1 Introduction

Osteosarcomas (OS) is the most popular primary malignancy that develops in bone and defined by the existing malignant mesenchymal cells generating immature or osteoid bone [1]. The incidence of OS in children and adolescence (especially 10–14 years old) [2] is quietly high that may be due to the rapid bone turnover and growth and its annual incidence peaks are at 8–11/million/year worldwide [3],. Osteosarcomas often occur in the long bones of extremities, near the metaphyseal growth plate including the humerus, femur and tibia, and less commonly in the pelvis, skull and jaw [4]. Following surgery, patients receive neo-adjuvant chemotherapy and chemotherapy treatment with a cocktail of chemotherapies [5], for instance, high-dose methotrexate (12 g/m2), ifosfamide and etoposide for younger patients in the sarcome-09 study [6], or combination of cisplatin, ifosfamide and doxorubicin with or without high-dose methotrexate [7]. This has been the standard of care for OS patients since the introduction of chemotherapies in the 1970s and the use of these therapeutic regimens has increased the overall 5-years survival rate of patients that had localized disease from 20 to 78% and the 10-year survival from 30–50% [8], but unfortunately, 20–30% patients are the recurrent and metastatic cases and their 5-year survival rate is less than 25% [9]. However, with the updated management strategies, such as several anti-carcinogens clinically application, the survival rate has not obviously improved for patients who has metastases or not over the last few decades [10]. A major reason of this poor prognosis is the high levels of tumor heterogeneity existing in the OS patients, and the complexity genetic and molecular mechanisms of osteosarcoma agenesis makes it quietly difficult to develop a singular effective therapeutic method in clinical practice. In addition, the stagnant survival rates of OS patients also indicate an urgent need for the better understanding of this disease, and the developing an effective approach of diagnosis and treatment by using the multiple combined, modern and interdisciplinary therapeutic regimens [11].

As a type of conserved serine/threonine kinase, mechanistic target of rapamycin kinase complex 1 (mTORC1) is mainly involved in the cell growth, proliferation, migration, immune responses, survival, autophagy, and metabolism regulation for maintaining cellular homeostasis [12]. MTORC1 forms two functionally distinct and structurally multi-subunit complexes of mTORC1 and mTORC2 that support mTOR signaling cascade [13]. mTORC1 consists of mLST8, Raptor, mTOR, PRAS40 and DEPTOR protein [14], and positively controls cell growth by stimulating protein and lipid synthesis, and its main downstream targets include the 4E binding protein 1 (4EBP1), insulin growth factor receptor (IGF-1R), transcription factor EB (TFEB), p70S6 kinase (S6K), protein kinase C and Unc-51-like autophagy-activating kinases [14, 15]. Dysregulations of mTORC1 signaling are associated with many disease, such as the neuronal disorders, diabetes, epilepsy and cancer [16]. Immunosuppressant rapamycin can inhibit the mTORC1 activity for tumor treatment, but is invalid for the mTORC1-dependent phosphorylation of 4EBP1 in some tumor cells [17]. The mTORC1 signaling can be activated by the EGFR/PI3K/Akt and IR/PI3K/Akt pathways [18] and phosphorylates S6K and 4EBP1 to initiate mRNA translation for cell growth and metabolism. Under physiological conditions, the mTORC1 signaling is tightly controlled, the loss of negative regulation caused the unrestrained cell growth in cancers [19], the recent clinical trials on tyrosine kinase inhibitors has revealed that the mTORC1 is a potential molecular target in osteosarcoma despite the single-agent of mTOR inhibitors is failure in osteosarcoma trials [20, 21]. Combinations of mTOR and other pathway inhibitors, such as the sorafenib with mTOR inhibition [22] suggested that molecular targets based on specific biomarkers are an advisable for the novel therapies developing.

