0042 Proteomic Biomarkers Of Circadian Time

Abstract

Introduction: Our ability to incorporate circadian timing into clinical decision-making is impaired by our inability to measure circadian phase efficiently. This impacts treatment for patients with suspected circadian rhythm disorders and insomnia or hypersomnia. The current Dim Light Melatonin Onset (DLMO) method for assessing circadian time is expensive and necessitates multiple samples over several hours. We utilized an aptamer approach to quantify circadian time. Methods: A high throughout aptamer array was used to profile 1013 plasma proteins in 28 individuals assigned to three multisite cohorts. Cohort1 had 6 participants sampled every 4 hours in a 24-hr interval in constant routine and after a night of inverted sleep. Cohort2 consisted of 9 individuals sampled every 2hrs over a 36hr constant routine. While 70% of Cohort1 and Cohort2 were used as a training set in the process of building a circadian time predictor, Cohort3, comprised of 13 individuals sampled at 12 timepoints in a 24hr interval was left untouched to be used as a validation cohort along with 30% of Cohort 1 and 2. Harmonic regression was used to find circadian regulated proteins. A machine learning multivariate circadian time predictor with elastic net regularization was built on the training data and validated in two independent cohorts. Results: Harmonic analysis revealed 129 plasma proteins (FDR p=0.005) to be circadian regulated; strikingly ACTH and pro-opiomelanocortin peaked in early morning and tapered as the day progressed, consistent with established circadian time stamps. In addition, we observed TSH, PTH, pancreatic hormone precursor and ghrelin proteins to show diurnal rhythmicity. A circadian time predictor that was trained on 70% of samples from Cohort1 and Cohort2 performed robustly, achieving an overall mean accuracy of 86% and median absolute error of 1.40hrs and 1.58hrs across the two validation cohorts, respectively. In 80% of the samples, the prediction error was less than 2hrs. Conclusion: A proteomic-based approach to quantify circadian time is a reliable and robust alternative to conventional DLMO and gene expression based methods.