A distinct metabolic signature in the saliva may help identify acute sleep deprivation. Researchers reported that a machine-learning model could be capable of distinguishing samples collected following total sleep deprivation from those collected following adequate sleep or sleep restriction with high precision, supporting the potential of oral-fluid metabolomics in future forensic and clinical applications.
In the randomized controlled crossover study, the researchers analyzed 440 oral-fluid samples collected from 20 healthy young male participants who completed three sleep conditions: one night of total sleep deprivation, 4 consecutive nights of sleep restriction to 6 hours per night, and a control condition with 8 hours of sleep. Using untargeted metabolomics and logistic regression approaches, the researchers sought to determine whether sleep loss could be identified from a single saliva sample without requiring a baseline sample from the same participant.
The primary analysis showed that acute sleep deprivation produced a distinct metabolic fingerprint. Linear discriminant analysis demonstrated clear separation between samples collected following a single night without sleep and samples collected during the control condition, whereas samples from the sleep-restriction and control conditions showed substantial overlap. The researchers reported that the metabolic differences associated with sleep deprivation were driven primarily by hydrophilic metabolites detected in the saliva samples.
The most clinically relevant classification model distinguished sleep deprivation from both the control and sleep-restriction conditions simultaneously. Following feature reduction, the model achieved an F0.5 score of 0.90 using just 12 molecular features. The model identified sleep-deprived samples with 94% precision and correctly detected 77% of samples collected during sleep deprivation against the other two groups. According to the researchers, these findings suggested that acute sleep deprivation may leave a measurable metabolic signature that can be detected without reference samples from the same individual.
The findings differed for moderate sleep restriction. Although the participants accumulated the same total sleep deficit across 4 nights of restricted sleep as in the total sleep-deprivation condition, the resulting metabolic changes were not sufficiently distinct to support robust classification. The researchers found considerable overlap between sleep-restriction and control samples and concluded that repeated restriction to 6 hours of sleep per night did not produce an exploitable metabolic signature under the study conditions.
The logistic regression model's performance also varied according to the time of sample collection. The metabolic fingerprint was most pronounced in samples collected during the morning and midday hours, with classification performance declining later in the day. The researchers reported that prediction margins were largest between 8:00 AM and noon and became narrower toward the evening, particularly at 11:00 PM. Nevertheless, correct classifications substantially outnumbered incorrect classifications across all time points.
The researchers emphasized that the findings should be interpreted with caution because the study population involved only healthy young male participants with habitual sleep schedules of seven to nine hours per night. The models were also developed under controlled experimental conditions. Whether the same metabolic signature can be detected in female participants; older adults; shift workers; patients with comorbidities; or those using medications, stimulants, or other substances remains unknown. The researchers also noted that the study was designed to identify sleep-deprived states rather than directly measure sleep-related impairment.
The researchers concluded that saliva-based metabolomics shows promise as a tool for identifying acute sleep deprivation but requires further validation prior to practical implementation.
"Metabolomics-based, reference-free sleep loss detection holds potential for applications in forensic, clinical, and occupational contexts," wrote lead study author Michael Scholz, of the Department of Forensic Pharmacology and Toxicology at the Zurich Institute of Forensic Medicine at the University of Zurich, and colleagues.
The study was funded by the Fund for Road Safety. Scholz and senior study author Thomas Kraemer of the University of Zurich reported filing a patent related to metabolic sleepiness markers in oral fluid. The study authors reported no other conflicts of interest.
Source: Journal of Proteome Research