Identification and Validation of Genes Involved in Energy Metabolism in Polycystic Ovary Syndrome.

In the two training sets, we identified a total of 2,327 DEGs in training set 1 (1,512 upregulated and 815 downregulated) and 1,322 DEGs in training set 2 (593 upregulated and 729 downregulated) ( Fig. 1A-1B ). By intersecting the DEGs from both datasets, we obtained 159 co-DEGs, including 49 downregulated and 110 upregulated genes ( Fig. 2A ). GO enrichment analysis of these 159 co-DEGs revealed 369 significant terms, which were categorized into 43 Cellular Components (CCs), 40 Molecular Functions (MFs), and 286 Biological Processes (BPs) ( Fig. 2B ). Among the most notable GO terms were ‘phagocytosis’ (BP), ‘specific granule’ (CC), and ‘peptidase regulator activity’ (MF). Additionally, KEGG pathway analysis identified 34 enriched pathways, including ‘Fc gamma R-mediated phagocytosis’, ‘Ras signaling pathway’, and ‘protein digestion and absorption’ ( Fig. 2C ). Identification of differentially expressed genes in PCOS granulosa cells. A) Volcano plot of the datasets GSE34526 and GSE193123 showing DEGs. Green triangles represent downregulated genes, red triangles represent upregulated genes, and gray triangles represent genes with no significant differential expression. B) Each column represents a dataset, and each row corresponds to a gene. Blue ones represent downregulated genes and red ones represent upregulated genes. Functional enrichment analysis of co-expressed DEGs in PCOS. A) DEGs identified through pairwise comparison between PCOS and control GCs. B) GO analysis of DEGs between PCOS and control groups. C) KEGG analysis of DEGs between PCOS and control groups. We identified three DE-ERGs as potential biomarkers: CD44, HS2ST1, and GPC4 ( Fig. 3A ). In both the GSE34526 and GSE193123 datasets, CD44 was upregulated, while HS2ST1 and GPC4 were downregulated ( Fig. 3B ). Chromosomal localization revealed that CD44 is located on chromosome 11, HS2ST1 on chromosome 1, and GPC4 on chromosome X ( Fig. 3C ). The diagnostic performance of these biomarkers was further validated by their area under the curve (AUC) values, which exceeded 0.8 in both training sets, suggesting strong predictive accuracy for PCOS ( Fig. 3D ). Validation of ERG biomarkers in PCOS. A) Comparative analysis of the two datasets showed that 3 DE-ERGs (CD44, HS2ST1 and GPC4) were deemed biomarkers. B) CD44 was an up-regulated gene, while HS2ST1 and GPC4 were downregulated in both the GSE34526 and GSE193123 datasets. C) Chromosomal localization of the differentially expressed genes for the three biomarkers. D) ROC curves were generated and AUC values were calculated to evaluate the diagnostic performance of these biomarkers in two training sets using the survivalROC package. Gene-Gene Interaction (GGI) analysis revealed the top 20 genes that most strongly interacted with these biomarkers, including ABCC5, GPC6, and HS3ST1. These genes were mainly involved in glucose metabolism processes, such as ‘glycosaminoglycan metabolic process’ and ‘aminoglycan catabolic process’ ( Fig. 4A ). Subcellular localization analysis predicted that CD44 is expressed on the plasma membrane, HS2ST1 in the Golgi apparatus, and GPC4 in both the endosome and plasma membrane ( Fig. 4B ). Furthermore, potential pathways associated with each biomarker were identified based on correlations with other genes ( Fig. 4C ). Specifically, CD44, with its high expression, was primarily linked to the toll-like receptor signaling pathway. Both CD44 and HS2ST1 were associated with the intestinal immune network for IgA production and asthma. Additionally, both HS2ST1 and GPC4 were involved in antigen processing and presentation, while GPC4, with low expression, was linked to the B-cell receptor signaling pathway. Moreover, GSVA revealed five upregulated KEGG pathways, including the toll-like receptor and B-cell receptor signaling pathways ( Fig. 4D ). Functional characterization of biomarkers. A) Subcellular localization predict of the biomarkers was located using Cell-PLoc 3.0. B) Potential pathways for each of the 3 biomarkers were obtained based on correlations of biomarkers and other genes. C) Differentially expressed KEGG up-regulated pathways were identified by GSVA. D) GGI analysis demonstrated the top 20 genes (ABCC5, GPC6, HS3ST1 etc.) that interacted most strongly with 3 biomarkers and mainly related to multiple glucose metabolism. The X2K network revealed that 13 transcription factors, such as ATF2, YY1, and TCF, were signi-ficantly associated with the targeted biomarkers. These transcription factors interacted with 40 proteins, including PML, SMAD4, and RB1, as well as 148 kinases, such as CDK1, MAPK14, and GSK38 ( Fig. 5A ). Additionally, the ceRNA network predicted 48 miRNAs, such as hsa-miR-106a-5p, hsa-miR-106b-5p, and hsa-miR-124-3p, along with 61 lncRNAs, including SNHG14, GAS5, and H19 ( Fig. 5B ). For instance, AL031428.1 regulates GPC4 through hsa-miR-27a-3p, while FGD5-AS1 controls CD44 and FGD5-AS1 via hsa-miR-302a-3p. Based on the CTD, predictions were made for 15 compounds ( Fig. 5C ). Notably, bisphenol A was predicted to interact with all three biomarkers simultaneously. Regulatory networks and therapeutic predictions for biomarkers. A) The X2K network analysis revealed that 13 transcription factors (TFs), including ATF2, YY1, and TCF, were the most significant TFs targeting the biomarkers, with 40 proteins interacting with 148 kinases. B) The ceRNA network predicted 48 miRNAs and 61 lncRNAs associated with the biomarkers. C) Biomarker-related compounds were predicted and a gene-compound network was constructed using the Comparative Toxicogenomics Database. Single sample gene set enrichment analysis (ssGSEA), which is designed for single sample cannot do GSEA, can be implemented by the GSVA R package. ssGSEA was used to analyze the samples in the training set, and the immune-related genes provided in the literature were used as the background gene set to obtain the