Data for this study were downloaded from the ADNI database on April 21, 2022. ADNI is a longitudinal observational study of aging that enrolls participants diagnosed as CN, subjective memory concerns (SMC), MCI (both early and late stages), and AD dementia. This study focused solely on data from CN individuals aged 55–90 years, collected between March 2011 and October 2021. CN is defined as having no impairment in cognitive function, with a Clinical Dementia Rating Global Score (CDR-SB) of 0, a Mini-Mental State Examination (MMSE) score ranging from 24 to 30, and normal memory functioning assessed using the Logical Memory II subscale (Fuller et al., 2020). A full description of the inclusion/exclusion criteria for the ADNI study can be found at https://adni.loni.usc.edu.

Participants provided demographic data upon enrollment, including age, sex, education level, and marital status. We also documented APOE ε4 genotype status, a known genetic risk factor associated with increased Aβ burden in the brain. Global cognitive function was assessed using the CDR-SB and MMSE.

The presence of sleep disturbance was determined using the Neuropsychiatric Inventory (NPI), a validated instrument covering 12 major behavioral domains with established interrater reliability and test–retest reliability (Cummings et al., 1994; Mega et al., 1996; Cummings, 1997). Component K of the NPI (NPI-K) assesses recent alterations in sleep patterns. Previous research has demonstrated associations between NPI-K scores and regional uptake of both 18F-flortaucipir and 18F-florbetapir uptake (Shokouhi, 2019). The NPI employs a structured hierarchical assessment approach, initiating with screening questions to identify symptoms within specific behavioral domains. According to previous studies (Elberse et al., 2024; Kim et al., 2023; Blackman et al., 2022), we established a specific screening protocol for sleep assessment in our study. Participants were categorized into two groups according to their sleep status. Those who reported sleep disturbance were classified into the sleep disturbance group, defined by an affirmative response to any of these key sleep questions: difficulty initiating sleep (K1), nighttime awakenings (excluding isolated bathroom visits with rapid sleep resumption; K2), or premature morning awakening relative to established sleep patterns (K6). Additionally, regarding the sleep assessment, we would like to clarify that the total severity score was calculated for NPI-K by summing up the severity ratings for all domains of sleep and nighttime behaviors. The sleep scores represent the sum of multiple items (higher scores indicate worse sleep quality).

We used the standardized uptake value ratio (SUVR) obtained using florbetapir-PET-AV45 to calculate Aβ burden. ADNI florbetapir PET scans were acquired using standardized ADNI PET protocols at the participating sites. FreeSurfer v7.1.1 delineated the regions of interest (ROIs). We used the structural MRI closest in time to each PET scan to rule out potential effects of brain atrophy associated with baseline MRI registration. Further details can be found in the ADNI_UCBERKELEY_AV45_Methods_01_14_21. The SUVR is defined as the ratio of the measured uptake in a target tissue ROI divided by the uptake in a reference ROI (Zasadny and Wahl, 1993). The choice of the reference ROI directly affects the sensitivity of SUVR quantification. The cerebellum has been widely used as a reference for florbetapir PET SUVR, especially in cross-sectional studies (Schwarz et al., 2017; Chiao et al., 2019). Here, we re-intensity-normalized the regional SUVR using a composite reference ROI of the whole cerebellum. The global 18F-Flortaucipir comprises frontal, anterior/posterior cingulate, lateral parietal, and lateral temporal regions.

At the time of data download, we identified 2,756 participants with available data from the three ADNI phases (ADNI-1, ADNI-2, and ADNI-3). Participants from all phases were eligible for inclusion, provided they met our study criteria. This cross-phase approach maximized our sample size by utilizing all available ADNI data. We applied systematic exclusion criteria as follows: 1,524 participants were excluded due to MCI or AD diagnoses. Additional exclusions comprised 8 participants with incomplete NPI-sleep questionnaires, 23 with missing APOE ε4 genotype data, 12 with incomplete neuropsychological assessments, and 168 individuals with non-zero CDR-SB scores. Subsequently, we applied Z-score standardization to the summary SUVR values (summarysuvr_wholecerebnorm, based on whole cerebellum reference region) and excluded 21 statistical outliers defined as values exceeding three standard deviations from the mean (|Z| > 3). Following these inclusion and exclusion procedures, 1,000 participants with complete summary SUVR data and regional measurements from 103 brain regions were retained for final analysis. A complete list of all 103 brain regions from FreeSurfer’s whole-brain segmentation, comprising the Desikan-Killiany cortical parcellation (68 regions) and FreeSurfer’s subcortical segmentation (35 additional structures), is provided in the Supplementary Table S7.

