Thermoneutral temperature exposure enhances slow-wave sleep with a correlated improvement in amyloid pathology in a triple-transgenic mouse model of Alzheimer’s disease

Abstract

Accumulation of amyloid‐β (Aβ) plays an important role in Alzheimer’s disease (AD) pathology. There is growing evidence that disordered sleep may accelerate AD pathology by impeding the physiological clearance of Aβ from the brain that occurs in normal sleep. Therapeutic strategies for improving sleep quality may therefore help slow disease progression. It is well documented that the composition and dynamics of sleep are sensitive to ambient temperature. We therefore compared Aβ pathology and sleep metrics derived from polysomnography in 12-month-old female 3xTg-AD mice (n = 8) exposed to thermoneutral temperatures during the light period over 4 weeks to those of age- and sex-matched controls (n = 8) that remained at normal housing temperature (22°C) during the same period. The treated group experienced greater proportions of slow wave sleep (SWS)—i.e. epochs of elevated 0.5–2 Hz EEG slow wave activity during non-rapid eye movement (NREM) sleep—compared to controls. Assays performed on mouse brain tissue harvested at the end of the experiment showed that exposure to thermoneutral temperatures significantly reduced levels of DEA-soluble (but not RIPA- or formic acid-soluble) Aβ40 and Aβ42 in the hippocampus, though not in the cortex. With both groups pooled together and without regard to treatment condition, NREM sleep continuity and any measure of SWS within NREM at the end of the treatment period were inversely correlated with DEA-soluble Aβ40 and Aβ42 levels, again in the hippocampus but not in the cortex. These findings suggest that experimental manipulation of SWS could offer useful clues into the mechanisms and treatment of AD.

Graphical Abstract

Graphical Abstract.

Statement of Significance.

Although the connection between disordered sleep and amyloid pathology in Alzheimer’s Disease (AD) has been actively investigated, very little is known about the effect of sleep enhancement. We demonstrate in a widely used mouse model of AD the feasibility of slow wave sleep (SWS) enhancement through thermoneutral temperature exposure, and further that the accumulation of amyloid-β (Aβ) in the brain varies inversely with the amount of SWS and the continuity of non-rapid eye movement sleep. The reduction in Aβ deposition with sleep enhancement appears selective to the hippocampus. This is of particular significance given the role of the hippocampus in memory and learning, which are known to deteriorate in AD, and reinforces the idea that sleep intervention could alleviate and perhaps even improve outcomes in those affected.

Research suggests that disordered sleep may be an early marker of AD and is associated with increased levels of Aβ [1]. This is consistent with findings over the past decade that sleep plays an important role in the active clearance of Aβ from brain tissue [2, 3]. The connection between the accumulation of Aβ in the brain and AD symptoms [4] suggests that maintaining low levels of Aβ in the brain may reduce plaque formation and thus slow down disease progression. Ongoing research into immunization therapy using monoclonal antibodies such as Lecanemab to target Aβ has shown some success in the reduction of amyloid markers in patients with early-onset AD [5]. The cost and potential for serious side effects associated with monoclonal antibody therapies warrant further investigation of alternative approaches such as non-pharmacological sleep enhancement for reducing Aβ accumulation.

To test the hypothesis that promoting sleep may reduce Aβ accumulation in AD, we conducted a study to enhance sleep through ambient temperature manipulation in an AD mouse model (3xTg-AD) and assessed the effects on Aβ pathology. Thermoregulation plays a critical role in sleep induction and ambient temperature is a key determinant of the composition of sleep [6, 7]. Mice and other laboratory animals are typically housed at temperatures that are below their thermoneutral zone (TNZ): i.e. a range of temperatures at which their basal metabolic rate is at a minimum [8]. Raising the ambient temperature to the TNZ is found to promote slow wave sleep (SWS) [9], the deepest phase of non-rapid eye movement (NREM) sleep, marked by large amplitude slow oscillations (0.5–2 Hz) in the electroencephalogram (EEG) that increase with sleep debt and are widely regarded as the currency of sleep homeostasis [10, 11]. Within the TNZ (approximately 26–30°C for mice [6]), metabolism slows down and deeper sleep is permitted absent the need for thermogenesis to maintain body temperature. A better understanding of how thermoneutral exposure impacts sleep (and in turn Aβ metabolism) may lead to better strategies for treating patients with AD.

Materials and Methods

Study design

Female 3xTg-AD mice (n = 19) were surgically instrumented for electroencephalography (EEG) and electromyography (EMG) to score sleep. Following a 2-week recovery period, mice were randomly assigned to experimental and control groups, and transferred to individual cages for a week-long baseline recording. Mice assigned to the experimental group (n = 10) were exposed to thermoneutral temperatures (as described below) during the light period every day for the next 4 weeks to promote sleep, while mice assigned to the control group (n = 9) remained at room temperature throughout. At the end of the experimental period, mice were euthanized and brain tissue was harvested for analysis of AD pathology. Conventional sleep metrics were derived from the EEG/EMG recordings for both groups and correlated with pathology using standard statistical methods.

