Overexpression of TRIM28 predicts an unfavorable prognosis and promotes the proliferation and migration of hepatocellular carcinoma

Mining high-throughput data sets

We mined HCC-correlated microarray and RNA-sequencing (RNA-seq) data in several databases, including Sequence Read Archive (SRA), ArrayExpress, and Gene Expression Omnibus (GEO). The search terms were as follows: (hepatocellular OR HCC OR hepatic OR liver) AND (tumor OR tumour OR carcinoma OR cancer OR neoplas*OR malignan*). We screened eligible studies according to the following criteria: (1) the subjects were Homo sapiens; (2) studies involving HCC and non-cancerous liver tissues; and (3) essential clinical information was provided. Further, RNA-seq data of HCC and non-HCC samples were downloaded from The Cancer Genome Atlas (TCGA) and Genotype-Tissue Expression (GTEx) databases. We classified the datasets included in the present study into 40 merged datasets according to the platform type using the “sva” algorithm in R. Log(x+1) conversion was applied to normalize the TRIM28 expression data.

Single-cell RNA-seq (scRNA-seq) analysis

ScRNA-seq data of 17,164 liver cancer cells and 35,625 non-cancerous cells were obtained from the GSE151530 dataset to assess gene expression patterns in cancerous and various non-cancerous cells. Then, t-distributed stochastic neighbor embedding (t-SNE) analysis was used for the visualization of high-dimensional data in two dimensions, and t-SNE-1 and t-SNE-2 were both used to demonstrate cell clustering. The t-SNE plots were taken from the Single-cell Atlas in Liver Cancer (scAtlasLC, https://scatlaslc.ccr.cancer.gov) [11].

Collecting TRIM28 protein expression data from public database

After identifying differentially expressed genes (DEGs) at the mRNA level, immunohistochemical staining (IHC) results of TRIM28 were obtained from the Human Protein Atlas (HPA) database (http://www.proteinatlas.org/). The link for downloading proteomics data was https://pdc.cancer.gov [12].

Statistical analysis

Comparisons between two groups were assessed using the student’s t-test, and Mann-Whitney-Wilcoxon test. The statistical results were exhibited in the form of mean(M) ± standard deviation (SD). p-values (p) less than 0.05 were considered significant. The standard mean difference (SMD) was calculated using Stata 12.0 (College Station, TX, USA). IF p<0.05 or I²>50 %, a random-effect model was calculated; otherwise, a fixed-effects model was selected. Begg’s test was performed to evaluate the publication bias.

To assess the potential of TRIM28 for distinguishing between the HCC and non-HCC groups, a summary receiver operating characteristic curve (sROC) was created in Stata 12.0 (College Station, TX, USA). The area under curve (AUC) of sROC was used to determine the capacity of TRIM28 for discriminating HCC from non-HCC. HCC cases with follow-up times no fewer than 30 days were employed to carry out a survival analysis using the “survival” package in R. Kaplan-Meier (K-M) curves with a log-rank test were generated to evaluate the overall survival (OS) status of HCCs with different TRIM28 levels. The Hazard ratio (HR) was used to assess the prognostic value of TRIM28 for the HCC group. The pooled HR was calculated using Stata 12.0 (College Station, TX, USA).

Screening for latent TRIM28-related dysregulated target genes

As mentioned above, TRIM28 is a transcription factor that is associated with immune infiltration in multiple tumors. We retrieved the chromatin-immunoprecipitation coupled with sequencing (ChIP-seq) data of TRIM28 and collected latent target genes and potential binding position from the Cistrome DB database (http://cistrome.org/db/#/). Scores>1.0 were used to filter the TRIM28 putative target genes. Subsequently, differentially expressed genes (DEGs) in the mRNA datasets were calculated via the “limma” package of R. DEGs with a SMD>0 and a lower 95 % confidence interval (CI) >0 were identified as upregulated DEGs (Up-DEGs) of HCC. In addition, we performed Pearson’s correlation analysis to examine the co-expressed genes derived from HCC-related datasets. Genes were considered positively co-expressed with HCC in a minimum of 15 mRNA datasets when the correlation coefficient (r) was greater than or equal to 0.4 and the p was less than 0.05. The latent target genes, Up-DEGs and the positively co-expressed genes were intersected, with genes that appeared three times selected as the candidate target genes for downstream analysis.

