Beyond organ age gaps: comparing plasma proteomic metrics of organ biology in relation to brain structure and neurodegenerative disease.

Abstract

Biological aging is heterogeneous across organ systems, and brain health may be influenced by biological processes extending beyond the central nervous system (CNS). This can be investigated using biological age measures that are increasingly used to estimate how biological systems deviate from chronological aging. However, organ level plasma proteomic metrics may capture different forms of biological variation depending on how they are defined. Here, we compared three complementary organ level proteomic metrics across 15 organ systems (and two global systems) in relation to age, organ-relevant physiological phenotypes, brain morphometry, and incident neurodegenerative disease, using plasma proteomics from 52,667 UK Biobank participants and brain MRI data from 5,078 of those individuals. We derived i) supervised age gap metrics using XGBoost models trained to predict chronological age, ii) unsupervised PC1 metrics capturing dominant organ-associated proteomic variation, and iii) phenotype guided PLS metrics linking plasma proteins to organ-relevant physiological traits. The three metric classes were related but not interchangeable. Global all-protein and all-organ-protein models showed strong age-related proteomic structure. However, organ-specific PC analyses showed that age was often but not always the dominant source of variation in protein signal across organs. Age gap metrics showed the most consistent associations with subsequent ICD-10-based neurodegenerative and vascular diagnoses, whereas all three metric classes showed distinct associations with brain structure and organ-relevant physiological phenotypes. Brain structure analyses revealed heterogeneous organ-brain association patterns across organs, suggesting that proteomic variation relates to brain structure in both a systemic and an organ-specific manner. Several systemic and organ-specific signals partially mediated the associations between disease specific genetic risk and incident diagnoses of Alzheimer's disease and stroke, albeit with small effect sizes (1-8%). Similarly, organ metrics provided modest incremental predictive improvement over baseline demographic models, with the largest increase in ROC-AUC observed for Alzheimer's disease when all organ age gap metrics were added jointly (delta ROC-AUC = 0.037). These results link brain related outcomes to proteomic variation across both CNS and non-CNS systems. Our work suggests that proteomic organ metrics capture complementary dimensions of biological variation rather than a single construct of organ aging. Although these measures may provide informative biological and potentially predictive signals, their incremental predictive value in the present analyses was modest.