Baseline demographic characteristics were compared using independent t-tests or analysis of variance (ANOVA) for continuous variables and Pearson χ² tests for categorical variables, as appropriate.

Regional SUVR values were compared between participants with and without sleep disturbance using independent t-tests. For descriptive purposes, unadjusted means and standard deviations are presented for the overall sample and each sleep group. Linear regression analyses were performed for each brain region with regional SUVR values as the dependent variable and sleep disturbance status as the primary predictor. Participants without sleep disturbance served as the reference group. All regression models were adjusted for age, sex, years of education, and marital status. Results are presented as regression coefficients (β) with standard errors (SE) and corresponding p-values. p-values from regression analyses were corrected for multiple comparisons using the Benjamini-Hochberg false discovery rate (FDR) method.

To examine the moderating effect of APOE ε4 status on the relationship between sleep disturbance and regional amyloid burden, we conducted moderation analyses using linear regression framework with separate models fitted for each brain region. These models included sleep disturbance as the primary predictor, APOE ε4 carrier status as the moderator, and their interaction term (Sleep disturbance × APOE ε4). Participants carrying ≥1 copy of the APOE ε4 allele (i.e., ε2/ε4, ε3/ε4, and ε4/ε4) were classified as APOE ε4+, while all others were classified as APOE ε4−. We also conducted stratified linear regression analyses by APOE ε4 carrier status to examine the association between sleep disturbance and regional SUVR values separately in carriers and non-carriers.

In our regression models, sleep disturbance was coded as a binary variable (0 = no sleep disturbance, 1 = sleep disturbance present) and APOE ε4 status was coded as a binary variable (0 = non-carrier, 1 = carrier). This coding scheme means that individuals with sleep disturbance and APOE ε4 carriers serve as the reference conditions in our analyses. Therefore, negative beta coefficients indicate that the comparison group (no sleep disturbance or non-carriers) has lower SUVR values compared to the reference group (sleep disturbance present or carriers).

Additionally, we investigated the association between sleep scores measured by NPI-K and SUVR values in the APOE ε4 + group using linear regression analyses with the aforementioned covariates. To visualize regional differences in amyloid burden associated with sleep disturbance in APOE ε4 + individuals, we used AFNI (Analysis of Functional NeuroImages) and SUMA open-source software. Forest plots were generated using R software to visualize regression coefficients and confidence intervals across brain regions.

We first conducted exploratory analyses to identify brain regions showing significant associations at the uncorrected p-value level (p < 0.05), derived from regression models adjusted for age, sex, years of education, and marital status. To address the multiple comparisons issue, we applied Benjamini-Hochberg false discovery rate (FDR) correction to the analytical results across all 103 brain regions. FDR correction is particularly appropriate for neuroimaging studies due to the inter-correlations among brain regions.

All statistical analyses were performed using IBM SPSS Statistics (version 28.0; IBM Corp., Armonk, NY, USA), Mplus version 8.3, and R software version 4.0.3 (R Foundation for Statistical Computing). Statistical significance was set at p < 0.05.

