Abstract 4468: Cross-species analysis and immunophenotyping using of a focused panel of immune-responsive genes

Immunotherapy has transformed the landscape of cancer therapy, unfortunately, not all patients can reap the benefits and identifying those patients that may benefit remains challenging. It is the consensus that patients with inflamed tumors are more likely to benefit from immunotherapies targeting immune checkpoints, and therapies aiming to stimulate or boost T cell anti-tumor immunity and modulate tumor inflammation are being investigated. Determining which type of immune therapy to administer depends largely on the patients immunophenotype. Various methods aimed at determining a tumor's immune profile are emerging or are available, of these, immune profiling by gene expression analysis is common. Several platforms are available, however, methods to analyze results vary. We previously used a transgenic mouse model of prostate cancer to identify target genes and developed a focused panel of immune-related genes. In addition, we designed a scoring system to profile the tumor's immune phenotype and assess tumor immune responses. Our aim is to extend the applicability of this analysis model to any platform capable of measuring these gene targets. As proof of concept for our analytical approach, we extended our assessment to human cancer by defining cancer specific immune profiles and assessing their clinical relevance. Our focused panel consisted of 115 genes (six reference/normalization genes and 109 immune-related genes) and our scoring system was based on an enrichment algorithm using various immune cell-specific signatures and immune related processes. We evaluated the robustness of this immunophenotyping approach using mRNA data from a cohort consisting of 8477 patients spanning 28 cancer types from The Cancer Genome Atlas (TCGA) Research Network. Our computational pipeline utilized unsupervised and machine learning approaches to identify and evaluate phenotype clusters from gene expression or immune signatures. In this study, Louvain clustering outperformed traditional hierarchical clustering. Using Louvain clustering, we identified distinct clusters from which immunophenotypes could be defined and compared these to various molecular and clinical features. Here, we summarize our findings and show that a minimal gene set effectively identified cancer immunophenotypes. This cross-species analysis corroborates our previous observation and paves the way for further development of our analysis model. Citation Format: Marco A. De Velasco, Yurie Kura, Kazuko S