Machine learning assisted immune profiling of COPD identifies a unique emphysema subtype independent of GOLD stage

Abstract

Summary: Chronic obstructive pulmonary disease (COPD) is a severe, progressive, and heterogeneous disease with a poor outcome. Inflammation plays a central role in disease pathogenesis; however, the interplay between immune changes and disease heterogeneity has been difficult to unravel. We performed a multilevel immunoinflammatory characterization of patients with COPD using flow cytometry, cytokine profiling, single-cell, or spatial transcriptomics in combination with machine learning algorithms. Our cross-cohort analysis demonstrated shared skewing of immune profiles in COPD lungs toward adaptive immune cells. We furthermore identified a subgroup of patients with COPD with a distinct immune profile, characterized by increased antigen-presenting cells, mast cells, and CD8+ cells, and circulating IL-1β, IFN-β, and GM-CSF, that were associated with increased emphysema severity and decreased gas exchange parameters independent of their GOLD-stage. Our findings suggest that unbiased immune profiling can refine disease classification and reveal inflammation-driven disease subtypes with potential relevance for prognosis and treatment strategies.

Keywords

• respiratory medicine
• machine learning