Identification and validation of ferroptosis-related key hub genes linking hypoxia and osteoporosis Authors: Yifan Gu, Qingsong Gu, Yiheng Li, Yuhu Chen, Yicong Wang, Jianqiao Li, Linhui Wang, Ziru Wang, Min Yang Published: 2025 ABSTRACT BackgroundOsteoporosis (OP) is a prevalent progressive metabolic bone disease in the middle-aged and elderly population. Previous research has indicated that physiological or pathological hypoxia may disrupt bone metabolism, ultimately contributing to OP development. However, the specific pathophysiological mechanisms remain unknown. Ferroptosis is a novel iron-dependent non-apoptotic cell death pathway. The objective of this investigation was to use bioinformatics analysis to identify key hub genes related to ferroptosis that may be linked to hypoxia and osteoporosis. Subsequently, we utilized experimental methods to validate these findings. MethodsDifferentially expressed genes (DEGs) in bone marrow mesenchymal stem cells (BMSCs) from primary osteoporosis patients and hypoxia-treated BMSCs were screened from the Gene Expression Omnibus (GEO) database. These DEGs were then intersected with the ferroptosis dataset (FerrDb), resulting in FerrDEGs. The potential functional roles of FerrDEGs were investigated through gene ontology (GO) and pathway enrichment analysis (e.g., KEGG, WikiPathways and Reactome). The STRING database was used to analyze protein-protein interaction networks, pinpointing hub genes within FerrDEGs. To further assess the predictive value of these hub genes, we conducted ROC analysis using the data from GSE230665 to identify key hub genes. Subsequently, NetworkAnalyst was employed to establish networks between target microRNAs(miRNAs) and key hub genes. Finally, we validated the key hub genes and miRNAs in an OVX rat model for further confirmation.ResultsA total of 50 FerrDEGs were identified in the overlap of hypoxia, OP, and ferroptosis pathways. These FerrDEGs were primarily associated with biological processes related to cellular response to chemical stress and oxidative stress. Pathway enrichment analysis revealed that these FerrDEGs played significant roles in processes such as viral carcinogenesis and epstein-barr virus infection. Multiple algorithms (CytoHubba, CytoNCA and Analyze Network) were employed to identify eight hub genes within FerrDEGs, including TP53, JUN​, SQSTM1, STAT3​, CDKN1A​​​, CAV1, CD44 and TGFBR1. Five key hub genes (JUN, SQSTM1, STAT3, CD44 and TGFBR1) were further screened by ROC analysis. Subsequently, hsa-miR-20a-5p was identified as a potential key microRNA through the construction of the miRNA-gene network. The five key hub genes and rno-miR-20a-5p were further validated in the OVX rat model.ConclusionIn summary, we first screened ferroptosis-related key hub genes linking hypoxia and osteoporosis. The findings suggest that JUN, SQSTM1​, STAT3​ ​, CD44​​ and TGFBR1 are significantly associated with OP and hypoxia, potentially serving as biomarkers for diseases linked to ferroptosis. Additionally, hsa-miR-20a-5p was identified as a crucial upstream regulator likely involved in the regulation of these genes simultaneously.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12891-025-09326-z. Introduction Osteoporosis (OP) is a chronic metabolic disease influenced by genetic, nutritional, and hormonal factors that leads to alterations in bone microstructure, decreased deposition of bone minerals and matrix components, and ultimately increased bone fragility [1]. Over the past 20 years, the global prevalence of osteoporosis has been about 18.3%, with about 23.1% for women and 11.7% for men [2]. Current clinical treatments for osteoporosis include lifestyle changes, calcium and vitamin D supplementation, and the use of anti-osteoporosis medications. However, the fracture rate and risk of re-fracture in osteoporosis patients remain high [3]. The etiology of osteoporosis is complex, and many factors have been found to be closely associated with it, but the specific mechanisms have not yet been elucidated [4]. Osteoporosis and its complications have become a worldwide health crisis. Studies have found that environmental or pathologic hypoxia due to high-altitude environments [5], obstructive sleep apnea syndrome [6], and a variety of respiratory disorders (including chronic obstructive pulmonary disease, asthma, and COVID-19) [7] are associated with osteoporosis. Hypoxia leads to imbalances in bone metabolism, inflammatory responses, and insufficient angiogenesis, which are closely associated with the development of osteoporosis [8]. Among them, the imbalance in bone metabolism mediated by osteoblasts and osteoclasts may be associated with the hypoxia-induced accumulation of iron ions, increased generation of reactive oxygen species (ROS), and mitochondrial dysfunction. This process is also referred to as ferroptosis [9]. Ferroptosis is a novel iron-dependent non-apoptotic cell death mode characterized by accumulation of lipid reactive oxygen species, which is distinct from apoptosis, necrosis and autophagy [10]. Iron overload and ROS accumulation are the main features of ferroptosis, and