Introduction

Osteosarcoma (OS), as the most prevalent tumor of bone origin, originates from primitive mesenchymal cells and mainly occurs in children and adolescents1,2. Chemotherapy, as the indispensable treatment option, has demonstrated a positive effect on the majority of patients with OS, but once chemoresistance strikes, it can lead to a dramatic reduction in patient survival3,4. The high susceptibility to chemotherapy resistance is also responsible for the lack of significant improvement in survival rates among OS patients over the past 20 years5. The specific mechanisms involved in the development of OS chemoresistance are not fully clarified, and protocols to address OS chemoresistance are urgently needed6,7.

Due to the fact that majority of tumor treatment regimens function by targeting apoptotic tumor cells, once tumor cells undergo apoptotic escape, treatment resistance will develop accordingly, leading to a disastrous prognosis8,9. Consequently, non-apoptotic forms of regulated cell death (RCD) associated with tumor therapy have come into the limelight and are gaining increased attention10. Distinct from apoptosis, ferroptosis, as a nontraditional RCD form featured by iron-dependent accumulation of lipid reactive oxygen species11, has been revealed to function as a pivotal role in tumor chemoresistance12. In particular, for OS, we identified through previous studies that ferroptosis may be tightly correlated with OS chemoresistance13. Ferroptosis is considered to be of great potential in antagonizing chemoresistance for OS14, however, further studies on the mechanisms associated with the action of ferroptosis on OS chemoresistance and the essential molecules involved are extremely rare.

In the present study, we obtained the list of ferroptosis-related genes and the characteristic of corresponding genes from the FerrDb database. In combination with the Gene Expression Omnibus (GEO) database, ferroptosis-related differentially expressed genes (DEGs) were acquired and followed by enrichment analysis and protein–protein interaction (PPI) network construction. Furthermore, hub genes with prognostic significance were identified through the TARGET database. The aim was to target critical ferroptosis-related DEGs by combining their own ferroptosis characteristics, expression characteristics and prognostic characteristics, and eventually construct an efficient prediction model whose prognostic value was individually validated by an independent cohort.

Materials and methods

Data sources and differential analysis

Through the GEO database (https://www.ncbi.nlm.nih.gov/geo/), we obtained RNA-seq data of 21 OS samples from the GSE87437 dataset based on GPL570 platform for differential analysis, including 11 chemoresistance samples and 10 non-chemoresistance ones. GEO2R (http://www.ncbi.nlm.nih.gov/geo/geo2r) was used to screen for DEGs meeting the criteria of p < 0.1, | log fold change (FC)|> 0.5. FerrDb (http://www.zhounan.org/ferrdb) is the first manually managed ferroptosis database covering regulatory factors and molecular markers for ferroptosis and ferroptosis-related diseases15. From the FerrDb database, 259 ferroptosis-related genes were downloaded, covering 108 drivers and 69 suppressors. In addition, 99 OS sample with survival information from the TARGET database (https://ocg.cancer.gov/ programs/target) were used for survival analysis and a prognostic model construction. For external validation of the prognostic value on the model and hub genes, through the GEO database, we obtained RNA-seq data and clinical information from the GSE21257 dataset based on GPL10295 platform, including 53 OS samples. All the above material was available from public databases and was free of ethical issue or informed consent.

Enrichment analyses and PPI networks construction

Gene Set Enrichment Analysis (GSEA), Gene Ontology (GO) enrichment analysis, Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis were conducted by the clusterProfiler package of R16. GO enrichment was applied to annotate and analyze genes involved in biological process (BP), cellular component (CC) and molecular function (MF)17. Moreover, we forecasted interactions between DEGs that achieved the combined score > 0.15 using Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database (https://string-db.org/)18 and visualized PPI networks using Cytoscape software, an open-source software for network analysis and visualization19. Top-ranked DEGs were calculated by cytoHubba20, a plug-in for Cytoscape to further screen hub genes.

Survival analyses and model construction

The median of gene expression was used to divide the low and high expression groups, and overall survival was selected as the survival parameter. Survival analyses involving log-rank test, univariate Cox regression and multivariate Cox regression were performed using the survival package of R and visualized using the survminer package of R. Variables that were significant in the univariate analysis were included in the multivariate analysis and those that were significant in the multivariate analysis were further incorporated into the prognostic model. The rms package of R was used to develop a nomogram and calibration curves in order to construct and assess the prognostic model separately21. In addition, time-dependent receiver operating characteristic (ROC) curves were structured using the timeROC package of R to validate the prognostic value of the model and hub genes.

