Enhancing early detection of lung cancer: Methylation anchor probe for low-signal enrichment (MAPLE).

Abstract

8057 Background: Non-Small Cell Lung Cancer (NSCLC) is among the most lethal cancers worldwide. Lung adenocarcinoma (LUAD) and lung squamous cell carcinoma (LUSC) represent approximately 80% of NSCLC cases in China. While low-dose computed tomography (LDCT) is widely used for annual screening, its high false-positive rate highlights the need for more accurate early detection methods. Circulating tumor DNA (ctDNA) analysis provides a promising non-invasive alternative for early cancer detection. However, conventional hybrid capture methods lack the sensitivity to detect low-abundance ctDNA in early-stage cancers. Previously, we developed ultra-sensitive Methylation Anchor Probes for Low signal Enrichment (MAPLE) that significantly improved the detection of colorectal cancer (Xie et al., 2024). Here, we have advanced this technology to develop a novel assay to enhance early detection of NSCLC and distinguish LUAD from LUSC, paving the way for more personalized treatment strategies. Methods: NSCLC-related methylation haplotypes were identified using in-house whole-genome bisulfite sequencing data from lung cancer tumor tissues and paired normal adjacent tissues (NATs). Haplotype selection was performed by filtering for those with a frequency difference greater than 0.1 between tumor and NATs and a frequency below 0.001 in healthy cfDNA, resulting in a panel targeting 12,904 methylation haplotypes. The panel was evaluated on 234 clinical samples, including 44 LUSC patients, 43 LUAD patients, 19 individuals with chronic obstructive pulmonary disease (COPD), and 128 healthy controls. All cfDNA samples underwent bisulfite conversion, library preparation, hybrid capture using the custom panel, and next-generation sequencing (NGS). The dataset was split into a 75% training set and a 25% validation set. A primary classifier was developed to identify cancer samples, and true positives were further analyzed with a subtype classifier to differentiate between LUAD and LUSC. Model performance was assessed for robustness using 40 resampling processes. Results: The methylation panel combined with a machine-learning classifier achieved an AUC of 0.93 (0.93–0.93) in the training set and 0.91 (0.91–0.92) in the validation set in the detection of NSCLC. Furthermore, the assay effectively distinguished COPD patients from cancer cases, with a specificity of 92.5% (90.0%–95.9%). Additionally, the subtype classifier accurately differentiated LUAD from LUSC with 100% accuracy. Conclusions: We developed a technique to specifically enrich NSCLC-related methylation haplotypes, improving sensitivity for early-stage NSCLC detection. The assay demonstrated strong performance in accurately distinguishing LUAD from LUSC, highlighting its potential to guide treatment decisions. The MAPLE platform is versatile and shows promise for broader applications in cancer early detection.