lipid peroxidation is a key process [11]. When there is iron overload, excess free Fe2+ surpasses the buffering capacity of ferritin and generates Fe3+ and a large amount of ROS through the Fenton reaction, thereby impacting a series of signaling pathways [12]. Many studies have suggested that ferroptosis may be a promising line of research for the prevention and treatment of OP [13–15]. Investigating the correlation between hypoxia, osteoporosis, and ferroptosis presents a promising research direction. Delving into the biological interconnections and underlying molecular mechanisms among these entities holds potential for shedding light on the pathogenesis of OP and unveiling prospective therapeutic strategies. In recent years, microarray technology and bioinformatics analysis have been widely used for screening gene alterations at the genomic level. Bioinformatics analysis of ferroptosis-related key hub genes linking hypoxia and osteoporosis have not been reported to date. In the present study, we identified DEGs in BMSCs from primary osteoporosis patients and BMSCs treated with hypoxia from the GEO database. We then overlapped OP-DEGs, hypoxia-DEGs, and FerrDb to obtain FerrDEGs. Through GO and pathway enrichment analysis, we examined the biological functions and pathways of these FerrDEGs. Subsequently, hub genes were determined using various algorithms on a PPI network followed by further screening out key hub genes through ROC analysis. Predictions were made for potential gene-miRNA interactions associated with these key hub genes. Finally, we validated these key hub genes as well as the top-scoring miRNA using an OVX animal model. The sequential workflow of our research is presented in Fig. 1. Fig. 1Schematic illustration of the overall general workflow of this study Schematic illustration of the overall general workflow of this study Materials and methods Collection of the datasets The relevant datasets were obtained from the Gene Expression Omnibus (GEO; http://www.ncbi.nlm.nih.gov/geo). The GEO accession ID for the OP dataset was GSE35958, which includes transcriptional profiling of 5 OP samples and 4 controls collected from BMSCs. Additionally, another OP dataset for ROC analysis validation, GSE230665, was derived from femoral tissue, which contains 12 OP patient samples and 3 controls. The dataset for this study on hypoxia was created by merging two microarray datasets (GEO accession ID: GSE32006 and GSE55875). The samples were derived from human BMSCs and consisted of 6 hypoxia-treated samples and 6 controls. Table 1 provides a detailed table of sample parameters. A post-hoc power analysis was conducted using the ‘pwr’ R package (v1.3-0) to assess the statistical power of the differential expression analysis given the limited sample sizes. Table 1Details of sample parametersGEO AccessionGEO PlatformSamplesPublication DateOPGSE35958GPL5705 OP and 4 healthFeb 22, 2012GSE230665GPL1033212 OP and 3 healthApr 29, 2023HypoxiaGSE32006GPL145503 hypoxia and 3 normoxiaNov 11, 2011GSE55875GPL145523 hypoxia and 3 normoxiaMar 14, 2014 Table 1Details of sample parametersGEO AccessionGEO PlatformSamplesPublication DateOPGSE35958GPL5705 OP and 4 healthFeb 22, 2012GSE230665GPL1033212 OP and 3 healthApr 29, 2023HypoxiaGSE32006GPL145503 hypoxia and 3 normoxiaNov 11, 2011GSE55875GPL145523 hypoxia and 3 normoxiaMar 14, 2014 Details of sample parameters Differentially expressed genes selection The hypoxia dataset was obtained by combining gene expression profiles from GSE32006 and GSE55875 based on clinical information. To minimize bias introduced by different batches of samples, batch correction of the combined data was performed using the ComBat method in the R software (version 4.3.2) ‘sva’ package (version 3.50.0). After batch correction using ComBat, the merged dataset was normalized using the ‘normalizeBetweenArrays’ function (limma package) with the quantile method. Probes were mapped to official gene symbols based on GPL platform annotation files. For probes mapping to the same gene, the median expression value was retained. ComBat parameters were set as: batch = GSE series, mod = NULL. Differential expression analysis (DEA) was performed on the hypoxia dataset and the OP dataset using the ‘limma’ software package (version 3.58.1). DEGs were filtered using the thresholds: |log2FC| > log2(1.2) and adjusted p-value (Benjamini-Hochberg FDR) < 0.05. Gene Set Enrichment Analysis (GSEA) was performed to assess the involvement of DEGs in the ferroptosis pathway. Additionally, we used the Ferroptosis Database (FerrDb V2; http://www.zhounan.org/ferrdb/) to obtain a dataset of 621 genes, including 264 driver genes, 238 suppressor genes, 9 marker genes, and 110 unclassified genes [16]. The Venn analysis of the ‘VennDiagram’ package (version 1.3.3) in the R software was used to overlay the OP-DEGs, Hypoxia-DEGs, and FerrDb, resulting in the identification of FerrDEGs. Functional enrichment analysis and visualization Gene enrichment analysis involves analysing a target genome to gain insight into the general biological functions of genes and chromosomal locations. A common and