Bioinformatics is a new interdisciplinary subject with huge development potential and can help to reveal the laws and mysteries of complex biological process through the comprehensive application of computer science, molecular biology and information technology. In this study, we developed a useful and reliable prognosis model for the risk stratification and personalized treatment decision of OS patients based on the mTORC1 signaling features. By WGCNA screening, we analyzed the gene module that associated with the mTORC1 score, these genes are enriched in the Glycolysis, Central carbon metabolism in cancer and response to hypoxia pathway. Patients were grouped into high and low risk groups using the RiskScore system, with high-risk patients having significantly poor prognosis. In addition, the higher RiskScore is prone to indicate the cancer metastasis and the lower immune infiltration of the Natural killer cell, Macrophage, activated CD8 T cell, Effector memory CD8 T cell, activated B cell, and Type 1 T helper cell. Finally, we developed a nomogram model combination of the RiskScore and metastatic feature for the precise prediction of patients’ probability of surviving. Our work provides a useful tool to optimize the treatment strategies.

2 Materials and methods

2.1 Data acquisition

The clinical information and RNA-seq of human osteosarcoma (OS) specimens were collected from the TARGET (https://ocg.cancer.gov/programs/target) database and it include 84 OS samples (TARGET-OS) with the encoding protein genes. Another dataset GSE39058 was downloaded from the Gene Expression Omnibus database (GEO, https:/ww.ncbinlm.nih.gov/geo/) and a total of filtering 42 OS samples were included in this study [23]. We obtained 200 mTORC1 signaling related genes from the Molecular Signatures Database (MSigDB, https://www.gsea-msigdb.org/gsea/msigdb).

2.2 The ssGSEA for MTORC1 score

The single sample gene set enrichment analysis (ssGSEA) is able to analyze each single tumor sample to be able to generate a score of the activity of a specific signaling pathway in that sample, which in turn helps to identify the specific activation of the signaling pathway in different samples [24]. The HALLMARK_MTORC1_SIGNALING.v2023.2 and h.all.v2023.2.Hs.symbols pathway score of patients were calculated by performing ssGSEA using the GSVA R package in the TARGET-OS cohort [25].

2.3 Weighted Gene Co-expression Network Analysis (WGCNA)

Gene module related to the mTORC1 singling signature was sectioned using WGCNA [26]. The pickSoftThreshold function was used to determine the soft threshold (β) ensuring the scale-free network, after that the hierarchical clustering was performed to screen the gene module that include at 60 genes. The correlation analysis between gene module and mTORC1 signature (score) were performed for the most correlated gene module.

2.4 Enrichment analysis

The function enrichment analysis of module genes was analyzed using the Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) with clusterProfiler R package [25].

2.5 Construction of RiskScore model and validation of prognostic value

To shrink the number of candidate gene, Least Absolute Shrinkage and Selection Operator (Lasso) Cox regression analysis was performed with the glmnet R package [25]. Subsequently, the multivariate Cox regression analysis was performed to determine the key risk and the regression coefficient for RiskScore construction, according to the formula: RiskScore =\(\:{\Sigma\:}{\upbeta\:}\text{i}\times\:\text{E}\text{x}\text{p}\text{i}\). (βi is the regression coefficient, Expi is the expression level of risk gene). The RiskScore of patients was calculated by using the RiskScore system and divided the patients into high- and low-risk groups based on the median value, then the Kaplan-Meier (KM) survival analysis was conducted by using the survival R package, the ROC was plotted by the timeROC R package for the classifier efficiency evaluation [27].

2.6 Tumor microenvironment (TME) analysis among various risk groups

The ssGSEA method of GSVA R package was used to analyze the immune infiltration score of 28 immune cells [28]. Subsequently, we used the ESTIMATE algorithm for use in assessing the composition of immune and stromal cells in the tumor microenvironment of OS patients. Specifically, by analyzing the gene expression data of the samples, immune scores, stromal scores and ESTIMATE scores were generated, which in turn helped to quantify the characteristics of the tumor microenvironment [29]. Meanwhile, the MCP-Count algorithm was used to analyze immune infiltration score of 10 immune cells. In addition, the progeny R package was used to calculate the tumorigenesis-related hallmark pathway score and the correlation between the RiskScore and these pathways was analyzed [30].

