Constructing a ferroptosis-related prognostic model for osteosarcoma based on scRNA-seq dataset and WGCNA analysis

Abstract

Osteosarcoma (OS) is a primary malignant bone tumor. Ferroptosis is closely related to the progression of osteosarcoma. The aim of this study is to explore the mechanism of ferroptosis-related genes in the progression of osteosarcoma. Methods: Utilizing the scRNA-seq dataset of osteosarcoma, differentially expressed genes (scRNA-DEGs) were identified between the osteosarcoma group and the control group, and ferroptosis-related genes in the TARGET-OS dataset were identified through WGCNA. By intersecting these two sets of ferroptosis-related genes, key candidate genes related to ferroptosis were obtained. The prognostic genes were selected from key candidate genes through univariate Cox and LASSO regression analysis. A prognostic model based on these genes was then constructed to investigate the relationship between ferroptosis and the prognosis of osteosarcoma patients. The correlation between the prognostic genes and immune cells, as well as immune checkpoint genes, was investigated through immune infiltration analysis. The drugs binding to prognostic genes were predicted. Results: We identified 48 ferroptosis-related genes in the scRNA-seq dataset, and 3859 ferroptosis-related genes were identified in the TARGET-OS dataset. After intersecting the two sets, 12 key ferroptosis-related genes were obtained, among which four genes (IFNG, HMOX1, CDKN1A, LGMN) were related to the prognosis of osteosarcoma patients. The prognostic model could accurately predict patient survival with good stability. Immune infiltration analysis revealed that the prognostic genes were significantly correlated with multiple immune cell types, and the expression of some immune checkpoint genes showed significant differences between control and osteosarcoma groups. Furthermore, we found that the prognostic genes were highly expressed in osteoblasts and macrophages. Conclusions: Four genes (IFNG, HMOX1, CDKN1A, LGMN) were closely related to the prognosis of osteosarcoma patients. These genes may serve as potential therapeutic targets for the treatment of osteosarcoma.

Introduction

Osteosarcoma (OS) is a primary malignant bone tumor, predominantly affecting the pediatric and adolescent population [1]. OS is characterized by significant genetic and phenotypic heterogeneity, making it challenging to identify consistent driver mutations across tumors or even within different regions of the same tumor [2]. The standard treatment approach includes early surgical resection and adjuvant chemotherapy. While patients with localized disease show significant improvements in treatment outcomes, with an event-free survival rate exceeding 60%, those with metastatic disease still exhibit poor prognosis, with an event-free survival rate below 30% [3]. Currently, OS is the most frequent primary bone malignancy, representing 35% of such tumors, with a male-to-female ratio of approximately 1.5:1 [4]. The majority of OS cases (80-90%) occur in the metaphyseal regions of long bones, particularly the distal femur, proximal tibia, and proximal humerus [5]. The five-year survival rate for OS remains dismal [5], and despite the development and implementation of new therapies, the prognosis for patients has not significantly improved since the 1980s. Approximately 40% of patients exhibit poor response to treatment [6,7]. Despite ongoing research, no substantial progress has been made in improving the efficacy of chemotherapy for OS, emphasizing the critical need for innovative approaches to treatment [8].

Ferroptosis, a programmed cell death mechanism, is driven by iron-catalyzed phospholipid peroxide accumulation [9]. As a distinct cell death pathway, ferroptosis plays a critical role in various diseases, including cancer, neurodegenerative disorders, and ischemic organ injury [10]. The salient features of ferroptosis are its dependence on iron and the accumulation of intracellular oxidative stress [11,12]. Isina et al. were the first to identify ferroptosis-like cell death in OS cell lines [13]. A mounting body of evidence supports the notion that inducing ferroptosis could serve as an effective strategy to impede the progression of OS [14]. For instance, Jiang et al. demonstrated that the expression of microRNA-144-3p promoted ferroptosis, reduced the viability of OS cells, and consequently inhibited tumor growth [15]. Furthermore, the activation of ferroptosis in murine models has been shown to significantly attenuate OS pulmonary metastasis [16]. In addition, multiple studies have confirmed that ferroptosis enhances chemosensitivity in OS. For example, METTL1-mediated modification of m7G regulates ferroptosis through the pri-miR-26a/FTH1 axis, which in turn affects OS cell sensitivity to the chemotherapeutic drugs doxorubicin and cisplatin [17]. Furthermore, Liu et al. have demonstrated that ferroptosis activation increases cisplatin sensitivity in OS cell lines [18]. These findings underscore the potential of ferroptosis as a promising therapeutic and prognostic target in OS. However, the precise molecular mechanisms underlying this association remain unclear. Therefore, this study utilizes bioinformatics methods such as WGCNA and scRNA-seq dataset analysis to identify prognostic genes related to ferroptosis in OS, aiming to discover new targets for the diagnosis and treatment of osteosarcoma.

