A novel risk score system based on immune subtypes for identifying optimal mRNA vaccination population in hepatocellular carcinoma

Abstract

Purpose

Although mRNA vaccines have shown certain clinical benefits in multiple malignancies, their therapeutic efficacies against hepatocellular carcinoma (HCC) remains uncertain. This study focused on establishing a novel risk score system based on immune subtypes so as to identify optimal HCC mRNA vaccination population.

Methods

GEPIA, cBioPortal and TIMER databases were utilized to identify candidate genes for mRNA vaccination in HCC. Subsequently, immune subtypes were constructed based on the candidate genes. According to the differential expressed genes among various immune subtypes, a risk score system was established using machine learning algorithm. Besides, multi-color immunofluorescence of tumor tissues from 72 HCC patients were applied to validate the feasibility and efficiency of the risk score system.

Results

Twelve overexpressed and mutated genes associated with poor survival and APCs infiltration were identified as potential candidate targets for mRNA vaccination. Three immune subtypes (e.g. IS1, IS2 and IS3) with distinct clinicopathological and molecular profiles were constructed according to the 12 candidate genes. Based on the immune subtype, a risk score system was developed, and according to the risk score from low to high, HCC patients were classified into four subgroups on average (e.g. RS1, RS2, RS3 and RS4). RS4 mainly overlapped with IS3, RS1 with IS2, and RS2+RS3 with IS1. ROC analysis also suggested the significant capacity of the risk score to distinguish between the three immune subtypes. Higher risk score exhibited robustly predictive ability for worse survival, which was further independently proved by multi-color immunofluorescence of HCC samples. Notably, RS4 tumors exhibited an increased immunosuppressive phenotype, higher expression of the twelve potential candidate targets and increased genome altered fraction, and therefore might benefit more from vaccination.

Conclusions

This novel risk score system based on immune subtypes enabled the identification of RS4 tumor that, due to its highly immunosuppressive microenvironment, may benefit from HCC mRNA vaccination.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13402-024-00921-1.

Introduction

Hepatocellular carcinoma (HCC), as one of the most common and lethal tumors, ranks the third leading cause of tumor-related death around the world, with a 5-year survival rate of only ~20% [1, 2]. Enhancing anti-tumor immunity using immune checkpoint blockades (ICBs), such as anti-PD-1/PD-L1 or anti-CTLA-4 antibodies, has been proven to prolong the survival of HCC patients in advanced stage, but therapeutic responses are observed only in a limited number of HCC patients [3–6]. Various strategies have been actively explored to complement ICBs therapy in HCC, mainly focused on the combination of ICBs and other existing therapies (e.g. anti-VEGF therapy, chemotherapy, and radiotherapy) [7–11]. Such combinations have been verified to enhance anti-tumor effects in vivo and improve the survival of HCC patients in clinical trials [12]. But an increasing number of patients were found non-responsive, and therefore, novel combinatorial therapeutic strategies are urgently needed.

Genetic or epigenetic alterations in tumorigenesis could produce tumor antigens, thereby inducing adaptive immune responses against tumor cells [13–15]. Tumor vaccination shows promise as a therapeutic approach to complement ICBs in malignancies, with the advantages of broad therapeutic window, minimal non-specific effect and induction of sustaining immune memory [16, 17]. Recently, the use of synthetic mRNA vaccine has again attracted the attention of oncologist and scientists under the background of COVID-19 pandemic, especially as exemplified by the recent clinical application of COVID-19 mRNA vaccine and the clinical trials of mRNA vaccine therapies for a variety of disease, including malignancies [18–23]. Generally speaking, an mRNA vaccine for tumor could deliver a specific transcript encoding one or more tumor-associated antigens or tumor-specific antigens to the cytoplasm of the host cell (especially in APCs), and enable the tumor antigens production, by which APCs increase the antigenic activity of T cells to enhance anti-tumor immune response [24, 25]. As an attractive alternative to DNA vaccines, mRNA vaccines require only cytosolic delivery for transient expression of encoded proteins, therefore largely avoiding host genome insertional mutation and inducing more predictable and more effective protein expression [26]. Numbers of mRNA vaccine based clinical trials demonstrated that mRNA vaccination could induce antigen-specific immunity and improve the clinical outcomes of patients with non-small cell lung cancer, glioblastomas, as well as prostate cancer [27–30]. However, the development of mRNA vaccine against HCC antigens is somehow lagged, and no HCC patient population suitable for mRNA vaccine therapies has been identified.

The present work aims to explore novel HCC antigens for mRNA vaccine development, and map HCC immune landscape to identify suitable populations for mRNA vaccine therapy. Twelve candidate genes associated with unfavorable prognosis and antigen presenting cell (APC) infiltration was identified from the pool of overexpressed and mutated genes in HCC. Three immune subtypes with distinct clinical and molecular characteristics were subsequently identified based on the profiles of the twelve HCC antigens. Finally, a risk score system based on the immune subtypes were introduced to stratify HCC patients into various risk subgroups, and proved to be an analogy to the immune subtypes. Each risk subgroup corresponded to distinct clinicopathological, cellular and molecular characteristics. Our studies elucidated a distinct tumor immune microenvironment in each patients with HCC, and preliminarily provided the theoretical basis for the development of HCC mRNA vaccines and identification of suitable populations for vaccination. We present the following article in accordance with the TRIPOD reporting checklist.

