Background: Early detection of lung cancer remains a challenge among high-risk individuals, particularly in low-middle income countries. Non-invasive highly stable biomarkers, such as circulating miRNAs, may distinguish high-risk cancer-free subjects from early-stage lung cancer in admixed populations.
Objective: To identify and validate a circulating miRNA signature to distinguish high-risk cancer-free individuals from patients with early-stage lung cancer.
Methods: This study employed a case–control design using plasma samples from admixed high-risk subjects and non-small cell lung cancer (NSCLC) patients from the Barretos Cancer Hospital, a non-profit cancer center that provides care exclusively through Brazil's public health system. High-risk subjects were recruited through the Barretos Cancer Hospital Mobile Lung Cancer Screening Program (BCH-MLCS), according to NLST and/or PLCOm2012 criteria and prospectively followed for five years to ensure absence of cancer. Plasma miRNA expression was quantified using NanoString nCounter Human v3 miRNA assay (770 targets), and miRNA counts were normalized using two selected housekeeping miRNAs. The discovery set included high-risk cancer-free subjects (n=42) and early-stage NSCLC patients (n=46). Machine learning approaches (LASSO, Boruta, and Random Forest) and binary logistic regression were applied to define the optimal miRNA combination. A miRNA-based risk score was derived from the logistic regression and applied to downstream analysis. The test set comprised high-risk subjects (n=21) and lung cancer patients (n=21) diagnosed through the BCH-MLCS. Both discovery and test sets were comprised by admixed individuals matched by age (±5 years), sex, and smoking status (current/former). Two independent external validation datasets were analyzed: (i) public dataset including pooled high-risk subjects (n=21; n=3/pool) and NSCLC plasma samples (n=37) from Canada (nCounter dataset); and (ii) public dataset including healthy subjects (n=51) and NSCLC cases (n=89) from Europe (RT-qPCR dataset). The miRNA signature performance was evaluated using receiver operating characteristic (ROC) curve analysis.
Results: Among 24 differentially expressed miRNAs identified in the discovery set, two were selected as the best miRNAs to distinguish high-risk cancer-free subjects from lung cancer patients (miRNA identities omitted due to intellectual property). The resulting 2-miRNA signature score differed significantly between high-risk cancer-free subjects and lung cancer patients (p<0.0001; AUC=0.726). The signature score did not differ according to the evaluated tobacco-related clinical characteristics or histological subtype. In the test set from the BCH-MLCS, the 2-miRNA signature score also discriminated high-risk cancer-free subjects from lung cancer patients (p=0.004; AUC=0.794). Despite being a real-world scenario in a public health setting, no significant differences were observed between the high-risk cancer-free subjects and cancer groups with respect to age (p=0.07), sex (p=0.22), or tobacco exposure (p=0.10). External validation using an independent public dataset generated with the same nCounter technology confirmed the discriminatory ability of the signature (p=0.047; AUC=0.712). Finally, to enhance the generalizability, the performance of the 2-miRNA signature was assessed in an independent RT-qPCR dataset, demonstrating consistent discrimination between non-cancer subjects and lung cancer patients (p=0.0026; AUC=0.609).
Conclusion: The 2-miRNA plasma signature distinguished high-risk cancer-free individuals from early-stage NSCLC patients across different ethnicities, suggesting potential utility in supporting early lung cancer detection and complementing current screening-based strategies in diverse healthcare settings.