Long‐term forgetting, sleep, and tau in autosomal‐dominant Alzheimer's disease

Abstract

INTRODUCTION

Accelerated long‐term forgetting (ALF) may represent one of the earliest cognitive changes in Alzheimer's disease (AD). However, whether ALF emerges in the preclinical stage of autosomal‐dominant AD (ADAD) is unclear, and its underlying mechanisms are poorly understood.

METHODS

We assessed ALF via long‐term memory testing, sleep physiology via overnight electroencephalography, and AD neuropathology burden via positron emission tomography in 28 non‐demented PSEN1 E280A mutation carriers and 24 healthy non‐carriers from a Colombian ADAD kindred.

RESULTS

Non‐demented carriers exhibited ALF despite having intact short‐term memory. They also demonstrated isolated reduction in parietal sleep spindle (SS) power, whereas other aspects of sleep remained intact. Reduced parietal SS power was associated with both ALF and greater tau burden in the precuneus.

DISCUSSION

ALF is among the earliest cognitive changes in preclinical ADAD, and the specific disruption of parietal SS power may be a key neurophysiologic mechanism linking early tau accumulation to ALF.

Highlights

• Accelerated long‐term forgetting (ALF) arises in preclinical Alzheimer's disease (AD).

• Preclinical AD shows reduced parietal sleep spindle (SS) power.

• Reduced parietal SS power is linked to both ALF and precuneus tau burden.

1. BACKGROUND

Alzheimer's disease (AD) is characterized by neuropathologic changes that begin years before cognitive symptoms emerge, ¹ , ² highlighting a critical window before widespread neurodegeneration when therapeutic interventions may be most effective. Therefore, accurately detecting early cognitive changes and identifying mechanisms that drive the transition from presymptomatic to symptomatic AD are crucial for advancing understanding of the disease and guiding the development of timely therapies. ³

One emerging marker of early cognitive impairment in AD is accelerated long‐term forgetting (ALF), in which individuals perform normally on standard short‐interval memory tests over minutes to hours but exhibit disproportionate memory loss over days to weeks. ⁴ ALF has been associated with apolipoprotein E ε4 carrier status, ⁵ , ⁶ cerebrospinal fluid markers of amyloid beta (Aβ) and tau, ⁶ , ⁷ and increased risk of cognitive decline. ⁸ , ⁹ These findings suggest that ALF could serve as a useful tool for early AD diagnosis and tracking of disease progression. However, whether ALF emerges in preclinical AD remains unclear, and the pathophysiologic mechanisms underlying its emergence in AD are also unknown.

One possible mechanism for ALF in the early stages of AD may involve derangements of sleep. Sleep plays a central role in supporting cognition, ¹⁰ and like ALF, sleep disturbances may be one of the earliest clinical manifestations of AD. ¹¹ , ¹² , ¹³ Specific disruptions in sleep microarchitecture, such as sleep spindles (SS) and slow wave activity (SWA), may be integral to AD pathophysiology, ¹⁴ as SS and SWA abnormalities demonstrate bidirectional relationships with brain Aβ and tau accumulation, ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² impair memory consolidation, ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ and increase dementia risk. ²⁹ , ³⁰ , ³¹ , ³²

Here, we studied ALF and sleep physiology in non‐demented carriers of the PSEN1 E280A mutation for autosomal‐dominant AD (ADAD) in a well‐characterized Colombian kindred, in whom mild cognitive impairment (MCI) begins on average at age ≈ 44 and dementia by age 49. ³³ , ³⁴ , ³⁵ ADAD offers a valuable model for examining early biomarker changes in AD, as ADAD mutation carriers are almost certain to develop AD dementia. We assessed whether non‐demented mutation carriers exhibited ALF by testing verbal memory over both 20‐minute and 7‐day periods. We also evaluated relationships between ALF and AD‐related sleep microarchitectural changes, focusing on parietal SS and prefrontal SWA power, given prior evidence for their roles in the pathophysiology of sporadic AD. ¹⁴ Finally, we examined how these changes were associated with Aβ and tau deposition. We hypothesized that ALF would be evident in preclinical ADAD and reflect underlying disruptions in sleep physiology as well as Aβ and tau accumulation.

