3.1 Cellular heterogeneity of LIHC tissues

The flow chart of this study was shown in Fig. 1. Quality control was carried out on the GSE149614 dataset. In total, 57,836 cells were obtained after filtering cells with > 20% mitochondrial gene content, features < 500, or features > 6000 and visualized by tSNE downscaling. The 57,836 cells were successfully classified into 21 independent clusters (Fig. 2A). SingleR was utilized to identify the cell clusters in a total of 8 cell types (Fig. 2B). Among them, clusters 0, 12, and 20 were annotated as T cells (21,443, 37.07%); clusters 8 and 17 were annotated as B cells (1675, 2.89%); cluster 11 was annotated as dendritic cells (564, 0.97%); clusters 4 and 18 were annotated as endothelial cells (3788, 6.54%); clusters 1, 5, 6, 9, 10, 14, 15, 16, 19 were annotated as hepatocytes (15,979, 27.62%); cluster 2 and13 were annotated as macrophages (8506, 14.70%). Cluster 3 was annotated as monocytes (4021, 6.95%); cluster 7 was annotated as smooth muscle cells (1860, 3.21%). The proportion of each cell between each sample is shown in Fig. 2C. The marker genes of eight cell types were used (T cells: CD3D; B cells: CD79A; Endothelial cells: PECAM1; Monocytes: IL1B; Macrophages: CD68; Smooth muscle cells: ACTA2; Dendritic cells: FLT3; Hepatocytes: ALB) were plotted in a bubble map (Fig. 2D). Each marker gene had a high expression and cellular expression ratio in cell subpopulations, indicating the good auto-annotation effect of SingleR. tSNE plots showed CD3D (Fig. 2E), CD79A (Fig. 2F), CD68 (Fig. 2G), CD14 (Fig. 2H), PECAM1 (Fig. 2I), ACTA2 (Fig. 2J), FLT3 (Fig. 2K), and ALB (Fig. 2L) expression at the cellular level.

Fig. 1

The study flow chart. scRNA Single-cell RNA-sequencing, PCA Principal Component Analysis, tSNE t-Distributed Stochastic Neighbor Embedding, FRGs FRGs, TCGA The Cancer Genome Atlas, LIHC Liver Hepatocellular Carcinoma Collection, DEGs Differentially Expressed Genes, WGCNA Weighted Gene Co-Expression Network Analysis, LASSO Absolute Shrinkage and Selection Operator

Fig. 2

Identification of tissue cell subpopulations in LIHC patients on the basis of single-cell RNA-seq data set GSE149614. A A total of 57,836 cells were clustered into 19 cell clusters by tSNE. B Cells were annotated by singleR into 8 cell types: B cells, T cells, endothelial cells, monocytes, macrophages, smooth muscle cells, dendritic cells, and hepatocytes. C Percentage stack graph showing the proportion of 8 cell types in each sample. D Expression levels of marker genes for the 8 cells are shown as bubble plots, with darker colors indicating higher expression levels and larger circles indicating a higher percentage of gene expression within the cell population. E–L tSNE plots showing the expression levels of CD3D (E), CD79A (F), CD68 (G), CD14 (H), PECAM1 (I), ACTA1 (J), FLT3 (K), and ALB (L) in the single-cell data set

3.2 Differential expression and scoring of FRGs in immune cells

The DEGs among immune cells from GSE149614 data were intersected with 20 FRGs, yielding 7 DEFRGs (ZFP36, NCOA4, FTH1, FTL, TNF, PCBP1, CYB561A3). The heat map was utilized to show the expression of 20 FRGs in immune cells (T cells, B cells, monocytes, macrophages, and dendritic cells) (Fig. 3A). ZFP36 was highly expressed in T cells, FTH1, FTL, PCBP1, NCOA4 in monocytes, FTL, FTH1, NCOA4, and TNF overexpressed in macrophages, and CYB561A3 was highly expressed in B cells. In addition, correlation analysis was performed for 20 FRGs. A considerably positive link was observed between FTH1 and FTL, and a considerably negative link between FTL and ZFP36 (Fig. 3B). Subsequently, seven DEFRGs were analyzed, including ZFP36 (Fig. 3C), NCOA4 (Fig. 3D), FTH1 (Fig. 3E), FTL (Fig. 3F), TNF (Fig. 3G), PCBP1 (Fig. 3H), and CYB561A3 (Fig. 3I) in the single-cell dataset. In addition to the highly expressed cell types shown in the heat map, ZFP36, FTH1, FTL, and PCBP1 were also widely expressed in other cell types, with FTH1 and FTL both having high expression in hepatocytes.

