Abstract 6595: Lineage markers and driver oncogenes coordinate through gene expression subtypes of lung adenocarcinoma

Abstract Background: EGFR-directed therapy marked the beginning of personalized medicine in lung adenocarcinoma (LUAD), where a single gene could be effectively targeted by a specific class of drugs, leading to remarkable treatment responses. After additional targets have been identified, the search for other druggable mutations has slowed despite the need to improve clinical outcomes, underscoring the need for new approaches. Hypothesis: We hypothesize that gene expression signatures will predict EGFR mutation status in patients and identify cases with similar signaling pathways but without canonical EGFR mutations. Deciphering the role of the EGFR signature in mutant (mt) and wild-type (WT) cases may provide potential therapeutic and classification strategies for understanding LUAD. Experimental design: To define an EGFR mutation signature (“mSig”) as a predictor of EGFR mutation status, we analyzed gene expression and clinical data from four lung adenocarcinoma cohorts. Microarray data from Memorial Sloan Kettering Cancer Center (MSKCC, n=192) served as the training set. For validation, we used combined microarray data from UNC (n=73) and TSP (n=41) as one dataset, and RNA-seq data from The Cancer Genome Atlas (TCGA, n=486) as another. Multiple model evaluation algorithms tested the performance of the predictor. Other key genetic mutations found in TCGA cohort were included in integrative analysis. Results: Comparing EGFR-mt to EGFR WT tumors identified 1, 020 differentially expressed genes in the training dataset as an EGFR mSig. Semi-supervised clustering visualized molecular profiles in both training and validation data. The EGFR mSig achieved an average negative predictive value (NPV) of 96.3% across datasets and was robust across multiple dataset combinations and six machine learning models (AUROC = 0.83-0.95). The EGFR mSig strongly correlated with LUAD subtypes (Bronchioid, Magnoid, Squamoid), particularly with Bronchioid [odds ratio (OR) = 9.2, p < 0.001]. EGFR mSig-positive EGFR-mt (“EGFR-mt/mSig(+)”) tumors were primarily Bronchioid (OR = 4.5, p < 0.001). Almost 50% of mSig(+) Bronchioid tumors in fact lacked EGFR mutations, often showing EGFR amplifications or RAS mutations. Two lung lineage markers and driver mutations associated with LUAD subtypes and EGFR activation patterns. Interestingly, EGFR-mt/mSig(-) tumors were generally non-Bronchioid, suggesting alternative pathways (e.g., RAS, TP53). RAS/RAF/RTK-mt tumors (EGFR WT or mSig(-)) showed high dual specificity phosphatase 4 (DUSP4) expression, a known repressor of multiple key activated RAS genes. Therefore, DUSP4 represents an attractive novel target in selected LUAD samples. Conclusions: Our results define a novel EGFR mSig that identifies EGFR-mt-like tumors. Identifying patients with driver gene mutations lacking EGFR mutations suggests a new approach of identifying new druggable targets. Citation Format: Minjeong Kim, Wisut Lamlertthon, Heejoon Jo, Yan Cui, Hyo Young Choi, Matthew D. Wilkerson, Liza Makowski, D. Neil Hayes. Lineage markers and driver oncogenes coordinate through gene expression subtypes of lung adenocarcinoma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 6595.