useful method for functional enrichment analysis is gene ontology (GO) analysis, which can be categorised into biological process (BP), cellular composition (CC), and molecular function (MF) [17]. The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway is utilised for metabolic pathway enrichment analysis and is widely used in genomic analysis [18]. We conducted GO and pathway enrichment analysis (e.g., KEGG, WikiPathways and Reactome) using the online analysis platform Metascape (http://metascape.org) [19] to investigate the functional roles of FerrDEGs. Metascape was used with its default background set (all protein-coding genes in the genome). Following this, we conducted molecular complex detection (MCODE) to identify interrelated central genes. The close interactions between each MCODE network and molecule were indicated using different colours. The networks ranged in size from 3 to 500, with a minimum and maximum size. Protein-protein interaction networks and hub genes extraction Using the online tool STRING (version 12.0, https://www.string-db.org/), a protein PPI network for FerrDEGs was constructed. This tool is designed to integrate all known and predicted associations between proteins, including physical interactions and functional associations. The current PPI network also uses a confidence score of >0.400. The results of the PPI analysis were imported into Cytoscape software (version 3.10.1, https://cytoscape.org/) to visualize the network. Nodes represent genes and edges represent connections between genes. Genes were sorted by size based on the Betweenness Centrality (BC) score analyzed by the CytoNCA plugin. The CytoHubba plugin was used to rank genes by a weighted score calculated using the Maximal Clique Centrality (MCC) algorithm, which has been demonstrated to achieve the highest accuracy in identifying topologically significant nodes among CytoHubba’s methods [20]. The hub genes are obtained by taking the intersection of the top 20 genes with the highest BC scores, the top 10 genes with the highest MCC scores, and the top 10 genes with the highest node degrees, based on multiple algorithms. Correlation analysis between hub genes To evaluate the relationship between the hub genes, we conducted correlation analysis using Pearson correlation. The R-value, which indicates the correlation coefficient and effect size, was used in the analysis. A correlation matrix heatmap was generated using the ‘corrplot’ R package. ROC analysis of the hub genes ROC analysis was performed using the R software package “pROC” (version 1.18.5) to obtain the area under the curve (AUC) values. The gene expression of the corresponding hub genes was obtained from the GEO dataset GSE230665. The ‘roc’ function of ‘pROC’ was used for ROC analysis, and the ‘ci’ function was used to evaluate AUC and confidence interval. The ROC curve plots sensitivity values on the Y-axis and false positive rates (1-specificity) values on the X-axis. Genes with AUC values > = 0.75 in the ROC curves are considered to have sufficient predictive value and are therefore considered key hub genes. Prediction of MiRNAs MicroRNAs (miRNAs) are crucial for molecular research. We used the NetworkAnalyst platform (https://www.networkanalyst.ca/) to construct key gene-miRNA regulatory networks for predictive analysis of associated miRNAs. NetworkAnalyst is an extensive online platform for meta-analysis of gene expression data, providing insight into biological mechanisms, roles, and interpretations. The key gene-miRNA network is based on the miRTarBase database (version 8.0). OVX model construction and cell culture Ten female Sprague-Dawley rats weighing approximately 230–250 g, provided by Anhui Provincial Laboratory Animal Center, were selected to prepare an experimental model of osteoporosis. This study was conducted in compliance with the ARRIVE guidelines (Animal Research: Reporting of In Vivo Experiments). Animals were randomly assigned to groups using a random number table. Both surgical procedures and data analysis were performed in a blinded manner. Based on preliminary experiments (effect size d = 1.8, α = 0.05, power = 0.8), the minimum sample size was determined using G*Power software, resulting in n = 5 per group. A total of 10 rats were included (5 in the OVX group and 5 in the Sham group). Anesthesia was induced via intraperitoneal injection of sodium pentobarbital (40 mg/kg). Postoperative analgesia was maintained with subcutaneous injections of buprenorphine (0.1 mg/kg) every 12 h for 48 h. Humane endpoints were defined as a greater than 30% decrease in femoral BMD or greater than 20% loss of body weight. After three months, blood was collected from the abdominal aorta and serum was obtained by centrifugation at 2000 r/m for 15 min. The serum was then frozen at −80 °C for future analysis. Micro-CT examination was performed on the left femur. Bone marrow stromal cells (BMSCs) were extracted from the right femur and cultured at 37 °C in a 5% CO2 incubator. For later study, third-generation cells were used. Micro-CT Following a 48-hour fixation in 4% paraformaldehyde, micro-CT scans were taken