Fig. 3

Expression levels and correlation analysis of FRGs among immune cells in the single-cell dataset GSE149614. A Heat map of the expression of 20 FRGs among immune cells. The top annotation bar indicates five immune cell types: T cells, macrophages, dendritic cells, B cells, and monocytes. Light-to-dark color gradient represents the progressively elevated expression level, with red color suggesting a positive relationship and blue color suggesting a negative relationship. B Heat map of the correlation of 20 FRGs, where red color denotes a positive relationship and blue color denotes a negative relationship. C–I tSNE plots showing the expression levels of ZFP36 (C), NCOA4 (D), FTH1 (E), FTL (F), TNF (G), PCBP1 (H), and CYB561A3 (I) in the single cell data set

3.3 Scoring and functional enrichment analysis of FRGs

The expression of DEFRGs was scored for each cell in the single-cell dataset GSE149614 using the AUCell package (Fig. 4A). Monocytes and macrophages had the highest gene scores for FRGs with mean scores of 0.52 and 0.55, respectively. Subsequently, DEGs between these two cell types were subjected to functional enrichment analysis.

Fig. 4

Single-cell data set GSE149614 FRG cell scoring and functional enrichment analysis of high-scoring cell populations. A FRG scores among cell subpopulations (AUC), with lighter colors indicating higher scores, where monocytes and macrophages had the highest mean scores. B Bubble plot of KEGG results for DEGs between monocytes and macrophages, with closer colors to red indicating smaller p and larger bubbles indicating more DEGs enriched within that pathway. C BP, CC, and MF enrichment results of GO analysis of DEGs between monocytes and macrophages presented as bubble plots, with colors closer to red indicating smaller p and larger bubbles indicating more DEGs enriched within that pathway. AUC area under the curve, KEGG Kyoto Encyclopedia of Genes and Genomes, DEGs differentially expressed genes, GO Gene Ontology, BP biological process, CC cellular component, MF molecular function

According to KEGG analysis results, DEGs of macrophages were primarily enriched in Coronavirus disease, ribosomes, phagosomes, rheumatoid arthritis, and lysosome pathways. DEGs of monocytes were primarily enriched in hematopoietic cell lineage, viral myocarditis, antigen processing and presentation, rheumatoid arthritis, and allograft rejection pathways (Fig. 4B). According to GO analysis results, DEGs of macrophages were primarily correlated with biological processes (BPs) such as cytoplasmic translation, leukocyte mediated immunity, and leukocyte cell–cell adhesion, cell components (CCs) such as cytosolic ribosome, cytosolic large ribosomal subunit, focal adhesion, and molecular functions (MFs) such as structural constitution of ribozyme, MHC protein complex binding, and MHC class II protein complex binding. DEGs of monocytes were primarily correlated with BPs, such as regulating the activation of T cells, cell–cell adhesion, and leukocyte mediated immunity, CCs such as endocytic vesicle membrane, endocytial vesicles, and tertiary granules, and MFs such as MHC protein complex binding, immune receptor activity, and MHC class II protein complex binding (Fig. 4C). Tables 2, 3 display the specific KEGG and GO enrichment results for Macrophage and Monocyte.