2.7 Identifying the independent prognostic factor and developing a nomogram

The decision-making tree model usually used for a nonparametric supervised learning algorithm for classification task and was constructed by using the rpart R package [31]. To select significant and independent prognostic factors, we performed univariate and multivariate Cox regression analysis. The rms R package was used to construct a developing nomogram model that combined various clinical factors, then the predictive accuracy of nomogram model was assessed by the calibration curve [32], and the decision curve was performed for the practical benefit evaluation in clinical practice.

2.8 Statistical tests

Statistical analysis was performed in the R software (version3.6.0). Difference between two sets of continuous variables was analyzed using the wilcoxon rank-sum test. The spearman method and log-rank test were used for correlation analysis and survival difference comparison among various risk patients. A p < 0.05 was statistically significant.

3 Results

3.1 Skyblue2 is the most correlated module with the mTORC1 signature

The mTORC1 score of each patient was computed based on the 200 mTORC1 related genes expression, and the WGCNA identified the optimal soft thresholdβ is 5 (Fig. 1A) and obtained 57 co-expression modules after hierarchical clustering and module merging, in which the grey is an invalid module that is not aggregated to other modules (Fig. 1B). The correlation analysis showed that the skyblue2 is the most correlated module with the mTORC1 score (p < 0.05, Fig. 1C) and contained 67 genes (Fig. 1D). Function enrichment analysis revealed that these 67 module genes were closely related to the Glycolysis/Gluconeogenesis, Fructose and mannose metabolism, Galactose metabolism pathway, Central carbon metabolism in cancer, HIF-1 signaling pathway, and Carbon metabolism in KEGG (Fig. 1E) and closely associated with the metabolism and hypoxia biological process (BP) including the the response to decreased oxygen levels process, glycolytic process through fructose-6-phosphate, canonical glycolysis, response to hypoxia, pyruvate metabolic process (Fig. 1F). These findings suggested that the mTORC1 signaling pathway involved in the cancer progression through altering the metabolism- and hypoxia-related pathways.

Fig. 1
figure 1

WGCNA for the mTORC1 signature related gene module. A Scale-free fitting index analysis of various soft threshold power (β). B Gene tree map based on dissimilarity measure (1-TOM) clustering. C The correlation heatmap between gene module and mTORC1 score. D The number of genes in each module. E KEGG enrichment analysis of skyblue2 module genes. F Biological process of GO enrichment analysis of skyblue2 module genes

Full size image

3.2 The established RiskScore is a reliable prognostic classification model

The univariate Cox regression analysis was used to determine the significant prognostic genes (p < 0.05) in 67 module genes, the lasso Cox regression analysis was used to reduce the numbers of candidate gene (Fig. 2A) and multivariate Cox regression analysis was used for RiskScore construction (Fig. 2B), Riskscore=\(\:\left(0.157\text{*}\text{A}\text{N}\text{K}\text{R}\text{D}37\right)+\left(0.226\text{*}\text{B}\text{N}\text{I}\text{P}3\right)+\left(0.556\text{*}\text{P}\text{D}\text{E}4\text{C}\right)+\left(0.162\text{*}\text{P}\text{D}\text{K}1\right)+\left(0.343\text{*}\text{S}\text{E}\text{R}\text{P}\text{I}\text{N}\text{E}2\right).\) The patients were classified into low- and high-risk groups, with the high-risk patients having noticeably poor prognosis (p < 0.05, Fig. 2C), the ROC analysis revealed that the AUC values of 1-, 3- and 5-years survival rate is high, which are 0.75, 0.77 and 0.76 respectively (Fig. 2D), suggesting the RiskScore is a good classifier in long- and short-term prognosis. In addition, the Principal Component Analysis (PCA) also exhibited that the high- and low-risk patients had clearly boundary (Fig. 2E), we visualized the expression of the model genes and found that these genes as risk factors were significantly overexpressed in the high-risk groups (p < 0.05, Fig. 2F). To verify the model robustness, we analyzed the prognosis and classifier efficiency of RiskScore in the validation set (GSE39058), which is important in the analysis biomarker [33]. The results showed that the patients in the high-risk groups had significantly poor prognosis (p < 0.05, Fig. 2G) and the AUC of 1-, 3- and 5- years survival rate is higher (more than 0.7), which are 0.74, 0.8 and 0.77 respectively (Fig. 2H), revealing the highly reliable of RiskScore model.

