The association of PRF1 with SKCM prognosis
The primary SKCM patients had shorter overall survival time the metastatic SKCM patients (P < 0.001, Supplementary table 1) (Figure 1A) with PRF1 expression upregulated in metastatic samples (Figure 1B). Patients in the PRF1 high expression group showed a better prognosis than that in the low expression group (Figure 1C). As one of the key marker genes of immune cytolytic activity (CYT), the PRF1 expression exhibited a strong positive correlation with lymphocyte infiltration score (Figure 1D). These results reflected a tight association between PRF1 and the metastasis and prognosis of SKCM, implying the remarkable varied CYT between primary and metastatic SKCMs.
The CYT-related biological processes in primary and metastatic SKCM
We identified 518 and 1332 co-expression genes in primary and metastatic SKCMs, of which 502 genes were shared by both (Supplementary figure 1), accounted 96.91% and 37.69% of the total co-expression genes respectively. Functional enrichment analysis revealed that both the primary and metastatic co-expression genes were enriched in the processes of T cell activation, lymphocyte activation, etc. (Figure 2A). For primary specific co-expression genes, they were mainly related to tumor necrosis, while immune response activation and inflammation response were the major enriched processes for metastatic specific co-expression genes (Figure 2B). The results of GSVA enrichment indicated that higher scores of LIPASE ACTIVATOR ACTIVITY, POSITIVE REGULATION OF PROTEIN LOCALIZATION TO CELL SURFACE and CELL CYCLE G2 M PHASE TRANSITION were enriched in primary SKCMs, whereas TOLERANCE INDUCTION, REGULATION OF B CELL PROLIFERATION and REGULATION OF ANTIGEN RECEPTOR MEDIATED SIGNALING PATHWAY obtained higher scores in metastatic SKCMs (Figure 2C).
Identification of C3 as the hub gene in PPI network
Since extremely few primary specific co-expression genes (n=16) were identified, the metastatic specific co-expression genes (n=830) were selected to construct a protein-protein interaction network based on the StringDB database. The network consists of 948 edges and 257 nodes (Figure 3A). The topological parameters of PPI network estimated by the NetworkAnalyzer plugin in Cytoscape indicated that C3 gene had the largest degree, the number of node connections (Figure 3B). The betweenness centrality of C3, which reflects the amount of control the node exerts over the interactions of other nodes in the network, ranked second (Figure 3C). In addition, the other four parameters including closeness centrality, clustering efficiency, average shortest path length and stress, also suggested the key role of C3 in the network (Supplementary figure 2).
The association of PRF1 with C3
The expression of the potential hub gene, C3, was found upregulated in the metastatic SKCMs (Figure 4A), and its high expression group exhibited a better prognosis than the low expression group (Figure 4B), suggesting that C3 might be a favorable prognostic predictor. Besides, the expression of PRF1 and C3 presented a moderate correlation (Figure 4C). Then, we divided SKCM samples into four groups, PRF1-high/C3-high, PRF1-high/C3-low, PRF1-low/C3-high, and PRF1-low/C3-low according to the median values of PRF1 and C3 expressions. Survival analysis indicated that PRF1-high/C3-high group showed the best prognosis and PRF1-L/C3-L group had the worst prognosis (Figure 4D). Additionally, most of the PRF1-L/C3-L group samples were primary SKCMs, while the PRF1-H/C3-H group samples were dominated by metastatic SKCMs (Supplementary Figure 3).
Validation of PRF1 and C3 in independent set
In the TCGA SKCM dataset, the potential synergy between PRF1 and C3 was observed that may contribute to a good prognosis of this disease. Then, we verified this relationship in an independent GEO dataset. Due to their small sample sizes, three GEO datasets including GSE22155 (n= 29), GSE54467 (n= 70) and GSE65904 (n= 206) were integrated into one set by employing Z-score method. The combined dataset showed no batch effect after Z-score normalization (Supplementary figure 4). The expression of PRF1 was positively associated with C3 in the validation set (Figure 5A), which was in line with that in TCGA dataset. Four groups divided by median values of PRF1 and C3 expressions also presented a significant difference in prognosis, with PRF1-H/C3-H group having the best prognosis and PRF1-L/C3-L group the worst prognosis (Figure 5B).
Over-expression of PRF1 in melanoma cell lines
The expression profiles of Cancer Cell Line Encyclopedia (CCLE) revealed that PRF1 and C3 were almost not expressed in SKCM cell lines. As a result, we attempted to overexpress PRF1 on the lung metastasis cell line, MALME-3M, and quantify the C3 expression to investigate their possible relationship. A recombinant vector containing the full fragment of PRF1 was first constructed based on the recombination clone technology (Supplementary figure 5). Then, the empty vector, 0.1ug and 1ug recombinant vector were transfected to MALME-3M cells. The relative expressions of PRF1 and C3 were quantified by qPCR with GAPDH as the control gene. Results indicated that the PRF1 expression was significantly up-regulated in cells transfected by recombinant vectors, and the expression of C3 was also increased (Figure 6A). The relative expression values of both were associated with the concentration of transfected recombinant vectors (Supplementary Table 5). Moreover, the relative expressions of PRF1 and C3 also showed a strong positive correlation (Figure 6B).
The association of PRF1/C3 with immunotherapy
PRF1 is well-known as the marker of immune cytolytic activity, and C3 is related to the inflammatory response of complement system. In the expression profiles of melanoma patients received adoptive T cell therapy, we investigated the expressions of PRF1 and C3 in four immune response groups including complete responders (CR, n=5), partial responders (PR, n=5), stable disease (SD, n=10), and progressive disease (PD, n=5). The PD group showed the lowest expressions of PRF1 and C3 in comparison with the other three groups (Figure 7A). A strong correlation between PRF1 and C3 was observed among the four group samples (Figure 7B).