A novel immunogenic cell death-related genes signature for predicting prognosis, immune landscape and immunotherapy effect in hepatocellular carcinoma

Abstract

Objective

Immunogenic cell death (ICD) has emerged as a promising strategy to activate the adaptive immune response, modulate the tumor microenvironment (TME) and enhance the efficacy of immune therapy. However, the relationship between ICD and TME reprogramming in hepatocellular carcinoma (HCC) remains poorly understood.

Methods

Transcriptional profiles and clinical spectrum of 486 HCC patients were obtained from TCGA and GEO databases. We utilized consensus clustering analysis to construct two distinct molecular subtypes and established an ICD-based scoring system (named ICD score) via WGCNA and LASSO Cox regression to predict the prognosis of the HCC cohort. Then we employed CIBERSORT and ESTIMATE methods to analyze the immune landscape of ICD score in HCC. Subsequently, the immunophenoscore (IPS) and tumor immune dysfunction and rejection (TIDE) analyses were performed to determine whether the ICD score could influence the immune therapeutic effect. Based on the ICD scoring system, a novel nomogram was generated to provide a numerical probability of HCC patients’ overall survival (OS).

Results

We identified two independent ICD clusters (cluster A/B), and cluster B possessed a worse prognosis and higher immune cell infiltration. Using ICD scoring system, the HCC patients were divided into high- and low-ICD-score groups. Through integrative analyses, the high-ICD cohort owned advanced TNM stage, high pathologic grade and increased suppressive immune cell enrichment. We developed a nomogram containing the ICD score, demonstrating a high predictive accuracy with a C-index of 0.703. We further discovered that PSMD2 and PSMD14 could serve as ICD-associated prognostic biomarkers and therapeutic targets in HCC.

Conclusion

The ICD score exhibits a high degree of reliability for predicting prognosis and may provide valuable guidance for the selection of immunotherapy for HCC patients. This novel scoring system enables the estimation of clinical immunotherapy response for HCC patients, offering new opportunities for personalized immunotherapy.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00432-023-05370-1.

Background

Hepatocellular carcinoma (HCC) is a major contributor to tumor-related deaths worldwide, ranking as the fifth most common malignancy(Llovet et al. 2023; Rong et al. 2023). HCC is commonly associated with cirrhosis caused by viral hepatitis B infection. Although great progress has been made in the diagnosis and treatment of HCC, the frequent recurrence or metastasis of HCC is difficult to prevent due to an insufficient understanding of its complex molecular pathogenesis(Devarbhavi et al. 2023; Cassese et al. 2022; Oura et al. 2023). Presently, immune therapy has gained significant attention, which has been applied to various malignancies. However, as for the HCC cohort, most of the patients showed limited response to such treatment as a result of suppressive TME reprogramming(Yu 2023). Because of lacking co-stimulatory signals, tumor cells could not trigger anti-tumor immunity due to insufficient immunogenicity. Then dendritic cells (DCs) also could not effectively transmit antigens to responsive cells, which resulted in immunological tolerance(Lee and Radford 2019; Pinato et al. 2020; Mittal et al. 2014).

Within the tumor microenvironment (TME), immunogenicity is a favorable clinical feature in part driven by the adaptive immune cells. However, malignancies often blockaded such anti-tumor activity by facilitating the phenotype of suppressive immune cells (including MDSCs, Tregs and TAMs), which could reduce the sensitivity of immunotherapy. At present, a novel kind of regulatory cell death was verified, which raised the immunogenicity in cancerous cells. And such a cell death method was concepted as immunogenic cell death (ICD)(Galluzzi et al. 2020; Fucikova et al. 2020; Petroni et al. 2021; Vaes et al. 2021). Activating the ICD process not only enables cancer cells to be sensitive to the cancer immune therapy but also leads to the increase in tumor-specific antigens which convert tumor condition from “immune-suppressive” to “immune-activated”(Lu et al. 2023). Recently, accumulating clinical and experimental data have revealed that ICD could prime an anticancer immune response within immunotherapy(Galon and Bruni 2019; Rabinovich et al. 2007; Hogg et al. 2020). In breast cancer, drug carriers containing indoleamine 2,3-dioxygenase-1(IDO1) siRNA and ICD inducer would be effective anticancer regimens to modulate the immunosuppressive TME by reversing the IDO1-mediated immunosuppression in a synergistic combination with ICD induction(Kim et al. 2022). The relief of the acidity in the TME promotes ICD alleviating the immunosuppressive microenvironment and synergistically enhancing the anti-tumor immune response(Li et al. 2023). Therefore, investigating the role of ICD in HCC may provide new insights into the development of effective immunotherapy strategies. Previous studies also revealed that ICD would trigger anti-tumor immune responses, particularly T cell responses. Moreover, ICD-induced danger-associated molecular patterns (DAMPs) could stimulate the maturation of DCs, leading to the activation of T cells and the generation of tumor-specific immune responses(Galluzzi et al. 2018; Kepp et al. 2014). Hence, it is reasonable to hypothesize that the process of ICD could potentially exert a substantial influence on the immune microenvironment of HCC. Investigating the correlation between ICD and the immune system could potentially yield novel perspectives on the design and implementation of immunotherapeutic strategies for HCC.

