Integrated analysis and validation of the TRIM28-H2AX-CDK4 diagnostic model assists to predict the progression of HCC

Upregulated expression of TRIM28 is predictive of a poor prognosis in HCC

To compare the expression levels of TRIMs in HCC, five subsets of microarray data were extracted from the GEO database, including GSE39791 (Platform: GPL10558), GSE36411 (Platform: GPL10558), GSE45267 (Platform: GPL570), GSE69715 (Platform: GPL570) and GSE87630 (Platform: GPL6947) (Figure 1A). Principal Component analysis was then performed, which showed that the quality and feasibility of the data met the criterion for further analyses (Supplementary Figure 1A). By taking the intersection of these five databases, three potential TRIMs, TRIM3, TRIM6 and TRIM28 (P < 0.01) (Figure 1B) were found to be significantly higher in HCC tissues. Data from Oncomine database showed that TRIM6 and TRIM28 expression was significantly higher in HCC compared with that in non-cancerous liver tissues (Supplementary Figure 1B).

Subsequently, we assessed the TRIM3, TRIM6 and TRIM28 expression profile in normal and HCC tissues collected from TCGA and GEPIA2 dataset cohorts, where the results indicated that TRIM6 and TRIM28 expression was significantly increased in HCC tissues compared with that in normal tissues according to the TCGA database (Figure 1C, Supplementary Figure 1C). However, only the expression of TRIM28 was significantly higher in GEPIA2 (P < 0.05). In addition, significant differences were also found among the TRIM28 levels and tumor staging groups (P = 0.000186) according to the GEPIA2 datasets (Figure 1C). Therefore, we then focused on TRIM28 for further analyses.

As shown in Figure 1D, patients with HCC expressing higher levels of TRIM28 were associated with reduced OS, DSS and PFS compared with those with lower TRIM28 expression levels. Subsequently, data containing the required clinical characteristics of 374 patients with HCC were acquired from TCGA. TRIM28 overexpression was found to associate significantly with age, BMI, pathological staging, histological grade, α-foetalprotein (ng/ml) and higher TNM staging. The detailed clinicopathological features are shown in Table 1. These results suggest that TRIM28 can contribute to HCC progression.

Relationship between methylation status and TRIM28 expression in HCC

DNA methylation is as an important mechanism of epigenetic modification that serves an important role in regulating gene expression. Since TRIM28 expression was found to be increased in HCC, we then assessed whether DNA methylation contributed to the regulation of TRIM28 expression. To assess the relationship between methylation status and TRIM28 expression in HCC, we first visualized 12 methylation sites (cg05663122, cg05025162, cg26981251, cg24109975, cg12528394, cg01339029, cg11909976, cg19476058, cg17840453, cg18397137, cg06180363 and cg05678175) in the DNA sequences of TRIM28, which were identified using MethSurv (Supplementary Table 1, Supplementary Figure 2A). We found that there was a negative correlation between TRIM28 expression and the methylation status of three methylation sites, cg11909976, cg19476058 and cg05678175 (Supplementary Table 1). Among these sites, only the methylation level of cg05678175 was significantly decreased in HCC (Supplementary Figure 2A). As shown in Supplementary Figure 2B, we found that higher TRIM28 expression was associated with its hypomethylation. Subsequent survival analyses also revealed that the hypomethylation of cg05678175 was positively associated with the OS of patients with HCC. These results suggest that cg05678175 is a potential functional methylation site on the TRIM28 gene.

