CDS2 Drives Tumor Progression in Epithelial Ovarian Cancer via the PI3K/AKT Pathway

ABSTRACT

Background

Epithelial ovarian cancer (EOC) frequently disseminates within the lipid‐rich peritoneal cavity, where tumor progression is accompanied by metabolic adaptations involving lipid utilization and mitochondrial function. Although these metabolic alterations are increasingly recognized as important features of EOC progression and peritoneal metastasis, the molecular factors linking mitochondrial–lipid metabolic remodeling with aggressive tumor behavior remain incompletely characterized. We therefore integrated mitochondrial‐ and lipid metabolism‐related transcriptomic analyses to identify candidate regulators and further investigated the clinical and biological features associated with CDS2.

Results

Seven key genes associated with mitochondrial and lipid metabolism were identified in EOC through comprehensive bioinformatic analyses and were used to construct the MALRS prognostic model. Further integration of FIGO stage, histologic grade, and MALRS scores yielded a nomogram whose calibration plots showed good agreement between predicted and observed 1‐, 3‐, and 5‐year survival probabilities. Subsequent screening of these candidate genes in chemotherapy‐resistant datasets identified CDP‐diacylglycerol synthase 2 (CDS2) as the most consistently upregulated candidate gene. Clinical validation confirmed that CDS2 was significantly overexpressed in tumor tissues and was associated with advanced disease stage, and survival analysis showed that higher CDS2 expression was associated with shorter progression‐free survival (PFS) in stage III/IV patients. Functional characterization demonstrated that CDS2 overexpression promoted both lipid droplet accumulation and migratory capacity in ovarian cancer cells. Pathway analysis identified significant enrichment of PI3K/AKT signaling, and experimental validation confirmed CDS2‐mediated hyperactivation of PI3K/AKT phosphorylation. Computational immune‐deconvolution analyses further identified exploratory associations between CDS2 expression, estimated T‐cell abundance, and immune‐checkpoint transcripts.

Conclusion

CDS2 is a lipid metabolism‐associated candidate biomarker linked to aggressive clinicopathological features in EOC. Our findings associate CDS2 with disease progression, PI3K/AKT‐related signaling, and immune‐related transcriptomic features, supporting its further evaluation as a candidate prognostic marker.

Abbreviations

ATCC

American Type Culture Collection

CDP‐DAG

cytidine diphosphate diacylglycerol

CDS2

CDP‐diacylglycerol synthase 2

CPT

Carnitine palmitoyltransferases

CTLA‐4

Cytotoxic T‐lymphocyte‐associated protein 4

DEGs

differentially expressed genes

EGFR

epidermal growth factor receptor

EOC

epithelial ovarian cancer

FABP4

fatty acid‐binding protein

GEO

gene expression omnibus

GO

Gene Ontology

GSEA

gene set enrichment analysis

GSVA

Gene set variation analysis

ICB

immune checkpoint blockade

IFN‐γ

interferon‐gamma

IHC

Immunohistochemistry

KEGG

Kyoto Encyclopedia of Genomes

LPA

lysophosphatidic acid

LPAR

Lysophosphatidic acid receptors

OS

overall survival

PA

phosphatidic acid

PCA

principal component analysis

PD‐1

programmed cell death protein 1

PFS

progression‐free survival

PG

phosphatidylglycerol

PI

phosphatidylinositol

PIP2

phosphatidylinositol‐4,5‐bisphosphate

PIP3

phosphatidylinositol‐3,4,5‐trisphosphate

PPI

protein–protein interaction

RTK

receptor tyrosine kinase

SREBF

Sterol regulatory element‐binding proteins

TCGA

the Cancer Genome Atlas

TMAs

tissue microarrays

TME

tumor microenvironment

TOM

topological overlap matrix

WGCNA

Weighted gene coexpression network analysis

1. Introduction

EOC, the most lethal gynecologic malignancy, is projected to experience a 55% [1] increase in incidence by 2035. Despite a 50.8% 5‐year overall survival rate, > 70% of patients present with advanced‐stage (III/IV) disease upon diagnosis characterized by extensive peritoneal dissemination [1, 2]. Although 80% of newly diagnosed patients initially respond to standard therapy, two‐thirds develop chemoresistance, culminating in relapse and metastatic progression [3]. Early peritoneal spread, diagnostic challenges, and high recurrence rates collectively render EOC a disease with exceptionally grim prognosis and dismal survival outcomes. This highlights the necessity of dissecting the mechanisms associated with metastasis and chemoresistance in ovarian cancer.

During disease progression, approximately 80% of epithelial ovarian carcinomas develop extensive abdominal metastases, with omental involvement representing the most frequent metastatic pattern [4]. Successful peritoneal metastasis, which is a major driver of ovarian cancer recurrence and mortality, depends on coordinated metabolic reprogramming in both neoplastic cells and associated stromal components within the tumor microenvironment [5, 6, 7]. In the event of nutrient deficiency, fatty acids are released from lipid droplets by lipolysis and metabolized by β‐oxidation in peroxisomes and mitochondria [8, 9]. This process provides the fuel for the tricarboxylic acid cycle and the respiratory chain complexes used for oxidative phosphorylation [10]. Mitochondria play critical roles in cellular energy metabolism, oxidative stress, and signaling and are closely connected with nearly all organelles within the cell [11]. The metabolic reprogramming of both tumor and stromal cells in the microenvironment modulates fatty acid transport and mitochondrial energy metabolism, ultimately impacting cellular energy generation and metabolic homeostasis.

Hyperproliferative cancer cells exhibit dysregulated lipid biosynthetic pathways that sustain membrane production and bioactive lipid synthesis [12]. The PI3K/AKT signaling cascade represents one of the most fundamental intracellular regulatory networks governing essential cellular processes [13, 14]. Accumulating evidence has demonstrated that the PI3K/AKT signaling axis promotes lipid biosynthesis, facilitates fatty acid uptake, and regulates metabolic enzymes in the fatty acid synthesis pathway through distinct molecular mechanisms [15, 16]. Current research has demonstrated that dysregulated PI3K/AKT signaling, despite its complex involvement in cancer progression, ultimately drives malignant transformation, metastatic competence, and acquired resistance to conventional chemoradiotherapies [17, 18]. These findings underscore its potential as a therapeutic target for metastatic cancer.

The PI3K/AKT signaling pathway is critically regulated by the relative abundance of phosphatidylinositol‐4,5‐bisphosphate (PIP2) and phosphatidylinositol‐3,4,5‐trisphosphate (PIP3). PI3K activation drives phosphoinositide metabolism, catalyzing the conversion of PIP2 to PIP3, which facilitates AKT membrane recruitment and subsequent phosphorylation [19]. The CDS2 gene is located on chromosome 20p12.3 and encodes CDP‐diacylglycerol synthase 2, which plays a central role in phospholipid metabolism. CDS2 orchestrates membrane phospholipid homeostasis by catalyzing the conversion of phosphatidic acid (PA) to CDP‐DAG, thereby governing the biosynthesis of key phospholipids, including phosphatidylglycerol (PG), phosphatidylinositol (PI), and their derivatives PIP2 and PIP3 [20, 21]. Moreover, CDS2 deficiency leads to reduced levels of PI and its metabolite PIP3, resulting in aberrant lipid accumulation and diminished organelle volume [22]. Beyond its canonical role in phospholipid homeostasis, emerging studies have implicated CDS2 in cancer‐associated signaling and metabolic vulnerabilities, including context‐dependent CDS2 dependency linked to CDS1 expression. Therefore, the contribution of the present study is not the initial identification of CDS2 as a cancer‐related molecule, but rather its integrated characterization within an EOC‐focused mitochondrial–lipid metabolic framework. Specifically, we examined whether CDS2 expression is associated with clinicopathological progression, intracellular lipid droplet accumulation, migratory behavior, PI3K/AKT‐associated signaling, and immune‐related features in EOC.