2. METHODS

2.1. Participant demographics and clinical information

Individuals from the PSEN1 E280A Colombian kindred cohort participated in this cross‐sectional study between 2020 and 2022 as part of the COLombia‐BOSton (COLBOS) biomarker study of ADAD. ³³ , ³⁴ We assessed the functional/cognitive status of each participant using the Functional Assessment Staging Tool (FAST), ³⁶ on which scores of 1 and 2 indicate normal cognition, 3 indicates MCI, and ≥ 4 indicates dementia. To be considered eligible for this study, participants must be non‐demented (i.e., FAST score 1–3). We also excluded individuals with a history of stroke, epilepsy, traumatic brain injury, kidney failure, human immunodeficiency virus, chronic psychiatric disorder, or substance abuse. All procedures were conducted in Spanish. Investigators were blind to mutation carrier status during data collection. Approval for this study was obtained from the Mass General Brigham Institutional Review Board (Boston, MA) and from the University of Antioquia Ethics Committee (Medellín, Colombia). All participants provided written informed consent prior to study engagement.

2.2. Cognitive testing

Participants underwent verbal learning and memory assessment via the 12‐Word List task from the NEUROPSI Atencion y Memoria neuropsychological battery. ³⁷ The learning component of the NEUROPSI Word List task included three learning trials of 12 words. A learning score was recorded as the average of the total words learned across the three learning trials. Memory of the word list was then assessed via free recall after delays of 20 minutes and 7 days. Indices of recall performance after 20‐minute and 7‐day delays were calculated as the difference between the number of words recalled correctly and the number of words originally learned (recalled – learned). Therefore, a recall index of zero would mean that a participant recalled the same number of words as they learned, and a positive versus negative recall index would mean that a participant recalled more versus fewer words than they learned, respectively. Our serial assessment of memory at 20 minutes and 7 days closely follows previously validated protocols examining ALF. ⁵ , ⁶ , ⁷ , ⁸ , ⁹ Participants also underwent general cognitive screening via Mini‐Mental State Examination.

RESEARCH IN CONTEXT

1. Systematic review: The authors reviewed the literature using PubMed, meeting abstracts, and conference presentations. Prior research suggests that accelerated long‐term forgetting (ALF) may be an early marker of cognitive impairment in Alzheimer's disease (AD). However, whether ALF arises during the preclinical stage of autosomal‐dominant AD (ADAD) remains unclear, and no studies have investigated the mechanisms underlying ALF.

2. Interpretation: In non‐demented ADAD mutation carriers, ALF emerges as one of the earliest cognitive changes and is specifically linked to diminished parietal sleep spindle (SS) power, despite otherwise intact sleep. Reduced SS power also reflects greater tau deposition in the precuneus.

3. Future directions: This work provides a framework for further investigation of ALF and its sleep‐related mechanisms. For example, future studies should clarify the time course of ALF's emergence in AD and evaluate whether SS power could serve as a novel neurophysiologic biomarker and/or therapeutic target.

2.3. Polysomnography

To record sleep, overnight polysomnography was conducted on the same night after assessment of word list learning and 20‐minute delayed recall. All polysomnograms were performed in a sleep laboratory in Medellín, Colombia, and included electroencephalography (EEG) with 19 scalp electrodes (placed according to the International 10–20 System and sampled at 250 Hz), as well as electromyography (EMG; chin, right deltoid, left deltoid), electrooculogram, and electrocardiogram electrodes, all recorded on a 32‐channel Cadwell Easy lll device. Participants were instructed to go to bed at their regular times. Participants underwent polysomnography in two batches, with the first batch (n = 30) recorded in 2020 and the second batch (n = 22) in 2022. Notably, there were no significant demographic or clinical differences between batches (Table S1 in supporting information).

2.4. EEG analysis

Polysomnographic EEG recordings were preprocessed in MATLAB R2022a using the EEGLAB ³⁸ and FieldTrip ³⁹ toolboxes. To identify stages 2 and 3 of non–rapid eye movement (NREM) sleep, we performed automated sleep staging using a published and publicly available sleep staging algorithm that performs at the level of human experts and has been validated in both cognitively normal individuals and in those with AD. ⁴⁰ We then calculated SWA power, SS power, SS density, SS peak frequency, and SS–SWA coupling strength from EEG data during NREM2 and NREM3, using custom scripts and toolboxes in MATLAB, described in further detail below.

