Beyond T Cells: The Role of B Lymphocytes Cancer immunology research has focused primarily on T cells and PD-L1 as determinants of immunotherapy response. However, tumor-infiltrating B lymphocytes (TILBs) are increasingly recognized as critical components of the tumor immune microenvironment, influencing tumor control and immunotherapy outcomes through antibody production, antigen presentation, and tertiary lymphoid structure formation.
The Knowledge Gap Despite emerging evidence for TILBs' importance, no validated prognostic index specifically quantifying the TILB compartment exists for lung adenocarcinoma (LUAD). Existing immune scoring tools either focus on T cells or treat immune cells as a composite without B-cell-specific resolution.
Machine Learning Approach This study used single-cell RNA sequencing (scRNA-seq) data from LUAD patients (GSE117570) to characterize B lymphocyte subtypes, then applied an exhaustive combination of 10 machine learning algorithms and 101 algorithm combinations to identify the optimal prognostic model.
Multi-Cohort Validation Strategy Beyond the discovery dataset, the B lymphocyte-related index (BRI) was validated in multiple independent external cohorts including TCGA and GEO datasets, and its association with immunotherapy response was assessed in immunotherapy-treated patient cohorts.
Single-Cell B Cell Characterization scRNA-seq data from GSE117570 (lung adenocarcinoma patients) were analyzed using the Seurat pipeline to identify and characterize distinct B lymphocyte subpopulations. Cell clustering, annotation with marker genes, and trajectory analysis were performed to understand B cell differentiation states within the tumor microenvironment.
Identification of 27 B Cell Signature Genes Differential expression analysis between B cell subpopulations and integration with published B cell biology literature identified 27 key genes associated with TILB function, activation, and differentiation. These genes formed the candidate feature set for the prognostic index.
10 ML Algorithms and 101 Combinations Tested To identify the most robust prognostic model, 10 machine learning algorithms were tested including LASSO, Ridge, Elastic Net, Random Survival Forest (RSF), Support Vector Machine, XGBoost, and others. Each algorithm was tested alone and in combinations (using ensemble stacking), yielding 101 total model variants evaluated by C-index.
RSF + superPC as Optimal Model The combination of Random Survival Forest (RSF) and supervised principal component analysis (superPC) achieved the highest average C-index of 0.65 across validation cohorts. The final BRI model was built using this approach, assigning each patient a continuous risk score based on the 27-gene signature.
Prognostic Stratification Patients stratified by BRI into high vs. low score groups showed significantly different overall survival and progression-free survival in the TCGA-LUAD cohort and multiple external validation datasets. High BRI scores correlated with worse prognosis, consistent with more aggressive tumor immune microenvironments or B cell dysfunction in high-risk tumors.
Immune Microenvironment Correlations High BRI scores were associated with specific immune microenvironment characteristics: altered ratios of B cell subtypes, differences in CD8+ T cell infiltration, and distinct macrophage polarization patterns. This suggests BRI captures a broader immune program rather than isolated B cell biology.
Immunotherapy Response Prediction In cohorts of NSCLC patients treated with ICIs, higher BRI scores were associated with better objective response rates and progression-free survival. This association, if confirmed prospectively, would make BRI a complement to PD-L1 and TMB for immunotherapy patient selection.
Drug Sensitivity Implications Analysis of drug response data showed BRI-stratified patients had differential sensitivity to specific chemotherapy agents and targeted therapies. High-BRI patients showed different IC50 profiles for certain drugs, suggesting BRI could guide not just immunotherapy but broader treatment selection.
Complementing PD-L1 and TMB Current immunotherapy biomarkers have significant limitations - PD-L1 is expressed in only a subset of responding patients, and TMB is impractical to measure in many settings. BRI provides an orthogonal signal based on the adaptive immune B cell compartment, potentially identifying responders missed by PD-L1/TMB testing.
Tertiary Lymphoid Structures The B cell signature likely captures signals related to tertiary lymphoid structures (TLS) - organized lymphoid aggregates within tumors that are strongly associated with immunotherapy response across multiple cancer types. BRI may serve as a transcriptomic surrogate for TLS presence.
Multi-Gene Panel Development The 27-gene BRI signature could be translated into an RNA-based clinical assay measurable from FFPE tumor tissue - similar to commercially available gene expression panels. This would make the test accessible in routine pathology practice.
Combination Biomarker Strategies The strongest clinical utility would likely come from combining BRI with PD-L1 TPS and TMB in a multi-biomarker model. Patients with high BRI + high PD-L1 + high TMB would be the most compelling immunotherapy candidates, while BRI-high/PD-L1-low patients might represent a distinct ICI-responsive subgroup.
Modest C-Index The optimal model achieved a C-index of 0.65, indicating modest rather than strong prognostic discrimination. While statistically significant and consistent across cohorts, C-index 0.65 suggests BRI captures one dimension of a multi-factorial prognosis and will need to be combined with other biomarkers for stronger performance.
Retrospective Dataset Limitations All validation was performed on retrospective genomic datasets (GEO, TCGA). These datasets have heterogeneous treatment histories, variable follow-up durations, and may have ascertainment bias. Prospective biomarker studies with standardized treatment protocols are needed.
From scRNA-seq to Clinical Assay The current BRI is derived from scRNA-seq data, which remains expensive and technically demanding. Converting the 27-gene signature into a validated, cost-effective bulk RNA or NanoString assay for FFPE tissue is the critical translation step.
Future Research Priorities Prospective studies should evaluate BRI as a companion diagnostic for ICI therapy selection, explore spatial transcriptomics to understand B cell distribution patterns within tumors, and investigate whether therapeutic strategies targeting B cell activation could be used in combination with ICIs for high-BRI patients.