Despite growing evidence that lipid metabolic adaptation and mitochondrial function contribute to EOC progression and peritoneal dissemination, the molecular factors connecting these metabolic programs with aggressive tumor phenotypes remain incompletely defined. We therefore adopted a biologically informed screening strategy that integrated stage‐associated mitochondrial genes, differentially expressed genes, and lipid metabolism‐related genes to prioritize candidate molecules associated with EOC progression. Through this framework, CDS2 emerged as a candidate of interest. We subsequently evaluated its clinicopathological associations and examined its relationships with intracellular lipid droplet accumulation, migratory behavior, PI3K/AKT‐associated signaling, and immune‐related features.

2. Results

2.1. Screening of Prognosis‐Related Differentially Expressed Genes in Ovarian Cancer

We downloaded RNA‐seq data for 429 TCGA ovarian carcinoma samples, which predominantly represent high‐grade serous ovarian carcinoma, and extracted the expression matrices of 1925 mitochondria‐related genes. WGCNA identified six modules (Figures 1A and S1A); the blue and green modules showed the strongest associations with pathological stage, and their 1448 genes were retained. For cross‐dataset candidate screening, differential expression was analyzed separately in GSE10971 and GSE27651 using p < 0.05 and |log2FC| ≥ 0.7 (Figure 1C,D). GSE10971 compares non‐malignant fallopian‐tube epithelium with high‐grade tubal/ovarian serous carcinoma, whereas GSE27651 includes ovarian surface epithelium, serous borderline tumors, low‐grade serous carcinoma, and high‐grade serous carcinoma. The datasets were not treated as a single uniform disease cohort; intersection of their differential‐expression results yielded 1194 genes for exploratory candidate prioritization. Epithelial ovarian cancer is markedly lipophilic, and there is substantial evidence that its lipophilic nature promotes its development and metastasis [23, 24, 25]. Therefore, 743 lipid metabolism‐related genes retrieved from the GeneCards database were used for candidate screening. Through Venn diagram analysis, we identified 7 candidate prognostic genes, including ACOT7, ALAS1, CDS2, NUDT19, PPP1CC, SLC44A1, and CYP2E1 (Figure 1E).

FIGURE 1.

Screening and identification of core regulatory genes involved in ovarian cancer. (A) Clustering dendrogram of genes, with dissimilarity based on topological overlap, together with assigned module colors. (B) Module–trait associations: Each row corresponds to a module eigengene, and each column corresponds to a trait. Each cell contains the corresponding correlation and p value. The table is color‐coded by correlation according to the color legend. Distribution of average gene significance and errors in the modules associated with the sex and ethnicity of cancer patients. Screening for differential genes between the ovarian cancer GSE10971 (C) dataset and the GSE27651 (D) dataset. (E) Venn diagram showing overlapping genes among the lipid metabolism‐related genes, DEGs and genes identified via WGCNA. (F) Prognostic models were developed by analyzing 7 DEGs in the TCGA ovarian cancer dataset via 62 machine learning algorithms. (G) Fusion of staging and grading to construct a more powerful nomogram model. (H) Expression of the seven DEGs in the three platinum resistance datasets. The blue color indicates downregulated expression in the resistant group, the red color indicates upregulated expression in the resistant group, the color shading indicates the degree, and the gray color indicates no statistical significance.

2.2. Construction and Evaluation of Prognostic Models for Ovarian Cancer

The 429 EOC patients in the TCGA database were randomly divided at an approximately 1:1:1:1 ratio into one training cohort (n = 108) and three nonoverlapping internal validation cohorts (n = 107 each). The candidate algorithm set was prespecified, and the seven genes screened above were applied to 62 machine‐learning algorithms for survival modeling. Model fitting and parameter estimation were performed in the training cohort, whereas the three held‐out cohorts were used to compare out‐of‐sample C‐indices (Figure 1F). This fixed‐seed, multi‐holdout internal validation enabled comparison of model performance across three nonoverlapping held‐out subsets within one reproducible partition. Four models—RSF + ridge, RSF + Enet (α = 0), Enet (α = 0), and Ridge—showed identical displayed C‐index values across the three validation cohorts (0.54, 0.53, and 0.65; mean 0.574). RSF + ridge was retained as a representative of these tied top‐performing models and was designated MALRS; the current results do not establish its superiority over the other tied models. The random seed was set to 123,456. The three held‐out cohorts provided separate internal evaluations of the fixed seven‐gene model‐development workflow. Patients in the four cohorts were categorized into high‐ and low‐risk groups on the basis of the median MALRS score (Figure S1B). We subsequently performed exploratory analyses to characterize the resulting risk groups. According to the PCA results, patients were significantly distributed in two separate orientations, demonstrating that the MALRS score was strongly correlated with the prognosis of EOC patients (Figure S1C). The results of the GO and KEGG enrichment analyses of the DEGs (p < 0.05, |log2FC| ≥ 0.5) between the high‐ and low‐scoring groups indicate that the MALRS model is associated with cell proliferation (Figure S1D). Furthermore, GSEA enrichment analysis revealed that PD‐1 expression and tumor evasion and tolerogenicity were enriched in the MALRS model. The results of these enrichment analyses revealed a strong correlation between the MALRS model and tumor invasion and immunity (Figure S1E). Friends analysis was used to rank the gene importance of 7 intersecting genes. To evaluate the prognostic association of the MALRS score, we performed univariate Cox regression analysis. The results showed that the MALRS score was associated with the prognosis of ovarian cancer patients (Figure 1G).

2.3. Construction of the Predictive Nomogram and Screening of Core Prognostic Genes

The choice of treatment and prognosis of patients with tumors is often influenced by various factors, such as FIGO stage, tumor grade, and lymph node metastasis. As a single risk model has limitations in predicting patient prognosis, we integrated FIGO stage, grade, and MALRS score data to construct an exploratory nomogram (Figure 1G). The calibration curves showed good visual agreement between the predicted and observed OS probabilities at 1, 3, and 5 years. The survival probability predicted by the nomogram tended to approach the actual survival probability (Figure S1H). To further prioritize the seven candidate genes, three platinum‐resistant ovarian cancer datasets (GSE45553, GSE58472, and GSE73935) were obtained from GEO, and differential expression analyses were performed. The datasets represent distinct resistance models: GSE45553 compares cisplatin‐sensitive and ‐resistant spheroids, GSE58472 includes platinum‐sensitive and ‐resistant IGROV‐1 derivatives, and GSE73935 includes multiple drug‐resistant sublines. CDS2 showed a consistent upregulation trend in the cisplatin‐resistant comparisons and was also upregulated in topotecan‐resistant sublines (Figures 1H and S1I,J). These concordant expression patterns supported candidate prioritization but do not establish a causal role in drug resistance.