Animal procedures

All animal procedures were carried out with prior approval from the Institutional Animal Care and Use Committee of the University of Kentucky. Experiments were conducted on adult female 3xTg-AD mice (12–15 months of age, Jackson labs, https://www.jax.org/strain/004807). The 3xTg-AD mouse expresses three human mutant genes (APP, Tau, and PS1), displays both Aβ and Tau pathology [12], and is widely used in preclinical AD research. The 3xTg-AD mouse shows AD symptoms such as age-dependent cognitive decline [13] and exhibits accumulation of Aβ, fibrillary tangles, and neuroinflammation in the brain as they age, with a relatively slow progression [13]. Extracellular Aβ deposits appear in the 3xTg-AD mouse by 6 months of age in the frontal cortex and become more extensive by 12 months [12]. We used 12–15-month-old mice with the expectation that the amount of Aβ accumulated and any effect of our treatment would be easier to quantify.

Before starting the study, mice were housed in groups of four in individually ventilated cages with bedding and free access to food and water. The room temperature during habituation and surgical recovery was ~22°C with a relative humidity of (50 ± 10%). Mice were housed on a 14:10-hour light:dark cycle with the lights on from 7 am to 9 pm.

To prepare for implantation surgery, the mice were moved to individual cages and given oral analgesia (Carprofen, 5 mg/kg/day; Medigel CPF, ClearH₂O) on the day before surgery. On the day of the procedure, hair on the scalp was shaved off under 2.5% isoflurane anesthesia and the anesthetized animal was fixed to a stereotactic frame using ear bars in a recumbent position. A midline scalp incision was made from behind the eyeline to the neck. The exposed skull was then swabbed with a povidone-iodine solution followed by 70% isopropyl alcohol, and a prefabricated rectangular headmount (No. 8202, Pinnacle Technology, Inc., Lawrence, KS) was glued to the skull with the center of the chip located above bregma. Four guide holes were tapped in the skull with an injection needle and stainless steel screws (No. 8209 and 8212, Pinnacle Technology) inserted into them to fix the headmount to the skull. The tips of these bone screws, which also serve as electrodes, rest on the dura mater. The screwheads were painted with silver epoxy to improve conductivity. The placement of the screws is designed to pick up hippocampal theta (6–9 Hz) oscillations as well as slow wave activity in the delta band (0.5–4 Hz) during sleep. Two stainless steel leads attached to the rear of the headmount were inserted into the nuchal muscle to sense muscle tone (bipolar EMG). After closing the incision with 2–3 sutures, dental acrylic was applied around the headmount to seal the surgery site and prevent infection. Oral analgesia (Carprofen) was administered daily for the next 3 days.

Implanted animals were housed in individually ventilated cages and given 2 weeks for recovery and for the implants to set. Sutures were then removed under 2.5% isoflurane anesthesia and the animals were moved to custom-built recording cages with a high, open ceiling to allow EEG/EMG acquisition through a tethered preamplifier plugged into the headmount. The tether was connected to a slip-ring commutator for strain relief and unhindered movement. Animals were given free access to food and water, with cages cleaned and bedding replenished at weekly intervals. All conditions in the cages (light:dark cycle, humidity) remained as maintained in the housing facility except for the cage temperature, which was manipulated for the experimental group as described below.

Data acquisition

EEG and EMG signals were monitored using a data acquisition and conditioning system (No. 8206, Pinnacle Technology, Inc., Lawrence, KS) through a wired preamplifier (10× or 100×, No. 8202, Pinnacle Technology, Inc., Lawrence, KS) plugged into the headmount, with a 100 Hz anti-aliasing lowpass filter. The data acquisition and conditioning system sampled the signals at 400 Hz and stored them on a computer for analysis using Pinnacle’s Sirenia acquisition software.

Thermoneutral treatment

To simulate normal diurnal cycles that favor sleep during the light period in nocturnal mice, the cage temperature was programmed to gradually increase, in hourly increments of 1°C, from 22°C (the ambient temperature in the facility) at the onset of the light period (zeitgeber time [ZT] 0) to 30°C by mid-day (ZT8), and then reverse course to fall back to 22°C by the onset of the dark period (ZT14). The temperature was not actively manipulated during the dark period. This cycle was repeated every day to simulate the gradual rise and fall in temperature in our natural environment and to promote sleep during the normal inactive phase, the light period.

Cage temperature control was achieved by housing the animal and recording apparatus within a glass terrarium (No. PT2612, ExoTerra). Two 100W infrared lamps (Ceramic Infrared Heat Emitter; Zoo Med) within the housing served as heat sources. A thermistor (STS-BTA, Vernier, Oregon) suspended in the cage monitored cage temperature (Ta), which was read into a computer as a calibrated voltage through a data acquisition board (NI USB-6211, National Instruments). A custom-written LabVIEW VI switched the heater on and off to maintain cage temperature close to the setpoint temperature profile with a tolerance of 0.5°C.

