Circulating metabolic subtypes and proteomic signatures associated with incident idiopathic pulmonary fibrosis: a UK biobank prospective cohort study

Abstract Background Idiopathic pulmonary fibrosis (IPF) is biologically heterogeneous, but whether circulating metabolic/biomarker heterogeneity and complementary proteomic information can prospectively identify individuals at different risks of incident IPF and improve risk prediction remains unclear. Methods We analyzed 9,443 UK Biobank participants with 308 baseline metabolic/biomarker (Met/Bio) traits and 2,909 plasma proteins. Participants were separated into training, internal validation, and geographically separated external validation sets before imputation and feature selection. Training-set LASSO selection, principal component analysis, and K-means clustering were used to derive Met/Bio subtypes. Incident IPF risk was assessed across subtypes. Within the highest-burden subtype, protein-wide associations were evaluated using complementary Cox approaches, and random-forest models integrating conventional factors, Met/Bio traits, and proteins were assessed at 5, 10, and 15 years. Results During 128,534 person-years of follow-up, 90 incident IPF cases occurred. Twenty-seven Met/Bio traits defined three reproducible subtypes. C3, characterized by a triglyceride-rich VLDL profile together with inflammatory features, had the highest observed IPF burden, with the overall risk pattern reproduced in both validation sets. Within C3, five proteins—CD80, ADA2, CDCP1, CCL7, and DPEP1—showed robust associations with incident IPF across unpenalized, ridge, and Firth Cox analyses. Compared with the Basic model, the integrated Basic+Met/Bio+Pro model increased 5-, 10-, and 15-year AUCs from 0.573, 0.659, and 0.705 to 0.851, 0.725, and 0.757 in internal validation and from 0.891, 0.650, and 0.692 to 0.971, 0.939, and 0.808 in geographically separated external validation. Conclusions Circulating Met/Bio subtyping identified reproducible heterogeneity in incident IPF risk and a risk-enriched subtype. Integrating Met/Bio and proteomic information showed proof-of-concept predictive improvement beyond conventional factors, warranting validation in larger independent cohorts.