A comprehensive analysis of ferroptosis-related metastasis genes in osteosarcoma

Abstract

Background

Osteosarcoma (OS) is an increasingly common primary bone tumor in children and adolescents, characterized by rapid local infiltration, distant metastasis, high invasiveness, and poor drug sensitivity. Ferroptosis is a programmed cell death caused by iron dependent, lipid peroxidation accumulation, ultimately leading to the destruction of cell membrane structure and function, resulting in cell death. However, the specific role of ferroptosis-related metastasis genes (FRMGs) in the pathogenesis of osteosarcoma is still unclear.

Methods

Utilize GEO database to obtain genetic and clinical information of osteosarcoma patients. Perform cluster analysis on FRMG using Consensus ClusterPlus. We conducted pathway enrichment analysis on DEGs from different subgroups using Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO). We used the Lasso regression model to construct line scores, and in order to improve the predictive ability of the model, we created a prediction bar chart and used a calibration curve. Finally, we used Oncoredist to compare drug sensitivity between high-risk and low-risk cohorts.

Results

In the clustering model, the survival time of patients in the C2 subgroup is higher than that of patients in the C1 subgroup. Pathway enrichment analysis showed that Glycan biosynthesis pathway, Spliceosome pathway, and Ubiquitin medtated proteoiysis pathway was enriched in osteosarcoma patients in group C1, while Glycosphingolipid biosynthsis pathway was enriched in osteosarcoma patients in group C2. In the Lasso regression model, the survival rate of high-risk scoring patients is significantly lower than that of low-risk scoring patients, and the risk score of C1 subgroup patients is significantly higher than that of C2 subgroup patients. The results of drug sensitivity analysis indicate that patients with low-risk scores are more sensitive to drugs. Patients with high expression of the FH gene in the risk model of osteosarcoma have a poorer prognosis, while those with high expression of the IL1B and SNCA genes in the risk model have a better prognosis. The single-cell analysis results showed that IL1B had the highest expression in monocytes/crops. The 3-year and 5-year survival values predicted by the ROC curve are 0.79 and 0.84, respectively, indicating the reliability of our prediction model.

Conclusion

This study proposes a clinically applicable prognostic model based on ferroptosis-related metastasis genes, which provides insights into risk stratification and treatment decision-making for osteosarcoma, and provides a potential reference for risk stratification and therapeutic decision-making in osteosarcoma.

Supplementary Information

The online version contains supplementary material available at 10.1007/s12672-026-05003-2.

Introduction

Osteosarcoma is the most common type of bone cancer in adolescents and teenagers, which usually develops in the metaphyses of long bones, particularly the proximal tibia, proximal humerus, and distal femur [1, 2]. The incidence of osteosarcoma is extremely low among all cancers, but osteosarcoma patients have a poor prognosis since they are particularly susceptible to metastases in their early stages [3–5]. Current standard treatments for osteosarcoma are surgery, chemotherapy, radiotherapy, and immunotherapy. However, owing to the histological characteristics of osteosarcoma, the curative ability of these treatment methods for patients with osteosarcoma is limited [6]. Therefore, it is crucial to construct a new model to predict the prognosis of osteosarcoma.

Ferroptosis is a unique programmed cell death mode that relies on iron ions to drive lipid peroxidation as its core. Its morphological characteristics include reduced cell volume, mitochondrial atrophy, increased membrane density, and no typical formation of apoptotic bodies [7, 8]. Ferroptosis not only plays a crucial role in tumor development, but also participates in the development of several other diseases, including cirrhosis, Parkinson’s disease, acute liver injury, non-alcoholic fatty liver disease, antimicrobial infections, multiple organ dysfunction syndrome, and rheumatoid arthritis [9–14].

Cells regulate ferroptosis by directing iron ion within the cell through transferrin receptor 1 (TFR1) [15]. Cells suppress ferroptosis by lowering the formation of lipid peroxides via the SLC7A11-GPX4 axis-mediated glutathione metabolism pathway [16, 17]. Existing research suggests that ferroptosis can regulate tumor metastasis to affect the prognosis of cancer patients. For example, LASS2 can inhibit the metastasis of thyroid, breast, and liver cancers by regulating the ferroptosis signaling pathway through TFRC interaction [18]. SGK2 improves the metastatic capacity of prostate cancer cells in vivo and in vitro by upregulating GPX4, which inhibits ferroptosis [19]. CST1 affects the stability of GPX4 protein to inhibit ferroptosis, that enhance the metastatic ability of gastric cancer [20]. Due to the limited study on how ferroptosis governs the prognosis of osteosarcoma by influencing metastatic ability, it is critical to investigate FRMGs’s capacity to predict the prognosis of osteosarcoma patients.