Relationship between the immune state and TRIM28 expression in HCC

Alterations of immune states have been suggested to be a typical feature in patients with HCC. Since TRIM28 was shown to be associated with the prognosis of HCC, we next investigated whether this phenomenon was at least partially attributed to changes to immune states. Using TCGA, we obtained the expression profiles of a series of genes associated with immune checkpoints that were found to be associated with TRIM28. As shown in Supplementary Table 2, programmed cell death protein-1 (PD-1), cytotoxic T lymphocyte antigen 4 (CTLA-4), T cell immunoglobulin and mucin domain-containing-3 (TIM3), T cell immunoreceptor with Ig and ITIM domains (TIGIT), CD8B, CD27, CD96, CD40 ligand (CD40LG) and TNF receptor superfamily member 4 (TNFRSF4) expression levels were upregulated, whilst adenosine A2a receptor (ADORA2A), CD33, killer cell lectin like receptor C1 (KLRC1), lymphocyte Activating 3 (LAG3), programmed death-ligand (PD-L1), Siglec 7 and Siglec 9 expression was downregulated in HCC. The values of these genes in predicting the various different survival outcome parameters are also shown in Supplementary Table 2. The general landscape showing the relationship between TRIM28 expression and immune checkpoint gene alterations in HCC was concisely visualized, including fusion, amplification, deep deletion, truncating and missense mutations (Supplementary Figure 3). The detailed relationship between TRIM28 expression and each representative immune checkpoint is individually presented in Supplementary Table 3. Of note, alterations in TRIM28 expression showed statistically significant co-occurrence instead of mutual exclusivity with a substantial quantity of immune checkpoints, such as Siglec 7, CD33 and Siglec 9.

Next, we conducted a series of gene expression analyses on the TIMER 2.0 database to evaluate the potential relationship between TRIM28 expression and immune infiltration in HCC. The analyses showed that alterations in TRIM28 copy numbers were associated with the degree of infiltration by several immune cell types, including CD4+ T cells, B cells, neutrophil and myeloid dendritic cells in HCC (Figure 2A). In addition, we further analysed the impact of immune infiltration on the clinical prognosis of patients with HCC, which showed that increased filtration by CD4+ T cells, macrophages and neutrophils were associated with a poorer prognosis in patients with HCC. Their survival period was shown to be <24 months (Figure 2B).

We next evaluated the correlation between TRIM28 expression and macrophage, T cells, B cells, neutrophil or dendritic cell infiltration using TIMER and TCGA. The results showed that the expression of TRIM28 was positively correlated with macrophage, T cells and B cell infiltration according to both TCGA and TIMER databases (Figure 2C, 2D). To further explore the relationship between TRIM28 and immune-infiltrating cells, we also validated the association of TRIM28 with a set of immune biomarkers representing seven different immune cell types. CD8A, CD8B, CD19, CD79A, interferon regulatory factor 5, integrin subunit α M, human leukocyte antigen (HLA)-DPB1, HLA-DQB1, HLA-DRA, HLA-DPA1, CD1C, neuropilin 1 and integrin subunit αX were found to be significantly correlated with TRIM28 expression in HCC (Table 2). We also performed enrichment analysis in the TCGA database using seGSEA, which found that increased B cell infiltration was observed in HCC with higher levels of TRIM28 expression (Figure 2E). Taken together, tumor-infiltrating immune cells, especially B cells, may serve influential roles in the clinical outcomes of patients with HCC with different levels of TRIM28 expression.

TRIM28/H2AX/CDK4 axis is involved in the poor prognosis of patients with HCC

To predict the potential function of TRIM28, we performed interaction and correlation analyses between TRIM28 expression and other candidate genes in HCC using the cBioPortal and GEPIA2 databases. We conducted a joint analysis in the high-TRIM28 and low-TRIM28 expression groups. The top 50 genes showing positive association with TRIM28 expression were presented in a heatmap and a circos plot as three clusters using Kmeans clustering (Figure 3A, 3B and Supplementary Figure 4A). We subsequently found four candidate genes, chaperonin containing TCP1 subunit (CCT)2, CCT7, H2AX and SWI/SNF-related, matrix-associated, actin dependent regulator of chromatin (SMARC), subfamily A, member 4, from the intersection of the top 50 genes (Figure 3C).