Our study revealed that the variation in ICD expression in HCC has an impact on both the immune microenvironment and the survival of patients. Based on the results obtained, we developed a prognostic scoring system that is associated with ICD and subsequently integrated it into a nomogram for the purpose of forecasting the survival of patients. Furthermore, downstream genes modulated by ICD were identified as potential targets for HCC treatment. The collective endeavors may reveal comprehension of the prognostic potential of genes associated with ICD and present immunotherapeutic approaches and remedies for patients with HCC, thereby facilitating the clinical decision-making process for HCC in the forthcoming times.

Materials and methods

Acquisition of datasets and ICD-related genes

In this study, we selected and analyzed multiple types of data from public libraries to investigate the potential role of ICD in HCC. Transcriptomic data in the form of FPKM values, somatic mutation data and clinicopathologic spectrum of 371 HCC patients and 50 normal tissue samples were obtained from The Cancer Genome Atlas (TCGA) database portal (https://portal.gdc.cancer.gov/). In addition, we also downloaded the GSE76427 dataset from the GEO database (https://www.nc-bi.nlm.nih.gov/geo/), which included 115 primary tumor tissue samples and 52 adjacent non-tumor tissue samples. To facilitate further analyses, all data having complete clinical information were transformed into log2 form. Besides, we utilized the GEPIA database (http://gepia.cancer-pku.cn/) consisting of HCC samples. To identify genes related to ICD, we referred to a previous study and identified 33 ICD-related genes for subsequent analysis(Garg et al. 2016).

Consensus clustering analysis

Using the "Consensus Cluster Plus" R package, we performed an unsupervised consensus clustering analysis to stratify patients into distinct subtypes based on ICD-related gene expressions. From 2 to 10, the optimal number of clusters (k) was determined, and the procedure was repeated 1,000 times to assure the stability of the results. The resulting clusters were characterized by strong intra-organizational links and weak inter-generational connections. With the assistance of “limma” and “ggplot2” packages, we performed PCA analysis to verify the validity of our prognostic model. We also examined the relationship between our model and ICD-related genes using the “heat map” package. Furthermore, Kaplan–Meier (K–M) survival analysis was performed to distinguish the differences in clinical outcomes between different clusters, utilizing the “survival” and “survminer” R packages.

Identification of DEGs and functional enrichment analysis

Differential expression analysis was conducted to identify genes that were significantly differentially expressed among the distinct ICD-related clusters. The “limma” and “Venn Diagram” R packages were utilized with a predefined threshold of |fold change|> 2 and an adjusted p-value filter < 0.05. To further explore the functional implications and biological effects associated with different ICD patterns, gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analyses were performed via the “cluster Profiler” R package. GO analysis results were categorized into three main domains, including biological process (BP), cellular component (CC) and molecular function (MF), and were screened with a p-value threshold of < 0.05.

WGCNA analysis and Hub genes identification

The transcriptomic profiling data and corresponding clinical information were incorporated into this study to construct a co-expression network using the “WGCNA” R package. This is a widely applied method that converts the gene expression matrix into unsigned co-expression networks, thereby revealing critical interacted gene modules and hub genes. Firstly, a gene expression similarity matrix was established by computing the Pearson correlation coefficient between any two genes. Next, soft-thresholding power β was increased based on the scale-free topology network criterion. The adjacency matrix was then clustered using topological overlap measure (TOM) and dissimilarity (1-TOM) between genes, and a dynamic tree-cut algorithm was applied to identify gene modules. Modules were considered valid if they contained at least 30 genes, and similar modules were merged at the cutoff value of 0.25. The relationships between modules and traits were determined, including survival, age, gender, grade, stage, T stage and ICD cluster (A = 0, B = 1), by constructing a dynamic pruning tree to divide genes from TCGA-LIHC and GSE76421 databases into different modules.

Construction and validation of the prognosis ICD score model

We performed the univariate Cox analyses to derive prognostic mRNAs and co-expressed lncRNAs with a p < 0.001 to establish the prognostic ICD score model by the LASSO Cox regression analysis based on the “glmnet” R package. Ultimately, we kept 10 genes and their coefficients by selecting the optimal penalty parameter (λ value) with the minimum criteria. The ICD score was calculated as follows: ICD score = ∑10i(EXPi*Coefi) (EXP: gene expression level, Coef: risk coefficients). We carried out the K–M survival analysis and principal component analysis (PCA) in the high- and low-ICD-score groups distinguished by the median value. Additionally, the Alluvial diagram was employed to evaluate the relation among ICD clusters, ICD scores and statuses. Meanwhile, the patients were randomly classified into training and testing groups with a proportion of 1:1. K–M survival analyses and receiver operating characteristic curves (ROC) were performed. To further certify the predictive ability of the model, survival distribution analyses were employed in training and testing groups, respectively.

