SUN-428 Identification Of Serum Metabolites That Predict Glucocorticoid Action In Adipose Tissue

Abstract Disclosure: K. Bennett: None. D. Chantzichristos: None. T. Garner: None. P. Jansson: None. J. McQueen: None. G. Johannsson: None. A. Stevens: None. Introduction: Personalising glucocorticoid (GC) replacement therapy in adrenal insufficiency (AI) remains challenging due to the lack of validated biomarkers that reflect GC action at the tissue level. Serum markers reflecting GC action on adipose tissue (AT) have not been defined. Measuring relationships between two or more metabolites may allow for better prediction of GC action; these can be modelled as higher order interactions (HOIs), defined by hypergraphs and topological data analysis (TDA). This clinical biomarker discovery study aimed to identify and validate HOIs between serum metabolites reflecting GC action in AT. Methods: Ten patients with primary AI participated in a randomized cross-over trial, receiving either intravenous saline (GC withdrawal) or a circadian hydrocortisone infusion (physiological GC exposure), with a washout period of more than two weeks. After 26 hours (8:00 AM) samples were collected, including adipose tissue (AT) biopsies for transcriptomic analysis, and serum and abdominal subcutaneous microdialysis for metabolomics profiling via LC&GC/MS. Multi-omic integration was performed using hypergraph-based modelling and TDA to explore relationships between circulating metabolites and AT molecular signatures. Predictive performance of serum metabolite ratio pairs was assessed using ROC curve analysis. Findings were validated in an independent 8 day cross-over study (n = 15), comparing high (1.25 mg) vs low (0.5 mg) doses of betamethasone. Results: Consistent matched metabolic signals and coordinated GC-driven changes were observed across serum and AT. Integrated networks revealed distinct topological features between GC exposure and withdrawal, with differences identified in the 1st and 2nd homological dimensions as defined by TDA. These structural changes highlight differences in network architecture under each condition, presenting a novel framework to investigate biomarkers that discriminate between GC states. Serum metabolite marker pairs were identified that outperformed single metabolites as predictive markers of GC state (AUC 0.95-1, FDR < 0.05). The top 15 pairs were investigated within our network models, where the ratio of phenylalanine:kynurenine was the most highly connected, supported by well-documented AT GC-response elements, such as TXNIP and KLF9. This marker pair was validated in the dose-response study, with a significant difference between high and low dose (paired t-test, p < 0.01). Discussion: Integrative models were generated using hypergraphs and TDA, revealing distinct molecular networks across GC states. Relationships between serum metabolites were identified as markers of GC state in AT, with the ratio of phenylalanine:kynurenine validated as a promising biomarker to support personalised GC replacement therapy. Presentation: Sunday, July 13, 2025