Quantifying lifespan trajectories and individual variability in multiple classes of spindle-like events in the sleep electroencephalogram

Abstract

The human brain undergoes many changes throughout the lifespan, making it critical to distinguish expected trajectories from pathological deviations in order to diagnose disease states. This thesis develops a new framework for understanding lifespan changes in the brain based on transient oscillatory activity during sleep, for the first time tracking individualized signatures of sleep across the lifespan and identifying novel lifespan dynamics in the activity of multiple classes of activity, providing a foundation for improved biomarkers of neurodevelopment and aging. Sleep offers a powerful window into health, with sleep dysfunction associated with nearly every neurological, psychiatric, neurodevelopmental, and neurodegenerative disorder. In particular, sleep spindles, 9-16Hz bursts of transient oscillatory activity that define stage N2 of sleep, have long been of interest as they are thought to be important for memory consolidation and display changes with age as well as across different disorders such as Schizophrenia, Autism and Alzheimer’s disease. Recent work has also suggested the existence of spindle-like events outside the sigma range (9-16 Hz), including theta-alpha ( 5–9 Hz) and delta ( 2–5 Hz) activity. To characterize how spindle-like classes appear across a large population and change with age, we characterized 1000s of spindle-like transient oscillatory events across 2-16Hz throughout the night for each of 725 subjects in a large cross-sectional dataset spanning 10-80 years (Cleveland Family Study, 55% female, average 41 years old). These spindle-like events formed individualized spindle- like classes within each individual and we extracted the features of these classes by fitting parametric models to each individual’s data. The features of these individually detected spindle-like classes formed clusters representing four population-level classes of activity which we termed σ_high (~12-16 Hz), deep-σ_low (~9-12 Hz), θ − α (~5-9Hz), and δ (~2-5 Hz). We built population level models for each class describing how the features changed as a function of age, sex, and electrode. We found statistically significant changes across multiple features (fre- quency, sleep depth, density, phase coupling to the slow oscillation) as a function of age, sex, and electrode. Importantly, some features changed continuously across the lifespan, while others shifted more sharply during specific developmental periods, suggesting neural markers of early life, mid-life, and late-life transitions. For example, early life ( 10-25 years) is marked by statistically significant steep increases in the frequency of theta spindle-like activity. This novel trajectory may serve as a neurodevelopmental biomarker. These findings highlight the robustness of spindle-like activity beyond the sigma range in a large population and the potential for previously unstudied biomarkers of development and aging. Despite clear population-level trends, individuals varied in which spindle-like classes they ex- pressed. By defining endophenotypes based on class co-expression, we uncovered structured inter- individual variability not captured by group-level models. These endophenotypes may offer a new axis for patient stratification and for linking neural activity to clinical outcomes. Given the observed heterogeneity in the spindle-like activity across individuals, it is possible that a more informative metric of change is relative to one’s self. To work towards a quantitative framework linking longitudinal change to health, we conducted preliminary analyses to assess the within- individual stability in about 1900 subjects with 2 visits 5 years apart from the Sleep Heart Health Study. Our preliminary findings found that about 30% of subjects were most similar to themselves over a five-year interval, and approximately 50% ranked within the top ten closest matches across a variety of pixel-based measures. This suggests that the spindle-like activity has stability that can be leveraged to link longitudinal change to outcomes within a quantitative framework and that simple pixel-based approaches represent a reasonable starting point for defining distance measures within such a framework. By quantifying individualized, multidimensional changes in sleep-related brain activity across the lifespan, this work establishes a foundation for identifying atypical neural trajectories and enables future efforts to detect early signs of pathology, personalize diagnostics, and develop precision sleep- based biomarkers for brain health