In this study, we demonstrated that ferroptosis-related metastasis genes (FRMGs) can predict the prognosis, immune cell infiltration, immune environment, and treatment sensitivity of patients with osteosarcoma. These findings add to our understanding of the role of FRMGs and their significance in clinical practice.

Materials and methods

Data collection and processing

The osteosarcoma datasets GSE21257 and GSE39058 have been downloaded from the GEO database (https://www.ncbi.nlm.nih.gov/geo/). The GSE21257 dataset contains clinical information (gender, age, survival status, survival time, metastatic and non-metastatic) and transcriptome mRNA expression levels of 53 patients with osteosarcoma. Similarly, the GSE39058 dataset contains clinical information (gender, age, survival status, survival time, metastatic and non-metastatic) and transcriptome mRNA expression levels of 42 osteosarcoma patients. We conducted a prognostic difference analysis between patients with osteosarcoma metastasis and those without metastasis.

Differential expression analysis

Limma [21] is a differential expression screening method based on a generalized linear model. Here, we use the R software package limma (version 3.40.6) for differential analysis to obtain differentially expressed genes between different comparison groups and control groups. Specifically, the expression profile dataset we obtained was subjected to multiple linear regression using the lmFit function, followed by computed moderated t-statistics using the eBays function, moderated F-statistic, and log-odds of differential expression by empirical Bayes moderation of the standard errors towards a common value, Ultimately obtaining the significance of differences for each gene. In our study, the comparison group consisted of non-metastatic patients with osteosarcoma, while the control group consisted of patients with osteosarcoma metastasis. According to the conventional differentially expressed genes (DEGs), we positioned the identification criteria for DEGs between the subgroup of osteosarcoma non metastatic patients and the subgroup of metastatic patients as | log2FC |>1.5 and P < 0.05.

Consensus clustering analysis

For cluster analysis, we used ConsensusClusterPlus [22], resampling 80% of the samples for 10 iterations and used aggressive PAM clustering with a 1-Pearson correlation distance. The ideal number of clusters was found using an empirical cumulative distribution function plot.

2.4 Pathway enrichment analysis

Conduct pathway enrichment analysis using DEGs from two subgroups in clustering groups. We utilized the KEGG Rest API (https://www.kegg.jp/kegg/rest/keggapi.html) to analyze KEGG pathways.

We use the R package org.Hs.eg.db (version 3.1.0) for GO analysis. We use GSEA software (version 3.0) for GSEA analysis [23].

Immune microenvironment analysis

The CIBERSORT algorithm was utilized to determine the level of immune cell infiltration in each sample. CIBERSORT is a tool employed to perform deconvolution analysis of immune cells [24]. We examined the variations between the groups in terms of ESTIMATEScore, ImmuneScore, StromalScore, and TumorPurity. ESTIMATE is an algorithm used to evaluate the purity of tumor samples [25].

Establishment of prognostic model

We selected FRMGs related to prognosis through LASSO regression analysis. The R software package glmnet was used to perform LASSO regression analysis. Risckscore is constructed based on the expression levels of genes and coefficients. Patients are classified as low or high risk based on the best cutoff value for the riskscore. We evaluate the survival curves for the high riskscore and low riskscore categories. ROC curve analysis is used to evaluate the accuracy of predictive models.

Single cell analysis

We used the tisch2 database (http://tisch.comp-genomics.org/) to perform single-cell clustering, grouping, annotation analysis, and gene expression analysis among different cell subgroups based on single-cell sequencing data of osteosarcoma.

Drug sensitivity analysis

We calculated the drug sensitivity among the different groups using Oncopredict. We applied CTRP2 data as the training set for drug sensitivity.

Nomogram analysis

Prognostic nomogram is a visualization tool based on multiple regression models, which integrates individual clinical characteristics of patients (such as age, tumor stage, pathological type, treatment plan, laboratory indicators, etc.) to quantitatively predict specific clinical outcomes (such as disease recurrence risk, survival probability, treatment response rate). We create a nomogram by combining riskscore and clinicalpathological data. The R package rms is used to construct nomogram.