2.4. CDS2 Is Highly Expressed in Ovarian Cancer and Predicts a Poor Prognosis

The results presented above suggest that CDS2 may be a key gene in the tumorigenesis and progression of ovarian cancer, and correlation analyses of various aspects of CDS2 will be carried out in the future. Box plot results revealed that CDS2 expression was significantly upregulated in tumor tissues compared with that in ovarian cancer paraneoplastic tissues (Figure 2A). Within the same discovery dataset, ROC analysis yielded an AUC of 0.753 (Figure 2B) as an exploratory measure of discrimination; because no independent diagnostic cohort was used, this result does not establish clinical diagnostic performance. The survival information of a large sample of ovarian cancer patients was further analyzed via K–M Plotter (https://kmplot.com/analysis/), and the expression of CDS2 was significantly negatively correlated with the PFS of ovarian cancer patients, which was more pronounced in patients with advanced ovarian cancer, suggesting that in advanced epithelial ovarian cancer, the expression of CDS2 can predict the survival prognosis of patients to some extent (Figure 2C). To further validate the expression of CDS2 in epithelial ovarian cancer, we conducted an IHC assay on 142 obtained ovarian cancer tissues. The results demonstrated that CDS2 expression was significantly higher in advanced‐stage tumor tissues than in early‐stage tumors. We subsequently obtained the clinical information of patients with ovarian cancer and analyzed their clinicopathological characteristics as well as their CDS2 expression levels. As shown in Table S1, high CDS2 levels were positively correlated with advanced tumor stage but were not significantly associated with patient survival status, likely because of the limited number of patients involved. Taken together, these observations support CDS2 as a candidate marker associated with ovarian cancer progression and prognosis that requires independent validation.

FIGURE 2.

CDS2 expression is associated with poor prognosis in ovarian cancer. (A) Comparative analysis of CDS2 levels between tumor sites and adjacent normal tissues in patients with ovarian cancer. (B) ROC analysis summarized exploratory discrimination within the discovery dataset. (C) Survival curves of patients in the CDS2 high‐ and low‐expression groups. PFS: Progression‐free survival. (D) Immunohistochemical staining of CDS2 in ovarian cancer tissue sections at different stages (Left). The stacked bar chart shows the percentage distribution of ovarian cancer stages under different CDS2 expression levels (right). The error bars represent the means ± SD. *p < 0.05, **p < 0.01.

2.5. Overexpression of CDS2 Enhances Lipid Droplet Accumulation and Migration in Ovarian Cancer

To further explore the potential function of CDS2 in the pathogenesis and progression of ovarian cancer, we constructed a protein–protein interaction (PPI) network to identify the top 20 genes that interact with the CDS2 protein (Figure 3A). The network placed CDS2 at the center of a tightly connected phospholipid/glycerophospholipid metabolic module, including enzymes involved in phosphatidic acid metabolism and remodeling, such as members of the AGPAT family (AGPAT1‐5), LPIN family (LPIN1‐3), and phospholipases (PLD1/2). Notably, the network also included several enzymes involved in mitochondrial phospholipid biosynthesis, such as TAMM41, PGS1, and CRLS1, as well as the phosphatidylinositol synthase CDIPT, suggesting that CDS2 is functionally linked not only to general lipid metabolism but also to mitochondrial phospholipid pathways. As shown in Figure 3A, most of these associated proteins are involved in lipid metabolic processes. To further explore the correlation between CDS2 and lipid metabolism, we performed a correlation analysis between CDS2 and the above lipid metabolism‐related genes. The results revealed a significant correlation between CDS2 and the expression of genes involved in lipid transport, such as CD36; lysophosphatidic acid (LPA) receptors, such as LPAR1 and LPAR2; de novo lipid synthesis, such as SREBF1, SREBF2, SCD, ACLY, GPAM, FASN, ACSS2, ACSL4, HMGCR, and HMGCS1; and fatty acid β‐oxidation, such as CPT1A, CPT1B, and AMACR (Figure 3B). GSVA of the HALLMARK database revealed that the fatty acid metabolic pathway presented a greater score in the CDS2 high‐expression group (Figure S2A). GSEA revealed that the genes highly associated with CDS2 were predominantly enriched in phospholipid metabolism, glycerophospholipid biosynthesis, and sphingolipid synthesis (Figure S2B–D). These results provide further evidence that CDS2 plays a crucial role in lipid metabolism in ovarian cancer. To elucidate the biological functions and pathways associated with CDS2, the DEGs between the high‐expression and low‐expression groups were subjected to GO enrichment and KEGG pathway analyses. The pathways were enriched in several invasion‐ and metastasis‐related pathways, such as the Rap1 signaling pathway and the PI3K‐Akt signaling pathway (Figure 5A). The OVCAR3 and SKOV3 cell lines were used in subsequent experiments to construct CDS2‐overexpressing cell lines (Figure 3C,D). To assess the effect of CDS2 on lipid metabolism in ovarian cancer cells, we detected the percentage of intracellular lipid droplets via Oil Red O staining. The Oil Red O‐positive area significantly increased after CDS2 overexpression (Figure 3E). Notably, patients were categorized into CDS2‐high and CDS2‐low groups according to their IHC staining scores. As demonstrated in Figure 4A,B, CDS2 expression was significantly positively correlated with both Ki67 and epidermal growth factor receptor (EGFR) expression levels. Furthermore, the results of Transwell and wound healing assays revealed that the migration capacity of ovarian cancer cells was significantly enhanced after CDS2 overexpression (Figure 4C,D). These findings confirm that CDS2 overexpression increases lipid droplet accumulation and promotes migration in ovarian cancer cells, thereby supporting ovarian cancer development.

FIGURE 3.

Overexpression of CDS2 enhances lipid droplet accumulation in ovarian cancer. (A) The network was generated via STRING (v11.5; https://string‐db.org) with high‐confidence interactions (combined score > 0.7). Nodes represent proteins, with CDS2 highlighted in red. The edges denote experimentally validated interactions extracted from the STRINGC. qPCR was used to verify the expression of CDS2 at the mRNA level following plasmid transfection. (B) Box plot of the correlation analysis between CDS2 and lipid metabolism‐related genes. (C) qPCR was used to verify the expression of CDS2 at the mRNA level following plasmid transfection. (D) Western blotting was used to verify the expression of CDS2 at the protein level after plasmid transfection. NC: Normal control; OE: Overexpression. (E) Oil Red O staining was used to assess intracellular neutral‐lipid accumulation of OVCAR3 and SKOV3 cells (Left). Quantification of the percentage of lipid droplets after Oil Red O staining (right). The error bars represent the means ± SD. *p < 0.05, **p < 0.01, ****p < 0.0001.

FIGURE 5.

Overexpression of CDS2 promotes the progression of ovarian cancer cells by regulating the PI3K/AKT pathway. (A) KEGG analyses of CDS2 between the high‐ and low‐expression groups. (B) GSVA analysis based on the REACTOME (upper) and HALLMARK (lower) databases. (C) Correlation analysis of CDS2 and PI3K‐AKT–mTOR pathway markers. (D) Western blotting was used to detect the expression of PI3K/AKT signaling pathway proteins in OVCAR3 and SKOV3 cells. (E) Western blot analysis of CDS2 and p‐AKT in 10 freshly collected EOC tissue samples (left). Linear regression analysis demonstrating their correlation (right). (F) Representative images of IHC staining for CDS2 and p‐AKT in EOC tissues (left). Correlation analysis of CDS2 and p‐AKT protein expression levels in clinical EOC samples (right). The error bars represent the means ± SD. *p < 0.05, **p < 0.01, *** p < 0.001, ****p < 0.0001.

FIGURE 4.

Overexpression of CDS2 enhances migration in ovarian cancer. (A,B) Correlation analysis between CDS2 immunohistochemical scores and Ki67 positivity rates (B) and EGFR (A) expression. (C) Migration ability of OVCAR3 and SKOV3 cells detected by a scratch assay (left). The quantification of wound closure is shown as histograms (right). (D) Transwell assays were used to detect the invasive ability of OVCAR3 and SKOV3 cells (left). The quantification of invaded cells is shown as histograms (right). The error bars represent the means ± SD. *p < 0.05, **p < 0.01, ****p < 0.0001.