Fig. 2
figure 2

Establishing and validating prognostic risk models A The trajectory of each independent variable changing with lambda and the confidence interval under lambda. B Distribution of coefficients of the prognostic gene signatures. C KM survival analysis of patients in TARGET-OS cohort. D ROC analysis of patients in TARGET-OS cohort. E Principal Component Analysis of patients in TARGET-OS cohort. F The expression difference of model genes in various risk groups. G KM survival analysis of patients in GSE39058 cohort. H ROC analysis of patients in GSE39058 cohort

Full size image

3.3 The higher RiskScore may be a cancer metastasis-related risk factor

We compared the distributional difference of RiskScore in various clinicopathologic features and found that the patients with ≤ 15 years old had higher ratio (Fig. 3A), the male low-risk patients had higher ratio (Fig. 3B) and most patients in low-risk are no metastatic (Fig. 3C). After that, further analysis revealed that the RiskScore is not affected the age (Fig. 3D) and gender distribution (Fig. 3E) of patients, but affected the metastatic distribution of patients, in which the patients with higher RiskScore are significantly prone to cancer metastasis (Fig. 3F). These results further suggest that the RiskScore may be an important indicator of poor prognosis in patients with OS.

Fig. 3
figure 3

Riskscore distribution characteristics among various clinicopathologic features A The age distribution of patients in high and low risk group. B The gender distribution of patients in high and low risk group. C The metastatic distribution of patients in high and low risk group. D The RiskScore difference among various age groups. E The RiskScore difference among various gender groups. F The RiskScore difference among various metastatic groups

Full size image

3.4 High-risk patients are related to a suppressive TME

Differences of the tumor microenvironment (TME) among high and low risk groups were evaluated, the ESTIMATE results demonstrated poor immune infiltration including the overall significantly lower ESTIMATE, stroma, immune scores in high-risk group (p < 0.05, Fig. 4A). The immune cells, such as the CD8 T cells, T cells, Monocytic lineage and endothelial cells are significantly high infiltration levels in the low-risk group (p < 0.05, Fig. 4B). In addition, the activated B and CD8 T cells, effector memory CD8 T cells, immature B cells, memory B cells, Type 1 T helper cells, macrophage, central memory CD8 T cell, MDSC, natural killer cells of others 28 immune cells also had significantly higher infiltration in the low-risk group (p < 0.05, Fig. 4C), suggested that the high-risk patients are associated with the suppressive TME.

Fig. 4
figure 4

Immune infiltration difference among high- and low-risk groups A The ESTIMATE of immune infiltration among high- and low-risk groups. B MCP-counter of 10 immune cell infiltration in the high- and low-risk groups. C ssGSEA of 28 immune cell infiltration in the high- and low-risk groups. (* p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001)

Full size image

3.5 Hypoxia and mTORC1 signaling are crucial for promoting OS progression

We calculated the tumorigenesis-related hallmark pathway score and their correlation with the RiskScore, and found that the Hypoxia pathway is significant positively correlated with the RiskScore, and the JAK-STAT pathway is significant negatively correlated with the RiskScore (Fig. 5A). The Wilcoxon rank sum test identified that the pathways of glycolysis, hypoxia and mTORC1 signaling had significant higher activity in the high-risk group (p < 0.05, Fig. 5B), indicating these pathways including the mTORC1 signaling are crucial for promoting OS progression.

Fig. 5
figure 5

The pathway characteristics analysis on RiskScore-related groups A The correlation analysis between the Riskscore and cancer-related hallmark signaling pathway. B The difference of HALLMARK pathway activity between high and low risk groups