Table 2 Results of KEGG analysis based on differentially expressed genes of Macrophage and MonocyteTable 3 Results of GO analysis based on differentially expressed genes of Macrophage and Monocyte3.4 Cellular communication

Cellular communication among 8 cell types was inferred and quantified by CellChat. In addition, the number (Fig. 5A) and intensity (Fig. 5B) of cellular communication were visualized by heat map and circle plot. Macrophages interacts with endothelial cells, hepatocytes, and smooth muscle cells in high numbers. In addition, hepatocytes and endothelial cells have high interaction intensity with macrophages and monocytes, respectively. In addition, all important receptor–ligand pairs (Fig. 4C) were counted when immune cells send/receive signals. MIF signaling pathway-related ligand-receptor pairs play a crucial role in this process (p < 0.01).

Fig. 5

Single-cell dataset GSE149614 LIHC cell subpopulation communication analysis. A Heat map of the number of interactions between the 8 cell types, the darker shades of red indicated a higher number of interacting ligand-receptor pairs. B Network diagram of the intensity of intercellular interactions between 8 cell types, where nodes indicate various cell types, arrows indicate from the signal source cells to the receiving cells, and the line thickness indicates the intensity of the intercellular interaction. The thicker it is, the higher the intensity of the interaction. Different colors represent different cell types. C Ligand–receptor pairs of intercellular communication relationships between immune cell populations, where the horizontal coordinates indicate cell types where cell communication occurs, and the vertical coordinates indicate ligand-receptor pairs. The figure only displays communication relationship results that are statistically significant (P < 0.05). The color of the circle, from blue to red, denotes a gradual increase in the communication probability of cellular interactions

3.5 Differential expression and correlation analyses of FRGs in the TCGA-LIHC dataset

The varied expression levels of DEFRGs between the tumor and control groups in the TCGA-LIHC liver cancer dataset were compared, where ZFP36, NCOA4, FTH1, FTL, TNF, and PCBP1 were matched with the TCGA transcriptome data. The expression of all six FRGs was significantly different (p < 0.05), with FTH1, FTL, and PCBP1 being overexpressed in the tumor group and ZFP36, NCOA4, and TNF expressed less in the tumor group (Fig. 6A). The expression levels of FRGs in different samples were presented as a heat map (Fig. 6B). The correlation matrix of FRGs expression levels was shown in Fig. 6C, where PCBP1 had a considerable positive association with ZFP36, NCOA4, and FTH1, respectively, and ZFP36 and NCOA4 and FTH1 and FTL had a considerable positive association, respectively. The correlation results with correlation coefficients greater than 0.2 were presented as scatter plots, in which FTL was positively linked with FTH1 (R = 0.59, P < 0.001), PCBP1 was positively associated with ZFP36 (R = 0.42, P < 0.001), PCBP1 was positively associated with FTH1 (R = 0.28, P < 0.001), PCBP1 was positively associated with NCOA4 (R = 0.60, P < 0.001), and NCOA4 was positively associated with ZFP36 (R = 0.34, P < 0.001).

Fig. 6

Expression of FRGs in TCGA-LIHC dataset and correlation analysis. A Box plot of the comparative expression levels of iron-related autophagy genes between LIHC tumors and paracancerous tissues. Group differences were analyzed by the Wilcoxon test, FDR-corrected p-values were annotated on the graph; B Differential expression of FRGs between different samples, shown as a heat map; orange represents tumor group, blue represents paracancer control group. The color of gene expression levels from light to dark indicates elevated expression levels, with the negative association in blue and positive in red. C Correlation matrix of differentially expressed FRG expression levels. Red denotes a positive correlation, and blue denotes a negative correlation. Darker colors indicate enhanced correlations, and non-significant ones are shown by black X. D–H Correlation analysis of expression levels of ferritinophagy with significant results, where correlation coefficients above 0.2 are indicated by dotted line plots, and correlation coefficients R and P values are labeled on the plots, respectively

