Osteosarcoma (OS) is the most widespread main bone malignancy in children and young adults (Rothzerg et al., 2020) as well as makes up for nearly two-thirds of the primary bone malignancies diagnosed in the first 20 years of one’s life (Yan et al., 2016; Chen et al., 2019). The combination of surgical resection skills and multiagent chemotherapy has extremely enhanced the prognosis of osteosarcoma patients (Bielack et al., 2002). Other studies have shown that over the last 30 years, neoadjuvant chemotherapy and surgical techniques have developed, and the survival rate of 5 years in these patients has increased to ∼70% (Bielack et al., 2002; Niu et al., 2019; Wakamatsu et al., 2019). On the other hand, the overall survival rate of 5 years in cases diagnosed with early pulmonary metastasis is less than 20%. Besides, another study shows that 20% of patients without early pulmonary metastasis have local recurrence or will eventually develop pulmonary metastasis (Bishop et al., 2016). However, the fact that these patients develop resistance to chemotherapy is a major challenge. Therefore, it is imperative to explore novel prognostic models to further improve the survival of osteosarcoma patients.

Cancer stem cells (CSCs), a small, special class of cells inside tumors, owns the capability of self-renewal and differentiation of different cell types that assemble the whole tumor (Reya et al., 2001). CSCs are tumor-derived cells that mutate from stem cells of adults and produce distinct subtypes of tumor cells via self-renewal and differentiation (van der Heijden and Vermeulen, 2019). CSCs were associated with the development, recurrence, metastasis, and resistance of cancers and facilitate tumor progression (Dawood et al., 2014; Nassar and Blanpain, 2016). The capability of CSCs to originate and maintain the proliferation of cancer cells is an essential factor for metastasis. CSCs are heterogeneous and own subgroups with metastatic ability (Nandy and Lakshmanaswamy, 2017; Aydemir Coban and Sahin, 2018). Besides, CSCs can stay in prolonged dormancy and have varieties of molecular mechanisms of chemotherapy resistance (Lytle et al., 2018). As CSCs act as crucial roles in tumor progression, CSC biomarkers and related pathways may become the prognostic biomarkers and underlying targets for treatment of osteosarcoma patients.

Herein, we mined the expression data of genes and the relevant clinical information of osteosarcoma samples, and CSC-related genes from the opening databases. Next, the prognostic CSC-related gene signature was built in the Therapeutically Applicable Research to Generate Effective Treatments (TARGET) cohort and then evaluated it in the Gene Expression Omnibus (GEO) dataset. Last, we explored the potential mechanisms underlying CSC-related genes using a functional enrichment analysis.

Figure 1 demonstrates a flow chart that explains the study. In this study, we mined data for a total of 95 patients with osteosarcoma and who had complete clinical information from the TARGET database, 15 normal samples, and 84 osteosarcoma samples in GSE33383 from the GEO database. Besides, the GSE21257 served as the validating cohort.

Osteosarcoma is a common malignant tumor of the bone with an elevated tendency of invasion and metastasis (Chen et al., 2021). Although there are multiple advances in comprehensive therapies for osteosarcoma, the prognosis of osteosarcoma patients is still poor. Thus, exploring the valid prognostic biomarkers and potential therapeutic targets is very critical for improving the prognosis of osteosarcoma. CSCs are a set of heterogeneous cells that own distinct differentiation abilities. Very little subpopulations of cancer cell have properties of stem cell in tumor tissues, but the potential of differentiation and proliferation as well as the abilities of self-renewal are the derived factors for tumorigenesis, development, invasion, recurrence, metastasis, and resistance. In this study, we explored the expression profile of 413 CSC-related genes between normal and osteosarcoma cohorts and obtained 210 genes that were differentially expressed among normal and osteosarcoma tissues in the GEO dataset. Then, we employed the univariate Cox regression analysis to explore the correlation between 210 differentially expressed CSC-related genes and the prognosis of osteosarcoma patients. Finally, an eight CSC-related gene signature was constructed by LASSO and multivariate Cox analysis. The curves of K-M and ROC confirmed that the predictive performance of the signature is stable and consistent with the TARGET internal and GEO external validation sets, and it was remarkably associated with the clinicopathological features and matrix score. What is more, the Cox analyses indicated that the eight-gene signature had an independent predictive capability for prognosis in two distinct cohorts. All of the results showed that the eight-gene signature had a significant potential in clinical application.