ESTIMATE analysis and CIBERSORT analysis

To evaluate the tumor microenvironment of HCC patients, two distinct analytical methods were employed: ESTIMATE analysis and CIBERSORT analysis. The ESTIMATE analysis was performed using the R package “estimate,” which utilizes gene expression signatures to estimate the proportion of immune and mesenchymal cells in tumor samples. The ESTIMATE score, which is calculated based on the immune and stromal scores, reflects the extent of immune infiltration in the tumor microenvironment. Additionally, the tumor purity was calculated as one minus the ESTIMATE score, representing the abundance of tumor cells in tumor tissue. On the other hand, CIBERSORT analysis was conducted to estimate the abundance of 22 immune cell types in each HCC sample. The percentage of each of the 22 immune cell types was represented on the ordinate of the results. Together, these two analytical methods provided a comprehensive evaluation of the immune landscape of HCC samples, which was used to identify potential relationships between immune cell infiltration and patient outcomes.

Immunotherapy prediction for ICD score

In this study, we aimed to evaluate the potential immunotherapy response between the high- and low-ICD-score groups. To achieve this, we utilized the Immune Portrait System (IPS) data from The Cancer Immunome Atlas (TCIA) (https://tcia.at/home) and applied the Submap algorithm (http://cloud.genepattern.org/gp) to predict the reaction of the groups to anti-PD-1 and anti-CTLA-4 therapy. Additionally, we analyzed differential gene PD-L1 expression (https://www.ncbi.nlm.nih.gov/gene) using the “limma” R package. Furthermore, we used the Tumor Immune Dysfunction and Exclusion (TIDE) analysis (http://tide.dfci.harvard.edu/) to compare the high- and low-ICD-score groups. We also investigated differential analyses in the aspects of HLA-related genes and immune checkpoints. To determine the correlation between the ICD score and immune cells, we conducted a Spearman correlation analysis. The significance level was set at p < 0.05.

Establishment and validation of a nomogram scoring system

We performed univariate and multivariate Cox regression of clinical indicators including age, gender, grade, TNM stage and ICD score to determine independent prognostic factors with a p value < 0.05 in the testing and the training groups. According to the results of multivariate analyses, the significant factors consisting of ICD score and TNM stage were subsequently used as input to develop a predictive nomogram by utilizing “rms” and “regplot” R packages. The nomogram scoring system could forecast HCC patients’ 1-, 3- and 5-year OS, quantitatively. The predictive accuracy of the nomogram was evaluated by c-index. Calibration curves were employed to depict consistency between the predicted survival events and the actual observations.

Statistical analysis

The statistical analyses and data analysis were performed using R 4.1.2. The Wilcox test was used to divide the groups into high and low ICD scores based on immune function and tumor-infiltrating immune cells. Univariate survival analysis was conducted using the K–M method and the log-rank test. Multivariate analysis was performed using the Cox regression model. Statistical significance was considered when p < 0.05 (* p < 0.05, ** p < 0.01, *** p < 0.001).

Results

Establishment of an ICD-related subtype

This study proceeded in accordance with the flowchart depicted in Fig. S1. The objective of this study was to identify molecular subtypes of HCC related to the ICD and to examine the clinical implications of these distinct clusters. In order to achieve this, an initial analysis was conducted on the expression of 33 genes related to ICD, as presented in Table S1, within HCC tissues in comparison to non-malignant samples. The study revealed that the expressions of 10 genes, namely IL1R1, CD4, PRF1, NLRP3, IL10, TLR4, IL6, MYD88, IL1B and NT5E, were down-regulated in tumor samples. Conversely, the expression of 13 genes, including IL17RA, PIK3CA, CXCR3, ENTPD1, ATG5, EIF2AK3, LY96, HMGB1, HSP90AA1, BAX, PDIA3, CALR and IFNGR1, was found to be elevated in HCC tissues (Fig. 1a).

Fig. 1.

Identification of ICD-related subgroups in HCC through consensus clustering analysis. a The boxplot displayed the differential expression of 33 ICD-related genes between HCC and non-cancerous specimens, with red representing tumor samples and blue representing normal samples. b On the basis of the consensus clustering analysis, we defined two distinct ICD-associated clusters (k = 2). c Then we conducted the principal component analysis of ICD-related gene expression profiles, which distinguished two ICD clusters: A (blue) and B (yellow). d This heatmap presented the correlation between the expression of ICD-related genes and the ICD clusters. e Utilizing the Kaplan–Meier analysis, it was identified that the HCC patients belonged to cluster B owned disadvantage in OS as compared to cluster A

Subsequently, a clustering of ICD genes was established based on the expression levels of the mentioned candidates. Survival analysis was then performed utilizing the TCGA and GEO databases. The findings indicated that the optimal number of clusters was determined to be two through consensus clustering analysis. Furthermore, the results of the principal component analysis (PCA) revealed that these two clusters were effective in distinguishing between different patients, as illustrated in Fig. 1b–c. Additionally, it was observed that patients categorized in cluster B exhibited raised levels of the majority of ICD-associated genes as compared to those in cluster A (Fig. 1d). Moreover, patients belonging to cluster B possessed notably inferior overall survival rates in contrast to those in cluster A, with a p-value of 0.020 (Fig. 1e). Furthermore, a comparative analysis was conducted to assess notable variations in survival, age, gender, pathologic grade and TNM stage among the two clusters (Fig. 2a).