enrichment scores of 28 immune cells. Wilcoxon test was used to analyze the differences in the enrichment scores of each immune cell between the PCOS and Control groups ( Fig. 6A ). There were significant differences in the enrichment scores of 19 immune cells between the groups (p < 0.05), and all of them were higher in the PCOS group than in the Control group. The Spearson correlation was used to analyze the correlation between hub genes and the enrichment scores of immune cells with differences between groups ( Fig. 6B ). Nineteen immune cells were negatively correlated with HS2ST1 and GPC4, and positively correlated with CD44. Immune infiltration analysis in PCOS granulosa cells. A) Wilcoxon test was used to analyze the differences in the enrichment scores of each immune cell between the PCOS and Control groups, and the results are shown in Fig. 1 - 2 . There were significant differences in the enrichment scores of 19 immune cells between the groups (p < 0.05). B) The Spearson correlation was used to analyze the correlation between hub genes and the enrichment scores of immune cells with differences between groups. Nineteen immune cells were negatively correlated with HS2ST1 and GPC4, and positively correlated with CD44. A total of 50 follicular fluid samples were collected. After RNA quantification, a total of 43 follicular fluid samples were finally used for reverse transcription and RT-QPCR analysis. Table 4 presents baseline characteristics and comparisons of ovarian stimulation parameters according to group variables. There was no significant difference in demographic parameters such as age (p=0.227) and body mass index (BMI, p=0.302) between the control group and the polycystic ovary syndrome (PCOS) group. In terms of endocrine indicators, basal luteinizing hormone (LH, 5.76 IU. L -1 vs. 4.43 IU.L -1 , p=0.024), total testosterone (TT, 0.41 ng. mL -1 vs. 0.29 ng. mL -1 , p<0.001) and anti-Mullerian hormone (AMH, P <0.001) were found in the PCOS group. 9.17 ng. mL -1 vs. 4.68 ng. mL -1 , p<0.001) were significantly higher than those in the control group. In the monitoring of ovulation induction cycle, the level of follicle stimulating hormone (FSH) on the day of human chorionic gonadotropin (HCG) injection in PCOS group was significantly lower than that in control group (9.21 IU. L -1 vs. 12.39 IU. L -1 , p=0.004). During the same period, the LH level was significantly increased (2.80 IU. L -1 vs. 1.26 IU. L -1 , p=0.010). The total dose of gonadotropin (Gn) in the PCOS group was significantly lower than that in the control group (1,437.50 IU vs. 1,800.00 IU, p=0.017), but there was no significant difference in the number of oocytes retrieved between the two groups (21.52 vs. 16.77, p=0.075). In terms of assisted reproductive technology, the use of intracytoplasmic sperm injection (ICSI) and conven-tional in vitro fertilization (IVF) showed a borderline significant difference between the groups (p=0.056), while the use of preimplantation genetic testing (PGT-A+M, PGT-A+SR, etc.) was lower than the baseline and there was no significant difference between the groups. Granulosa cell samples were obtained from patient basic data ( x __ ±s, n(%)) Note: The normal distribution data of continuous variables were expressed as mean ± standard deviation, and the non-normal distribution data were expressed as median (interquartile range). Categorical variables are represented as n(%). Independent sample T test (normal distribution) or Wilcoxon rank sum test (non normal distribution) were used for comparison between groups. Categorical variables were compared using Fisher’s exact test. The non-parametric test analysis of CD44, GPC4 and HS1ST2 in PCOS group (PCOS) and control group (CON) found that: The expression of CD44 ( Fig. 7A ) was significantly up-regulated in PCOS group (median PCOS=0.8976 vs. CON=0.3622, P=0.0079, two-sided Mann-Whitney U test), the actual median difference was 0.5354, and the Hodges-Lehmann effect size was 0.4181. It suggests that it may participate in the pathological process of PCOS by regulating cell adhesion or inflammatory response. Although there was no significant difference in GPC4 ( Fig. 7B ) expression (P=0.1306), there was a moderate effect size difference (HL=0.1138) in the median expression of GPC4 in the PCOS group (0.3732) compared with the control group (0.1927), combined with its biological function in the Wnt signaling pathway. A larger sample size (currently n=16/14) is needed to further verify the potential association. The expression difference of HS1ST2 ( Fig. 7C ) was the smallest between the two groups (median PCOS=0.1281 vs. CON=0.1081, P=0.2945, HL=0.03502), and the sample size was limited (n=15/13), so the current data did not support its significant correlation with PCOS. Validation of biomarker expression in clinical granulosa cell samples. The expression of A) CD44, B) GPC4, C) HS1ST2 in luteinized granulosa cells from different patients.CD44, PCOS group (n=25) and control group (n=17), The median expression level in PCOS group (0.8976) was significantly higher than that in the control group (0.3622), and the actual median difference was 0.5354. The Hodges-Lehmann effect size was estimated to be 0.4181. There was a statistically significant difference (P = 0.0079*, two-sided Mann-Whitney U test).; CPG4 PCOS group n=16, control group n=14,PCOS group median expression level (0.3732) was slightly higher than that of control group (0.1927), but the actual median difference was small (0.1805), Hodges-Lehmann effect size was 0.1138, no statistically significant difference (P = 0.1306,P = 0.1306,P = 0.1306,P = 0.3732,P =14). Bilateral Mann-Whitney U test). HS1ST2: PCOS group n=15, control group n=13). The median expression level of PCOS group (0.1281) was close to that of control group (0.1081), the actual median difference was only 0.01994, and the Hodges-Lehmann effect size was 0.03502. There was no statistically significant difference (P = 0.2945, two-sided Mann-Whitney U test).