Sleep scoring

Sleep was scored using a previously developed automated algorithm [14] based on conventional criteria from the archived polygraphic recordings. EMG was used as an indicator of sleep/wake state as animals have significantly lower muscle tone during sleep compared to wakefulness. The EEG was used to distinguish NREM sleep from rapid eye movement (REM) sleep. The EEG frequency bands associated with different stages of sleep are delta (0.5–4 Hz) and theta (6–9 Hz): NREM sleep is characterized by high delta power and REM sleep by high theta power.

The recordings were segmented into 4-second-long epochs for sleep scoring as diagramed in Figure 1. Three features were computed in each epoch: delta/theta band EEG power ratio, high/low-frequency band EEG power ratio (i.e. ratio of 9–80 Hz to 0.5–9 Hz power), and EMG power. In each 24-hour recording, a Gaussian Observation Hidden Markov Model (GO-HMM) algorithm [14] was used to separate all epochs into three clusters based on the features. The cluster with the highest mean EMG power was first labeled as wake. The two remaining clusters were then labeled by comparing the mean value of their delta/theta band EEG power ratio: high for NREM and low for REM [14].

Figure 1.

Mouse sleep scoring. (A) The flowchart indicates the general process used for scoring vigilance state based on EEG and EMG feature values observed in 4-second-long epochs. An unsupervised three-state hidden Markov model was fitted to the sequence of feature values in each 24-hour-long recording. The state with the highest average EMG power corresponded to wakefulness (wake), while the other two states were labeled REM or NREM depending on whether their average theta band (6–9 Hz) EEG power was high or low, respectively. A subset of NREM epochs were further classified as slow wave sleep (SWS) if their EEG power in the low-delta band (0.5–2 Hz) exceeded a threshold determined from baseline recordings in all animals included in the study. (B) Snapshot showing EEG and EMG features used by the HMM classifier to track changes in vigilance state.

After each epoch was scored as wake, REM sleep, or NREM sleep, a subset of NREM sleep epochs were further identified as SWS, the deepest phase of NREM sleep, depending on the frequency and strength of delta oscillations in the EEG. This was done to quantitatively analyze the effect of thermoneutral exposure on sleep quality. SWS is characterized by a surge in EEG power in the low-delta band (0.5–2 Hz) during NREM sleep [15, 16]. To set a threshold for this distinction, the proportion of low-delta EEG power relative to delta (0.5–4 Hz) was computed for all NREM sleep epochs in each animal’s baseline recording and the 70th percentile value was estimated. Then the highest 70th percentile value of all baselines was used as the threshold to differentiate between light NREM sleep and SWS in NREM epochs in the rest of the experimental and control recordings [9]. This procedure limits the proportion of SWS in any baseline recording to 30% of NREM sleep and serves as a frame of reference for evaluating SWS observed in the experimental phase of the study. NREM epochs with a low-delta EEG power fraction greater than the SWS threshold are thus scored as SWS (Figure 1).

Sleep analysis

After scoring sleep using the GO-HMM algorithm for all recordings, we reviewed the quality of scoring and the corresponding EEG/EMG signals. Some animals were found to have poor EEG/EMG signal quality—excessive power line noise, or brief intermittent spikes and saturation artifacts that would influence the feature values fed to the scoring algorithm and corrupt its performance. Data from two animals in the experimental group and one in the control group had to be excluded due to pervasive artifacts of this nature, bringing the sample size for sleep analysis down from the implanted 19 to 16, with eight in each group. Of the sixteen remaining recordings, a few specific days in which artifacts affected the accuracy of sleep scoring were also excluded.

For recordings with satisfactory signal quality and reliable sleep scores, we did a detailed analysis of sleep metrics associated with each vigilance state. Each 24-hour recording was split into non-overlapping 30-minute intervals (48 per day), and the percent time spent in wake, NREM, REM, and SWS computed for that interval along with the number of bouts and mean bout duration. A bout of a particular state was marked by when the animal entered and left that state. SWS bouts were filtered to exclude isolated one-epoch-long bouts and merge neighboring bouts separated by a single epoch of another state.

Biochemical assays

All mice were euthanized either at the desired endpoint of the experiment, i.e. after the 5-week study period, or earlier if there was an adverse event such as the headmount coming loose. Following CO₂ euthanasia, brains were collected, rapidly frozen, and stored at −80°C until assayed. Mice that reached the desired endpoint (seven experimental and eight control animals) had their brains harvested for biochemical assays. Soluble Aβ was serially extracted in 0.2% diethylamine (DEA, in 50 mM NaCl) followed by RIPA buffer (50 mM Tris-HCl, 150 mM NaCl, 1% Triton X-100, 0.1% SDS, and 0.5% sodium deoxycholate), and 70% formic acid (FA); sandwich ELISA measurement of Aβ was performed by Ab42.5 capture followed by detection using end-specific antibodies for Aβ40 (13.1.1) or Aβ42 (12F4). A standard protease inhibitor cocktail (ThermoFisher) with EDTA was included in extraction buffers. Total tau (KHO0631) and phosphorylated tau pT181 (KHB0041) were measured in the RIPA fraction (Invitrogen). Assays were performed by personnel blinded to treatment conditions according to procedures previously described [17–19].

