Recognition of molecular clusters and a novel prognostic signature based on natural killer cell-related genes in skin cutaneous melanoma.

Background: Skin cutaneous melanoma (SKCM) is the third most common type of cutaneous malignant tumor with a poor prognosis. This research aimed to recognize molecular clusters and develop a novel prognostic signature based on natural killer (NK) cell-related genes (NKCRGs) in SKCM.

Methods: The data were obtained from public databases, including ImmPort, TCGA, GEO, GTEx and GEPIA2. The crucial NKCRGs in SKCM were determined by using a Venn diagram to intersect NKCRGs, differentially expressed genes and prognosis-related genes. The "clusterProfiler" software was employed to perform KEGG and GO analyses of crucial NKCRGs. The molecular subtypes were recognized based on crucial NKCRGs by consensus cluster analysis, and Kaplan-Meier survival curves of samples in different subtypes were performed by the "survival" package. Tumor microenvironment, drug sensitivity and somatic mutation analyses were conducted among different subtypes. A prognostic signature was constructed based on crucial NKCRGs by multiple machine learning algorithms. The core NKCRGs were identified by uni- and multi-variate Cox analyses, quantitative real-time PCR experiment, overall survival, immune cell infiltration, single-cell RNA sequencing and pan-cancer analyses.

Results: 32 crucial NKCRGs were identified in SKCM, and KEGG and GO analyses exhibited that these crucial NKCRGs were primarily related to NK cell-mediated cytotoxicity and immune system process. Two distinct clusters (C1 and C2) in TCGA-SKCM were recognized based on 32 crucial NKCRGs. Compared with C1, C2 presented higher expression levels of 32 crucial NKCRGs and higher overall survival (Log-rank, p < 0.0001). There were significant disparities between two clusters in both drug sensitivity and tumor microenvironment. TTN (78.7%) and MUC16 (72.7%) genes exhibited the highest mutation frequency and the RTK-RAS pathway had the highest proportion of affected samples in C1 and C2. A 12-NKCRG optimal prognostic signature was constructed by 13 combinations of 7 machine learning algorithms utilizing 32 crucial NKCRGs. Two core NKCRGs, CD247 and KIR2DL4, were identified in SKCM.

Conclusion: This research demonstrated a novel molecular classification and prognostic signature based on NKCRGs in SKCM, which might be used to forecast the prognosis of SKCM and assist clinicians in making therapeutic strategies, and our results suggested that CD247 and KIR2DL4 might be valuable prognostic biomarkers and potential therapeutic targets for SKCM patients.