Fig. 2.

Clinicopathologic characteristics and immune infiltration between different clusters. a Here, we validated the relationship between the clinicopathologic features and ICD-related clusters in HCC cohort. b This network diagram demonstrated the influence of ICD regulatory genes on the survival of HCC patients. c Employing the ssGSEA algorithm, we examined the infiltration levels of different immune cells in ICD cluster A/B. d Then we screened out the DEGs between cluster A and cluster B. Then the GO and KEGG enrichment analyses were performed based on these DEGs. The statistical significance of the findings is denoted by p-values

Furthermore, a univariate COX regression analysis was performed on the 33 ICD-related genes. The results indicated that 23 of these genes were significantly associated with the prognosis of HCC (Fig. S2). A network diagram was created to analyze the interrelationship between the 32 ICD-related genes (excluding IL17A) and their impact on the prognoses of HCC patients (Fig. 2b). The infiltration degree of 23 immune cells in the two subtypes was estimated through ssGSEA analysis to establish the correlation between ICD-related genes and the TME. It was observed that the mRNA levels of 21 immune cells exhibited a higher value in cluster B as compared to cluster A (Fig. 2c).

DEGs identification and functional enrichment analysis

Subsequently, a differential analysis was performed on the two clusters, and a total of 30 genes were clarified to be differentially expressed. Then we conducted GO enrichment analyses to investigate the possible biological roles of the differential expressed genes (DEGs). The results indicated that these DEGs were mainly associated with biological processes such as positive regulation of cytokine production, cellular response to biotic stimulus and interleukin-1 production. Besides, these DEGs exhibited enrichment in the cellular component aspect, specifically in the external side of the plasma membrane, endocytic vesicle and plasma membrane signaling receptor complex. Furthermore, the DEGs exhibited significant enrichment in molecular functions such as cytokine receptor binding, cytokine activity and growth factor receptor binding. The DEGs were found to be significantly enriched in the NOD-like receptor signaling pathway, Toll-like receptor signaling pathway, C-type lectin receptor signaling pathway and T cell receptor signaling pathway as visualized by the KEGG analysis (Fig. 2d).

Structure of an ICD scoring system via WGCNA

We obtained the transcriptional profiles of HCC from TCGA and GEO databases. WGCNA algorithm was conducted to identify co-expression modules based on these RNA-seq profiles. Soft-thresholding was established using the soft connectivity function of WGCNA, with a power of β = 5 chosen as the soft-thresholding value based on the scale-free topology fit index, which was up to 0.8 (Fig. 3a–b). Then we chose an average linkage clustering to identify distinct clusters based on topological overlap, resulting in the identification of five co-expression modules using the dynamic tree-cut algorithm (Fig. 3c). To explore the relationship between the identified co-expression modules and clinical traits, including survival, age, gender, grade, stage and T stage, we integrated the ICD cluster defined A = 0 and B = 1 established previously into the co-expression modules. We found that the MEyellow module was negatively correlated with the HCC ICD cluster (r =  – 0.14, p = 0.005), while the MEblue (r = 0.61, p = 6e-39), MEbrown (r = 0.53, p = 1e-27), MEturquoise (r = 0.62, p = 6e-41) and MEgrey (r = 0.17, p = 0.001) modules had a positive correlation with the HCC ICD cluster (Fig. 3d). Subsequently, we extracted genes from the established modules and found that 28 immune-related genes were associated with a high risk for HCC prognosis (Fig. 3e, p < 0.001). Additionally, we also verified 20 long noncoding RNAs (lncRNAs) that were co-expressed with immune-related genes and showed a correlation with the prognosis of HCC patients (Fig. 3f, p < 0.001).

Fig. 3.

Construction of an ICD scoring system in HCC. In order to establish the ICD scoring system, we carried out scale-free topology analysis (a–b). The clustering dendrograms were used to cluster genes with similar expression patterns into co-expression modules (c), and the heatmap showed the correlation between module eigengenes and clinical traits of HCC (d). Uni-Cox analysis was used to analyze the prognosis of genes extracted from WGCNA and lncRNA genes in HCC patients (e–f). The cvfit and lambda curves of 10-time cross-validation were used for tuning parameter selection by LASSO regression (g–h). The heatmap revealed the expression patterns of 10 ICD genes with the ICD score (i), and the ICD score and 10 signature genes were positively correlated with literature that reported ICD markers (j). Statistical significance is indicated by *p < 0.05, **p < 0.01, ***p < 0.001

