Abstract Background TP53 mutation is the most common mutation in hepatocellular carcinoma (HCC), and it affects the progression and prognosis of HCC. We investigated how TP53 mutation regulates the HCC immunophenotype and thus affects the prognosis of HCC. Methods We investigated TP53 mutation status and RNA expression in different populations and platforms and developed an immune prognostic model (IPM) based on immune-related genes that were differentially expressed between TP53^WT and TP53^MUT HCC samples. Then, the influence of the IPM on the immune microenvironment in HCC was comprehensively analysed. Findings TP53 mutation resulted in the downregulation of the immune response in HCC. Thirty-seven of the 312 immune response-related genes were differentially expressed based on TP53 mutation status. An IPM was established and validated based on 865 patients with HCC to differentiate patients with a low or high risk of poor survival. A nomogram was also established for clinical application. Functional enrichment analysis showed that the humoral immune response and immune system diseases pathway represented the major function and pathway, respectively, related to the IPM genes. Moreover, we found that the patients in the high-risk group had higher fractions of T cells follicular helper, T cells regulatory (Tregs) and macrophages M0 and presented higher expression of CTLA-4, PD-1 and TIM-3 than the low-risk group. Interpretation TP53 mutation is strongly related to the immune microenvironment in HCC. Our IPM, which is sensitive to TP53 mutation status, may have important implications for identifying subgroups of HCC patients with low or high risk of unfavourable survival. Fund This work was supported by the International Science and Technology Cooperation Projects (2016YFE0107100), the Capital Special Research Project for Health Development (2014-2-4012), the Beijing Natural Science Foundation (L172055 and 7192158), the National Ten Thousand Talent Program, the Fundamental Research Funds for the Central Universities (3332018032), and the CAMS Innovation Fund for Medical Science (CIFMS) (2017-I2M-4-003 and 2018-I2M-3-001). Keywords: TP53, Mutation, Immune prognostic model, Immune profile, Hepatocellular carcinoma __________________________________________________________________ Research in context. Evidence before this study We searched PubMed through Feb 20, 2019, for research articles containing the terms “immune prognostic model AND hepatocellular carcinoma” without language or date restrictions. This search did not find any previous high-throughput studies that had investigated the potential prognostic role of immune prognostic models in hepatocellular carcinoma. In addition, the same search method was used to identify articles containing the terms “immune prognostic model AND TP53”. This search also identified no previous high-throughput studies that had investigated the relationship between immune prognostic models and TP53. Added value of this study We found that the immune phenotype was related to TP53 mutation and developed and validated an immune prognostic model for hepatocellular carcinoma that was affected by TP53 mutation status. This model is based on the expression of 2 immune genes that differentiate patients with a low or high risk of poor survival in both the training and validation cohorts. Our study included 865 patients with hepatocellular carcinoma to establish and validate an immune prognostic model, and to our knowledge, it is the largest prognostic model discovery project for hepatocellular carcinoma. Our results suggest that this immune prognostic model is more accurate than clinicopathological risk factors alone. We further developed a nomogram to predict patient prognosis, and it consisted of the immune prognostic model, vascular tumour invasion and hepatitis C status. Implications of all available evidence For the first time, we identified and validated an immune prognostic model based on 2 immune genes. This model has independent prognostic significance for patients with hepatocellular carcinoma and directly quantifies mRNA expression; thus, it has considerable potential for use in future clinical trials and could be implemented for determining the prognoses of individual patients in clinical practice. Moreover, the model reflects the intensity of the immune response triggered by TP53 status in the microenvironment of hepatocellular carcinoma. This study is also the first to describe an immune prognostic model associated with TP53 mutations and can be used as a reference for understanding other cancers. Alt-text: