Abstract 2044: Using tumor immune microenvironment physiologic profiling to tailor immune checkpoint inhibitor diagnostic classification

Abstract

Abstract Introduction: We recently described an approach using TCGA bladder data to validate the tumor-immune microenvironment (TIME) classification function of a 101-gene expression profile that has previously been shown to be associated with immune checkpoint inhibitors (ICI) efficacy in TNBC and NSCLC. This algorithm, which distinguishes inflammatory (IM), immunosuppressive (MSL), and mesenchymal (M) components of the TIME, was then used to test the existing threshold of the related diagnostic, DetermaIO (DTIO), a 27-gene expression signature that assesses efficacy of ICI. Here we apply the same approach to eight tissues to evaluate the current threshold, or identify tissue tailored thresholds. Methods: The TCGA RNA expression data for eight different solid tumor carcinomas were downloaded, and each patient was classified as IM, MSL, or M, based on their correlation to one of three centroids as previously described [1]. The DTIO score was calculated for each patient and used to determine new thresholds by comparing Cohen’s kappa in patients with IM signatures which were called positive, while those with an M or MSL phenotype were called negative. Alternative thresholds were calculated by fitting a smoothing algorithm to the continuous score and determining local peaks of higher thresholds. Thresholds whose kappa scores were in the 80th quantile were evaluated. Results: Each of the eight tissues provided at least three alternate thresholds. Four tissues—colorectal, gastric, renal cell and esophageal cancers had thresholds that did not differ significantly from the previously established threshold. However, head and neck, ovarian, pancreatic, and prostate cancers all had thresholds that were significantly higher and thus captured a lower fraction of patients as likely responders. Conclusions: The association of DTIO with efficacy to ICI therapy has previously been validated in TNBC, NSCLC, bladder, colon, renal cell, and gastric cancers using an established threshold. These results support the previously selected uniform threshold for these cancers and suggest that it is appropriate for esophageal cancer. H&N, ovarian, pancreatic and prostate cancers have lower ICI efficacy rates and with the exception of H&N cancer, have struggled in unstratified studies to achieve adequate efficacy for drug approval. The correlation between tissue specific tailored thresholds, which identify fewer patients as likely responders in these low efficacy rate tumors, suggests that the natural TIME physiology distinguished by tissue specific DTIO thresholds may be an appropriate classifier for ICI efficacy in these tissue types.