Introduction

Hepatocellular carcinoma (HCC), the prevalent sub-type of liver cancer and the fourth leading cause of cancer-related mortality globally, often impacts individuals suffering from chronic liver disease and cirrhosis1. Consequently, it claims the dubious distinction of being one of the deadliest cancers globally2. Illnesses commonly arise from viral hepatitis infections, chronic liver conditions, alcohol abuse, and metabolic disorders such as diabetes3. Although patients can be cured by liver transplantation, surgical resection and thermal ablation, long-term outcomes are compromised by high recurrence rates (up to 70% of cases 5 years after treatment)4. Clinical statistics reveal that a significant number of patients are often diagnosed at advanced stage, resulting in the inability to achieve a complete cure5. Even though multiple biomarkers for HCC have been identified to aid in surveillance and early detection, there are still a considerable proportion of individuals with poor survival6,7. Therefore, it is crucial to investigate new therapeutic targets to improve the treatment efficacy of HCC.

Research suggest that rapid progression of HCC and poor survival rates are frequently attributed to the presence of heterogeneous sub-populations of tumor cells8.

A key factor influencing the heterogeneity of HCC is the tumor micro-environment (TME)9. Inflammation in the liver micro-environment plays a pivotal role in cancer invasion and metastasis, and pro-inflammatory cytokines secreted by immune cells implicated in steatohepatitis can also promote cancer progression10. Neutrophils, as the most numerous type of white blood cells in human immune system, play a complex and important role in the TME11. Studies have shown that the accumulation of tumor-associated neutrophils (TANs) in tumors can be triggered by external stimuli from TME, leading to a switch between pro-tumor and anti-tumor phenotypes12.

Neutrophil extracellular traps (NETs) are web-like structures released by activated neutrophils, primarily composed of decondensed chromatin and intracellular granule proteins13. Initially, it was discovered to capture and eliminate extracellular pathogens, playing a protective role in antibacterial defense14. Tumor-associated neutrophils have been shown to readily form NETs in HCC, indicating that NETs play a significant role in the progression and metastasis of liver cancer15. Zhan et al.‘s mechanistic study revealed that the levels of NETs are markedly increased in hepatitis B virus (HBV)-positive HCC. Suppressing NETs via the TLR4/RAGE-ROS signaling pathway could potentially be advantageous for HCC patients16. Additionally, studies have found that NETs trigger tumorous inflammatory response and fuel HCC metastasis. Focusing on NETs, rather than the neutrophils directly, may present a viable strategy to combat HCC metastasis17. One study also proved that NETs can effectively predict the immunotherapy response and prognosis of HCC patients as imaging genomics biomarkers18. Therefore, analyzing the levels of NETs and their relationship with the survival of HCC patients is of great value. Moreover, the new and reliable prognostic model developed by NETs offers enhanced guidance for targeted therapy.

This study aims to identify prognostic genes associated with NETs through TCGA database, and to explore the correlation between NETs and TME. Thereafter, the nomograms were constructed using the prognostic genes. Studies on enrichment analysis, immune infiltration, and drug sensitivity provides a deeper understanding of the complex mechanisms of these prognostic genes. The ultimate aim is to offer potential guidance for targeted and immunotherapeutic treatments of HCC, as well as to provide fresh perspectives for prognostic evaluations.

Materials and methods

Data source

TCGA-HCC was downloaded on August 19, 2024, from the cancer genome atlas (TCGA, https://portal.gdc.cancer.gov), contained 374 LIHC tissue samples and 50 adjacent tissue samples (control cohort), among which existed 368 tumor tissue samples with survival information. The validation cohort GSE14520 (GPL3921) tracked to gene expression omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) database, including 221 LIHC tumor samples with survival information. The count and FPKM data from TCGA were logarithmically transformed (using 2 ^ x − 1 and log2(x + 1)) to compress the data range. Additionally, the data were further standardized using the Z-score normalization method, so that the mean of the data became 0 and the standard deviation was 1.

The na.omit function was directly used to delete the rows containing missing values. If the data information was not desired to be lost, the mean imputation method could be employed, that is, the missing values were replaced with the mean value of the corresponding column. The quality of the data and the pertinence of the analysis were ensured by removing the duplicate gene information and screening the samples according to the sample types. From a previous study19, a total of 87 neutrophil extracellular traps-related genes (NETs-RGs) were obtained. Referring to the methods of a previous study20, 1793 immune-related genes (IRGs) were acquired.

Differentially expressed genes (DEGs) identification

DESeq2 package (v 1.38.0)21 was engaged to pinpointed DEGs between cancerous and control tissues samples with survival information in the TCGA-LIHC (p.adj < 0.05 and |log2fold change (FC)| >1). Subsequently, ggplot2 package (v 3.4.1)22 was applied to create a volcano plot of DEGs, while the top 10 up-regulated and down-regulated DEGs were annotation in volcano plot ranked by |log2FC|. Furthermore, we utilized the ComplexHeatmap package (v 2.14.0)23 to generate a heatmap for these 20 DEGs.

Weighted gene co-expression network analysis (WGCNA)

First, a preliminary gene screening of NETs-RGs was conducted via univariate Cox regression (p < 0.05). Subsequently, we calculated the NETs-RGs scores for both the tumor cohort with survival information and the normal cohort in TCGA-LIHC using ssGSEA algorithm from GSVA package (v 1.42.0)24. Then, differences between the two cohorts were tested (p < 0.05). After that, 368 LIHC patients were cohorted via median NETs-RGs score and a survival analysis utilizing the Kaplan-Meier (K-M) was performed on the two cohorts with the aid of the survival package (v 3.5.3)25 (p < 0.05).