Unlabelled Box 1. Introduction Hepatocellular carcinoma (HCC) ranks sixth among the most common types of cancer and has one of the highest mortality rates among cancers [[39]1,[40]2]. Currently, there are a number of established treatments for HCC, including chemotherapy with sorafenib, vascular catheterization, radiofrequency ablation, surgical resection, and liver transplantation [[41]3,[42]4]. However, the recurrence rate is high, even for patients who have received treatment in the early stage, and the survival rate of patients with advanced cancer, including those who receive treatment, is poor [[43]4]. Tumour-promoting immune diseases are considered to enable the development of HCC. HCC cells stimulate a significant immune response, which yields the proper microenvironment for their development [[44]5]. Because of the poor prognosis after standard treatment, immunotherapy is being studied in depth as an additional treatment [[45]6]. In addition, a number of immune-related parameters have been reported to predict the prognosis of patients with HCC, further emphasizing the significance of immune status for determining the outcomes of HCC [[46]5,[47]7]. Nevertheless, few studies have systematically investigated the immune phenotype within the HCC microenvironment and its relationship with prognosis. Sentences similar to “In human cancer, TP53 is the most commonly mutated gene” have appeared in the introductions of thousands of publications dating back to 1990, one year after the first TP53 mutation was described in colorectal and lung cancer [[48]8,[49]9]. This discovery was followed by the identification of hundreds of new cancer genes, although none of them surpassed the importance of the discovery of TP53. Although thousands of cancer genomes have been sequenced, candidates of paramount importance have not been found. TP53 mutation is often observed and is among the five most conspicuous mutations in common human cancers [[50]10,[51]11]. The wild-type TP53 protein plays important roles in apoptosis after DNA damage and in cell cycle regulation [[52]12]. However, in the event of TP53 mutation, cells with DNA damage can escape apoptosis and transform into cancer cells. Furthermore, the mutant TP53 protein loses its wild-type function and accumulates in the nucleus [[53]13]. This accumulation is considered to be a highly specific marker of malignant tumours [[54]13]. A study covering 12 tumour types with a total of 3281 tumours found that the average mutation frequency of TP53 was approximately 42% [[55]11]. The high mutation rate of TP53 makes its genetic alteration a very attractive potential therapeutic target. Gene therapy, targeted tumour vaccines, and anticancer drugs targeting TP53 mutations, including APR-246, MK-1775, ALT-801, and Kevetrin, are in the early stages of clinical trials. TP53 mutation is also the most common mutation in HCC [[56]14]. This gene plays an important role in maintaining genomic stability, and its functional deletion can cause centrosome amplification, aneuploid cell proliferation and chromosomal instability (CIN) [[57]15]. In particular, when TP53 mutations are combined with functional defects in the tumour suppressor pRb or with spindle checkpoint defects, they are more likely to cause high-level CIN and genomic instability [[58]16]. Considerable data have shown that mutant TP53 proteins simultaneously lose their tumour-suppressive functions and obtain new capacities to advance tumourigenesis [[59]17]. In HCC, TP53 alterations are correlated with serum alpha-fetoprotein (AFP) levels, tumour stage, vascular invasion, tumour differentiation and Child-Pugh class [[60][18], [61][19], [62][20], [63][21]]. Compared with HCC patients with wild-type TP53, those with tumour TP53 mutations have shorter overall survival (OS) and relapse-free survival times [[64]22]. Thus, understanding the exact effects of TP53 on the pathogenesis of HCC and other forms of cancer is critical. Interestingly, one of the most recent studies suggested that different immune responses are related to TP53 mutational status [[65]23,[66]24]. Therefore, we speculate that the shorter OS of HCC patients with TP53 mutation may be partly caused by the specific influences of these mutations on the cancer-associated immune system. In this study, we conducted a comprehensive analysis of TP53 mutation status and RNA expression to study the relationship between TP53 mutations and immune responses in HCC. The results showed that the immune response of HCC without TP53 mutation (TP53^WT) was markedly stronger than that of HCC with TP53 mutation (TP53^MUT). Importantly, our immune prognostic model (IPM) including immunological genes whose expression is affected by TP53 mutations can be used as an important prognostic model and has potential for use in patient management, and the included genes can serve as potential therapeutic biomarkers for HCC. 2. Materials and methods 2.1. RNA-sequencing data The somatic mutation status for 364 HCC samples (workflow type: VarScan2 Variant Aggregation and Masking), and gene expression data and the corresponding clinical datasheets for 374 HCC samples were obtained from the Cancer Genome Atlas (TCGA) website ([67]https://portal.gdc.cancer.gov/repository) (up to September 10, 2018) [[68]14]. Surgical resection samples were collected from patients diagnosed with HCC, and these patients did not receive prior treatment for their disease [[69]14]. Among these HCC samples, 359 HCC samples with RNA-sequencing data and TP53 mutation information were subjected to subsequent analyses. Sequence data were obtained using the Illumina HiSeq_RNA-Seq and Illumina HiSeq_miRNA-Seq platforms. The study reported herein fully satisfies the TCGA publication requirements ([70]http://cancergenome.nih.gov/publications/publicationguidelines). The gene symbols were annotated based on the Homo_sapiens.GRCh38.91.chr.gtf file ([71]http://asia.ensembl.org/index.html). Log[2] transformations were performed for all gene expression data. The function of the trimmed mean of M values (TMM) normalization method of the edgeR R package (Version 3.24.3; [72]http://www.bioconductor.org/packages/release/bioc/html/edgeR.html) in R software (Version 3.5.2; [73]https://www.r-project.org/) was applied to normalize the downloaded data [[74]25]. The average RNA expression value was used when duplicate data were found. Genes with an average expression value >1 were retained, and low-abundance RNA-sequencing data were removed. 2.2. Microarray data The gene expression profile matrix files from [75]GSE54236 based on platform [76]GPL6480 (including 78 HCC samples and 77 adjacent noncancerous samples), [77]GSE76427 based on platform [78]GPL10558 (including 115 HCC samples and 52 adjacent noncancerous samples), and [79]GSE14520 based on platform [80]GPL571 (including 225 HCC samples and 220 adjacent noncancerous samples) were downloaded from the Gene Expression Omnibus (GEO) database ([81]https://www.ncbi.nlm.nih.gov/geo/). Among these datasets, only gene expression data for [82]GSE76427 were subjected to log[2] transformation. The average RNA expression value was taken when duplicate data were found. Genes with an average expression value >1 were retained, and low-abundance RNA-sequencing data were removed. Three datasets ([83]GSE54236 (n = 78), [84]GSE76427 (n = 115), and [85]GSE14520 (n = 221)) with survival information were integrated into the meta-GEO HCC cohort (n = 414) to validate the IPM. The sva package (Version: 3.30.1; [86]http://bioconductor.org/packages/release/bioc/html/sva.html) was used to eliminate batch effects, and the scale method of the limma R package (Version 3.38.3; [87]http://www.bioconductor.org/packages/release/bioc/html/limma.html) was used to normalize the data [[88]26]. The obtained data were used according to the TCGA and GEO data access policies. Both mRNA profile data and clinical feature data for HCC are publicly obtainable and open access. All analyses were carried out based on pertinent guidelines and regulations. 2.3. Patients in the Peking HCC cohort and sample collection From 2004 to 2015, 101 patients who underwent surgery and were diagnosed with HCC at Peking Union Medical College Hospital (Beijing, China) participated in this study in accordance with the provisions of the Helsinki Declaration ([89]Table S1). These patients did not undergo neoadjuvant therapy before surgery. Two experienced pathologists examined all haematoxylin and eosin (H&E)-stained slides of each tumour sample. All final diagnoses were based on the morphology of the tumour samples after staining with H&E. Informed consent forms were signed by all patients. One-hundred thirty-one formalin-fixed paraffin-embedded HCC samples were collected to examine the protein levels of immune genes. 