Materials and methods

HCC data extraction and preprocessing

Publicly available HCC RNA-seq data and corresponding clinical follow-up information of patients were obtained from the International Cancer Genome Consortium (ICGC, https://icgc.org/), The Cancer Genome Atlas (TCGA, https://cancergenome.nih.gov/), and the Gene Expression Omnibus (GEO, https://www.ncbi.nlm.nih.gov/geo/) databases. Of the 371 HCC cases in the TCGA HCC cohort, 343 were cases with overall survival (OS) longer than one month, and were selected as training cohort for risk score system development. And external validations were conducted using RNA-seq from the ICGC HCC cohort (validation cohort 1; n = 229) and GSE14520 cohort (validation cohort 2; n = 242). All datasets are freely available as public resources. Consequently, local ethics approval was not required.

Identification of candidate targets for mRNA vaccination in HCC

First, using ANOVA test, the significantly overexpressed genes of HCC were obtained using Gene Expression Profiling Interactive analysis (GEPIA, http://gepia2.cancer-pku.cn) under the threshold of log2FC > 1 and FDR < 0.05, which possibly encoded tumor antigens in HCC [31]. Then, the genome alteration status of HCC in the TCGA cohort was also investigated using the cBioPortal database (http://www.cbioportal.org). And the mutated genes of HCC were subsequently intersected with the aforementioned overexpressed genes. And the overlapping genes were likely encode tumor-specific antigens. And univariate Cox regression analysis was conducted to investigate the prognostic values of these overexpressed and mutated genes (P value < 0.05). Given the crucial roles of APCs (e.g. B cell, macrophage and dendritic cell) in the anti-tumor immunity of mRNA vaccine, we further investigated the relationship between the abundance of APCs and the prognosis-associated overlapping genes of HCC through Tumor Immune Estimation Resource (TIMER, https://cistrome.shinyapps.io/timer/) (Pearson correlation coefficient > 0.4, P value < 0.05) [32]. And these prognosis-associated overlapping genes significantly associated with one or more types of APCs were considered as possible candidates for mRNA vaccine development.

Immune subtypes identification based on the candidate genes in HCC

Using R package ‘ConsensusClusterPlus’, Partitioning Around Medoids (PAM) algorithm was applied to classify patients with HCC into several immune subtypes (ISs) based on the mRNA expression of the candidate genes. The clustering efficacy was assessed by principal component analysis (PCA). KM survival curves were conducted to evaluate the difference in OS between different ISs. Then, the associations between ISs and clinicopathological characteristics, such as histologic grade, TNM stage and vascular invasion, were also statistically analyzed with Chi-square test and Fisher’s test.

Cellular and molecular characterization of HCC

The single-sample GSEA (ssGSEA) algorithm was performed to estimate the enrichment levels of 21 previous reported immune-related terms, which were used represented the relative abundance of each type of immune cell [33–35]. ESTIAMTE algorithm was employed to produce an immune score, which reflected the immune infiltration level in the tumor microenvironment [36]. Moreover, 42 molecular signatures by Wolf et al. were also calculated for subsequent analysis [37]. Then, to explore the difference of molecular characterizations in differential samples, we employed R package ‘GSVA’ to quantify the 23 pivotal oncogenic pathways, which were obtained from MSigDB.

Construction and validation of the risk score system based on the immune subtypes

For better clinical utility, we developed a risk score system based on the immune subtypes. First, we figured out the differentially expressed genes (DEGs) among different ISs under the threshold of log2 fold change (FC) >1 or <−1 and P value < 0.05. The overlapping DEGs were subsequently analyzed for their associations with OS using univariate Cox regression analysis. Then, R package ‘glmnet’ was employed to conduct LASSO regression analysis for critical prognosis-associated DEGs. The risk score system was produced by multivariate Cox regression analysis as followed: risk score = ∑ the critical prognosis-associated gene expression × the corresponding coefficient. For external validation, the risk scores were further calculated for the GSE14520 cohort ICGC HCC cohort. Then, patients were divided into four risk subgroups (e.g. RS1, RS2, RS3 and RS4) on average for each cohorts, according to the risk score from low to high. KM survival analyses were conducted to evaluate the differences in OS among different risk groups. And the C-index and ROC curves were used to assess the predictive capability of the risk score system.