Next, in all samples of the TCGA-LIHC dataset, WGCNA analysis was completed through WGCNA package (v 1.7.1)26. Initially, LIHC samples were clustered to identify and remove outliers. PckSoftThreshold was employed to determine the soft value required to construct the scale-free network (R2 = 0.85). Subsequently, a hierarchical clustering dendrogram was constructed using a dynamic tree cut method (minModuleSize = 500). Finally, by calculating the correlation coefficient matrix and p-values between gene modules and NETs-RGs scores, we selected the modules most correlated with NETs-RGs scores (|correlation (cor)| > 0.3 and p < 0.05), the module genes were named key module genes.

Pinpointed candidate genes and function enrichment evaluation

Candidate genes were obtained through DEGs, IRGs and key module genes by ggvenn package (v 0.1.9). Subsequently, the clusterProfiler package (v 4.7.1.3)27 was utilized to perform gene ontology (GO) and kyoto encyclopedia of genes and genomes (KEGG) to research the potential pathways and functions of these candidate genes (p.adj < 0.05)28,29,30. An the enrichment results were sorted in ascending order by p.adj value, displaying the top 10 most significant functions; for KEGG, the top 5 most significant pathways were displayed. Then, a protein-protein interaction (PPI) network was generated by STRING (https://string-db.org/) (interaction score > 0.4) to gain insights into the interconnections between these genes and was visualized using Cytoscape (v 3.5.2), and genes with interactions were retained for further analysis.

Construction and validation of prognostic models

To further screen for genes with prognostic value from the candidate genes, we first took the following steps in TCGA-HCC: univariate Cox regression analysis (hazard ratio (HR) ≠ 1 and p < 0.05) and proportional hazards (PH) assumption test (p > 0.05) were performed via survival package. A HR value greater than 1 indicated that the gene was a risk gene for CC, an HR value less than 1 indicated that the gene was a protective gene for HCC, and an HR value of 1 indicated that the result was meaningless. Meanwhile, the 95% confidence interval (CI) values were calculated. Least absolute shrinkage and selection operator (LASSO) was used to select prognostic genes, which was accomplished by glmnet package (v 4.1.4)31. The LASSO analysis, through 10-fold cross-validation, could compress the regression coefficients of some unimportant variables to 0, thus achieving variable selection. In traditional regression methods, when dealing with multicollinearity, the estimated regression coefficients were unstable, and the standard errors were increased, resulting in inaccurate judgments of the significance of variables. The regularization feature of LASSO could make the coefficients of some of the independent variables with collinearity become 0, and only one or a few independent variables that had a greater impact on the dependent variable were retained. In this way, the problem of multicollinearity was effectively alleviated, and the stability of the model was improved32.Then, the Bootstrap method was used to evaluate the reliability of the model. Each gene in the model was sampled 1000 times with replacement. After each sampling, the Cox regression model was fitted, and the regression coefficients were recorded. Finally, the mean value, standard deviation, and 95% confidence interval of the regression coefficients for each gene were calculated. For small samples, the exact Cox regression was used instead of the asymptotic Cox regression, which could estimate the parameters and test the hypotheses more accurately. In order to ensure that the model was not overfitted and that sufficient statistical power was maintained when the sample size was small, the exact Cox regression was used instead of the asymptotic Cox regression for small samples, as it could estimate the parameters and test the hypotheses more accurately. After that, a predictive risk model was established by integrating these selected prognostic genes. The methodology for determining the risk score is outlined follow:\(\:\text{r}\text{i}\text{s}\text{k}\:\text{s}\text{c}\text{o}\text{r}\text{e}=\sum\:_{\text{j}=1}^{\text{n}}\text{E}\text{x}\text{p}\text{r}(\text{g}\text{e}\text{n}\text{e}\text{j})\times\:\text{C}\text{o}\text{e}\text{f}\left(\text{g}\text{e}\text{n}\text{e}\text{j}\right)\), where Expr genej and Coef genej represent prognostic gene expression levels and prognostic gene coefficients, respectively.

Then, scores indicating risk were computed for all LIHC specimens, and the median risk score was determined to stratify the samples in the TCGA-LIHC. Subsequently, risk curve plots and survival status plots were drawn to display the distribution of samples in two risk cohorts. Then, the survminer package (v 0.4.9) was used for K-M to ascertain the distinctions in survival status between the two cohorts (p < 0.05). After that, survivalROC package (v 0.4) was utilized to generate the receiver operating characteristic (ROC) curves to assess 1, 3, and 5 year survival probabilities. The external validation cohort GSE14520 was used to validate the risk model.

Independent prognostic analysis and nomogram construction

Subsequently, an independent prognostic analysis was conducted by integrating risk score and clinical characteristics (stage (stage I/II, stage III/IV), age (≤ 60, > 60), gender, stage T (T1/T2, T3/T4), stage N (N0, N1), stage M (M0,M1)) of tumor samples in TCGA-HCC, Wilcoxon test was done by stats package (v 3.6.2)33 for samples with two or fewer clinical characteristics and the risk score. For samples with three or more clinical characteristics and risk score, Kruskal-Wallis test was conducted using the Kruskal.test function. The data lacking clinical characteristics were deleted. This step was aimed to examine the distribution differences in risk scores among LIHC patients with survival information who had different clinical characteristics (p < 0.05).

Subsequently, based on TCGA-HCC, Cox regression analysis (p < 0.05) and PH assumption test (p > 0.05) were employed to integrate risk score and clinical characteristics, thereby identifying independent prognostic factors associated with LIHC.

