Plasma Lipidomic Signatures Across the Healthy-Pre-COPD-COPD Continuum Identified by Machine Learning

Area of Science:

• Pulmonary Medicine
• Metabolomics
• Biomarker Discovery

Background:

• Chronic obstructive pulmonary disease (COPD) presents a significant global health challenge.
• Pre-COPD represents a critical early, high-risk phase with distinct phenotypes like small airway dysfunction (SAD), emphysema, and preserved ratio impaired spirometry (PRISm).
• Current spirometry and imaging methods do not fully capture the molecular heterogeneity of early Pre-COPD stages.

Purpose of the Study:

• To analyze plasma lipidomic profiles across the continuum from healthy individuals to Pre-COPD and COPD.
• To identify shared and phenotype-specific lipid signatures using machine learning.
• To assess the diagnostic and discriminative potential of these lipid signatures.

Main Methods:

• A cross-sectional study involving 124 participants: 30 healthy controls, 63 Pre-COPD individuals (SAD, emphysema, PRISm), and 31 COPD patients.
• Untargeted plasma lipidomics using UHPLC-high-resolution mass spectrometry.
• Regularized machine learning (elastic-net logistic regression) for feature selection and model validation via nested cross-validation.

Main Results:

• 171 differential lipids were identified, with models incorporating lipids outperforming clinical-only models (AUCs ranging from 0.83 to 0.92).
• A core set of lipids (ceramides, PG, LacCer) was associated with disease stage and pulmonary function.
• Phenotype-specific lipids were identified, including PI and Cer for SAD, PE for emphysema, and GM3 for PRISm.

Conclusions:

• Plasma lipidomic profiles can effectively distinguish between healthy, Pre-COPD, and COPD stages, including specific phenotypes.
• Identified lipid signatures show potential as biomarkers for disease stage and pulmonary function.
• Further external and prospective validation is required for clinical application of these candidate lipid markers.