Abstract
Metastasis is a formidable challenge in the prognosis of melanoma. Accurately predicting the metastatic potential of non-metastatic melanoma (NMM) and determining effective postoperative adjuvant treatments for inhibiting metastasis remain uncertain. In this study, we conducted comprehensive analyses of melanoma metastases using bulk and single-cell RNA sequencing data, enabling the construction of a metastasis score (MET score) through diverse machine-learning algorithms. The reliability and robustness of the MET score were validated using various in vitro assays and in vivo models. Our findings revealed a distinct molecular landscape in metastatic melanoma characterized by the enrichment of metastasis-related pathways, intricate cell-cell communication, and heightened infiltration of pro-angiogenic tumor-associated macrophages compared to NMM. Importantly, patients in the high MET score group exhibited poorer prognoses and an immunosuppressive microenvironment, featuring increased infiltration of regulatory T cells and decreased infiltration of CD8(+) T cells, compared to the low MET score patient group. Expression of PD-1 was markedly higher in patients with low MET scores. Anti-PD-1 (aPD-1) therapy profoundly affected antitumor immunity activation and metastasis inhibition in these patients. In summary, our study demonstrates the effectiveness of the MET score in predicting melanoma metastatic potential. For patients with low MET scores, aPD-1 therapy may be a potential treatment strategy to inhibit metastasis. Patients with high MET scores may benefit from combination therapies.