Materials and methods

Data preparation

scRNA-seq datasets of bone tissue were downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/), including datasets GSE169396 and GSE162454. GSE169396 comprises four normal tissue datasets, while GSE162454 includes six osteosarcoma tissue datasets. After correcting for batch effect, the two datasets were merged to form a training set for the identification of scRNA-seq differentially expressed genes (scRNA_DEGs).

Gene expression profile and clinical data of TARGET-OS were obtained from the GDC database using the R package “TCGABiolinks”. TARGET-OS, containing 88 osteosarcoma samples, was utilized to construct the WGCNA (Weighted Gene Co-expression Network Analysis) network and prognostic model. Transcriptome dataset GSE21257, which contains 53 osteosarcoma samples, was obtained for the validation of the prognostic model. 484 ferroptosis-related genes were obtained from the FerrDb (version 2) database (http://www.zhounan.org/ferrdb/current/).

Osteosarcoma single-cell data processing

Quality control was performed on the four control samples from the GSE169396 dataset and the six OS samples from the GSE162454 dataset using the R package “Seurat” (v4.3.0.1). The NormalizeData function was applied to normalize the data, and highly variable genes were identified using the FindVariableFeatures function (with nfeatures set to 2000). To correct for batch effects between samples and avoid interference in downstream analyses, Canonical Correlation Analysis (CCA) was used. Subsequently, scale transformation was applied to the data using the ScaleData function for linear conversion, followed by dimensionality reduction with the RunPCA function. The number of dimensions for further analysis was determined by generating an elbow plot using the ElbowPlot function, with the top 50 principal components selected for downstream analysis. Cell clustering analysis was performed using the FindNeighbors and FindClusters functions (with dims = 1:50 and resolution = 0.1). Cell type annotations were conducted as described by the study [19].

Finally, we used the FindMarkers function with the thresholds |avg_logFC| > 0.1, p-value < 0.05, and min.pct > 0.25 to screen for differential genes between OS and the control groups, and intersected them with ferroptosis-related genes to obtain ferroptosis-related differential genes (Fer-DEGs).

Construction and module selection of co-expression network (WGCNA)

The single-sample gene set enrichment analysis (ssGSEA) method from R package “GSVA” (v1.46.0) was employed to calculate the enrichment scores of ferroptosis-related genes in the TARGET-OS dataset. A weighted gene co-expression network was constructed using R package “WGCNA” (v1.72-1). The dynamic tree cutting algorithm was applied to transform the network into a topological overlap matrix (TOM), categorizing all genes into distinct modules based on similar expression patterns. Pearson correlation coefficients were then calculated between the enrichment scores of each module and the ferroptosis-related genes. Modules significantly correlated with ferroptosis enrichment scores (|cor| > 0.3, p-value < 0.05) were identified, and the genes within these modules were designated as ferroptosis-related module genes (Fer-RMs). These genes were intersected with the previously identified ferroptosis-related differentially expressed genes (Fer-DEGs) to determine the key ferroptosis-related genes (Fer-KGs) that play critical roles in osteosarcoma.

Identification and functional enrichment of key ferroptosis-related genes in osteosarcoma

Gene Ontology (GO) enrichment analysis was conducted as a functional annotation method to elucidate the roles of genes and proteins, focusing on three main categories: Molecular Function (MF), Biological Process (BP), and Cellular Component (CC). Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis provided insights into Fer-KGs related biological pathways, based on KEGG database. We used the R package “clusterProfiler” (v4.6.2) to perform GO and KEGG pathway enrichment analysis on the Fer-KGs to identify the specific functions and pathways involved in the Fer-KGs.