After that, a nomogram was built using the rms package (v 6.5-1)34 based on the prognostic genes. Calibration plot and decision curve analysis (DCA) curves were generated using rms package.

Gene set enrichment analysis (GSEA)

Subsequently, relevant signaling pathways and bioinformatics of LIHC patients with survival information in two risk cohorts were explored. The reference gene set was c2.cp.kegg.v2023.1.Hs.symbols.gmt from MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/). In all samples of the TCGA-LIHC, firstly, DEGs between two risk cohorts were identified by DESeq2 package. Then, reliant on these DEGs, cor function was adapted for calculating the Spearman correlation coefficient, using the correlation coefficient as the sorting standard. Subsequently, GSEA was performed by clusterProfiler package (p < 0.05, false discovery rate (FDR) < 0.25, |normalized enrichment score (NES)| > 1). GSEA results were sorted in descending order based on the NES values and the top 5 pathways were selected for visualization.

Analyses of immune microenvironment

The ssGSEA algorithm was intended for determining the infiltration of 28 immune cell types35 between two risk cohorts in the TCGA-LIHC (excluding samples with p > 0.05). Differential immune cells obtained by comparing 28 immune cell infiltrations (p < 0.05). Subsequently, the cor function was designated to analyze the correlations between the differential immune cells, as well as between prognostic genes, risk score and these immune cells (|cor| > 0.3 and p < 0.05).

Drug sensitivity analysis

The relevant data on 138 standard chemotherapeutic drugs were collected to review the differences in responses between two risk cohorts by querying cancer drug sensitivity genomics (GDSC, https://www.cancerrxgene.org) database. Within TCGA-LIHC, pRRophetic package (v 0.5) was utilized to estimate the 50% inhibitory concentration (IC50) values for the drugs, and Wilcoxon test was applied to assess the differences in drug IC50 between two risk cohorts (p < 0.05). Subsequently, the top 10 drugs with the most significant differences in two risk cohorts were selected for presentation.

RNA extraction and reverse transcription-quantitative PCR (RT-qPCR)

To confirm the reliability of the identified prognostic genes, RT-qPCR was conducted using clinical specimens from individuals with LIHC and those from healthy controls. A sum of 5 tissue samples from LIHC patients with verified diagnoses, along with an equal number of 5 samples from the control group, were gathered from the First Hospital of Lanzhou University. All participants had provided their written informed consent. Medical Ethics Committee of the First Hospital of Lanzhou University granted approval for this study (LDYYLL2024-759). Then, RNA from each sample was extracted with 1 ml TRIzol reagent and 300 ul chloroform, and its concentration was quantified with NanoPhotometer N50. The cDNA was reverse transcribed using the SweScript First Strand cDNA Synthesis Kit (Servicebio, Wuhan, China). RT-qPCR amplification was accomplished by the CFX96 Real-Time Fluorescence Quantitative PCR Instrument (Bio-Rad, California, USA) for a total of 40 cycles. One cycle consisted of 1 min of pre-denaturation at 95 °C, denaturation at 95 °C for 20 s, annealing at 55 °C for 20 s and extension at 72 °C for 30 s. Finally, the amplification curve, melting curve, and Cq value were analyzed. The relative expression levels were determined by 2−△△CT method and statistical analysis and visualization were conducted using Graphpad Prism 5 (v 8.0)36. Within the scope of this research, GAPDH served as a housekeeping gene to validate the precision of our findings. Details regarding the RT-qPCR reaction mixture, the sequences of the primers, and the specific conditions for amplification were provided in supplementary Tables 1, 2, and 3, respectively.

Statistical analysis

Statistical analyses were conducted in R (v 4.3.2), with Wilcoxon test engaged to differentiate cohort pairs and p < 0.05 indicating significance.

Results

Recognition DEGs and key module genes

Within TCGA-HCC, 8,945 DEGs were identified (7,011 genes up and 1,934 genes down) (p.adj < 0.05) (Fig. 1a and b). Subsequently, a gene co-expression network was established based on NETs-RGs through WGCNA. Initially, we scored NETs-RGs in both tumor and normal samples, and found a significant distinction in NETs-RGs scores between the two cohorts (p = 0.024) (Fig. 1c). Then, based on the median NRGs-RGs score, high and low scoring cohorts were divided, K-M curve showed a significant survival difference between two scoring cohorts (p = 0.00049) (Fig. 1d). Next, hierarchical clustering analysis was performed on all samples and identified no obvious outliers (Fig. 1e). Subsequently, the optimal soft-thresholding power was determined to be β = 10 (R² > 0.85) (Fig. 1f). Then, a hierarchical clustering dendrogram was constructed, dividing the genes into 11 co-expression modules (Fig. 1g). Ultimately, MEbrown module (cor = -0.78, p = 7e-77) was selected as the key module (Fig. 1h), and the 1,870 genes within MEbrown module were designated as key module genes.

Fig. 1
figure 1

Screening for DEGs. (a) Volcano plot of DEGs in TCGA-HCC. (b) Heatmap of DEGs in TCGA-HCC. The numerical range was from − 1 to 3, representing values related to gene expression levels. The legend on the right side showed that colors corresponded to expression densities, with red indicating high expression density and blue indicating low expression density. The black lines represented different quantiles, such as 0%, 25%, 50%, 75%, and 100%, and the data distribution was presented. (c) NETs-RGs scores between tumor and normal samples. (d) The survival probability of the two risk groups. Blue represents patients in low risk group and red represents patients in high risk group (p-value = 0.00049) LIHC = 374, control = 50, NETs-RGs = 87. (e) The hierarchical clustering analysis among all samples. (f) Determination of the soft thresholding power. The y-axis of the left panel represents the scale-free fit index and the y-axis of the right panel represents the mean connectivity. The x-axis represents power value. (g) Dendrogram of differentially expressed genes clustered. The colors represent the identified gene modules. (h) Correlations of gene modules with NETs-RGs scores.

