Tri-marker model integrating BCL2A1 expression in CD8⁺ T cells, PD-L1, and HOT score as a predictor of immune checkpoint blockade response in lung adenocarcinoma: A multi-cohort integrative analysis.

291

Background: Despite advances in immune checkpoint blockade (ICB) for advanced lung adenocarcinoma (LUAD), predictive biomarkers beyond PD-L1 remain limited. This retrospective correlative study characterizes BCL2A1 expression in CD8⁺ T cells as a novel biomarker and evaluates its integration into a tri-marker predictive model. Methods: Publicly available bulk and single-cell RNA-seq datasets from ICB-treated LUAD patients were analyzed (discovery cohort, n = 60; five validation cohorts, n = 126). Inclusion required pathologically confirmed LUAD and pre-treatment tumor transcriptomes. Subgroups included ICB-treated vs. non-ICB cohorts. Primary endpoint: overall survival (OS); secondary: predictive AUC. This exploratory retrospective bioinformatics study used existing datasets without formal power calculation (assumed detectable HR < 1; null HR = 1). Assay: RNA sequencing; expression quantified as TPM/log2 counts, primarily continuous. Dichotomization cut-points for BCL2A1, PD-L1, and HOT score were optimized by Youden's index in discovery and locked for validation. Validation performance used leave-one-cohort-out (LOCO) cross-validation with macro-AUC averaging. No multiple testing correction beyond benchmarking. Statistics: differential expression, Cox proportional hazards, logistic regression, AUC, and Spearman correlation; full cohorts analyzed. Results: Single-cell analyses revealed BCL2A1 enrichment in tissue-resident memory and proliferating CD8⁺ T-cell subsets, with inferred increased outgoing signaling via the MIF pathway (p = 0.0278). In bulk cohorts, high BCL2A1 expression was significantly associated with improved OS in ICB-treated patients (HR = 0.43; 95% CI: 0.21–0.87; p < 0.05), but not in non-ICB cohorts, indicating treatment-specific relevance. The tri-marker model (BCL2A1, PD-L1 expression, and 27-gene HOT score) demonstrated robust predictive performance (discovery AUC = 0.826; validation macro-AUC = 0.774), outperforming PD-L1 alone (AUC = 0.706) and established signatures (TIDE, IPS, TIS, IFNG). Cross-platform in-silico reproducibility of marker expression and model scores was high (Spearman ρ = 0.982–0.993). Conclusions: BCL2A1 characterizes a CD8⁺ T-cell state linked to favorable ICB outcomes in LUAD. The tri-marker model offers a reproducible and superior predictive tool for patient stratification, supporting prospective clinical validation as a transcriptional biomarker.