Immune classifier-based signatures provide good prognostic stratification and predict the clinical benefits of immune-based therapies for hepatocellular carcinoma

The normal human liver cell line L02 and the HCC cell line Hep3B were purchased from the Chinese Academy of Sciences (Shanghai, China) and cultured in DMEM (Gibco, Carlsbad, CA, USA) supplemented with 10% foetal bovine serum and 1% penicillin. Cells were cultured at 37 °C in a 5% CO₂ atmosphere.

Total RNA was extracted from cells using an RNeasy Mini Kit (Qiagen, Valencia, USA) and then reverse transcribed into cDNA using the PrimeScript™ RT reagent Kit according to the manufacturer’s instructions. Relative mRNA expression levels were determined on an ABI 7500 Fast PCR instrument. GAPDH was used as the internal control. Relative expression levels of IL6, CCR3, SAA1, and GCG were quantified using the 2^−ΔΔCt method. The primers are listed in Additional file 1: Table S1.

We collected gene expression profile data (RNA-seq) from HCC patient samples in The Cancer Genome Atlas (TCGA) database (https://​cancergenome.​nih.​gov/​). In total, data from 371 patients with follow-up, status, and gene expression data were included in the liver hepatocellular carcinoma (LIHC) cohort. An additional 212 HCC datasets were downloaded from the International Cancer Genome Consortium (ICGC-LIRI-JP, https://​dcc.​icgc.​org/​projects/​LIRI-JP) for further validation. The ICGC-LIRI-JP dataset was used to verify the prognosis-related immune biomarker genes in the HCC cohort.

We downloaded immune cell-related gene profiles from the ImmPort database [21] (http://​www.​immport.​org). Immune cell-related genes were classified into different subgroups based on different immune cell characteristics, including cell type, functions, and associated pathways [22].

To evaluate immune cell infiltration characteristics in HCC patient samples, we performed ssGSEA to evaluate the degree of immune cell enrichment in the different samples. ssGSEA was performed using the gene set variation analysis (GSVA) package in R software [23].

We utilized the Estimation of STromal and Immune cells in MAlignant Tumour tissues using Expression data (ESTIMATE) method [24], as well as the CIBERSORT tool [25] and other algorithms to estimate cell type abundance from bulk tissue transcriptomes and to assess tumour immune scores and immune cell purities. The Microenvironment Cell Populations-counter (MCP-counter) [26], accessed through Connectivity Map at https://​portals.​broadinstitute.​org/​cmap/​, was used to predict potential drug sensitivities. In addition, the DESeq package [27] in R software was applied to determine differential gene expression in HCC tumour tissue and adjacent normal tissue from the raw CGA datasets. We defined the false discovery rate (FDR) to be no more than 0.05 and the |log₂ (fold change)| to be no less than one as the cut-off value.

To identify immune cell-related pathways, we utilized Kyoto Encyclopedia of Genes and Genomes (KEGG) and Gene Ontology (GO) analyses. Biological process (BP), molecular function (MF), and cellular component (CC) pathway analyses were included in the GO analysis. These analyses were performed using the clusterProfiler package in R software [28, 29].

To investigate the key regulatory genes and validate the connections between differentially expressed genes, a PPI network was constructed from gene expression datasets from the HCC cohort. We constructed the PPI network based on known and predicted PPIs. Furthermore, we defined the confidence score as  ≥  0.7, and the maximum number of interactors was identified as 0. These selected genes were input into Cytoscape (version 3.6.1) under the following parameters: degree cut-off  =  2, haircut on, node score cut-off  =  0.2, k-core  =  2, and max. depth  =  100.

We used single-factor Cox regression analysis to identify prognosis-related genes and to assess the effects of the immune status on patient prognoses [30]. We set a p value  <  0.01 as the significance level for immune cell-related genes. Multivariate Cox analysis was performed to calculate the risk scores. The median was defined as the dividing line between the high and low scores.

Both a submap and the Tumour Immune Dysfunction and Exclusion (TIDE) algorithm were used to predict potential responses to ICIs [31].

A dataset from the Genomics of Drug Sensitivity in Cancer (GDSC, https://​www.​cancerrxgene.​org/​) [32] database was used to screen for potential chemotherapeutics for the treatment of HCC. We estimated the half-maximal inhibitory concentrations (IC50s) for these drugs using the pRRophetic package in R software.