No significant difference in global SUVR was detected between individuals with and without sleep disturbance (1.10 ± 0.15 vs. 1.11 ± 0.17, p = 0.256). Further analysis of regional SUVR values using unadjusted independent t-tests revealed significant differences in 8 ROIs between participants with and without sleep disturbance, including the central corpus callosum, mid-anterior corpus callosum, left banks of superior temporal sulcus, left cuneus, left lingual gyrus, left pericalcarine, left superior temporal, and left precuneus (Supplementary Table S2). These regions represent functionally distinct neural networks with well-established roles in cognitive processing. The corpus callosum regions facilitate critical interhemispheric connectivity and information transfer between cerebral hemispheres. The temporal regions, specifically the banks of superior temporal sulcus and superior temporal cortex, are primarily involved in language processing and auditory function—cognitive domains that are characteristically impaired during AD progression. The visual cortex areas, including the cuneus, lingual gyrus, and pericalcarine cortex, are responsible for visual information processing, representing another functional domain known to be compromised with AD advancement. Notably, the precuneus serves as a pivotal hub within the default mode network (DMN), a brain network that is particularly vulnerable to early AD pathology. This selective pattern of regional amyloid deposition suggests that sleep disturbance may preferentially target specific neural networks that are known to be vulnerable in AD, rather than causing indiscriminate amyloid accumulation across the entire brain. Such network-specific effects support the hypothesis that sleep-related amyloid deposition follows established pathways of AD-related neurodegeneration.

However, after adjusting for sex, age, education, and marital status, only the left cuneus region remained statistically significant (β = 0.024, SE = 0.012, p = 0.04; Supplementary Table S2), indicating that participants with sleep disturbance showed higher Aβ deposition in the left cuneus compared to those without sleep disturbance. As a key component of the DMN that is highly active during resting state, the cuneus has been consistently implicated in AD pathogenesis, with Aβ accumulation preferentially starting in several core DMN regions. Recent studies have further identified the cuneus/precuneus as a central hub for brain functional connectivity alterations in sleep-related cognitive impairment (Mattioli et al., 2021; Horovitz et al., 2009; Lunsford-Avery et al., 2020). These findings collectively suggest that sleep disturbance may preferentially target Aβ deposition in vulnerable DMN regions such as the left cuneus, potentially representing an early marker of sleep-related neurodegeneration risk in CN older adults.

Given the established role of genetic factors in AD pathogenesis, we further examined whether individual genetic susceptibility modulates the relationship between sleep disturbance and Aβ deposition. APOE alleles represent established genetic risk factors for AD, with multiple studies demonstrating that APOE ε4 facilitates Aβ seeding and accelerates Aβ aggregation in cerebral tissues (Liu et al., 2017; Dodart et al., 2005). We found that the interactions between sleep disturbance and APOE ε4 were associated with regional SUVR. Complete results for all brain regions are presented in Supplementary Table S3, while Table 2 summarizes the significant findings with corresponding FDR-corrected p-values. We initially conducted exploratory analyses without FDR correction to identify potential regions of interest. The negative beta coefficients indicate lower SUVR values in individuals without sleep disturbance and APOE ε4 non-carriers compared to their respective reference groups, consistent with our hypothesis that sleep disturbance is associated with increased regional Aβ deposition.

We reported effects of sleep disturbance and presence of the APOE ε4 on regional SUVR, presenting brain regions showing significant associations in covariate-adjusted regression analyses. Before FDR correction, 13 brain regions showed significant associations with sleep disturbance or sleep disturbance*APOE ε4 interactions after adjusting for age, sex, years of education, and marital status, including multiple corpus callosum regions (anterior, central, mid-anterior, and mid-posterior), left caudate nucleus, bilateral cerebellar cortex, bilateral choroid plexus, left hippocampus, bilateral inferior lateral and lateral ventricles, and white matter hypointensities.