Multiplex immunofluorescence staining and imaging

Two tumor arrays with tumor and adjacent normal samples from 90 HCC patients (HLivH180Su09-T-001 and HLivH180Su09-T-002) were purchased from Shanghai Outdo Biotech (Co. Ltd., Shanghai, China.). All tissues were collected according to the ethical standards from Shanghai Outdo Biotech (No.: SHYJS-CP-1607007). And informed consent was obtained from all participants. For multi-color immunofluorescence staining, the PanoPANEL Kits (Panonvue, cat.No. TSA-RM-2759) was applied according to the manufacturer’s instructions. The molecules panel 1, including SLC2A1 (Proteintch, cat.NO. 66290-1-Ig, 1:300 dilution), S100A9 (Proteintch, cat.NO. 26992-1-AP, 1:200 dilution) and SPP1 (Proteintch, cat.NO. 22952-1-AP, 1:200 dilution), was conducted on the slide 1 (tumor array HLivH180Su09-T-001). And the molecules panel 2, including CDCA8 (Proteintch, cat.NO. 12465-1-AP, 1:50 dilution), DFNA5 (Proteintch, cat.NO. 13075-1-AP, 1:200 dilution), G6PD (Proteintch, cat.NO. 66373-1-Ig, 1:100 dilution) and PGF (Proteintch, cat.NO. 10642-1-AP, 1:200 dilution), was also conducted in the slide 2 (tumor array HLivH180Su09-T-002). In short, the slides were baked for 1 h at 60 °C, following by deparaffinization with xylene for 3 min three times. Then, microwave treatment was performed for antigen retrieval, and the blocking buffer was used to block the nonspecific binding for 15 min at room temperature. After blocking, the slides were incubated with primary antibodies for 1 h at 30 °C, following by incubation with secondary antibodies for 10 min at room temperature. After that, a specific fluorochrome from the PanoPANEL Kits for each primary antibody was used for the visualization of the corresponding antigens. For every consecutive antibody staining, the microwave treatment and antigen retrieval should be done like prior operation. Imaging was performed at 4× and 20× magnifications using the PanoVIEW VS200 Imaging System (Panonvue, China). The H-scores of each antibody staining were obtained based on the extent and intensity of the staining. To calculate the risk score, each protein expression score was normalized with a z-score.

Weighted gene co-expression network analysis (WGCNA)

The R package ‘WGCNA’ was applied to construct co-expression modules among 1590 immune-related genes in HCC [38]. Then, GSVA was performed to calculate the enrichment scores of these co-expressed immune-related genes modules, the differences of which were also investigate among tumors in various ISs or RSs [33]. And the KEGG and GO functional enrichment analyses were also performed for modules associated with HCC prognosis [39].

Statistical analysis

Statistical analysis was performed using the R software version 4.1.1 (http://r-project.org/) and GraphPad Prism 8.0 software (GraphPad Software, Inc.). Group differences analysis were performed using Wilcoxon rank-sum test or student’s t-test, and expressed as means ± standard deviation (SD). Chi-square test and Fisher’s test were performed for comparison of clinical characteristics. A two-tailed P value < 0.05 was considered as statistically significant. And all the work has been reported in line with the REMARK criteria.

Results

Screening of candidate targets for mRNA vaccination in HCC

First, GEPIA2 analysis identified 2218 aberrantly expressed genes in HCC, of which 1485 were overexpressed genes that possibly encoded tumor antigens (Fig. 1A; Table S1). Next, we detected 13105 genes that mutated in HCC, and the mutation counts and altered genome fraction in individual tumors were shown in Fig. 1B, C (Table S2). Of note, TTN, TP53, CTNNB1 and MUC16 were the most frequently mutated genes considering both mutation counts and altered genome fraction (Fig. 1D, E). Combine the mutation and expression data of HCC, 714 overexpressed and frequently mutated tumor-specific genes were identified to potentially encode tumor antigen in HCC. We subsequently identified 372 genes associated with both OS and DFS from the 714 aforementioned genes (Fig. 1F; Tables S3 and S4). Notably, of the 372 genes, 12 were identified notably associated with least one types of APCs (e.g. B cells, macrophages and dendritic cells) based on Pearson correlation coefficient > 0.4, including ATP1B3, CFL1, CMTM7, DYNC1H1, FABP5, IFT52, PAPSS1, PPT1, RIPK2, SCPEP1, TNFRSF4 and YWHAZ, and thus considered as suitable targets for mRNA vaccine development, the concept of which was first proposed by Huang et al (Fig. 1G–I; Table S5) [40]. Interestingly, for dendritic cells, the correlations of these 12 candidate genes were all greater than 0.4, whereas all less than 0.4 but greater than 0.3 for B cells, which in some way reflected that the tumor antigens encoded by these 12 candidate genes would be likely to be presented by the APCs (especially for dendritic cells) to T cells to trigger an anti-tumor immune response in HCC. And KM survival curves for OS were generated for these 12 candidate genes respectively based on the optimal cutoff points from X-tile software (Version 3.6.1) [41], and patients with higher expression of these 12 candidate genes exhibited shorter OS than those with lower expression (Fig. S1). In addition, with the same cutoff value respectively, all except FABP5, the selected candidates also associated with shorter DFS (Fig. S2). mRNA vaccines using the transcripts of these 12 candidate genes may likely activate the cognate interactions between professional APCs and T cells, thereby inducing anti-tumor immunity in HCC. These findings suggested these 12 genes as promising candidates for mRNA vaccine development.

Fig. 1.

