findings are derived from in silico analyses and therefore require experimental validation at the protein and functional levels, particularly to confirm pathway activity and miRNA–target interactions. Additionally, the absence of phosphorylation data limits direct inference regarding mTOR complex activation, and the relatively limited availability of EAOC-specific datasets may affect the generalisability of the results.

5. Conclusions

This study provides an in silico integrated transcriptomic and post-transcriptional overview of the mTOR signalling pathway in EAOCs. We identified hsa-miR-30a-5p, hsa-miR-30d-5p, hsa-miR-7-5p, and hsa-miR-26a-5p as central hub nodes within the mTOR signalling network in this gynaecological malignancy. The concurrent downregulation of IRS1, GRB10, DDIT4, and PIK3CD by these hub miRNAs suggests coordinated transcriptional reprogramming of the PI3K/AKT/mTOR network. Together with the observed miRNA co-regulation of Wnt, MAPK and mTOR simultaneously suggest epigenetic control of oncogenic pathways cross talk in EAOCs. This work provides a framework for future studies aimed at functional validation and may inform the development of targeted therapeutic strategies that account for both transcriptional and post-transcriptional regulatory mechanisms in these histotypes.