Mapping systemic inter-organ metabolic networks across glycemic continuum using whole-body [

purposeGlucose homeostasis relies on coordinated interactions among multiple organs, and its disruption relates to diabetes development. This study investigated how inter-organ metabolic coordination, assessed by whole-body [¹⁸F]FDG PET/CT, is altered across the glycemic continuum at both the population and individual levels, and whether individualized dysfunction features can improve early diabetes risk stratification.

methodsWe analyzed whole-body [¹⁸F]FDG PET/CT scans from 1,149 adults across two independent centers, classified into normoglycemic, pre-diabetic, and diabetic groups based on fasting glucose. Standardized uptake values normalized by lean body mass were extracted from 20 major organs. These values were further adjusted for age, sex, and BMI using general linear models to reduce demographic confounding. Group-level metabolic connectivity networks were constructed using bootstrapped correlation matrices with False Discovery Rate correction. To capture individual-level dysregulation, we generated deviation networks by quantifying how each subject's inter-organ coordination diverged from demographically matched normative reference patterns. These personalized network features were used to train and validate a machine learning model for classifying pre-diabetes versus diabetes.

resultsAt the population-level, network topological analysis revealed a decline in inter-organ metabolic connectivity from normoglycemia to pre-diabetes to diabetes. Pre-diabetes was marked by widespread but modest weakening of connections across multiple organs, while diabetes showed fewer but more concentrated disruptions, indicating a shift toward localized network breakdown. Specific alterations included reduced coordination between the kidneys in pre-diabetes, and disrupted connectivity between the brain and liver in diabetes. Individualized deviation networks captured subject-level differences in metabolic connectivity, with greater heterogenity observed in pre-diabetes. A machine learning model trained on these personalized features successfully distinguished diabetes from pre-diabetes (AUC = 0.75, external validation), with brain-peripheral connections emerging as the most informative predictors.

conclusionThis study reveals distinct patterns of inter-organ matbolic connectivity breakdowns across glycemic states and demonstrates that individualized network features can effectively capture subject-specific dysregulation.