Construction and validation of prognostic model

For transcriptome dataset TARGET-OS including 88 OS samples, survival data were available for 85 OS samples. We randomly divided these 85 samples into two groups at a ratio of 7:3, with the group accounting for 7/10 (n = 60) serving as the training set, and the group accounting for 3/10 (n = 25) serving as the test set. Transcriptome dataset GSE21257 including 53 OS samples was used as an independent validation set.

Based on the expression profile of Fer-KGs from training set, these genes associated with prognosis of OS patients were screened out through univariate Cox analysis (p-value < 0.1) and Proportional Hazards (PH) assumption testing (p-value > 0.05) using R package “survival” (v0.12.4). Then, prognostic genes were identified through least absolute shrinkage and selection operator (LASSO) COX regression analysis using R package “glmnet” (v4.1-7), meanwhile regression coefficients of prognostic genes were obtained. The results of the LASSO regression analysis were visually represented using a forest plot, which was generated with the R package “forestplot” (v3.1.1).

Using the following equation (1), the risk scores of OS samples in training set, test set and validation set were calculated. “Coef” represents the LASSO regression coefficient of a prognostic gene and “expr” represents the expression value of a prognostic gene.

$$\mathrm{R}\mathrm{i}\mathrm{s}\mathrm{k} \mathrm{S}\mathrm{c}\mathrm{o}\mathrm{r}\mathrm{e}=\sum _{\mathrm{i}=1}^{\mathrm{n}}(\mathrm{C}\mathrm{o}\mathrm{e}\mathrm{f}\left(\mathrm{i}\right)\times \mathrm{e}\mathrm{x}\mathrm{p}\mathrm{r}\left(\mathrm{i}\right))$$

The OS samples were stratified into high- and low-risk groups by median risk score across training, test, and validation cohorts. Kaplan-Meier (K-M) survival curve analysis was performed using the R package “survminer” (v0.5.9) for the training, test, and validation sets, respectively. 1-, 3-, and 5-year ROC curves, generated using R package “survivalROC” (v1.0.3.1), were utilized to assess the capability of the prognostic model in predicting OS patients’ survival rates. R package “ggrisk” (v1.3) was used to visualize the distribution of prognostic gene expression levels in the training, test, and validation set.

Analysis of immune infiltration and tumor immune microenvironment

To explore the infiltration of immune cells in OS samples, the ESTIMATE algorithm (Estimation of Stromal and Immune cells in Malignant Tumor tissues using Expression data) was implemented in the package “estimate” (v1.0.13). Generating the ImmuneScore (immune infiltration), ESTIMATEScore (stromal score + immune score), and Tumor Purity for OS samples in dataset TARGET-OS. The Wilcoxon test was used to examine differences in these scores across risk groups (p-value < 0.05). Next, we obtained 28 gene sets of immune cells from the TISIDB database (http://cis.hku.hk/TISIDB/download.php), and calculated the ssGSEA scores of these 28 gene sets using R package “GSVA” (v1.46.0) based on dataset TARGET-OS. A significant difference analysis (Wilcoxon test) was also performed in the ssGSEA scores of 28 immune cells to find significantly different immune cells. Spearman correlation analysis between ssGSEA scores of significantly different immune cells and expression levels of prognostic genes was conducted.

To investigate the immune therapeutic effects in OS, we employed the Wilcoxon test to analyze the expression levels of common immune checkpoint genes (including CD274, CTLA4, LAG-3, LGALS9, HAVCR2, PDCD1, PDCD1LG2, TIGIT) between the control and OS groups and subsequently calculated the correlation between the expression levels of these immune checkpoints and the risk score.

Drug prediction

For many anticancer drugs obtained from the Genomics of Drug Sensitivity in Cancer (GDSC) database (https://www.cancer rxgene.org/), we used the R package “oncoPredict” (v0.2) to predict the IC₅₀ (half-maximal inhibitory concentration) values of these drugs based on the expression profile of dataset TARGET-OS. We performed a Wilcoxon test to analyze the significant difference in the IC₅₀ values between the high-risk and low-risk groups, and analyze the Spearman correlation between the IC₅₀ values and the risk scores.