Potential molecular mechanism of TRIM28 in HCC

To explore the underlying signaling pathways of TRIM28 potential target genes in HCC, we utilized the “clusterProfiler” package to analyzed and visualized these genes [13]. If adjusted p-value<0.05, the items from the gene ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were identified. Additionally, protein-protein interaction (PPI) networks were constructed utilizing the Search Tool for the Retrieval of Interacting Genes (STRING) database (http://string-db.org/).

The relationship between TRIM28 and immune infiltration levels in HCC

A previous study revealed that immune cell infiltration was closely related to tumor progression [14]. Therefore, we leveraged the TIMER algorithm to calculate patients’ immune scores based on TCGA data. The violin plots were popped out to present the infiltration scores of six types of immune cells in the high-TRIM28 and low-TRIM28 expression groups. We performed the Spearman correlation to investigate the immune correlation of TRIM28 in HCC. Scatter plots were drawn to show the correlation between TRIM28 expression and immune infiltration levels employing the SangerBox platform (http://vip.sangerbox.com/), which is a publicly available website for analyzing TCGA data [15].

Drug sensitivity analysis

Gene expression and drug sensitivity data were obtained from the CellMiner database (http://discover.nci.nih.gov/cellminer/) [16]. When drugs without clinical trials or FDA approval were excluded, we utilized the Pearson algorithm in R to evaluate the associations between TRIM28 expression and drug sensitivity IC50 values, and p<0.05 was the threshold for significance. The “impute”, “limma”, and “ggpubr” algorithms in R were employed for data processing.

Molecular docking

The three-dimensional structure of TRIM28 (database ID: 2RO1) was downloaded from the RCSB Protein Data Bank (http://www.rcsb.org/). The structure of the drug ligand was retrieved from PubChem database. AutoDock Tools 1.5.6 (Scripps Research Institute, USA) were used for the docking analysis. Finally, visualization was performed using PyMOL software 2.2.0. The active ligand on the target was denoted by a threshold lower than −6.0 kcal/mol.

In vitro experiments

Up-regulated TRIM28 expression at both the mRNA and protein levels in HCC

Of the 40 merged datasets, 26 showed that the expression of TRIM28 in the HCC group was higher than in the non-HCC group (p<0.05). Representative cohorts were GPL570, GSE25097-GPL10687, and TCGA_GTEx. Additionally, the findings of the integrative analysis utilizing the random effects model provided further evidence of up-regulated expression of TRIM28 in HCC. (pooled SMD=0.97, 95 %CI: 0.83–1.12, Figure 1A). Based on the funnel plot, no significant publication bias was observed (p=0.258, Figure 1B). The sensitive analysis revealed no significant differences among the enrolled datasets.

Figure 1:

Comprehensive analysis of TRIM28 expression level in HCC and non-HCC groups. (A) Forest plot of pooled standard mean deviation values for TRIM28 in HCC. (B) Begg’s funnel plot of the enrolled studies. No significant publication bias exists (p=0.258).

The scRNA-seq analysis showed the broad existence of up-regulation of TRIM28 in the HCC cells, whereas TRIM28 overexpression rarely occurred in the non-malignant cells except for T cells (Figure 2A). It was further verified that highly expressed TRIM28 in HCC indeed came from the malignant cells. A proteomics analysis indicated that TRIM28 protein expression was highly expressed in the tumor compared to the control (0.2642 ± 0.4597 vs. −0.2652 ± 0.1237, p<0.0001). Based on publicly available immunohistochemical staining, we found that positive TRIM28 staining was conspicuous in the HCC samples but not in the non-HCC samples (Figure 2B). Therefore, TRIM28 was observed to exhibit higher expression levels at both the mRNA and protein levels in HCC.