Polycystic ovary syndrome (PCOS) is a common endocrine disorder affecting women of reproductive age, presenting with a spectrum of symptoms including hyperandrogenism, ovulatory dysfunction, and polycystic ovarian morphology ( 1 , 2 ). It is estimated to affect 6-15% of women worldwide, making it a leading cause of infertility and metabolic issues ( 3 ). Despite progress in understanding PCOS, current diagnostic criteria, which rely primarily on clinical signs, hormone levels, and ultrasound imaging, often fail to capture the full complexity of the disorder ( 4 ). The reliance on these criteria can lead to underdiagnosis or misdiagnosis, potentially delaying appropriate interventions and leaving patients without effective care ( 5 ). Furthermore, traditional therapeutic approaches may not adequately address the complexities of the syndrome, highlighting the urgent need for innovative diagnostic tools and personalized therapeutic approaches tailored to the unique needs of each patient ( 6 , 7 ). Identifying new biomarkers could significantly enhance diagnostic accuracy and enable targeted therapies that address both reproductive and metabolic aspects of the condition, improving outcomes and quality of life for those affected ( 8 , 9 ). Advances in omics technologies offer the potential to identify novel biomarkers and therapeutic targets for PCOS, moving towards personalized medicine approaches. Granulosa cells, essential components of the ovarian follicle, are essential for oocyte development and maturation ( 10 ). These cells not only support oocyte growth but also regulate steroidogenesis and maintain a proper microenvironment for follicular development ( 11 ). Alterations in granulosa cell function can disrupt normal folliculogenesis, potentially leading to ovulatory dysfunction characteristic of PCOS ( 12 ). Emerging evidence has suggested a connection between granulosa cell activity and energy metabolism, emphasizing the sensitivity of these cells to metabolic changes ( 13 ). In women with PCOS, dysregulated energy metabolism pathways in granulosa cells have been implicated in the development of insulin resistance and metabolic disturbances, linking reproductive and metabolic dysfunctions ( 14 ). However, the specific roles of energy metabolism-related genes (ERGs) in the pathophysiology of PCOS remain poorly understood. Understanding the specific ERGs involved in granulosa cell dysfunction could facilitate the development of targeted therapies using cell-based assays and in vitro models. This study aims to use bioinformatics approaches to analyze publicly available data on PCOS, focusing on identifying biomarkers associated with energy metabolism. This study also aims to identify and validate ERG-associated biomarkers with potential for trans-lation into diagnostic tools or therapeutic interventions for PCOS. By exploring the functional roles and regulatory mechanisms of these ERGs, our research seeks to provide a deeper understanding of PCOS pathogenesis and open new avenues for improving diagnosis, prognosis, and personalized treatment strategies for women with this complex disorder.

In conclusion, our study identified and validated three ERG-associated biomarkers (CD44, HS2ST1, and GPC4) in PCOS granulosa cells. The clinical validation of CD44 upregulation further supports its potential as a diagnostic biomarker. These findings provide novel insights into the molecular mechanisms of PCOS and may lead to the development of more effective diagnostic and therapeutic strategies for this complex disorder.