Statistical analysis

R software (v4.3.1) was used for all statistical analyses and data processing. Differences between the two groups were compared using the Wilcoxon rank-sum test (for non-normally distributed variables) and the independent Student’s t-test (for normally distributed variables). P-value < 0.05 were considered statistically significant.

Results

Ferroptosis-related gene screening based on scRNA-seq data analysis

The scRNA-seq datasets GSE169396 and GSE162454 were subjected to quality control and data integration, yielding a total of 71,164 cells and 25,163 genes. By t-SNE based dimensionality reduction clustering, we used the top 50 principal components (PCs) to identify 14 distinct cellular clusters (Figure 1A). Cell type annotation revealed 9 unique cell types: Endothelial cells (n = 1,246), Macrophages (n = 17,417), Mast cells (n = 2,146), Monocytes (n = 18,063), Osteoblasts (n = 14,585), Osteoclasts (n = 2,747), Plasmocytes cells (n = 2,253), B cells (n = 1,830), and T cells (n = 10,877) (Figure 1B). Osteoblasts and Macrophages showed a higher proportion in the disease group than in the control group and served as key cell types, while Monocytes were more abundant in the control group (Figure 1C). The differential analysis between the OS group and the control group yielded a total of 812 DEGs, of which 48 were ferroptosis-related DEGs (Figure 1D), denoted as scRNA_DEGs for subsequent analysis.

Figure 1.

scRNA-seq data analysis and ferroptosis-related gene screening. A. t-SNE (t-distributed stochastic neighbor embedding) distribution plots of different clustered samples. B. Annotated t-SNE clustering graph. C. Plot of cell ratio in control and disease groups. D. Differential expression analysis between two key cell types (Osteoblasts and Macrophages).

WGCNA Co-expression network analysis

Based on the TARGET-OS dataset, ssGSEA analysis was conducted on the ferroptosis-related gene set to calculate ssGSEA scores for each OS sample. The association between ssGSEA scores and survival rates was assessed using survival curves (Figure 2A). The results indicated that high expression of ferroptosis genes may play an important role in osteosarcoma. Subsequently, WGCNA analysis was performed on the TARGET-OS dataset. A scale-free network was constructed using a soft-thresholding power of 7, followed by topological overlap matrix (TOM) transformation via the dynamic tree-cutting algorithm, clustering the genes into 13 modules (Figure 2B). After merging similar modules, 11 distinct modules were obtained. Spearman’s Correlation analysis was used to assess the relationship between the 11 modules and ssGSEA scores (Figure 2C). Three modules that showed significant correlations with ferroptosis were selected for further analysis: the negatively correlated MEbrown module (cor: -0.41) and the positively correlated MEred (cor: 0.36) and MEpink (cor: 0.38) modules (Figure 2D). These three modules contained a total of 3,859 genes.

Figure 2.

WGCNA analysis. A. Survival curves, with high-risk group curves in red and low-risk group curves in blue. B. Soft threshold screening. Scale-free fit index (left), the end point of each vertical line corresponds to one sample. By calculating the Euclidean distance between samples from bottom to top, more similar samples are clustered into one class, the vertical axis indicates the distance between samples, the smaller the value of the distance the closer the greater the similarity. The average degree of connectivity (middle), the horizontal axis is the soft threshold, the vertical axis is the scale-free topology model fit index. Hierarchical Cluster Tree (right), the horizontal axis is the soft threshold, and the vertical axis is the mean connectivity. Vertical axis is scale-free topological model vertical axis is average connectivity?. C. Module-trait correlation plot, red is positive correlation, blue is negative correlation. The horizontal axis is the trait, the vertical axis is the different modules, and the main part is the correlation heatmap, with blue indicating negative correlation and red indicating positive correlation. The values in each individual grid are the correlation coefficients, with larger absolute values of the coefficients indicating stronger correlations, and the significance p-values in parentheses, with smaller p-values indicating more significant results. D. Scatterplot of GS and MM correlation strengths for key modules. E. Venn diagram showing the intersection between key module genes and scRNA_DEG, highlighting the final screened iron death-related genes.