Then we employed the prognostic immune genes and co-expressed lncRNAs to construct a prognostic ICD score signature through LASSO Cox regression analysis. The cvfit and lambda curves are shown in Fig. 3g and h, respectively. The ICD score was calculated for each sample based on five hub mRNA and lncRNA expressions. Moreover, all enrolled HCC cases were randomly divided into training and validation phases. In these two cohorts, we revealed that the OS time got shorter followed by the rise of ICD score in HCC patients. Moreover, when focusing on every single gene, the OS was significantly lower in the high-expression group than in the low-expression group (Fig. S3). Furthermore, the study found that the ICD score and the expression levels of the 10 genes were highly and positively linked with a cluster of the collected ICD markers (Fig. 3i and j). Overall, the results suggest that the established ICD score has prognostic significance for HCC patients and that the 10 genes included in the ICD score may serve as potential therapeutic targets for HCC treatment.

Validation of the accuracy and stability of the ICD scoring system

In order to evaluate the reliability and consistency of our model, a PCA analysis was conducted (Fig. S4a). Furthermore, the significant differences in overall survival between the high- and low-ICD-score groups (p < 0.001) indicated the effectiveness of our model (Fig. S4b). An alluvial diagram was also created to visually demonstrate the relationship among the ICD clusters, ICD score and living status (Fig. S4c). The analysis showed that ICD cluster A had a higher ICD score than ICD cluster B, and that the high-ICD-score group had a higher survival rate, consistent with the results in Fig. S4e (p = 4.3e-07). These findings suggest that the high-ICD-score group is associated with more favorable clinical features and better survival outcomes compared to the low-ICD-score group. Moreover, our results also demonstrated a significant correlation between the ICD score and T and B cells, indicating that the ICD score may impact immune cells and subsequently influence the development of HCC (Fig. S4d).

We examined the stability and predictability of the ICD scoring system in the training and validation phases. Firstly, we analyzed expressions of 10 prognostic genes derived from the LASSO COX analysis and observed that they expressed more highly in the high-ICD-score group (Fig. 4a). Secondly, the K–M survival curve revealed that the OS rate of HCC patients in the low-ICD-score group was significantly higher than those in the high-ICD-score group (Fig. 4b). Thirdly, based on the ROC curve, the AUCs of the training and validation groups were 0.771 and 0.841, respectively, which represented our model has good predictive efficacy (Fig. 4c). In addition, from the visualized heatmap results, we concluded the expression differences of 10 prognostic genes in low- and high-ICD-score groups. In the validation group, the expression of the 10 prognostic genes appeared to be increased in the high-ICD-score group than the low-ICD-score group, as well as in the training group. The distribution curve of the ICD score showed that with the increase in the ICD score, survival times shortened significantly. The scatter dot plot showed that the survival time of HCC patients was inversely correlated with the ICD score (Fig. 4d).

Fig. 4.

Validation of the ICD score model in terms of the prognosis of HCC patients. Panel (a) displays the differential expression analysis of the 10 prognostic genes used to establish the ICD score in high- and low-ICD-score groups. Panel (b) depicts the Kaplan–Meier analysis of the overall survival (OS) between the two groups. Panel (c) shows the ROC curves of the established model with AUC values of 0.771 and 0.847 for the training and testing groups, respectively. Panel (d) presents the ICD score distribution, survival status of each patient and heatmaps of prognostic 10 gene signature. Overall, these results demonstrate the effectiveness of the ICD score model in predicting the prognosis of HCC patients

The clinical correlation analysis and the immune landscape of ICD score

In this study, we analyzed the correlations between the ICD score and various clinicopathologic spectra (including survival status, age, gender, pathologic grade and TNM stage) in patients with HCC. The survival status, pathologic grade and TNM stage (especially T stage) showed differential distributions in the low-ICD-score and high-ICD-score groups (Fig. 5a–b, p < 0.05). Interestingly, we also verified that the patients who belonged to the high-ICD-score subgroup owned worse OS than those of the low-ICD-score subgroup.

Fig. 5.

Correlations between ICD score and different clinicopathologic characteristics. a The proportions of subsets divided by survival status, age, gender, grade, TMN stage, T stage, M stage and N stage are compared in different ICD score groups. b Then these boxplots revealed the correlation of various clinicopathologic characteristics and ICD score in HCC patients. These analyses provide insight into the relationship between ICD score and clinical features of HCC patients, which may help to identify potential biomarkers and therapeutic targets for personalized treatment

Using the ESTIMATE methodology, we also examined the ability of the immune-related prognostic signature to predict the TIME. As presented in Fig. 6a, our results indicated that the low-ICD-score group was associated with a higher stromal score (p = 1.5e-07) and estimate score (p = 0.011). The CIBERSORT algorithm was used to investigate the immune infiltration of patients in both ICD score groups. We found that the abundance of 22 types of immune cells differed significantly between the two groups, with CD4 memory resting T cells and monocytes being more enriched in the low-ICD-score group (Fig. 6b–c).

