Bioinformatics-based screening of key genes for transformation of liver cirrhosis to hepatocellular carcinoma

Genes were screened using the GEO (http://​www.​ncbi.​nlm.​nih.​gov/​geo) database, specifically the GSE89377 series on the GPL6947 platform (Illumina HumanHT12 V3.0 expression beadchip), the GSE17548 series [8] on the GPL570 platform (Affymetrix Human Genome U133 Plus 2.0 Array), the GSE63898 series [9] on the GPL13667 platform (Affymetrix Human Genome U219 Array), and the GSE54236 series [10] on the GPL6480 platform (Agilent-014850 Whole Human Genome Microarray 4x44K G4112F), the basic clinical info of the patients selected is showed in Additional file 1: Table S1. According to the annotation information in each platform, probes were converted into corresponding gene symbols. The GSE89377 dataset contained 40 HCC tissue samples and 12 liver cirrhosis samples, the GSE17548 dataset contained 18 HCC tissue samples and 19 liver cirrhosis samples, the GSE63898 dataset contained 228 HCC tissue samples and 168 liver cirrhosis samples, and the GSE54236 dataset contained 81 HCC tissues samples and 80 liver cirrhosis samples.

GEO2R (https://​www.​ncbi.​nlm.​nih.​gov/​geo/​geo2r/​) was used to identify DEGs between HCC and liver cirrhosis samples. GEO2R is an interactive network tool that allows users to compare two or more datasets in the GEO series to identify DEGs under experimental conditions [11]. Genes without a corresponding gene symbol, and genes with more than one probe set are separately removed, and fold change (FC) > 2 and adjusted p < 0.05 are the threshold criteria for statistical significance. In order to identify significant DEGs, the Venn online tool (http://​bioinformatics.​psb.​ugent.​be/​webtools/​Venn/​) was used to draw a Venn map, and overlapping DEGs were retained for further analysis.

The String online database (http://​string-db.​org Version:11.0) [12] was used to build a PPI network. Analysis of functional interactions between proteins can be helpful for understanding mechanisms related to disease occurrence and development. Herein, a comprehensive Gt score > 0.4 was considered statistically significant. The molecular composite detection plug-in MCODE within Cytoscape [13] was used to cluster the resulting network in order to reveal closely connected regions. The most important module in the PPI network was identified by MCODE. The selection criteria were MCODE score > 5, cut-off value = 2, node cut-off value = 0.2, maximum depth = 100, and k-score = 2.

Correlations between hub genes was analysed which came from TCGA using cBioPortal (http://​www.​cbioportal) [14]. The online database the Wurmbach liver dataset [15] which comes from Oncomine (http://​www.​oncomine.​com) [16] was used to analyse expression levels of key genes under normal, cirrhotic, hepatocellular and hepatocyte dysplasia conditions. Overall survival related to hub genes was analysed using the Kaplan–Meier curve feature of Kaplan–Meier Plotter which includes 364 patients [17]. With the Wurmbach liver dataset, Oncomine was used to analyse differential expression of core genes in different studies, and to analyse the relationships between expression level and liver cancer tumour grade, hepatitis virus infection status, satellites, and vascular invasion, and to identify core genes most closely related to the development of HCC [16]. DEGs closely related to core genes were further screened by cBioPortal [18], and KEGG pathway analysis of genes related to DEGs was performed using DAVID [19]. The resulting pathway map was drawn using KEGG [20]. The relevant positions of corresponding genes in the pathway were coloured red.

DEGs identified in the four microarray datasets (382 in GSE17548, 870 in GSE54236, 1094 in GSE63898, and 210 in GSE89377) were screened after the chip results were normalised. As shown in the Venn map, 58 genes overlapped in the four datasets (Fig. 1a, Additional file 2: Table S2), comprising 12 and 46 up- and down-regulated genes. STRING database screening and PPI network construction were performed, and visualization was carried out using Cytoscape software (Fig. 1b). The PPI network was constructed and significant modules were identified, with 93 edges and 39 nodes in the PPI network. MCODE plugin used to find the top hub genes, the most closely connected module was identified (Fig. 1c); it includes 13 nodes and 66 edges, and genes in this region were upregulated in HCC tissues.

Three hub genes (seed genes) were identified by Cytoscape software, for which the names, abbreviations, and functions are listed in Table 1. Furthermore, there was a significant correlation among the three hub genes (p < 0.05) according to the cBioPortal database (Fig. 2a–c), demonstrating a specific interaction between hub genes. Additionally, the Oncomine database was used to analyse the expression of the hub gene under normal, cirrhosis, hepatocellular carcinoma and hepatocyte dysplasia conditions(Fig. 2d–f). There were no significant differences in CDKN3 between normal, cirrhotic and hepatocyte dysplasia, but expression was increased significantly in HCC (Fig. 2d). Furthermore, there were no significant differences between CYP2C9 and LCAT in normal, liver cirrhosis and hepatocyte dysplasia, but levels were decreased in HCC (Fig. 2e–f). This indicates that the hub gene plays an important role in the development of cirrhosis into HCC. However, it does not play an important role in normal, liver cirrhosis and hepatocyte dysplasia. Survival-related hub genes were analysed using the Kaplan–Meier curve feature within the Kaplan–Meier Plotter database which included 364 cases of hepatocellular carcinoma (Fig. 3). We noted that HCC patients with elevated CDKN3 levels were associated with a decrease in overall survival (p < 0.05) (Fig. 3a). And, patients with high expression of CYP2C9 or LCAT have a higher survival rate(p < 0.05) (Fig. 3b–c). We also noted that in the Wurmbach liver dataset, the mRNA expression level of CDKN3 was associated with tumour grade, hepatitis virus infection status, and satellite and vascular invasion (Fig. 4), but CYP2C9 and LCAT were not associated with tumour development (Additional file 3: Fig. S1 and Additional file 4: Fig. S2). These results suggest that CDKN3 may be a key gene in the transformation of liver cirrhosis to liver cancer, and is closely related to the occurrence and development of HCC.

Meanwhile, Oncomine analysis showed that CDKN3 was significantly up-regulated in anaplastic oligodendrocytoma, leukaemia, hepatocellular carcinoma and sarcoma (Fig. 5).

In order to understand the expression of CDKN3 in different tumours, Oncomine analysis was performed, and the results showed that CDKN3 was significantly elevated in anaplastic oligodendrocytoma, leukaemia, hepatocellular carcinoma and sarcoma (Fig. 5). In order to further analyse the potential mechanism of the influence of CDKN3 on HCC, the co-expression genes of CDKN3 in TCGA HCC transcriptome data were screened by cBioProtal data analysis platform, Pearson and Spearman scores are more than 0.3. The top 20 genes are listed in Table 2. Correlations with expression levels were evaluated by Pearson and Spearman scores. KEGG pathways related to the top 500 related genes were analysed by DAVID online software, and the top 8 hits are shown in Table 3. The results showed that changes in CDKN3 expression mainly affected pathways related to HCC, such as the cell cycle, DNA replication, oocyte meiosis and mismatch repair. We observed that most genes with strong correlations to CDKN3 expression were involved in cell cycle regulation, and these genes were used to generate a KEGG pathway map (Fig. 6), which relevant positions of the corresponding genes in the pathway are coloured red. The results showed that ORC, MCM, Chk1 and 2, and some other genes occupy important positions, mainly located in the S or G2/M phases of the cell cycle, and they regulate the G1 phase of the gene. Thus, CDKN3 mainly inhibits the transition from the G 1 to the S phase, and plays an important role in inhibiting cancer progression.