Identification and functional enrichment analysis of key genes

Intersection of the scRNA_DEGs with three WGCNA modules identified 12 key genes: TFRC, HMOX1, CDKN2A, IFNG, IL1B, CTSB, LGMN, EGR1, TIMP1, CDKN1A, SLC16A10, and GDF15 (Figure 2E). Functional enrichment analysis of these key genes revealed significant enrichment in functions related to cytokines, chemokines, and NF-κB signaling, as indicated by GO enrichment analysis (Figure 3A). KEGG pathway enrichment analysis demonstrated significant enrichment in pathways such as the HIF-1 signaling pathway, Antigen processing and presentation, Ferroptosis, and various cancer-related pathways (Figure 3B).

Figure 3.

GO/KEGG enrichment analysis. A. GO enrichment analysis. B. KEGG pathway enrichment analysis. The color indicates the enriched entries, and the redder the color, the higher the degree of enrichment.

Identification prognostic gene and prognostic model construction

Using the training set of 60 samples from the TARGET-OS dataset, univariate Cox analysis (P < 0.1) and the proportional hazards (PH) assumption test were conducted on the 12 key genes (Figure 4A), resulting in the identification of four genes associated with prognosis (IFNG, HMOX1, CDKN1A, LGMN) (Figure 4B). These four genes were determined to be prognostic genes through LASSO regression analysis. Subsequently, a prognostic model associated with ferroptosis was constructed based on the risk score. The performance of the model was evaluated across the training set, test set and an external validation set. The K-M survival curve indicated that patients in the high-risk group had significantly lower survival rates compared to those in the low-risk group (Figure 4C). Additionally, as the risk score increased, the same time point was significantly higher (Figure 4C), indicating that the risk score effectively distinguishes patient populations with different survival probabilities. The area under the curve (AUC) values at 1, 3, and 5 years were 0.719, 0.824, and 0.818, respectively (Figure 4C), indicating strong accuracy and predictive capability of the model. The results from the test set (Figure 4D) and external validation set (Figure 4E) were consistent with those observed in the training set, with K-M survival curves, risk scores, and ROC curves showing similar trends. AUC values for 1, 3, and 5 years exceeded 0.6 across all sets, highlighting the model’s generalizability and reliability in predicting prognosis across different patient populations.

Figure 4.

Prognostic gene identification and prognostic model construction. A. One-factor cox forest plot, the dark blue line represents the 95% confidence interval of HR, and the red squares represent the HR value of each variable. B. Lasso regression screening of prognostic genes. Left panel: the horizontal axis is the value of log (lamabda), the vertical axis is the regression coefficient, and different colors indicate the changes of regression coefficients of different genes with the change of log (lamabda). Right panel: log (lamabda) values on the horizontal axis and partial likelihood deviations on the vertical axis. The red dots show the changes of partial likelihood deviation with log (lamabda) values. C. Training set risk factor linkage plot (left), red is high risk group, blue is low risk group, red circle is death, blue circle is survival, and the horizontal axis is sample. Test set 1-, 3-, and 5-year ROC curves (center); and test set KM curves (right), with survival time on the horizontal axis and survival rate on the vertical axis. D. Test set risk factor linkage plots (left), red is high-risk group, blue is low-risk group, red circle is death, blue circle is survival, and the horizontal axis is sample. Training set 1, 3, and 5-year ROC curves (center); training set KM curves (right), the horizontal axis is survival time, and the vertical axis is survival rate. E. Validation set risk factor linkage plots (left), red is high-risk group, blue is low-risk group, red circle is death, blue circle is survival, and the horizontal axis is sample. Validation set 1-, 3-, and 5-year ROC curves (center); validation set KM curves (right), the horizontal axis is survival time, and the vertical axis is survival rate.