Statistical analysis

Statistical models were constructed to estimate the main effects of time (i.e. time of day in 30-minute intervals) and experimental group along with their two-way interaction on the mean bout duration, number of bouts, and percent time spent in each vigilance state as dependent variables. To account for repeated measures, generalized estimating equations [20] were used. Generalized estimating equations account for the inherent nesting in repeated measures designs and incorporate a working correlation matrix when the true correlation over time is unknown. The current study used an exchangeable correlation matrix. Baseline values (the first week of data) were averaged for each of the 48 daily measurement occasions and used as a time-varying predictor of the outcome. Significant group-by-time interactions were followed up to assess for between-group differences at each time point. The Benjamini-Hochberg linear step-up procedure [21] was used to control the false discovery rate, using a significance level of α = 0.05. All analyses were conducted using PROC GENMOD as part of SAS v 9.4 (SAS Institute, Inc.).

A test of correlation (Pearson coefficient; see Supplementary Material for Spearman rank coefficient) was performed between sleep metrics (% time, mean bout duration, number of bouts) for NREM, REM, wake, and SWS during the light period of the last experimental week and the levels of Aβ40 and Aβ42 in hippocampal and cortical tissue fractions without regard to treatment condition. More specifically, the sleep metrics used for this test were recomputed for the portion of the light period (ZT03-12 hours) during which the temperature setpoint for the experimental group was in the neighborhood of the TNZ for mice (26–30 °C). Only animals that had both measurable sleep metrics and tissue assays, and completed the full 4 weeks of the protocol, were included in this particular analysis (five control, six experimental).

The effect of treatment on each vigilance state was evaluated on the basis of percent time and mean bout duration. For each of these metrics, a one-way ANOVA was performed to compare measurements in each state (NREM, REM, wake, and SWS) derived from the baseline week and the last week of treatment in experimental and control animals, respectively. In the event of a significant overall outcome (type I error probability, p < 0.05), Tukey’s Honestly Significant Difference test was used to evaluate pairwise differences. In addition, Wilcoxon’s signed rank test was used to separately compare baseline metrics to the last week of treatment for control and experimental groups, respectively, since the measurements are repeated in each animal and therefore produce matched samples.

Results

Effect of thermoneutral treatment on sleep metrics

As the cage temperature increased from 22 to 30°C during the light period, the mean percent time spent in SWS increased significantly for the experimental group compared to controls, although the mean percent time spent in REM, NREM, and total sleep were unaffected (Figure 2). This difference is especially pronounced during the hours spent in the TNZ.

Figure 2.

Diurnal profile of mouse sleep metrics under experimental and control conditions. The plots depict the percent time spent in total sleep, NREM, REM, and SWS as a function of time of day averaged over the 4 weeks of treatment for control and experimental mouse cohorts (n = 8 each). The percent time spent in each vigilance state was first averaged for each animal over all days of recording available within a week, then over the available weeks of treatment, and finally over all animals within the cohort. Error bars represent the sample standard deviation of this measure over all animals in a cohort. Time of day is given in terms of Zeitgeber Time (ZT), with the lights on from 7:00 am (ZT 0 hours) to 9 pm (ZT 14 hours). Each data point represents values measured in the previous 30 minutes (i.e., the first data point covers ZT 0 to 0.5 hours). Cage temperature is raised in steps from 22 to 30°C during the light period for the experimental group. Asterisks (*) indicate times at which a statistically significant difference (p < 0.05) was detected between the experimental and control groups for a given metric. The mean percent time in SWS is significantly greater for experimental mice in the peak hours of thermoneutral exposure and significantly lower for some of the dark period at ambient temperature. Three isolated incidents when control mice have significantly more REM sleep than experimental mice were also found. However, no significant differences were found for NREM or total sleep between the two groups.

When averaged separately over the entire light or dark period (Figure 3), both mean percent time and mean bout duration of SWS in the light period increased significantly from baseline to the last week of treatment for experimental animals but not for controls. This was accompanied by a significant decrease in total sleep % time and REM bout duration in the dark period. However, no differences in percent time or bout duration between experimental and control groups were statistically significant, either at the end of treatment or in the baseline, in the light or in the dark, for SWS, REM sleep, NREM sleep, or total sleep.

Figure 3.

Effect of thermoneutral exposure on sleep metrics. These bar plots represent mean bout duration (A) and percent time (B) averaged over experimental (EXPT) and control (CTRL) animals in the baseline week (solid bars) and in the last week of treatment (hatched bars) for total sleep, NREM sleep, REM sleep, and SWS in the light (left) and dark (right) phases of the day. Asterisks (*) indicate a significant difference (p < 0.05) between the pairs indicated based on Wilcoxon’s signed rank test for pairwise comparisons. All data are shown as means on a logarithmic scale with error bars representing standard deviation (n = 8 in each group). EXPT mice had greater % time in SWS and mean SWS bout duration in the last treatment week compared to the baseline week during the light period. EXPT mice also had less % time in total sleep and REM mean bout duration in the last treatment week compared to the baseline week during the dark period.

