Date of Award

January 2026

Document Type

Thesis

Degree Name

Medical Doctor (MD)

Department

Medicine

First Advisor

Tore Eid

Abstract

Authors: Ami Mange1,2, Kan Yaovatsakul2, Ryan Bose-Roy2, Haley Cox3, Adebayo Efunnuga3, Joshua Hobbs2, Yuqing Huang2, James Poe1, Haleh Nadim1, Roni Dhaher1, Aline Herlopian2, Robert Duckrow2, Dennis Spencer3, Hitten Zaveri2, Tore Eid1Affiliations: Departments of 1Laboratory Medicine, 2Neurology, 3Neurosurgery; Yale School of Medicine, New Haven, CT Background: Epilepsy affects over 70 million people worldwide, and the unpredictability of seizure onset is among its most disabling features – ranked by patients above seizure severity. Prior work has established that seizures cluster at characteristic phases of circadian and multidien biological cycles, and that neurochemical changes, including increases in extracellular isoleucine and decreases in GABA, precede spontaneous seizures by several hours in humans and animal models. Saliva is a biochemically rich, non-invasive biofluid that mirrors plasma chemistry and contains amino acids, organic acids, steroid hormones, and bile acids relevant to epilepsy pathophysiology, making it well suited to longitudinal monitoring. Because peripheral metabolites are reflected in brain extracellular chemistry, saliva represents a promising non-invasive window into the neurochemical environment underlying seizure susceptibility; however, distinguishing epileptic seizures from psychogenic non-epileptic seizures (PNES) remains a major clinical challenge that a salivary biomarker approach could also address. A critical prerequisite for reliable salivary biomarker research – the systematic characterization of metabolite stability under variable pre-analytical conditions – has not previously been reported for the analyte classes of interest. Aims: This thesis addresses two specific aims: (1) to identify salivary metabolite signatures that distinguish epilepsy from PNES in patients undergoing inpatient video-EEG monitoring at the Yale Epilepsy Monitoring Unit (EMU); and (2) to characterize the pre-analytical stability of key salivary metabolite classes under variable storage temperatures and durations. Hypothesis: We hypothesize that a latent biochemical variable for seizure likelihood can be identified using serial salivary sampling in people with epilepsy, and that this signal can be used to develop individualized, non-invasive seizure forecasting algorithms. Methods: Saliva samples were collected by passive drool from 30 adults undergoing inpatient video-EEG monitoring in the Yale EMU; 18 contributed samples to analysis (14 epilepsy, 4 PNES). Samples were collected six times daily and within one hour of any clinical seizure, stored immediately at -20°C at the bedside, and transferred to -80°C at the conclusion of each admission. Amino acids (45 analytes) and organic acids (73 analytes) were quantified by liquid chromatography-tandem mass spectrometry (LC-MS/MS). Samples were classified by temporal proximity to seizures as baseline, pre-seizure, or post-seizure. LASSO-penalized logistic regression and decision tree classifiers were applied to 130 baseline samples to distinguish epilepsy from PNES. For the stability experiment, saliva was collected from a healthy volunteer on two independent dates and aliquoted into conditions spanning room temperature (~22°C), 4°C, and -20°C across timepoints from 1 hour to 4 weeks, with immediately frozen -80°C samples serving as the reference. Amino acid concentrations were quantified by LC-MS/MS and analyzed descriptively. Results: LASSO logistic regression achieved a cross-validated AUC of 0.97 and sample-level accuracy of approximately 90% (sensitivity 91%, specificity 95%) for epilepsy classification. Key discriminating features included tryptophan and glutamine (enriched in epilepsy) and alpha-hydroxy isovaleric acid, leucine, citramalic acid, and threonine (enriched in PNES). Results were robust across normalization strategies, with the majority of normalization pairs yielding AUC > 0.80. Decision tree classifiers identified citramalic acid and glutamic acid as the most consistently selected features across all tree depths and normalization schemes. At the patient level, 18 of 18 participants were correctly classified when sample counts were balanced. In the stability experiment, room temperature storage beyond 6 hours produced large, directional concentration increases in the majority of amino acids, with some exceeding 10-fold by 24 hours, while phosphoethanolamine degraded by more than 90%. Refrigeration at 4°C attenuated but did not fully prevent these changes. Storage at -20°C preserved the majority of analytes through 4 weeks. Impact: This work demonstrates for the first time that salivary metabolomics can distinguish epilepsy from PNES with high accuracy using a non-invasive, easily collected biofluid, and establishes evidence-based guidelines for salivary sample handling broadly applicable to clinical metabolomics research. Together, these findings provide a methodological and scientific foundation for saliva-based seizure forecasting. The ultimate translational goal of this research is the development of an implantable oral biosensor capable of continuously monitoring salivary metabolites and providing patients with real-time, personalized estimates of seizure risk – a tool that could transform the lives of people living with the daily burden of unpredictable seizures.

Comments

This thesis is restricted to Yale network users only. This thesis is permanently embargoed from public release.

Download

Share

COinS