We used male and female Tg2576 (Hsiao et al., 1996) mice (aged 6 months old for early intervention, and 11 months old for late intervention) overexpressing a mutant form of amyloid precursor protein (APP), APPK670/671 L, linked to early-onset familial AD, and non-transgenic WT mice from the same strain (Taconic Biosciences; Cologne, Germany). The animal room temperature was constant at 21–23°C, with a 12:12 h light:dark cycle (8 am/9 am ON, 8 pm/9 pm OFF, according to daylight saving time). Mice were housed in groups, had access to food and water ad libitum, and received daily routine health checks throughout the study. The study was conducted with the approval of the Cantonal Veterinary Office Zurich under license ZH210/17.

All mice underwent electroencephalography/electromyography (EEG/EMG) implantation surgery two weeks prior to the start of the intervention periods (early and late intervention, Figure 1). Subjects were randomly allocated to the placebo or the SO treatment group, within each genotype. Baseline (BL) cognitive ability and 24-h sleep/wake behavior were assessed before the treatment period started. SO (Xyrem®, UCB; 300 mg/kg, p.o.; SO half-life time = 20–40 min (Robinson and Keating, 2007)) and placebo (Xyrem®-placebo—no sodium oxybate in the formula, UCB; 300 mg/kg, p.o.) treatments were administered two times daily (ZT1 and ZT9) for 2 weeks [weekdays ON, weekends OFF regime (Kollarik et al., 2023; Morawska et al., 2021)]; 24-h sleep/wake recordings were re-evaluated during treatment, and cognitive ability was re-tested after the treatment period was concluded. Finally, the mice were euthanized via intraperitoneal injection of sodium pentobarbital followed by cervical dislocation, and their brains were harvested for further molecular and histological analyses.

We performed all surgical procedures under deep inhalation anesthesia with isoflurane (4–4.5% for induction in anesthesia box; 2–2.5% for maintenance using a nose cone fitting) and applied lidocaine, a local anesthetic (Xylocaine, Zurich) on the surface of the head skin prior to surgery. At the end of the surgery, we administered a combination of an anti-inflammatory (5 mg/kg, s.c., Metacam®, Zurich) and a pain relief drug (0.1 mg/kg, s.c., Temgesic®, Zurich), to further prevent postoperative inflammation and pain.

The animals were implanted as described before (Buchele et al., 2016). Briefly, we placed two stainless steel screws (Bossard, #BN650, 1,421,611), one for each hemisphere, located 2 mm posterior to Bregma, 2 mm lateral from midline. We then inserted two gold wires bilaterally into the neck muscles and applied sutures to close the skin around the implant. We connected the peridural screws and the muscle leads to a pin header (Farnell, #M80-8530445) for EEG and EMG recording, and fixed the structure with dental cement. Postsurgical analgesia was administered throughout the 3 days following the surgery during both light (0.1 mg/kg, s.c., Temgesic®, Zurich) and dark (1 mg/kg via drinking water, Temgesic®, Zurich) periods. We monitored wound healing, body weight, and home cage activity on a daily basis during the first week after the surgery and one time per week thereafter.

A 24-h EEG/EMG recording took place after a ~ 48-h habituation to recording conditions in all mice. The mice were tethered to a rotating swivel and had access to every corner of the recording chamber. Signals were amplified using an N700 polysomnography (PSG) amplifier (Embla, Ontario, Canada) with a bipolar montage, digitized and collected at a sampling rate approximately 200 Hz and an input range of ±200 mV using a 16-bit digital-to-analog-converter. Data were collected via Somnologica Science, version 3.3.1 (ResMed, Saint-Priest, France). The scoring of the vigilance states was determined using SPINDLE (Miladinović et al., 2019). Post-processing of EEG signals was carried out using MATLAB (ver. R2019b) as described before (Kollarik et al., 2023). Briefly, we first removed artifacts by detecting clipping events, followed by a three-point moving average and a basic Fermi window function. Then, we resampled EEG signals at 200 Hz and filtered them between 0.5 and 30 Hz using low- and high-pass zero-phased equiripple FIR filters (Parks–McClellan algorithm; applied in both directions, filtfilt). We inspected the signal for regional artifacts not detected during automatic scoring or the above-mentioned preprocessing steps. Within scored NREMS, brief portions of the signal (<7 sample points at 200 Hz) > ± [8× interquartile range] were reconstructed by piecewise cubic spline interpolation from the neighboring points. Finally, we extracted measures of power in specific bandwidths, by performing a spectral analysis of consecutive 4-s epochs (FFT routine, Hamming window, 2-s overlap, and resolution of 0.25 Hz) and normalized the data indicating the percentage of each bin with reference to the total spectral power between 0.5 and 30 Hz.