Effect of thermoneutral treatment on amyloid neuropathology

Thermoneutral treatment induced a robust reduction in the most soluble (DEA) forms of Aβ40 and Aβ42 in the hippocampus; in contrast, no change in Aβ40 or Aβ42 was found in the cortex (Figure 4A). No significant experimental differences in FA Aβ, typically thought to represent plaque deposits, were found in either hippocampus or cortex (Figure 4B). Similarly, no significant experimental differences in tau or phosphorylated tau were found in hippocampus or cortex (Figure 4C). As such, all of the following discussion concerns only DEA-soluble Aβ unless otherwise specified.

Figure 4.

Effect of thermoneutral treatment on Aβ and Tau levels in the hippocampus and cortex. Bar plots of mean levels of DEA-soluble Aβ40 and Aβ42 (A), FA-soluble Aβ (B), total Tau, phosphorylated tau (pTau), and pTau to Tau ratio (C) in the hippocampus and cortex are shown for experimental (EXPT) and control (CTRL) groups. All data are shown as mean ± SD (n = 7 for CTRL; n = 8 for EXPT). Values are standardized to total protein concentration as determined by BCA assay, and any significant difference between the control and experimental groups is highlighted (* = p < 0.05; Mann–Whitney U-test). The treatment significantly reduced both DEA-soluble Aβ40 and Aβ42 in the hippocampus, but not in the cortex. Nor did the treatment have an effect on FA-soluble amyloid (representative of the least soluble pool of amyloid in the brain, including plaques), total Tau, or phosphorylated tau, in either hippocampus or cortex. The intermediately soluble Aβ fraction (RIPA) was similarly unaffected (not shown).

Correlation between sleep metrics and amyloid neuropathology

A statistically significant (p < 0.05) or nearly significant negative correlation was found between hippocampal levels of DEA-soluble Aβ and percent time, mean bout duration, and number of bouts of SWS (during ZT03-12 hours, the TNZ portion of the light period) in the final week of the study (Table 1). That is, mice that experienced more SWS in the last week of the experiment had lower Aβ levels in hippocampus. The correlations remained the same for percent time spent in SWS regardless of whether it was measured for the TNZ period alone (ZT03-12 hours) or for the entire light period (ZT0-14 hours), or whether SWS was measured relative to total time or only relative to time in NREM sleep (Table 2, Figure 5).

Table 1.

Correlation Between Aβ Levels and Various Sleep Metrics.

		Percentage		Mean bout duration		Bout count
Variables being correlated		Cerebral cortex	Hippocampus	Cerebral cortex	Hippocampus	Cerebral cortex	Hippocampus
Tissue Amyloid Level	Vigilance State	R (P-value)	R (P-value)	R (P-value)	R (P-value)	R (P-value)	R (P-value)
Aβ40	SWS	0.172 (0.613)	−0.712 (0.014)	0.205 (0.545)	−0.748 (0.008)	0.183 (0.590)	−0.741 (0.009)
Aβ40	Total sleep	0.032 (0.926)	−0.335 (0.314)	0.137 (0.688)	−0.456 (0.159)	−0.017 (0.960)	0.253 (0.453)
Aβ40	NREM	−0.100 (0.771)	−0.401 (0.221)	0.033 (0.924)	−0.726 (0.011)	−0.124 (0.717)	0.656 (0.028)
Aβ40	REM	0.171 (0.615)	0.271 (0.421)	−0.070 (0.837)	−0.249 (0.459)	0.104 (0.761)	0.435 (0.181)
Aβ42	SWS	−0.228 (0.500)	−0.687 (0.019)	−0.119 (0.728)	−0.706 (0.015)	−0.150 (0.660)	−0.680 (0.021)
Aβ42	Total sleep	0.099 (0.772)	−0.280 (0.405)	0.257 (0.446)	−0.338 (0.310)	−0.034 (0.921)	0.142 (0.677)
Aβ42	NREM	−0.320 (0.338)	−0.419 (0.200)	−0.276 (0.412)	−0.695 (0.018)	0.087 (0.800)	0.581 (0.061)
Aβ42	REM	0.547 (0.081)	0.345 (0.298)	0.212 (0.532)	−0.177 (0.603)	0.253 (0.452)	0.438 (0.178)
Aβ42/Aβ40	SWS	−0.639 (0.034)	0.427 (0.190)	−0.509 (0.110)	0.454 (0.161)	−0.552 (0.079)	0.521 (0.100)
Aβ42/Aβ40	Total sleep	0.080 (0.816)	0.284 (0.398)	0.048 (0.889)	0.758 (0.007)	0.073 (0.831)	−0.591 (0.056)
Aβ42/Aβ40	NREM	−0.366 (0.268)	0.209 (0.537)	−0.576 (0.064)	0.569 (0.068)	0.396 (0.229)	−0.644 (0.032)
Aβ42/Aβ40	REM	0.596 (0.053)	−0.042 (0.903)	0.343 (0.302)	0.378 (0.251)	0.281 (0.403)	−0.336 (0.313)