Fig. 6.

Immune landscape of ICD score. a Using estimate algorithm, we calculated the immune scores, stromal scores and estimate scores in low-/high-ICD-score groups. b–c Besides, the enrichments of 22 kinds of immune cells in different ICD score subtypes were accessed upon the CIBERSORT analysis. d–e Through Kaplan–Meier analysis, we identified that high levels of CD8 T cells and plasma cells are associated with better overall survival. f–j These results indicated the correlations between the risk score and immune cell infiltration

We then estimated the relationship between immune cell infiltration and patient survival. Higher infiltration of CD8 T cells (p = 0.007) and plasma cells (p = 0.027) was associated with a better prognosis (Fig. 6d–e). Lastly, we conducted a Spearman correlation analysis to estimate the relationship between immune cell infiltration and ICD score. We found that macrophages M0 and T cells follicular helper were positively related to the ICD score, while B-cell naïve, T-cell CD4 memory resting and monocytes were negatively related (Fig. 6f–j, p < 0.05). Totally, we considered that the ICD score was associated with the infiltrations of various immune cells in HCC, suggesting ICD score was a potential statistical index to evaluate the TME construction in malignancies.

ICD score predicts immune therapeutic responses in HCC

Here, we compared the efficacy of immune therapy in the different ICD score clusters based on IPS scores. Treatment with anti-CTLA4 and anti-PD-1, either in monotherapy or combination therapy, was more likely to affect HCC patients in the high-ICD-score group (Fig. 7a). This finding suggested that HCC patients with a high ICD score could have a better response to immune therapy. Furthermore, HCC patients with a high ICD score possessed a higher mRNA level of CD274, indicating that anti-PD-1/PD-L1 immune therapy could be an effective treatment option for these patients (Fig. 7b, p = 0.00011).

Fig. 7.

Prediction of the therapeutic response of the ICD score model. a IPS comparison of the high- and low-ICD-score groups. b It was found out that PD-L1 was up-regulated in high-ICD-score group. c Boxplots showed significantly reduced ICD scores in the immune response group compared with the non-response group in HCC cohort. d The patients in L-TMB presented better OS than those of H-TMB group. e Kaplan–Meier curves for HCC patients stratified by both TMB and ICD score. f Different expression of HLA-related genes in different groups. g Different expression of checkpoint-related genes in different groups

To assess the capability of the established ICD scoring system to predict immune therapy response, we carried out a TIDE analysis. Our analysis revealed that the ICD score was significantly raised in the non-responsive group than in the responsive group (Fig. 7c, p = 0.016), indicating that immune therapy would potentially be more beneficial for HCC patients with a low ICD score.

Tumor mutational burden (TMB) is a useful biomarker for identifying patients who may benefit from immune therapy, particularly in terms of drug response and prognosis prediction (Fumet et al. 2020; Yarchoan et al. 2017). Our survival analysis based on the median TMB values divided into high- and low-TMB groups showed that the high-TMB group had a worse survival outcome than the low-TMB group (Fig. 7d, p < 0.001). We also combined the TMB and ICD score groups into four subgroups for a survival analysis, which revealed that the low-TMB and low-ICD-score group had the best prognosis (Fig. 7e, p < 0.001).

Moreover, we observed that the expressions of HLA-related genes, except for HLA-B, HLA-DRB5, HLA-E, HLA-F and HLA-G, were different between the ICD score groups. The high-ICD-score group showed increased expression of HLA-related genes (Fig. 7f). Additionally, our differential analysis of immune checkpoint-associated genes showed that LAG3, CD274, HAVCR2, CTLA4, TIGIT and PDCD1 were expressed more highly in the high-ICD-score group than in the other group, indicating the potential significance of immune therapy for HCC patients in the high-ICD-score group (Fig. 7g). Overall, these results suggest that the ICD score is a valuable predictor of immune therapy response in HCC patients and could be used to guide treatment decisions.

Development of a nomogram to predict survival

Survival analysis indicated that HCC patients with different ICD scores had distinct prognoses, as shown in Fig. S4b. To further estimate whether the predictive signature was an independent prognostic factor for HCC patients, we performed univariate and multivariate Cox analyses with ICD score and clinical variables including age, gender, grade and stage in both testing and training groups. In the testing group, stage (p = 0.002) and ICD score (p < 0.001) were significantly associated with patient prognosis based on the results of the univariate Cox analysis. The same result was observed in the training group, with a p-value < 0.001 (Fig. 8a). The multivariate Cox regression analysis revealed that both stage and ICD scores were independent determinants of OS in HCC patients in both testing and training groups (Fig. 8b, p < 0.05). We constructed a nomogram integrating stage and ICD score to predict the probability of prognostic survival at 1, 3 and 5 years of HCC patients, in order to inform clinical decision-making (Fig. 8c). The C-index of the nomogram was 0.703, indicating good discrimination ability (Fig. 8d). The calibration curves showed that the 1-, 3- and 5-year OS predictions by the nomogram were highly consistent with actual observations (Fig. 8e).