Following FDR correction for multiple comparisons, several regions showed significant sleep disturbance*APOE ε4 interactions, including the left choroid plexus (β = 0.120, SE = 0.032, p = 1.50 × 10⁻⁴, q = 0.004), bilateral lateral ventricles (left: β = 0.141, SE = 0.034, p = 3.00 × 10⁻⁵, q = 0.002; right: β = 0.136, SE = 0.033, p = 4.00 × 10⁻⁵, q = 0.002), mid-posterior corpus callosum (β = 0.075, SE = 0.034, p = 2.70 × 10⁻⁴, q = 0.006), right choroid plexus (β = 0.112, SE = 0.029, p = 1.10 × 10⁻⁴, q = 0.004), and white matter hypointensities (β = 0.118, SE = 0.035, p = 7.50 × 10⁻⁴, q = 0.013). Notably, a significant sleep disturbance*APOE ε4 interaction was observed in the left hippocampus (β = 0.049, SE = 0.020, p = 0.015, q = 0.169). This finding is particularly meaningful given the hippocampus’s critical role in memory formation and its susceptibility in early AD pathogenesis.

The distribution pattern of these regions suggests that sleep disturbance may preferentially affect DMN-related structures and their supporting systems. Specifically, DMN core impairment is evidenced by Aβ deposition in the hippocampus, a key DMN node directly involved in memory processing. DMN connectivity disruption may result from corpus callosum Aβ deposition, which could compromise interhemispheric connections critical for DMN function. Additionally, the observed Aβ deposition in the ventricular system and choroid plexus suggests potential clearance system impairment, which may affect glymphatic clearance mechanisms that are crucial for Aβ removal during sleep.

The significant interaction analysis results prompted us to perform subgroup analyses, and participants were divided into two groups based on APOE ε4 status (Table 3; Supplementary Tables S4, S5). Among APOE ε4 non-carriers, bilateral lateral ventricle SUVR was significantly decreased in those with sleep disturbance (left: β = −0.039, SE = 0.017, p = 0.019, q = 0.140 and right: β = −0.034, SE = 0.017, p = 0.042, q = 0.140) after adjusting for age, sex, education, and marital status (Supplementary Table S4). Although these associations did not remain significant after FDR correction, this finding was contrary to our expected results. In contrast, when analyzing the entire population without considering APOE status and focusing only on sleep disturbance (Table 2), both brain regions showed a trend toward increased SUVR. Moreover, among APOE ε4 carriers, both regions demonstrated significantly elevated SUVR that remained significant even after FDR correction (Supplementary Table S5). This pattern suggests different responses to sleep disturbance between APOE ε4 carriers and non-carriers in lateral ventricular regions.

Among APOE ε4 + participants, sleep disturbance was associated with significantly elevated Aβ SUVR across multiple brain regions after controlling for the same confounding variables (Figure 1). APOE ε4 carriers with sleep disturbance exhibited significantly higher Aβ deposition in multiple corpus callosum regions, including the anterior (β = 0.078, SE = 0.031, p = 0.011), central (β = 0.077, SE = 0.026, p = 0.004), mid-posterior (β = 0.119, SE = 0.028, p = 0.00003), and posterior (β = 0.072, SE = 0.028, p = 0.012) segments. Elevated SUVR values were also observed in cortical regions including the left banks of the superior temporal sulcus (β = 0.080, SE = 0.036, p = 0.025), left cuneus (β = 0.058, SE = 0.021, p = 0.007), left entorhinal cortex (β = 0.029, SE = 0.014, p = 0.037), left lingual gyrus (β = 0.057, SE = 0.021, p = 0.006), left pericalcarine cortex (β = 0.088, SE = 0.028, p = 0.002), left superior parietal lobule (p = 0.05), and right entorhinal cortex (p = 0.035). The left precuneus showed a trend toward significance (β = 0.068, SE = 0.035, p = 0.052). Notably, increased SUVR was also found in the bilateral hippocampus (left: β = 0.039, SE = 0.015, p = 0.013 and right: β = 0.025, SE = 0.015, p = 0.096).