We measured SWA (0.6–1 Hz) power in the bilateral prefrontal (Fp1/Fp2), frontal (F3/F4), temporal (T3/T5/T4/T6), central (C3/C4), parietal (P3/P4), and occipital (O1/O2) regions using multitaper spectral analysis. Spectrograms were computed using the Chronux toolbox ⁴¹ , ⁴² across non‐overlapping 6‐second windows, with three tapers and a time‐bandwidth product of 2, resulting in a spectral resolution of 0.33 Hz. EEG artifacts were automatically identified based on thresholds for excessively high amplitudes (≥ 400 uV), EMG artifact (standard deviation [SD] of the bandpass‐filtered signal between 30 and 50 Hz exceeding 5 uV), and flat channels (≤ 0.01 uV); windows containing excessive artifacts (averaging 6.6% of total recording) were discarded prior to spectrogram calculations. For this study, we analyzed relative SWA power, calculated as the ratio of spectral power in the 0.6 to 1 Hz range to the total delta power (0.6–4 Hz) of each participant. This measure of relative SWA power has been validated in prior studies of sleep in AD and shown to be specifically associated with AD neuropathology. ¹⁶ , ¹⁷

We measured the power, density, and peak frequency of SS (11–16 Hz) events and the strength of SS–SWA coupling in the bilateral prefrontal (Fp1/Fp2), frontal (F3/F4), temporal (T3/T5/T4/T6), central (C3/C4), parietal (P3/P4), and occipital (O1/O2) regions using a previously published algorithm that detected SS events based on duration and amplitude thresholds (SS signal > 1.5 SDs from baseline EEG signal for at least 300 ms and > 2.5 SDs at its amplitude peak). ⁴³ For each SS event, we calculated relative SS power by using the continuous wavelet transform to obtain the ratio of spectral power in the 11 to 16 Hz range to the total power spectrum (0.1–30 Hz) of each participant. This ratiometric approach controls for individual differences in absolute EEG power, allowing for more robust comparisons across participants. ⁴³ We computed SS density as the number of SS events per minute. We also extracted the peak frequency of each SS event. To determine SS–SWA coupling strength, we used the Hilbert transform to compute the instantaneous phase angles of the filtered SWA signal. We then extracted SWA phases corresponding to SS amplitude peaks and used circular statistics to compute the resultant vector length (i.e., phase‐locking strength). ⁴⁴ Topoplots of SS power were generated using FieldTrip.

Despite using relative measures of SWA and SS power, we observed significant differences in SWA (P = 3 × 10⁻¹⁰) and SS (P = 5 × 10⁻⁸) power between the two batches of participants who underwent polysomnography in 2020 versus 2022. The reason for these differences was unclear, as EEGs from both batches appeared similar upon visual inspection, and there were no significant demographic or clinical differences between the batches (Table S1). To improve data interpretability, we converted relative SWA and SS power values within each batch to batch‐specific z scores prior to carrying out additional statistical analyses on these measures. Notably, after performing batch‐specific z scoring, we no longer observed batch‐related differences in SWA (P = 1) or SS (P = 0.39) power. To further mitigate batch‐related confounding, we incorporated batch as a covariate in our multivariate statistical models (Section 2.6) and conducted batch‐stratified analyses of all primary outcomes to confirm that the directions of effects and associations within individual batches were consistent with those observed in the main combined analyses (Supplementary Results in supporting information).

2.5. Positron emission tomography

A subset of participants underwent ¹¹C‐Pittsburgh compound B (PiB) positron emission tomography (PET) imaging to assess brain Aβ burden (n = 31) and ¹⁸F‐flortaucipir (FTP) PET imaging to assess brain tau burden (n = 32). All PET imaging was acquired at the Massachusetts General Hospital within 6 months of polysomnography and preprocessed and analyzed using previously described protocols. ⁴⁵ , ⁴⁶ To briefly summarize, global cortical Aβ deposition was measured via the distribution volume ratio of PiB uptake across a large neocortical target region that included the frontal, lateral temporal, and retrosplenial cortices. Regional tau deposition in the entorhinal cortex and precuneus was measured via the standardized uptake volume ratio of FTP. Tracer uptake in the cerebellar gray matter was used as the reference for both PiB and FTP. ⁴⁵ , ⁴⁶

2.6. Statistical analysis

Table S2 in supporting information details the statistical tests used in this study. All mixedeffects and generalized linear models considered age, sex, education, depression (measured by Geriatric Depression Scale), functional status (measured by FAST), sleep apnea diagnosis, and polysomnography batch as potential covariates. For each model, covariates were selected for inclusion using a multi‐model inference algorithm that maximized goodness of fit while avoiding overfitting. ⁴⁷ Incorporation of selected covariates into each multivariate model was accomplished using inverse probability weighting. Statistical tests were two tailed with α = 0.05, and the Holm method was used to adjust for multiple comparisons where indicated. Statistical calculations were performed using R 4.5.1.