2.11 Survival analysis

We analyzed the prognostic differences between the two groups using the R package Survival, and we utilized the log-rank test method to determine the significance of prognostic differences across samples from various groups.

Statistical analysis

For date analysis, we utilized R (version 4.1.1) software. Pearson was employed to conduct correlation analysis. Wilcoxon or T-test analysis is used to analyze the differences between two groups. P < 0.05 is considered statistically significant.

Results

Screen of FRMGs

Based on the incidence of distant metastases, we classified patients with osteosarcoma in the GSE21257 dataset into groups that were metastatic and non-metastatic. Patients with metastatic osteosarcoma had a far worse outcome than those with non-metastatic osteosarcoma (Fig. 1A). A differential analysis conducted between the metastatic and non-metastatic groups of patients with osteosarcoma using the screening criteria of | log2FC |>1.5 and P < 0.05 obtained 472 DEGs. A heat map displayed the top 40 genes (Fig. 1B, C). We gathered information on 453 ferroptosis genes (http://www.zhounan.org/ferrdb), and FRMGs were identified by examining the intersection of these ferroptosis genes and DEGs (Fig. 1D). FRMGs were shown to be correlated (Fig. 1E).

Fig. 1.

Construction of FRMGs. A Survival analysis of metastasis and non-metastasis groups of patients with osteosarcoma. B Volcano plot of differential analysis. C heatmap of differential genes. D Venn plot of ferroptosis genes and metastasis genes. E Correlation analysis of FRMGs

Consensus clustering analysis of FRMGs

The offline area needed to be maintained as large as possible while evaluating the downward trend of the CDF Delta, and we attempted to maintain the slowest decrease in delta. Considering these two factors, K = 2 was chosen as the number of clusters (Fig. 2A–C). After splitting the osteosarcoma patients into two groups, C1 and C2, we discovered that group C2’s prognosis was superior to that of group C1’s (Fig. 2D). Finally, we found that the SNCA, IL1B, CYBB, and CTSB in theFRMGs were highly expressed in the C2 group, while PEX6, CDCA3, YAP1, GJA1, FH, ACSL1, ELOVL5, and MTDH were highly expressed in the C1 group (Fig. 2E).

Fig. 2.

Consensus clustering analysis. A–C Consensus clustering analysis. D survival analysis of patients in groups C1 and C2. E expression of FRMGs in patients in groups C1 and C2

Clinical features of C1 and C2

We compared the clinical information of osteosarcoma patients in groups C1 and C2 in Table 1. We found that the proportion of metastasis in patients with osteosarcoma in group C1 was higher than that in group C2. We also found no significant differences in gender and age (adult and minor) between C1 and C2 groups of osteosarcoma patients.

Table 1.

Clinical information of osteosarcoma patients

Characteristics	C1 (N = 25)	C2 (N = 28)	Total (N = 53)	pvalue
OS (months)
Median [min-max]	31.00 [4.00,189.00]	70.00 [10.00,246.00]	45.00 [4.00,246.00]
Status				0.36
Alive	12 (22.64%)	18 (33.96%)	30 (56.60%)
Dead	13 (24.53%)	10 (18.87%)	23 (43.40%)
Gender				0.4
Female	7 (13.21%)	12 (22.64%)	19 (35.85%)
Male	18 (33.96%)	16 (30.19%)	34 (64.15%)
Age (months)				1.0
≥216	9 (16.98%)	10 (18.87%)	19 (35.85)
<216	16 (30.19%)	18 (33.96%)	34 (64.15%)
Metastases
No	4 (7.55%)	15 (28.30%)	19 (35.85%)	0.01
Yes	21 (39.62%)	13 (24.53%)	34 (64.15%)

Enrichment analysis of DEGs between C1 and C2 groups of patients

To find genes that were expressed differently between the C1 and C2 groups, we performed differential analysis on the C1 and C2 groups using the screening criteria of | log 2FC |>1.0 and P < 0.05. The top 40 genes in C1 and C2 are shown as circular plots (Fig. 3A, B). We conducted KEGG and GO pathway enrichment analyses on the DEGs between groups C1 and C2. These genes were primarily enriched in DNA metabolic activity, cell motility, and the T-cell pathway. (Fig. 3C, D). For GSEA analysis, we discovered that Glycan biosynthesis pathway, Spliceosome pathway, and Ubiquitin medtated proteoiysis pathway was enriched in osteosarcoma patients in group C1, while Glycosphingolipid biosynthsis pathway was enriched in osteosarcoma patients in group C2 (Fig. 3E).