We trained the mice in a forced alternation T-maze test adapted from a previous study (Deacon and Rawlins, 2006). First, we fed the mice with the reward in the home cages to habituate them to their taste and eliminate hyponeophagia. From the day prior to the test until the end of the test, all animals were restricted to eating 1.5 g/mouse/day of regular chow to increase the motivation for the reward. We split the 1.5 g of food into small pieces and recorded the weights of the mice daily throughout the experiments in order to monitor and prevent potential monopolization of the food by the dominant mouse in the cage. On the second day, we habituated the mice to the enclosed maze by raising all doors, filling the food wells with rewards, and placing at once an entire home cage group in the maze for 3 min. After the habituation day, we trained each subject in three trials per day for 2 days, with an inter-trial interval of ca. 20 min for each mouse. Each trial consisted of two parts: the first one was a forced attempt, in which one of the arms was blocked by a door constraining the mouse to choose the open arm. The second part of the trial consisted of a choice attempt that started after the mouse consumed the reward in the open arm—or explored the arm for 2 min—and was placed back in the starting point facing away from the goal arms, and the forcing door was raised. Thus, in the second half of the trial, both goal arms were opened, but only the arm that had been previously blocked contained a food reward. Entering this previously unexplored, now rewarded arm was recorded as a “correct” response, while choosing the initially sole-available arm was recorded as an “incorrect” response. On the experimental day, each mouse was subjected to four trials. Testing timing before and after treatment was kept constant for each mouse, and all tests were conducted between ZT2 and ZT5.

After fresh harvesting, the brains were split into left and right hemi-brains. A randomly chosen hemi-brain was post-fixed in 4% paraformaldehyde in phosphate saline buffer (PBS), and the remaining hemi-brain was placed in a 1.5-ml Eppendorf tube and immediately snap frozen in liquid nitrogen. The post-fixed hemi-brains were dehydrated in sequential 15 and 30% sucrose in PBS solutions, then embedded in 30% sucrose molds, and then cut into 40-μm-thick sagittal sections using a freezing stage-equipped microtome (Leica SM2000 R). Every sixth section was selected and mounted onto gelatinized SuperFrost Ultra Plus slides and dried at 37°C overnight, amounting to a total of five sections per mouse.

We mechanically homogenized the frozen hemi-brains and sequentially extracted soluble and insoluble proteins with diethylamine (DEA) and formic acid (FA), respectively (Casali and Landreth, 2016). Aβ content in soluble and insoluble brain extracts was assessed by sandwich ELISA as indicated by the manufacturer of the kit (Aβ-40: KMB3481, Aβ-42: KMB3441, Thermo Fisher). To serve as controls, the placebo-treated Tg2576 brains were blindly processed in an identical fashion and in parallel to SO-treated Tg2576 brains throughout the entire process. Soluble and insoluble Aβ-40 and Aβ-42 were normalized to brain tissue mass and expressed in picogram per milligram.