3. RESULTS

3.1. Study population characteristics

Characteristics of the study population are provided in Table 1. The study population consisted of 52 individuals (28 PSEN1 E280A mutation carriers and 24 healthy, non‐carrier family members). Participants had a mean (standard deviation) age of 36.8 (6) years, received a median (interquartile range) of 11 (8.5–14) years of education, and included 25 (51.9%) females. Most carriers (n = 24) were in the preclinical stage of AD, as they did not demonstrate objective evidence of cognitive or functional impairment (FAST 1 or 2), while 4 carriers were in the early clinical stage of AD with MCI (FAST 3). The proportion of individuals with sleep apnea was similar between carriers and non‐carriers (46.4% versus 50%, P = 1). Carriers and non‐carriers exhibited similar sleep macroarchitectural characteristics, with the exception that carriers spent a higher percentage of time in NREM2 sleep than non‐carriers (66.4% versus 60.1%, P = 0.005).

TABLE 1.

Participant demographic and clinical characteristics.

All participants (n = 52)
Characteristic	Carriers (n = 28)	Non‐carriers (n = 24)	p value
Age, mean ± SD, years	36.2 ± 5.6	37.4 ± 6.5	0.48
Sex, # (%), female	15 (53.6)	12 (50)	1
Education, median (IQR), years	11 (5–13)	13 (11–15)	0.01
GDS, median (IQR)	1 (0–2)	0 (0–3)	0.9
MMSE, median (IQR)	28 (28–29)	29 (28–29)	0.07
FAST, median (IQR)	1 (1–2)	1 (1–1)	0.11
Sleep apnea, # (%), diagnosed	13 (46.4)	12 (50)	1
Sleep macroarchitecture
TST, median (IQR), minutes	391 (342–428)	375 (326–396)	0.11
Sleep efficiency, median (IQR), %	82.2 (66.3–89.4)	76.1 (71.2–86.2)	0.24
Arousal index, median (IQR), # per hour	5.1 (4.4–7.1)	6.3 (4.8–7.1)	0.2
% time in NREM1, median (IQR)	9.6 (6.8–13.8)	12 (9.4–14.3)	0.19
% time in NREM2, median (IQR)	66.4 (61.3–71.1)	60.1 (54–63.5)	0.005
% time in NREM3, median (IQR)	1.8 (0.1–6.1)	5.3 (1.3–9.7)	0.16
% time in REM, median (IQR)	19.4 (16.6–23.4)	20.1 (14.5–25.1)	0.73
Participants with PiB PET imaging (n = 31)
PiB PET imaging, # (%)	15 (53.6)	16 (66.7)	0.4
Participants with FTP PET imaging (n = 32)
FTP PET imaging, # (%)	16 (57.1)	16 (66.7)	0.57

FAST, functional assessment staging tool; FTP, 18F‐flortaucipir; GDS, Geriatric Depression Scale; IQR, interquartile range; MMSE, Mini‐Mental State Examination; NREM, non–rapid eye movement sleep; PET, positron emission tomography; PiB, 11C‐Pittsburgh compound B; REM, rapid eye movement sleep; SD, standard deviation; TST, total sleep time.

3.2. Cognitively unimpaired and mildly impaired mutation carriers demonstrate ALF

We conducted a time‐dependent analysis of word list recall performance across 20‐minute and 7‐day delay intervals in carriers versus non‐carriers and observed a significant time x carrier status interaction (β: −3, 95% confidence interval [CI: −4.8, −1.2], P = 0.001, Figure 1A). Post hoc analyses at each time point revealed similar performance between carriers and non‐carriers on learning (β: −0.51, 95% CI: [−1.2, 0.2], P = 0.16, Figure 1B) and 20‐minute recall (β: −1, 95% CI [−2.3, 0.28], P = 0.19, Figure 1C), but significantly worse performance by carriers on 7‐day recall (β: –1.9, 95% CI: [−3.2, −0.56], P = 0.0008, Figure 1D). These results indicate the presence of ALF in carriers. Findings were similar even after excluding carriers with MCI from the analysis (Table S3 in supporting information).

FIGURE 1.

