A spatial architecture-embedding HLA signature to predict clinical response to immunotherapy in renal cell carcinoma

An important challenge in the real-world management of patients with advanced clear-cell renal cell carcinoma (aRCC) is determining who might benefit from immune checkpoint blockade (ICB). Here we performed a comprehensive multiomics mapping of aRCC in the context of ICB treatment, involving discovery analyses in a real-world data cohort followed by validation in independent cohorts. We cross-connected bulk-tumor transcriptomes across >1,000 patients with validations at single-cell and spatial resolutions, revealing a patient-specific crosstalk between proinflammatory tumor-associated macrophages and (pre-)exhausted CD8 + T cells that was distinguished by a human leukocyte antigen repertoire with higher preference for tumoral neoantigens. A cross-omics machine learning pipeline helped derive a new tumor transcriptomic footprint of neoantigen-favoring human leukocyte antigen alleles. This machine learning signature correlated with positive outcome following ICB treatment in both real-world data and independent clinical cohorts. In experiments using the RENCA-tumor mouse model, CD40 agonism combined with PD1 blockade potentiated both proinflammatory tumor-associated macrophages and CD8 + T cells, thereby achieving maximal antitumor efficacy relative to other tested regimens. Thus, we present a new multiomics and spatial map of the immune-community architecture that drives ICB response in patients with aRCC. Multiomics and spatial mapping of tumor samples derived from a real-world cohort of patients with advanced renal cell carcinoma, as well as integration of transcriptomics and human leukocyte antigen genotyping data, provides a machine learning-derived signature of response to immune checkpoint blockade.