Data analyses were carried out using IBM® SPSS® Statistics 25 software, and the data were visualized by GraphPad Prism 9 (GraphPad Software, Inc., San Diego, CA). Outliers were detected with boxplots, normality was assessed using skewness and kurtosis for each variable, and homogeneity of variances was assessed with Levene’s test. Delta activity gain was analyzed with multiple t-tests for each hour, and p-values were corrected with the false discovery method in GraphPad Prism 9. NREMS proportion was analyzed with two-way ANOVA (genotype*treatment). Analyses of the main effect of genotype and treatment were separately run and followed by pairwise comparisons. All pairwise comparisons were reported with 95% confidence intervals (CI) and p-values. The cognitive performance of the mice was assessed in two steps. First, we conducted independent t-tests to investigate the baseline differences between genotypes in each age group, separately. Second, we ran a paired t-test to evaluate the change in cognitive performance from baseline to after treatment in Tg2576 mice. Biserial correlation analysis was used to assess the relationship between delta activity gain and cognitive performance in Tg2576 mice after cognitive performance was assigned to two categories: fail (≤ 50) or success (> 50). Briefly, we first computed point-biserial correlation in SPSS and then transformed the point-biserial correlation coefficient (r_pb) into the biserial correlation coefficient (r_b) as it has been described before (Field, 2012). The difference in levels of soluble and insoluble Aβ-40 and Aβ-42 between placebo- and SO-treated Tg2576 mice was assessed with an independent t-test. The plaque burden in the hippocampus and cortex was calculated by dividing the total area covered with plaques by the total area of ROI. Then, the data were analyzed with a two-way ANOVA (genotype*treatment), followed by analyses of the main effects of genotype and treatment, separately. Subsequently, pairwise analyses were run to assess whether there was a difference in plaque burden between placebo- and SO-treated mice within the same genotype and whether there was a difference between Tg2576 and WT mice within the same treatment. Pearson’s correlation analysis was run to evaluate the relationship between delta power gain and plaque burden in the hippocampus. Quantitative estimates of amyloid plaque burden in WT and Tg2576 young and aged mice identified through Congo red staining were analyzed using independent-samples t-test.

Decreased delta power during sleep is associated with AD (Lee et al., 2020) and may be one of the factors that worsen disease symptoms (Bianchetti et al., 1995; Wennberg et al., 2017). Considering that Tg2576 mice present with reduced delta activity in NREMS (Kollarik et al., 2023), we first aimed to rescue the impaired delta activity by SO pharmacotherapy. We administered placebo or SO (300 mg/kg, p.o.) at ZT1 and ZT9 during the light period for all groups. As the first step, we evaluated whether such a pharmacological regime exerted an increase in the time spent in NREMS compared to the baseline. Our results demonstrate that administration of SO (300 mg/kg, p.o.) did not significantly alter the time spent in NREMS in any of the groups in comparison with their own baseline (Figures 2A,B). Delta activity gain, i.e., increased SWA, from baseline was especially observed after the administration of ZT9 in all groups treated with SO when compared with the placebo-treated groups (Figures 2C–F). A difference in delta activity gain between SO and placebo after the first hour of administration was apparent only in Tg2576 mice in the late intervention cohort (Figure 2F).

Studies showed that memory loss starts at approximately 6 months of age in Tg2576 mice, which coincides with the appearance of detergent-insoluble Aβ aggregates (Westerman et al., 2002). Therefore, we first assessed the cognitive performance of Tg2576 mice and WT controls in the T-maze forced alternation test in both age groups, separately, to determine the existence of treatment-naïve (baseline, BL) differences in memory (Figures 3A,B). We observed that Tg2576 mice in both age groups (early intervention: n = 17, M = 42.65, SD = 14.99; late intervention: n = 20, M = 53.75, SD = 29.43) performed poorly [early intervention: t(26.21) = 4.884, p < 0.001, Cohen’s d = 1.64; late intervention: t(36) = 2.753, p = 0.009, Cohen’s d = 0.91] compared to age-matched WT controls (early intervention: n = 18, M = 79.63, SD = 28.18; late intervention: n = 18, M = 75.93, SD = 18.28) in the T-maze test.