Word list learning and recall across time in carriers versus non‐carriers. A, Time‐dependent analysis of performance on word list recall across two delay intervals. Datapoints and lines represent the trajectories of estimated marginal means derived from mixed effects modeling, and shaded areas represent 95% confidence intervals. B–D, Post hoc individual time point analysis of learning, 20‐minute delayed recall, and 7‐day delayed recall in carriers versus non‐carriers. Multiple comparison adjustments were applied to p values derived from these three post hoc tests: **p < 0.01. ***p < 0.001. ns, non‐significant.

3.3. Carriers exhibit decreased parietal SS power

Having demonstrated the presence of ALF in preclinical AD, we next evaluated whether this finding was associated with changes in sleep physiology, focusing first on group differences in sleep microarchitectural features. We began by visualizing the topography of SS power across the brain surface in carriers versus non‐carriers (Figure 2A). Upon visual inspection, there was strikingly less SS power over the parietal region in carriers relative to non‐carriers. Indeed, subsequent statistical analysis revealed reduced parietal (P3/P4) SS power in carriers (β: −0.48, 95% CI: [−0.88, −0.08], P = 0.02, Figure 2B). Because participants underwent polysomnography in two separate batches, we also performed batch‐stratified analyses to confirm that decreases in parietal SS power among carriers were independently observed within each batch (Supplementary Results and Figure S1 in supporting information). This finding remained significant even after excluding carriers with MCI (β: −0.46, 95% CI: [−0.89, −0.03], P = 0.04) and appeared specific to the parietal region, as SS power in other brain regions did not differ between carriers and non‐carriers (Table S4 in supporting information). We also found no significant differences in SWA power (Table S5 in supporting information), SS density (Table S6 in supporting information), SS frequency (Table S7 in supporting information), or SS–SWA coupling strength (Table S8 in supporting information) between carriers and non‐carriers. Together, these findings indicate that reduced parietal SS power reflects a specific regional disruption in sleep physiology that emerges in preclinical AD, absent other obvious changes in sleep microarchitecture.

FIGURE 2.

Comparison of SS power between carriers and non‐carriers. A, Group‐averaged topographic representation of SS power in carriers and non‐carriers. Labeled electrodes P3 and P4 indicate the locations where parietal SS power was derived for subsequent statistical analyses. B, Quantification of parietal SS power in carriers versus non‐carriers. Participants underwent polysomnography in two separate batches. To reduce batch‐related confounding, SS power values were standardized within each batch, yielding batch‐specific z scores. *P < 0.05. ns, non‐significant; SS, sleep spindle.

3.4. Parietal SS power is associated with both short‐ and long‐term memory function in carriers

As sleep plays a key role in memory consolidation, we next investigated whether differences in sleep microarchitecture between carriers and non‐carriers were associated with short‐ and long‐term memory function. Examining the relationship between parietal SS power and word list recall over time, we detected a significant time x SS power interaction in carriers (β: 2.5, 95% CI: [0.61, 4.3], P = 0.009), but not non‐carriers (β: 0.31, 95% CI: [−1.5, 2.2], P = 0.74). This carrier‐specific interaction corresponded to the associations of parietal SS power with both 20‐minute (β: 1, 95% CI: [0.29, 1.7], P = 0.02) and 7‐day (β: 1.4, 95% CI: [0.27, 2.6], P = 0.03) recall performance in carriers (Figure 3). Batch‐stratified analyses confirmed that these positive associations for 20‐minute and 7‐day recall were independently observed within each polysomnography batch (Supplementary Results and Figure S2 in supporting information). Notably, parietal SS power showed the strongest association with performance on 7‐day recall (Cohen f ²  = 0.27), relative to 20‐minute recall (f ²  = 0.13) and learning (f ²  = 0.005). In other words, carriers with less parietal SS power were more likely to demonstrate ALF. Importantly, the associations of parietal SS power with 20‐minute and 7‐day recall performance in carriers remained significant even after excluding carriers with MCI (Table S9 in supporting information). Given the robust relationship between parietal SS power and verbal memory function in preclinical AD, we also investigated whether other parietal SS properties, such as SS density (Table S10 in supporting information), SS frequency (Table S11 in supporting information), and SS–SWA coupling (Table S12 in supporting information), were associated with word list task outcomes, but found no significant relationships.

FIGURE 3.