As a result, we analyzed the expression profile of CCT2, CCT7, H2AX and SMARCA4 from the GSE39791, GSE36411, GSE45267, GSE69715 and GSE87630 datasets. H2AX and SMARCA4 expression was significantly higher according to the GSE45267 dataset (Figure 4A), but there were no differences in the GSE39791, GSE36411, GSE69715 and GSE87630 datasets (Supplementary Figure 4A). In order to find the target gene of TRIM28 more accurately, we also analyzed the expression data of CCT2, CCT7, H2AX and SMARCA4 in a dataset obtained from TCGA, which found that only H2AX expression was upregulated in HCC (Figure 4A). Subsequently, we also confirmed that there was a strongly positive correlation between TRIM28 and H2AX expression in HCC according to the TCGA dataset (Figure 4B). To evaluate the value of TRIM28 or H2AX expression for predicting the prognosis of patients with cancer, the association between TRIM28 or H2AX expression and OS, PFS and DSS were analyzed in the same Kaplan Meier-Plotter cohort. Higher TRIM28 or H2AX expression was significantly associated with decreased OS, RFS, PFS and DSS in patients with HCC (Figure 1D, Supplementary Figure 4B). These results suggest that the TRIM28/H2AX axis is involved in HCC tumor progression to result in the poor prognosis of patients with HCC.

To further unravel the downstream signaling pathways of TRIM28 and H2AX, we used cBioPortal and found that the p53, TGF-β, Wnt, mitogen-activated protein kinase and oxytocin receptor-mediated signaling pathways were associated with TRIM28 and H2AX expression (Supplementary Figure 5A). By comparing the top five genes that were associated with TRIM28 and H2AX expression, we found that four genes, CDK4, SMARCD1, SMARCB1 and Bcl-2-associated X, apoptosis regulator (BAX), were potentially correlated with TRIM28 and H2AX expression (Supplementary Figure 5B). We first found that high levels of CDK4, SMARCD1, SMARCB1 and BAX expression were associated with a reduced OS in patients with HCC (Supplementary Figure 6A). Using Spearman’s correlation analysis, we noted that there was a strong correlation between TRIM28 or H2AX expression and the expression these four candidate genes in HCC according to TCGA (Supplementary Figure 6B). We then analyzed the association between the expression of these genes and patient prognosis using univariate or multivariate Cox analysis in patients with HCC (Tables 3.1–3.3 and Supplementary Tables 4–15). Univariate Cox analysis demonstrated that higher expression levels of CDK4, SMARCD1, SMARCB1 and BAX were significantly associated with reduced OS. However, only the expression of three genes (CDK4, SMARCD1 and SMARCB1) were significantly associated with DSS and only CDK4 and SMARCD1 were significantly associated with PFS (Supplementary Tables 1–12). Multivariate Cox analysis confirmed that CDK4 overexpression was an independent risk factor for reduced survival in patients with HCC (Tables 3.1–3.3). Therefore, we speculated that the TRIM28/H2AX/CDK4 axis may serve a crucial role in HCC.

Construction and validation of the TRIM28/H2AX/CDK4 diagnostic model in HCC

We conducted Receiver Operating Characteristic (ROC) curve analysis of TRIM28, H2AX and CDK4 to evaluate their diagnostic value in patients with HCC. The Area under the receiver operating characteristic curve (AUC) was calculated to be 0.945, 0.945 and 0.940 for TRIM28, H2AX and CDK4, respectively (Figure 5A). Subgroup analysis also demonstrated the diagnostic value of TRIM28/H2AX/CDK4 in various important clinical characteristics of HCC, with AUC values of 0.925, 0.932 and 0.929 for T1, respectively; 0.965, 0.959 and 0.952 for T2/3/4, respectively; 0.926, 0.938 and 0.926 for stage I, respectively; and 0.963, 0.954 and 0.949 for stage II/III/IV, respectively (Figure 5A).

Based on clinicopathological factors, we constructed a prognostic nomogram with the aim of providing a quantitative analytical tool that can be used to predict the 1-, 2-, or 4-year OS, DSS and PFS rate of individual patients by combining the expression levels of TRIM28/H2AX/CDK4 in the entire TCGA cohort. The calibration curves of the nomogram showed consistencies between the predicted and observed OS/DSS/PFS probabilities (Figure 5B).