2.6. CDS2 Overexpression Promotes the Progression of Ovarian Cancer Cells by Regulating the PI3K/AKT Pathway

Pathway enrichment analysis indicated an association between CDS2 expression and PI3K/AKT signaling (Figure 5A). Consistently, GSVA (Figure 5B) and GSEA (Figure S3A) showed enrichment of PI3K/AKT/mTOR‐related signaling in the CDS2‐high expression group. Furthermore, exploratory correlation plots showed positive associations between CDS2 and PI3K‐AKT–mTOR pathway markers (Figure 5C). Because these multi‐gene correlations used nominal, unadjusted p values, they were interpreted as hypothesis‐generating. As shown in Figure 5D, CDS2 overexpression led to increased phosphorylation of PI3K and AKT in OVCAR3 and SKOV3 cells. Western blot analysis of fresh ovarian cancer tissue samples revealed that CDS2 expression was significantly correlated with the phosphorylation levels of AKT and PI3K (Figure 5E). Furthermore, IHC analysis of 30 ovarian cancer tissue samples revealed elevated AKT phosphorylation in tumors with high CDS2 expression (Figure 5F). These findings collectively suggest that CDS2 may promote ovarian cancer progression and metastasis by activating the AKT pathway through lipid metabolism regulation.

2.7. CDS2 Expression Levels Are Correlated With Different Immune Cell Abundances in Ovarian Cancer

To investigate the relationship between CDS2 expression and the immune microenvironment, immune cell abundance in ovarian cancer samples was estimated using the CIBERSORT and xCell algorithms (Figure 6A,B). Furthermore, the associations between CDS2 expression levels and different immune cell types were evaluated using Spearman rank correlation coefficients and corresponding p values. The results obtained via the CIBERSORT algorithm revealed a negative correlation between high CDS2 expression and activated CD4+ T cells, CD8+ T cells, and regulatory T cells. The results obtained via the xCell algorithm revealed a clear negative correlation between high CDS2 expression and several immune cell populations, including CD4+ memory T cells, CD4+ naïve T cells, CD4+ Tcm, CD8+ Tcm, mast cells, CLPs and macrophages. Across both algorithms, estimated CD8+ T‐cell and activated memory CD4+ T‐cell abundance showed negative associations with CDS2 expression (Figure 6C). These concordant computational estimates identify exploratory candidates for further study but do not establish a regulatory effect. Further scatter plots revealed a significant negative correlation between CDS2 expression and CD8+ T cells, as well as activated memory CD4+ T cells (Figure S4A). Furthermore, we conducted a survival analysis to ascertain the correlation between immune cell activity and the probability of survival in EOC patients. The results demonstrated a significant correlation between survival in EOC patients and the presence of both CD8+ T cells and activated CD4 memory T cells (Figure 6D). Given the close association between CDS2 and immune cells, we conducted an additional analysis to explore its correlation with immune checkpoints. As shown in Figure 6E, exploratory analyses using nominal, unadjusted p values indicated positive correlations between CDS2 expression and BTN3A1, CD226, CD274, CD276, CTLA‐4, ICOSLG, PDCD1LG2, PVR, and SIRPA expression. Among these exploratory findings, positive correlations were observed between CDS2 expression and the immune checkpoint molecules programmed cell death protein 1 (PD‐1) and cytotoxic T lymphocyte antigen‐4 (CTLA‐4) (Figure 6F). IHC analysis of 142 ovarian cancer tissue sections further revealed a statistically significant correlation between CDS2 and PD‐1 expression levels (Figure S4B). Overall, these computational and tissue‐level associations are hypothesis‐generating and do not establish immune suppression or immunotherapy response.

FIGURE 6.

CDS2 expression is associated with computationally estimated immune‐cell abundances in ovarian cancer. (A) Immune infiltration results based on the CIBERSORT algorithm. (B) Immune cell infiltration results based on the xCells algorithm. (C) Venn diagram showing the intersection results of the CIBERSORT and xCell immune infiltration algorithms. (D) Kaplan–Meier curves for the OS of patients in the high and low immune cell expression groups. CD8+ T cells (Left); CD4+ memory T cells (Right). (E) Correlation analysis of CDS2 with various immune checkpoints. F. Scatter plots showing the correlation between PD‐1 (upper) and CTLA‐4 (lower) expression levels and CDS2 expression levels. (F) Scatter plots showing the correlation between PD‐1 (upper) and CTLA‐4 (down) and CDS2 expression levels.

3. Discussion

Ovarian cancer ranks as the eighth most prevalent malignancy worldwide [26]. EOC represents the predominant pathological type, accounting for approximately 80% of all ovarian malignancies. The majority of EOC cases are thought to originate from the ovarian surface epithelium or the fimbriated end of the fallopian tube [27]. Serous ovarian cancer represents the most common subtype of EOC. Approximately 80% of serous ovarian cancer patients present with metastatic disease at diagnosis. Unlike malignancies that predominantly spread to peritoneal fluid‐exposed organs, both primary and recurrent high‐grade serous ovarian cancer exhibit a distinct tropism for adipose‐rich tissues, including the omentum, mesentery, and appendix. These sites are characterized by abundant fat deposits enveloping intestinal surfaces, endocrine activity, and a high concentration of energy‐dense lipids, which may contribute to their role as preferential metastatic niches [28, 29]. The molecular mechanisms underlying this adipotropic preference remain poorly understood. Therefore, systematic investigations are urgently needed to identify key drivers of ovarian cancer progression and peritoneal metastasis.

Other than insidious onset and poor prognosis, EOC is also known for its obvious colonization tendency to peritoneum which is lipid abundant [30, 31]. It's widely observed that among the majority of patients diagnosed with high‐grade serous ovarian cancer (HGSOC), which is the most representative subtype of EOC, widespread metastasis to omentum majus and mesenterium tissues that are fat‐abundant is common [30]. These unique metastatic sites tend to provide tumor cells with sufficient substrates containing enough endogenous energy such as fatty acid and paracrine growth signal, supporting its survival and metastasis [29, 32] Hence, fat tissue originated signaling and metabolic substrate may significantly reinforce invasive capacity and resistance of ovarian cancer cells [33]. The biological behaviors of EOC suggest that its malignancy was not only empowered by endogenous carcinogenic process, but was also closely coherent with its adaptation to metabolism in peritoneum microenvironment [34]. It is worth noting that mitochondria is deeply involved in the process of the progression and therapeutic response in ovarian cancer, by coordinating the flexibility of biological energy, participating in the exploitation of fatty acid and stabilization of redox homeostasis. Thereby, the key factor enabling this process might be enriched in mitochondrial metabolic network, and acting as a pivotal role in its crosstalk with lipid remodeling programme [34, 35, 36]. Given the above mentioned issues, consequently, other than establishing conventional screening protocol using nonspecific genes, we elaborated the algorithm by constructing a framework based on mitochondrial metabolism and integrating lipid remodeling, which could simultaneously recognize key molecular factors in metabolic adaptation and peritoneum metastasis. Unfortunately, the molecular mechanism of lipid‐favored metastasis preference of ovarian cancer remains undermine. Thus, this article aims at identifying and illustrating key genes that are responsible for the linkage of metabolic reprogramming and peritoneum metastasis.