This study identified and validated three ERG-associated biomarkers (CD44, HS2ST1, and GPC4) in PCOS granulosa cells using an integrated bio-informatics and experimental approach. PCOS is the most common reproductive and metabolic disorder, affecting approximately 6-15% of women of reproductive age ( 25 , 6 , 26 ). Previous studies have suggested that the DEG in granulosa cells from PCOS patients and normal controls correlates with oocyte activity to some extent. However, due to the high degree of heterogeneity and numerous subtypes of PCOS, a clear and reliable biomarker for its diagnosis remains elusive. In this study, we identified three potential biomarkers: CD44, HS2ST1, and GPC4. Among these, CD44, a non-kinase cell surface transmembrane glycoprotein, has been extensively studied as a cancer stem cell (CSC) marker in various cancers ( 27 ). It has been shown that elevated CD44 expression significantly contributes to key tumorigenic mechanisms, including cell proliferation, metastasis, invasion, migration, and stemness ( 28 ). More recent findings indicate that CD44, as a widely expressed adhesion molecule, plays a crucial role in promoting pathological angiogenesis and related diseases by regulating endothelial cell functions ( 29 ). Furthermore, CD44 serves as the primary receptor for hyaluronic acid (HA) in the extracellular matrix, with molecular sizes ranging from approximately 80 to 250 kDa across different cell types ( 30 ). The function of CD44 is multifaceted, as it can bind to extracellular matrix components, such as hyaluronic acid (HA) and osteopontin (OPN), as well as to various growth factors, thereby playing a pivotal role in tumor progression. HA, a key component of the cumulus cells’ extracellular matrix, accumulates transiently, leading to the expansion of cumulus-oocyte complexes in preovulatory mammalian follicles ( 31 , 32 ). This interaction is known to suppress apoptosis, promote tumor cell mobility and metastasis, and activate lymphocytes. Notably, prior research has suggested that the interaction between HA and CD44 may also impact fertility and oocyte quality, further underscoring its relevance in reproductive health ( 33 ). The upregulation of CD44 in PCOS granulosa cells observed in our study is consistent with previous reports linking CD44 to cell proliferation and inflammation, both of which are implicated in PCOS pathogenesis. Paravati R et al . ( 34 ) found that insufficient level of OPN-CD44 adhesion complex in endometrial epithelium of ovulatory PCOS patients leads to endometrial infertility. Furthermore, the levels of OPN, CD44, and the inflammatory cytokines TNF-α and IFN-γ exhibit a cyclical pattern in infertile PCOS patients. Specifically, compared to infertility patients, PCOS patients with a secretory endometrium showed reduced levels of OPN and CD44. However, elevated levels of CD44 in serum, as well as OPN and inflammatory cytokines, were observed in the endometrial biopsy medium. This suggests that the alterations in CD44 and OPN in infertility patients are primarily localized to the endometrium. Moreover, it was observed that elevated TNF-α and IFN-γ levels in the blood of PCOS patients seem to directly regulate the expression of CD44 and OPN in the endometrium via the STAT1 and NF-κB pathways ( 35 ). In a related study, elevated levels of OPN and CD44 in biopsy medium and serum inversely correlated with their expression in endometrial tissue ( 36 ). These findings further underscore the complex regulatory mechanisms involved in endometrial dysfunction in PCOS. In experimental models, CD44 deficiency has been shown to alleviate insulin resistance and adipose tissue inflammation in diabetic mice. Treatment with anti-CD44 antibodies also reduced blood glucose levels and adipose tissue macrophage accumulation in mice fed a high-fat diet. Similarly, in humans, CD44 expression was detected in inflammatory cells within obese adipose tissue, and serum CD44 levels were found to be positively correlated with insulin resistance and poor glycemic control. These observations suggest that CD44 plays a key pathogenic role in the development of adipose tissue inflammation and insulin resistance in both rodents and humans ( 37 ). In patients with PCOS, the presence of inflammatory cytokines in the pelvic peritoneal fluid creates a proinflammatory environment in the uterus, which further exacerbates the inflammatory response driven by elevated levels of TNF-α and IFN-γ in the bloodstream. This proinflammatory milieu may also contribute to increased testosterone levels in PCOS patients, which is a hallmark feature of the condition ( 38 ). Heparan sulfate proteoglycans (HSPGs) are glyco-proteins found on cell surfaces and within the extra-cellular matrix (ECM), where they carry unbranched polysaccharide chains known as glycosaminoglycans (GAGs) ( 39 ). Among these, heparan sulfate (HS) is a major component of HSPGs, which are strategically positioned on the cell membrane and play crucial roles in the ECM. The gene HS2ST1 (Heparan sulfate 2-O-sulfotransferase 1), a member of the sulfatase family, encodes a protein responsible for performing a variety of vital biological functions. Specifically, the HS2ST1 enzyme catalyzes the transfer of sulfate groups to the C2 position of the iduronate residue in HS. HSPGs, including HS, are involved in numerous biological processes such as development, cell signal-ing, morphogenesis, cell communication, and tissue repair, underscoring their essential roles in maintain-ing cellular and tissue homeostasis. In recent years, the functional roles of HS2ST1 as a potential drug target and biomarker have garnered significant attention in the biomedical field. For example, studies have shown that dry eye patients exhibit reduced expression of HS2ST1 and two other HS glycotransferase genes in the conjunctiva; however, the underlying mechanism responsible for this downregulation remains unclear ( 40 ). GPC4, a member of the heparan sulfate proteoglycan family, is an adipocytokine that plays key roles in both the cell membrane and extracellular environment ( 41 ). GPC4 is anchored to the outer surface of the cell membrane via glycosylphosphatidylinositol (GPI) and can be cleaved from the membrane into the extracellular space by GPI lipase. This dual localization suggests that GPC4 may be involved in a variety of physiological processes beyond its membrane-bound role. Research has revealed that GPC4 is expressed in both visceral and subcutaneous adipose tissues, with higher expression levels observed in visceral adipose tissue. This difference in expression could help explain the association between increased waist-to-hip ratio (WHR), body mass index (BMI), and the higher incidence of cardiovascular and metabolic diseases in individuals with PCOS ( 42 ). Ussar et al . first demonstrated that glypican-4 can be secreted from adipocytes in mice, with fluctuations in its levels being associated with human insulin sensitivity, confirmed a link between GPC4 levels, BMI, and insulin sensitivity in humans, with notable gender-based variations ( 43 ). In particular, higher levels of GPC4 were observed in women with nonalcoholic fatty liver disease, and these levels were correlated with factors such as body fat distribution, insulin resistance, and arterial stiffness ( 44 ). Additionally, Zheng et al . identified a potential association between GPC4 and insulin resistance in PCOS patients. They found that GPC4 directly interacts with insulin receptors in the liver, skeletal muscle, and adipose tissue, enhancing insulin signaling. Inhibition of insulin-induced