Immunogenic cell death and tertiary lymphoid structures jointly define dysregulated antitumor immune states and...

Immunogenic cell death (ICD) and tertiary lymphoid structures (TLSs) are biologically distinct yet interconnected components of antitumor immunity. ICD promotes danger-associated immune activation, whereas TLSs organize local adaptive immune responses. However, how these processes jointly affect immune heterogeneity, prognosis, and therapeutic responsiveness in lung adenocarcinoma (LUAD) remains unclear.

To define the coordinated roles of TLS- and ICD-related programs in LUAD, we used a multi-cohort and multi-layer analytical framework. This framework included transcriptomic analysis, Mendelian randomization, consensus machine-learning prognostic modeling, immune landscape characterization, single-cell RNA sequencing, multiplex immunohistochemistry, immunotherapy response assessment, exploratory drug screening, structural analyses, and multilayer network topology analysis.

TLS- and ICD-associated genes were transcriptionally dysregulated in LUAD. By integrating 13 machine-learning algorithms, we developed 801 candidate prognostic models and identified optimal TLS- and ICD-related models that stratified patients effectively into high- and low-risk groups. These models showed strong performance in TCGA and remained robust across six external cohorts. Low-risk tumors had higher TLS+ICD scores than high-risk tumors. Immune landscape analysis indicated that low-risk tumors had elevated immune scores, enhanced antigen presentation, cytotoxic T/NK-cell signatures, checkpoint activation, and lower Tumor Immune Dysfunction and Exclusion (TIDE) scores, suggesting a more immunotherapy-responsive microenvironment. Consistently, higher TLS+ICD scores correlated with improved survival and greater immunotherapy responsiveness in independent cohorts. Mendelian randomization and survival analyses prioritized CD40, CD5, and GFI1 as candidate TLS-ICD immune-regulatory genes, with single-cell and mIHC analyses defining their major cellular origins, spatial protein phenotypes, and prognostic relevance. Drug prediction, molecular docking, and molecular dynamics simulations identified CAY10576, SCH772984, and SZ4TA2 as potential therapeutic agents for high-risk LUAD. Multilayer gene regulatory network reconstruction and GLMY homology analysis revealed progressive disruption of higher-order immune-regulatory organization from normal to tumor tissues and between low TLS-risk/low ICD-risk (LL-risk) and high TLS-risk/high ICD-risk (HH-risk) states.

TLS- and ICD-related programs jointly delineate clinically significant immune states in LUAD through convergent transcriptomic, genetic, single-cell, spatial, pharmacological, and systems-level evidence, and are associated with immune microenvironment remodeling, immunotherapy responsiveness, prognosis, and therapeutic vulnerability. These findings establish a biomarker-guided, machine learning-supported framework for LUAD risk stratification and may inform more precise immunotherapy-oriented therapeutic strategies.