Relationship between parietal SS power and word list task performance. Regressions of parietal SS power with word list (A) learning, (B) 20‐minute delayed recall, and (C) 7‐day delayed recall in carriers versus non‐carriers. Solid lines and shaded areas represent the slopes ± standard errors of significant associations. Dotted lines represent non‐significant (ns) regressions. Cohen f2 values are shown for carriers and indicate the strength of association between parietal SS power and performance on each component of the word list task. Participants underwent polysomnography in two separate batches. To reduce batch‐related confounding, SS power values were standardized within each batch, yielding batch‐specific z scores. Multiple comparison adjustments were applied to P values from these three regression analyses, which were conducted after detection of a carrier‐specific time x SS power interaction on word list task performance in the main time‐dependent analysis. ns, non‐significant; SS, sleep spindle.

3.5. Prefrontal SWA power is associated with short‐term memory function in carriers

Although we did not detect group differences in SWA power between carriers and non‐carriers, we hypothesized that prefrontal SWA power was associated with memory function in carriers, based on studies in sporadic AD that highlight the role of SWA in memory consolidation. ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ Indeed, we observed a significant time x prefrontal SWA power interaction in carriers on word list recall over time (β: 1.2, 95% CI: [0.2, 2.3], P = 0.02, Figure 4), which corresponded to a carrier‐specific association of prefrontal SWA power with 20‐minute recall (β: 0.95, 95% CI: [0.3, 1.6], P = 0.02), and a trend toward association with 7‐day recall (β: 1, 95% CI: [−0.06, 2.1], P = 0.06). Batch‐stratified analyses showed consistent results across polysomnography batches (Supplementary Results and Figure S3 in supporting information). However, this association was no longer significant after excluding carriers with MCI (Table S13 in supporting information).

FIGURE 4.

Relationship between prefrontal SWA power and word list task performance. Regressions of prefrontal SWA power with word list (A) learning, (B) 20‐minute delayed recall, and (C) 7‐day delayed recall in carriers versus non‐carriers. In (B), the solid line and shaded area represent the slope ± standard error of a significant association. In all panels, dotted lines represent non‐significant regressions. Cohen f2 values are shown for carriers and indicate the strength of association between prefrontal SWA power and performance on each component of the word list task. Participants underwent polysomnography in two separate batches. To reduce batch‐related confounding, SWA power values were standardized within each batch, yielding batch‐specific z scores. Multiple comparison adjustments were applied to P values from these three regression analyses, which were conducted after detection of a carrier‐specific time x SWA power interaction on word list task performance in the main time‐dependent analysis. SWA, slow wave activity.

3.6. Associations of Aβ and tau burden with disruptions in sleep physiology

Finally, we examined the relationship between sleep physiology and AD neuropathology, using Aβ and tau PET imaging. We focused on global cortical Aβ and regional tau deposition in the precuneus and entorhinal cortex (EC), based on previous research emphasizing the importance of these regions in AD pathophysiology. ⁴⁵ , ⁴⁶ Indeed, carriers exhibited increased deposition of global Aβ (β: 0.34, 95% CI: [0.17, 0.51], P = 0.0004, Figure S4A in supporting information) and precuneus tau (β: 0.24, 95% CI: [0.02, 0.47], P = 0.03, Figure S4B) compared to non‐carriers. Carriers also trended toward increased tau deposition in the EC (β: 0.28, 95% CI: [−0.01, 0.56], P = 0.06, Figure S4C). These results are consistent with our prior work showing that carriers exhibit the fastest rate of tau accumulation in the parietal region. ⁴⁶

We found that parietal SS power was inversely associated with tau burden in both the precuneus (β: −0.36, 95% CI: [−0.6, −0.1], P = 0.008, Figure 5A) and EC (β: −0.2, 95% CI: [−0.4, −0.01], P = 0.04, Figure 5B). Batch‐stratified analyses showed results consistent with the main findings (Supplementary Results and Figure S5 in supporting information). Consistent with these findings, elevated levels of tau in the precuneus and EC were associated with worse performance on both 20‐minute and 7‐day recall (Table S14 in supporting information). We found no significant relationship between parietal SS power and Aβ deposition (β: −0.24, 95% CI: [−0.64, 0.17], P = 0.24, Figure 5C).

FIGURE 5.

Relationship between parietal SS power and burden of AD neuropathology. Regressions of parietal SS power with (A) precuneus tau, (B) EC tau, and (C) global cortical Aβ deposition. Solid lines and shaded areas represent the slopes ± standard errors of significant associations. Participants underwent polysomnography in two separate batches. To reduce batch‐related confounding, SS power values were standardized within each batch, yielding batch‐specific z scores. Aβ, amyloid beta; AD, Alzheimer's disease; DVR, distribution volume ratio; EC, entorhinal cortex; ns, non‐significant; SS, sleep spindle; SUVR, standardized uptake value ratio.