Fig. 8.

Construction of a nomogram predicting OS based on the ICDscore system in HCC. a–b Univariate and Multivariate cox analyses of clinical characteristics and ICD score in the training and testing phase. c Nomogram integrated the ICD score and TNM stage for predicting the 1-,3- and 5-year OS in HCC patients. d The c-index of the established nomogram. e Calibration curves of the nomogram for predicting of 1-,3- and 5-year OS

PPI network and exploration of the downstream genes regulated by ICD

In this study, we utilized the Search Tool for the Retrieval of Interaction Genes (STRING) to construct a protein–protein interaction (PPI) network based on the 29 immune genes associated with prognosis (Fig. 3e). The resulting network was visualized using Cytoscape software, and the top ten genes were screened out using the degree method. These ten genes included HSP90AA1, PSMD14, PSMD2, PSMD6, ACTG1, PSMD1, PSMD11, PSME3, PSMD13 and EED. We then performed differential analyses of these ten genes between HCC samples and normal individuals obtained from the GEPIA database, which revealed that PSMD2 and PSMD14 were significantly up-regulated in HCC patients (Fig. S5).

To investigate the impact of PSMD2 and PSMD14 on patient survival, we conducted K–M analyses in HCC patients with high and low expressions of PSMD2 and PSMD14. Our results showed that patients with low expressions of PSMD2 and PSMD14 had significantly better overall survival (OS) rates with p < 0.05 (Fig .S6). Moreover, we found that PSMD2 and PSMD14 were positively correlated with several ICD-related genes, including CALR, CASP1, CASP8, HMGB2 and HSP90AA1, indicating that they were downstream genes regulated by ICD (Fig. S7). These findings suggest that PSMD2 and PSMD14 may play a crucial role in the pathogenesis of HCC and could serve as potential prognostic biomarkers and therapeutic targets.

Discussion

The role of immunotherapy which activates the body's immune system and kill tumor cells effectually for a long time while harboring fewer side effects has been extensively studied(Alzeibak et al. 2021; Tan et al. 2020). A low immunogenicity of the tumor itself as well as an immune-suppressive microenvironment resulting in escaping from immune surveillance and disease progression attenuates the effect of immunotherapy(Seliger and Massa 2021). The previous study reveals that the release of danger signals by tumor cell death improves immunogenicity and induces ICD(Galluzzi et al. 2017). ICD refers to a form of cell death triggered by certain anticancer treatments that leads to the release of endogenous danger signals, such as ATP, high-mobility group box 1 protein (HMGB1) and calreticulin. These danger signals, in turn, activate the immune system and induce an anti-tumor immune response(Guo et al. 2021; Galluzzi et al. 2020). ICD is now recognized as a crucial mechanism for the success of various cancer therapies, including chemotherapy, radiotherapy and photodynamic therapy(Sen Santara et al. 2023; Liu et al. 2023; Jung et al. 2023). Consequently, comprehending the mechanisms and clinical ramifications of ICD is imperative for the advancement of innovative cancer therapies.

In this study, we aimed to establish an ICD-related subtype in HCC and investigate the clinical significance of these distinct clusters. We first analyzed the expression of 33 ICD-related genes in HCC tissues compared to non-cancerous tissues and verified that about 10 genes were down-regulated in tumor samples, while the expression of 13 genes was elevated in HCC tissues. Interestingly, the up-regulated genes included cytokines such as IL17RA, growth factor receptors such as PIK3CA and stress response genes such as HSP90AA1. According to the expression patterns of these ICD regulatory genes, we established two clusters (cluster A/B) using consensus clustering and PCA analysis. We observed that patients in cluster B possessed higher expressions of most ICD-associated genes compared to cluster A, and patients in cluster B had significantly worse overall survival than those in cluster A. These results suggest that the expression of ICD-related genes is associated with the prognosis of HCC patients. Moreover, we found that the infiltration degree of immune cells was higher in cluster B than in those cluster A. This finding is consistent with previous studies, showing that immune infiltration is associated with HCC prognosis(Yang et al. 2021; Liu et al. 2020). We also conducted a differential analysis between the two clusters, resulting in 30 DEGs. These genes were primarily involved in cytokine receptor binding, cytokine activity and growth factor receptor binding in the molecular function aspect. The KEGG analysis revealed that the DEGs were predominantly enriched in the NOD-like receptor signaling pathway, Toll-like receptor signaling pathway, C-type lectin receptor signaling pathway and T cell receptor signaling pathway. These results suggest that these pathways play an important role in the ICD-associated subtype of HCC and may be potential targets for therapy.

