OS is the most common primary malignant tumor of bone, and the susceptible population is adolescents. Its prognosis is very poor, and early metastases often occur. 20% of patients died of tumor metastasis or unresectable tumors, and the remaining 80% of patients had small metastases at the time of diagnosis. Many patients develop lung metastases within 1 year, and the 5-year survival rate is only 15%. Like many other malignancies, its etiology remains unknown. Recently, a large number of new diagnostic techniques and effective chemotherapy methods have been developed, and the current 5-year survival rate has risen to 55%-70%. Preoperative adjuvant chemotherapy followed by radical resection is still the most effective treatment. If surgical resection is not possible, radiotherapy may be beneficial for controlling local tumors. As a generality, metastasis is the most adverse factors at diagnosis among known prognostic factors . There are large differences in survival between patients with metastatic OS (10–20%) and non-metastatic OS (50–78%) [26, 27]. Moreover, metastatic OS are still very difficult to control and there are few effective therapeutic targets. Kinase targets, immune checkpoint inhibitors and cell surface marker GD2 have been actively investigated in multiple current clinical trials, but are inadequately evaluated. Therefore, further studies on early diagnosis or prediction of metastasis are warranted.
In our study, multiple bioinformatics analysis tools were used to identify 4 key genes related to metastasis and prognosis of OS patients, thus we constructed a risk score model which may benefit the treatment and prognosis evaluation of OS.
Using GO, KEGG and GSEA, we annotated the function of genes in the key module most related with metastasis, and clarified the underlying mechanism of metastasis in OS. Our results revealed that these genes were found to be enriched in cell cycle, cell proliferation, cell migration and immune response. Some researchers had demonstrated the functional link between cell cycle disorder and cancer cell invasion and metastasis [13, 23, 24]. Several small pilot studies have reported that expression of molecules of tumor cell immune response, particularly HLA class II, can induce anti-tumor T cell responses, which may affect tumor progression and survival time of patients [25–27]. Hence, we suggested that genes in blue module probably involved in the development and metastasis of OS through cell cycle pathway and immune response pathway.
After screening and filtering, we obtained four genes that may predict OS metastasis and have prognostic effects, and were evaluated in a cox regression model, indicating that it is an independent prognostic factor. The 4 key genes consist of ALOX5AP, HLA-DMB, HLA-DRA and SPINT2. ALOX5AP, also called 5-LO-activating protein (FLAP), which plays an important role in Synthesis of leukotriene and associates with prognosis of primary neuroblastoma patients  and esophageal squamous cell carcinoma patients . HLA-DMB, one of the HLA class II beta chain paralogues, is expressed in antigen-presenting cells. Previous studies have confirmed that mRNA and protein levels of HLA-DMB are highly expressed in tumor samples from patients with advanced serous ovarian cancer with a large number of tumor-penetrating CD8 T lymphocytes, which can significantly prolong the median survival time . HLA-DRA, a component of MHC II, alpha chain paralogues, also are expressed in antigen presenting cells. Both at transcription and protein levels, reduced expression of HLA-DRA has been shown to predict poor overall survival and progression-free survival in diffusive large B-cell lymphoma . Moreover, enrichment analysis revealed up-regulation of immune response gene sets, including antigen presentation (HLA-DMB and HLA-DRA). SPINT2 gene expression was down-regulated, altering dysregulation of the HGF/MET signaling pathway, which contributes to cancer development and progression [32–34]. Whether these genes play the same role in the development and metastasis of OS deserves further study.
However, there were still some limitations in our work. Firstly, there are relatively small numbers of patients in two datasets obtained from publicly available database. In order to verify the stability and accuracy of the risk prediction model, more expression data and corresponding clinical information need to be collected, especially independent cohorts from multiple centers to further evaluate the applicability of the model. Secondly, our analysis is completely based on bioinformatics analysis, we need to accumulate more comprehensive experimental evidence, including in vivo and in vitro experiments Secondly, our analysis was entirely based on bioinformatics analysis to clarify the effect and possible molecular mechanisms of 4 genes on OS.