of the left femur samples and three-dimensional histomorphometry images were constructed using software. Micro-CT scanning was performed using the following parameters: voltage 50 kV, current 200 µA, resolution 10 μm, with calibration against hydroxyapatite standards. The osteoporosis model was assessed using bone mineral density (BMD), bone volume (BV), trabecular number (Tb.N), bone volume/tissue volume (BV/TV), and trabecular spacing (Tb.sp). Lipid peroxidation assay The Lipid Peroxidation Malondialdehyde (MDA) Assay Kit (Beyotime, China) was used to determine lipid peroxidation levels. The MDA assay can be performed directly on serum. To prepare BMSCs for the assay, they were homogenized with cell lysis buffer on ice and the resulting supernatant was collected by centrifugation at 12,000 ×g for 10 min. To determine the MDA content, 100 µl of the sample was mixed with 200 µl of MDA assay work solution in a centrifuge tube. The mixture was heated in boiling water for 15 min and then cooled to room temperature in a water bath. After centrifuging at 1000×g for 10 min at room temperature, 200 µl of the supernatant was transferred to a 96-well plate. The absorbance was measured at 532 nm using an enzyme marker. The MDA content was determined using the standard curve. Three repetitions were set for the tests, with a coefficient of variation (CV) within the batch being less than 5%. Iron level measurement Serum iron levels were measured using the Serum Ferri Ion Content Assay Kit (Solarbio, China). To determine iron levels, 100 µl of serum was mixed with 250 µl of the working solution and incubated for 5 min in a boiling water bath. The samples were then cooled to room temperature in a water bath, mixed with 62 µl of chloroform, and centrifuged at 10,000×g for 10 min at room temperature. 210 µl of supernatant from each sample was pipetted into a 96-well plate, and the absorbance was measured at 520 nm. The Cell Iron Content Assay Kit (Solarbio, China) was used to measure the iron level in cells. The BMSCs were homogenized in cell lysis buffer on ice and then centrifuged at 12,000×g for 10 min. Next, 20 µl of supernatant was collected and mixed with 180 µl of working buffer for 10 min at 25 °C. Finally, the absorbance was measured at 510 nm. Iron ion levels in serum and cells were determined using standard curves. Three repetitions were set for the tests, with a CV within the batch being less than 5%. RT-qPCR Total RNA and miRNA were isolated from BMSCs using the Animal RNA Isolation Kit (Beyotime, China) and Small RNA Isolation Kit (Beyotime, China), respectively. RT-qPCR was performed using the SYBR Green One-Step qRT-PCR Kit (Beyotime, China) for mRNA and miRNA detection. The expression of target genes was analyzed using the 2−ΔΔCt method. Primers for rno-miR-20a-5p and U6 were purchased from RiboBio Co., while other relevant primers are listed in Supplementary Table S1. Statistical analysis The mean ± standard deviation was used to express quantitative data. The t-test was used to compare differences between the two groups. For all quantitative analyses, each test was repeated five times. A statistically significant result was defined as P Value < 0.05. Results Identification of FerrDEGs DEGs in OP and control were screened based on the GSE35958 dataset. The MA plot (Fig. 2A), volcano plot and heatmap (Supplementary Figure S1) revealed a total of 4642 DEGs, with 2362 genes down-regulated and 2280 genes up-regulated. PCA and RLE analyses were conducted to identify dimensions and screen samples based on quality control criteria and normalization of raw data. After batch correction, the differences between hypoxia datasets were significantly reduced (Supplementary Figure S2A). The RLE plots indicated acceptable normalization (Supplementary Figure S2B). Based on the combined dataset of GSE55875 and GSE32006, the DEGs in the hypoxic and normoxic groups were screened. The volcano plot (Fig. 2B) and heatmap (Supplementary Figure S1B) showed a total of 4949 DEGs, including 2585 down-regulated genes and 2364 up-regulated genes. GSEA revealed significant enrichment of DEGs from both the OP and hypoxia datasets in the ferroptosis pathway (OP: NES = 1.13, FDR = 0.122, Fig. 2C; Hypoxia: NES = 1.17, FDR = 0.064, Fig. 2D. Among the OP DEGs, Hypo DEGs and FerrDb, the 50 FerrDEGs were then identified. These results suggest that there are mechanistic commonalities and interactions between OP, hypoxia, and ferroptosis (Fig. 2E).Fig. 2A-B MA plots of DEGs. C-D Results of GSEA analysis showing enrichment in ferroptosis pathways. E The Venn diagram illustrates the common DEGs among OP, hypoxia, and ferroptosis A-B MA plots of DEGs. C-D Results of GSEA analysis showing enrichment in ferroptosis pathways. E The Venn diagram illustrates the common DEGs among OP, hypoxia, and ferroptosis Enrichment analysis of FerrDEGs GO and pathway enrichment analysis were performed using the Metacape online platform to further understand the biological functions of FerrDEGs. The results showed that the significantly enriched genes were involved in the ferroptosis, cellular responses to stress, Hepatitis C