We also established an ICD scoring system using the prognostic immune genes and co-expressed lncRNA to predict the prognosis of HCC patients. This scoring system showed good predictive ability in both the training and validation cohort. Patients with different ICD scores displayed heterogeneous clinicopathologic characteristics, indicating that the low-score group had a better prognosis, a lower-grade classification and a less advanced TNM stage, especially the T stage. A significantly different immune landscape was observed in the two score groups. In the low-ICD-score group, the proportion of activated immune cell infiltration consisting of CD4 memory resting T cells and monocytes was up-regulated. This result resembles the former studies, indicating that CD4 + memory resting T cells and CD8 + T cells which significantly increased in HCC were protective factors for HCC(Viveiros et al. 2019). HCC patients who had a high percentage of cytotoxic CD4 + T cells harbored a strong prognosis in their OS(Fu et al. 2013). As a key component of the HCC matrix, the mononuclear macrophage system integrates the proliferation pressure and metabolic adaptation of the tumor microenvironment, mediates the response to induce tumor-promoting inflammation and coordinates immune status. It is becoming an important target for collaborative tumor immunotherapy(Netea-Maier et al. 2018).

The combination of oxaliplatin with immune checkpoint inhibitors (ICIs) like anti-PD-1 had a synergistic effect in inhibiting tumor growth and in inducing DCs maturation and CD8 + T-cells activation by inducing ICD(Chiaravalli et al. 2022; Zhu et al. 2020). In our study, we can see that the high-ICD-score HCC patients who had higher expressions of checkpoint-related genes consisting of CD274, CTLA4, PDCD1 and so on, showed more sensitivity to treatment with anti-CTLA4 and anti-PD-1, either in monotherapy or in combination therapy. Currently, the most commonly used staging system for HCC is the Barcelona Clinic Liver Cancer (BCLC) staging system, which incorporates variables such as tumor size, number of nodules, vascular invasion and liver function(Llovet et al. 2018). However, the BCLC staging system does not take into account the immune microenvironment or the expression of immune-related genes. Our ICD scoring system, which is based on the expression of immune-related genes, could be used in conjunction with the BCLC staging system to improve patient stratification and treatment decisions. An ICD scoring system also had a good performance in assessing the prognosis of patients with gastric cancer and predicting treatment response(Gan et al. 2023). Additionally, we explored downstream genes regulated by ICD, suggesting that PSMD2 and PSMD14 are potential targets for HCC. PSMD2 regulated by ICD, could promote the proliferation of HepG2 cells by facilitating cellular lipid droplet accumulation(Tan et al. 2019). Overexpression of PSMD14 correlated with vascular invasion, tumor number, tumor recurrence and poor tumor-free and overall survival of patients with HCC(Lv et al. 2020).

However, our study still exhibited various constraints that warrant recognition. First, our study was retrospective in nature, and further prospective studies are needed to validate our findings. Second, this investigation only included HCC patients from publicly available datasets, and our findings should be confirmed in larger cohorts and diverse populations. Third, more HCC immunotherapy data need to be excavated to evaluate the effects of immunotherapy estimated by our ICD scoring system. Fourth, we focused on ICD-related genes and did not investigate other aspects of the immune microenvironment, such as T cell infiltration or the expression of immune checkpoints. Further studies are needed to investigate the interactions between these factors and the expression of ICD-related genes in HCC.

Ultimately, our research has successfully identified the distinct clusters of HCC which were associated with the expression of immune-related genes in relation to ICD. We showed that the expression of ICD-related genes is associated with patient prognosis and immune cell infiltration. We also developed an ICD scoring system that could be used to predict patient prognosis. Our findings provide new insights into the role of ICD in HCC and have important clinical implications for patient stratification and treatment decisions. Further studies are needed to validate our findings and investigate the interactions between ICD-related genes and other aspects of the immune microenvironment in HCC.

Conclusion

In summary, we identified two molecular subtypes based on the expression of ICD-associated genes and established an ICD core signature with a good performance in assessing the prognosis of patients with HCC, evaluating the immune landscape and predicting immunotherapy response. A nomogram integrating the ICD score revealed a strong prognostic power for HCC patients, which is worthy of clinical promotion. Additionally, we observed PSMD2 and PAMD14 were latent targets for HCC. Consequently, the ICD score model could be applied as a novel reference to guide treatment and predict prognosis in HCC patients.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

JW, DYC and GMX designed/planned the study. GMX, YFJ, YL, JZG and XFX acquired and analyzed data, performed computational modeling. GMX, YFJ, DYC and JW wrote and revised the manuscript. DYC and JW supervised the study. All authors participated in imaging analysis and discussion of related data and approved the submitted version.

Funding

This work was supported by the Science Technology Department of Zhejiang Province (2023C03063), Huadong Medicine Joint Funds of the Zhejiang Provincial Natural Science Foundation of China (LHDMD22H310005), the Health Commission of Zhejiang Province (JBZX-202004 and 2023RC013) and National Natural Science Foundation of China Grant (NO. 82073144 and NO. 82202974).

Data availability

The data used to support the findings of this study are available from the public databases, including TCGA database (https://gdcportal.nci.nih.gov/) and GEO database (http://www.ncbi.nlm.nih.gov/geo/, containing datasets of GSE76427).

Declarations

Conflict of interest

The authors declare no conflict of interest.