Transcriptomic Profiling Identifies a Prognostic and Predictive Inflammatory Signature in Patients With Lung Adenocarcinoma

Document Type

Conference Proceeding

Publication Date

10-1-2025

Publication Title

J Thorac Oncol

Keywords

biological marker, diazepam, immune checkpoint inhibitor, interleukin 7, transcriptome, cohort analysis, conference abstract, controlled study, diagnosis, diagnostic test accuracy study, differential gene expression, extracellular matrix, female, functional enrichment analysis, gene expression, gene expression profiling, gene ontology, human, human tissue, inflammation, least absolute shrinkage and selection operator, lung adenocarcinoma, machine learning, major clinical study, male, non small cell lung cancer, overall survival, prediction, predictive value, progression free survival, RNA sequencing, signal transduction, survival analysis

Abstract

Introduction: Circulating inflammatory markers such as neutrophilto- lymphocyte ratio are promising biomarkers in non-small cell lung cancer (NSCLC) patients. However, inflammatory transcriptomic biomarkers have not been well studied. Here, we constructed a transcriptomic- based inflammation signature and validated its prognostic and predictive value in LUAD. Methods: We analyzed RNA-seq data from 507 lung adenocarcinoma (LUAD) patients and 59 normal controls from The Cancer Genome Atlas (TCGA). Inflammation-related genes (IRGs) were obtained fromgeneCards, and differentially expressed genes (DEGs) were identified using DESeq2 with a threshold of |Log2FC| > 0.5 and adjusted p-value < 0.05. Prognostic biomarkers were selected using LASSO regression, followed by univariate and multivariate Cox proportional hazard models. We devel- oped a risk score using significant genes, and patients were stratified into high- and low-score groups using the median as the cut-off value. Survival analysis for Overall Survival (OS) and progression-free Sur- vival (PFS) was performed using Kaplan-Meier curves and log-rank tests. Functional enrichment analysis was conducted using Gene Ontology (GO). External validation was performed using independent GEO cohorts (GSE29013, GSE30219). Additionally, we used multiple machine learning models to assess the predictive value of inflamma- tion-related genes for immune checkpoint inhibitor (ICI) response across four GEO datasets. Results: We identified 3,083 DEGs between LUAD and normal samples, of which 872 were inflammation-related. LASSO regression and Cox analysis revealed three independent prognostic biomarkers: ADM, LOX, and NEFL, all associated with poor survival (P-value < 0.05). The risk score model effectively grouped patients into high-score and low-score groups. The high-score group had significantly worse PFS and OS than the low-riskgroup (P-value < 0.05). Functional enrichment analysis highlighted pathways related to neuronal signaling and extracellular matrix remodeling in tumor cells compared to normal tissue. Molecular analysis revealed distinct pro- tein and gene expression profiles between riskgroups, with proteins such as ERBB3 and MAPK9 overexpressed in the low-riskgroup. In contrast, SERPINE1 and SMAD3 were overexpressed in the high-riskgroup. External validation confirmed the prognostic utility of the risk model in independent cohorts. A machine learning model was devel- oped using inflammation-related DEGs to predict ICI response based on transcriptomic data from four GEO cohorts. After training multiple models, LightGBM demonstrated the best performance when applied to a merged dataset of 68 patients and 42 genes, achieving an accuracy of 0.89 and an AUROC of 0.8. Using SHAP (Shapley Additive Expla- nations) analysis, we identified the key genes driving the model's predictions. The results showed that IL7 (SHAP Importance: 0.2722, p = 0.0018) and SLC2A1 (SHAP Importance: 0.2058, p = 0.0130) are significant predictors of ICI response, suggesting their potential as biomarkers in LUAD. Conclusions: We constructed an inflammation score based on gene expression in LUAD patients. The signature depended on 3 inflammation genes significantly associated with survival. The high-inflammation group was associated with better OS, PFS, and DSS. Machine learning models demonstrated that inflammation- related genes could predict ICI response with high accuracy across multiple datasets.

Volume

20

Issue

10

First Page

S442

Find It @ Sladen

Share

COinS