Figure 1 displayed the flow chart of present study. The included information in the present study consisted of 84 osteosarcoma patients’ information in TARGET database and 53 osteosarcoma patients’ information in GSE21257. Detailed Clinical characteristics of osteosarcoma patients were summarized in Table 1.
Table 1
clinical charactics of TARGET and GEO cohorts.
|
TARGET cohort
|
GEO cohort
|
Total number
|
84
|
53
|
Age
|
|
|
Age < = 18
|
71
|
40
|
Age > 18
|
13
|
13
|
Gender
|
|
|
Male
|
47
|
34
|
Female
|
37
|
19
|
Metastasis
|
|
|
Yes
|
21
|
34
|
No
|
63
|
19
|
Status
|
|
|
Alive
|
57
|
30
|
Dead
|
27
|
23
|
Identification of Prognostic IRG in TARGET database
17 IRG were correlated with OS according to univariate Cox analysis result (Fig. 2A). Figure 2B displayed the correlation between these IRG (Fig. 2B).
Establishment of prognosis signature in TARGET database
LASSO-Cox regression analysis was utilized to calculate the expression profiles of prognostic IRG (Fig. 2C, 2D). According to LASSO-Cox regression analysis result, we constructed the prognostic IRG signature. The risk score formula of each patient was listed as follows: risk score = 3.259*expression level of IL12B − 0.252*expression level of CCL2–0.001*expression level of CD82–0.292*expression level of CMKLR1–0.971*expression level of FPR1–0.618*expression level of MMP14 + 2.084*expression level of MYC + 0.045*expression level of PDPN + 0.380*expression level of RGS16–0.114*expression level of SELL + 1.089*expression level of SLC7A1.
Median risk score value was used to divide all the osteosarcoma patients into high and low risk subgroups. Patients in low-risk subgroup had a significantly better OS than patients in high-risk according to Kaplan-Meier curve result (Fig. 3A). Figure 3B and 3C displayed IRG heatmap and distribution in the signature. The area under the curve (AUC) of prognosis signature were 0.870 at 1 year, 0.818 at 2 years, and 0.816 at 3 years (Fig. 3D). PCA (Fig. 3E) and t-SNE (Fig. 3F) analysis results demonstrated that patients in high and low risk groups were distributed in different directions.
Validation of the prognosis Signature in the GEO Cohort
To assess the stability of prognosis signature established from TARGET database, osteosarcoma patients in GSE21257 were also categorized into high and low subgroups. Patients in the low risk subgroup had a longer survival time compared with the high risk subgroup (Fig. 3G). Figure 3H and 3I displayed IRG heatmap and distribution in the signature in GEO cohort. And AUC of the prognosis signature in GSE21257 were 0.684 at 1 year, 0.708 at 2 years, and 0.763 at 3 years (Fig. 3J). PCA (Fig. 3K) and t-SNE (Fig. 3L) analysis results indicated a discrete distribution of osteosarcoma in different risk groups which was similar to the results in TARGET database.
Independent prognostic value of prognosis signature
To investigate whether risk score in IRG signature was an independent prognostic factor for osteosarcoma, univariate and multivariate Cox analysis of clinical characteristics in the TARGET cohort were performed. Risk score of IRG signature (HR = 6.067, 95% CI = 3.135–11.740, P < 0.001) and metastasis status (HR = 4.764, 95% CI = 2.220–10.220, P < 0.001) were significantly correlated with OS of osteosarcoma according to univariate Cox analysis result (Fig. 4A). And age (HR = 0.713, 95% CI = 0.335–1.520, P = 0.382) and gender (HR = 0.999, 95% CI = 0.919–1.086, P = 0.984) were no correlated with OS of osteosarcoma. In multivariate Cox analysis, risk score of IRG signature (HR = 6.154, 95% CI = 3.013–12.567, P < 0.001) and metastasis status (HR = 4.045, 95% CI = 1.856–8.813, P < 0.001) were still independent prognostic factors for OS after correction of other confounding factors (Fig. 4B). Heatmap of clinical characteristics and IRG signature was display in Fig. 4C. Next, GSVA was performed to elucidate the potential biological mechanisms between different risk subgroups. GSVA heatmaps (Fig. 4D) showed that cytokine receptor interaction, pathways in cancer, toll like receptor signaling pathway was dynamically correlated to risk stratification. These signaling pathways might be participate in the different risk subgroups. Percent weigh of different clinical characteristics in different risk score were display in Fig. 5. Risk score in metastasis status (P < 0.01) was significantly higher than that in non-metastasis status. Risk score in dead status (P < 0.01) was significantly higher than that in alive status. And no apparent significances of risk score in different age subgroups and gender subgroups were found. In the stratification survival analysis, no apparent significances of different risk score in age > 18 subgroup. And other clinical subgroups, patients in the low risk subgroup had a longer survival time compared with the high risk subgroup (P < 0.05).
Immune Status and Tumor Microenvironment Analysis
ssGSEA was utilized to quantify the enrichment scores of immune cell subpopulations, related immune functions. In TARGET cohort, the fractions of B cells, CD8 + T cells, DCs, Macrophages, Neutrophils, NK cells, pDCs, T helper cells, Tfh, Th2 cells, TIL and Treg in high risk subgroup were lower than that in low risk subgroup (Fig. 6A). While in GEO cohort, the fractions of aDCs, B cells, CD8 + T cells, DCs, Macrophages, Neutrophils, pDCs, Tfh, Th1 cells, Th2 cells, TIL, Treg in high risk subgroup were lower than that in low risk subgroup (Fig. 6B). The results of fractions of B cells, CD8 + T cells, DCs, Macrophages, Neutrophils, pDCs, Tfh, Th2 cells, TIL, Treg were similar in both TARGET and GEO cohorts between different risk subgroups (adjusted P < 0.05, Figs. 6B, D). In both TARGET and GEO cohorts, function score of T cell co-inhibition, T cell co-stimulation, Type I IFN response, APC co-inhibition, APC co-stimulation, CCR, check point, cytolytic activity, HLA, inflammation promoting, parainflammation and MHC class I in high risk subgroup were significantly lower than that in low risk subgroup (Fig. 6D, 6E). To investigate the association between immune score, estimate score and risk score, the estimate formula was utilized to analyze immune and estimate score of osteosarcoma patients in TARGET cohort. And our results indicated negative correlations of risk scores with immunescore (R=-0.68, P < 0.001, Fig. 6C) and estimate score (R=-0.54, P < 0.001, Fig. 6F).