and hepatocellular carcinoma, cellular response to chemical, Gastrin signaling pathway, etc. The terms were plotted as a network (Fig. 3A and B) to better visualize the interrelationships between the terms. The table in Fig. 3C shows the top 20 terms in detail. The respective results of BP, CC, MF and pathway enrichment analysis are shown in detail (Supplementary Figure S3).Fig. 3Functional enrichment analysis of FerrDEGs: (A) Network of enriched terms. B A bar chart of top 20 biological pathways based on the P-value and the percentage of genes. C Top 20 enriched terms from GO Biological Process and pathway databases (KEGG, Reactome, WikiPathways) Functional enrichment analysis of FerrDEGs: (A) Network of enriched terms. B A bar chart of top 20 biological pathways based on the P-value and the percentage of genes. C Top 20 enriched terms from GO Biological Process and pathway databases (KEGG, Reactome, WikiPathways) The MCODE algorithm was applied to explore two highly correlated modules (Fig. 4A-C). Module 1 genes were associated with autophagy of mitochondria, autophagy in pancreatic ductal adenocarcinoma, and organelle disassembly; and module 2 genes were associated with cellular senescence, PID Smad2/3 nuclear pathway, and androgen receptor signaling pathway (Fig. 4D).Fig. 4Construction of interactive network: (A) GO enrichment analysis was applied to each MCODE network. The same color nodes represent an interactive network and perform similar biological functions. B–C Two MCODE components were constructed with the screened hub genes. D Details of the 2 clusters were shown in table Construction of interactive network: (A) GO enrichment analysis was applied to each MCODE network. The same color nodes represent an interactive network and perform similar biological functions. B–C Two MCODE components were constructed with the screened hub genes. D Details of the 2 clusters were shown in table PPI network construction and hub genes identification Since 10 of the 50 genes are not associated with other molecules and do not form molecular networks, the resulting PPI network has 40 nodes and 130 edges. The top 10 genes, including Tumor Protein P53 (TP53), Jun Proto-Oncogene (JUN), Signal Transducer and Activator of Transcription 3 (STAT3), Cyclin Dependent Kinase Inhibitor 1 A (CDKN1A), Caveolin 1 (CAV1), Sequestosome 1 (SQSTM1), CD44 Molecule (CD44), Early Growth Response 1 (EGR1), Transforming Growth Factor Beta Receptor 1 (TGFBR1), Transcription Factor (SP1), were ranked according to the degree of clustering of the proteins analyzed by STRING (Supplementary Table S2). The PPI network was visualized with Cytoscape and constructed using CytoNCA plug-in. The top 20 genes are shown inside the circle and are sorted by size based on the Betweenness Centrality (BC) score (Fig. 5A). Additionally, the top 10 genes with the highest MCC scores were obtained using CytoHubba plug-in, as shown in Fig. 5B. The table (Fig. 5C) shows the descriptions and functions of the 8 hub genes selected by intersecting the top 20 genes with highest BC scores, top 10 genes with highest MCC scores, and top 10 genes with highest node degree. To further support the robustness of the hub gene selection, a random forest analysis (using the R ‘randomForest’ package) was performed. The top 8 genes identified were consistent with those obtained from the PPI network analysis (Supplementary Figure S4). Fig. 5Hub gene identification: (A) FerrDEGs PPI network constructed using Cytoscape, top 20 genes were shown inside the circle, genes were ranked by size according to Betweenness Centrality (BC) score. Node color represents the gradient of Betweenness Centrality (BC) values: red (high) → green (low). B Top 10 genes with the highest degree values were identified using CytoHubba and the depth of the color correspond to the weighted score. C The description and function of 8 hub genes were shown in table Hub gene identification: (A) FerrDEGs PPI network constructed using Cytoscape, top 20 genes were shown inside the circle, genes were ranked by size according to Betweenness Centrality (BC) score. Node color represents the gradient of Betweenness Centrality (BC) values: red (high) → green (low). B Top 10 genes with the highest degree values were identified using CytoHubba and the depth of the color correspond to the weighted score. C The description and function of 8 hub genes were shown in table Correlation analysis among the hub genes A Pearson correlation analysis was conducted to evaluate the relationship between the hub genes (Fig. 6A). The TP53-TGFBR1 pair showed the largest positive correlation, with an R-value of 0.9667 and a P-value of 0.0002 (Fig. 6B). Conversely, the SQSTM1-CAV1 pair showed the largest negative correlation, with an R-value of −0.8000 and a P-value of 0.0138 (Fig. 6C). Fig. 6Correlation analysis among the hub genes: (A) Correlation analysis among the 8 hub genes by Pearson analysis. B Correlation analysis between TP53 and TGFBR1 by Pearson analysis. C Correlation analysis between SQSTM1 and CAV1 by Pearson analysis Correlation analysis among the hub genes: (A) Correlation analysis among