Immune profiling analysis: microenvironment, infiltration, and checkpoint

Utilizing the ESTIMATE algorithm to analyze the immune microenvironment revealed significant differences in Immune score, ESTIMATE score, and Tumor Purity for each sample (Figure 5A). Immune infiltration analysis further identified significant differences between high-risk and low-risk groups across 12 immune cell types, including Activated B cell, Activated CD8+ T cell, CD56 bright natural killer cell, Central memory CD8+ T cell, Effector memory CD8+ T cell, Gamma delta T cell, Immature B cell, Mast cell, MDSC (Myeloid-Derived Suppressor Cell), Monocyte, Regulatory T cell, and Type 1 T helper cell (Figure 5B). Spearman correlation analysis between the 12 differential immune cells, risk scores, and four prognostic genes revealed a significant negative correlation between the risk scores and the differential immune cells (correlation coefficient < -0.3 and p-value < 0.05). In contrast, the differential immune cells showed a positive correlation with the prognostic genes (Figure 5C), suggesting that upregulation of these genes is associated with the activation or recruitment of immune cells, which may positively impact patient prognosis. Additionally, this study analyzed the correlation between immune therapy-sensitive OS patients in high and low-risk groups and immune checkpoint genes, finding differences in 8 immune checkpoint genes (CD274, CTLA4, LAG3, LGALS9, HAVCR2, PDCD1, PDCD1LG2, and TIGIT) between the high and low-risk groups (Figure 5D), and a high correlation with the risk scores (Figure 5E). In addition, in the single-cell dataset, the prognostic genes were highly expressed in macrophages, suggesting that these genes play a significant role in the immune regulatory function of macrophages, which may be closely associated with immune responses and prognosis in the tumor microenvironment.

Figure 5.

Immune infiltration analysis. A. ESTIMATE immune microenvironment. B. Box plots of differences in expression of immune cells between high and low risk groups, ns (not significant): P > 0.05, *: P < 0.05, **: P < 0.01, ***: P < 0.001. C. Heatmap of correlation of genes with 12 immune cells. D. Box plots of differences in immune checkpoint genes. E. Correlation plots of immune checkpoint genes with risk score correlation plot.

Drug sensitivity analysis

Utilizing the GDSC database, the response of each OS patient to anticancer drugs was assessed through the half-maximal inhibitory concentration (IC50), determining the sensitivity to various chemotherapy drugs (Table S1). Additionally, the response of specific biomarkers to anticancer drugs was evaluated between the high-risk and low-risk groups (Table S2). Spearman’s correlation analysis revealed a significant association between 79 drugs and the risk score (Figure 6A), with notable differences in drug responses observed between the high-risk and low-risk groups (Figure 6B). This analysis underscores the potential of these biomarkers in predicting therapeutic outcomes and guiding personalized treatment strategies.

Figure 6.

Drug sensitivity analysis. A. Correlation plot between medications and risk scores. B. Box plot of medication differences between high and low risk groups.

Discussion

As one of the primary skeletal tumors, OS exhibits a notably low survival rate, with less than 20% of patients surviving five years post-metastasis [20]. Despite advancements in medical treatments that have improved outcomes for OS, approximately 40% of patients continue to experience poor prognoses [6,7]. Ferroptosis, a form of programmed cell death associated with iron metabolism and oxidative stress [9], has been implicated in the progression of osteosarcoma [15]. However, the molecular mechanisms regulating ferroptosis in OS remain poorly understood. Therefore, it is imperative to investigate potential therapeutic targets related to ferroptosis in order to improve patient survival outcomes. In this study, 4 prognostic genes were identified through bioinformatics analysis and their prognostic implications were evaluated alongside an analysis of their immune characteristics. These 4 genes (IFNG, HMOX1, CDKN1A, and LGMN) were significantly correlated with prognosis and immune infiltration, and they appear to play protective roles in osteosarcoma.