Full size image

3.6 RiskScore is key independent prognostic factor and supported a reliable nomogram model

Finally, we constructed a decision-making tree model based on the age, gender, metastatic and Risktype features in the TARGET-OS cohort, the model showed that the classification ability of RiskType is more effective than other clinical features, followed by the metastatic feature, and four risk subgroups were obtained (Fig. 6A), in which the C1 subgroup is the significant good prognosis (p < 0.05, Fig. 6B), the C1 and C2 group were categorized into the high-risk group (Fig. 6C). Subsequently, the univariate Cox regression analysis showed that the metastatic and RiskScore features are the significant prognostic factors (p < 0.05, Fig. 6D) and the multivariate Cox regression analysis showed that they are independent prognostic factors (p < 0.05, Fig. 6E). To quantify the survival probability of patients, we combined the metastatic and RiskScore features to develop a nomogram model (Fig. 6F), which indicated that the RiskScore has the greatest influence on survival prediction. The calibration curve showed that the prediction curve of 1-, 3- and 5-years calibration points is closely coincided with the standard curve (Fig. 6G), suggesting that the nomogram has excellent prediction accuracy. The decision curve revealed that the net benefit of RiskScore and nomogram model is obvious higher than the extreme curve (Fig. 6H), indicating that our model had excellent clinical practicability. ROC analysis also demonstrated that the nomogram model is an excellent short- and long-term prognostic classifier.

Fig. 6
figure 6

Analysis of independent prognostic factors and a nomogram developing A Decision-making tree including Risktype, Metastatic, and age for risk stratification. B KM survival analysis among four risk subgroups. C The risk distribution among four risk subgroups. D Univariate Cox regression analysis for significant prognostic factors. E Multivariate Cox regression analysis for independent prognostic factors. F A developing nomogram model was constructed. G The 1-, 3- and 5-year calibration curves of the nomogram. H The decision curve of nomogram model. I ROC analysis of nomogram model

Full size image

4 Discussion

Osteosarcoma (OS) is a popular primary malignant bone tumor with early metastasis, high-grade aggressive and poor prognosis features in adolescents [34], meanwhile its complex heterogeneity generated a various of patients with different subtypes that exhibited unique immune response pattern, tumor microenvironment, genotypes and phenotypes at the macroscopic and microscopic aspects [35]. At present, the surgery with adjuvant chemotherapy is the primary means for osteosarcoma, but not benefit to all patients that highlighted the fact of many patients do not respond to the standard therapies [36]. In addition, several chemo-resistant cell models have been well constructed and many differentially expressed markers and noncoding RNAs were identified [37, 38]. Hashimoto et al. NY-ESO-1 and MAGE-A4 expression may be associated with the immune status in the tumor microenvironment and highlighted that identification of validated biomarkers in OS has a key role in patient prognostic assessment and treatment stratification [39]. This study analyzed the actual variation among osteosarcoma patients with differing intrinsic mechanisms and outcomes to develop novel therapeutic targets and strategy for those with a dismal prognosis. This study constructed a useful and reliable prognostic model to screen the high-risk patients and offer the tailoring supplementary treatments for improving prognosis based on the mTORC1-related signature.

Bone sarcomas, especially the osteosarcoma develop in a highly dynamic bone microenvironment consisting of bone cells (osteocytes, osteoblasts, and osteoclasts), stromal cells (mesenchymal stem/stromal cells (MSCs), fibroblasts), vascular cells (pericytes and endothelial cells), a mineralized extracellular matrix (ECM), and immune cells (lymphocytes, macrophages) [40]. A fine-tuned orchestrated activity of stromal, vascular and bone cells maintain the bone homeostasis through the cellular communications (autocrine and paracrine). However, the chromosomal aneuploidy and gene mutation initiate tumorigenesis, the tumor cells will highjack bone physiological pathways for their survive and grow [41]. Osteosarcoma also is highly vascularized bone tumors and was surrounded by an acidic and hypoxic bone microenvironment [42], the RiskScore and the genes in the skyblue2 module are closely associated with the Hypoxia signaling and the hypoxia-induced factor-1 (HIF-1) pathway, which may be further promote the angiogenesis [43], thus the drugs of anti-angiogenesis may benefit to the high-risk OS patients, such as the sorafenib alone [44], or combined with everolimus (an mTOR2 inhibitor) [22]. In addition, the tumor-associated macrophages (TAMs) is another major component of OS microenvironment, with a number of myeloid, dendritic and lymphoid cells [45]. Increased infiltration of TAMs is related to less active metastasis and longer survival as compared to epithelial tumors [46, 47]. The mechanism of TAMs inhibiting metastasis in OS is unclear, different teams in precise heterogeneous population case found and suggested that M2 macrophages exert an anti-metastatic effect rather anti-inflammatory [47], the macrophage-activating agent mifamurtide significantly improved the 6-year overall survival in a clinical trial [48]. MSCs and MSC-derived osteoblasts are sensors of their microenvironment, they not only express multiple signaling receptors, but also secrete ECM components, such as the cytokines and metalloproteinases (MMPs) to modulate tumor microenvironment [49]. The tumor infiltration of immune cells is strictly dependent on the proteolysis and plasticity of ECM [50], Nicolas-Boluda et al. demonstrated that T cells can cross the blood vessel but was trapped by the tumor nodules [51]. T cells do not release the MMPs to lyse the matrix and cannot progress through the dense and tight fibers [52], this may be explained that the T cells are significantly decrease in the high-risk patients, thus induced proteolysis may be a key event to achieving immune-cell therapy success in osteosarcoma.