Fig. 3.

Differential enrichment analysis. A Differential gene analysis of patients in groups C1 and C2. B circle diagram of DEGs. C, D KEGG and GO enrichment analysis. E GSEA pathway enrichment analysis

Analysis of Immune microenvironment in patients of C1 and C2 groups

(Fig. 4A) shows the proportion of all immune cells in osteosarcoma patients. While the infiltration degrees of NK cells resting, T-Cell CD4 memory-resting, and T-Cell CD4 naive were lower in the C2 group than in the C1 group, the infiltration degrees of T-Cells_CD8, T-Cell_follicular helper, and T-Cells_gamma delta were higher in the C2 group than in the C1 group (Fig. 4B). Due to the fact that the cell score calculated by CIBERSORT is zero in both C1 and C2 groups, it is not shown in the Fig. 4B. Furthermore, The ESTAMATE analysis revealed that the C2 group had higher ESTIMATEScore and ImmuneScore than the C1 group, but a lower TumorPurity (Fig. 4C).

Fig. 4.

Analysis of Immune microenvironment. A The proportion of immune cells in patients with osteosarcoma. B types of immune cell infiltration in patients in groups C1 and C2. C ESTAMATE analysis in C1 and C2.* P < 0.05, **P < 0.01, ***p < 0.001

Establishment of prognostic model

To further investigate the impact of FRMGs on the prognosis of osteosarcoma patients, we used the Lasso regression algorithm to identify FRMGs. In the LASSO regression analysis, we obtained three genes. RiskScore=-0.00115 * expression of IL1B+− 0.27754 * expression of SNCA + 0.12631 * expression of FH (Fig. 5A, B). We analyzed the relationships among riskscore, OS, overall survival, survival status, and the expression of genes. We found that IL1B and SNCA are highly expressed in the low riskscore group, while FH is highly expressed in the high riskscore group (Fig. 5C). Among patients with osteosarcoma, the high riskscore group’s prognosis was considerably worse than the low riskscore group’s (Fig. 5D). We generated time-dependent ROC curves for 12, 36, and 60 months (Fig. 5E). Similarly, it was found in the validation set GSE39058 that low-risk patients had better prognosis, the 12, 36, and 60 months survival values predicted by the ROC curve are 0.94, 0.87, and 0.87 (Supplementary material). This result confirms the accuracy of the prognostic model. Finally, we picked up that the riskscore for Group C1 was higher than that of Group C2 (Fig. 5F). Due to the higher riskscore of C1 subtype patients compared to C2 subtype patients, this may be the reason why the prognosis of C2 subtype patients is higher than that of C1 subtype patients.

Fig. 5.

Lasso regression analysis. A, B LASSO analysis model construction. C impact of riskscore genes on patient prognosis. D survival analysis of high and low riskscore groups. E ROC curve analysis. F comparison of riskscore between the groups C1 and C2 of patients.****P < 0.0001

Survival analysis of prognostic FRMGs

We investigated the effects of the IL1B, SNCA, and FH genes on the prognosis of patients with osteosarcoma. Patients with higher FH gene expression in osteosarcoma had a worse prognosis, whereas patients with higher IL1B and SNCA gene expression had a better prognosis (Fig. 6A–F). We observed similar results in the validation set GSE39058.

Fig. 6.

Survival analysis. A–F Impact of IL1B, SNCA, and FH genes on the prognosis of patients with osteosarcoma

Comparing RiskScore in with and without osteosarcoma metastasis

We found that the expression levels of SNCA and IL1B in patients with osteosarcoma metastasis were significantly lower than those in patients without metastasis, while the expression level and risk score of FH in patients with osteosarcoma metastasis were significantly higher than those in patients without metastasis (Fig. 7A–D).

Fig. 7.

Compare the differences in gene expression levels and riskscore. A–D Compare the expression levels of SNCA, IL1B, FH, and riskscore between patients with osteosarcoma metastasis and non-metastasis. **P < 0.01, ***p < 0.001, ****P < 0.0001

Single cell sequencing analysis of patients with osteosarcoma

Currently, many researchers consider single-cell analysis as a necessary means of studying the relationship between genes and the immune microenvironment [26–28]. Therefore, We use the TISCH2 database to analyze single cells. We identified 29 populations and 8 cell subpopulations in the osteosarcoma single-cell dataset GSE162454 (Fig. 8A, B). (Fig. 8C) depicts the proportions of eight types of cells. (Fig. 8D) shows the proportion of eight types of cells in each sample of the single-cell dataset GSE162454. We found that IL1B had the highest expression in monocytes/macrophages (Fig. 8E).