To comprehensively evaluate the clinical significance of TRIM28, H2AX and CDK4 expression in patients with HCC, we analyzed the expression levels of TRIM28, H2AX and CDK4 in a set of HCC tissue microarrays using IHC staining. In 90 pairs of primary HCC tissue samples and corresponding adjacent non-cancerous tissue samples, the staining intensities of TRIM28, H2AX and CDK4 were found to be significantly higher in the HCC tissues (Figure 6A, 6B). Additionally, we found that TRIM28 is positively correlated with H2AX and CDK4 expression (Figure 6C), which is similar to the results obtained using data from TCGA. In these 90 HCC cases containing the clinicopathological information, the staining intensities of TRIM28, H2AX and CDK4 in tissues from late-stage HCC (stage III) were stronger compared with those from early-stage HCC (stage I and II). In addition, their expression in tissues at stages II and III according to American Joint Committee on Cancer were higher compared with those at stage I. Their expression levels were also found to associate with T, N and M staging (Supplementary Tables 16–18). Univariate analysis showed that TRIM28, H2AX and CDK4 expression were significant independent factors for OS and PFS (Supplementary Tables 19 and 20). Subsequently, multivariate analysis revealed that the expression of CDK4 was an independent predictor for OS and PFS in HCC (Supplementary Tables 16 and 17). Kaplan-Meier survival analysis suggested that higher expression levels of TRIM28, H2AX or CDK4 expression were associated with poorer prognoses in patients with HCC (Supplementary Figure 7A, 7B). The high expression levels of all three proteins combined was associated with reduced OS compared with higher expression levels of either of the three proteins alone, whereas higher PFS rates were observed in patients that did not express high levels of these three proteins (Figure 6D (a, b)). As shown in Table 4, high expression levels of TRIM28, H2AX or CDK4 were significantly associated with tumor recurrence in an independent cohort of HCC. To detect the diagnostic value of TRIM28, H2AX and CDK4 in patients with HCC, ROC analysis was performed using the “roc” function in the pROC R package (Version 1.17.0.1) and “ggplot2” R package (Version 3.3.3). In the 90 HCC cases, the AUC was calculated to be 0.921, 0.795 and 0.890, respectively (Figure 6E (a)), which was consistent with the results from TCGA. The AUC of the integrated analysis of these three targets was 0.933, which was higher compared with either of the three proteins alone (Figure 6E (b)). These results implicate the potential roles of TRIM28, H2AX and CDK4 in evaluating the prognosis of patients with HCC.

Gene expression analysis

TRIM expression data were obtained using GSE36411, GSE39791, GSE45267, GSE69715, and GSE87630, a computational method that is used to determine whether a previously defined set of genes show statistically significant differences in expression levels [38]. The different TRIM3, TRIM6 and TRIM28 expression datasets were available from TCGA data portal (http://www.cancer.gov/tcga, v30.0; September 23, 2021), GEPIA2 web server (http://gepia2.cancer-pku.cn/#analysis) and Oncomine gene expression array datasets. Additionally, we obtained violin plots showing TRIM3, TRIM6 and TRIM28 expression among the different pathological stages (stages I–IV) of all TCGA tumors using the “Pathological Stage Plot” module of GEPIA2.

DNA methylation analysis

We explored the association between TRIM28 and their methylation state using the MethSurv tool (https://biit.cs.ut.ee/methsurv/). Spearman correlation coefficients were applied to estimate the correlation between TRIM28 expression and their gene methylation levels in TCGA-HCC.

Immune checkpoints and infiltration analysis

Immune checkpoint data were derived from PubMed. Kaplan-Meier survival analysis was performed using the selected gene expression dataset from the HCC patient samples using Kaplan-Meier plotter (https://www.kmplot.com). The significance of co-occurrence or mutual exclusivity was calculated in cBioPortal web (https://www.cbioportal.org/). Q < 0.05 was considered to indicate a statistically significant result. To calculate the correlation of the expression levels of TRIM28 with tumor-infiltrating immune cells, Tumor Immune Estimation Resource (TIMER 2.0, https://cistrome.shinyapps.io/timer/) was used. We calculated the association between TRIM28 expression and a variety of tumor-infiltrating immune cell types, including CD8+ T cells, CD4+ T cells, B cells, macrophages, neutrophils and myeloid dendritic cells. In addition, the association between cumulative survival and TRIM28 copy numbers in HCC was also calculated in both TIMER 2.0 [39] and TCGA [40, 41].