Identification of potential candidate targets for mRNA vaccination in HCC. A Chromosomal distribution of overexpressed and under-expressed genes in HCCs. B–E Identification of genes encode HCC-specific antigens. B, C Samples overlapping in B mutation count and C altered genome fraction groups. D, E Genes with highest frequency in D mutation count and E altered genome fraction groups. F Potential HCC-specific antigens with high expression and mutation significantly associated with OS and DFS (total 371 candidates). G, H Identification of 12 potential candidate genes significantly associated with APCs infiltrations. G Volcano plot to show the association between the 12 candidate genes and B cell infiltrations. I Volcano plot to show the association between the 12 candidate genes and macrophage infiltrations. H Volcano plot to show the association between the 12 candidate genes and dendritic cell infiltrations

Immune subtypes identification based on the 12 candidate genes in HCC

Given the close association between the 12 aforementioned candidate genes and APCs, we further investigated whether the 12 candidate genes would generate an immunophenotyping for HCC, aiding to identification of suitable patients for mRNA vaccine. Therefore, PAM algorithm was conducted for consensus clustering of 371 samples from TCGA cohort based on the mRNA expression of the 12 candidate genes. Then, k = 3 was selected for stable clustering efficiency and three immune subtypes were obtained, namely immune subtype 1 (IS1), IS2 and IS3 in the TCGA cohort (Fig. 2A, B). And PCA validated that the 12 candidate genes worked well with significant clustering efficacy in the TCGA cohort (Fig. 2C). And the IS3 exhibited highest expression of these 12 candidate genes, whereas the IS2 the lowest (Fig. S3). KM survival curves also revealed significant different survival among these three subtypes, in which IS2 had the best survival probability whereas IS3 had the worst survival probability in the TCGA cohort (Fig. 2D). Furthermore, IS3 was notably associated with an advanced TNM stage and histologic grade as well as increased vascular invasion when compared with IS1 and IS2 (Fig. 2E–G). Similar results were also found in the ICGC cohort (Fig. 2H–L). And the mutation profiles involved 15 highly mutated genes in HCC were showed respective for different subtypes, indicating that IS3 tumors exhibited highest mutation rate of TP53 compared with IS1 and IS2 (Fig. 2M). In addition, IS2 had lowest fraction of genome alteration, though no significant difference in genome alteration fraction between IS1 and IS3 (Fig. 2N). Overall, the immune subtype could predict the clinicopathological outcomes and somatic mutation rates (e.g. TP53 and genome alteration fraction) in HCC patients.

Fig. 2.

Identifying a novel immune subtypes of HCC based on the 12 candidate genes. A Relative change in area under consensus clustering cumulative distribution function curve for k = 2 to 9. B The TCGA HCC cohort was divided into three distinct immune subtypes (e.g. IS1, IS2 and IS3) when k = 3. C PCA suggested the reliable clustering efficacy of the 12 candidate genes in the TCGA HCC cohort. D Significant differences in OS among these three subtypes were found in the TCGA HCC cohort. E–G Distribution ratio of clinicopathological characteristics across different subtypes in the TCGA HCC cohort. IS3 tumor exhibited an advanced TNM stage (E) and histologic grade (F) as well as increased vascular invasion (G) when compared with IS1 and IS2. H Similar immune subtyping were also identified in the ICGC HCC cohort when k = 3. I PCA suggested the reliable clustering efficacy of the 12 candidate genes in the TCGA HCC cohort. J Significant differences in OS among these three subtypes were also found in the ICGC HCC cohort. K, L Distribution ratio of clinicopathological characteristics across different subtypes in the ICGC HCC cohort. IS3 tumor exhibited an advanced TNM stage (K) and histologic grade (L). M The ten most commonly mutated genes in HCC were shown in the waterfall plot. IS3 tumors exhibited highest mutation rate of TP53 compared with IS1 and IS2. N IS2 had lowest fraction of genome alteration, though no significant difference in genome alteration fraction between IS1 and IS3

DEGs among different immune subtypes reflected the immunosuppressive phenotype of IS3 tumor

To investigate the immune subtype more deeply, differential expression analysis were conducted among the three subtypes, through which 223 genes were identified as the DEGs among these three subtypes (Fig. 3A–D, Table S6). GO enrichment analyses for the 223 DEGs were shown in Fig. 3E, and they were significantly enriched in immune-related biological processes, such as leukocyte aggregation, regulation of lymphocyte anergy, regulation of granulocyte macrophage colony-stimulating factor production, regulation of T cell tolerance induction, regulation of T-helper cell differentiation, antigen processing and presentation of exogenous peptide antigen via MHC class II, etc., most of which were closely related to immunosuppressive phenotypes in malignancies (Fig. 3E). Pathway enrichment analysis were also notably enriched in immune-related pathways, such as leukocyte transendothelial migration, Th1 and Th2 cell differentiation, neutrophil extracellular trap formation, B cell receptor signaling pathway, etc. (Fig. 3F). In addition, multiple oncogenic pathway associated with immunosuppression were also identified, including PI3K-Akt signaling pathway, HIF-1 signaling pathway, TNF signaling pathway, NOD-like receptor signaling pathway, Toll-like receptor signaling pathway, etc. (Fig. 3F). Interestingly, the IS3 exhibited highest expression of the DEGs involved in the immunosuppressive biological processes (e.g. regulation of lymphocyte anergy, regulation of granulocyte macrophage colony-stimulating factor production, regulation of T cell tolerance induction and antigen processing and presentation of exogenous peptide antigen via MHC class II), whereas the IS2 the lowest (Figs. 3A–C and S4). These findings further indicated that the immune subtype could reflect the immune suppression or evasion in HCC.

Fig. 3.