2.4. Immunohistochemistry (IHC) Paraffin-embedded HCC samples were serially sectioned at 4-μm intervals and subsequently mounted on glass slides. The slides were then baked in the oven at 60 °C for 1 h, deparaffinized, and rehydrated. Heat-mediated antigen retrieval was conducted in a pressure cooker in 10 mmol/L Tris-citrate buffer (pH: 6.0). Endogenous peroxidase activity was blocked by incubating the sections with 3% hydrogen peroxide at room temperature for 10 min. After washing with phosphate-buffered saline (PBS) and incubation with goat serum at room temperature for 30 min, the slides were incubated with primary antibodies overnight at 4 °C. After washing with PBS, each slide was incubated with the appropriate peroxidase-labelled AffiniPure goat anti-rabbit IgG (H + L) (111–035-0030, 1:200, Jackson) secondary antibody for 30 min. Each section was washed with PBS and then developed with 3.3′-diaminobenzidine (DAB) solution for 5 min. Each section was washed with water before counterstaining with haematoxylin. The results of IHC staining were evaluated and scored by two pathologists. For EXO1 (exonuclease 1) expression analysis, a primary anti-EXO1 antibody (LS-C408381, 1:100; LifeSpan) was used. EXO1 is localized in the nucleus of tumour cells. The proportion of stained tumour cells was counted by two pathologists. Scores for the intensity of staining were determined as follows: 0 (negative), 1 (weak), 2 (moderate), and 3 (strong). The staining index (SI) for EXO1 was calculated as staining intensity × the proportion of positive tumour cells. For TREM-1 (triggering receptor expressed on myeloid cells-1) expression analysis, a primary anti-TREM-1 antibody (ab225861, 1:200; Abcam) was used, and two pathologists counted the number of TREM-1-positive infiltrating lymphocytes (TILs). Images were obtained using a NanoZoomer S210 C13239-01 scanner. 2.5. Gene set enrichment analysis (GSEA) To determine how the immunological pathways and corresponding immune genes differ between HCC samples without (n = 249) and with (n = 110) TP53 mutations in the TCGA HCC cohort, GSEA (Version: 3.0; [90]http://software.broadinstitute.org/gsea/index.jsp) was performed [[91]27]. An annotated gene set file ([92]c5.bp.v6.2.symbols.gm) was selected for use as the reference gene set. The threshold was set at P < 0.05. 2.6. Differentially expressed gene (DEG) analysis We compared 249 HCC samples without TP53 mutations and 110 HCC samples with TP53 mutations to identify DEGs using the edgeR R package, and the thresholds were |log[2]-fold change (FC)| > 2.0 and FDR < 0.01 [[93]25]. 2.7. Construction and validation of an immune-related prognostic model Among the 359 HCC samples with RNA-sequencing data and TP53 mutation information, 350 HCC samples with survival information were subjected to subsequent analyses. The expression profiles of the DEGs from 350 HCC patients with survival information were analysed via univariate Cox regression analysis. The prognostic value of the DEGs for OS was defined by univariate Cox regression analysis. In this analysis, genes were regarded as significant at P < 0.001. For highly correlated genes, the traditional Cox regression model cannot be used directly; thus, least absolute shrinkage and selection operator (LASSO) with L[1]-penalty, which is a popular method for determining interpretable prediction rules that can handle the collinearity problem, was used [[94]28]. Among the immune genes that were significant in the univariate Cox regression analysis, key immune genes were selected by the LASSO method. In this approach, a sub-selection of immune genes involved in HCC patient prognosis was determined by shrinkage of the regression coefficient via the imposition of a penalty proportional to their size. Finally, a relatively small number of indicators with a weight of nonzero remained, and most of the potential indicators were shrunk to zero. Therefore, LASSO-penalized Cox regression was implemented to further reduce the number of immune genes. In this analysis, we subsampled the dataset 1000 times and chose the immune genes that were repeated >900 times [[95]29]. LASSO Cox analysis was performed by using the glmnet R package (Version: 2.0–16; [96]https://cran.r-project.org/web/packages/glmnet/index.html). Finally, an immune-related prognostic model was constructed utilizing the regression coefficients derived from multivariate Cox regression analysis to multiply the expression level of each immune gene. X-tile 3.6.1 software (Yale University, New Haven, CT, USA) was applied to determine the best cutoff for HCC patients classified as low risk and high risk. The log-rank test and Kaplan-Meier survival analysis were used to assess the predictive ability of the prognostic model. 