All statistical analyses were performed using SPSS 25 (IBM Corporation, Armonk NY) and R software (version 3.6.1). All statistical results with p  <  0.05 were considered significant.

To construct an immune cell gene-based classifier, we performed ssGSEA and consistent cluster analysis. Based on the different gene scores, we obtained four immune cell gene subgroups (Groups 1–4) (Fig. 1A). Advanced-stage HCC was mainly found in Group 4. A total of 16 immune cell-related gene clusters with high scores were also clustered in Group 4, corresponding with the activated immune subgroups. To further explore the prognoses associated with these four different subgroups, we conducted Kaplan–Meier analysis of overall survival, and the results demonstrated that the prognoses of patients in Group 4 were worse than those of patients in the other groups (Fig. 1B). Specifically, the prognoses of patients in Group 1 were better than those of patients in Group 4 (p  =  0.00023) (Fig. 1D); Group 2 had more favourable overall survival times than Group 4 (p  =  0.00018) (Fig. 1D); and the prognoses of patients in Group 3 were better than those of patients in Group 4 (p  =  0.0015) (Fig. 1E). We identified four HCC sample subgroups based on the immune scores, and Group 4 was associated with the most advanced clinical stages and significantly poorer prognoses than the other groups.

To further investigate immune cell infiltration in the four subgroups, we applied the CIBERSORT algorithm to the data and confirmed that the different subgroups also had significantly different levels of immune cell infiltration. There were significant differences in the proportions of naïve B cells, plasma cells, M0 macrophages, and resting dendritic cells. Group 4 had the lowest proportions of naïve B cells, plasma cells, and resting dendritic cells but the highest proportion of macrophages (Fig. 2A). The MCP-counter algorithm was used to assess the infiltration levels of immune cells in the HCC samples. The results showed that the infiltration levels of CD8⁺ T cells and cytotoxic lymphocytes were significantly different. Group 1 had higher degrees of CD8⁺ T cell and cytotoxic lymphocyte infiltration than the other groups (Fig. 2B). We also used the ESTIMATE method and found that Group 2 had the highest immune purity, with significant differences from that of Group 4 (Fig. 2C). Group 2 also had the lowest immune scores, and these scores were significantly different than those of Group 4 (Fig. 2D). Overall, the four-group classifier of immune system-related genes showed significant differences in immune scores, immune cell infiltration levels, and immune-based prognoses. Group 4 had higher levels of immune cell infiltration, was associated with advanced HCC stages, and had shorter overall survival times. Therefore, tumour immune escape might occur in samples from Group 4, with high immune infiltration, representing disabled and exhausted immune functions.

Immune checkpoint blockade (ICB) has been widely applied to treat a variety of tumours and shown significant favourable therapeutic effects [33]. HLA molecules are crucial for immune system function and are very clinically significant in immunotherapy [34]. The functional diversity of HLA genes is closely related to cancer genomics and cancer progression [35]. Importantly, responses to ICIs rely on the evolved efficiency of HLA-mediated immunity [36, 37]. We next investigated the internal relationship between the four groups of immune subgroups and HLA genes and identified several immune regulatory molecules, including HLA-A, HLA-E, and HLA-DRB5 (Fig. 3A). To further support correlations between immune checkpoint molecules and our immune-based classifier, we also determined that CD244, ICOS, ADORA2A, CD70, PD-L1 (CD274), and TIGIT molecules showed significant differences among the different immune-based subgroups (Fig. 3B). These results indicate that HLA genes and immune checkpoint molecules play crucial roles in the different subgroups and that they may be valid therapeutic targets for cancer therapies.

The tumour microenvironment is crucial for tumourigenesis and tumour development, and the CD8⁺ T cell-mediated anti-tumour response has been shown to be significantly increased through the production of cytokines such as INF-γ [38]. Accumulating evidence has shown that immune cells, such as CD8⁺ T cells, can facilitate the upregulation of immune checkpoints and enhance anti-tumour immune responses [39]. To annotate IFN-γ and T helper cell-related genes, we performed cluster analysis and found that the immune response has a crucial relationship with IFN-γ-related regulatory pathways (Fig. 4A). More importantly, Group 4 had significantly high expression levels of IFN-r-, IFNGR2-, IFNGR1-, JAK1-, and JAK2-related genes, with IFNG showing significant differences among the subgroups (Fig. 4B). Group 4 had higher IFNG expression levels than Group 2 (Fig. 4C).