Correlation (Pearson coefficient) between brain Aβ levels and sleep metrics in each vigilance state during the TNZ period (ZT03-12 hours) of the last week of the experiment. combinations with p values under 0.05 are shown in boldface. A strong and significant negative correlation was found between all SWS metrics and Aβ levels (but not their ratio) in hippocampus. in the cortex, only the Aβ ratio shows a significant negative correlation with SWS %. A significant negative correlation was found between Aβ levels in hippocampus and NREM mean bout duration and Aβ40 levels in hippocampus. A significant positive correlation was found between Aβ40 levels in hippocampus and the number of NREM bouts. No significant correlation was found between cortical Aβ and any sleep metrics.

Table 2.

A Closer Examination of the Correlation Between SWS % and Aβ Levels Under Various Conditions.

			Cerebral cortex	Hippocampus
Aβ	SWS	Time period	R (P-value)	R (P-value)
Aβ40	SWS in NREM	Thermoneutral	0.205 (0.546)	−0.711 (0.014)
Aβ40	SWS in NREM	Light period	0.141 (0.680)	−0.725 (0.012)
Aβ40	SWS in total	Thermoneutral	0.172 (0.613)	−0.712 (0.014)
Aβ40	SWS in total	Light period	0.125 (0.714)	−0.715 (0.013)
Aβ42	SWS in NREM	Thermoneutral	−0.183 (0.590)	−0.674 (0.023)
Aβ42	SWS in NREM	Light period	−0.239 (0.480)	−0.684 (0.020)
Aβ42	SWS in total	Thermoneutral	−0.228 (0.500)	−0.687 (0.019)
Aβ42	SWS in total	Light period	−0.265 (0.430)	−0.682 (0.021)
Aβ42/Aβ40	SWS in NREM	Thermoneutral	−0.610 (0.047)	0.435 (0.182)
Aβ42/Aβ40	SWS in NREM	Light period	−0.633 (0.037)	0.486 (0.130)
Aβ42/Aβ40	SWS in total	Thermoneutral	−0.639 (0.034)	0.427 (0.190)
Aβ42/Aβ40	SWS in total	Light period	−0.649 (0.031)	0.464 (0.151)

Combinations with p values under 0.05 are shown in boldface. A strong and significant negative correlation was found between Aβ levels (but not their ratio) in hippocampus and SWS % during the TNZ period (ZT03-12 hours) as well as for the entire light period (ZT0-14 hours). This correlation held for net SWS % as well as SWS % within NREM sleep. In contrast, the ratio of Aβ levels in cortex, but not their individual levels, was significantly negatively correlated with net SWS % or SWS % within NREM.

Figure 5.

Correlation between Aβ levels and percent time in SWS. Scatter plots showing levels of Aβ40 (left), Aβ42 (middle), and the ratio of the two (right) in cortex (upper) and hippocampus (lower) against % time spent in SWS during the TNZ period (ZT3-12 hours) in the final week of treatment for control and experimental animals. A significant correlation (p < 0.05; Pearson coefficient) was found between SWS % and DEA-soluble Aβ levels in the hippocampus but not in the cerebral cortex. Animals with more SWS in the final week of treatment were found to have lower Aβ levels in hippocampus. Conversely, the correlation between SWS % and the ratio of Aβ42 levels relative to Aβ40 was weak in the hippocampus but significantly negative in the cortex. This suggests that gains in SWS affect both Aβ40 and Aβ42 indiscriminately in the hippocampus. Only mice that had both assays and sleep metrics available at the end of treatment, and completed the full course of intervention, were included (five control, six experimental).

Hippocampal Aβ showed a significant negative correlation with mean NREM bout duration, but a significant positive correlation with the number of NREM bouts (Table 1). Taken together, more SWS (by any metric), as well as more continuous NREM sleep (fewer but longer bouts) are correlated with reduced amyloid pathology in hippocampus. While the ratio of Aβ42 to Aβ40 in the hippocampus was not correlated with any measure of SWS, it did show correlations with NREM sleep bout duration and number of bouts (but not percent time) that indicated a reduction in the Aβ42/Aβ40 ratio with greater NREM and total sleep continuity (similar to the correlations with Aβ42 and Aβ40 individually).

In contrast, no significant correlations, either positive or negative, were found between any sleep metric and Aβ40 or Aβ42 levels in cerebral cortex (Table 1). However, the ratio of Aβ42 relative to Aβ40 was significantly negatively correlated with SWS % in the cortex (but not in hippocampus). This correlation was consistent whether SWS was measured during the TNZ period or the entire light period, or for total SWS or SWS as a proportion of NREM sleep (Table 2), which suggests an effect of SWS on amyloid pathology in the cortex consistent with that observed in the hippocampus.