3.6 Assessment of the immune microenvironment and correlation analysis of FRGs

The infiltration of different immune cells in the TCGA-LIHC dataset was analyzed using CIBERSORTx. The infiltration of 22 immune cells among different subgroups is illustrated in Fig. 7A. M2 macrophages, neutrophils, monocytes, memory B cells, and gamma delta T cells were less infiltrated in the LIHC case set, while M0 macrophages, Tregs, resting dendritic cells, and activated mast cells were more infiltrated in the LIHC case group. Correlations between the degree of infiltration between different immune cells were analyzed, and the correlation matrix is shown in Fig. 7B. where cells with correlation coefficients greater than 0.3 or less than − 0.3 were selected. CD8 T cells and activated memory CD4 T cells were positively correlated (coefficient = 0.40, P < 0.001), and naïve B cells and plasma cells were negatively associated (coefficient = 0.33, P < 0.001), M2 macrophages and M0 macrophages were negatively associated (coefficient = − 0.45, P < 0.001), resting memory CD4 T cells and CD8 T cells were negatively associated (coefficient = − 0.44, P < 0.001), naïve B cells and monocytes were negatively associated (coefficient = − 0.33, P < 0.001), resting and activated NK cells were negatively associated (coefficient = − 0.38, P < 0.001). Finally, the association of DEFRGs with different immune cells was assessed separately (Fig. 7C). Most FRGs were observed to have a significant positive correlation with M0 macrophages and Tregs (P < 0.001).

Fig. 7

Immune cell infiltration analysis of TCGA-LIHC dataset. A Comparison of different levels of immune cell infiltration in the LIHC case/control group. Differences between groups were analyzed by the Wilcoxon test, and the statistical significance of differences is indicated by the "*" sign, where "*" indicates P < 0.05, "**" indicates P < 0.01, "***" indicates P < 0.001, "****" indicates P < 0.0001; B correlation matrix between immune cells, where red denotes positive association, blue denotes a negative association and darker color indicates increased association. Non-statistical significance is indicated by black X's; C Correlation matrix between immune cells and FRGs, where red denotes a positive association, blue indicates a negative correlation, and darker color indicates enhanced correlation, and correlation coefficients and p-values are marked in squares

3.7 WGCNA

In total, 3882 DEGs, including 1588 overexpressed genes and 2294 genes with low expression, were screened from the TCGA-LIHC gene expression matrix utilizing the limma package of R language. The screened DEGs were accepted by WGCNA (Fig. 8). Seven modules were calculated, which had some correlation with FRGs and immune cell infiltration, respectively. MEbrown had a considerably positive link with ZFP36, NCOA4, and PCBP1 (P < 0.001), MEgreen had a considerably positive link with TNF (P < 0.001), MEturquoise had a considerably positive link with PCBP1 (P < 0.001), and MEbrown and MEyellow had a considerably positive association with Tregs and macrophages, respectively. M2 had a considerable negative correlation (P < 0.001), and MEturquoise had a considerable positive correlation with M0 macrophages (P < 0.001).

Fig. 8

WGCNA of TCGA-LIHC dataset. A, B The power parameter screening process of WGCNA, where the values of clustering tree Connectivity A and model fitting R-square B with increasing power and slowing down the rate of change is the best power value, suggesting that 4 is the best power value; C TCGA-LIHC dataset C tree diagram of cluster analysis of samples and corresponding FRGs-related gene expression and immune cell infiltration; D gene cluster analysis of WGCNA illustrated as a tree diagram, where the modules of gene classification are demonstrated by distinct colors; E correlation matrix between the module scores retrieved by WGCNA and the obtained FRGs-related gene expression and the level of differential immune cell infiltration, with red indicating positive correlation, green indicates negative correlation. Darker colors indicate an enhanced correlation, and correlation coefficients and p-values are shown in the cells within the matrix. WGCNA Weighted Gene Co-Expression Network Analysis, FRGs ferritinophagy-related genes

3.8 Correlation between DEFRGs and patient prognosis

The association between DEFRGs in the TCGA-LIHC dataset and the patient prognosis was analyzed using univariate Cox regression. Patients with higher expression of FTH1 (HR = 1.47, 95% CI 1.19–1.81), FTL ((HR = 1.26, 95% CI 1.08–1.48), and PCBP1 (HR = 1.57, 95% CI 1.15–2.15) had a poorer prognosis (Fig. 9A). Individuals were categorized into high- and low-expression groups as per the gene expression using the maxstat package, and the results were in line with the outcomes of COX regression analysis with continuous variables. Individuals with high expression of FTH1, FTL, and PCBP1 had a considerably poorer prognosis than those with low expression of these genes (P < 0.001).