Firstly, RNA‐seq expression data from 429 ovarian cancer samples were obtained from The Cancer Genome Atlas (TCGA). An expression matrix comprising 1925 mitochondria‐relevant genes acquired from publicly available sources was then constructed. Subsequently, Weighted Gene Co‐expression Network Analysis (WGCNA) was hereby applied to identify the blue and green modules that exhibited the strongest correlations with pathological stage, hence establishing a biologically informed framework for downstream analyses. What's more, we screened out 7 prognostic genes in EOC by integrating WGCNA authenticated genes from staging, DEGs of independent EOC cohort, and lipid metabolism related genes. Therefore, 62 sorts of machine learning algorithms were then exploited to conduct survival models on a basis of the above mentioned 7 genes; specifically, the combination of RSF and ridge exhibited the best predictive efficacy. As a result, we developed a MALRS model for predicting the prognosis of ovarian cancer patients using machine learning algorithms. Further illustration of PCA, enrichment analysis, and Cox regression analysis indicated that MALRS is strongly correlated to the prognosis of EOC patients, along with proliferation, invasion, and TIME features. In addition to the precise prediction in conventional clinical variants such as staging, grading, and lymph node metastasis of the MALRS model, it is noteworthy that it is capable of capturing subtle information containing mitochondria and pathological staging related risk factors. Moreover, considering the limitation of monofactor risk model in the evaluation of individual prognosis, we constructed a nomogram utilizing the staging and grading of MALRS and FIGO by combining risk genes and pathology, which furthered its adaptation in clinic. Aiming at further regocize the core regulatory factor among candidate genes, a multilayer priority screening was then carried out on the basis of the correlation of prognosis, ranking of contributions, and the transcriptome of 3 independent cisplatin‐resistant cohorts. The result illustrated that CDS2 was upregulated in all cisplatin‐resistant datasets and Topotecan resistant group, which indicates CDS2 as the most steady candidate. Considering CDS2 was highly expressed in tumor tissues, the relationship with adverse prognosis, and its high abundance of IHC in late stage tumors, we identified CDS2 as the ultimate core candidate of ovarian cancer progression.

CDS2 serves as the key enzyme that catalyzes the transformation of Phosphatidic Acid (PA) to CDP‐Diacylglycerol (CDP‐DAG). Other than being considered a fully characterized tumor oncogenic molecule, CDS2 was previously reckoned as the pivotal regulatory factor in maintaining homeostasis of PA and CDP‐DAG. It's been determined that the lack of CDS2 could induce the remodeling of de novo synthesis of PA [37]. It was reported that lacking CDS2 could destruct PIP2/PIP3 dependent signal transduction in endothelial cells, inducing VEGFA mediated blood vessel contraction and tumor suppression [20]. Moreover, CDS2 may be involved in the regulation of mitochondrial function and energy metabolism in a METTL3 dependent m6A modification pattern [38]. It's reported that an interaction that competitive synthesis between CDS1 and CDS2 could be vulnerable when it comes to therapeutic strategy against uveal melanoma and other sorts of tumors [39]. Conclusively, the above mentioned research did provide solid biological evidence of CDS2 in regulating lipid metabolism and mitochondria, which was chosen after our specific screening tactics. Nevertheless, most research would concentrate on explaining the crosstalks of CDS2 and general phenotypes such as homeostasis of lipid metabolism, vessel biology, and pan‐cancer. As a result, the biological meaning and interactive mechanism in signaling of CDS2 remains to be systemically elucidated. To some extent, this article indicated CDS2 as a candidate that is both clinically relevant and biologically meaningful other than solely a regulator of phospholipid metabolism, which compensates for the deficits in this area. By further analyzing and integrating CDS2 and our established models such as MALRS multi‐dimensionally, CDS2 was hereby recognized to serve as a core hub. CDS2 was firstly proved to participate in the interactions between mitochondria‐lipid metabolism reprogramming and immune suppressive microenvironment in TME. Furthermore, CDS2 was found intensively closely related to the activation of PI3K‐AKT signaling so as to provide late stage EOC with novel target and intertwined metabolic section that is of a sort not previously recognized.

Accumulating evidence has led to a conclusion where characteristics of lipid metabolism exert significant potential to serve as biomarkers utilizing as early carcinoma detection markers, specifically in EOC [40, 41, 42]. As EOC tends to be lipophilic, multiple mechanisms are exploited to enhance lipid intake so as to meet its metabolic demands of proliferation and metastasis [29, 43]. This unique biological behavior of EOC suggests that lipid metabolism reprogramming is pivotal in the adaptation of the peritoneum microenvironment that is lipocyte abundant, also impacting maintaining malignant progression [44, 45]. On one hand, EOC cells may reshape their lipid metabolism features to survive in nutrient‐lacking TME. For instance, Fatty Acid Binding Protein 4 (FABP4) in EOC would be upregulated to induce lipolysis of surrounding lipocytes, providing energy support by intaking released fatty acids [29]. On the other hand, this unique metabolic reprogramming could produce and magnify carcinoma signals, which could actuate proliferation and metastasis. For instance, the upregulation of carnitine palmitoyltransferase 1A (CPT1A) was certified to promote tumor progression [46]. Consistently, CDS2 is tightly related to the adaptation of EOC in lipid metabolism, acting as an important metabolic section in the crosstalk of lipid metabolism reprogramming and invasion. Our results illustrated that the role of CDS2 was not solely a participator of lipid synthesis, but a broad lipid metabolism reprogramming. Specifically, other than significantly correlated with CD36, LPAR1/2 and SREBF1/2 that are responsible for lipid intake, signaling transduction and de novo lipid synthesis, respectively, CDS2 is also positively relevant with fatty acid β oxidation modulators such as CPT1A/B and AMACR. In metastatic ovarian cancer, CD36 demonstrates significant overexpression, which markedly enhances fatty acid internalization capacity and promotes tumor progression [47]. Moreover, in hepatocellular carcinoma (HCC), CD36 upregulation activates the Src/PI3K/AKT signaling axis, promoting cellular proliferation, invasion, and metabolic reprogramming [48]. What's more, in vitro silencing of LPAR significantly impairs tumor cell invasiveness and migratory capacity. Furthermore, LPAR induces angiogenesis and promotes tumor cell invasion through extracellular matrices and basement membranes via activation of multiple intracellular signaling pathways, including the Ras/MEKK1‐MAPK and PI3K/Akt pathways [49]. The SREBP family serves as master regulators of lipid homeostasis, with multiple studies demonstrating their significant overexpression in metastatic ovarian carcinomas [50, 51, 52]. Mechanistically, SREBPs drive oncogenic progression by inducing cancer‐specific lipid metabolic reprogramming that promotes both proliferative and metastatic capacities. Notably, the PI3K/AKT oncogenic signaling pathway frequently activated in cancers stabilizes SREBP proteins, thereby modulating the expression of key enzymes involved in lipid biosynthesis [53]. Under metabolic stress, ovarian cancer cells upregulate carnitine palmitoyltransferase to enhance fatty acid β‐oxidation capacity. Importantly, the inhibition of CPT1A synergizes with cisplatin treatment, significantly improving therapeutic efficacy in ovarian cancer models [54]. In this study, we identified CD36, LPAR, SREBP, and CPT1A as key genes strongly correlated with CDS2 expression, all of which showed significant associations with ovarian cancer patient prognosis. These genes exhibited distinct functional roles in lipid metabolic reprogramming of ovarian cancer. Notably, CDS2 upregulation in two ovarian cancer cell lines markedly enhanced lipid droplet accumulation and cellular migratory capacity, suggesting potential coregulation through these associated genes. Simultaneously, Oil Red O staining manifested that the overexpression of CDS2 led to increased accumulation of lipid droplets. It's noteworthy that CDS2‐related lipid metabolism reprogramming is not simply just about energy storage, more importantly exerting phospholipid remodeling. Among the pathway enrichment analyses, both GSVA and GSEA demonstrated that elevated CDS2 expression was associated not only with enhanced fatty acid metabolism, but also with significant enrichment of pathways involved in phospholipid metabolism, glycerophospholipid biosynthesis, the phosphatidylinositol signaling system, and sphingolipid synthesis (Figure S2B–D). As a result, CDS2 catalyzes the conversion of phosphatidic acid (PA) to CDP‐diacylglycerol (CDP‐DAG) and plays a central role in phosphatidylinositol precursor supply as well as lipid homeostasis maintenance [37]. The metabolic alterations driven by the aberrant expression of CDS2 may extend beyond the accumulation of neutral lipids and may additionally affect the availability of phospholipids and related lipid species, thereby reshaping membrane lipid composition and membrane‐associated signaling states. In tumor cells, membrane lipid remodeling is not only essential for the dynamic renewal of biological membranes, but may also provide the structural basis required for membrane fluidity, receptor signaling platform organization, and invasive migration. In the canonical Class I PI3K/AKT signaling cascade, PI3K utilizes PIP2 as a substrate to generate PIP3, the accumulation of which at the inner leaflet of the plasma membrane serves as a critical lipid platform for AKT recruitment and activation [55].

