As the leading cause of cancer-related deaths worldwide, lung cancer makes non-invasive biomarkers critically important for its diagnosis and monitoring. However, traditional plasma protein detection methods, such as mass spectrometry for liquid biopsy, have limitations including limited depth, the masking of low-abundance proteins by high-abundance proteins, and a focus on a single endpoint. This study aimed to employ a NaY-based low-abundance protein enrichment technique for plasma proteomic analysis, exploring its potential application in the prediction of multiple events in lung cancer, spanning diagnosis, lymph node metastasis detection, and TNM staging.

Researchers prospectively collected samples from 287 patients with lung cancer and benign lung diseases at the First Affiliated Hospital of Guangzhou Medical University, the National Center for Respiratory Medicine, and the Tianjin Key Laboratory of Clinical Multi-Omics, strictly defining inclusion and exclusion criteria, and confirming lung cancer and benign lung diseases through histopathology, while also collecting patients' clinicopathological data and blood samples. Blood samples underwent low-abundance protein enrichment-based proteomic testing, and a multi-task biomarker study was conducted using feature selection (based on ROC curve evaluation and LASSO regression), random forest model construction and validation, and the SHAP algorithm for model interpretation.

Schematic diagram of the study workflow

The three cohorts differed in age, sex, and underlying diseases, but UMAP dimensionality reduction analysis revealed no significant inter-group differences. After rigorous quality control and quantitative analysis, expression profiles of 4,703 plasma proteins were obtained, and a large number of differentially expressed proteins were successfully identified. These differential proteins could effectively distinguish benign from malignant patients, as well as patients at different stages, and showed high overlap across different tasks.

Analysis of differentially expressed proteins between groups

Lung cancer diagnosis: A 10-protein panel (such as BAK1, CTSW, etc.) was selected, achieving an internal validation AUC of 0.91 (0.84-0.97) and an external validation AUC of 0.87 (0.77-0.97), which increased to 0.91 (0.85-0.98) when combined with clinical factors;

Lymph node metastasis detection: A 9-protein panel (such as XPO1, NPNT, etc.) achieved an AUC of 0.88 (0.80-0.96), which increased to 0.90 (0.85-0.97) when combined with clinical factors;

TNM staging: The AUCs of the 10-protein panel varied across different stages, and after combining with clinical factors, the AUCs for each stage improved; for example, stage I increased from 0.86 (0.69-0.93) to 0.88 (0.74-0.96), and stage IV from 0.86 (0.75-0.93) to 0.99 (0.98-0.99). Through the SHAP method, the key proteins and their risk or protective roles in each task were clarified; for example, in task #1, CTSW was a protective factor and BAK1 a risk factor, and most of these biomarkers were closely related to tumor progression.

Multi-task biomarker discovery and validation

This study adopted a novel plasma proteomics technique based on low-abundance proteomic enrichment and successfully developed a highly accurate multi-task prediction model for lung cancer. Compared with previous studies on lung cancer detection based on plasma protein measurement, this study differs in technical methods, biomarker screening, and model performance, highlighting the enormous potential of plasma proteomics in biomarker research. Based on the results of this study, directions for further optimizing and validating the prediction model are proposed, including further validation in larger-scale and more diverse populations, and exploring combined applications with other detection technologies or biomarkers.