In contrast, prefrontal SWA power was inversely associated with global Aβ (β: −1.1, 95% CI: [−2, −0.17], P = 0.02, Figure 6A) and EC tau burden (β: −0.42, 95% CI: [−0.76, −0.09], P = 0.01, Figure 6B), but showed no significant association with precuneus tau deposition (β: −0.45, 95% CI: [−1.1, 0.16], P = 0.14, Figure 6C). Batch‐stratified analyses showed results consistent with the main findings (Supplementary Results and Figure S6 in supporting information). Together, these findings demonstrate that regional alterations in sleep microarchitecture have distinct relationships with Aβ and tau pathology, with reduced SS power in the parietal region specifically tied to precuneus tau accumulation.

FIGURE 6.

Relationship between prefrontal SWA power and burden of AD neuropathology. Regressions of prefrontal SWA power with (A) global cortical Aβ, (B) EC tau, and (C) precuneus tau deposition. Solid lines and shaded areas represent the slopes ± standard errors of significant associations. Participants underwent polysomnography in two separate batches. To reduce batch‐related confounding, SWA power values were standardized within each batch, yielding batch‐specific z scores. Aβ, amyloid beta; DVR, distribution volume ratio; EC, entorhinal cortex; ns, non‐significant; SUVR, standardized uptake value ratio; SWA, slow wave activity.

4. DISCUSSION

This study examined the relationships between verbal memory performance over a 7‐day interval, sleep physiology, and AD neuropathology in non‐demented carriers of the PSEN1 E280A mutation for ADAD. We found that these carriers demonstrated ALF despite having intact learning and 20‐minute recall. Furthermore, we identified reduced parietal SS power as a potential neurophysiologic mechanism linking ALF to tau accumulation in preclinical ADAD.

Our detection of ALF in non‐demented carriers using 7‐day recall aligns with another study that observed ALF in a British ADAD cohort, ⁸ suggesting its potential as a sensitive cognitive marker of preclinical AD. However, important questions remain regarding how early ALF emerges in ADAD, and whether shorter testing intervals (e.g., < 7 days) are sufficient for its detection. Future studies assessing recall performance across multiple delay intervals and age ranges in carriers are needed to address these questions and refine ALF's clinical utility.

We also identified unique associations between regional changes in sleep microarchitecture and memory performance in carriers across different time intervals. Our investigation focused on two key aspects of sleep microarchitecture—SS and SWA—which play pivotal roles in memory consolidation and AD progression. Prior studies have tied SS and SWA abnormalities to elevated Aβ and tau burden, ¹⁵ , ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² hippocampus‐dependent memory deficits, ²³ , ²⁴ , ²⁵ , ²⁶ accelerated cognitive decline, ²⁷ , ²⁸ and increased dementia risk. ²⁹ , ³⁰ , ³¹ , ³² However, these studies focused solely on short‐term memory spanning minutes to hours, without addressing whether more subtle cognitive deficits such as ALF would manifest over longer intervals. They also did not investigate the mechanisms contributing to long‐term memory deficits in AD.

Our findings provide new evidence that ties specific sleep‐related processes to long‐term memory in AD. We found that parietal sleep spindle power correlated with both 20‐minute and 7‐day recall, but was more strongly associated with 7‐day recall, suggesting that individuals with lower parietal SS power were more prone to ALF. By contrast, prefrontal SWA power was only associated with 20‐minute recall. These patterns suggest that in the early stages of AD, prefrontal SWA may reflect local physiologic mechanisms that support short‐term memory, whereas parietal SS activity may reflect the integrity of a broader network that supports both short‐ and long‐term memory, thereby linking it to ALF. This hypothesis aligns with prior neuroanatomic and functional connectivity studies showing that: (1) the prefrontal cortex, by interacting with medial temporal lobe structures, is critical for new memory formation and consolidation, ⁴⁸ and (2) the parietal cortex, as a primary integration hub within the default mode network, facilitates both consolidation of new memories and reactivation of older memories. ⁴⁹ , ⁵⁰