Establishment and validation of the risk score system based on the immune subtype. A Differential expression analysis between IS3 and IS1. B Differential expression analysis between IS3 and IS2. C Differential expression analysis between IS2 and IS1. D Overlapping DEGs among these three immune subtypes. E Dot plot showing the immune-related GO terms for the 223 DEGs. F Dot plot showing the KEGG terms for the 223 DEGs. G, H Lasso regression analysis identified 7 unfavorable prognostic genes (CDCA8, DFNA5, G6PD, PGF, S100A9, SLC2A1 and SPP1) for risk score system establishment. I–K Significant differences in OS of HCC patients in different risk subgroups in the TCGA HCC cohort (I), ICGC HCC cohort (J) and GSE14520 cohort (K). L–N ROC curve analysis suggested the reliable predictive ability of the risk score system in the TCGA HCC cohort (L), ICGC HCC cohort (M) and GSE14520 cohort (N)

Establishment and validation of the risk score system based on the immune subtype

Next, we conducted univariate Cox regression analysis on the 223 DEGs and 115 genes were notably associated with OS in HCC (P < 0.01) (Table S7). Then, lasso regression analysis were performed on these 115 prognosis associated DEGs, and finally identified seven unfavorable prognostic genes (CDCA8, DFNA5, G6PD, PGF, S100A9, SLC2A1 and SPP1; Fig. 3G, H), KM survival curve of which were shown in Fig. S5. A risk score system was developed based on the seven genes in the TCGA cohort (training cohort), the formula of which was shown as followed: risk score = (0.23111 × CDCA8 expression) + (0.04020 × G6PD expression) + (0.05146 × DFNA5 expression) + (0.09111 × S100A9 expression) + (0.03219 × SPP1 expression) + (0.07995 × SLC2A1 expression) + (0.12048 × PGF expression). And for external validation, the risk scores were also respectively calculated for the ICGC cohort and GSE14520 cohort. KM survival curves demonstrated that the risk score system had significant prognostic prediction efficacy, and increased risk score was closely associated with decreased survival in HCC (Fig. 3I–K). Of note, the risk score system exhibited an impressive predicting ability for OS prediction, with a C-index of 0.735 (95% CI, 0.690–0.780) in the TCGA cohort, 0.695 (95% CI, 0.613–0.777) in the ICGC HCC cohort, and 0.613 (95% CI, 0.556–0.670) in the GSE14520 cohort. In addition, the AUC values of the risk score for 1, 2, 3, and 4 year OS prediction were 0.811, 0.741, 0.744, and 0.746 in the TCGA cohort, 0.751, 0.716, 0.746, and 0.788 in the ICGC cohort, and 0.613, 0.661, 0.638, and 0.640 in the GSE14520 cohort (Fig. 3L–N).

Identification of the risk subgroup as an analogy to the immune subtype

It can be found that patients with IS3 tumors had highest expression of the 12 candidate genes (Figs. S6 and S7A). Besides, IS3 tumors also exhibited significantly highest risk scores, whereas IS2 lowest (Figs. 4A and S7B). And as the risk score increased, the proportion of IS3 tumors significantly increased, and IS2 tumors accounted for 88% of tumors in RS1, while IS3 tumors 60% of RS4 in the TCGA cohort (Fig. 4B), which were much consistent with that in the ICGC cohort (Fig. S7C). Then, to achieve better association with the three immune subtypes, we merged the RS2 and RS3 into one group, namely RS2+RS3, more than 60% of which were IS1 tumors (Fig. 4B). Besides, the AUC for IS3 tumor prediction of the risk score system in the TCGA cohort was 0.908 (95%CI, 0.874–0.943), while 0.935 (95%CI, 0.907–0.963) for IS2 tumors (Fig. 4C, D). And similar results were also found in the ICGC cohort (AUC = 0.927 (95%CI, 0.891–0.963) for IS3 tumors; AUC = 0.928 (95%CI, 0.888–0.968) for IS2 tumors) (Fig. S7D, E). KM survival curve indicated significant differences of OS among RS1, RS2+RS3, and RS4, in which tumors in RS4 had the worst survival, whereas tumors in RS1 the best (Fig. 4E). In addition, increased risk scores were closely associated with advanced clinicopathological outcomes, such as higher TNM stage and histologic grade, and increased vascular invasion, in the TCGA cohort (Fig. 4F–H). Similar results were also observed in the ICGC cohort and GSE14520 cohort (Fig. 4I–K). Increased risk scores were also significantly associated with advanced BCLC stage and CLIP stage as well as higher level of AFP in GSE14520 cohort (Fig. 4L–N). Besides, mutation profiles involved the top four mutated genes (e.g. TP53, CTNNB1, TTN and MUC16) in HCC were showed respective for different risk subgroup (Fig. 4O). Similar to tumors with IS3, tumors in RS4 exhibited highest mutation rate of TP53 compared with the other two subgroups (Fig. 4O). In addition, highest fraction of genome alteration was also observed in RS4 (Fig. 4P). Taken together, these findings suggested the reliable prognostic prediction efficacy of the risk score system. To a certain extent, the risk subgroup may be used as an analogy to the immune subtype.

Fig. 4.