phosphorylation of CCAAT/enhancer binding protein-β (C/EBPβ) was shown to impair adipocyte differentiation in vitro. HS2ST1 and GPC4, which were both found to be downregulated in our study, are involved in biological processes such as heparan sulfate biosynthesis and adipocyte insulin sensitivity regulation. Their reduced expression may contribute to the metabolic dysfunction and insulin resistance commonly observed in PCOS, particularly through impaired extracellular matrix signaling and adipokine pathways ( 45 ). Their downregulation may contribute to the metabolic dysfunction observed in PCOS. Although HS2ST1 has not been extensively studied in the context of PCOS, GPC4 has been the subject of several investigations. Furthermore, CD44 has been linked to PCOS research, particularly concerning the endometrium. However, there remains a gap in understanding the roles of these three genes in granulosa cell research. Our preliminary findings indicate differential expression of these genes in PCOS patients, warranting further investigation into their potential involvement in the disease. Our ceRNA network indicated hsa-miR-302a-3p as a key miRNA targeting CD44 and FGD5-AS1. MicroRNAs (miRNAs), which are single-stranded small non-coding RNA molecules, have been increasingly implicated in the pathophysiology of PCOS. The ceRNA network analysis revealed that hsa-miR-302a-3p targets CD44, suggesting a potential mechanism for post-transcriptional regulation of CD44 expression in PCOS. Research has shown that miR-483-5p inhibits the activity of Notch3 and MAPK3 in human granulosa cumulus cells by directly interacting with the 3’ untranslated regions (3’UTRs) of Notch3 and MAPK3 mRNA ( 46 ). Similarly, a study conducted by Lauren W et.al , involving microarray analysis and PCR validation of human follicle fluid, revealed a significant upregulation of five specific miRNAs (miR-32, miR-34c, miR-135a, miR-18b, and miR-9) in the PCOS group ( 47 ). In addition to identifying diagnostic biomarkers, we explored the potential pharmacological relevance of CD44, HS2ST1, and GPC4. Among the predicted compounds from the CTD database, bisphenol A was associated with all three biomarkers. This finding suggests that BPA, an endocrine-disrupting chemical widely present in plastics, may contribute to the development or exacerbation of PCOS through its regulatory effects on energy metabolism-related genes. The identification of bisphenol A as a shared modulator highlights a potential environmental factor in PCOS pathogenesis. Furthermore, several other compounds predicted to interact with the biomarkers may hold therapeutic promise. Further studies are warranted to investigate the potential of these candidate agents to modulate gene expression and improve metabolic and reproductive outcomes in PCOS. The subcellular localization analysis revealed that CD44 is predominantly expressed on the plasma membrane, HS2ST1 in the Golgi apparatus, and GPC4 in both the endosome and plasma membrane. Functionally, CD44 was strongly associated with the toll-like receptor signaling pathway, aligning with its known role in inflammation and innate immune response. Both CD44 and HS2ST1 were enriched in pathways related to IgA production and asthma, whereas HS2ST1 and GPC4 were linked to antigen processing and presentation. Notably, GPC4 downregulation was associated with suppression of B-cell receptor signaling, suggesting impaired humoral immune responses in PCOS. Our GSVA and ssGSEA analyses further supported immune dysregulation: 19 immune cell types exhibited significantly higher enrichment scores in the PCOS group, and Spearman correlation analysis demonstrated that CD44 positively correlated, while HS2ST1 and GPC4 negatively correlated, with immune infiltration. These findings reinforce the immunometabolic roles of these biomarkers in PCOS granulosa cells. From a pharmacological perspective, the identification of bisphenol A (BPA) as a compound associated with all three biomarkers suggests a potential en-vironmental contributor to PCOS pathogenesis. BPA is a well-known endocrine disruptor that may alter energy metabolism and immune signaling by modulating genes such as CD44, HS2ST1, and GPC4. Further studies are needed to evaluate whether BPA exposure directly contributes to PCOS development or exacerbation. Clinically, our RT-qPCR results confirmed significant CD44 upregulation in PCOS granulosa cells (P = 0.0079), consistent with its role in inflammation and insulin resistance. Although GPC4 and HS2ST1 did not reach statistical significance (P = 0.1306 and P = 0.2945, respectively), moderate effect size differences and their known roles in insulin signaling and heparan sulfate modification suggest functional relevance that warrants further investigation in larger cohorts. Overall, CD44 emerges as a strong candidate for further mechanistic exploration, while GPC4 may represent a promising metabolic biomarker requiring validation. Our findings should also be considered in the context of broader research on potential therapeutic targets for PCOS. In a bioinformatics-based analysis, Liu et al . (2024) identified several genes as potential therapeutic targets for PCOS, emphasizing the importance of a systems biology approach to understanding this complex disorder ( 48 ). While our study focused on ERGs, future research could explore the overlap between our identified biomarkers and the targets identified by Liu et al . (2024) to identify synergistic therapeutic opportunities. Furthermore, considering the metabolic aspects of PCOS, the impact of metformin on visfatin levels, as highlighted by Reddy et al . (2025), provides a crucial clinical perspective ( 49 ). Although our study did not directly investigate the effects of metformin, the link between ERGs, metabolic dysregulation, and potential therapeutic interventions like metformin underscores the need for integrated approaches. Given that PCOS is associated with an increased risk of ovarian and breast malignancies, Chen et al . (2023) emphasizes the importance of long-term monitoring and risk management in women with PCOS ( 50 ). While our study primarily focused on the identification of diagnostic biomarkers, further research could investigate the link between ERGs, cancer-related pathways, and the potential for targeted prevention strategies in PCOS patients. However, several limitations exist within this study. First, the sample size of the public transcriptomic datasets ( GSE34526 and GSE193123 ) was relatively small, which may limit the statistical power and generalizability of the bioinformatics findings. Second, although we performed RT-qPCR validation using granulosa cells from PCOS patients and controls, the sample sizes for GPC4 (n = 16/14) and HS2ST1 (n = 15/13) were modest, potentially underpowered to detect subtle but biologically meaningful differences. This limitation may explain the lack of statistical significance despite moderate effect sizes and biological plausibility. Therefore, we acknowledge that the current clinical validation is preliminary and requires expansion. Future studies with larger and multicenter clinical cohorts are planned to confirm these findings and assess the robustness of biomarker expression across different PCOS phenotypes and populations. Additionally, single-cell or spatial transcriptomic technologies may be employed to validate gene expression in situ and further elucidate the granulosa cell-specific roles of these biomarkers in vivo. Ultimately, prospective clinical trials will be necessary to confirm the diagnostic and therapeutic utility of these targets.