On the other hand, increases in high-amplitude SWA during NREMS may be associated with improved cognition (Wilckens et al., 2016). Thus, we analyzed cognitive performance of Tg2576 mice in both age groups after treatment (AT) with placebo or SO (Figures 3C,D). Our analyses revealed that cognitive performance in the early intervention SO group improved AT (Figure 3C) [BL: M = 48.81, SD = 18.90; AT: M = 70.24, SD = 9.45; t(6) = 3.286, p = 0.017, Cohen’s d = 1.43], while treatments with placebo did not affect this outcome [BL: M = 38.33, SD = 10.54; AT: M = 42.50, SD = 28.18; t(9) = 0.460, p = 0.657, Cohen’s d = 0.20]. There were no significant effects of either placebo [BL: M = 52.50, SD = 28.61; AT: M = 46.67, SD = 27.27; t(9) = −0.771, p = 0.460, Cohen’s d = 0.16] or SO [BL: M = 55.00, SD = 31.72; AT: M = 63.33, SD = 31.23; t(9) = 0.905, p = 0.389, Cohen’s d = 0.26] over Tg2576 mice cognitive ability in the late intervention group (Figure 3D).

After determining increased delta activity and improved cognitive performance in Tg2576 mice after 2 weeks of oral SO administration, we examined whether there is an association between both measures. To this end, we first assigned the cognitive performance scores as fails (≤ 50%) or successes (> 50%) and then ran a biserial correlation between treatment-elicited delta activity gain after the second administration (delta gain in the 9th + 10th hours) in both treatments groups, i.e., placebo and SO, and cognitive performance in Tg2576 mice of both age groups, separately. Our analyses showed a trend toward a positive correlation between delta activity gain and cognitive performance in the early intervention group [n = 8, r_pb(6) = 0.653, r_b = 0.819, p = 0.079, Figure 4A], while there was no apparent correlation in the late intervention group [n = 9, r_pb(7) = 0.563, r_b = 0.786, p = 0.114, Figure 4B].

Despite the progressiveness of AD pathology in Tg2576 mice has been widely reported before (Kawarabayashi et al., 2001), differences in the time onset suggested across individual colonies may lead to uncertain baseline conditions for longitudinal histopathological assessments. To determine the exact pathological stage in terms of plaque formation in the 6- and 11-month-old Tg2576 mice used in our study, as the basis for additional assessments of the effect of our treatment on pathology, we performed Congo red staining of sagittal brain sections from non-treated WT and Tg2576 mice across both age groups and quantitatively assessed plaque burden in the hippocampus (Figure 5). Representative photomicrographs from 11-month-old Tg2576 mice (Figures 5A–E) demonstrated a high level of plaque deposition in hippocampal areas. Stereological analyses confirmed a very low and indifferent plaque burden in WT and Tg2576 mice at the early disease stage [t(6) = 0.169, p = 0.871, Cohen’s d = 0.12, Figure 5F], while evidencing a higher plaque burden in Tg2576 at moderately advanced disease stage [t(7) = −3.850, p = 0.006, Cohen’s d = 2.50, Figure 5G] than age-matched WT littermates.