Identification of the risk subgroup as an analogy to the immune subtype. A IS3 tumors also exhibited significantly highest risk scores, whereas IS2 lowest. B Distribution ratio of immune subtypes across different risk subgroups in the TCGA HCC cohort. C ROC curve analysis to assess the diagnostic capability of the risk score system for IS3 tumors. D ROC curve analysis to assess the diagnostic capability of the risk score system for IS2 tumors. E Significant differences in OS were found among these three risk subgroups (e.g. RS1, RS2+RS3 and RS4) in the TCGA HCC cohort. F–H Distribution ratio of clinicopathological characteristics across different risk subgroups in the TCGA HCC cohort. RS4 tumor exhibited an advanced TNM stage (F) and histologic grade (G) as well as increased vascular invasion (H) when compared with RS1 and RS2+RS3. I, J Distribution ratio of clinicopathological characteristics across different risk subgroups in the ICGC HCC cohort. RS4 tumor exhibited an advanced TNM stage (I) and histologic grade (J) when compared with RS1 and RS2+RS3. K–N Distribution ratio of clinicopathological characteristics across different risk subgroups in the GSE14520 cohort. RS4 tumor exhibited an advanced TNM stage (K), BCLC stage (L) and CLIP stage (M) as well as increased AFP level (N) when compared with RS1 and RS2+RS3. O The four most commonly mutated genes HCC were shown in in the waterfall plot. RS4 tumor exhibited highest mutation rate of TP53 compared with RS1 and RS2+RS3. P RS4 had highest fraction of genome alteration compared with RS1 and RS2+RS3

Validation of the risk score system at translational level

Of note, to validate the risk score system at translational level for clinical application, multi-color immunofluorescence histochemistry was applied to evaluate the protein expression level of the seven risk score system genes on a HCC tissue microarray (Fig. 5A, Table S8). Consistent with the aforementioned transcriptional analysis, tumors in RS4 had shortest OS and DFS (Fig. 5B, C), and ROC curve analysis also suggested the reliable predictive ability of the risk score system (Fig. 5D, E). Taken together, the risk score system is promising in predicting clinical survival of HCC patients by feat of the multi-color fluorescent immunohistochemistry analysis.

Fig. 5.

Validation of the risk score system at translational level. A Multi-color immunofluorescence histochemistry for the risk score system genes in tissue microarray of HCC. B KM survival curve showed the significant differences of OS among different risk subgroups in the tissue microarrays of HCC. C KM survival curve showed the significant differences of DFS among different risk subgroups in the tissue microarrays of HCC. D ROC curve showed the predictive capability of the risk score system for OS of HCC patients. E ROC curve showed the predictive capability of the risk score system for DFS of HCC patients

Cellular and molecular characteristics of risk subgroups

The therapeutic efficiency of mRNA vaccine depends on the tumor immunity. Therefore, we subsequently estimated the abundance of 21 immune cells using ssGSEA algorithm. As shown in Fig. 6A, B, RS4 exhibited significant increases in the infiltration of pro-tumor immune cells such as MDSC and regulatory T cells, as well as dendritic cells, in all three HCC cohorts. Besides, RS4 also appeared increased immune scores (Fig. 6C–E). RS4 also showed the highest scores for lymphocyte, tumor-associated macrophage, MHC-II, immune checkpoint, TGF-β signaling and tumor cell proliferation and migration signatures, indicating the malignant and immunosuppressive phenotype of RS4 (Fig. 6F). In addition, we also observed stepwise increased enrichment scores of multiple oncogenic pathways from low to high risk score groups (e.g. RS1, RS2+RS3 and RS4) (Fig. 6G). For example, RS4 displayed notable activation of multiple malignance and immunosuppression-related pathways, such as KRAS signaling, TGF-β signaling, PI3K-AKT-mTOR signaling, epithelial-mesenchymal transition, MYC signaling, and WNT-β-catenin signaling (Fig. 6G). Then, ImmuneCellAI (http://bioinfo.life.hust.edu.cn/ImmuCellAI/#!/) was utilized to assess the therapeutic response to ICBs [42]. Notably, 86% in RS4 were non-responsive to ICB, whereas 60% in RS1 exhibited best therapeutic response to ICB (Fig. 6H). Taken together, these results further suggested patients in RS4 may benefit more from mRNA vaccination due to the highly immunosuppressive TME.

Fig. 6.

Cellular and molecular characteristics of risk subgroups. A Heatmap to show the enrichment levels of 21 immune cell signatures among the three HCC risk subgroups. B Differential enrichment scores of MDSC, Treg cells, activated CD4+ T cells and dendritic cells among the three HCC risk subgroups in the TCGA HCC cohort, GSE14520 cohort and ICGC HCC cohort. C–E RS4 had markedly higher immune score than RS1 and RS2+RS3 in the TCGA HCC cohort (C), ICGC HCC cohort (D) and GSE14520 cohort (E). F Enrichment scores of 42 previously defined molecular signatures by Wolf et al. in various HCC risk subgroups. G Differential enrichment scores of multiple oncogenic pathways among HCC risk subgroups. H The proportion of patients with different therapy response to ICBs among different HCC risk subgroups

Identification of immune gene co-expression modules and hub gene of HCC

Immune gene co-expression modules were developed using WGCNA with a soft threshold of 5 for the scale free network (Fig. 7A, B). And six gene modules were identified (Fig. 7C, D, Table S9). We then investigated the distribution of different immune subtypes or risk subgroups in the eigengenes of six modules, and identified significant differences in four modules, including green, red, turquoise and grey modules (Fig. 7E, F). And IS3 and RS4 showed the highest eigengenes in the red and turquoise modules, whereas IS2 and RS1 the green and grey modules. In addition, KM survival curves showed that higher enrichment scores of the green module closely associated with superior OS of HCC, while higher enrichment scores of the red module decreased OS of HCC (Fig. 7G, H). Of note, the green module was closely related to anti-tumor immune processes, such as immune effector process, activation of immune response, cytolysis, etc., which, to some extent, verified its role as a favorable prognostic parameter of HCC (Fig. 7I). On the contrary, the red module significantly associated with TGF-β signaling pathway, mTOR signaling pathway, PD-L expression and PD-1 checkpoint pathway in cancer, T cell receptor signaling pathway, B cell receptor signaling pathway and antigen processing and presentation (Fig. 7J).

