The melanocytic transcriptomic state is independently associated with poor overall survival in long term follow up of patients with metastatic melanoma

Introduction: Advanced cutaneous melanoma shows substantial heterogeneity in clinical outcomes, even among patients classified within the same clinical stage. Understanding the molecular drivers underlying this heterogeneity is critical for advancing treatment strategies. Previous studies have identified distinct melanoma transcriptomic states - Tirosh et al. and Balderson et al. have agreed on a four-subtype model defining “Undifferentiated/AP1”, “Neural Crest/AP2”, “Transitory”, and “Melanocytic” states. However, their biological and clinical relevance remains unclear. Here we profiled in-transit melanoma (ITM) using digital spatial RNA profiling to associate melanoma transcriptomic states with overall survival (OS) and acral melanoma (AM) status. Methods: Digital spatial profiling (Nanostring GeoMx Whole Transcriptome Atlas) was performed across a tissue microarray constructed from patients with ITM diagnosed from 1990-2020. After filtering poor quality areas of interest (AOIs), we processed the data using noise correction and quantile normalization. We applied Principal Component Analysis (PCA) to define melanoma subtype signatures by ranking genes contributing to each of the first four PCs. We performed gene set enrichment analysis using GSEA with significance defined as an adjusted p-value < 0.05. Association between gene or gene-set expression and overall survival was evaluated using optimal expression cutoffs to define high and low groups, with a minimum group size of 20% to avoid overfitting. Cox proportional hazards models were then used to assess survival differences between groups. Results: We analyzed a cohort of 84 patients (116 AOIs) with ITM passing spatial transcriptomic QC. PCA of transcriptional profiles revealed distinct transcriptional states. The PC1 axis differentiated Transitory from Undifferentiated melanoma, PC2 reflected immune cell infiltration, PC3 corresponded to stromal cells and Neural crest-like melanoma, and PC4 associated with Melanocytic melanoma. We derived gene sets from the PCA results and associated the expression of each gene set with OS. Across a cohort of treatment-naïve ITM, high expression of the melanocytic state conferred a median overall survival difference of 7.59 years and independently associated with poor survival in multivariate analysis. AMs showed higher melanocytic state gene expression compared to non-acral cases. These findings were validated in external datasets, supporting that the melanocytic state predicts poor prognosis. Conclusion: The melanocytic transcriptional state was independently associated with worse overall survival in patients with metastatic melanoma and was enriched in acral melanoma, suggesting that assessment of the melanocytic state may have value for clinical risk stratification.