Interestingly, in addition to proteins directly involved in lipid metabolic processes, the PPI network also identified several key mitochondrial enzymes, including TAMM41, PGS1, and CRLS1, as well as phosphatidic acid metabolism–related enzymes such as AGPAT1–5 and LPIN1–3. Among these, TAMM41, a CDP‐diacylglycerol synthase required for mitochondrial cardiolipin biosynthesis, suggests a potential role for CDS2—classically characterized as an endoplasmic reticulum–resident protein—in coordinating lipid trafficking and exchange between the endoplasmic reticulum and mitochondria. Moreover, GSEA further revealed significant enrichment of pathways related to mitochondrial biogenesis and transcriptional activation in the CDS2‐high expression group (Figure S2D), indicating that CDS2 may be involved not only in lipid remodeling but also in concurrent mitochondria‐associated metabolic adaptation.

Furthermore, integrated analysis of immunohistochemistry (IHC) results and KEGG pathway analysis revealed significant enrichment of EGFR expression and PI3K/AKT signaling pathway activation in the high‐risk patient cohort. EGFR, a member of the receptor tyrosine kinase (RTK) family, plays a pivotal role in numerous solid malignancies [56]. EGFR overexpression predominantly regulates cellular proliferation, differentiation, and survival through activation of the STAT signaling pathway, PI3K/AKT cascade, and PLC‐γ/PKC axis [57]. Dysregulation of the fundamental PI3K/AKT signaling pathway invariably promotes malignant progression, driving tumor formation, metastatic potential, and therapeutic resistance to conventional cancer treatments [58, 59]. Both Western blot and IHC assessments revealed that CDS2 overexpression is associated with enhanced AKT activation Collectively, our findings position CDS2 as a potential metabolic switch upstream of the PI3K/AKT pathway. We propose that CDS2 may functionally synergize with lipid metabolic genes (e.g., CD36, LPAR) to activate PI3K/AKT phosphorylation, thereby stabilizing SREBP and modulating lipid biosynthesis. Importantly, targeting CDS2 expression emerges as a promising lipid‐focused anticancer strategy capable of suppressing both tumor growth and metastatic dissemination.

Recent studies have established that efficient peritoneal metastasis serves as the primary determinant of the high recurrence and mortality rates observed in patients with ovarian cancer [5, 6, 7]. This metastatic process requires intricate crosstalk between malignant cells and stromal components within the tumor microenvironment [60]. Elucidating these cellular interactions is critical for developing targeted immunotherapeutic strategies. Immune cells, particularly CD8+ T cells, serve as critical players in TME [61, 62]. Notably, our study revealed a strong correlation between CD8+ T cells and activated CD4 memory T cells and CDS2 expression. Upon antigen activation, they mediate cytotoxic effects through the release of interferon‐gamma (IFN‐γ), granzymes, and perforin, thereby eliminating tumor cells. Notably, we identified significant correlations between CDS2 expression and immune checkpoint molecules PD‐1 and CTLA‐4, suggesting their potential interplay in modulating CD8+ T cell functionality within the ovarian cancer immune microenvironment. However, within the tumor microenvironment, elevated expression of immune checkpoint receptors (e.g., PD‐1, CTLA‐4) frequently drives CD8+ T cells into a state of functional anergy or exhaustion [61]. These immune checkpoint blockade (ICBs) function by attenuating immunosuppressive signaling within the tumor microenvironment, potentiating antigen‐presenting cell activation and effector T‐cell responses [63]. Despite being a revolutionary cancer treatment, immune checkpoint blockade as monotherapy achieves objective responses in merely 6%–15% of patients, while 85% develop resistance [7, 43]. The limited clinical efficacy suggests tumors can evade immune destruction through multiple unidentified mechanisms. Elucidating these immune escape pathways is therefore critical for improving immunotherapy outcomes. Recent studies have revealed the crucial role of intratumoral T cell lipid metabolic reprogramming in T cell activation. Hu et al. demonstrated that constitutive activation of the AKT/mTORC1/SREBP1 signaling axis in CD8+ T cells drives metabolic reprogramming, resulting in aberrant cholesterol biosynthesis and subsequent T cell exhaustion [64]. Our study revealed a significant negative correlation between CDS2 expression and CD8+ T cell infiltration abundance, while demonstrating positive correlations with PD‐1 and CTLA‐4 expression levels. Immunohistochemical analysis further identified robust associations between CDS2 expression and multiple clinically relevant markers, including EGFR, and immune checkpoint molecule PD‐1. Collectively, these computational and tissue‐level associations are hypothesis‐generating. They do not establish regulation of CD8+ T cells or response to immune‐checkpoint blockade and may be influenced by tumor purity, histology, stage, stromal composition, and batch effects in bulk transcriptomic data.

From a translational perspective, these findings nominate CDS2 as a candidate metabolic marker for further investigation rather than an immediately actionable clinical target. Its associations with advanced clinicopathological features, lipid metabolic changes, PI3K/AKT‐related signaling, and immune‐related characteristics may provide a basis for future studies evaluating whether CDS2 can contribute to molecular stratification or prognostic assessment in EOC. Mechanistically, subsequent studies incorporating loss‐of‐function and rescue approaches, direct lipidomic measurements, and pathway‐intervention experiments will be required to determine whether these associations reflect a causal role of CDS2. Independent clinical cohorts will also be necessary to assess the reproducibility and potential clinical utility of these observations.

Taken together, our findings suggest associations between CDS2 expression, lipid‐related cellular phenotypes, PI3K/AKT‐associated signaling, and immune‐related features in ovarian cancer. However, several study limitations warrant consideration. The transcriptomic datasets include different serous histologic and reference‐tissue contexts, and MALRS was internally evaluated in three held‐out subsets from one fixed‐seed TCGA partition. Further resampling, histology‐specific assessment, and independent external validation will be valuable for confirming generalizability. Mechanistically, while our data suggest CDS2 potentially coordinates both CD36‐mediated exogenous fatty acid uptake and SREBP‐dependent de novo lipogenesis to fulfill the energetic and biomembrane synthetic demands of metastatic tumor cells, direct experimental validation of this coordinated metabolic reprogramming requires further investigation. Furthermore, although our immune profiling implicates CDS2 in regulating CD8+ T cell infiltration through PD‐1/CTLA‐4 upregulation, the precise metabolic alterations imposed on CD8+ T cells remain undefined. Addressing these knowledge gaps through subsequent mechanistic studies will be essential for comprehensively understanding ovarian cancer progression and developing targeted therapeutic strategies.

4. Conclusion

In summary, this study integrated transcriptomic data to develop the MALRS model for prognostic risk stratification in ovarian cancer. Through multi‐step exploratory prioritization, CDS2 was identified as a candidate showing increased expression in advanced ovarian cancer and drug‐resistance‐associated datasets. Functionally, it promoted intracellular lipid droplet accumulation and enhanced the migratory capacity of EOC cells, while its expression was positively associated with activation of the PI3K/AKT signaling axis. Furthermore, computational immune profiling identified exploratory associations between CDS2 expression, estimated T‐cell abundance, and immune‐checkpoint transcripts. Collectively, these findings identify CDS2 as a phospholipid‐metabolism‐associated candidate linked to aggressive tumor behavior and immune‐related transcriptomic features in an HGSOC‐enriched context and support its independent biological and clinical validation.

