Association with immune checkpoint inhibitor efficacy of a 27-gene classifier in renal cell cancer.

Abstract

4575 Background: The 27-gene immuno-oncology (IO) signature that incorporates expression from activated inflammatory cells, cancer associated fibroblasts, and tumor cells to produce a binary classifier has been shown to be associated with efficacy to immune checkpoint inhibitors (ICIs) in breast, lung, and bladder cancers. Here we created clustered heat maps using data from The Cancer Genome Atlas (TCGA) to confirm the classifier function and diagnostic threshold in renal cell carcinoma (RCC), then applied the predefined algorithm to RNAseq data from a community RCC cohort treated with ICI therapy. Methods: Previously, we described the selection of 939 genes from the TCGA breast and lung datasets that comprise mesenchymal (M), mesenchymal stem-like (MSL), and immunodulatory (IM) gene expression patterns centered upon the twenty-seven genes selected for the IO score (AACR, 2021). We created an expression dataset using these genes in clear cell (n = 403) and papillary (n = 203) RCC and used k-means clustering to organize the genes and cases (k=3). We assessed the 27-gene classification of cases by utilizing area under the curve for phenotypic classification and determining the sensitivity and specificity of the previously established threshold compared to optimal accuracy for quantitating the fraction of cases enriched into the IM+ cluster (likely sensitive to ICIs) as opposed to the M or MSL clusters (likely insensitive). Finally, the IO score was evaluated in a small multi-institutional RNAseq dataset of forty-three RCC patients treated with an ICI for which there was definitive one-year progression free survival (PFS) data. Results: The 27-gene IO signature applied to the TCGA sample data had an AUC of 90.3 for stratification of cases into IM+ as opposed to M and MSL clusters while the established threshold for likely sensitive enriched 90% of cases into the appropriate IM cluster as opposed 28% into the M and MSL. Efficacy was defined by PFS. Given this result, the 27-gene IO signature was applied with the predefined threshold to the forty-three ICI treated patients. Patients who had a IO+ score by the 27-gene signature had significantly better one-year PFS compared to patients with a negative IO score (hazard ratio = 0.235, 95% CI = 0.069 - 0.803, p < 0.01). Median PFS was 5.2 months for patients classified as IO score negative versus 8.6 months for those classified as IO score+. Conclusions: The 27-gene IO signature has been validated across multiple tumor types and here in RCC to classify the tumor immune microenvironment without changing the algorithm or threshold. Results demonstrate that the 27-gene classifier has a strong correlation with efficacy of ICI therapy in RCC. This is the fourth tumor type in which the same algorithm has been validated as a predictor of ICI efficacy. These data support this assay as a strong pan-cancer immune system classifier worthy of further prospective study for ICI therapy.