Search results and study characteristics
According to the inclusion criteria, three GEO datasets and TCGA dataset were obtained in our study: GSE15605, GSE46517, GSE7553 and TCGA skin cutaneous melanoma data. 1078, 407, 892 and 2148 DEGs from the expression profile datasets GSE15605, GSE46517, GSE7553 and TCGA dataset were extracted, respectively. 160 consistently expressed genes were identified by Venn analysis (Figure 1). Among them, 144 genes were up-regulated while 16 were down-regulated compared to normal skin tissue (Table 1).
Gene ontology analysis
Gene ontology describes gene function and relationships between these concepts. P < 0.01 was used as the cut-off criterion. DEGs were classified into the biological process: pathways and larger processes made up of the activities of multiple gene products. After GO enrichment analysis, we found that 160 DEGs were enriched in 17 GO terms (biological process). Among them, the most enriched GO terms were epidermis development (p=1.88E-17), keratinocyte differentiation (p = 9.90E-10), keratinization (p = 3.40E-06) and establishment of skin barrier (p = 1.49E-05) (Figure 2).
Pathway enrichment analysis
Pathway enrichment analysis was carried out by online websites of KEGG, a database was applied to assign sets of DEGs to specific pathways. p < 0.05 was used as the cut-off criterion. After pathway enrichment analysis, we found that 160 DEGs were enriched in 11 pathways. Among them, DEGs were mainly enriched in the pathways in cancer (p = 0.03), transcriptional misregulation in cancer (p = 0.01), Rap1 signaling pathway (p = 0.02) and Ras signaling pathway (p = 0.03) (Figure 3).
PPI network analysis
We obtained a PPI network from STRING to describe protein interactions (Figure 4). Based on the information obtained from the STRING database, a PPI framework with 160 nodes and 385 edges was generated, and its local clustering coefficient was 0.45. The results of computed hub genes were shown in the Table 2, including LOR, FLG, KRT5, CDSN, DSG1, DSG3, KRT1, IVL and EGFR.
Module analysis of the PPI network
To study and identify the function of the overlapping DEGs in detail, cluster analysis of the PPI network was conducted based on the ClusterONE Cytoscape plugin, an important tool for the analysis of densely connected and possibly overlapping regions within the Cytoscape network, which would contribute to the classification of protein network and relevant analysis. There were a total of 23 functional modules given and the most significant module (node = 29, density = 0.4335, p-value = 1.01E-8, Figure 5A) was selected for further analysis of functions and pathways to deeply understand the melanoma progression. To further verify the accuracy of this inference, the module genes were submitted into DAVID to perform the KEGG pathway enrichment analysis. The results showed that they were significantly enriched in the renin secretion signaling pathway, epidermal development, keratinocyte differentiation, peptide cross-linking, keratinization and single biological cell adhesion, among other processes shown, p < 0.05 (Figure 5B).
Hub genes expression level and survival analysis
Using UALCAN, we verified gene expression level of hub genes in 1 normal tissue, 104 primary tissues, and 368 metastasis tissues from TCGA database. Through this analysis, we found that LOR, FLG, KRT5, CDSN, DSG1, DSG3, KRT1 and IVL were closely related to the metastasis of melanoma (p < 0.01) (Figure 6A–I). Kaplan–Meier survival analyses showed that FLG, KRT5, DSG1, DSG3, and IVL expression levels were significantly associated with melanoma patient survival (p < 0.01) (Figure 6J–R).
Detection of hub genes related protein expression by immunohistochemistry
In order to determine the expression of hub genes in human melanoma, we used immunohistochemical methods to detect the expression of the protein corresponding to hub genes. We detected the expression level of hub genes in 63 pairs of melanoma specimens (melanoma and adjacent normal tissue) by immunohistochemistry (Figure 7). The results showed that the expressions of FLG, DSG1, DSG3, IVL and EGFR were considerably higher than those of adjacent normal tissues (P <0.05) (Figure 7B, 7E, 7F, 7H and 7I).