Fig. 7.

Immune gene co-expression modules of HCC. A Sample cluster. B Scale-free fit index and mean connectivity for various soft-thresholding powers. C Dendrogram of all immune genes clustered according to a dissimilarity measure. D Gene counts in each module. E Differential enrichment level of each module in HCC immune subtypes. F Differential enrichment level of each module in HCC risk subgroups. G Differential prognosis of tumors with high, medium and low enrichment score of the green module. H Differential prognosis of tumors with high, medium and low enrichment score of the red module. I Circular barplot showing KEGG terms and GO terms (CC cellular component; BP biology process; MF molecular function) in the green module. J Circular barplot showing KEGG terms and GO terms in the red module

Discussion

With the development of technique, mRNA vaccines have emerged as highly effective strategies in the prophylaxis and treatment of diseases [43]. Recently, analysis of potential tumor antigen in pancreatic adenocarcinoma and cholangiocarcinoma provided novel insights for mRNA vaccine development of solid tumors [40, 44]. Promising candidate for mRNA vaccine and suitable patients’ group are the key of successful vaccine treatment [45–47]. In this study, we analyzed the aberrantly expressed and mutational landscape of HCC, and screened out a series of potential candidate genes of mRNA vaccination, including ATP1B3, CFL1, CMTM7, DYNC1H1, FABP5, IFT52, PAPSS1, PPT1, RIPK2, SCPEP1, TNFRSF4, and YWHAZ. Overexpression of these antigens was not only correlated with poor OS and DFS, but also increased infiltration of APCs, especially for dendritic cells and macrophages. Consequently, these antigens may be crucial modulators for hepatocellular carcinogenesis and tumor progression, and may be directly processed and presented to cytolytic T lymphocytes to induce anti-tumor immunity.

Although these candidate antigens have to be functionally verified in future works, some of them were supported for mRNA vaccine development by previous studies. For instance, CFL1 is identified as a novel and frequent immunogenic tumor associated antigen in IDH^mut gliomas [48]. PPT1 is identified as an unfavorable biomarker in a variety of cancers, including HCC, and its inhibitor DC661 could promote dendritic cell maturation and CD8⁺ T cell activation, thereby enhancing anti-tumor immunity in HCC [49, 50]. Of note, Sharma et al. also reported that PPT1 inhibition reduced infiltration of MDSCs and enhances macrophage M2 to M1 polarization, which promoting the anti-tumor activity of CD8⁺ T cells in melanoma [51]. RIPK2 is a crucial modulator of both innate immunity and adaptive immunity, functioning mainly through the NOD/NF-кB signaling, MAPK signaling and c-MYC signaling [52, 53]. Zhou et al. also reported a RIPK2-dependent manner of NF-кB, MAPK and STAT3 activation, by which NOD2 enhances liver inflammatory response and hepatocarcinogenesis [54]. TNFRSF4 (OX40, CD134), one of the TNFR superfamily members, functions as a co-stimulatory receptor on T cells, induction of which through OX40L or agonist antibodies could enhance the proliferation, survival and activation of cytolytic T lymphocytes and reverses immune suppression within the tumor [55–59]. A recent study by Song et al. demonstrated that the combination of a TME-responsive penetrating nanogels with OX40 agonist antibody effectively enhances the anti-tumor effects of cytolytic T lymphocytes, and suppresses Treg cell, thereby tuning the “cold” phenotype of tumor to “hot” [60]. Similarly, Hong et al. also reported that the combinatory therapy of the synthetic TLR9 ligand and OX40 agonist antibody results in a local T cell anti-tumor immunity and eradicates cancer throughout the body [61]. Taken together, these findings, to some extent, suggest the values of these 12 candidate genes as vaccine targets for HCC.

Given that only a fraction of cancer patients may benefit from mRNA vaccine, we classified HCC patients into three immune subtypes based on the expression of the 12 candidate genes for the suitable population selection for mRNA vaccination. Interestingly, the three immune subtypes exhibited distinct clinicopathological, cellular and molecular characteristics. IS3 tumors showed worst prognosis and advanced clinicopathological features (e.g. TNM stage, histologic grade and vascular invasion) compared to the other two immune subtypes (e.g. IS1 and IS2), suggesting the immune subtypes as a potential prognostic biomarker for HCC. And increased mutation rate of TP53 was observed in tumors in IS3 compared with those in the other two immune subtypes. TP53 dysfunction results in immune suppression and evasion in TME, to some extent, suggesting the immunosuppressive phenotype of IS3 tumors [62, 63]. Of note, the combination of characteristics (e.g. regulation of lymphocyte anergy, regulation of T cell tolerance induction, regulation of granulocyte macrophage colony-stimulating factor production, antigen processing and presentation of exogenous peptide antigen via MHC class II), also suggests IS3 an immunosuppressive tumor and potential population suitable for immune response reactivation and enhancement through exogenous vaccines.