Fig. 8.

Single cell analysis. A–D Clustering, grouping, annotation analysis. E expression analysis of IL1B, SNCA, and FH genes in different cell subgroups in the tumor microenvironment of osteosarcoma patients

Correlation analysis between riskscore and immune cell infiltration

We investigated the association between the SNCA, IL1B, FH, riskscore and immune cell abundance in osteosarcoma. We found that the expression of SNCA is positively correlated with the infiltration of T_cells_follicular_helper cells and negatively correlated with the infiltration of Macrophages_M0 cells. The expression of IL1B is positively correlated with the infiltration of T_cells_regulatory_(Tregs) cells and negatively correlated with the infiltration of T_cells_CD8 cells. The expression level of FH is negatively correlated with the infiltration of T_cells_CD8 cells and Monocytes cells. Riskscore is positively correlated with infiltration of Macrophages_M0 cells and negatively correlated with infiltration of T_cells_CD8 cells, T_cells_follicular_helper cells, and Monocytes cells (Fig. 9).

Fig. 9.

Correlation analysis. (Fig. 8) Correlation analysis between riskscore, expression levels of SNCA, IL1B, and FH, and degree of immune cell infiltration. *P < 0.05, **P < 0.01, ***p < 0.001, ****P < 0.0001

Correlation analysis between riskscore and ESTIMATE SCORE

SNCA and IL1B expression levels correlate positively with ESTIMATEScore, ImmuneScore, and StromalScore, but adversely with TumorPurity cell infiltration (Fig. 10A, B). FH’s expression level and riskscore are positively connected with TumorPurity E but negatively correlated with ESTIMATESCore, ImmuneScore, and StromalScore (Fig. 10C, D).

Fig. 10.

Correlation analysis. A–D Correlation analysis between riskscore, expression levels of SNCA, IL1B, and FH, and ESTIMATE SCORE. *P < 0.05, **P < 0.01, ***p < 0.001, ****P < 0.0001

Drug sensitivity analysis

Using Oncopredict, we determined the Durg sensitivity of patients with osteosarcomas. We found that patients with low riskscore had significantly higher sensitivity to the drugs GSK461364, BI.2536, SB743921, paclitaxel, leptospirin.B, vincristine and than patients with high riskscore osteosarcoma (Fig. 11A–F).

Fig. 11.

Drug sensitivity analysis. A–F Analysis of drug sensitivity in the high- and low riskscore groups of patients with osteosarcoma.**P < 0.01, ***p < 0.001, ****P < 0.0001

Construction nomogram

We included gender, age, and riskscore of osteosarcoma patients to construct a nomogram to predict the 36 and 60-month survival of osteosarcoma patients (Fig. 12A). The calibration curve indicates strong model consistency (Fig. 12B). The ROC AUC values for the 36 and 60-month survival periods of the model are 0.79 and 0.84. suggesting that the model is accurate. The calibration curve indicates strong model consistency (Fig. 12C). This result indicates that the ROC model is highly accurate. The risk score is statistically significant in the model variables, while the age and gender of the patients are not statistically significant, indicating that the risk score is an independent risk factor affecting the survival rates of patients at 36 months and 60 months. These results suggest that clinical doctors should detect risk genes (IL1B, SNCA, FH) to focus on the 36 month and 60 month survival rates of patients, without paying too much attention to gender and age differences.

Fig. 12.

Nomogram analysis of the riskscore in patients with osteosarcoma. A Nomogram, B calibration curve, and C ROC curve. ⁻P > 0.05, * P < 0.05

Discussion

We first created FRMGs to predict prognosis, immune cell infiltration, and drug sensitivity in osteosarcoma patients. We performed consensus clustering on FRMGs and divided them into C1 and C2 groups to determine their expression. Similarly, we found that patients with osteosarcoma in group C2 had higher levels of immune cell infiltration and better survival prognosis than those in group C1. Furthermore, we developed a Lasso regression model to analyze and construct the riskscore genes of patients with osteosarcoma and evaluated their prognosis and drug sensitivity. These findings enhance our understanding of the regulation of distant metastasis of osteosarcoma by ferroptosis genes and provide critical support for improving the accuracy of prognosis and optimizing clinical treatment for patients with osteosarcoma.