Based on the quantification results of the Congo red staining showing now discernible plaque load in the Tg2576 mice at 6 months of age, agreeing with previous literature (Kawarabayashi et al., 2001), we proceeded to evaluate the effect of SO and placebo treatments onto amyloidosis in this premorbid (plaque free) stage via determinations of soluble and insoluble Aβ-40 and Aβ-42 levels in brain tissue from Tg2576 mice via ELISA. We observed that soluble Aβ-40 [t(8) = 0.638, p = 0.791, Cohen’s d = 0.18] and Aβ-42 [t(8) = 0.280, p = 0.864, Cohen’s d = 0.11, Figures 6A,B] did not differ between treatments. Insoluble Aβ-40 level, on the other hand, was higher [t(8) = 2.846, p = 0.022, Cohen’s d = 1.80] in the brains of SO-treated Tg2576 mice (n = 5, M = 4.982, SD = 0.379) in comparison with what was observed in the brains of placebo-treated Tg2576 mice (n = 5, M = 4.459, SD = 0.158, Figure 6D). Insoluble Aβ-42 levels did not differ [t(8) = 0.597, p = 0.567, Cohen’s d = 0.38, Figure 6E] between treatments. Aβ-42/Aβ-40 ratio is considered a more accurate way of reflecting the progress of AD (Wiltfang et al., 2007) than Aβ-40 and Aβ-42 expressed separately. Therefore, we followed our analyses with the assessment of treatment-dependent changes in Aβ-42/Aβ-40 ratio. We did not detect a difference in the soluble Aβ-42/Aβ-40 ratio between treatments [t(8) = 0.695, p = 0.723, Cohen’s d = 0.23, Figure 6C], while our results indicated a trend with a large effect size toward a lower insoluble Aβ-42/Aβ-40 ratio in the brains of SO-treated Tg2576 mice [t(8) = 2.009, p = 0.079, Cohen’s d = 1.27, Figure 6F] compared to placebo-treated controls.

Tg2576 mice have been reported to develop Aβ plaques as they age (Benzing et al., 1999; Hsiao et al., 1996; Ishii et al., 2014; Kawarabayashi et al., 2001), as confirmed by the significant difference in the load of Congo red-stained plaques between WT and Tg2576 mice at 11 months of age. In fact, the hippocampus and cortex are two primary regions vulnerable to plaque accumulation in AD (Bero et al., 2011). Therefore, we investigated plaque burden in both these key areas in brains from the late intervention cohort treated with placebo or SO. We included WT mice in our analyses as negative controls. We found strong evidence of reduced plaque burden in the hippocampal (Figures 7A–C) and cortical (Figures 7D–F) brain areas of Tg2576 mice after 2 weeks of treatment with SO. The interaction effect between genotype and treatment on plaque burden in the hippocampus (Figure 7C) was a trend with a large effect size [F(1, 28) = 3.807, p = 0.061, η_p² = 0.120]. The main effect of genotype on plaque burden was significant [F(1, 28) = 10.820, p = 0.003, η_p² = 0.279], while the main effect of treatment was a trend [F(1, 28) = 3.634, p = 0.067, η_p² = 0.115]. Following pairwise comparisons within the genotype, we observed a significant decrease in plaque burden in SO-treated Tg2576 mice, while there was no change in the placebo treatment group. There was no effect of treatment on the naturally low plaque burden in the hippocampus of WT mice. In particular, after 2 weeks of SO treatment, plaque burden in the hippocampus of Tg2576 mice was at a similar level to the plaque burden in WT mice. In the cortex, plaque burden analyses revealed similar results to those in the hippocampus (Figure 7F). The interaction effect between genotype and treatment on plaque burden in the cortex was a trend with a large effect size [F(1, 28) = 4.054, p = 0.054, η_p² = 0.126]. There was a significant main effect of genotype [F(1, 28) = 13.965, p = 0.001, η_p² = 0.333] but not of treatment [F(1, 28) = 2.951, p = 0.097, η_p² = 0.095] on plaque burden. We followed the analyses with pairwise comparisons and observed a greater plaque burden in placebo-treated Tg2576 mice in comparison with SO-treated mutants. Within the placebo-treated mice, Tg2576 mice showed a higher plaque burden in the cortex than WT, while SO-treated mice did not differ between genotypes.

Our findings of increased delta activity gain and decreased plaque burden in Tg2576 mice after treating with SO led us to investigate the relationship between both parameters. Therefore, we ran a Pearson’s correlation between the plaque burden in the hippocampus and cortex (Figure 8), and delta activity gain after the second administration of the treatment (delta activity gain in the 9^th + 10^th hours). Our findings demonstrated that mice that had higher delta activity gain showed less plaque burden in the hippocampus (r = −0.882, p = 0.020; Figure 8A), while no association was found between cortical plaque burden and delta activity gain (r = −0.065, p = 0.902; Figure 8B).