Furthermore, sleep disturbance was significantly associated with increased SUVR in subcortical structures, including the bilateral caudate nuclei (left: β = 0.064, SE = 0.022, p = 0.004, q = 0.046 and right: β = 0.060, SE = 0.024, p = 0.011, q = 0.297), bilateral cerebral white matter (left: β = 0.059, SE = 0.025, p = 0.016 and right: β = 0.042, SE = 0.053, p = 0.026, q = 0.040), bilateral choroid plexus (left: β = 0.094, SE = 0.025, p = 0.00019, q = 0.005 and right: β = 0.088, SE = 0.024, p = 0.00025, q = 0.005), bilateral inferior lateral ventricles (left: β = 0.067, SE = 0.026, p = 0.009 and right: β = 0.048, SE = 0.023, p = 0.036), bilateral lateral ventricles (left: β = 0.096, SE = 0.026, p = 0.0003, q = 0.005 and right: β = 0.098, SE = 0.026, p = 0.0003, q = 0.005), and white matter hypointensities (β = 0.103, SE = 0.028, p = 0.00026, q = 0.005).

Here we present the 20 brain regions demonstrating the most pronounced differences in regional amyloid-β burden between APOE4 carriers experiencing sleep disturbances and those without sleep complaints, highlighting the selective vulnerability of specific neural networks in this genetically at-risk population (Figure.2). This widespread pattern of Aβ accumulation in APOE ε4 carriers encompasses key regions of the DMN (cuneus, precuneus, entorhinal cortex and hippocampus), interhemispheric connections (corpus callosum), and cerebrospinal fluid circulation pathways (lateral ventricles, choroid plexus), indicating that sleep disturbance is associated with Aβ deposition across multiple anatomically and functionally distinct brain systems in APOE ε4 carriers.

We investigated whether subjective sleep disturbance severity in APOE ε4 carriers demonstrated linear associations with regional Aβ burden (Table 4; Supplementary Tables S6). All analyses were adjusted for age, sex, education, and marital status.

This widespread pattern of Aβ accumulation in APOE ε4 carriers encompasses key regions of the DMN (precuneus, cuneus, entorhinal cortex, hippocampus), interhemispheric connections (corpus callosum), and cerebrospinal fluid circulation pathways (lateral ventricles, choroid plexus), indicating that poor sleep quality is linked to Aβ deposition across multiple brain regions in APOE ε4 carriers (Figure 3).

The distribution pattern of these regions suggests that poor sleep quality may preferentially affect DMN-related structures and their supporting systems. Specifically, DMN core impairment is evidenced by significant Aβ deposition in the left precuneus (β = 0.026, SE = 0.011, p = 0.015), a critical posterior DMN hub essential for self-referential processing and episodic memory retrieval. The left posterior cingulate cortex, another core DMN node, also demonstrated significant Aβ accumulation (β = 0.023, SE = 0.01, p = 0.028), suggesting potential disruption of the network’s metabolic and connectivity functions. Memory-related DMN components showed robust associations, with bilateral entorhinal cortex displaying strong significance (left: β = 0.014, SE = 0.004, p = 0.001, q = 0.009; right: β = 0.017, SE = 0.005, p = 5.40E-04, q = 0.009) and left hippocampus showing significant Aβ deposition (β = 0.014, SE = 0.005, p = 0.005, q = 0.037). Visual cortex DMN components, including the cuneus (β = 0.017, SE = 0.007, p = 0.01) and pericalcarine cortex (β = 0.026, SE = 0.009, p = 0.003, q = 0.024), also exhibited significant associations, indicating potential disruption of visual–spatial processing within the DMN framework. DMN connectivity disruption may result from extensive corpus callosum Aβ deposition across multiple segments (anterior: β = 0.026, SE = 0.01, p = 0.008; central: β = 0.025, SE = 0.008, p = 0.003; mid-posterior: β = 0.038, SE = 0.009, p = 1.49E-05), which could compromise interhemispheric connections critical for DMN synchronization and function. Additionally, the observed Aβ deposition in the bilateral lateral ventricles (left: β = 0.027, SE = 0.008, p = 0.001, q = 0.009; right: β = 0.027, SE = 0.008, p = 0.001, q = 0.009) and choroid plexus (left: β = 0.028, SE = 0.008, p = 0.0004, q = 0.008; right: β = 0.027, SE = 0.007, p = 0.0002, q = 0.005) suggests potential clearance system impairment, which may affect glymphatic clearance mechanisms that are crucial for Aβ removal during sleep.