Full size image

Recognition and functional annotation of candidate genes

By intersecting DEGs, IRGs and key module genes, 78 candidate genes were obtained (Fig. 2a, supplementary Table 4). The overlapping situations of these genes with NET-RGs were examined, identified 5 overlapping genes including MMP9, FPR1, IL1B, IL6 and FPR2 (Supplementary Fig. 1). This could help to highlight the potential role of NETs in hepatocellular carcinoma. GO analysis found that these candidate genes enriched in 316 terms, including 31 molecular functions (MF), 7 cellular components (CC) and 278 biological processes (BP), with the top 10 enriched terms comprising immunoglobulin production and leukocyte mediated immunity (p.adj < 0.05) (Fig. 2b). Additionally, KEGG enrichment identified 22 signaling pathways, including cytokine-cytokine receptor interaction and cytokine receptor (p.adj < 0.05) (Fig. 2c). Subsequent PPI results showed that 30 genes had no interaction relationships, while CETP, PTGER2, GBP2, and UTS2 each had only one interaction (interaction score > 0.4) (Fig. 2d). After removing these 30 genes without interaction relationships, 48 candidate genes were used for further research (supplementary Table 5).

Fig. 2
figure 2

Acquisition and functional enrichment analysis of candidate genes. (a) Venn diagrams of candidate genes, DEGs = 8945, IRGs = 1793, and key module genes = 1870. (b) The GO outcomes are displayed with a histogram. (c) A histogram was constructed to illustrate the KEGG outcomes. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; BP, biological process; CC, cellular component; MF, molecular function. (d) Screening 48 candidate genes by PPI network.

Full size image

Prognostic model development and evaluation

Based on all tumor samples with survival information in TCGA-HCC, five candidate prognostic genes were determined through univariate Cox regression analysis including HMOX1 (HR = 1.22, 95% CI: 1.07–1.40, p = 0.003), MMP9 (HR = 1.14, 95% CI: 1.03–1.26, p = 0.011), TNFRSF4 (HR = 1.33, 95% CI: 1.06–1.66, p = 0.01), MMP12 (HR = 1.16, 95% CI: 1.02–1.32, p = 0.02), FLT3 (HR = 0.23, 95% CI: 0.07–0.81, p = 0.02), among which only FLT3 was a protective gene (HR < 1) (Fig. 3a). Univariate Cox regression analysis was conducted to evaluate the association strength and statistical significance between each gene and the prognosis of HCC patients by calculating the HR values, their 95% CIs, and p-values. The p-values of these five genes were all less than 0.05, indicating that there were statistically significant associations between them and the survival of HCC patients, and the null hypothesis of no association between gene expression and survival outcome was rejected. Meanwhile, the HR values of HMOX1, MMP9, TNFRSF4, and MMP12 were greater than 1, while the HR value of FLT3 was less than 1, and the 95% CIs of each did not contain 1, which further supported the correlations between these genes and the prognosis of HCC. This demonstrated that obtaining these five genes through univariate Cox regression analysis was reasonable and statistically grounded. Subsequently, the PH assumption test was conducted for these five genes, and all five genes passed the test (supplementary Table 6). Then, LASSO analysis was performed on these five candidate prognostic genes, and it was found that when lambda = 0.002, the penalty values for all five candidate prognostic genes were not zero (Fig. 3b). The Bootstrap results showed that the mean value of the coefficient of HMOX1 was 0.2009, with the hazard ratio (HR) approximately equal to 1.22 (exp(0.2009) ≈ 1.22), and the mean value of the coefficient of FLT3 was − 1.5377, with the HR approximately equal to 0.215 (exp(-1.5377) ≈ 0.215). The small standard deviations indicated that the coefficient estimations were stable. For example, the maximum standard deviation (SD) of TNFRSF4 was 0.124, which showed that the stability of its coefficient estimation was relatively low. The minimum SD of MMP9 was 0.05, suggesting that its coefficient estimation was relatively stable. The 95% confidence intervals (CIs) of all genes did not contain 0, which was consistent with the result of the univariate Cox regression analysis with p < 0.05, verifying the significance(supplementary Table 7). Therefore, HMOX1, MMP9, TNFRSF4, MMP12, and FLT3 were retained as prognostic genes for further research.

Fig. 3
figure 3

Construction of a prognostic model in TCGA-HCC. (a) Univariate Cox regression analysis in TCGA-HCC. (b) Coefficient curve. Different colors represent different genes. No zero values were selected as a penalty coefficient (Left). The minimum lambda of the lasso model was selected via 10 folds of cross-validation. Lambda was determined when the partial likelihood deviance was smallest (Right). (c) The risk score of the different risk groups. (d) The survival probability of the different risk groups. Blue represents high-risk patients and red represents low-risk. The patients were divided according to their median risk score. (e) Prognostic genes in the TCGA-HCC dataset were analyzed using ROC curves.

