Differentially expressed of iron metabolism-related genes
Heatmap was generated to visualize the expression level of m6A methylation-related genes between tumor tissue and normal tissue (Figure 1). The expression levels of SLC11A2, FLVCR1, FTH1, TFRC, SLC25A38, and FTL were significantly overexpressed in tumor samples than normal tissue. The expression levels of HMOX1 were significantly low expressed in tumor samples.
Consensus clustering analysis of iron metabolism-related genes
The consensus clustering analysis showed that the most appropriated selection was divided HCC patient into two clusters (Figure 2A-D). The Kaplan–Meier curve showed that 5 year-OS of cluster 1 was significant longer than cluster 2 (P = 0.006) (Figure 3A). Furthermore, the relationship between the clustering and clinicopathological features were evaluated. Tumor T status, and stage were significant difference between the cluster 1 and cluster 2 (Figure 3B).
Prognostic model base on iron metabolism-related genes
Total of 4 iron metabolic-related genes (FLVCR1, FTL, HIF1A, HMOX1) were identified to construct the prognostic model. Our prognostic model that based on the 4 iron metabolic-related genes was formed using the following formula: risk score = (the means expression of FLVCR1 * 0.16713) + (the means expression of FTL * 2.52E-05) + (the means expression of HIF1A * 0.014361) + (the means expression of HMOX1 * 0.002315).
According the median level of the risk score, the HCC patients divided into low-risk and high-risk groups. Kaplan–Meier survival curves showed the low-risk group had a better survival rate than high-risk group (HR = 1.329, 95% CI = 1.210– 1.461, P < 0.001) (Figure 4A). In addition, we evaluated the accuracy of the OS-related prognostic model by constructed the ROC curve, and the AUC of risk score was significant than other clinicopathological parameters (Figure 4B). Finally, we ranked HCC patients by prognostic model to analyses the survival distribution. We could determine the mortality of HCC patients base on their risk scores. Moreover, with the increase of risk score, the patient’s mortality rose (Figure 4C, D).
The relationships between prognostic model, clinicopathological parameters and tumor-infiltrating immune cells
The univariate and multivariate Cox regression analysis showed that the prognostic model can be as an independent prognostic factor for HCC patients (Figure 5A, B). Moreover, the prognostic model was significantly related to clinicopathological features such as survival state (p=0.001), stage(p=0.002), grade(p=0.001), and tumor T status(p=0.002) (Figure 6A-D).
We next evaluated the association of prognostic model and tumor-infiltrating immune cells in HCC microenvironment using CIBERSORT algorithm. The CIBERSORT analysis showed that the composition of 22 immune cell types in High-risk group and low-risk group of HCC varied significantly (Figure 7A, B). Macrophages M0 were more enriched in High-risk group, while T cells CD8, NK cells activated were more concentrated in low-risk group (Figure 7C).
The expression of the iron metabolism-related genes in Oncomine database, The Human Protein Atlas and Kaplan Meier plotter
We analyze the expression level of FLVCR1, FTL, HIF1A, HMOX1 in liver cancer by Oncomine database. The expression level of FLVCR1, FTL, HIF1A in different hepatocellular carcinoma was higher than the normal group in the Wurmbach Liver (222906_at), Chen Liver (IMAGE: 1575419), Roessler Liver (200989_at) (Figure 8A-C). However, the expression level HMOX1 was no value in Oncomine. In addition, we verify the histological level of FLVCR1, FTL, HIF1A, HMOX1 by the Human Protein Atlas database, and the results showed that FLVCR1, FTL, HIF1A, HMOX1 is upregulated in HCC tissues and downregulated in normal tissue (Figure 8D-G). The prognostic significance of FLVCR1, FTL, HIF1A, HMOX1 were identified by Kaplan Meier-plotter server. The results showed that the FLVCR1, FTL, HIF1A, HMOX1 were closely related to the OS of HCC patients (Figure 9 A-D).
Validation of the prognostic model
We calculated the risk score of each HCC patient in the ICGC data portal project Liver Cancer - RIKEN, JP (LIRI-JP) as an independent external validation by the same formula. The HCC patients divided into high- and low-risk groups based on the median level of risk score. The Kaplan–Meier survival curves showed the prognostic value of our prognostic model (P < 0.001) (Figure 10A). In addition, the ROC curve also showed a good ability of the OS-related prognostic model to predict prognosis of HCC patients (Figure 10B). And with the increase of risk score, the mortality rate of patients rose (Figure 10C, D). Therefore, these validation results confirmed the predict prognosis ability of our prognostic model.