5. Experimental Procedures

5.1. Data Collection

RNA‐seq and survival data for 429 TCGA ovarian carcinoma samples, which predominantly represent high‐grade serous ovarian carcinoma, were downloaded from the TCGA database (https://portal.gdc.cancer.gov/repository). For cross‐dataset differential‐expression screening, GEO data comprised 56 serous tumor/borderline samples and 30 nonmalignant reference samples (https://www.ncbi.nlm.nih.gov/geo/). Specifically, GSE10971 includes non‐malignant fallopian‐tube epithelium and high‐grade tubal/ovarian serous carcinoma, whereas GSE27651 includes ovarian surface epithelium, serous borderline tumors, low‐grade serous carcinoma, and high‐grade serous carcinoma. These datasets were analyzed separately and intersected for candidate prioritization; they were not treated as a biologically uniform clinical cohort. Furthermore, 1925 mitochondria‐related genes were obtained from the published literature. In August 2024, GeneCards (https://www.genecards.org/) was searched using the terms “lipid metabolism,” “fat metabolism,” and “fatty acid metabolism.” The returned gene symbols were combined and deduplicated, yielding 743 nonredundant lipid metabolism‐related genes; no additional relevance‐score cutoff was applied. The study was conducted in accordance with the TCGA and GEO access policies, relying exclusively on databases that are publicly accessible and adhering strictly to publication guidelines. This approach obviated the necessity for approval from the local ethics committee.

5.2. Weighted Gene Coexpression Network Analysis

The TCGA‐OC cohort was used as the training cohort, and the expression matrix of 1925 mitochondria‐related genes was extracted. After standardized preprocessing, the correlations between genes were calculated to construct a similarity matrix. The optimal soft‐thresholding power β was determined on the basis of the scale‐free topology criterion, and the similarity matrix was transformed into a weighted adjacency matrix, where the elements were the β powers of the similarity coefficients. The topological overlap matrix (TOM) was then computed and converted into a dissimilarity matrix. Subsequent clustering analysis was performed to identify modules, and the correlation between module eigengenes and phenotypic data was analyzed to reconstruct the coexpression network. Finally, the two most relevant module genes were identified.

5.3. Differentially Expressed Gene Identification

Differentially expressed genes between ovarian cancer tumor tissues and paracarcinoma tissues were identified via the ‘limma’ R software package, with criteria for DEG identification set at a false discovery rate (FDR) < 0.05 and |log2FC| ≥ 0.7.

5.4. Construction and Evaluation of Machine Learning Models

The 429 TCGA‐OC cases were randomly divided at an approximately 1:1:1:1 ratio into a training cohort (n = 108) and three nonoverlapping internal validation cohorts (n = 107 each). The candidate algorithm set was prespecified, and the seven core genes were applied to 62 machine‐learning algorithms for survival modeling. Model fitting and parameter estimation were confined to the training cohort; the held‐out cohorts were used to compare out‐of‐sample C‐indices across three nonoverlapping subsets within the fixed, reproducible partition. Across the three validation cohorts, RSF + ridge, RSF + Enet (α = 0), Enet (α = 0), and Ridge yielded the same displayed C‐index values (0.54, 0.53, and 0.65; mean 0.574). RSF + ridge was retained as a representative of the tied top‐performing models and was named MALRS; the present results do not demonstrate superiority over the other tied models. The seven‐gene candidate set defined above was used as the fixed input for model comparison. The random seed was set to 123,456, and model fitting and parameter estimation were confined to the training subset. Each patient's model score was subsequently calculated, and patients were categorized into high‐ and low‐scoring groups on the basis of the median risk score. Moreover, the predictive accuracy of the model utilized principal component analysis (PCA), gene set enrichment analysis (GSEA), and other analytical techniques. Finally, univariate analyses were used to assess the ability of the MALRS model to independently predict the prognosis of patients with ovarian cancer.

5.5. Construction and Evaluation of a Predictive Nomogram Integrating the MALRS Signature and Clinical Data

As the predictive validity of the MALRS model alone may not be sufficient, we incorporated stage and grade into the construction of the nomogram. The nomogram was designed via the R package “rms” to integrate the risk score of the model as a prognostic factor. This enabled the evaluation of the predictive probability of 1‐, 3‐, and 5‐year OS. In addition, calibration plots were used to facilitate a visual comparison between the actual and nomogram‐predicted results.

5.6. Functions and Pathway Enrichment Analysis of DEGs

The gene expression levels were stratified into high‐ and low‐level categories for CDS2 via the “limma” package. The R packages “enrichplot”, “org.Hs.eg.db” and “ggplot2” were subsequently used for Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses to identify enriched pathways. Additionally, gene set enrichment analysis (GSEA) was performed via the “clusterProfiler” package to elucidate the activities and pathways that are closely associated with CDS2.

5.7. Gene Set Variation Analysis (GSVA)

Gene set variation analysis (GSVA) was also performed to evaluate the correlation between CDS2 and the lipid metabolism‐based signature and signaling pathway. The R package “GSVA” was utilized to conduct GSVA, and the outcomes were subsequently visualized via the R package “pheatmap”.

5.8. Tumor Microenvironment and Immune Infiltration Analysis

CIBERSORT and xCell were used as complementary algorithms to estimate immune‐cell abundance from bulk transcriptomic data. Spearman rank correlation and the Wilcoxon rank‐sum test were used to evaluate associations with key genes. These computational findings were interpreted as exploratory associations rather than direct evidence of immune‐cell regulation or immunotherapy response.

5.9. Tissue Specimens

Ovarian cancer tissues were collected from 30 EOC patients at Guangdong Provincial People's Hospital. All patients provided written informed consent, and the study was approved by the Ethics Committee of Guangdong Provincial People's Hospital (KY‐Z‐2021‐188‐02). In addition, formalin‐fixed, paraffin‐embedded ovarian cancer tissue microarrays (TMAs) were obtained from the National Engineering Center for Biochip (Outdo Biotech, Shanghai, China), with approval from the Outdo Biotech Clinical Research Ethics Committee (SHYJS‐CP‐1804029). The corresponding clinical information is provided in Table S1.

5.10. Cell Culture

The human ovarian cancer cell lines OVCAR3, SKOV3, A2780, ES‐2, and CAOV3 and the normal ovarian epithelial cell line HOSEPIC were obtained from the American Type Culture Collection (ATCC). All the cells were maintained in Dulbecco's modified Eagle's medium (Thermo‐Scientific, USA) supplemented with 10% fetal bovine serum and 1% penicillin–streptomycin. All the cells were incubated according to the manufacturer's instructions at 37°C in a water‐saturated atmosphere of 5% CO₂/95% air. Mycoplasma tests were negative, and STR analysis authenticated all lines.

5.11. Immunohistochemistry (IHC)

The tissue sections were processed and stained according to standard protocols at the Biopathology Institute (Servicebio, Wuhan, Hubei, China) via primary antibodies against CDS2 (1:1000). Immunostaining intensity was scored as follows [65]: negative (0), weak (1), moderate (2), and strong (3). The extent of staining was evaluated on the basis of the percentage of positively stained cells: 0 (< 10%), 1 (10%–25%), 2 (25%–50%), 3 (50%–75%), and 4 (> 75%).

