Prediction of immunotherapy responsiveness in melanoma through single-cell sequencing-based characterization of the tumor immune microenvironment

Abstract

Immune checkpoint inhibitors (ICB) therapy have emerged as effective treatments for melanomas. However, the response of melanoma patients to ICB has been highly heterogenous. Here, by analyzing integrated scRNA-seq datasets from melanoma patients, we revealed significant differences in the TiME composition between ICB-resistant and responsive tissues, with resistant or responsive tissues characterized by an abundance of myeloid cells and CD8+ T cells or CD4+ T cell predominance, respectively. Among CD4+ T cells, CD4+ CXCL13+ Tfh-like cells were associated with an immunosuppressive phenotype linked to immune escape-related genes and negative regulation of T cell activation. We also develop an immunotherapy response prediction model based on the composition of the immune compartment. Our predictive model was validated using CIBERSORTx on bulk RNA-seq datasets from melanoma patients pre- and post-ICB treatment and showed a better performance than other existing models. Our study presents an effective immunotherapy response prediction model with potential for further translation, as well as underscores the critical role of the TiME in influencing the response of melanomas to immunotherapy.

Keywords

• Immunotherapy
• Machine learning
• Melanoma
• Single-cell sequencing
• Tumor micro-environment