Abstract Introduction/Rationale High-resolution computed tomography (CT) remains the gold standard for diagnosing and quantifying pulmonary emphysema, yet robust predictors of disease progression are lacking. We hypothesized that baseline CT low-attenuation patterns can forecast the rate of tissue degeneration and that rapid progression associates with distinct immune responses in lung tissue. Methods We developed LobTe (Lobe-based Transformer Encoder), a machine-learning model trained on baseline and 5-year follow-up CT images from 4,612 smokers with or without COPD in the COPDGene cohort. LobTe was then validated in an independent in-house cohort to stratify patients by progression rate. To characterize the pulmonary immune landscape, we performed single-cell RNA sequencing and GeoMx spatial transcriptomics, using single-cell signatures to deconvolute spatial data and quantify immune cell populations across parenchyma and airways. The integrated dataset included 62 patients with COPD and controls, classified by LobTe as fast or slow progressors, with or without baseline emphysema. Results LobTe accurately separated fast from slow progressors in both emphysematous and nonemphysematous groups. In emphysematous fast progressors, pulmonary macrophages were decreased, while adaptive T and B lymphocytes expanded. In contrast, nonemphysematous fast progressors exhibited increased macrophages with reduced CD4+ and CD8+ T cells. Central memory CD4+ T cells were enriched in emphysematous fast progressors, whereas slow progressors showed higher proportions of cytotoxic CD8+ T cells. Spatial transcriptomics confirmed adaptive immune cell enrichment in both parenchymal and airway regions of fast progressors, consistent with persistent inflammatory activation. Conclusion Baseline CT low-attenuation signatures predict emphysema progression and reveal immune shifts across disease stages. Early progression involves macrophage expansion, while advanced disease is driven by adaptive immune activation. The enrichment of central memory CD4+ T cells may underlie chronic, non-resolving inflammation in fast progressors, whereas cytotoxic CD8+ T cells could confer protection in slower progressors. These results identify imaging-linked immune endotypes that may inform precision therapies in COPD. Future Directions Next steps include expanding LobTe training with larger longitudinal CT datasets and validating single-cell and spatial transcriptomic findings in independent cohorts. Integrating imaging and molecular data will facilitate immune-based endotyping and the discovery of circulating biomarkers for precision COPD treatments. This abstract is funded by: None
Zhang et al. (Fri,) studied this question.