Plasma proteomic profiles in diverse populations identify biomarkers predicting metabolic dysfunction-associated steatotic liver disease up to 16 years before onset

Early identification of metabolic dysfunction-associated steatotic liver disease (MASLD) is crucial for preventing disease progression, yet reliable predictive tools remain lacking. We analyzed proteomics data of 37,966 participants without baseline MASLD from southern United Kingdom, using multivariable Cox regression to conduct protein selection and model development. The top-ranking proteins were combined with clinical data and polygenic risk scores to develop models. The models were internally validated through 1,000 times bootstrap resampling within the Southern UK cohort, and externally validated in a preclinical Northern UK cohort (n = 14,986) and an inception Southern China cohort (n = 100). Cumulative MASLD incidences were compared across baseline protein concentration quintiles, and temporal trends in protein levels were assessed. Of the 2,737 proteins analyzed, five emerged as robust predictors for MASLD incidence within 5 years: CDHR2 (AUC = 0.812), GGT1 (AUC = 0.797), FUOM (AUC = 0.791), KRT18 (AUC = 0.789) and ACY1 (AUC = 0.777). Participants with higher baseline levels of these proteins faced markedly increased risks of MASLD (HRs = 7.05 - 9.81). In the Northern UK cohort, the combination of 5 proteins showed prominent validated predictive accuracy across different time frames: 5-year (AUC = 0.838), 10-year (AUC = 0.772), over 10-year (AUC = 0.742), all-time (AUC = 0.756). Combined with clinical predictors, the predictive performance is further enhanced: 5-year (AUC = 0.904), 10-year (AUC = 0.848), over 10-year (AUC = 0.799), all-time (AUC = 0.822). In the Southern China Inception cohort, the 5-Protein Panel also showed an outstanding performance (AUC = 0.912). Our study identified five key proteins, established and validated prominent predictive models of MASLD up to more than 16 years before onset, offering novel potential strategies for ultra-early prediction and intervention.