Fig. 9

Survival analysis of risk groups for the TCGA-LIHC dataset. A Forest plot of the outcomes of COX regression analysis for the six DEFRGs. B–G Survival curves (K–M method) for FTH1 (B), FTL (C), NCOA4 (D), PCBP1 (E), ZFP36 (F), and TNF (G) in high- and low-expression groups. Cutoff values were determined by the maxstat package, where orange reaches the high-risk group and purple color denotes the low-risk group. FRGs. FRGs, K–M Kaplan–Meier

3.9 Prognostic marker screening and risk score construction

LASSO regression analysis was utilized to screen three FRGs as prognostic markers, including FTH1, FTL, and PCBP1 (Fig. 10A, B).

Fig. 10

LASSO regression screening for prognosis-related FRGs. A, B Screening of prognostic markers using LASSO logistic regression models; partial likelihood deviation with tenfold cross-validation used to calculate the optimal λ; C ROC curves and AUC values for risk score prediction of 1-, 3- and 5-year survival of patients; D Survival curves for high and low expression groups according to risk scores (K–M method); E Other COX regression analysis of the impact of clinical features on patient prognosis, presented as a forest plot; F Multivariate COX regression analysis of significant clinical features in C, presented as a forest plot. LASSO, absolute shrinkage and selection operator, FRGs ferritinophagy-related genes, ROC receiver operating characteristic AUC area under the curve, K–M Kaplan–Meier

The coefficients of the candidate prognostic markers were found based on the results of the analysis of the LASSO regression model. Subsequently, the risk score RS was measured by means of the following equation:

\(\mathrm=0.1846*\mathrm1+ 0.0391*\mathrm+0.1618*\mathrm1\).

The ROC curves for risk score prediction of 1-, 3- and 5-year survival of diseased individuals are shown in Fig. 10C, with the best predictive power for 1-year survival (AUC = 0.687). The best cutoff for the predictive ability of the risk score for survival time in individuals with LIHC was determined using the maxstat package was 5.9135. Individuals with LIHC were categorized into high- and low-risk groups as per their cutoff values. In addition, individuals with no survival data were eliminated. Individuals with high risk scores had significantly shorter prognostic survival duration than those with low risk scores (Fig. 10D).

3.10 Prognostic model construction

The outcomes of univariate COX regression demonstrated that both age and tumor stage affected patient survival except for risk score/group (Fig. 10E). A multivariate prognostic prediction model was constructed using the Cox regression model (Fig. 10F). Nomogram (Fig. 11A) and calibration curves (Fig. 11B–D) were drawn with the rms package for predicting 1-, 3- and 5-year survival in individuals with LIHC. Tumor stage, age (c-index = 0.633), and risk score (c-index = 0.635) were used to construct the nomogram, which demonstrated their value as predictors. With a c-index of 0.678, the survival prediction model constructed by integrating age, stage, and risk scores demonstrated superior prognosis predictive performance.

Fig. 11

Prognostic risk model for LIHC. A Nomogram of the multivariate COX regression model for risk score prediction of survival in individuals with LIHC from the TCGA-LIHC dataset. B–D Calibration curves for 1-(B), 3-(C), and 5-(D) year survival prediction

3.11 Immunophenotypes of risk groups

A comparison of immune cell abundance between the risk subgroups in the TCGA-LIHC dataset is shown in Fig. 12A. M0 macrophages, plasma cells, and Tregs had higher abundance in the high-risk group (P < 0.05). Resting mast cells, neutrophils, and activated memory CD4 T cells had higher abundance in the low-risk group (P < 0.05).