Through DEGs enrichment analysis based on C1 and C2, we discovered that DEGs are related to the T lymphocytes pathway and its association. T lymphocytes are classified primarily into CD4(+)T and CD8(+)T subsets. Research has reported that ferroptosis can also regulate the activity of T cells. For example, compound N6F11 specifically degrades GPX4 to induce ferroptosis in a mouse model of pancreatic cancer, thus triggering an anti-tumor immune response [29]. Bevacizumab induces ferroptosis and enhances CD8 T cell immune activity in liver cancer by regulating HAT1 and increasing IL-9 [30]. The delivery of miR-21-3p nanoparticles promotes ferroptosis, making melanoma sensitive to PD-1 immunotherapy [31]. Based on this, our constructed FRMGs related typing may predict the prognosis of osteosarcoma patients through the T cell pathway.

We found in drug sensitivity analysis that low riskscore patients with osteosarcoma are more sensitive to tumor chemotherapy drugs paclitaxel and vincristine. Research has shown that the activation drug of PSME2, irinotecan, can synergistically promote the death of osteosarcoma cells when used in combination with the chemotherapy drug paclitaxel [32]. Through single-cell sequencing and network pharmacology analysis, vincristine may form drug targets with RUNX2, OMD, and CD4 for the treatment of osteosarcoma [33]. These results indicate that the riskscore we established may have some feasibility in indirectly predicting the prognosis of osteosarcoma patients by predicting drug sensitivity. In addition, the IL1B, SNCA, and FH genes that we have identified as constituting riskscore may become key targets for enhancing drug sensitivity in osteosarcoma.

Existing research indicates that macrophages secrete interleukin-1B (IL1B), which decreases osteosarcoma’s sensitivity to neoadjuvant chemotherapy treatments [34]. SNCA blocks EMT and metastasis in breast cancer, and its expression may help predict the chemical sensitivity of breast cancer cells and evaluate their immune microenvironment [35]. Knocking down the neutrophil extracellular trap gene FH inhibits osteosarcoma metastasis and recurrence [36]. Furthermore, we discovered that individuals with osteosarcoma who expressed high levels of FH had a better prognosis than those who expressed high levels of IL1B and SNCA. According to these investigations, ferroptosis-related metastasis genes may be a good indicator of an osteosarcoma patient’s prognosis. However, the mechanisms by which these genes regulate osteosarcoma metastasis require further validation through in vitro and in vivo experiments.

Research has shown that ferroptosis can to some extent affect the sensitivity of osteosarcoma patients to drugs. NGR modified CAF derived targeted tumor vascular system induces ferroptosis and overcomes chemotherapy resistance in osteosarcoma [37], Eriodictyol-cisplatin coated nanomedicine synergistically promote osteosarcoma cells ferroptosis and chemosensitivity [38], The induction of ferroptosis by impairing STAT3/Nrf2/GPx4 signaling enhances the sensitivity of osteosarcoma cells to cisplatin [39].

GPX4 is easily post-translational modified by fumarate accumulated under FH conditions, leading to a decrease in its activity and inducing ferroptosis [40]. In FH deficient cells, the remaining acetyl CoA after TCA cycle is used for fatty acid synthesis. The peroxidation of membrane phospholipids by polyunsaturated fatty acids can lead to ferroptosis. Research has shown that the abundance and localization of polyunsaturated fatty acids in cells can affect the degree of lipid peroxidation, thereby affecting ferroptosisi [41, 42]. In FH deficient cells, fumarate salts complexed with GSH accumulate, depleting NADPH and ultimately increasing ROS levels [43]. FH deficient cells increase the antioxidant transcription factor NRF2 [44], and tumor cells lacking NRF2 are sensitive to ferroptosis [45].

Ferroptosis plays an extremely important role in regulating immune cell responses. Research has shown that cancer cells undergoing ferroptosis hinder dendritic cell-mediated anti-tumor immunity [46], and the occurrence of fesrroptosis significantly reduces the number of myeloid derived suppressor cells (MDSCs) and tumor associated M2 like macrophages (M2 TAMs) in the tumor microenvironment. At the same time, there was an increase in CD4 and CD8 T cells infiltrating the tumor [47]. Mitochondrial targeted brequinar liposomes enhance mitochondrial associated ferrotosis to promote blocking immunotherapy at bladder cancer checkpoint [48]. Dihydroartemisinin inhibits pancreatic cell growth by inducing ferroptosis and activating immune cell populations of CD8 T cells, NK cells, and NKT cells [49].