Then, to facilitate the clinical application of the immune subtype, we developed a risk score system based on the immune subtypes, which exhibited a reliable capability for OS prediction with satisfactory C-indexes and AUC values. The risk score system exhibited an impressive discrimination capability for IS2 and IS3 tumors with AUC values greater than 0.9. RS4 mainly overlapped with IS3, RS2+RS3 with IS1, and RS1 with IS2. In addition, RS4 tumor also exhibited a significant immunosuppressive and malignant phenotype. For example, increased infiltration of pro-tumor immune cells (e.g. MDCS and Treg cells) in RS4 tumors suggest an immunosuppressive TME of HCC, and may impair the therapeutic effect of in HCC, which was also evidenced by the low responsivity of ICBs in RS4 tumors. Previous studies also demonstrated that MDSCs exhibit immunosuppressive activities in TME, facilitating immune escape and non-response to ICB of tumor cells [64–67]. MDSCs are also able to facilitate the pro-tumor activities of Treg cells, thereby dampening anti-tumor immunity of effective T cells in HCC [68, 69]. In addition, aberrant activations of oncogenic signalings (e.g. KRAS signaling, MYC signaling, TGF-β signaling, etc.) notably lead to the immunosuppressive TME by recruiting pro-inflammation immune cells (e.g. MDSCs, plasmacytoid dendritic cells and Treg cells), thereby limiting the infiltration and antitumor activities of cytolytic T lymphocytes and therapeutic effect of ICBs to tumors [70–74]. The similarity in immunological characters between IS3 tumor and RS4 tumor was also observed through the distribution in immune gene co-expression modules that both IS3 and RS4 exhibited the highest eigengenes in the red module, but lowest in the green module. The immunosuppressive phenotype of the red module and the anti-tumor immune phenotype of the green module, to some extent, suggested that both IS3 and RS4 corresponded to tumors with anti-tumor immune anergy or suppression, needing mRNA vaccine to reactivate or enhance anti-tumor immunity, while IS2 or RS1 tumors may be therapeutically responsive to ICBs, due to the decreased infiltration of pro-tumor immune cells (e.g. MDCS and Treg cells) and less activation of immunosuppressive biological processes. Taken together, these findings suggested RS4 as an analogy to IS3, and the risk score system is promising and reliable in the identification of potential population suitable for mRNA vaccine.

As far as we know, no other study so far establishes a novel risk score system based on immune subtypes to select suitable HCC population for mRNA vaccination based on multi-omics analysis. However, several limitations to this study should be pointed out. First, the candidate genes for mRNA vaccination identified in this work should be experimentally validated in future works. Second, the association between immunotherapy effects and the immune subtype and risk subgroup remains further evaluation in a real world HCC cohort of immunotherapy.

In conclusion, this study identified ATP1B3, CFL1, CMTM7, DYNC1H1, FABP5, IFT52, PAPSS1, PPT1, RIPK2, SCPEP1, TNFRSF4 and YWHAZ as potential HCC antigens for mRNA vaccine development. HCC patients with RS4 tumors are suitable candidates for tumor vaccination. The current study provides a theoretical foundation for mRNA vaccine development, patient prognosis prediction and suitable population selection for HCC mRNA vaccination.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: Z.H.K., C.B., C.Y.J., and S.C.Z.; Methodology: Z.H.K., T.C.W., L.H., and Z.Z.D.; Investigation: Z.H.K., and T.C.W.; Writing—Original Draft: Z.H.K., T.C.W., L.H., and Z.Z.D.; Writing—Review & Editing: Z.H.K., C.X.M., T.C.W., W.W.T., T.W.L., Y.L., X.Z.Q., W.B.K. and W.Q.B.; Visualization: Z.H.K.; Supervision: Z.H.K., C.B., C.Y.J., and S.C.Z.; Funding Acquisition: C.B., C.Y.J., S.C.Z., and T.W.L.

Funding

This study was supported by National Natural Science Foundation of China (project NO.: 81972263, 82072714 and 82103221), the program of Guangdong Provincial Clinical Research Center for Digestive Diseases (2020B1111170004), China Postdoctoral Science Foundation (2020M683094), the Science and Technology Program of Guangzhou (202201011427), the Excellent Young Talent Program of Guangdong Provincial People’s Hospital (KY012021190) and the High-level Hospital Construction Project (DFJH201921).

Data availability

All data generated and described in this article are available from the corresponding web servers, and are freely available to any scientist wishing to use them for noncommercial purposes, without breaching participant confidentiality. Further information is available on reasonable request from the corresponding author.

Declarations

Ethics approval and consent to participate

The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All datasets are freely available as public resources. Therefore, local ethics approval was not required. Two tumor arrays with tumor and adjacent tumor samples from 90 HCC patients (HLivH180Su09-T-001 and HLivH180Su09-T-002) were purchased from Shanghai Outdo Biotech (Co. Ltd., Shanghai, China.). All tissues were collected according to the ethical standards from Shanghai Outdo Biotech (No.: SHYJS-CP-1607007). And informed consent was obtained from all participants.

Competing interests

The authors declare no competing interests.