With the advancements in bioinformatics and the growing use of single-cell sequencing (scSeq) for disease analysis, these methods can efficiently identify marker genes relevant to various conditions, providing foundational support for targeted therapies. Bioinformatics analyses have revealed that ANK1 inhibits OS cell proliferation, migration, and invasion of OS cells by promoting ferroptosis, suggesting it as a novel therapeutic target for OS treatment [21,22]. Through analysis of transcriptome datasets, Christofides et al. identified hub genes and discovered that myofascial fibrosarcoma homologous gene B serves as a key transcriptional regulator within OS contexts. Meanwhile, scRNA-seq data analysis demonstrated that macrophages contribute significantly within the tumor microenvironment by presenting antigens, secreting cytokines, and participating in immune regulation [23]. Our scRNA-seq data analysis indicated that ferroptosis-related genes are highly expressed in specific cell types within the tumor microenvironment, particularly in osteoblasts and macrophages. Osteoblasts and macrophages are integral components of the osteosarcoma (OS) microenvironment and play multiple roles in OS initiation and progression [24]. Although the cellular origin of OS remains controversial, accumulating evidence identifies osteoblasts as OS progenitors [25,26]. Furthermore, osteoblasts promote OS metastasis and invasion by modulating the bone microenvironment and interacting with other cellular components [27,28]. Macrophages, which are central to bone homeostasis and the “osteoimmune” system [29], play critical roles in regulating inflammatory responses, mediating chemotherapy resistance, facilitating OS metastasis, and orchestrating microenvironmental crosstalk, although the mechanisms are incompletely understood [30]. Several studies confirm that the degree of macrophage infiltration significantly correlates with the prognosis of OS patients [31,32]. Additionally, the expression of these ferroptosis-related genes was significantly correlated with patient prognosis, suggesting their critical involvement in OS progression. Previous studies have corroborated this hypothesis, demonstrating that ferroptosis modulates the advancement of osteosarcoma and the effectiveness of treatments [33].

We further employed WGCNA to isolate key genes related to ferroptosis. Functional enrichment analyses illustrated that these genes are predominantly enriched in regulating cytokine/chemokine signaling pathways, which are crucial for immune responses. Cytokines, including interleukins and interferons etc., play pivotal roles in modulating immune responses and cellular behaviors by influencing the tumor microenvironment to either promote or suppress neoplastic progression, thereby influencing therapeutic efficacy [34]. KEGG enrichment analysis revealed significant enrichment of key genes in the HIF-1 signaling pathway, which exhibits multifaceted roles in osteosarcoma (OS). Experimental evidence indicates that HIF-1 signaling supports hypoxia-induced OS cell metastasis, while its central mediator HIF-1α drives tumor cell proliferation, migration, and invasion in OS through activation of the AKT/Cyclin D1 signaling cascade [35,36]. Comprehensive studies confirm that sustained HIF-1 pathway activation induces therapy-resistant phenotypes via apoptosis evasion and cellular senescence mechanisms, ultimately exacerbating therapeutic resistance [37]. This suggested that ferroptosis not only impacted tumor cell mortality but may also influenced immune evasion strategies and therapeutic responses by modulating immune cells within the tumor microenvironment [38,39]. Our analysis further examined how the 4 ferroptosis-related prognostic genes (IFNG, HMOX1, CDKN1A, and LGMN) interact within the immunological landscape of OS by assessing immune infiltration patterns and checkpoint analysis. We found notable disparities across multiple infiltrating immune cell types positively correlating them specifically towards twelve distinct categories.

These 4 prognostic genes may play pivotal roles in enhancing immunosurveillance and improving immunotherapy outcomes for OS. Subsequently, based on these four prognostic genes, we constructed robust and reliable prognostic models and validated their predictive power in two additional independent datasets. IFNG (interferon gamma) is a well-known cytokine that enhances both innate and adaptive immune responses, playing a critical role in antitumor immunity [40]. S100A4/IFN-γ-mediated apoptosis relies heavily on reactive oxygen species (ROS) formation, and human-derived OS cell lines expressing S100A4 exhibit heightened sensitivity to IFN-γ-induced apoptosis compared to other cell types [41]. Emerging evidence suggests that IFNG plays a critical role in the chemotherapeutic efficacy of osteosarcoma (OS), with significant correlations with favorable clinical outcomes in OS patients [42,43].