Full size image

Next, a risk model was established using the expression levels of these five prognostic genes and coefficients derived from LASSO analysis to determine the risk scores for LIHC patients: Risk Score = 0.20 × HMOX1 + 0.12 × MMP9 + 0.154 × TNFRSF4 + 0.06 × MMP12–2.60 × FLT3. Then, the risk scores for tumor samples in TCGA-HCC (0.94) and GSE14520 (-6.83) were calculated, and patients were categorized into two distinct risk cohorts based on the median risk score. The distribution pattern of risk scores and the associated survival outcomes were graphed based on the patients’ risk scores in ascending order, demonstrating that a higher risk score was significantly associated with a higher mortality rate (Fig. 3c and Supplementary Fig. 2a). K-M analysis revealed that the high-risk group had a markedly lower survival rate across both datasets (p < 0.05) (Fig. 3d and Supplementary Fig. 2b). In the TCGA-HCC, ROC analysis further confirmed its accuracy in predicting the survival rates at 1 (0.69), 3 (0.63) and 5 years (0.70) (for GSE14520, the values were 0.68, 0.68 and 0.63) (Fig. 3e and Supplementary Fig. 2c).

Nomogram construction and clinical characteristic analysis

Cox analysis and PH assumption test indicated that risk score emerged as the sole independent predictor of prognosis for LIHC (p < 0.05) (Fig. 4a and b and supplementary Table 8). Then, a nomogram was built via prognostic genes (Fig. 4c). The calibration plots exhibited that the nomogram’s predictions for 1, 3, and 5 year survival probabilities closely matched the actual outcomes, aligning well with the reference line, indicating a high level of predictive accuracy (Fig. 4d). Following that, the results from DCA portrayed a net benefit above zero for the nomogram, signifying its effective construction (Fig. 4e). Clinical characteristic analysis revealed significant differences in risk scores among stages and T stages (p < 0.05) (Fig. 4f). The risk score was higher in the T3/T3 group than in the T1/T2 group, and it was higher in the stage III/IV group than in the stage I/II group. This indicated that the risk score was closely associated with the tumor stage of LIHC. A higher risk score might have predicted a more advanced tumor stage. It was suggested that clinicians could incorporate the risk score into the disease assessment system for LIHC patients to assist in judging the disease progression degree of patients. This would enable timely adjustment of treatment regimens and adoption of more targeted intervention measures to improve patient prognosis.

Fig. 4
figure 4

Construction of nomogram and analysis of clinical characteristic with riskScore. (a) Univariate Cox regression and multivariate Cox regression (b) analyses of independent predictor of prognosis for LIHC. (c) Nomogram was constructed based on the expression of prognostic genes. (d) Calibration curves of nomogram for predicting OS at 1-year, 3-year and 5-year in the TCGA-HCC dataset. (e) DCA curves were constructed based on prognostic genes. (f) Box plots of clinical characteristic (age, T_stage, N_stage, M_stage, stage, and gender) between subgroup in the TCGA-HCC cohort. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns p > 0.05.

Full size image

Enrichment, immune landscape and drug sensitivity in two risk cohorts

The DEGs in two risk cohorts were primarily enriched in retinol metabolism and drug metabolism cytochrome p450 (Fig. 5a). The heatmap of 28 different immune cells showed in the Supplementary Fig. 3 between two risk cohorts, with statistical difference in 16 immune cells, including regulatory T cells and MDSCs (Fig. 5b). Subsequently, Spearman’s correlation analysis indicated that regulatory T cells were statistically positively correlated with MDSCs (cor = 0.89, p = 9.13e-126) (Fig. 5c and supplementary Table 9). The correlation analysis between differential immune cells and prognostic genes showed that MDSCs had notable link to HMOX1 (cor = 0.54), MMP9 (cor = 0.62), TNFRSF4 (cor = 0.48), MMP12 (cor = 0.43), and FLT3 (cor = 0.43) (Fig. 5d and supplementary Table 10).

Fig. 5
figure 5

Analysis of GSEA, immune microenvironment, and drug sensitivity. (a) Significantly enriched pathways in the KEGG gene sets between the two risk groups in TCGA-HCC cohort. (b) Box plot of the difference in immune cell content between two risk groups. (c) Correlations between infiltrating immune cells. Red and blue colors indicate positive and negative correlations, color intensity represents the degree of correlation. (d) Correlations between infiltrating immune cells and prognostic genes. (e) The differences of sensitivity of patients between two risk groups. Lower IC50 is equal to better sensitivity. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns p > 0.05.

Full size image

In drug sensitivity analysis, 74 drugs showed substantial discrepancies between the two risk cohorts (p < 0.05). Axitinib and AS601245 had higher IC50 values in the high-risk cohort, while ABT.888 and A.443,654 had lower IC50 values (Fig. 5e).

Experimental validation of prognostic genes

As shown in Fig. 6, the results of RT-qPCR analysis indicated that the expression levels of HMOX1 (p < 0.01) and FLT3 (p < 0.05) in LIHC tissues were significantly decreased compared to the control, while the expression of MMP9 (p < 0.05) and TNFRSF4 (p < 0.0001) were the opposite. Although the expression of MMP12 (p > 0.05) was elevated in LIHC, there was no significant difference compared to the control group. In LIHC tissues, the expression levels of HMOX1 and FLT3 were significantly decreased compared to the control group. The significant changes in gene expression implied a close association with the disease state. When LIHC occurred in the body, the down - regulation of these two genes might have affected a series of key physiological processes within the cells. The expression of MMP9 and TNFRSF4 in LIHC tissues showed an opposite trend to that of HMOX1 and FLT3, and was significantly higher than that in the control group. The significantly up - regulated expression of these two genes in LIHC tissues reflected their active roles in the process of tumorigenesis and development. Therefore, they could be used as potential biomarkers for monitoring the progression of LIHC or evaluating the prognosis of patients. The expression of MMP12 was elevated in LIHC tissues, but there was no significant difference compared to the control group. However, its expression trend still indicated a certain connection with the LIHC state. Although the current results did not reach statistical significance, under a larger sample size or different research conditions, the changes in its expression might provide more information related to LIHC, and it still had the potential to be used as an auxiliary biomarker.

