Perivascular Dynamics, Sleep Apnea, and Memory: Interpreting a Moving Target

In this issue of the Journal, Lee and colleagues (pp. 2382–2392) present longitudinal evidence from a large, community-based cohort linking changes in obstructive sleep apnea (OSA) severity to changes in a diffusion magnetic resonance imaging (MRI) index of perivascular water movement, diffusion tensor imaging (DTI) analysis along the perivascular space (ALPS), and, in turn, changes in visual memory performance over 4 years (1). These findings extend earlier cross-sectional reports by demonstrating temporal coupling of disease burden, perivascular signal, and cognition in the same individuals, within a design that incorporates repeat polysomnography, advanced neuroimaging, and detailed neuropsychological testing.

The authors’ approach has several strengths. More than 1,000 adults underwent home polysomnography, DTI, and a comprehensive cognitive battery at baseline and 4-year follow-up. OSA severity was defined categorically and continuously using the apnea–hypopnea index (AHI), and associations were examined in cross-sectional and longitudinal frameworks. Across analytic strategies, higher AHI correlated with lower DTI-ALPS index; worsening OSA over time corresponded to greater DTI-ALPS index decline, whereas improvement was accompanied by relative increases. Change in DTI-ALPS index was associated with change in visual memory and mediated the relationship between AHI and visual memory. Adjustment for mean diffusivity, an important potential confounder, did not materially alter these results.

As acknowledged by the authors, the DTI-ALPS metric itself warrants careful interpretation. Analysis along the perivascular space estimates water diffusivity orthogonal to major white-matter tracts in deep white matter (2). Although sometimes described as a proxy for “glymphatic” function, it is more accurately regarded as an index of perivascular diffusivity influenced by white-matter microstructure, vascular pulsatility, and age. The authors explicitly avoid overinterpreting DTI-ALPS index as a direct clearance measure, a wise choice given the ongoing debate about whether the glymphatic model, as originally formulated, fully explains perivascular physiology in humans. Although large-scale cerebrospinal fluid and vascular oscillations occur during non-REM sleep, a debate is ongoing, with some authors suggesting that net solute clearance may not be globally enhanced during sleep and that local fluid dynamics can depend on vascular tone, arousal state, and anesthetic condition (3). By framing the DTI-ALPS index in neutral mechanistic terms, Lee and colleagues preserve the validity of their inferences while keeping the door open to multiple explanatory pathways.

An alternative, and perhaps complementary, framework views these findings through the lens of sleep microstructure and arousal–vascular coupling. Brief, infraslow noradrenergic surges during non-REM sleep microarousals, not unlike cyclic alternating patterns (CAPs) (4), have been argued to coordinate changes in sleep spindles, vasomotion, cerebral blood volume, and perivascular fluid movement (5). In OSA, repeated respiratory events produce frequent arousals, often shifting the CAP from more synchronized, potentially restorative subtypes to more desynchronized, wake-promoting subtypes as severity increases (6, 7). Such a shift may impair the coupling between arousal-related vascular pulsatility and astroglial water handling, resulting in lower daytime DTI-ALPS index without invoking a single, sleep-specific clearance mechanism. This “arousal– vascular decoupling” hypothesis is consistent with the observed longitudinal associations and can be tested directly by combining high-density polysomnography with perivascular imaging.

The specificity of the cognitive association reported by Lee and colleagues (1) is particularly interesting. Visual memory, rather than verbal memory or executive function, was linked to DTI-ALPS index change. OSA has previously been linked with memory deficits, increased biomarkers of Alzheimer’s disease, and heightened risk of cognitive decline and dementia (8, 9). In Alzheimer’s disease, early deficits are typically verbal-amnestic, even though visual memory impairment can antedate diagnosis in some individuals and is associated with occipitotemporal and medial temporal pathology. Vascular cognitive impairment often presents with executive and processing speed deficits, whereas dementia with Lewy bodies is characterized by visuospatial impairments. The pattern here, preferential association with visual memory, perhaps suggests potential medial temporal involvement with occipitotemporal contributions, compatible with a mixed Alzheimer–vascular pathway. This is not proof of causality, but it offers a direction for hypothesis generation and for future phenotyping of OSA-related cognitive change.

Other work strengthens this line of reasoning. In older adults, REM-predominant hypoxemia has been associated with greater white-matter hyperintensity burden in frontal and parietal lobes, which in turn relates to entorhinal cortical thinning and impaired mnemonic discrimination after sleep (10, 11). Such findings suggest that stage-specific hypoxemia may accelerate small-vessel injury with downstream medial temporal effects, a pattern that sits comfortably alongside the mediation pathway demonstrated by Lee and colleagues.

The strengths of this study include its large size, community-based sampling, repeated multimodal assessments, and converging analytic approaches. Adjusting for mean diffusivity strengthens the inference that the DTI-ALPS index association is not simply a surrogate for microstructural injury. The mediation analysis, although modest in effect size, is methodologically rigorous and cautiously interpreted. Limitations are also clear: single-night home polysomnography risks misclassification of OSA severity, reduced EEG montage precluded analysis of slow-wave activity and CAP subtypes, daytime MRI acquisitions cannot isolate sleep-state effects, and continuous positive airway pressure exposure was minimal, preventing treatment inference. Finally, the DTI-ALPS index’s lack of specificity for clearance processes remains an interpretive constraint.

Clinically, this work underscores the value of domain-specific cognitive assessment in OSA, particularly for visual memory in mid- to late-life adults. It also prompts reconsideration of what constitutes meaningful disease burden; modest overall AHI may still confer cognitive risk if hypoxemia or arousals are concentrated in vulnerable sleep stages such as REM. From a research perspective, the findings argue for future studies that integrate detailed sleep microstructure analysis, including arousal phenotyping, with harmonized perivascular imaging pipelines and complementary MRI measures such as perivascular space volumetry and free-water mapping. Stratification by age, sex, and genetic risk factors may clarify susceptibility patterns. Given accumulating evidence that OSA is associated with neuroinflammation (12), mechanistic work should also examine antiinflammatory strategies, potentially also including metabolic interventions such as glucagon-like peptide-1 receptor agonists, which have shown promising, if yet preliminary, neuroprotective and cognitive benefits in other populations (13, 14), alongside established OSA therapies.

Lee and colleagues have provided robust longitudinal evidence linking OSA burden, perivascular imaging, and cognition in a general population sample. Whether this reflects impaired glymphatic clearance, disrupted arousal–vascular coupling, or a combination of the two, the signal is a tractable target for mechanistic and interventional work. Their careful design and measured interpretation set a high standard, and their findings open a space for nuanced, physiology-informed strategies to protect brain health in OSA.