2.8. Estimation of immune cell type fractions CIBERSORT is an approach to characterizing the cell composition of complex tissues based on their gene expression profiles, and it is highly consistent with ground truth estimations in many cancers [[97]30]. A leukocyte gene signature matrix consisting of 547 genes, which was termed LM22, was used to distinguish 22 immune cell types, and these types contained myeloid subsets, natural killer (NK) cells, plasma cells, naive and memory B cells and seven T cell types. We utilized CIBERSORT in combination with the LM22 signature matrix to estimate the fractions of 22 human haematopoietic cell phenotypes between HCC samples with and without TP53 mutations. The sum of all estimated immune cell type fractions is equal to 1 for each sample. 2.9. Functional enrichment analysis The Database for Annotation, Visualization and Integrated Discovery (DAVID) (Version: 6.8; [98]https://david.ncifcrf.gov/) and the KO-Based Annotation System (KOBAS) (Version: 3.0; [99]http://kobas.cbi.pku.edu.cn/) were used to perform functional and pathway enrichment analyses to assess the biological implications of the prognostic model [[100]31,[101]32]. Significant biological processes and pathways were visualized using the GOplot (Version: 1.0.2; [102]https://cran.r-project.org/web/packages/GOplot/index.html) and ggalluvial (Version: 0.9.1; [103]https://cran.r-project.org/web/packages/ggalluvial/index.html) R packages, respectively. 2.10. Independence of the IPM from traditional clinical features Among 350 HCC samples with survival information, 213 HCC samples with complete clinical information, including AFP, gender, weight, age, pathologic stage, vascular tumour invasion, weight, histologic grade, hepatitis B status, hepatitis C status, alcohol consumption status and non-alcoholic fatty liver disease status, were subjected to subsequent analyses. To validate whether the predictions of the prognostic model were independent of traditional clinical features (including AFP, gender, weight, age, pathologic stage, vascular tumour invasion, weight, histologic grade, hepatitis B status, hepatitis C status, alcohol consumption status and non-alcoholic fatty liver disease status) for patients with HCC, univariate and multivariate Cox regression analyses were conducted. 2.11. Construction and evaluation of the nomogram To individualize the predicted survival probability for 1 year, 3 years and 5 years, a nomogram was constructed based on the results of the multivariate analysis. The rms R package (Version: 5.1–3; [104]https://cran.r-project.org/web/packages/rms/index.html) was used to generate a nomogram that included significant clinical characteristics and calibration plots. Calibration and discrimination are the most commonly used methods for evaluating the performance of models. In this study, the calibration curves were graphically assessed by mapping the nomogram-predicted probabilities against the observed rates, and the 45° line represented the best predictive values. A concordance index (C-index) was used to determine the discrimination of the nomogram, and it was calculated by a bootstrap approach with 1000 resamples [[105]33]. In addition, the predictive accuracies of the nomogram and separate prognostic factors were compared using the C-index and receiver operating characteristic (ROC) analyses. All statistical tests were two-tailed with a statistical significance level set at 0.05 in this study. 3. Results 3.1. Association between immune phenotype and TP53 mutations in HCC In HCC, TP53 mutation is the most common type of mutation ([106]Fig. 1A). Pioneering investigations demonstrated that TP53 mutation is associated with OS in patients with HCC [[107]22]. Although the pathogenetic role of TP53 mutations in the prognosis of patients with HCC has been well documented, their specific influences on immune profiles in HCC have not been thoroughly investigated. Hence, for the first time, we utilized gene expression data and clinical information on HCC patients in TCGA to find immune-related biological processes linked to TP53 status. GSEA analysis of HCC samples without (n = 249) and with (n = 110) TP53 mutations was performed. The results showed that TP53^WT HCCs were significantly enriched in 414 biological processes, and 4 immune-related biological processes were selected: REGULATION_OF_HUMORAL_IMMUNE_RESPONSE (normalized enrichment score, NES = 2.201, size = 47), NEGATIVE_REGULATION_OF_DEFENSE_RESPONSE_TO_VIRUS (NES = 1.722, size = 17), NEGATIVE_REGULATION_OF_IMMUNE_EFFECTOR_PROCESS (NES = 1.681, size = 95), and HUMORAL_IMMUNE_RESPONSE (NES = 1.586, size = 153) (P < 0.05) ([108]Fig. 1B) ([109]Table S2). In contrast, TP53^MUT HCCs were not enriched in any immune-related biological processes ([110]Table S3). Fig. 1. [111]Fig. 1 [112]Open in a new tab Gene set enrichment analysis of TP53 in the TCGA dataset. (A) Genomic landscape of HCC and the mutational signatures in the TCGA dataset, which were assayed on the FireBrowse platform. (B) Significant enrichment of the immune-related phenotype in TP53^WT HCC patients compared with that in TP53^MUT HCC patients. 