Plasma proteome-based machine learning effectively predicted ectopic fat content across multiple organs, with liver fat prediction achieving an R2 of 0.74 and AUC of 0.94.
Observational (n=666)
Plasma proteomic profiling combined with multi-organ MRI identifies tissue-specific molecular drivers of ectopic fat accumulation, offering potential novel targets for cardiometabolic diseases.
ABSTRACT Ectopic fat deposition is a major risk factor for cardiometabolic diseases, but its systemic molecular drivers and disease associations remain poorly understood, owing to the reliance on correlation studies, a single‐organ perspective, and limited disease coverage. Plasma proteomic data (1454 proteins) and multi‐organ MRI from 666 UK Biobank participants were integrated. A knockoff‐based causal inference framework was applied to identify plasma proteins with putative causal effects on fat accumulation in the liver, pancreas, skeletal muscle, and pericardium. Tissue‐specific machine learning models, functional enrichment analysis and systematic assessment of associations with 70 diseases were performed. Results showed that ectopic fat distribution varied substantially across tissues. Plasma proteome‐based machine learning effectively predicted fat content across multiple organs, with a consistent performance hierarchy in both regression and classification tasks: liver ( R 2 = 0.74, AUC = 0.94) > pancreas > muscle > pericardium ( R 2 = 0.20, AUC = 0.67). Causal inference identified 94 organ‐specific driver proteins, encompassing regulators of metabolism, inflammation, and signaling. Among these, PLA2G1B emerged as the strongest driver for pancreatic fat; ERBB2 and IGFBP2 acted as key positive and negative regulators of liver fat, respectively; CA14 and muscle fat formed a clear negative feedback loop; pericardial fat was primarily driven by NCAM2 and TNFRSF10B. Enrichment analysis revealed distinct, tissue‐specific biological programs: pancreatic fat was linked to angiogenesis and humoral immunity; liver fat was associated with cytokine activity and PI3K‐Akt signaling; muscle fat was connected to BMP/TGF‐β signaling and extracellular matrix organization. Prognostically, muscle fat broadly correlated with cardiometabolic, neurological, and digestive diseases; liver fat was strongly linked to diabetes and hypertension; pancreatic fat was associated with metabolic conditions including diabetes and liver disease. Ectopic fat accumulation is driven by tissue‐specific molecular mechanisms with distinct disease associations. This causal proteomic atlas provides novel regulatory circuits and protein targets, advocating for tissue‐targeted diagnostic and therapeutic strategies.
Zhang et al. (Mon,) conducted a observational in Ectopic fat distribution (n=666). Plasma proteins was evaluated on Fat accumulation in the liver, pancreas, skeletal muscle, and pericardium. Plasma proteome-based machine learning effectively predicted ectopic fat content across multiple organs, with liver fat prediction achieving an R2 of 0.74 and AUC of 0.94.