the 8 hub genes by Pearson analysis. B Correlation analysis between TP53 and TGFBR1 by Pearson analysis. C Correlation analysis between SQSTM1 and CAV1 by Pearson analysis Validation of key hub genes in the GSE230665 dataset ROC analysis was performed on the GSE230665 dataset to evaluate the predictive value of the hub genes. The resulting ROC curves indicated that the AUC values of five key hub genes, namely JUN, SQSTM1, STAT3, CD44 and TGFBR1, were greater than or equal to 0.75 (Fig. 7). ROC curve analysis was performed, and the results are presented with DeLong 95% confidence intervals, PR-AUC values, and optimal cut-off values determined by the Youden index (Supplementary Table S4). Given the limited sample sizes of the public datasets, a post-hoc power analysis was performed for the five key hub genes. In dataset GSE35958, all key hub genes exhibited Cohen’s d effect sizes greater than 0.8, indicating a high probability of detecting a true effect of this magnitude. Effect sizes (Cohen’s d) for the key hub genes are reported in Supplementary Table S5. Fig. 7Validation of these hub genes based on the GSE230665 dataset. The X-axis represents the (1-Specificity), and the Y-axis represents the sensitivity Validation of these hub genes based on the GSE230665 dataset. The X-axis represents the (1-Specificity), and the Y-axis represents the sensitivity Target MiRNAs prediction and integrated MiRNAs-Targets network construction Potential miRNAs targeting the five key hub genes were predicted using the miRTarBase v9.0 database within the Gene Regulatory Networks module of NetworkAnalyst. hsa-mir-20a-5p was selected as the top candidate based on its highest Betweenness Centrality score, with three connections (to SQSTM1, STAT3, and TGFBR1). Although hsa-miR-16-5p and hsa-miR-93-5p also targeted three genes, their betweenness scores were considerably lower than that of hsa-miR-20a-5p (Fig. 8). Fig. 8Identification of miRNAs: Interactions among key hub genes and miRNAs Identification of miRNAs: Interactions among key hub genes and miRNAs Validation of the key hub genes and rno-mir-20a-5p in vivo The OVX rat model was used to validate the key hub genes and rno-mir-20a-5p. The model was successfully established in Sprague-Dawley rats 12 weeks postoperatively. The microCT results indicated a decrease in BMD, BV, BV/TV, and Tb.N, and an increase in Tb.SP in the OVX group (Fig. 9). To detect the expression of these 5 key hub genes, the mRNA of BMSCs isolated from femur was used. The results showed that the OVX group had significantly higher relative mRNA expression levels of TGFBR1, STAT4, and CD44 compared to the Sham group (Fig. 10A-C). Conversely, the expression levels of JUN and SQSTM1 were lower in the OVX group than in the Sham group (Fig. 10D-E). These results are consistent with the up- and down-regulation results in the sequencing dataset (Supplementary Table S3). In addition, we evaluated the predetermined nodal points and the miRNA with the highest score, rno-mir-20a-5p, by PCR assay. The results showed that the expression level of rno-mir-20a-5p in the OVX group was significantly lower than that in the Sham group (Fig. 10F).Fig. 9Micro-CT results and analysis. A Representative 3D micro-CT reconstruction images of the OVX and Sham groups. B-F Quantitative bone analysis results of micro-CTFig. 10Validation of key hub genes in vivo. A–E The mRNA expression level of 5 hub genes (TGFBR1, STAT3, CD44, JUN and SQSTM1) in OVX and Sham. F The miRNA expression level of rno-mir-20a-5p in two groups Micro-CT results and analysis. A Representative 3D micro-CT reconstruction images of the OVX and Sham groups. B-F Quantitative bone analysis results of micro-CT Validation of key hub genes in vivo. A–E The mRNA expression level of 5 hub genes (TGFBR1, STAT3, CD44, JUN and SQSTM1) in OVX and Sham. F The miRNA expression level of rno-mir-20a-5p in two groups Lipid peroxidation and iron overload are characteristic manifestations of ferroptosis. To further demonstrate that the development of osteoporosis is related to ferroptosis, we measured lipid peroxidation levels and iron concentrations in serum and BMSCs of OVX rats. The results showed that the levels of MDA and the iron concentrations were significantly increased in the OVX group (Fig. 11A-D). Fig. 11Validation of lipid peroxidation and iron concentration. A-B Relative MDA expression in serum and BMSCs. C-D Relative iron concentration of serum and BMSCs Validation of lipid peroxidation and iron concentration. A-B Relative MDA expression in serum and BMSCs. C-D Relative iron concentration of serum and BMSCs Discussion The 2019 Nobel Prize in Physiology or Medicine was awarded to William G. Kaelin Jr. (USA), Gregg L. Semenza (USA) and Sir Peter J. Ratcliffe (UK) for their significant discoveries on cellular oxygen sensing mechanisms [21]. Cellular metabolism and physiology are affected by oxygen levels, and the response to hypoxia is crucial for the treatment of a variety of diseases such as cancer, cardiovascular diseases and osteoporosis. The hypoxia pathway, primarily regulated through hypoxia-inducible factor (HIF), plays a