The Problem of Radiotherapy Resistance Radiotherapy is a critical treatment for NSCLC, particularly for patients with inoperable tumors. However, many tumors develop or harbor inherent resistance to radiation, where cancer cells adapt their biology to withstand radiation's DNA-damaging effects. This resistance leads to treatment failure and poor survival outcomes.
The Immune Connection to Radioresistance Beyond direct DNA repair mechanisms, the immune microenvironment plays a crucial role in whether radiotherapy succeeds. Radiation can stimulate immune responses against tumors, but an immunosuppressive tumor microenvironment can neutralize this benefit. Understanding which immune factors predict resistance is essential for designing better combined approaches.
Study Approach This bioinformatics study used publicly available gene expression datasets from radiation-resistant versus radiation-sensitive NSCLC cell lines and patient samples, combined with TCGA (The Cancer Genome Atlas) clinical data, to identify genes and immune signatures associated with radiotherapy resistance in NSCLC.
Key Innovation: Multi-Algorithm Gene Filtering By intersecting results from three different machine learning approaches (LASSO regression, SVM-RFE, and Random Forest), the study identified the most robustly predictive genes from an initial list of 103 radioresistance-associated candidates - a stringent approach that reduces false positives and increases confidence in the identified markers.
Primary Gene Expression Datasets Two GEO (Gene Expression Omnibus) datasets were analyzed: GSE197236, comparing radiation-resistant vs. control A549 lung adenocarcinoma cells using microarray technology; and GSE253564, comparing samples treated with anti-PD-L1 alone vs. anti-PD-L1 plus radiotherapy using RNA sequencing. Using two datasets with different experimental designs increases the robustness of identified genes.
Differential Expression Analysis Differentially expressed genes were identified using limma (for microarray data) and DESeq2 (for RNA sequencing data), both widely used statistical tools for gene expression analysis. Criteria of p-value less than 0.05 and fold-change magnitude greater than 2-fold were applied. The overlap between the two datasets yielded 103 genes associated with radiotherapy resistance.
TCGA-LUAD Validation Cohort The Cancer Genome Atlas Lung Adenocarcinoma (TCGA-LUAD) dataset, comprising 539 primary tumor samples with clinical outcome data, was used for prognostic model development and validation. Samples were split 7:3 into training and validation sets. ComBat harmonization corrected for batch effects between datasets.
Pathway Enrichment Analysis GO and KEGG pathway enrichment analyses identified biological processes over-represented among the 103 radioresistance genes. Enriched pathways included blood coagulation, complement activation, growth factor activity, cytokine signaling, DNA damage repair, and the p53 signaling pathway - revealing the diverse biological mechanisms underlying radioresistance.
LASSO Regression for Feature Selection LASSO regression was applied to the 103 candidate genes to identify those most predictive of radiotherapy resistance while penalizing model complexity. LASSO shrinks redundant predictors to zero, automatically selecting a parsimonious and robust gene signature. Cross-validation with 10 folds was used to select the optimal regularization parameter.
SVM-RFE for Recursive Feature Elimination Support Vector Machine Recursive Feature Elimination (SVM-RFE) trains SVM models on subsets of features, iteratively removing the least informative features based on their coefficient weights. This approach identifies features that maximally separate resistant from sensitive samples.
Random Forest for Ensemble Feature Importance Random Forest analysis ranked all 103 genes by their importance using Mean Decrease Accuracy (MDA) and Mean Decrease Gini (MDG) metrics across 500 decision trees. The top 50 genes by both metrics were selected as candidates.
Four Genes at the Intersection By taking the intersection of genes selected by all three algorithms, four genes were identified as robustly associated with radioresistance: TGFBI (Transforming Growth Factor Beta Induced), FAS (also known as CD95, a cell death receptor), PTK6 (Protein Tyrosine Kinase 6), and FA2H (Fatty Acid 2-Hydroxylase). Each gene was independently validated by all three distinct analytical approaches.
Diagnostic Nomogram Performance A diagnostic nomogram integrating all four key genes was developed and showed strong predictive capability for NSCLC radiotherapy outcome. Calibration curves confirmed close alignment between nomogram-predicted probabilities and actual observed outcomes. ROC analysis showed AUC greater than 0.7 for all four genes individually, confirming their diagnostic value.