Notably, the associations between sleep microarchitectural alterations and memory function were observed only in carriers, pointing to a specific link between these relationships and AD pathophysiology. ²⁵ Indeed, we found that regional changes in SS and SWA were both associated with AD neuropathology, though each reflected distinct components: decreased parietal SS power was a stronger indicator of tau burden, whereas decreased prefrontal SWA power tracked more closely with Aβ burden. Collectively, these results support the hypothesis that AD neuropathology is closely tied to specific changes in sleep physiology, which in turn may disrupt memory consolidation and contribute to deficits in both short‐ and long‐term recall. ¹⁴ In particular, decreased parietal SS power may represent a critical sleep‐dependent mechanism that links worsening tau pathology to subtle cognitive changes in preclinical AD, given that it was the only sleep feature altered in cognitively unimpaired carriers and showed the strongest association with ALF. By contrast, prefrontal SWA power may play a more prominent role in later disease stages, because unlike parietal SS power, prefrontal SWA power did not significantly differ between non‐demented carriers and non‐carriers, and its association with memory function was no longer significant after excluding carriers with MCI.

Ultimately, understanding which brain networks are most vulnerable to AD pathophysiology in the earliest stages of disease may inform the development of targeted therapies that preserve network integrity and promote cognitive resilience against AD. ²⁶ , ⁵¹ In this context, studying ADAD offers a unique opportunity to probe mechanisms underlying the transition from preclinical to clinical AD, given the near certainty of progression to AD dementia among mutation carriers. Notably, the clinical trajectory of ADAD is remarkably similar to that of sporadic AD, ³³ , ³⁴ , ³⁵ suggesting shared mechanisms of disease progression. Consistent with our findings in ADAD, we have recently shown that sporadic AD is also characterized by SS derangements that correspond with cognitive decline. ⁵² At the same time, sporadic AD differs from ADAD in important ways. For example, sporadic AD shows earlier temporal lobe tau accumulation and more prominent age‐related and vascular pathology than ADAD. ¹ , ² How these differences interact with sleep physiology and cognition warrants further investigation. Moreover, our results should be cautiously interpreted in the context of other neurological conditions that commonly coexist with AD, such as epilepsy, Lewy body disease, and TAR DNA‐binding protein 43 pathology. ⁵³ , ⁵⁴ , ⁵⁵

Strengths of this study include a uniquely well‐characterized ADAD cohort and the integration of multimodal assessments (e.g., PET, sleep EEG, cognitive) to address important questions regarding AD neuropathology, sleep physiology, and cognitive performance. Limitations of this study include modest sample size, single‐ rather than multiple‐night polysomnography, absence of assessments targeting other cognitive domains, and lack of information on treatment status for sleep apnea or sleep medication use. Given the limited sample size, mediation analysis was unfortunately not possible, which precluded assessment of potential causal pathways. The cross‐sectional design of this study may have also limited our ability to detect associations of SS density and SS–SWA coupling strength with memory performance, as longitudinal studies in sporadic AD have shown significant relationships between these sleep microarchitectural features and cognitive function over longer periods of time. ²⁷ , ²⁸ Future longitudinal studies with larger cohorts and repeated sleep measurements are needed to further examine these relationships.

In summary, our findings demonstrate that ALF is detectable in the preclinical stages of ADAD and is closely associated with both regional changes in sleep physiology and AD neuropathology. Parietal SS power most notably emerges as a promising biomarker for early detection of ADAD‐related cognitive and pathologic changes, highlighting a possible sleep‐related mechanism linking early cognitive symptoms to underlying brain pathology. Future studies are needed to assess the generalizability of these results and to investigate potential causal relationships among sleep disturbances, AD neuropathology, and cognitive decline.

CONFLICT OF INTEREST STATEMENT

Dr. Quiroz has served as a consultant for Biogen. Dr. Lam has served as a paid consultant for Neurona Therapeutics, Acadia Pharmaceuticals, and UCB. The institution of Dr. Lam has received research funding from Neurona Therapeutics and Sage Therapeutics. The other co‐authors report no competing financial interests or conflicts. At the time of this publication, Dr. Pardilla‐Delgado is employed by Vanda Pharmaceuticals, but the current work is unrelated to his position there. Author disclosures are available in the supporting information.

CONSENT STATEMENT

All research participants provided written informed consent prior to study engagement.

Supporting information

Supporting Information

Supporting Information

You JC, Bender AC, Berezuk C, et al. Long‐term forgetting, sleep, and tau in autosomal‐dominant Alzheimer's disease. Alzheimer's Dement. 2026;22:e71235. 10.1002/alz.71235