crucial role in the development of osteoporosis. Research has shown that the BMP signaling pathway responds to HIF in order to regulate bone homeostasis [22]. Additionally, the activation of the HIF pathway leads to the expression of the downstream gene VEGF, resulting in bone regeneration in osteoporosis [23]. In recent years, an increasing number of researchers have started studying the ferroptosis signaling pathway in relation to hypoxia and osteoporosis. A study revealed that HIF-1α inhibits RANKL-induced ferritinophagy and protects osteoclasts from ferroptosis [24]. Another study found that inhibiting HIF-1α overexpression significantly increased GPX4 expression and inhibited osteoblast ferroptosis [25]. Furthermore, a bioinformatics analysis confirmed HIF-1α as a potential regulatory gene for ferroptosis in primary osteoporosis [26]. The mechanism linking ferroptosis, hypoxia, and osteoporosis remains incompletely understood. Therefore, investigating their molecular mechanisms and identifying and intervening early undoubtedly holds important clinical significance. The aim of this study was to identify the key genes responsible for hypoxia-ferroptosis-osteoporosis by crossing FerrDb with microarrays of BMSCs derived from OP patients as well as hypoxia-induced BMSCs. A series of bioinformatics methods were used to obtain 50 FerrDEGs. GO enrichment analysis showed that FerrDEGs were mainly enriched for cellular response to chemical stress, multicellular organismal-level homeostasis, negative regulation of cell population proliferation, response to oxidative stress and negative regulation of phosphate metabolic process, which is consistent with previous study [26]. In addition, Pathway enrichment analysis showed that the FerrDEGs were mainly enriched in Viral carcinogenesis, Epstein-Barr virus infection, Hepatitis B, Human T-cell leukemia virus 1 infection and Fluid shear stress and atherosclerosis. It is important to note that these processes are mainly related to immunity and inflammation. Numerous studies have shown that immune-inflammatory processes and atherosclerosis play a critical role in osteoporosis [27–29]. Additionally, the MCODE analysis indicates a possible correlation between FerrDEGs and mitochondrial autophagy. Recent findings propose that irregular levels of autophagy and mitochondrial autophagy disturb the balance of bone metabolism [30–32]. The interaction between ferroptosis and mitochondrial autophagy deserves further investigation. Based on Cytoscape’s CytoNCA plug-in, CytoHubba plug-in, and Analyze Network tool, we identified eight overlapping DEGs as hub genes in the PPI network. These genes include TP53, JUN, SQSTM1, STAT3, CDKN1A, CAV1, CD44, and TGFBR1. These eight genes may play an important role in the pathogenesis of hypoxia-ferroptosis-osteoporosis. To test the hypothesis based on the GSE230665 dataset, we conducted a ROC analysis to compare the AUC values of 8 hub genes. We identified 5 key hub genes, including JUN, SQSTM1, STAT3, CD44, and TGFBR1, with an AUC value greater than or equal to 0.75. These genes were considered as an initial reference to determine whether they are differentially expressed during osteoporosis pathogenesis. Although the discovery cohort was based on BMSCs and the validation cohort utilized bone tissue, both share key cellular components of the bone microenvironment (e.g., osteoblasts). The hub genes identified through ROC analysis (AUC >0.8) demonstrated strong predictive performance, suggesting their relevance across these related biological contexts. JUN encodes a protein that is a component of the AP-1 transcription factor and plays a crucial role in regulating cell proliferation, differentiation, and apoptosis [33]. Recent studies suggest a direct correlation between JUN and bone metabolism. JUN is closely associated with the RANKL signaling pathway, which triggers the formation of osteoclasts and bone resorption, leading to bone loss [34, 35]. Additionally, JUN initiates the expression of BGP, BALP, and type I collagen and promotes bone formation through AP-1 [36, 37]. SQSTM1, also referred to as p62, is a vital protein involved in selective autophagy [38]. Its aberrant expression is closely linked to neurodegenerative diseases, tumors, infectious diseases, and the development of chronic diseases [39]. In RANKL-induced osteoclastogenesis, SQSTM1/p62 acts as an adaptor protein for autophagy. During RANKL-induced osteoclast differentiation, a negative correlation was observed between the expression and localization of p62/SQSTM1 and LC3 accumulation, as well as F-actin ring formation, which regulates autophagic activation [40]. In Paget’s disease of bone, SQSTM1/p62 plays a crucial role in regulating the turnover of ubiquitinated proteins through autophagy and the ubiquitin-proteasome system. Furthermore, it enhances NF-κB signaling and activates the Keap1/Nrf2 pathway in response to oxidative stress [41, 42]. STAT3 is involved in signaling and transcriptional activation, regulating various functions such as cell differentiation, proliferation, immune response, survival, apoptosis, and angiogenesis. The