The study aimed to investigate the role of energy metabolism-related genes (ERGs) in polycystic ovary syndrome (PCOS) by leveraging bioinformatics approaches to identify key biomarkers and unravel their regulatory mechanisms. Specifically, the objectives include: 1) analyzing gene expression datasets ( GSE34526 and GSE193123 ) to identify differentially expressed genes (DEGs) in PCOS granulosa cells; 2) intersecting common DEGs with ERGs from Reactome to prioritize candidate biomarkers; 3) characterizing their functional roles through enrichment analysis and subcellular localization predictions; 4) mapping transcriptional (X2K) and post-transcriptional (ceRNA) regulatory networks to uncover molecular mechanisms; and 5) predicting potential therapeutic compounds targeting these biomarkers. By integrating these analyses, the research seeks to validate novel ERG-associated biomarkers (CD44, HS2ST1, GPC4) and provide mechanistic insights into PCOS pathogenesis, ultimately contributing to improved diagnostic strategies and precision medicine approaches for metabolic and reproductive dysfunction in PCOS. Additionally, the study aims to validate the bioinformatics findings through clinical sample analysis. Specifically, granulosa cells collected from PCOS patients and healthy controls will undergo RT-qPCR to quantify CD44 expression levels. This experimental validation will directly confirm the upregulation of CD44 identified in silico, bridging computational predictions with clinical evidence. By integrating transcriptomic data analysis and wet-lab verification, the research seeks to establish CD44 as a robust candidate biomarker for PCOS pathogenesis and provide empirical support for its functional role in metabolic and reproductive dysfunction.

Two datasets ( GSE34526 and GSE193123 ) utilized in this study were retrieved from the GEO database. GSE34526 ( GPL570 ) served as Training Set 1, comprising 7 PCOS and 3 control granulosa cell samples. GSE193123 ( GPL24676 ) was used as Training Set 2, with 3 PCOS and 3 control granulosa cell samples. ERGs were acquired from the Reactome database ( 15 ). These datasets were selected based on two main criteria: 1) availability of granulosa cell transcriptomic profiles from PCOS patients and healthy controls, and 2) adequate sample quality and annotation for differential expression analysis. Differentially expressed genes (DEGs) between PCOS and control samples were identified using Limma (v3.52.4) for GSE34526 and DESeq2 (v1.38.3) for GSE193123 , with thresholds of |log2FC| ≥ 0.5 and P < 0.05 ( 16 , 17 ). Overlapping DEGs (co-DEGs) were further analyzed using Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses via the clusterProfiler package (v4.4.4) ( 18 ) ( P < 0.05) to determine the biological functions of the co-DEGs. To identify energy metabolism-related biomarker genes, we first intersected the list of co-DEGs from both datasets with a curated set of ERGs obtained from the Reactome Pathway Database (accessed January 2024). Overlapping genes were determined based on matching official gene symbols, ensuring consistency across platforms and annotations. The DE-ERGs were identified as potential biomarkers by overlapping co-DEGs with ERGs. The expression levels of these biomarkers were analyzed in both the PCOS and control groups across the two training datasets. Subsequently, receiver operating characteristic (ROC) curves were generated, and the area under the curve (AUC) values were calculated to assess the diagnostic performance of the biomarkers in the two training datasets, using the R package survivalROC (v1.0.3) ( 19 ). To explore the biological roles of candidate biomarkers, we constructed a gene-gene interaction (GGI) network using the GeneMANIA plugin ( 20 ) in Cytoscape (v3.9.1). The network was generated with default parameters, integrating multiple interaction types including co-expression, co-localization, genetic inter-actions, and shared pathways, with equal weights assigned to each data source. Subcellular localization of biomarkers was predicted using Cell-PLoc 3.0 ( 21 ), a multi-label predictor based on functional domain composition and GO information. Default settings were used with input sequences derived from UniProt human protein entries, ensuring high specificity for eukaryotic localization patterns. Additionally, Gene Set Enrichment Analysis (GSEA) was performed based on the Spearman correlation of biomarkers and other genes referring to KEGG gene set from the Molecular Signatures Database (MSigDB) (P < 0.05) in GSE34526 . Gene Set Variation Analysis (GSVA) was performed by GSVA package (v1.49.8) ( 22 ) to detect differentially functional gene sets with KEGG gene sets as background set (|t| ≥ 2 and P < 0.05) in GSE34526 . In addition, we applied Gene Set Variation Analysis (GSVA) using the GSVA package (v1.49.8) to evaluate the sample-wise enrichment of KEGG pathways in GSE34526 . Unlike GSEA, which assesses enrichment at the group level, GSVA provides an unsupervised, non-parametric method to estimate pathway variation at the individual sample level, enabling better characterization of subtle functional differences in small sample sizes and heterogenous groups. This dual approach ensured robustness and cross-validation of functional enrichment