Fig. 6
figure 6

Relative mRNA expression of the prognostic genes was analyzed by qRT-PCR. *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001; ns p > 0.05, n = 6.

Full size image

Discussion

HCC is one of the most common cancers worldwide and represents a major challenge for healthcare worldwide37. Hundreds of thousands of people die from HCC globally each year, with most patients diagnosed at a late stage, missing the optimal window for treatment38. Effective and reliable prognostic biomarkers are crucial for clinical treatment. Currently, new studies have emphasized that NETs facilitated immune escape from HCC39. This study examined the expression levels of genes reported in the literature to be associated with NETs19 and immune in HCC and healthy samples and found that most of these genes were differentially expressed. Subsequently, a new prognostic model consisting of HMOX1, FLT3, MMP9, MMP12, TNFRSF4 and was constructed using univariate Cox and LASSO, followed by an in-depth analysis of the immune microenvironment.

In biological research, various molecules play diverse and crucial roles in different immune cell types, having a significant impact on aspects such as cell function, disease development, and immune responses. Specifically, HMOX1 is expressed in macrophages, where it assumes a pivotal role in intracellular oxidative stress responses and immune regulation, contributing to the maintenance of normal macrophage function40. FLT3 is predominantly expressed in dendritic cells, where it is closely associated with their development, maturation, and antigen - presenting functions, thus influencing the body’s immune response41. Neutrophils are capable of expressing MMP9, which is released during inflammation and tissue repair to degrade the extracellular matrix, facilitating the migration of inflammatory cells and tissue remodeling42. MMP12 is highly expressed in alveolar macrophages, where it takes part in lung inflammation and tissue repair by degrading the extracellular matrix of alveolar walls, and is related to the development of lung diseases such as chronic obstructive pulmonary disease43,44. Monocytes may also express MMP12 during their differentiation into macrophages, participating in inflammatory responses and immune regulation, and playing a part in the body’s innate immune defense45. Besides CD4 + T cells, other T - cell subsets, such as CD8 + T cells, can also express TNFRSF4, which is involved in T - cell activation, proliferation, and immune regulation, affecting the intensity and duration of T - cell - mediated immune responses46,47. Antigen - presenting cells like macrophages and dendritic cells also express TNFRSF4, where it interacts with T cells during antigen presentation to regulate the balance between immune activation and immune tolerance48.

Heme Oxygenase 1 (HMOX1) is a catalytic enzyme that facilitates the degradation of heme into iron, carbon monoxide and biliverdin, playing a significant role in cellular response to inflammation49. Studies have indicated that in HCC, the tumor microenvironment is marked by a high level of oxidative stress. The up-regulation of HMOX1 might change the activation pattern of neutrophils, thus promoting the release of NETs48,50. Nevertheless, the chronic and excessive release of NETs can result in an imbalance within the inflammatory microenvironment, creating conditions that are conducive to the proliferation and metastasis of tumor cells51. This could act as a double-edged sword in the progression of HCC, and the regulatory role of HMOX1 on NETs is one of the crucial factors affecting this balance. In our study, the results of the univariate Cox regression analysis indicated that the HR for HMOX1 was 1.225, with a p value of 0.003. Specifically, within a 95% confidence interval of 1.069–1.403, a one - unit increase in HMOX1 expression was significantly associated with a 22.5% increase in the risk of disease progression or death in HCC patients. Since HMOX1 has been identified as a risk factor for HCC, it is hypothesized that HMOX1 might be involved in the biological processes promoting the growth, invasion, or metastasis of HCC, thus playing a crucial role in the development and prognosis of HCC52. Park et al. found that HMOX1 expression was not associated with patient survival or recurrence, indicating that HMOX1 does not appear to play a role as a prognostic factor in HCC patients53. Nevertheless, Zheng et al. confirmed through chromosome conformation capture experiments that Donafenib and GSK-J4 synergistically induce ferroptosis in HCC by up-regulating HMOX1 expression, which represented a novel synergistic lethal interaction54. Over-expression of HMOX1 reduced the proliferation of human HBV-HCC cells and inhibited their migratory and invasive capabilities55. There were also suggested that HMOX1 may be regarded as a favorable prognostic factor in HBV-HCC patients undergoing liver resection56.

Fms-like tyrosine kinase 3 (FLT3) is a significant member of the class III receptor tyrosine kinase (RTK) family, playing a crucial role in the proliferation of hematopoietic cells and lymphocytes57. Although current research regarding the direct connection between FLT3 and NETs is scarce, considering that neutrophils are derived from hematopoietic stem cells, abnormal FLT3 signaling might change the differentiation pathway of hematopoietic stem cells58. Consequently, this could impact the quantity and functional status of neutrophils, thus indirectly affecting the generation of NETs. Aydin discovered that FLT3 is pivotal in the initiation and progression of HCC. Targeting FLT3 may represent a promising strategy for combating this tumor59. In our study, the HR for FLT3 was 0.231, with a p value of 0.022, suggesting that FLT3 is a protective gene in HCC. An increase in FLT3 expression is associated with a decreased risk of HCC. The 95% confidence interval of 0.066–0.809 further validates this finding. Thus, FLT3 might be involved in pathways that inhibit the growth of HCC, induce apoptosis of cancer cells, or enhance the immune response against HCC cells60. Lai found that novel FLT3 / AURK multikinase inhibitors were effective against sorafenib-resistant HCC57. Further studies have shown that sorafenib can slow down the progression of tumors in HCC patients who exhibit high levels of FLT3 expression61. In this study, the expression of HMOX1 and FLT3 in the HCC group was notably lower compared to healthy groups, signifying its positive factors in the prognosis of HCC.