Figure 2:

(A) Single-cell transcriptomic atlas of malignant cells and various non-malignant cells in liver cancer. (B) The IHC-based protein expression of TRIM28 in HCC tissues and normal liver tissues. All the IHC staining images were obtained from the HPA database (https://www.proteinatlas.org/). IHC, immunohistochemistry.

Clinical potential of TRIM28 in HCC

We generated an sROC curve with an AUC of 0.84 (95 % CI: 0.81–0.87), sensitivity of 0.64 (95 % CI: 0.58–0.70), and specificity of 0.88 (95 % CI: 0.84–0.92), demonstrating the high accuracy of increased TRIM28 expression in discriminating HCC from non-HCC tissues (Figure 3A). We estimated the prognostic capacity of TRIM28 and found that the low TRIM28 group exhibited superior OS compared to the high TRIM28 group for HCC patients based on three microarrays, TCGA, and the proteomics cohort (all p<0.05). Additionally, the increased TRIM28 mRNA levels corresponded to the shortened OS in HCC, as confirmed by the online web-tool, Gene Expression Profiling Interactive Analysis (GEPIA2) (http://gepia.cancer-pku.cn/) (Figure 3B, p<0.05). A forest plot with a pooled HR of 2.36 (95 % CI: 1.81–3.09) demonstrated that high expression of TRIM28 may be a risk factor of HCC (Figure 3C). Additionally, by utilizing the TCGA dataset, we examined the association between TRIM28 mRNA levels and clinicopathological features in individuals with HCC. Notably, as shown in Table 3, TRIM28 mRNA expression showed a significant correlation with age (p=0.0306), grade (p<0.0001), stage (p=0.0002) and pathologic T (p=0.0002), while it did not show significant relevance to other clinical parameters.

Figure 3:

The clinical significance of TRIM28 expression in liver cancer. (A) SROC curve demonstrating performance of TRIM28 in discriminating HCC from non-HCC tissues. (B) The GEPIA database was used to demonstrate the impact of high TRIM28 expression on the prognosis of overall survival in HCC patients. (C) Pooled HR based on five datasets. SROC, summary receiver operating characteristic; AUC, area under the curve; HR, hazard ratio.

Enrichment analysis

Through the Cistrome DB database, we obtained five TRIM28 ChIP-seq datasets, which revealed 4,063 TRIM28 putative targets that were found in a minimum of three datasets. By investigating the links between all genes and TRIM28 across all data sources, including TCGA and gene chips, we identified 8676 Up-DEGs and 1754 positively co-expressed genes. These genes were then overlapped to obtain 629 potential dysregulated target genes related to TRIM28. Subsequently, we conducted GO and KEGG enrichment analyses using these overlapping genes. The KEGG pathway analysis showed enrichment in the spliceosome, RNA transport, and cell cycle pathways (Figure 4A). The results of the GO analysis suggested that TRIM28 played a role in HCC by regulating biological processes including the cell cycle, cell cycle process, as well as chromosome organization. The most significantly enriched GO cell components were nuclear protein containing complex, chromosome, and the catalytic complex. In terms of molecular function, the most enriched GO terms were RNA binding, enzyme binding, and hydrolase activity acting on acid anhydrides (Figure 4B–D). Additionally, to further explore the correlation among these proteins, we constructed PPI networks based on the related genes enriched in the top three clustered pathways (Figure 4E–G). Using Cytoscape 3.7.2, we identified seven hub genes, including SNRPA1, SNRPF, SNRPD1, SF3B2, SNRPB, SNRPE, and EFTUD2. Further analysis using the Cistrome DB database revealed binding peaks of these hub genes at the transcription initiation sites of TRIM28 (Figure 5 and S1).