HMOX1 (heme oxygenase 1), primarily involved in heme metabolism, plays a crucial role in the recovery from iron stress and can lead to ferritin-associated cell death, thereby reducing oxidative stress [44]. HMOX1 exhibits a dual role in osteosarcoma (OS), demonstrating anti-tumor potential through its capacity to mitigate oxidative damage and suppress OS cell proliferation/metastasis [45]. Experimental evidence reveals that HMOX1 induction at the cellular level restores osteoblast proliferation (PMID: 24224047). Zoledronic acid has been shown to increase lipid peroxidation in osseous tissues through the upregulation of HMOX1, ultimately inhibiting OS cell proliferation [46]. Conversely, another study demonstrates that HMOX1 downregulation inhibits OS cell proliferation and metastasis while promoting apoptosis and cell cycle arrest [47]. Paradoxically, as a pro-angiogenic mediator, HMOX1 has been shown to potentiate tumor progression and metastasis in multiple malignancies, including colorectal carcinoma, lung cancer, and renal cell carcinoma [48-50]. Our findings identify HMOX1 as a protective factor in OS (HR = 0.71) with significant positive correlations to various immune cell populations, including MDSC, mast cells, T cells, etc. These collective findings position HMOX1 as a promising therapeutic target for OS. However, the precise mechanistic underpinnings require further experimental validation, which will constitute a primary focus of our subsequent investigations.

CDKN1A (cyclin-dependent kinase inhibitor 1A), commonly designated as p21, primarily functions as a cell cycle regulator by inhibiting cyclin-dependent kinases, thereby controlling cellular proliferation and apoptosis [51]. Research indicates suppression occurring via miR-106b-5p inhibition coupled together rising concentrations noted concerning CDKN1A leads arresting phases witnessed G0/G2 halting any further divisions taking place effectively stunting malignant expansions. However, previous studies have paradoxically shown that CDKN1A downregulation can actually promote tumor progression under certain microenvironmental conditions [52,53]. Previous experimental results have shown that miR-95-3p knockdown suppresses OS cell growth through regulation of CDKN1A expression [54]. Taken together, these findings confirm that CDKN1A is a critical tumor suppressor in the pathogenesis of OS. Therapeutic strategies targeting CDKN1A may thus offer new therapeutic avenues for the treatment of osteosarcoma.

LGMN (Legumain), a member of the C13 cysteine peptidase family, drives malignant aggressiveness and metastasis in diverse cancers including breast and ovarian carcinomas [55]. Accumulating evidence has established LGMN as a reliable prognostic biomarker, as evidenced by its elevated expression in colorectal cancer tissues correlating with poor clinical outcomes [56]. Furthermore, melanoma patients with high LGMN expression exhibit enhanced local tumor invasiveness and reduced survival rates [57]. In gastric cancer, LGMN overexpression predicts increased peritoneal metastasis risk and diminished survival probability [58]. Notably, recent work reveals LGMN overexpression in osteosarcoma (OS) tissues, associated with decreased overall/progression-free survival and elevated M2 macrophage infiltration [59]. Our findings further demonstrate significant positive correlations between LGMN expression and infiltration levels of multiple immune cell populations, suggesting its pivotal role in orchestrating the OS immune microenvironment. These collective findings underscore LGMN’s dual potential as a prognostic indicator and therapeutic target in OS. Elucidating the precise molecular mechanisms underlying these observations constitutes an essential direction for future investigation.

This study confirmed that high expression of the previously identified ferroptosis-related genes provides protective benefits in individuals with osteosarcoma, consistent with earlier findings. These results emphasized the importance of understanding the interactions between these biomarkers and their impact on patient outcomes during the course of osteosarcoma treatment. Moving forward, this research opened new avenues for exploring the potential of these biomarkers in developing innovative therapeutic solutions.

There are some unavoidable limitations to this study. First, this study was based on the analysis of retrospective data from public databases and the limited sample size of osteosarcoma as a rare malignancy. Second, although our prognostic model showed good predictive performance, it still needs to be validated in more real-world data to clarify its clinical value. In addition, our results were all based on bioinformatics analysis, and further ex vivo experiments are needed to explore and validate these findings. Therefore, clarifying the roles and mo-lecular mechanisms of these four prognostic genes (IFNG, HMOX1, CDKN1A, and LGMN) in OS will be the focus of our future research.

Disclosure of conflict of interest

None.