Gene enrichment analysis

We searched the cBioPortal website by using protein name ‘TRIM28’ and ‘Homo sapiens’ as the search terms. Parameters (evidence; experiments; low confidence 0.150 and ≤50 interactors) were set to obtain the potential TRIM28-binding proteins. In addition, GEPIA2 was used to obtain the top 50 genes associated with TRIM28 or H2AX based on the TCGA tumor and normal tissue datasets. A single-gene co-expression heat map was then made to show the differentially expressed genes (cBioPortal and GEPIA2) in the low- and high-expression tissue groups. A circos plot was also produced to show the Spearman correlation between TRIM28 expression and other mRNAs in HCC.

Survival analysis

Survival plots analyzing TRIM28 expression in HCC were obtained using the ‘Survival Analysis’ module in TCGA. We then used the ‘Survival Map’ module of The Kaplan-Meier Plotter [42] to obtain the association between overall survival (OS), relapse-free survival (RFS), progression-free survival (PFS) or disease-specific survival DSS and TRIM28 or H2AX expression. Log-rank test was used to assess statistical significance.

Tissue microarrays

Tissue microarrays (cat. no. HLivH180Su16) were purchased from Shanghai Outdo Biotech Co., Ltd., to detect the expression of TRIM28, H2AX and CDK4 gene.

Immunohistochemical analysis (IHC)

Briefly, the microarrays were dewaxed in xylene and rehydrated in a descending ethanol gradient and PBS. Endogenous peroxidase was inactivated using 3% H₂O₂ and antigen retrieval was performed using the Tris-EDTA antigen retrieval buffer for TRIM28, H2AX and CDK4 staining. Sections were blocked with 10% normal goat serum at room temperature for 15 min and incubated with primary antibodies against TRIM28 (Santa Cruz Biotechnology, Inc.), H2AX (ab124781, Abcam) or CDK4 (ab108357, Abcam) at 4°C overnight. The sections were then incubated with corresponding secondary antibodies conjugated with horseradish peroxidase at room temperature for 30 min. The staining was visualized using a DAB kit (Beijing Zhongshan Golden Bridge Biotechnology Co., Ltd.; OriGene Technologies, Inc.). All sections were examined and scored independently by two investigators in a double-blinded manner. Staining and scoring were performed as previously described [43].

Statistical analysis

The expression level of TRIM28 in patients with HCC was evaluated using scatter/box/stage plots GraphPad Prism 7 Software. A Mann-Whitney U test and Wilcoxon signed-rank sum test were used to compare the difference between two groups of data, while a Kruskal-Walli’s test was utilized to evaluate the difference among ≥ three groups. Further exploration was analyzed by post-hoc Bonferroni test or adjusted Bonferroni correction for multiple comparisons. A time-dependent ROC curve was used to calculate the diagnostic value of TRIM28, H2AX or CDK4 gene expression levels. Categorical variables are expressed in the form of quantity (percentage), and a χ² test or Fisher’s exact test were used for comparison. Cox proportional-hazards model was used for univariate and multivariate survival analyses. Volcano plots were plotted using the Limma statistical package (version 3.52.2). Nomograms were constructed using the R package (Version 3.6.3) and calibration plots were generated to evaluate the performance of the nomogram. P < 0.05 was considered to indicate a statistically different difference.

Limitation

In the present study, we have determined the role of TRIM28/H2AX/CDK4 associated with immune status in HCC. And the expression and prognosis of TRIM28, H2AX and CDK4 were validated further in an independent cohort of patients with HCC via IHC. However, there is still a lack of in vivo study for the expression of these three genes and the interaction between TRIM28, H2AX and CDK4.

Availability of data and materials

The data and material can be found from correspondence author.