These model genes were defined as risk factors in the multivariate Cox regression analysis. Ankyrin repeat domain protein 37 (ANKRD37) is a hypoxia-associated protein and was up-regulated in colon cancer indicating poorer survival rate [53]. BNIP3 is another hypoxia-associated mitophagy protein, Vara-Pérez el. reported that the elevated BNIP3 levels as pro-tumorigenic regulator in melanoma is associated with poorer patient’s survival, the depletion of BNIP3 can compromise tumor growth in vivo [54]. However, the deletion or silencing of BNIP3 in the chronic kidney disease can significantly increase the mitochondrial damage, the production of mitochondrial ROS and activation of the NLRP3 inflammasome, demonstrating that the BNIP3- mediated mitophagy played a crucial protective role against hypoxia-induced cell injury [55], these evidences indicated that the protective role of BNIP3 is specific to some disease. Phosphodiesterase type 4 C (PDE4C) is a key messenger that specifically hydrolyze cAMP in cell signaling systems, Wright el. reviewed the deleterious or protective role in different cancer, where PDE activity can facilitate cancer via avoiding cAMP-triggered cell cycle arrest or the regulation of cAMP prevent cancer initiation [56], and it can be a potential cancer treatment target. The PDK1 was thought as master kinase with autophosphorylation to involve in the cell growth, invasion, metastasis and apoptosis regulation through activating the PI3K-AKT signaling pathways, the overexpression of PDK1 was observed in a plethora of cancers, such as, while the silenced PDK1 can inhibit tumor growth and proliferation through modulating tumor microenvironments and affecting tumor immunotherapies [57]. SERPINE2 can promote the radio-resistance of lung cancer cells and knockdown of SERPINE2 improve the tumor radio-sensitivity in vitro and in vivo [58], Chen el. reported that SERPINE2 could be acted as a target for metastasis in advanced renal cell carcinoma [59]. These results suggest that the key genes we identified may be aberrantly expressed in OS, which in turn may promote aggressiveness and drug resistance in OS and worsen patient prognosis by affecting hypoxia response, metabolic regulation and signaling pathways. Meanwhile, the regulatory effects of these genes on the tumor microenvironment may further support the progression of OS, suggesting their importance as potential therapeutic targets in OS research and treatment.

There also had some limitations in our study. First, the small sample size of this study may not be sufficient to represent the diversity of the entire OS patient population; in the future, we will add data from multiple centers and different populations to improve the reliability and generalizability of the results. In addition, this study lacks in-depth experimental validation to explore the specific mechanism of action of these genes in OS. Therefore, we will further combine functional experiments, including gene knockout and animal models, to be able to provide a potential molecular basis for OS-targeted therapy.

5 Conclusion

The inappropriate diagnosis and sub-optimal therapy can be irrevocably reduce the survival probability, the prospective clinical trials and prognostic tool will perfect the standard practice of OS treatment. This study developed a mTORC1 signaling-related risk prognostic model that acted as an independent prognostic factor for osteosarcoma. Our study offers a useful tool to predict the prognosis and indicate the immune infiltration, and may be provide potential treatment targets for immunotherapy.