mRNA and its transcription play an important role in immune-related gene regulation [40, 41]. The m⁵C methylation of RNA modifies the transcription of multiple genes, regulates protein expression, and affects cell phenotypes, and nucleoside modifications have been shown to suppress the potential for RNA to activate innate immune cells [42]. Here, m⁵C regulators were associated with all four groups. The NSUN7 regulator was expressed at low levels in all of these subgroups (Fig. 5B), and the overall levels of m⁵C regulator expression showed obvious differences between Group 2 and Group 4 (p  <  0.01) and less obvious differences between Group 1 and Group 4 (p  <  0.05) (Fig. 5B). The mRNA expression of NSUN4 was remarkably different among the immune subgroups (Fig. 5C).

We used the DESeq package in R software to identify and classify immune system-related DEGs. The upregulated DEGs in Group 4 were positively correlated with the immune classifier-based genes, and the upregulated DEGs in Groups 1, 2, and 3 were negatively correlated with the immune classifier-based genes (Fig. 6A). To further explore the regulatory roles of these DEGs, we used the clusterProfiler package in R software to characterize the associated genes. GO pathway analysis indicated that these DEGs are involved in a variety of cell regulation pathways, including the regulation of membrane potentials, neurotransmitter transport, and those for the regulation of blood pressure. CC analysis indicated that these DEGs play crucial roles in synaptic membranes, the presynapse, the ion channel complex, and transmembrane transporter pathways. MF analysis showed that the regulatory mechanisms for the DEGs are mainly enriched in receptor-ligand activity, signalling-receptor activator activity, and other such pathways (Fig. 6B). KEGG pathway annotations showed that Group 4 was enriched in WNT signalling pathway elements. The downregulated DEGs were enriched in immune response-related pathway elements (Fig. 6C).

To gain a deeper contextual understanding of their possible interactions, we constructed a PPI network and identified four hub genes (IL6, CCR3, SAA1, and GCG; Fig. 7A). To further identify HCC prognosis-related genes, Cox regression analyses were performed and yielded 45 DEGs. After applying stringent screening conditions (p  <  0.01), we identified five immune-related genes that demonstrated significant differences related to prognosis (SEMA3A, TNFRSF11B, GUCA2A, SAA, and CALCR). In addition, the expression levels of IL6, CCR3, SAA1, and GCG in the HCC cell line (Hep3B) and control cell line (L02) were determined by qRT-PCR. As shown in Fig. 7B, C, the expression of IL6 and CCR3 was higher in the HCC cell line (Hep3B) than in the control cell line (L02). The expression level of SAA1 was lower in Hep3B cells than in L02 cells (Fig. 7D). There was no significant difference in GCG expression between the HCC cell line and the control cell line (Fig. 7E). Using multivariate Cox regression for further analysis, we constructed a risk-coefficient model using these five genes. This model demonstrated that the subgroup with the highest risk scores had poorer prognoses than the subgroup with the lowest risk scores (p  =  0.014; Fig. 7F).

To further validate these five immune gene-based biomarkers, Cox regression analyses were applied to datasets from the International Cancer Genome Consortium (ICGC-LIRI-JP). These alternate independent analyses showed that the group with the highest risk scores also had poorer prognoses than the group with the lowest risk scores (Fig. 7G). Area under the curve (AUC) analyses (Fig. 7H), which were used to quantify and evaluate diagnostic accuracy, showed that the AUC value for the five gene-based prognostic model was 0.71 (95% CI  =  61.01–81.94).

Further studies were performed to explore the relationship between the four immune subgroups and immunotherapy responses by evaluating the clinical immunotherapy responses to PD-1 and CTLA4. Interestingly, Group 4 was characterized as T cell-deficient and had the highest immune deficiency scores (Fig. 8A). These studies confirmed that immune deficiency is involved in tumour immune escape, especially in Group 4.

To comprehensively examine the responses of the different immune subgroups to chemotherapeutic agents, we used the pRRophetic algorithm in R software. The results indicated that the half-maximal inhibitory concentration (IC50) values of Group 4 were significantly different than those of the other subgroups (the corresponding drugs are listed in Fig. 8B–S), but not all of these drugs exhibited highly sensitive responses. Only EHT 1864, FH535, and lapatinib exhibited higher sensitivities in Group 4, demonstrating the limited selection of drugs for treating advanced-stage HCC tumours.