Abstract 4217: Prognostic gene expression signatures of immune responses in the colon cancer microenvironment

Abstract

Abstract Colorectal cancer remains the 2nd leading cause of cancer deaths in the United States. This suggests that traditional prognostic factors are not optimally refined for predicting survival outcomes and guiding therapeutic decisions for some patients. Mounting evidence supports that quantifying the strength and diversity of host immune responses in the tumor microenvironment may improve prognostication and clinical decision-making; however, standard pathological assessment of T cell infiltration is time-consuming and difficult to standardize for clinical utility. The goal of this study was to develop a molecular classifier associated with CRC prognosis based on the expression of 770 immune-related genes measured on the Nanostring (NS) nCounter PanCancer Immune Profiling Panel. This panel includes markers of immune cell types, common cancer antigens, and diverse categories of immune responses (e.g. T cell function, cytokines). We also aimed to assess the validity of combining gene expression data derived from different tissue types (FFPE, fresh frozen) and mRNA profiling platforms (NS, Rosetta/Merck human RSTA Custom Affymetrix 2.0 microarray). FFPE (N=24) and fresh frozen tumor tissues (N=28) from 50 primary stage II colon cancers from the Moffitt Cancer Center Total Cancer Care cohort were profiled using the NS platform, and microarray data were generated on frozen tissues from all patients. Geometric mean-normalized NS data of FFPE and frozen tumor tissues were merged by the ComBat algorithm that adjusted for different RNA source types. 634 (87%) genes in the NS dataset had expression values that positively correlated with those of the microarray data. A 2-way hierarchical cluster analysis of these genes in NS data revealed two clusters of patients with non-overlapping overall survival (OS) curves, but no statistically significant difference due mainly to a lack of events (Log-rank P=0.12; 5-year OS probability=91.3% vs 74.1% for cluster 1 (N=23) and cluster 2 (N=27)). To examine cross-platform predictability of the 2 clusters, a 5-gene classifier was trained on NS data using penalized logistic regression. Applying this classifier to microarray data on the same patient set (N=49) significantly discriminated the clusters (AUC=0.8, P<0.01). Functional annotation of the 5 genes (CD27, CD37, ITGAL, KLRG1 and LAG3) revealed enrichment for T cell receptor signaling, hematopoietic cell lineage, and natural killer cell mediated cytotoxicity (FDR P<0.001). This pilot study provides early evidence that an immune gene expression panel may capture the prognostic value of intratumoral host immune responses. It also supports the feasibility of combining different RNA sources and expression platforms. Expanded future studies that pool across sample types and publicly-available expression datasets are needed to validate the 9-gene classifier from this study and examine its broader prognostic impact. Citation Format: Youngchul Kim, Hannah J. Hoehn, Yunyun Chen, Mollie E. Barnard, Amanda Bloomer, Sean Yoder, Domenico Coppola, Stephanie L. Schmit. Prognostic gene expression signatures of immune responses in the colon cancer microenvironment [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 4217.