In the eight-gene signature identified and validated on the foundation of CSC-related genes, DLL1, FOLR1, and MEF2C were risk variables, while EOMES, ERCC2, PSMA5, PTN, and SPI1 were protective factors. Delta-like ligand 1 was highly expressed in metastatic osteosarcoma cells compared with osteoblasts and non-metastatic osteosarcoma cells (Hughes, 2009). Besides, Pu et al. (2017) found that the expression of DLL1 is negatively associated with multichemoresistance of osteosarcoma. MiR-34a-5p enhances multichemoresistance of osteosarcoma by repressing DLL1. Folate receptor 1 is overexpressed in varieties of solid tumors (Necela et al., 2015; Huang et al., 2018; Kim et al., 2018) and is closely associated with the prognoses of multiple malignant tumors, such as triple-negative breast cancer (Ginter et al., 2017), ovarian cancer (Kobel et al., 2014), and non-small-cell lung cancer (Iwakiri et al., 2008). Myocyte enhancer factor 2C was overexpressed in hepatocellular carcinoma. MEF2C-mediated VEGF-induced vasculogenic mimicry, angiogenesis, and invasion, as well as inhibition of β-catenin-induced tumor growth (Bai et al., 2015). Besides, MEF2C was upregulated in gefitinib-resistant cancer tissues and associated with gefitinib resistance in hepatic cancer cells (Zhang et al., 2018). He et al. (2021) found that EOMES was down expressed in tissue samples and related to the advanced stage and poor prognosis of hepatocellular carcinoma. Moreover, EOMES is associated with the disease-free survival time in gastric cancer (Cui et al., 2020) and also associated with better overall survival in triple-negative breast cancer (Wang et al., 2018). The gene polymorphisms of ERCC2 and ERCC1 were related to chemotherapy sensitivity and overall survival in osteosarcoma patients (Cao et al., 2015). PSMA5 is dysregulated in prostate cancer and associated with bortezomib resistance as well as disease-free survival (Fu et al., 2019). PTN is highly expressed in osteosarcoma, and by target PTN, miR-627-3p inhibits proliferation and migration of osteosarcoma cells (He et al., 2019). LncRNA OIP5-AS1 is related to doxorubicin resistance of osteosarcoma via mediating miR-137-3p/PTN axis (Sun et al., 2020). Besides, the whole transcriptome RNA-seq also proved that PTN is differentially expressed in osteosarcoma (Reimann et al., 2014). Bioinformatics analysis showed that SPI1 was a prognosis-related gene in esophageal squamous cell carcinoma (Ho et al., 2017; Zhang et al., 2020). Most of these genes were related to prognosis in the other cancers, but whether these genes relate to the prognosis of osteosarcoma by altering the molecular mechanism of stem cells still is unclear because only a few correlational researches about these genes have been described.

Next, we assessed the association between risk score, gender, age, and metastasis in osteosarcoma patients through Cox regression analysis. All of the outcomes demonstrated that the risk score could be the independent prognostic variables in patients with osteosarcoma. Besides, based on the risk score, age, gender, race, tumor site, and metastasis, we established a nomogram for forecasting the prognosis of the osteosarcoma patients. From this nomogram, we were able to predict the 1-, 3-, and 5-year survival of osteosarcoma cases. The calibration plots verified that the signature could exactly appraise the survival of the osteosarcoma patients. Finally, in the past years, the underlying mechanisms of tumor susceptibility to stem cells have become a hot topic for research. On the foundation of the DEGs between two risk groups (low and high), we performed GO and KEGG analysis, which demonstrated that several tumor-related biological functions were very affluent, such as the MAPK signaling pathway, Ras signaling pathway, chemokine signaling pathway, and apoptosis.

Despite the many advantages outlined above, our study has some drawbacks. First, there was lack of adequate cases and other clinicopathological information. Second, the performance of our prognostic signature lacked validation in more independent databases. Last, all of our results were on the foundation of the existing opening databases and lacked basic experimental verification as well as clinical information support.

In summary, we established a new prognostic signature based on eight stem cell-related genes. It proves that it is an independent factor that correlates with overall survival, hence, offering a novel orientation to forecast the prognosis of the osteosarcoma patients.

The datasets presented in this study can be found in online repositories. The names of the repository/repositories and accession number(s) can be found in the article/supplementary material.

BX designed the research study. BX, LL, ZC, YX, and YZ performed the literature search and statistical analysis. BX, AL, YX, PW, and CX interpreted the data and drafted the manuscript. BX and HL critically revised the manuscript. All authors read and approved the final manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.