Matrix metalloproteinase-9 (MMP9) is a crucial mediator of tumor initiation and progression62. In our study, univariate Cox regression analysis demonstrated that the HR for MMP9 was 1.140, with a p value of 0.011. This finding suggests that MMP9 serves as a risk factor for HCC. An elevated level of MMP9 expression is correlated with an increased risk of HCC and might facilitate the biological processes related to its growth, invasion, or metastasis63In HCC, MMP9 expression is significantly elevated in the gain of function (GOF) CTNNB1 mutations (CTNNB1 GOF). so CTNNB1 GOF mutations promoted the formation of an inhibitory tumor micro-environment by triggering MMP9 secretion64. In HBV-infected HCC patients, Chen et al. found that HBV stimulated peripheral blood mononuclear cells (PBMCs) and macrophages to express MMP-9. More interestingly, MMP9 in turn promotes HBV replication by inhibiting the effects of IFN-α65. Meanwhile, previous studies have suggested that VEGFA signal could activate tumor-associated macrophages (TAM) _ SPP1-derived MMP9 to promote the invasion and metastasis of HCC, which may provide new insights into the clinical treatment of HCC patients66. Moreover, targeted drug studies have revealed that indocurcumin effectively combats metastatic HBV-positive HCC by inhibiting the activity of MMP967.

Matrix metalloproteinase-12 (MMP12), also known as macrophage metalloelase, was first identified in human alveolar macrophages68. Univariate Cox regression analysis disclosed that the HR for MMP12 was 1.163, with a p - value of 0.02. This finding implies that MMP12 is a risk factor for HCC. An elevated expression of MMP12 is associated with an increased risk of HCC and may be implicated in biological processes promoting the growth, invasion, or metastasis of HCC69. Many studies have shown that the over-expression of MMP12 is associated with the poor prognosis of HCC, which may be an important factor affecting the efficacy of TACE70,71,72. Furthermore, He et al. found that MMP12 expression was associated with tumor FOXP3 regulatory T cell infiltration and poor prognosis in HCC68. However, some studies have proposed that blood MMP-12 has good sensitivity and higher accuracy but poor specificity in the diagnosis of HCC69. MMP9 and MMP12 might degrade the extracellular matrix around NETs, thus influencing the structural stability and distribution of NETs69. In HCC, the heightened levels of MMP9 and MMP12 secreted by tumor cells not only facilitate tumor cell invasion and metastasis by breaking down the basement membrane but also potentially modify the existence of NETs in the tumor microenvironment73. During the invasion and metastasis process in HCC, the interaction among MMP9, MMP12, and NETs seems to be especially intricate and significant74.

Tumor Necrosis Factor Receptor Superfamily Member 4 (TNFRSF4), named OX40, is an important costimulatory molecule for T cell proliferation and survival, plays a key role in regulating the immune response, cell survival and the inflammatory response75. Research findings indicate that the activation of the TNFRSF4 signaling pathway can impact neutrophil function, thereby influencing the generation of NETs76. In HCC, tumor cells may disrupt the normal functioning of neutrophils and modulate the release and activity of NETs by regulating the TNFRSF4 signaling pathway77. Univariate Cox regression analysis revealed that the HR for TNFRSF4 was 1.327, with a p-value of 0.014. This indicates that TNFRSF4 is a risk factor for HCC. Elevated expression of TNFRSF4 is associated with an increased risk of HCC and may be involved in biological processes that promote the growth, invasion, or metastasis of HCC77. Interestingly, Yan et al. found that the expression of TNFRSF4 in HCC was significantly higher than in normal tissues, which was identified as a risk factor for overall survival in male patients77. Notably, Wang et al. found that the over-expression of TNFRSF4 affected the immune cell infiltration in HCC76. Studies have proposed the combined intratumoral administration of CpG and anti-TNFRSF4 antibodies as a promising immunotherapy approach for HCC78. In this study, we also observed that the expression of TNFRSF4 and MMP9 in the HCC group was notably higher compared to the healthy group, which confirmed that their high expression was associated with poor prognosis.

Furthermore, the enrichment analysis revealed that those DEGs in two risk groups were primarily enriched in ‘drug metabolism cytochrome p450’ signaling pathway. Cytochrome P450, a group of enzymes predominantly expressed in the liver, plays a pivotal role in numerous biochemical processes79. One study suggested that cytochrome P450 2E1 plays a dynamic role in the pathogenesis of HCC, which could serve as a potential therapeutic target for liver carcinogenesis80. In this study, we hypothesized that the expression of these five biomarkers affects the progression of HCC by participating in ‘drug metabolism cytochrome p450’ signaling pathway. Additionally, we found that Axitinib had higher IC50 values in the high-risk cohort, which confirmed that the combination of Axitinib proposed by Yang et al. was well tolerated in the treatment of advanced HCC81.