findings. We performed an expression2 kinase (X2K) analysis to investigate the interactions between transcription factors (TFs), kinases, and related proteins ( 23 ). The top 10 TFs, kinases, and their associated proteins were included in the X2K network. Additionally, miRNAs and lncRNAs associated with biomarkers were predicted using the Starbase database, with miRNA prediction scores ≥ 5 (from miRanda and PITA) and lncRNA scores ≥ 20, forming a competing endogenous RNA (ceRNA) network. Finally, compounds related to the biomarkers were identified from the Comparative Toxicogenomics Database (CTD) to construct a gene-compound network. Compounds were selected based on their ability to (1) downregulate upregulated genes, (2) upregulate downregulated genes, or (3) be associated with genes in multiple studies. Single sample gene set enrichment analysis (ssGSEA), which is designed for single sample cannot do GSEA, can be implemented by the GSVA R package. ssGSEA was used to analyze the samples in the training set, and the immune-related genes provided in the literature were used as the background gene set to obtain the enrichment scores of 28 immune cells. To compare immune cell enrichment between PCOS and control groups, we used the Wilcoxon rank-sum test with continuity correction, and adjusted P-values were reported for each immune cell type. Immune cell types with P < 0.05 were considered significantly differentially enriched. Correlation analyses between biomarker expression and immune cell infiltration levels were conducted using Spearman’s correlation coefficient, and correlation heatmaps were generated to illustrate the results. Nineteen immune cells were negatively correlated with HS2ST1 and GPC4, and positively correlated with CD44. All analyses were performed using the R (V4.2.2), and network visualization was conducted using Cytoscape (v3.9.1,) ( 24 ). P < 0.05 was considered statistically significant. The following R packages were used throughout the study: limma (v3.52.4) and DESeq2 (v1.38.3) for differential gene expression analysis; clusterProfiler (v4.4.4) for functional enrichment analysis (GO and KEGG); GSVA (v1.49.8) for ssGSEA and GSVA; survivalROC (v1.0.3) for ROC curve and AUC calculation; ggplot2 (v3.4.2), pheatmap (v1.0.12), and corrplot (v0.92) for data visualization; psych (v2.3.6) and stats (base R) for correlation and statistical tests. This study collected granulosa cells from 60 infertile women undergoing routine IVF-ET treatment at our center between October 2020 and (Month) 2021, with clinical data recorded (including age, BMI, basal hormone levels). All participants underwent controlled ovarian hyperstimulation using a long-acting follicular phase long protocol. The PCOS group inclusion criteria were defined as follows: Women diagnosed with PCOS according to the 2003 Rotterdam diagnostic consensus, meeting at least two of the following three criteria: 1) oligo- or anovulation; 2) clinical and/or biochemical hyperandrogenism; 3) polycystic ovarian morphology (ovarian volume >10 mL and/or ≥12 follicles measuring 2–9 mm in diameter per ovarian cross-section), with exclusion of other disorders mimicking PCOS phenotypes. This study collected discarded follicular fluid and supernatant from the first single-follicle puncture during oocyte retrieval in PCOS patients and control subjects undergoing assisted reproductive therapy at our gynecological reproductive center between November 2023 and February 2025. A total of 60 cases were ultimately included for analysis. The control group inclusion criteria were: 1) Infertility due to tubal or male factors; 2) Regular ovulatory cycles. Exclusion criteria included: 1) Female age >35 years; 2) Premature ovarian insufficiency (POI); 3) Comorbid endocrine disorders or infectious diseases in females; 4) Endometriosis or adenomyosis; 5) History of ovarian surgery, radiotherapy, or chemotherapy; 6) Autoimmune diseases. This study was approved by the Ethics Committee of the First Affiliated Hospital of Sun Yat-sen University. Granulosa cells were isolated using density gradient centrifugation as follows: 1. Transfer follicular fluid collected from culture flasks into 1–3 labeled 50 mL sterile nuclease-free centrifuge tubes. Balance and centrifuge at 2000 rpm for 15 min at 4 °C. 2. Carefully discard the supernatant. Combine the precipitated follicular fluid mixtures from the same patient into a single tube. Add PBS to the pellet to a total volume of 7–8 mL and mix thoroughly. 3. Prepare a new 15 mL centrifuge tube with 10 mL of 50% Percoll separation medium. Using a Pasteur pipette, slowly layer the cell suspension from step 2) onto the Percoll medium along the tube wall to form a distinct interface. Balance and centrifuge at 2500 rpm for 25 min at room temperature. 