Fig. 12

Immunophenotypes of risk groups. A Infiltration of immune cells in the risk subgroups. B Expression of common immune checkpoint genes in the risk subgroups. Differences between groups were analyzed by the Wilcoxon test. Statistically significant differences are indicated by "*" signs, where P < 0.05 is indicated by "*", P < 0.01 by "**", and P < 0.001 by "***", P < 0.0001 by "****", insignificant by ns

In both the risk groups, the expression of 14 common immune check loci (BTLA, CD40, CD70, CTLA4, HAVCR2, IDO1, LAG3, LMTK3, PDCD1, TIGIT, TJP1, TNFRSF14, TNFRSF18, and TNFRSF9) was observed. CD40, CD70, CTLA4, HAVCR2, IDO1, LMTK3, TIGIT, TNFRSF14, TNFRSF18, and TNFRSF9 had substantially different levels of expression between the two risk groups (P < 0.05) (Fig. 12B).

3.12 Drug sensitivity prediction

According to drug sensitivity analysis, a significant difference was observed in drug sensitivity to erlotinib (Fig. 13B) and selumetinib.BRD.A02303741 (Fig. 13C), BRD.A02303741.navitoclax (Fig. 13D), dasatinib (Fig. 13E), PD318088 (Fig. 13F), navitoclax.PLX.403 (Fig. 13G), navitoclax.piperlongumine (Fig. 13H), decitabine.navitoclax (Fig. 13I), UNC0638.navitoclax (Fig. 13J), ABT.737 (Fig. 13K), tretinoin.navitoclax (Fig. 13L), alisertib.navitoclax (Fig. 13M), navitoclax.birinapant (Fig. 13N), myriocin (Fig. 13O), and GSK.J4 (Fig. 13P) between both risk groups (Fig. 13A, adj.p value < 0.05 and |log2FC|> 0.5). Individuals in the group with high risk were more sensitive to GSK.J4, dasatinib, myriocin, and selumetinib. BRD.A02303741, erlotinib, and PD318088 individuals in the group with low risk. However, individuals in the group with high risk were less sensitive to tretinoin.navitoclax, navitoclax.birinapant, UNC0638.navitoclax, BRD. A02303741. Navitoclark, decitabine.navitoclax, alisertib.navitoclax, ABT.737, navitoclax.piperlongumine, and navitoclax.PLX.4032 than those in the low-risk group.

Fig. 13

Drug sensitivity variation in high- and low-risk groups. A Drugs with differences are shown in volcano plots, with red and green indicating higher and lower drug sensitivity, respectively, in the group with high risk. In addition, adj.p value < 0.05 and |log2FC|> 0.5 suggest substantial variations in drug sensitivity among the risk groups; B–P Box plots showing variation in drug sensitivity between the risk groups for erlotinib (B), selumetinib. Selumetinib.BRD.A02303741 (C), BRD.A02303741.navitoclax (D), dasatinib (E), PD318088 (F), navitoclax.PLX.4032 (G), navitoclax.piperlongumine (H), decitabine.navitoclax (I), UNC0638.navitoclax (J), ABT.737 (K), tretinoin.navitoclax (L), alisertib.navitoclax (M), navitoclax. birinapant (N), myriocin (O), and GSK.J4 (P), where orange and purple represent the groups with high and low risk, respectively

3.13 Somatic mutation analysis of risk groups

Fisher's exact test for somatic mutations was performed to detect differentially mutated genes in tumor samples between groups with high- and low-risk in the TCGA-LIHC dataset (2 cases with no mutation data and 366 cases with analyzed data). The genes with the most significant differences were TP53, ARID1B, TNRC18, HIPK3, and PDZRN4 (P < 0.01, Fig. 14A). Subsequently, mutations in FRGs were counted, with the highest mutation frequency being HERC2 (23/366, 6.28%), followed by USP24 (8/366, 2.18%), ATG5 (6/366, 1.63%), and PCBP1 (4/366, 1.09%) (Fig. 14B). Moreover, the number of Tmbs was counted between the risk groups and between FRG mutation/non-mutation groups. The number of Tmbs did not differ considerably between the risk groups (Fig. 14C), whereas the number of Tmbs was substantially larger in the FRG mutation group than that in the non-mutation group (Fig. 14D).

