P1.24-09 Spatially Defined Immune Response Signatures in 5μm Tumor Section of Resected T1-2N0M0 Lung Cancer Predict Clinical Outcome

Abstract

<h2>Introduction</h2><p>The risk of recurrence in early-stage (T1-2N0M0) non-small cell lung cancer (NSCLC) is 20-50%. Although the immune system contributes to essential processes in tumor growth and metastasis, the effects of intratumoral immune signatures on recurrence risk, treatment responses, and survival outcomes are poorly understood. A compounded challenge in practice is that the distinction recurrence (REC) and a second primary tumor (2P) is not always clear, making treatment difficult. We hypothesize that the immune profile in resected, formalin-fixed paraffin-embedded (FFPE) tumor, can provide insight in the antitumor capability of each patient’s immune system, and hence how well it can control micro-metastasis and suppress recurrence or metastasis. In this pilot study, we employed the NanoString GeoMx Digital Spatial Profiler (GeoMxDSP) to detect intratumoral immuno-RNA and protein abundance by region of interest (ROI), and report on RNA-based analyses to identify immuno biomarkers in predicting REC and/or 2P.</p><h2>Methods</h2><p>Pathologically reviewed hematoxylin and eosin stained tissues were annotated to identify inflamed/hot and non-inflamed/cold tumor ROIs (morphology). Using the GeoMx solid tumor microenvironment morphology kit, tumor (PanCK+) and immune (CD45+) cell populations were identified within each ROI and individually profiled using the GeoMxDSP human immune pathways gene expression panel (73 genes, 5 housekeepers, 6 control probes). R software (v4.1.2) was used to conduct statistical analyses, starting with univariate Cox proportional hazard (PH) ratio model analysis of all ROIs clustered by sample ID with age, sex, smoking status, tumor histology/stage, morphology, and segment as covariates; PH tests to confirm biomarkers meeting the assumption; false discovery rate (FDR at 0.05) applied for multiple testing correction; and hazard ratios (HRs) with 95% confidence intervals estimated for each biomarker adjusted for significant covariates (Mod1). Cox models were run separately for REC and 2P groups with various pre-defined outcome events. A contrasting model using 38 patients, where all ROIs were combined for each sample to mimic bulk-tissue averaged RNA expression levels (Mod2). Parallel analyses of functional proteins and validation in independent samples are ongoing.</p><h2>Results</h2><p>A total of 311 ROIs in 38 T1-2N0M0 NSCLC patients comprimsed the cohort, with a mean diagnosis age of 66 years, 45% female, 21% never-smokers, 74% adenocarcinoma; 8 had REC, 6 had 2P, 2 both, and 5 equivocal between REC and 2P. Among the top 10 significant immuno-RNAs identified by Mod1 or Mod2 (14 total), 5 (36%) were identical with similar predictive effects: CXCL10, CXCR6, PDCD1LG2, TIGIT predict 2P; and MKI67 predicts REC. Nine (64%) differed in biomarkers or predicted event, e.g.: ITGAV (Mod1 predicts REC), DKK2 (Mod2 predicts REC+2P), and CTNNB1 (Mod1 predicts 2P, Mod2 predict REC). For the common markers signified in both models, Mod1 had more power and stable HR estimates than Mod2, e.g., HRs of 3.7-7.7 for TIGIT in Mod1 but 13.2-180.1 in Mod2. Covariates adjustment in Mod1 removed 8 markers (B2M, BATF3, CCND1, HAVCR2, HLA.DQ, KRT, PDCD1, STAT3) as significant predictors for REC or 2P.</p><h2>Conclusions</h2><p>We demonstrate the feasibility of this method of analyzing the immune mileu of a patient’s tumor and differential correlation with REC or 2P.<br></p>