Statistical analysis

Statistical analysis was conducted by version 3.6.3 of the R software and version 3.8.2 of the Cytoscape software. The analysis results were considered statistically significant at p < 0.05.

Results

Ferroptosis-related DEGs identification in chemoresistance OS patients

Via the GEO database, we acquired RNA-seq data of 21 OS samples from the GSE87437 dataset for differential analysis, including 11 chemoresistance samples and 10 non-chemoresistance ones. The volcano plot presented the results of differential analysis via the GEO2R (Fig. 1A). Compared to non-chemoresistance samples, there were 1292 up-regulated DEGs and 828 down-regulated DEGs in chemoresistance ones. Among them, 22 ferroptosis-related DEGs were identified, including 16 up-regulated and 6 down-regulated ones (Table 1) (Fig. 1B).

Figure 1
figure 1

Ferroptosis-related DEGs identification in chemoresistance OS patients. (A) The volcano plot of differential analysis in the GSE87437 dataset. (B) The Venn diagram obtained by intersecting the DEGs in the GSE87437 dataset and ferroptosis-related genes in the FerrDb database.

Full size image
Table 1 Ferroptosis-related DEGs in OS.
Full size table

GSEA of DEGs in chemoresistance OS patients

The results of differential analysis in the GSE87437 dataset were subjected to GSEA (Table 2). The top-three most significant-enriched gene sets based on normalized enrichment score (NES) value, which negatively correlated with the DEGs, were integrin1 pathway, assembly of collagen fibrils and other multimeric structures, response to metal ions (Fig. 2A). Besides, the top-three most significant-enriched gene sets based on NES value, which positively correlated with the DEGs, were TNFRSF members mediating non canonical NF-kappaB pathway, CTL pathway, tcytotoxic pathway (Fig. 2B).

Table 2 Top-ten most significant-enriched gene sets according to NES value ranking of GSEA results positively and negatively correlated with DEGs, respectively.
Full size table
Figure 2
figure 2

GSEA of ferroptosis-related DEGs in chemoresistance OS patients. (A) The top-three most significant-enriched gene sets that negatively correlated with the ferroptosis-related DEGs based on NES value. (B) The top-three most significant-enriched gene sets that positively correlated with the ferroptosis-related DEGs based on NES value.

Full size image

GO and KEGG enrichment analyses of ferroptosis-related DEGs

GO and KEGG enrichment analyses were performed on 22 ferroptosis-related DEGs. GO analysis suggested that these genes mainly functioned in response to iron ion, extrinsic component of membrane, and oxidoreductase activity. The top-three most significant-enriched terms in each of the BP, CC, and MF entries were identified for GO visualization network construction (Fig. 3A). In addition, KEGG analysis indicated that corresponding genes were associated with ferroptosis, autophagy, mineral absorption, arachidonic acid metabolism, and biosynthesis of amino acids. The top-five most significant-enriched pathways were identified for KEGG visualization network construction (Fig. 3B).

Figure 3
figure 3

GO and KEGG enrichment analyses of ferroptosis-related DEGs. (A) GO enrichment analysis visualization network based on the top-three most significant-enriched terms in each of the BP, CC, and MF entries. (B) KEGG enrichment analysis visualization network based on the top-five most significant-enriched pathways.

Full size image

PPI networks of ferroptosis-related DEGs

Interactions between ferroptosis-related DEGs were predicted using STRING database with a combined score > 0.15 and visualized using Cytoscape software. The initial PPI network with 20 nodes and 43 edges was developed (Fig. 4A). Furthermore, we applied cytoHubba to identify the top-ranked genes and recognized a 12-node, 24-edge PPI network based on the initial network (Fig. 4B).

Figure 4
figure 4

PPI networks of ferroptosis-related DEGs. (A) The initial PPI network covering 20 nodes and 43 edges using Cytoscape. (B) The PPI network covering 12 nodes and 24 edges after filtering by cytoHubba.

Full size image

Hub genes identification

Among these genes, we discovered that the expression of cystathionine beta-synthase (CBS), suppressor of cytokine signaling 1 (SOCS1) and epidermal growth factor receptor (EGFR) influenced the prognosis of OS patients in TARGET database. High expression of CBS (Fig. 5A), and low expression of SOCS1 (Fig. 5B), EGFR (Fig. 5C) were significantly associated with poor survival probability. Meanwhile, CBS, a suppressor of ferroptosis, was up-regulated in chemoresistance samples according to previous differential analysis, while SOCS1 and EGFR, as drivers of ferroptosis, were correspondingly down-regulated in chemoresistance samples. We targeted these three genes as hub genes for further analyses.