The role of the tumor TME was receiving heightened attention due to its substantial impact on clinical outcomes and therapeutic responses82. In this study, 28 different types of immune cells were specifically selected for analysis. This is because these cells represent multiple key aspects of the tumor immune response. For example, Tregs, as crucial regulators of the immune response, can inhibit the activity of effector T cells, thereby impeding the body’s immune surveillance and killing of tumor cells83,84. MDSCs also play a central role in immunosuppression. They can inhibit the functions of immune cells through various mechanisms, creating conditions for the immune escape of tumor cells85. We found that MDSC (myeloid derived suppressor cell) had notable link to HMOX1, MMP9, TNFRSF4, MMP12 and FLT3. Higher infiltration of MDSCs in HCC compared to controls86. Li et al. proposed that MDSC- targeted therapy may be the best choice for HCC patients with higher neutrophil counts87. In detail, MDSCs exhibit a positive correlation with HMOX1 (cor = 0.54). This correlation indicates that MDSCs may up-regulate the expression level of HMOX1, assisting HCC cells in withstanding oxidative stress, thereby enhancing their survival ability and promoting tumor progression88. MMP9 has the function of degrading the extracellular matrix, creating conditions for the migration and invasion of HCC cells89,90. Meanwhile, there is a strong correlation between MDSCs and MMP9 (cor = 0.62), suggesting that MDSCs may enhance the invasion and metastasis ability of HCC cells by promoting the expression of MMP9, exerting a crucial impact on the malignant development process of HCC91. TNFSRF4 is involved in the regulation of immune cell activation and proliferation, and MDSCs are associated with it (cor = 0.48). MDSCs may inhibit the functions of immune cells by interfering with the TNFSRF4 related immune-regulatory signaling pathways, thus helping HCC cells achieve immune escape92. MMP12 is also associated with MDSCs (cor = 0.43). Its function is similar to that of MMP9 and can promote the metastasis of HCC cells by degrading the extracellular matrix93. FLT3 plays an important role in the proliferation and differentiation of hematopoietic stem cells. In HCC, when FLT3 is abnormally activated, it can promote the growth and survival of tumor cells94. The association between MDSCs and FLT3 (cor = 0.43) implies that MDSCs may be involved in regulating the proliferation and survival signaling pathways of HCC cells95. Therefore, we believed that these five biomarkers may influence the occurrence of HCC by regulating the expression of MDSC to regulate the immune microenvironment of HCC. This finding further supports our belief that HMOX1, MMP9, TNFRSF4, MMP12 and FLT3 have potential as biomarkers in HCC and we remain keenly interested in exploring their specific molecular mechanisms.

Drug - sensitivity analysis showed that 74 drugs exhibited significant differences between the two risk groups (p < 0.05). The IC50 value, a key indicator of drug potency, varied significantly for specific drugs across different risk cohorts. Specifically, Axitinib and AS601245 had higher IC50 values in the high - risk group, indicating that patients in the high-risk group require higher concentrations of these two drugs to achieve 50% cell - growth inhibition96. This suggests that tumor cells in the high-risk group may have stronger drug-resistance mechanisms, such as enhanced drug efflux pump activity or altered drug targets97. For these patients, conventional doses of Axitinib and AS601245 may not be effective. Clinicians may need to consider increasing the drug doses, which may lead to more severe side effects, or explore combination - drug regimens to enhance the efficacy of these two drugs using other agents98. In contrast, ABT.888 and A.443,654 had lower IC50 values in the high - risk cohort, indicating that the high - risk group is more sensitive to these two drugs. ABT.888, as a poly (ADP-ribose) polymerase (PARP) inhibitor, can specifically act on tumor cells with DNA damage repair defects99. It is speculated that tumor cells in the high - risk group may have abnormal DNA repair pathways, making ABT.888 and A.443,654 more effective, so low - dose administration can be used to enhance efficacy and reduce side effects. Based on the above findings, targeted therapy has broad prospects. For example, high - risk patients who are resistant to Axitinib but sensitive to ABT.888 can try ABT.888 targeted therapy. Similarly, the low-risk cohort can also select drugs according to IC50 values and tumor characteristics. Looking ahead, clinical research should conduct large-scale trials for different patient subgroups, use techniques such as whole - genome sequencing to clarify the associations between drug sensitivity and genetic and clinical characteristics, establish prediction models, and study the dynamic changes during treatment to optimize treatment strategies and improve the clinical outcomes of cancer patients.

Currently, the samples used in this study were sourced from bulk RNA-sequencing samples in public databases, and it may not be possible to precisely distinguish the signals of each cell type. In follow-up research, single-cell RNA sequencing or other cell-isolation techniques will be employed to further validate which cell types these signals specifically originate from. In addition, the small sample size of RT-qPCR may lead to large result deviations and insufficient representativeness, making it difficult to reflect the overall real situation. Clinical samples will be increased for research in the future. Moreover, research on 28 types of immune cells also lacks experimental support. Techniques such as flow cytometry and immunohistochemical experiments should be used to verify their characteristics and functions. The role network of HMOX1, MMP9, TNFRSF4, MMP12 and FLT3 in the progression of HCC requires further constructed in cell experiments and animal models for functional exploration. In future research, specific markers can be employed to sort neutrophils, and cell - lines can be utilized to investigate the expression patterns of TNFRSF4 in different immune cells. Meanwhile, experiments should be conducted to verify the types of immune cells. Moreover, the dataset used in this study lacks treatment history and comorbidity data. Future research should collect more of such data and carry out in - depth investigations into treatment history and comorbidities, so as to provide more evidence for personalized treatment. We plan to further explore these mechanisms through molecular approaches in the future.

Conclusion

The risk model constructed by the signature composed of HMOX1, MMP9, TNFRSF4, MMP12 and FLT3 can predict the prognosis of HCC patients, and these biomarkers were related to tumor immunity. This study provides a novel direction, theoretical foundation and potential target therapy for HCC. Our findings confirm the up-regulation of MMP9, TNFRSF4 and down-regulation of HMOX1, FLT3 in HCC patients, which presents a fresh perspective for the treatment of HCC.