The significant correlations observed appear to be independent of treatment since both experimental and control mice were pooled together for the test, and animals that experienced more SWS in general, and more continuous NREM sleep, also seemed to have less Aβ40 and Aβ42 in hippocampus. This is a strong indication that Aβ levels are correlated specifically with the characteristics of NREM and SWS, and not through some other putative physiological changes associated with thermoneutral treatment (Tables 1 and 2).

Discussion

Sleep and AD

It is well-established that there is a bidirectional relationship between sleep and AD [22]. Overnight sleep deprivation causes Aβ levels in the cerebrospinal fluid to increase by 25–30% in adult humans [23]. Studies conducted on the APP transgenic mouse model of AD found that Aβ levels in hippocampal interstitial fluid are significantly higher during chronic sleep deprivation [24, 25], and that the mice experience increased wakefulness following the formation of Aβ plaques [26]. Sleep disturbances in some form occur in up to 65% of patients with AD [27]. Changes in sleep, including decreased total sleep, NREM sleep, and REM sleep as well as sleep fragmentation, often occur before clinical signs of cognitive decline associated with AD manifest themselves [28]. This bidirectional relationship suggests that early management of sleep disturbances may slow or prevent AD. Slow wave activity during NREM sleep decreases with time in AD patients [29]. SWS disruption is also found to lead to increased Aβ levels in human CSF [30, 31]. A reduction in slow wave activity is also found in multiple mouse models of AD [32–34]. Conversely, optogenetic increases in slow oscillations (<1 Hz) stop amyloid plaque deposition in APP mice [35], which lends support to the idea of SWS enhancement as a therapeutic option for AD [36].

Our findings

Our results show that there is significant enhancement in SWS in experimental mice exposed to thermoneutral temperatures during the day compared to control mice even though the total time spent in NREM sleep is similar for both groups. This discovery indicates that temperature control is a viable method for titrating SWS in this mouse model. The accompanying (slight) reduction in mean total sleep time in the dark period by the end of the treatment regimen appears consistent with the notion that NREM sleep in the light period was more restorative and of better “quality.”

Sleep quality in general is determined not only by the proportion of SWS and total sleep time but also by sleep continuity (in other words, less fragmented sleep). In our experiments, we observed no significant effect of thermoneutral exposure on metrics that characterize sleep continuity such as mean bout duration and number of bouts in 3xTg-AD mice. This finding further indicates that the effect of thermoneutral warming on sleep is highly specific to the time spent in SWS. However, this perspective is altered by examining the correlation between various sleep metrics and amyloid pathology, as we discuss below.

In Tables 1 and 2, we examine the correlation of amyloid levels assayed in hippocampus and cortex with various sleep metrics (also see Figure 5). Mainly, we tested these correlations for sleep metrics measured specifically for the duration of the light period that experimental animals were exposed to thermoneutral temperatures (ZT3-12 hours) in the final week of treatment, pooling data for all experimental and control animals (n = 11) that reached the desired endpoint of the study. In doing so, we found that a consistent trend emerged in which all metrics of SWS (% time, mean bout duration, number of bouts) had a significant negative correlation with both Aβ42 and Aβ40 (but not the Aβ42/Aβ40 ratio) in hippocampus but not in cortex; conversely, only SWS % was negatively correlated with the Aβ42/Aβ40 ratio but not with the individual levels of Aβ42 and Aβ40. Furthermore, NREM mean bout duration and the mean number of bouts showed negative and positive correlations, respectively, with both Aβ42 and Aβ40; but not % time spent in NREM. Taken together, these outcomes suggest that: 1. Less fragmented (more continuous) NREM sleep, and not the net duration of NREM, is associated with amyloid clearance; and 2. More SWS overall, and a greater proportion of SWS within NREM (i.e. greater NREM sleep efficiency) is critical. This is strikingly consistent with the growing evidence that sleep fragmentation and reduced levels of slow wave activity in sleep EEG are associated with and even predictive of Alzheimer’s pathology.

Thermoneutral warming for 4 weeks selectively increased SWS and reduced only DEA-soluble Aβ levels in the hippocampus but not the cortex. This finding is especially interesting when juxtaposed with our recent finding that chronic sleep fragmentation increased the accumulation of RIPA-soluble Aβ42, selectively in the hippocampus but not in the cortex of female 3xTg-AD mice (Note: RIPA-soluble Aβ40 and DEA-soluble Aβ40 and Aβ42 were spared) [37]. Together, these findings suggest that AD-like neuropathology in the hippocampus is more sensitive to sleep regulation and further reinforces the bidirectional relationship between sleep and AD. Not only can sleep disorders accelerate the accumulation of Aβ, but sleep enhancement can slow (and perhaps arrest?) disease progression. At the same time, it is interesting to note that thermoneutral warming did not affect the FA-soluble fraction of Aβ, suggesting that plaque deposition was not affected in this 4-week-long exposure study.