Figure 4:

Enrichment analyses. (A) The KEGG enrichment analysis. (B) GO enrichment analysis in terms of biological process. (C) GO enrichment analysis in terms of cellular component. (D) GO enrichment analysis in terms of molecular function. (E) PPI in the spliceosome pathway, which is the most significantly clustered KEGG pathway. Seven hub genes (SNRPA1, SNRPF, SNRPD1, SF3B2, SNRPB, SNRPE, and EFTUD2) were identified. (F) PPI in the RNA transport pathway. (G) PPI in the cell cycle pathway. KEGG, Kyoto encyclopedia of genes and genomes; GO, gene ontology; PPI, protein-to-protein internet.

Figure 5:

Peaks of hub genes binding to TRIM28 in ChIP-seq based on the Cistrome DB database. (A) SNRPA1; (B) SNRPF; (C) SNRPD1; (D) SF3B2; (E) SNRPB; (F) SNRPE; (G) EFTUD2. ChIP-seq, Chromatin-immunoprecipitation coupled with sequencing.

The immune relevance of TRIM28

Violin plots revealed significant differences between the high- and low-TRIM28 groups in various immune cell types, including B cells, CD4+ T cells, CD8+ T cells, neutrophils, macrophages, and dendritic cells (p<0.05, Figure 6A). The high TRIM28 group exhibited higher immune scores compared to the low-TRIM28 group for B cells, CD4+ T cells, macrophages, neutrophils, and dendritic cells in HCC patients. As shown in Figure 6B, we observed a positive correlation between TRIM28 expression levels and five immune cells, namely, B cells (p=8.4e-22), CD4+ T cells (p=4.6e-12), macrophages (p=0.01), neutrophils (p=2.9e-8), and dendritic cells (p=3.7e-13), except for CD8+ T cells.

Figure 6:

(A) The relationship between TRIM28 expression and infiltration levels of immune cells; the blue “violin” refers to the low-TRIM28 expression group, while the red “violin” refers to the high-TRIM28 expression group. (B) Relevance between TRIM28 expression with infiltration levels of all the six immune cells.

Correlation between TRIM28 and drug sensitivity

As shown in Table 4, 24 anticancer drugs that showed a significant correlation with TRIM28 expression were extracted. Surprisingly, we observed that TRIM28 expression was highly negatively correlated with the IC50 values of pluripotin (r = −0.309; p=0.018), where lower IC50 values indicated higher drug sensitivity. Further, the results showed that higher TRIM28 expression was linked to enhanced sensitivity of tumor cells to pluripotin.

Molecule dynamics simulation

To investigate the potential of pluripotin as a target for TRIM28, we performed molecular docking between pluripotin and TRIM28. As shown in Figure 7, the docking analysis revealed stable interactions between the target molecules of pluripotin and TRIM28, with a binding energy of −10.46 kcal/mol. This value that was within the activity threshold of less than −6.0 kcal/mol indicating the presence of robust binding forces. These findings suggest that pluripotin holds promise as a potential agent for targeting TRIM28.

Figure 7:

Pluripotin two-dimensional structure and TRIM28 three-dimensional structure molecular docking diagram.

TRIM28 contributes to cell proliferation and migration in HCC

We used siRNA to knock down TRIM28 expression in MHCC97-L and Huh7 cells. The RT-qPCR results suggested that TRIM28 expression was down-regulated by TRIM28-siRNA (Figure 8A and B). As expected, the CCK-8 and transwell migration experiments showed that down-regulation of TRIM28 suppressed the proliferation and migration of liver cancer cells (Figure 8C–F).

Figure 8:

TRIM28 knockdown in human liver cancer cell lines. (A) RT-qPCR verification of the TRIM28 siRNAs transfection efficiency in MHCC97-L cells. (B) RT-qPCR verification of the TRIM28 siRNAs transfection efficiency in Huh7 cells. (C) Line chart of OD450 values for different groups of MHCC97-L cells. (D) Line chart of OD450 values for different groups of Huh7 cells. (E) The effect of TRIM28 knockdown on the migration ability of MHCC97-L cells detected by transwell migration assay. (F) The effect of TRIM28 knockdown on the migration ability of Huh7 cells detected by transwell migration assay. NC, negative control; OD, optical density. *p<0.05, **p<0.01, ***p<0.001.