3.2. Identification of differentially expressed immune-related genes between HCC samples with and without TP53 mutations To identify the correlations between TP53 status and 4 immune-related processes, 312 immune-related genes were obtained from the 4 immune-related processes. To identify differentially expressed immune-related genes between TP53^WT HCC and TP53^MUT HCC tissues, we performed differential expression analysis using the edgeR package [[113]25]. Of the 312 immune-related genes investigated, 37 genes were differentially expressed between TP53^WT and TP53^MUT HCCs (FDR < 0.05 and |log[2] FC| > 1) ([114]Table S4). 3.3. Construction of an IPM and evaluation of its predictive ability in the TCGA HCC cohort Taking the differences in immune status between TP53^WT and TP53^MUT HCCs into consideration, we attempted to assess the predictive ability of the DEGs. Univariate Cox regression analysis was performed, and it revealed that 7 of the 37 DEGs were significantly related to OS ([115]Table S5). To find the genes with the greatest prognostic value, we applied Cox-proportional hazards analysis based on the L1-penalized (LASSO) estimation, and two genes (TREM1 and EXO1) that appeared >900 times out of 1000 repetitions were selected [[116]29,[117]34]. We used LASSO because it is suitable for constructing models when there are a large number of correlated covariates [[118]34]. To obtain a uniform cutoff value to stratify the patients into high- and low- risk groups, we conducted normalization of the expression levels of TREM1 and EXO1 in the TCGA, meta-GEO and Peking HCC cohorts with mean value = 0 and standard deviation (SD) = 1 [[119]35]. Then, by weighting the normalized expression level of each immune gene to the regression coefficients of the multivariate Cox regression analysis, we established a risk score model to predict patient survival (risk score = normalized expression level of TREM1 * 0.336 + normalized expression level of EXO1 * 0.392). We calculated the risk score for each patient and categorized the patients into high-risk or low-risk groups according to the optimal cutoff point (1.37) obtained from X-tile software. The cutoff point (1.37) in the TCGA HCC cohort served as the cutoff to assign patients into high- and low- risk groups across all the HCC cohorts. As shown in [120]Fig. 2A, the high-risk patients had a shorter OS than their low-risk counterparts. In addition, the high-risk group showed a 3.17-fold higher risk (95% confidence interval (CI): 2.02–4.98, P < 0.001) than the low-risk group. The risk score distribution and gene expression data are shown in [121]Fig. 2B. [122]Fig. 2C shows the predictive potential of the IPM using time-dependent ROC curves. The area under the ROC curve (AUC) of the prognostic model for OS was 0.7048 at 0.5 years, 0.7388 at 1 year, 0.7119 at 2 years, 0.7276 at 3 years and 0.6558 at 5 years. Fig. 2. [123]Fig. 2 [124]Open in a new tab Prognostic analysis of the IPM. Kaplan-Meier survival, risk score and time-dependent ROC curves of the IPM for the TCGA HCC cohort (A-C) and meta-GEO HCC cohort (D-F). (A and D) OS was significantly higher in the low-risk score group than in the high-risk score group. (B and E) Relationship between the risk score (upper) and the expression of two prognostic immune genes (bottom) is shown. (C and F) Time-dependent ROC curve analysis of the IPM. (G-H) The C-index was used to evaluate prognostic performance for survival prediction. Performance was compared between the IPM and 3-gene signature_2018 by calculating the C-index in the TCGA and meta-GEO HCC cohorts. (I) Kaplan-Meier survival of the IPM for the Peking HCC cohort by using immunohistochemistry. 