At present, the international mainstream treatment for osteosarcoma is centered around the “Methotrexate+Doxorubicin+Cisplatin” (MAP regimen), and some patients may combine with ifosfamide (IE regimen) due to tumor progression or resistance. However, the drug sensitivity analysis in this manuscript did not screen for the above-mentioned chemotherapy drugs. Previous studies have shown that in a retrospective study of osteosarcoma patients receiving standard chemotherapy (SC, n = 104, doxorubicin+cisplatin+ifosfamide and high-dose methotrexate) and cisplatin+paclitaxel+high-dose methotrexate (CPM, n = 104), it was found that compared with the SC group, the CPM group had significantly prolonged PFS (16.85 ± 3.40 months vs. 15.72 ± 3.21 months, P = 0.015) and a higher 5-year OS rate (54.81% vs. 40.38%, P = 0.037) [50]. Other studies have shown that, ¹²⁴I-labelled BMSC-Derived Extracellular Vesicles Deliver CRISPR/Cas9 Ribonucleoproteins with a GFP-Reporter System to Inhibit Osteosarcoma Proliferation and Metastasis [51], There is a KCNJ2/HIF1 α positive feedback loop in osteosarcoma tissue, which promotes the metastasis of osteosarcoma cells [52]. These studies contribute to the diagnosis and treatment of OS. Bioinformatics is also a commonly used method for predicting the prognosis of osteosarcoma, The hexosamine biosynthesis pathway-related gene signature correlates with immune infiltration and predicts prognosis of patients with osteosarcoma [53], Identification of endothelial INSR as an osteosarcoma-related biomarker and therapeutic target based on weighted gene co-expression network analysis [54]. Similarly, this manuscript supplements the diagnosis and treatment of osteosarcoma using bioinformatics methods. Therefore, based on our model, low-risk bone cancer patients were found to have higher sensitivity to paclitaxel than high-risk patients. This suggests that using chemotherapy regimens containing paclitaxel based on risk scores instead of standard chemotherapy regimens may be more beneficial in improving patient survival.

The model we constructed can effectively predict the prognosis of osteosarcoma patients and to some extent know the chemotherapy and immunotherapy drugs used in osteosarcoma patients. In summary, our model can provide new treatment ideas for patients with osteosarcoma in clinical practice.

Conclusion

Ferroptosis-related metastatic genes predict an osteosarcoma patient’s prognosis, immune cell infiltration, and treatment susceptibility. This study provides a unique and useful reference for the clinical management of osteosarcoma.

Limitation

This manuscript is an analytical study, therefore we only used information from osteosarcoma patients in public databases to construct a prognostic model for osteosarcoma and draw preliminary conclusions. Due to the extremely small sample size of osteosarcoma patients in our hospital, there is no suitable sample size for clinical validation. In addition, limitations in the experimental site, equipment, and technology have hindered us from making further progress in in vitro experiments. At the same time, the insufficient sample size in hospitals may lead to the risk of overfitting in the model, and the lack of additional experiments may also limit the promotion of the model. We have to admit that bioinformatics analysis alone, even coupled with validation datasets, cannot fully substitute for the direct confirmation of molecular mechanisms that functional experiments provide. Experimental validation is crucial for establishing reliable causal relationships and supporting the robustness of the conclusions. Our goal is to collaborate with multiple hospitals and research institutions in future research to collect clinical information from patients with osteosarcoma for further validation studies.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Da Lei analyzed the data and wrote the manuscript, Peiyou Zhao conducted the statistical analysis and wrote the manuscript, Xinwei Chen Funding support and wrote the manuscript, Yanan Yang designed the study. Shun Gao Funding support and designed the study.

Funding

This study was supported by Guiding Project of Meishan Science and Technology Bureau (project number is 2023KJZD065).

Data availability

The osteosarcoma data comes from GSE21257 and GSE39058 in the GEO database (https://www.ncbi.nlm.nih.gov/geo/).

Declarations

Ethics approval and consent to participate

The data used in this article for ethical approval and participation consent are from public databases and therefore do not involve ethics. Not applicable.

Consent for publication

The consent form is not applicable.

Competing interests

The authors declare no competing interests.