0041 Preliminary Identification and Validation of a Plasma Metabolome-Based Biomarker for Circadian Phase in Humans. (12th April 2019)

• Record Type: Journal Article
• Authors: Cogswell, Dasha; Markwald, Rachel R; Cruickshank-Quinn, Charmion; Quinn, Kevin; Melanson, Edward L; Reisdorph, Nichole; Wright, Kenneth P; Depner, Christopher M
• Is Part Of: Sleep. Volume 42(2019)Supplement 1
• Page Start: A17
• Page End: A17
• Publication Date: 2019-04-12
• Subjects: Sleep -- Physiological aspects -- Periodicals; Sleep disorders -- Periodicals; Sommeil -- Aspect physiologique -- Périodiques; Sommeil, Troubles du -- Périodiques; Sleep disorders; Sleep -- Physiological aspects; Sleep -- physiological aspects; Sleep Wake Disorders; Psychophysiology; Electronic journals; Periodicals; 616.8498
• DOI: 10.1093/sleep/zsz067.040
• Languages: English
• ISSNs: 0161-8105

Abstract

Abstract: Introduction: Identifying a reliable predictor of circadian phase is increasingly important to diagnose circadian disorders and circadian misalignment, inform treatment schedules, and support personalized medicine. Findings using blood transcriptomics to predict circadian phase show promise, but no published findings have utilized a metabolomics approach. Therefore, the potential utility of metabolomics to predict circadian phase is unknown. Here, we analyzed the plasma metabolome during adequate and insufficient sleep to identify a circadian phase biomarker. Methods: 16 (8M/8F) healthy participants aged 22.4±4.8y (mean±SD) completed a randomized cross-over in-laboratory study with 3 baseline days (9h sleep opportunity/night), followed by control (9h sleep) and insufficient sleep (5h) conditions, each lasting 5 days. Circadian phase was determined by dim light melatonin onset (DLMO). Blood was collected every 4 hours across 24 hours on the final day of each condition and used for aqueous and lipid LCMS metabolomics analyses. Two models were built to predict DLMO using Partial Least Squares Regression using the full dataset and a rhythmic metabolite-only (determined by MetaCycle R package) dataset. Each model was created by randomly splitting the data, with 66% used to train the model and 33% used as validation samples. Results: Using Leave-One-Out Cross-validation (LOOCV), R 2 for the full dataset was 0.58 using 7 components and 20 features. When validated on the holdout samples, R 2 was 0.37 with a median error of 3.5h±4.2 (median±IQR), and 31% of the validation samples had an error <2h. Using LOOCV, R 2 for the rhythmic metabolite-only dataset was 0.62 using 11 components and 50 features. When validated on the holdout samples, R 2 was 0.53 with a median error of 2.4±3.9, and 43% of the validation samples had an error <2h. During insufficient sleep, the rhythmic data set was non-significantly (P = 0.051, Wilcoxon t-test) better at predicting DLMO versus the full data set. Conclusion: Our preliminary findings show promising trends for metabolomics, especially the rhythmic metabolite dataset, however additional analyses with more subjects are required. Support (If Any): NIH-R01HL085705, NIH-R01HL109706, NIH-R01HL132150, NIH-F32DK111161, and NIH-UL1TR000154; and Sleep Research Society Foundation 011-JP-16