Integrative modeling of malignant epithelial programs in EGFR-mutant LUAD via single-cell transcriptomics and multi-algorithm machine learning

TL;DR

This study uses single-cell data and machine learning to identify malignant programs in EGFR-mutant lung cancer cells and proposes a new prognostic tool called EGFRmERS.

Contribution

The novel EGFR Mutation-Associated Malignant Epithelial Cell-Related Signature (EGFRmERS) is developed for improved prognosis and treatment prediction in EGFR-mutant LUAD.

Findings

EGFR-mutant epithelial cells show subcluster heterogeneity in malignant potential and pathway enrichment.

EGFRmERS outperforms existing models in predicting prognosis and immunotherapy response.

PERP is identified as a key gene linked to poor prognosis and cancer progression.

Abstract

Lung adenocarcinoma (LUAD) is the most common subtype of non-small cell lung cancer, with EGFR mutations serving as key oncogenic drivers. However, patients harboring EGFR mutations exhibit considerable heterogeneity in clinical outcomes and treatment responses. Characterizing the malignant features of EGFR-mutant epithelial cells may facilitate improved stratification and personalized therapeutic strategies.

Using publicly available single-cell RNA sequencing data, malignant epithelial cells were identified in EGFR-mutant LUAD samples via inferCNV and k-means clustering. Pseudotime trajectories were constructed using Monocle2, and branch-specific genes were extracted for functional analysis. Differentially expressed genes were integrated with TCGA bulk transcriptomic data, and ten machine learning algorithms were applied to construct the EGFR Mutation-Associated Malignant Epithelial Cell-Related Signature (EGFRmERS). The prognostic value of EGFRmERS was validated across multiple independent cohorts. Associations between EGFRmERS and immune infiltration, immunotherapy response, tumor mutation burden (TMB), and copy number variations (CNVs) were systematically assessed. The performance of EGFRmERS was also benchmarked against previously published LUAD prognostic signatures. Finally, the core gene PERP was selected for in vitro functional validation, including qRT-PCR, Transwell migration/invasion, and colony formation assays.

EGFR-mutant epithelial cells were classified into subclusters with varying malignant potential, enriched in pathways such as cell cycle regulation and DNA repair. The EGFRmERS signature robustly predicted patient prognosis across multiple cohorts and outperformed existing models. High EGFRmERS scores were associated with an immunosuppressive microenvironment, reduced immunotherapy responsiveness (as indicated by TIDE and IPS scores), elevated TMB, and increased genomic instability. PERP was identified as a key gene, highly expressed in LUAD and associated with poor prognosis. Functional assays confirmed its role in promoting cell migration, invasion, and clonogenic capacity.

This study delineates key malignant programs in EGFR-mutant epithelial cells at the single-cell level and proposes a robust prognostic scoring system, EGFRmERS, with strong predictive power for survival and immunotherapy benefit. PERP was identified as a potential therapeutic target, offering novel insights for precision stratification and treatment in EGFR-mutant LUAD.

Figures