TGFBI Shows Strongest Prognostic Correlation Among the four genes, TGFBI demonstrated the strongest correlation with NSCLC prognosis. TGFBI encodes a protein involved in cell adhesion and the extracellular matrix, and its overexpression promotes cancer cell migration, invasion, and resistance to cell death - mechanisms that could explain radiation resistance.
Prognostic Risk Model Performance A LASSO Cox regression prognostic model was developed using the four genes and TCGA-LUAD survival data. The model successfully stratified patients into high-risk and low-risk groups with significantly different overall survival (log-rank p < 0.05), validated in both the training set and the independent validation set.
Regulatory Network Analysis The four key genes were connected to broader regulatory networks. SP1 transcription factor was found to simultaneously regulate both FAS and PTK6. RNA-binding protein interactions and miRNA-mediated regulation (via the ceRNA network) for each gene were mapped, revealing complex regulatory relationships that could serve as additional therapeutic targets.
Immune Microenvironment Analysis ssGSEA (single-sample Gene Set Enrichment Analysis) was used to estimate the relative abundance of 22 immune cell types in each patient's tumor, computed from gene expression profiles. This revealed significantly different immune landscapes between high-risk (radioresistant) and low-risk patients.
Naive B Cells and M0 Macrophages as Key Differentiators The most significant immune cell differences between high-risk and low-risk groups involved naive B cells and M0 macrophages. High-risk patients showed different infiltration patterns of these cell types, suggesting that the immune microenvironment composition contributes to or reflects radiotherapy resistance.
Macrophage Polarization Implications The study found significant differences in macrophage subtypes between risk groups. Since M1 macrophages can enhance radiation-induced tumor killing while M2 macrophages promote immune suppression and tissue repair (potentially protecting tumor cells), targeting macrophage polarization toward M1 phenotype could enhance radiotherapy effectiveness.
B Cell Activation Strategies Naive B cells may play roles in shaping the adaptive immune response to radiation-damaged tumor cells. Enhancing naive B cell activation or promoting their differentiation into effector B cells producing tumor-targeting antibodies could complement radiotherapy by promoting immune-mediated tumor clearance.
Predictive Biomarker Application The four-gene signature (TGFBI, FAS, PTK6, FA2H) could be assessed from pre-treatment tumor biopsy using standard RNA expression profiling to predict which patients will respond to radiotherapy. High-risk patients identified by the model could be offered alternative or enhanced treatment strategies rather than standard radiotherapy alone.
Combination Therapy Rationale Understanding that radioresistance correlates with specific immune cell patterns provides rational basis for combining radiotherapy with immunotherapy. High-risk patients with M2 macrophage enrichment might benefit from macrophage-reprogramming agents. Those with low naive B cell infiltration might benefit from approaches that enhance B cell activation such as CD40 agonists.
Drug Target Identification DGIdb drug-gene interaction analysis identified potential drugs targeting the key genes. EDELFOSINE and VB-111 were identified as potential FAS-targeting agents, suggesting existing drugs might be repurposed as radiosensitizers for patients with radioresistance signatures.
Monitoring Radiotherapy Response The identified gene signature and immune infiltration patterns could potentially be monitored during treatment using liquid biopsy approaches measuring circulating tumor DNA or tumor-derived RNA, enabling dynamic adjustment of treatment based on whether the molecular resistance profile is changing.
In Vitro and Database Limitations The primary gene identification was performed in radiation-treated cell lines and patient transcriptomic databases rather than in prospective clinical radiotherapy trials. Cell line behavior does not perfectly mirror human tumor biology, and gene expression databases may not capture the full complexity of clinical radioresistance.
Retrospective TCGA Data The prognostic model was validated in TCGA-LUAD data, which is retrospective and may not reflect modern radiotherapy protocols. TCGA was not specifically designed to study radiotherapy outcomes, and radiation treatment details are incompletely captured in the database.
Prospective Validation Needed The four-gene signature requires prospective validation in a dedicated clinical cohort where NSCLC patients are followed through radiotherapy with pre-treatment biopsy gene expression profiling and standardized outcome assessment. Such a trial would definitively establish the clinical utility of the signature.
Mechanistic Studies Required The biological roles of TGFBI, PTK6, and FA2H in radioresistance need experimental validation. Cell-based and mouse model studies knocking out or overexpressing these genes, then measuring radiation sensitivity, would confirm their causal roles and potentially identify druggable mechanisms.