It is possible that increasing the ambient temperature to the thermoneutral range reduced hippocampal Aβ levels by some mechanism related to thermoregulatory physiology other than sleep, such as relief from cold stress or a reduction in metabolic demand, for instance. However, as we see in Figure 5, even control mice that did not receive thermoneutral treatment but had higher amounts of SWS, and more continuous bouts of NREM sleep, were found to have lower levels of Aβ in hippocampus (and lower Aβ42/Aβ40 in cortex) at the end of the treatment period. Although we cannot altogether reject the possibility that thermoneutral treatment lowers Aβ levels due to its metabolic or physiological effects rather than sleep effects per se, the correlation between SWS and hippocampal Aβ levels across both control and experimental animals suggests that SWS does play a mechanistic role in the regulation of Aβ.

Finally, in the study, we made an arbitrary decision to apply a 4-week-long intervention, Whether similar outcomes would be obtained with a longer intervention or if the effect would plateau after a certain duration is an open question; the effects on behavior and cognition are also unknown. In order to explore these possibilities, a future study in which cohorts are euthanized for assays at various serial time points throughout treatment is warranted.

Significance

The strong correlation between sleep disruption and AD pathology suggests that sleep intervention can be a promising approach for AD prevention and treatment. A wide variety of sleep intervention strategies have been studied. In humans, behavioral strategies in the form of adult day health care [38], bright light therapy, [39] and continuous positive air pressure [40] have achieved various degrees of success in improving sleep quality. In an AD mouse model, optogenetic stimulation to restore slow oscillations also halts Aβ deposition [35]. Intervention methods specifically targeting SWS like transcranial direct current stimulation (tDCS) and transcranial magnetic stimulation are also being developed [36]. Due to the long interval between the early pathology of AD and clinical diagnosis, it is hard to perform studies on human participants on the effect of early sleep intervention.

Thermoneutral treatment as a means of SWS intervention is new and unexplored, especially in the treatment of AD. There is one study on the effect of temperature on Aβ levels in an AD mouse model that focuses on thermoregulation [41]; but the effect on sleep and SWS was not explored. With SWS possibly being the agent of change in AD pathology in our thermoneutral exposure model, we hope that our study leads to new noninvasive and practical methods of SWS intervention for AD.

Limitations

Due to limited resources, our study only used mice of one age group (12–15 months old). Furthermore, only female mice were used. However, it is important to note that AD is approximately twice as prevalent in women as in men [42].

It is to be expected that the 12–15 months old 3xTg-AD mice will show the plaques and tangles associated with amyloid and tau pathology, and immunohistochemical analysis of the tissue would have further clarified whether our treatment affected these features in any way. Unfortunately, due to resource constraints at the time of the experiments, we were unable to sequester and stain tissue for such an analysis, a limitation that we intend to address in future studies.

Our study has a limited sample size. Of the 19 animals used in this study, only brain tissues from 15 animals (seven control, eight experimental) were available in the end for biochemical assays. EEG/EMG data from the later weeks of treatment in certain animals also had to be excluded due to signal deterioration and artifacts that made sleep scoring inaccurate and unreliable, after which data from only 16 animals (eight control, eight experimental) were available. The difficulty of recording high-quality signals over a prolonged period limits our ability to collect data for a bigger sample size. The overall outcome of the study, including the reduction in DEA-soluble Aβ levels found by biochemical assays could possibly have been reinforced by a larger sample size.

All mice used in this experiment were instrumented for EEG/EMG-based sleep analysis with a head-mounted chip held to the skull by bone screws that served as EEG electrodes. Though minimally invasive, the bone screws making contact with the dura may provoke an inflammatory response, a potential confounding variable in our analysis of amyloid deposition. Unfortunately, we were unable to perform immunohistochemical assays in this preliminary study and therefore cannot comment on these potential effects; nor did we have enough mice of the desired age to perform the treatment on a larger sample without EEG. However, both experimental and control cohorts suffer from the same limitation and we assume that any effects of implantation will even out. In future studies, a subset of mice in each treatment group will undergo treatment without EEG analysis in order to clarify any implant-related effects.

Finally, our experimental design does not address whether the decrease in Aβ pathology induced by thermoneutral warming is directly mediated by SWS or indirectly mediated by a cascade of nonspecific physiological changes. Future studies to explore this will need to be conducted.

Conclusion

Increasing the ambient temperature to the thermoneutral range was found to be effective at promoting SWS in female, middle-aged 3xTg-AD mice. Thermoneutral treatment also reduces DEA-soluble Aβ levels in the hippocampus, a critical neural substrate for learning and memory that exhibits profound pathological changes in AD. We also found a significant correlation between the time spent in SWS and levels of Aβ found in hippocampus but not the cortex; the same was true of the correlation between NREM sleep continuity (but not the net amount) and Aβ. These findings were independent of whether mice received thermoneutral treatment and indicated that sleep depth, continuity, and efficiency all contribute to amyloid clearance in the brain. These outcomes not only reinforce previous findings indicating an important link between SWS and Aβ metabolism but also that thermoneutral exposure might be a promising strategy for improving and testing the benefits of SWS in experimental models of neurodegenerative disease.

Disclosure Statement

Financial disclosure: none. Nonfinancial disclosure: none.

Data Availability

The data underlying this article will be shared on reasonable request to the corresponding author.