Figure 5
figure 5

K-M curves of hub genes for OS patients in the TARGET database. (A) High expression of CBS was significantly associated with poor survival. (B) Low expression of SOCS1 was significantly associated with poor survival. (C) Low expression of EGFR was significantly associated with poor survival.

Full size image

Expressions of CBS, SOCS1 and EGFR might be independent prognostic factors for OS patients

Through the TARGET database, clinical characteristics of OS patients (containing age, gender, race, surgery, metastasis, tumor side, tumor region, primary site progression) together with hub gene expressions were included in univariate Cox regression analysis. The results showed that metastasis, high expression of CBS, low expression of SOCS1 and EGFR were risk factors for poor prognosis in patients with OS. Furthermore, these variables with significant results were integrated into multivariate Cox regression analysis, which revealed that expressions of CBS, SOCS1 and EGFR might be independent prognostic factors for OS patients (Table 3).

Table 3 Cox regression analyses of overall survival for OS patients.
Full size table

Validation for the prognostic value of the hub genes

Internal validation was performed through the TARGET database to detect the prognostic value of the 3 hub genes with time-dependent ROC curves (Fig. 6A). Moreover, external validation for the prognostic value of CBS, SOCS1 and EGFR was implemented in an independent cohort of 53 OS samples from the GSE21257 dataset, which demonstrated the results consistent with the internal validation (Fig. 6B).

Figure 6
figure 6

Time-dependent ROC curves of hub genes for OS patients. (A) Internal validation via a cohort of 99 samples from the TARGET database. (B) External validation via a cohort of 53 samples from the GSE21257 dataset.

Full size image

A prognostic model construction and evaluation

Variables with significant results in multivariate Cox regression analysis were incorporated into the construction of the prognostic model. Metastasis and expressions of CBS, SOCS1, EGFR were aggregated in a nomogram to predict 1-, 3-, and 5-year survival probability in patients with OS (Fig. 7A). The C-index of this prognostic model reached 0.788 (0.756–0.820), and the predicted and actual outcomes in the 1-, 3-, and 5-year calibration curves were nearly identical (Fig. 7B).

Figure 7
figure 7

Visualization of the prognostic model for OS patients. (A) Construction of a nomogram to predict 1-, 3-, and 5-year survival probability. (B) Construction of 1-, 3-, and 5-year calibration curves to evaluate predictive efficacy of the nomogram.

Full size image

Validation for the prognostic value of the model

An independent cohort of 53 OS samples from the GSE21257 dataset was conducted to externally validate the model. The time-dependent ROC curve demonstrated the model's outstanding predictive efficacy for prognosis of patients with OS, and was higher than that of individual hub genes (Fig. 8).

Figure 8
figure 8

Based on an independent external validation cohort of 53 OS samples from the GSE21257 dataset, the time-dependent ROC curve demonstrated the model's outstanding predictive efficacy for prognosis of patients with OS.

Full size image

Discussion

We have previously reported that ferroptosis can impact chemotherapy resistance in patients with OS13. Chemotherapy, a common management for cancer, provides varying degrees of therapeutic efficacy for most human cancers22. However, since the majority of the treatment regimens involved target the apoptotic effect of tumor cells, once apoptotic escape of tumor cells occurs, it can lead to a reduction in the sensitivity of the original treatment and thereby resulting in a poor prognosis23. Distinguished from apoptosis, ferroptosis, as an emerging type of RCD with unique morphological and biological characteristics, is recognized as promising in modifying cancer chemoresistance and has demonstrated corresponding value in OS chemotherapy24. Fu et al. observed that OS cells that underwent ferroptosis behaved more sensitively to adriamycin treatment25. Liu et al. revealed a significant decrease of ferroptosis activity in cisplatin-resistant OS cells, while non-resistant ones gradually developed a hyposensitivity to cisplatin along with the addition of ferroptosis inhibitors, and this hyposensitivity could be ameliorated by ferroptosis agonists12. In the present study, we identified 22 ferroptosis-related DEGs between chemoresistance and non-chemoresistance samples of OS, from which CBS, COCS1, EGFR were detected to be associated with overall survival in OS patients, and thus they were classified as hub genes for further study to explore the potential molecular mechanism and prognostic value. The aim was to target critical ferroptosis-related DEGs by combining their own ferroptosis characteristics, expression characteristics and prognostic characteristics, and eventually construct an efficient prediction model whose prognostic value was individually validated by an independent cohort.