phosphorylation of STAT3 is a crucial event in several signaling networks that regulate osteogenesis and osteoblastogenesis [43]. When P-STAT3 is translocated to the nucleus, it binds to the Runx2 promoter, enhancing transcription and triggering osteoblastic differentiation in BMSCs [44, 45]. It is shown that osteoclast formation is generally stimulated indirectly by these pathways through JAK1 and STAT3 actions in inflammatory and other accessory cells, including osteoblasts [46]. Additionally, STAT3 regulates NFATc1 transcription, which controls osteoclast differentiation and bone homeostasis [47]. CD 44 is a transmembrane glycoprotein receptor that is widely distributed on the cell surface. It has multiple functions related to inflammation, including heterogeneous adhesion, cell migration, and modulation of the immune response [48]. During the process of osteoclast differentiation, CD44 is involved in the fusion of pro-osteoclasts, resulting in the formation of multinucleated mature osteoclasts [49]. The adhesion of osteoclasts to bone matrix, mediated by CD44-OPN, affects the process of bone resorption [50, 51]. CD44-induced cell signaling activates αvβ3 integrin, enhancing osteoclast motility. OPN stimulates osteoclast migration through αvβ3- and CD44-mediated cell signaling, which further increases CD44 expression on osteoclasts [52, 53]. TGFBR1 serves as a receptor for the cytokine TGF-β. Its role is to facilitate the translocation of TGF-β signaling from the cell surface to the cytoplasm, which is essential for cell proliferation and differentiation [54]. The TGF-β signaling pathway plays a crucial role in maintaining skeletal homeostasis by coupling the bone-forming and bone-resorbing activities of osteoblasts and osteoclasts [55, 56]. The TGF-β pathway aids in the differentiation of osteoblasts in BMSCs by signaling through specific type I and type II serine/threonine kinase receptors [57]. Additionally, TGF-β alters RANKL-dependent osteoclast activation and osteolysis [58]. Typically consisting of 18–25 nucleotides, miRNAs are involved in the regulation of many important biological processes, including cell development, differentiation, proliferation, metabolism and cell cycle regulation [59]. In this study, we constructed a miRNA-target gene network and identified the three miRNAs (hsa-mir-20a-5p, hsa-mir-16-5p, and hsa-mir-93-5p) that had the highest interaction with key hub genes. Among them, hsa-mir-20a-5p has the highest betweenness score and may be a crucial regulator of hypoxia-ferroptosis-osteoporosis. The hsa-mir-20a-5p gene is located on chromosome 12 and is included in the mir-17–92 cluster. Several studies have shown that hsa-miR-20a-5p is a key regulator of bone formation and bone resorption. Specifically, mir-20a promotes osteogenic differentiation by upregulating BMP/Runx2 signaling [60] and promotes osteogenesis by targeting BMP and activin BAMBI [61].In addition, anti-osteoclastic effects were induced in denuded rats by upregulating the expression of rno-miR-20a-5p in circulating and bone tissues [62], suggesting that rno-mir-20a-5p can effectively inhibit osteoclasts. To further validate our findings, an osteoporosis model was successfully established in OVX rats. Our study revealed that the OVX group had lower expression of JUN and SQSTM1 in BMSCs compared to the control group. Conversely, the OVX group had higher expression of TGFBR1, STAT3, and CD44 than the control group, which is consistent with the bioinformatics analysis. The involvement of ferroptosis in the development of osteoporosis was confirmed by detecting lipid peroxidation and iron levels at both serum and cellular levels. Finally, our results may provide new insights into the relationship between ferroptosis-related genes, hypoxia, and osteoporosis. We identified five key hub genes and hsa-mir-20a-5p as potential biomarkers for osteoporosis diagnosis and treatment. This study is limited by the relatively small sample sizes of the publicly available GEO datasets. Future studies should aim to validate these findings in larger, independent cohorts. Conclusion In summary, through multiple bioinformatics approaches, our study successfully identified five key hub genes associated with ferroptosis, hypoxia, and osteoporosis, namely JUN, SQSTM1, STAT3, CD44, and TGFBR1, as well as their potential regulator hsa-mir-20a-5p. In addition, mitochondrial autophagy may also be involved in the pathogenesis associated with ferroptosis-hypoxia-osteoporosis. In the future, we intend to collect sufficient bone tissue from patients with primary osteoporosis for further study. Supplementary Information Supplementary Material 1 Supplementary Material 1 Supplementary Material 2 Supplementary Material 2 Supplementary Material 3 Supplementary Material 3 Supplementary Material 4 Supplementary Material 4 Supplementary Material 5 Supplementary Material 5 Supplementary Material 6 Supplementary Material 6 Supplementary Material 7 Supplementary Material 7 Supplementary Material 8 Supplementary Material 8 Supplementary Material 9 Supplementary Material 9 Supplementary Material 10 Supplementary Material 10