Fig. 14

Somatic mutation analysis of the TCGA-LIHC dataset. A Fisher's exact analysis of the differences in somatic mutations between both risk groups. B Waterfall plot of mutations in FRGs. C Violin plot of the number of Tmbs in the risk groups, where the horizontal coordinate denotes the two risk groups and the vertical coordinate denotes the TMB. The variations between groups were analyzed by the Wilcoxon test. D Violin plot of the number of Tmbs in the FRG mutation/non-mutation groups, where the horizontal coordinate denotes the risk group and the vertical coordinate denotes the TMB. The Wilcoxon signed-rank test was utilized to observe group variations

3.14 Somatic mutation features of risk groups

A total of eight mutation features were extracted from the TCGA-LIHCLIHC somatic mutation data (Fig. 15A). Among them, Sig1 is similar to SBS40 in the COSMIC database (unknown pathogen), Sig2 is similar to SBS46 (sequencing artifact), Sig3 is similar to SBS6 (MMR disorder), Sig4 is similar to SBS22 (aristolochic acid), Sig5 is similar to SBS49 (sequencing artifact), Sig6 is similar to SBS16 (unknown pathogen), Sig7 is similar to SBS17b (unknown pathogen), and Sig8 is similar to SBS28 (unknown pathogen). Sig1 and Sig2 had a higher prevalence of somatic mutations in most samples (Fig. 15B). The expression of mutant characteristics was then compared between the risk groups as well as between the FRG mutation/non-mutation groups. No difference in the expression of the mutation features was observed between the high- and low-risk groups (P > 0.05) (Fig. 15C). However, the FRG mutation group exhibited considerably greater levels of Sig1, Sig2, Sig3, Sig5, and Sig7 expression than the non-mutation group (P < 0.01) (Fig. 15D).

Fig. 15

Somatic mutation feature analysis of the TCGA-LIHC dataset. A Mutation patterns of somatic mutation features of LIHC samples and similar mutation features of the COSMIC database. B Composition of the mutation features of the samples. C Expression of each mutation feature in both risk groups. D Expression of each mutation feature in FRG mutation and non-mutation subgroups. Wilcoxon test was utilized to examine variations between groups. Statistically significant differences are indicated by "*" signs, where *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001, and ns represents insignificant. In addition, red and blue denote the high– and low-risk groups, respectively. COSMIC, the Catalogue of Somatic Mutations in Cancer

3.15 FTH1,FTL is highly expressed in hepatocellular carcinoma cells and hepatocellular carcinoma tissues.

We further validated the FTH1, FTL genes that were significantly associated with poor prognosis. We conducted PCR experiments using a normal liver cell line (LO2) and three liver cancer cell lines (HepG2, LM3, and 97-H). As shown in Fig. 16A, B, FTH1, FTL were significantly overexpressed in the three liver cancer cell lines compared with normal liver cells.

Fig. 16

A expression of FTH1 mRNA in different liver cancer cells and normal liver cells; B expression of FTL mRNA in different liver cancer cells and normal liver cells; C the HE staining results of the patient tissue samples. D expression of FTH1, FTL in tumor tissue; *P < 0.05, **P < 0.01, ***P < 0.001

We first performed HE staining on the patient tissue specimens we collected to distinguish between cancer tissue and adjacent normal tissue. (Fig. 16C). We further detected 60 pairs of liver cancer tissues by immunohistochemistry, and found that the expression of FTH1, FTL in tumor tissues was significantly higher than that in para-tumor tissues. This is consistent with previous bioinformatics results (Fig. 16D).