Clustering by antigen-presenting genes reveals immune landscapes and predicts response to checkpoint immunotherapy

Abstract

Immune checkpoint blockade (ICB) has demonstrated efficacy by reinvigorating immune cytotoxicity against tumors. However, the mechanisms underlying how ICB induces responses in a subset of patients remain unclear. Using bulk and single-cell transcriptomic cohorts of melanoma patients receiving ICB, we proposed a clustering model based on the expression of an antigen-presenting machinery (APM) signature consisting of 23 genes in a forward-selection manner. We characterized four APM clusters associated with distinct immune characteristics, cancer hallmarks, and patient prognosis in melanoma. The model predicts differential regulation of APM genes during ICB, which shaped ICB responsiveness. Surprisingly, while immunogenically hot tumors with high baseline APM expression prior to treatment are correlated with a better response to ICB than cold tumors with low APM expression, a subset of hot tumors with the highest pre-ICB APM expression fail to upregulate APM expression during treatment. In addition, they undergo immunoediting and display infiltration of exhausted T cells. In comparison, tumors associated with the best patient prognosis demonstrate significant APM upregulation and immune infiltration following ICB. They also show infiltration of tissue-resident memory T cells, shaping prolonged antitumor immunity. Using only pre-treatment transcriptomic data, our model predicts the dynamic APM-mediated tumor-immune interactions in response to ICB and provides insights into the immune escape mechanisms in hot tumors that compromise the ICB efficacy. We highlight the prognostic value of APM expression in predicting immune response in chronic diseases.