4. After centrifugation, distinct layers will form: an upper PBS layer, a narrow cloudy granulosa cell layer, a transparent separation medium layer, and a bottom erythrocyte layer. Carefully aspirate the granulosa cell layer into a new sterile nuclease-free centrifuge tube. Wash the cell pellet twice with PBS, followed by centrifugation at 1500 rpm for 5 min at 4 °C. 5. Discard the supernatant, resuspend the pellet in 1× PBS, and homogenize by gentle pipetting. Filter the suspension through a 40 μm cell strainer. Collect the filtrate and centrifuge at 1500 rpm for 5 min at 4 °C. 6. Discard the supernatant and add 2–3 mL of red blood cell lysis buffer. Mix gently and incubate at 4 °C for 10 min. Centrifuge at 1500 rpm for 5 min at 4 °C. 7. After washing, discard the supernatant. The white cellular pellet at the tube bottom represents granulosa cells. Add 1 mL of RNAiso Plus or 200 μL of RIPA protein lysis buffer and store at −80 °C for subsequent use. Prepare TRIzol reagent, PBS, isopropanol, nuclease-free water, chloroform substitute, 75% ethanol, 1.5 mL sterile nuclease-free tubes, and nuclease-free pipette tips in advance. Pre-cool the centrifuge to 4 °C. Thaw frozen cells, wash gently twice with ice-cold PBS, and lyse cells in 1 mL TRIzol per well by pipetting 3–5 times. Transfer the lysate to a nuclease-free EP tube and incubate on ice for 5 min. Add 200 μL chloroform substitute, invert vigorously for 1 min until emulsified, and incubate on ice for 15 min. Centrifuge at 12,000 rpm for 15 min at 4 °C. Three phases will form: a colorless aqueous phase (top), a white protein interphase, and a colored organic phase (bottom). Transfer 400 μL of the aqueous phase to a new EP tube. Add an equal volume of pre-cooled isopropanol, invert to mix, and incubate on ice for 10 min. Centrifuge at 12,000 rpm for 10 min at 4 °C. Discard the supernatant, wash the RNA pellet with 1 mL 75% ethanol, and centrifuge at 7500 rpm for 10 min. Repeat washing 2–3 times. Air-dry the pellet on ice, resuspend in 20 μL nuclease-free water, and measure RNA concentration and purity. Total RNA was extracted from granulosa cells using TRIzol Reagent (Invitrogen, USA) following the manufacturer’s protocol with minor optimizations to ensure RNA purity and integrity. Reverse transcription was performed using the One-Step gDNA Removal and HiScript III Reverse Transcriptase Kit (Nanjing Novizan Biotechnology Co., Ltd.) with RT-qPCR-specific premix. Prepare the reverse transcription reaction system as shown in the table below: To formulate a retrotranscriptional response system ( Table 1 ): Reverse transcriptional response system A total of 6 pairs of primers were obtained by consulting literature, browsing PrimerBank, GEO database, and using Primer6 and other primer design software. According to the product description of biotechnology company, the primers were diluted 10 times, 100 times and 1000 times to calculate the amplification efficiency of the primers. Finally, primers with stable efficiency between 95% and 100% were used for qPCR reaction. The primers used are shown in Table 2 . The internal reference gene was GAPDH. MelloPrimer sequences The primer is synthesized by Shenggong Co., LTD., and the specific sequence is as follows ( Table 2 ): The Real Time PCR reaction was performed, and the standard procedure of two-step PCR amplification was used according to the manufacturer’s instructions. The reaction conditions were as follows: 1) Predenatura-tion: 95 °C,30 s; 2) PCR reaction was repeated 40 cycles at 95 °C for 5 s and 60 °C for 30 s; 3) Melting curve: 95 °C, 15 s; 60 °C, 1 min; 95 °C for 15 s. Reaction product specificity was determined by melting curves. . Real-time PCR reaction was performed using TB Green® Premix Ex Taq™ II (Tli RNaseH Plus) kit with 20 μL reaction system configured as follows ( Table 3 ): Real-time fluorescent quantitative PCR configuration system After gently blowing and mixing with pipette, PCR reaction was carried out: pre-denaturation at 95 °C for 30 s, and then reaction program was set for 10 s at 95 °C and 30 s at 60 °C. A total of 40 cycles were carried out. Finally, the dissolution curve was collected: reaction at 95 °C for 15 s, reaction at 60 °C for 1 minute, reaction at 95 °C for 15 s. In this study, BM SPSS software (version 25.0, IBM Corp, Armonk, NY, USA) was used for baseline data statistical analysis. Continuous variables with normal distribution were expressed as mean (standard deviation), and continuous variables without normal distribution were expressed as median (interquartile range). Categorical variables were expressed as frequencies and percentages. Data distribution was assessed using normality tests, and appropriate descriptive statistics were used for normally and non-normally distributed variables, respectively. Normally distributed continuous variables were compared between groups using the Welch t test, and non-normally distributed continuous variables were compared using the Wilcoxon rank-sum test. For between-group comparisons of categorical variables, Fisher’s exact test was used when the expected frequency was less than 5, and chi-square test was used otherwise. (Statistic staststic, then t test shows t value, ANOVA shows F value, chi-square test shows chi square value, Fisher’s test does not show any statistic (shown in blank), Wilcoxon rank sum test shows W value. The qRT-PCR results were analyzed using GraphPad Prism8 (GraphPad Software, Boston, USA): one-way repeated measures ANOVA was used to analyze the mRNA expression levels of differentially expressed genes in granulosa cells obtained from patients with polycystic ovary and illuminated patients. RT-qPCR used GAPDH gene as internal reference, and used 2-∆∆C value method to quantitatively calculate the relative GAPDH of target gene.