Through the TARGET database, high expression of CBS and low expression of SOCS1, EGFR suggested poor survival probability for patients with OS. CBS, shown to be up-regulated in chemoresistant samples by differential analysis, is regarded as a suppressor of ferroptosis26,27. CBS is the rate-limiting enzyme as well as the first enzyme in the transsulfuration pathway, which serves a critical role for humans in the maintenance of health and the development of disease28. The typical biochemical function of CBS under physiological conditions is to catalyze the transition of serine and homocysteine to cystathionine and water. Additionally, CBS can also catalyze substitution reactions which generate hydrogen sulfide29,30. Suppression of CBS has been reported to initiate ferroptosis in hepatocellular carcinoma27. A similar response was measured by Liu et al. in ovarian cancer cells, whereby knockdown of CBS enhanced ferroptosis susceptibility of cancer cells, while overexpression of CBS correspondingly promoted ferroptosis resistance26. In addition, SOCS1 and EGFR, shown to be down-regulated in chemoresistant samples by differential analysis, are considered as drivers of ferroptosis31,32. SOCS1 belongs to a family of 8 intracellular proteins that under physiological conditions limit type I Interferon and Interferon-γ receptor activation, serving as a classical negative feedback loop regulator of the Janus kinase and – signal transducer and activator of transcription pathway33,34. SOCS1 has exhibited significant potential in multiple diseases, and its related gene therapy and biologic application are active fields of study worldwide35. For a wide range of cancers, SOCS1 is recognized as a tumor suppressor and may act in a cell context-dependent manner36. As for OS, it has been confirmed that expression of SOCS1 can sensitize U-2 OS cells to ferroptosis inducer, which may be linked to the reduced glutathione level32. EGFR is a transmembrane glycoprotein belonging to the tyrosine kinase receptor family, whose binding to cognate ligands result in its autophosphorylation and subsequent activation of signal transduction pathways involved in regulating cell proliferation, differentiation and survival37,38. In terms of EGFR in ferroptosis, Poursaitidis et al. discovered that cell death in activated EGFR mutant cells occurred by ferroptosis and repression of EGFR signaling pathway corresponded to the rescue of cell viability31. Apart from its broad role in the genesis and progression of tumors, EGFR also influences therapeutic efficacy and is engaged in chemoresistance, which has recently been described as involving ferroptosis effect39. Combining the results of hub genes in differential analysis, the actions of hub genes in ferroptosis, and the prognostic relevance of hub genes in patients, we considered that CBS, SOCS1 and EGFR might play essential roles in OS and its chemoresistance with potential research and clinical value.

Furthermore, we included hub gene expressions along with clinical variables for patients with OS in Cox regression analyses and identified variables with statistically significant results, containing expressions of CBS, SOCS1, EGFR and metastasis, for incorporation into the construction of a prognostic model. OS tends to indicate a high malignancy degree and a poor survival period, and given its complexity and heterogeneity, accurate prediction of patient prognosis is in high demand40. More clinical models of OS have been reported in successive studies. Cao et al. developed the prognostic model for OS on the basis of immune-related genes41. Zhang et al. used genes identified by co-expression network analysis as the predictive tool for OS lung metastasis42. Xiao et al. assessed the prognosis of OS patients in terms of key macrophage-associated genes43. In the present study, the value of both the model and the adopted genes were validated in an independent cohort, and the near overlap between predicted and actual outcomes in the 1-, 3-, 5-year calibration curves suggested high efficiency. This is one of the few studies on OS chemoresistance involving ferroptosis and the first to construct the prognostic model for OS patients with ferroptosis-related genes. However, there remain certain limitations to our study. On the one hand, the low incidence of OS and the restrictive condition of chemoresistance have led to a relatively small number of samples for differential analysis, which may result in the omission of some valuable genes. On the other, numerous clinical studies as well as cellular and animal experiments are required to support our findings.

In conclusion, we revealed CBS, COCS1, and EGFR as ferroptosis-related genes that might be critically involved in OS chemoresistance and applied them to construct an efficient predictive model for the prognosis of OS patients.