3.4. Validation and evaluation of the IPM in the meta-GEO HCC cohort and Peking HCC cohort To determine whether the IPM was robust, the performance of the IPM with the TCGA HCC cohort was assessed in the meta-GEO HCC cohort, which consisted of 414 HCC patients. With the same formula and the same cutoff obtained from the TCGA HCC cohort, the patients in the meta-GEO HCC cohort were divided into a high-risk group and a low-risk group. Consistent with the outcomes of the TCGA HCC cohort, patients who were assigned to the high-risk group had significantly worse OS than those who were assigned to the low-risk group ([125]Fig. 2D). The risk in the high-risk group was 1.97-fold higher than that in the low-risk group (95% CI: 1.37–2.83, P < 0.001), demonstrating the applicability of the developed IPM in different platforms. The risk score distribution and gene expression data are shown in [126]Fig. 2E. Furthermore, the IPM achieved an AUC of 0.6781 at 0.5 years, 0.5657 at 1 year, 0.6111 at 2 years, 0.6260 at 3 years and 0.6028 at 5 years ([127]Fig. 2F). Recently, Yang et al. proposed a prognostic model including 3 genes (secreted phosphoprotein 2 (SPP2); cell division cycle 37-like 1 (CDC37L1); and enoyl-CoA hydratase domain containing 2 (ECHDC2)) to predict the prognosis of patients with HCC [[128]36]. They first integrated 7 HBV-associated HCC datasets to identify DEGs. Second, weighted gene co-expression network analysis (WGCNA) was performed on those DEGs to identify the most significant module. Third, a protein-protein interaction (PPI) network was constructed for the most significant module to identify hub genes. Finally, a three-gene prognostic signature (risk score = expression of SPP2 * - 0.1941 + expression of CDC37L1 * - 0.5466 + expression of ECHDC2 * - 0.4714) for these hub genes was established by univariate and multivariate Cox regression analysis in the [129]GSE14520 dataset. We calculated the C-indexes to compare the prognostic values of their model and our IPM. The C-index is the most commonly used performance measure for survival models; it ranges from 0.5 to 1 and is equal to the AUC [[130]37]. The higher the value of the C-index is, the better the predictability of the model. The C-index of the IPM for 1 to 5-year OS exceeded that of the previous model in both the TCGA and meta-GEO HCC cohorts, suggesting that our IPM had favourable efficacy for predicting both short- and long-term prognosis ([131]Fig. 2G and H ). To further examine the robustness and practical application of the IPM, we validated the prognostic power of the IPM using protein values for immune genes and survival information for patients with HCC in our cohort recruited from Peking Union Medical College Hospital. This cohort consisted of 101 HCC patients. We detected the protein levels of two immune genes (TREM1 and EXO1) with IHC. The results revealed that the IPM consisting of these two immune genes at the protein level can differentiate HCC patients with a low or high risk of poor survival based on the same formula and the same cutoff obtained from the TCGA HCC cohort. Representative staining images of TREM1 and EXO1 were demonstrated in [132]Fig. S1. The patients in the high-risk group exhibited poorer OS than the patients in the low-risk group (hazard ratio (HR): 3.22; 95% CI: 0.73–14.24, P = 0.02) ([133]Fig. 2I). Overall, our results demonstrated that the IPM is robust across different molecular levels, platforms and datasets. Fig. S1. [134]Fig. S1 [135]Open in a new tab Representative images of immunostaining at high-risk and low-risk patients. 3.5. Stratification analyses of OS for the IPM according to TP53 status in the TCGA HCC cohort Consistent with the IPM, TP53 status was also significantly related to the prognosis of patients with HCC ([136]Fig. 3A). Stratification analyses were performed to test whether the prognostic value of the IPM was independent of TP53 status. Therefore, patients in the TCGA HCC cohort were divided into two groups according to TP53 status. Stratification analyses suggested that the IPM was significantly related to OS in the TP53^WT and TP53^MUT TCGA HCC cohorts ([137]Fig. 3B and C). In addition, correlation analyses suggested that the risk score was significantly negatively associated with OS in the TP53^WT and TP53^MUT TCGA HCC cohorts ([138]Fig. 3D). Furthermore, univariate and multivariate Cox regression analyses showed that the predictive power of the IPM for the OS of patients with HCC is independent of TP53 status ([139]Fig. 3E). Fig. 3. [140]Fig. 3 [141]Open in a new tab Prognostic analysis of TP53 mutation. (A-C) Kaplan-Meier survival of TP53 status (A), TP53 mutation subgroup (B), and TP53 wild-type subgroup (C). (D) Analysis of the correlation between risk score and survival time according to TP53 status. (E) Univariate and multivariate regression analysis of the relation between the IPM and TP53 status regarding prognostic value. Red indicates no statistical significance, and green indicates statistical significance. (F) Kaplan-Meier survival of the different types of TP53 mutations. (G) Kaplan-Meier survival of the TP53 missense mutation subgroup. (For interpretation of the references to colour in this figure legend, the reader is referred to