5.12. Cell Transfection

OVCAR3 cells and SKOV3 cells in the logarithmic growth phase were inoculated in 6‐well plates. When confluence reached 60%–80%, the cells were transfected with plasmid‐containing RNA overexpressing CDS2 via Lipofectamine 2000 (Invitrogen; Shanghai, China) according to the manufacturer's instructions. The cells were cultured for 48–72 h and collected for subsequent experiments.

5.13. RT–qPCR

An EZ‐press RNA Purification Kit (EZBioscience; USA) was used to isolate total RNA from OVCAR3 and SKOV3 cells following the manufacturer's protocol. The RNA concentration and quality were assessed with a microspectrophotometer (KAIAO; Beijing, China). cDNA synthesis was performed via HiScript III RT SuperMix (Vazyme; Nanjing, China). GAPDH was used as a housekeeping gene. SYBR Green Master Mix (Vazyme; Nanjing, China) was used to detect the products. The PCR cycling conditions were as follows: 95°C for 30 s; 95°C for 5 s; and 60°C for 30 s for 40 cycles. The relative RNA expression was calculated via the 2^−ΔΔCt method.

5.14. Western Blot

The cells were collected, and the proteins were extracted from the cells via RIPA lysis buffer. The protein concentration was determined via a BCA protein assay kit (Thermo Scientific; USA). Total protein (60 μg) was separated via 10% SDS–PAGE and then transferred onto a 0.45‐μm PVDF membrane. The membranes were blocked with 5% nonfat dry milk in TBS for 1 h at room temperature. CDS2 antibody (1:1000, AUSBIO), PI3K antibody (1:1000, Proteintech), P‐PI3K antibody (1:1000, Cell Signaling), AKT antibody (1:1000, Cell Signaling), P‐AKT antibody (1:1000, Cell Signaling), and β‐actin antibody (1:10,000, Proteintech) were incubated overnight at 4°C. The next day, after washing, secondary antibodies (1:5000, Proteintech) were added, and the samples were incubated at room temperature for 2 h. The relative density of the bands was visualized via Image Quant software.

5.15. Wound Healing Assay

Following transfection for 48 h, the cells were collected and seeded into 6‐well plates at a density of 5 × 105 cells/well. When the cell confluence reached 90%, a pipette tip (10 μL) was used to form a scratch. Images were taken at 0 and 24 h, and the migration area was analyzed with ImageJ.

5.16. Transwell Assay

Transwell chambers (BD Falcon; USA) were used to assess the invasive potential of the cells. Following transfection for 48 h, 4 × 10⁵ cells were prepared, resuspended in serum‐free DMEM and then added to the upper chamber. The lower chamber was filled with 500 μL of DMEM containing 20% FBS. After incubation at 37°C for 48 h, the cells were fixed with formaldehyde and acetic acid (1:1) at room temperature for 30 min and then stained with crystal violet for 30 min. The invading cells were counted via a microscope.

5.17. Oil Red O Staining

After 48 h of transfection, the cells were treated with 1.2 mM oleic acid for 24 h and finally stained with an Oil Red O kit (Senbeijia Biological Technology; Nanjing, China). ImageJ was used to quantify the Oil Red O‐positive area fraction in the cells.

5.18. Statistical Analysis

Bioinformatic and statistical analyses were performed in R 4.4.1 using RStudio 2024.04.2 + 764. Key packages included WGCNA 1.72–5, limma 3.60.2, survival 3.7–0, randomForestSRC 3.3.1, glmnet 4.1–8, rms 6.8–2, clusterProfiler 4.12.0, enrichplot 1.24.0, GSVA 1.52.2, pheatmap 1.0.12, ggplot2 3.5.1, and org.Hs.eg.db 3.19.1. GSEA Desktop 4.3.3 and GraphPad Prism 9.0 were also used. The results are reported as the means ± SD. The two groups were compared by Student's t‐test. For differential expression analyses, false discovery rate correction was applied as specified above. Unless otherwise stated, p values in exploratory multi‐gene correlation panels were nominal and unadjusted; these findings were interpreted as hypothesis‐generating. A p value < 0.05 was considered statistically significant.

Author Contributions

Lixue Cao: conceptualization, investigation, funding acquisition, writing – original draft, writing – review and editing, methodology, supervision, validation, project administration. Yanzhen Chen: investigation, validation, formal analysis, data curation. Yingqi Liu: investigation, validation, formal analysis, data curation, funding acquisition. Changnian Yang: writing – original draft, writing – review and editing, investigation, visualization, validation, methodology, funding acquisition. Zhentao Zhang: conceptualization, validation, visualization, methodology, software, formal analysis. Yongqing Lin: investigation, methodology, validation. Yiyuan Wang: software, formal analysis. Siman Chen: software, formal analysis. Fei Yu: validation, formal analysis. Haiyan Ye: writing – review and editing, investigation, supervision. Shanshan Wu: funding acquisition, resources, supervision, data curation, conceptualization, investigation, writing – original draft, writing – review and editing, validation, methodology, formal analysis, software, project administration.

Funding

This work was supported by the National Natural Science Foundation of China (grant number 82103534), Guangdong Basic and Applied Basic Research Foundation (2025A1515220088) to L.C. This work was supported by the Guangdong Province Science and Technology Innovation Strategy Special Fund for College Students’ Scientific and Technological Innovation Cultivation (No. pdjh2025bk102) to Y.L., C.Y., and S.W. This work was also supported by the 2024 Undergraduate Innovation and Entrepreneurship Education Base Project of Guangdong Medical University (2JD24227) to S.W. We would like to express our gratitude to these funding sources for their support.

Ethics Statement

All procedures of this study were approved by the Ethics Committee of Guangdong Provincial People's Hospital (approval number KY‐Z‐2021‐188‐02) and the Outdo Biotech Clinical Research Ethics Committee (SHYJS‐CP‐1804029).

Consent

Written informed consent was obtained from all participants.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: Construction and evaluation of prognostic models for ovarian cancer. (A) Graphs of scale independence mean connectivity and scale‐free topology the appropriate soft‐power was 6. (B, C) PCA analysis between high‐ and low‐risk score groups in ovarian cancer. (D, E) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses and Gene Set Enrichment Analysis (GESA) of differentially expressed MALRS‐related genes. (F) Functional Friends analysis of the seven model genes was performed, with genes ranked from top to bottom based on their pathway enrichment significance. (G) Univariate analysis of different clinical factors and the MALRS model for ovarian cancer prognosis prediction. (H) Quantitative assessment of patient survival prediction in the TCGA‐OC cohort. The calibration plots for consistency between predicted and observed 1‐year, 3‐year, and 5‐year survival. (I) Box plots showing CDS2 expression in three ovarian cancer platinum‐resistance datasets. (J) CDS2 expression upregulated in topotecan resistant group.

Figure S2: CDS2 expression in ovarian cancer is closely related to metabolic pathways. (A)The expression profiles of lipid‐related genes between the two cohorts. (B, C) GSEA enrichment analysis of lipid metabolism pathway showed that CDS2 was associated with phospholipid metabolism, glycerophospholipid biosynthesis and sphingolipid synthesis. (D) CDS2 is associated with transcriptional activation and mitochondrial biosynthesis.

Figure S3: CDS2 expression is closely related to AKT signaling pathway in ovarian cancer. (A)GSEA enrichment analysis of the PI3K pathway.

Figure S4: CDS2 is associated with immune cell infiltration and immune checkpoint expression. (A) Scatter plots demonstrate a correlation between CD8+ T (Left) cells and CD4+ memory T cells (Right) of immune cells and CDS2 expression levels. (B) Correlation analysis between CDS2 immunohistochemical scores and PD‐1.

Table S1: The table shows the relationship between CDS2 protein and clinicopathological parameters of epithelial ovarian cancer.

Data Availability Statement

The data are available from the corresponding author upon reasonable request.