Large-scale plasma proteomics identifies proteins and pathways associated with incident interstitial lung disease

Reliable biomarkers for early interstitial lung disease (ILD) risk stratification are limited; therefore, we aimed to identify prediagnostic proteins associated with incident ILD and develop a long-term prediction model.

We analyzed 52,743 UK Biobank participants without baseline ILD and measured 2,923 plasma proteins. Cox models were used to identify proteins associated with incident ILD. Key predictors were selected using LASSO and sequential forward selection within a LightGBM framework. Model performance was assessed using the area under the curve (AUC) and SHAP. Prediagnostic protein trajectories were evaluated using LOESS and clustering. Single-cell transcriptomic data from human idiopathic pulmonary fibrosis lungs were used to explore cellular sources.

During a mean follow-up period of 14.4 years, 814 ILD cases were identified in the cohort. A total of 277 proteins were significantly associated with incident ILD and enriched in immune activation and cytokine signaling pathways. Five proteins (GDF15, LAMP3, SFTPD, EDA2R, and WFDC2) were identified as dominant predictors. Trajectory analysis showed that several ILD-associated proteins deviated from control levels years before diagnosis and could be grouped into four distinct temporal clusters. GDF15, LAMP3, SFTPD, WFDC2, and EGFR showed detectable prediagnostic alterations more than a decade before ILD onset. Furthermore, connective tissue disease (CTD)-ILD displayed a distinct inflammatory proteomic profile, including CXCL13, CCL7, PDCD1, and TNFRSF4. When the five protein biomarkers were combined with clinical variables, the integrated model achieved relatively good predictive accuracy for ILD, with AUCs of 0.938 and 0.908 for 5-year and 10-year ILD risks, respectively. Single-cell analysis indicated that these proteins may predominantly originate from dysregulated alveolar and airway epithelial cells.

Plasma proteomics identified prediagnostic protein signatures associated with future ILD risk and improved long-term prediction when combined with clinical variables.