We used 4-month-old APP/PS1 transgenic mice purchased from the Model Animal Research Center of Nanjing University. Transgenic mice were randomly divided into five groups: APP/PS1, Aricept, PNS, GP, and NP groups (n = 14). Their littermate wild-type mice were the control (n = 14). Mice were kept in cages (two mice per cage) and maintained at a constant temperature on a 12-h light–dark cycle with access to food and water freely. All procedures concerning the care, treatment and dissection are following the Animal Ethics Committee of Beijing University of Chinese Medicine (No. BUCM-4-2016040301-2001).

Based on previous studies (Yang, et al., 2014), the dosage of these drugs was converted according to the body weight index between human beings and mice. PNS was 17 mg/kg.d, GP 19 mg/kg.d, and Aricept 0.65 mg/kg.d. NP is the combination of PNS and GP. Drugs were given once a day for 3 months. WT and APP/PS1 groups have equal volume vectors. To minimize the stress effect caused by intragastric administration, we developed a three-step self-feeding system to guide mice to feed (Figure 1): 1) Food ball or drug ball preparation. Food pallets were ground into powder, and food ball was produced by mixing with 200 µL saline or drug solution. 2) Feeding habit management. We trained the mice of feeding habit for 1 week. From 7–9 a.m., we removed food pallets to drive hunger. During that time, we put a food box at the edge of the cage. After 2 hours of fasting, the food ball or drug ball was applied in the food box. Due to hunger, the mice promptly grabbed the food ball and ate it up. One hour later, food supply was provided as normal. 3) Drug administration. After the training, we found that mice ate the food ball within 5 min.

Golgi staining was performed using the Hito Golgi-Cox OptimStain^TM Kit, according to the manufacturer’s instructions. In brief, the fresh brain was kept in the mixture impregnation solution at room temperature in the dark and replaced into the impregnation solution on the next day. The brain was stored for 2 weeks, following the replacement of solution 3 for 48 h at 4°C and preparing 150-μm tissues sections. Neurons were stained with the mixture of solution 4, 5, and distilled water (1:3:5 ratio) for 10 min. The slides were dehydrated by continuous processes and mounted using a mounting medium. For a dendritic structure analysis, we selected neurons (5 per animal) located in the CA1 region of the hippocampus. These neurons must be visually inspected for the integrity of dendritic branches and isolated from neighboring neurons to avoid interference of synaptic tracing by Image-Pro Plus. The dendritic length from the soma and the number of intersecting radii every 10 μm were used to estimate dendritic arborization complexity (Figures 2A,B). Furthermore, we randomly chose the second dendritic branches at both the basal and apical segments. Images of dendritic spines were taken using the confocal microscope (Olympus; magnification, 100 ×) and calculated.

Western blot and analysis were carried out, as described previously (Liu et al., 2011). In brief, proteins (5–10 μg of total protein) were separated by SDS-PAGE, transferred to PVDF membranes, and blocked in 5% skimmed milk. The primary antibodies were rabbit anti-APP (1:2000, ab241592, Abcam), rabbit anti-PSD-95 (1:1000, ab238135, Abcam), or mouse anti-β-actin (1:4000, #3700, CST). The secondary antibodies were corresponding horseradish peroxidase–conjugated ones. A chemiluminescent substrate kit (Thermo Fisher Scientific, United States) was used and developed by using an X-ray film (Kodak) in the dark room. The developed films were scanned, and the percentage of the band relative intensity was analyzed by ImageJ software.

Brains were dissected and processed for serial paraffin sections (6 µm). By conventional dewaxing and rehydration, sections were immersed for 20 min in 0.1% Thioflavine-S (Sigma-T1892, Sigma-Aldrich, United States) PBS solution at room temperature. The green fluorescence-stained plaques were visualized by fluorescence microscopy. The number of plaques in the cerebral cortex and hippocampus was analyzed respectively.

The hippocampal and cortical homogenates were collected for supernatant fraction and allowed to precipitate. The pellet was further re-suspended in 70% formic acid, received sonication for 10 s, and was centrifuged at 17,000 rpm for 1 h at 4°C. The supernatant was collected for the insoluble Aβ ELISA assay. The insoluble levels of Aβ1-40 and Aβ1-42 were determined by using commercial kits (KHB3481 and KHB3441, Invitrogen, United States), according to the manufacturer’s instructions.

MWM is a classical method to evaluate hippocampal-dependent spatial learning and memory. Before the behavioral test, mice were acclimatized to the laboratory room for at least 30 min. The test took place between 8 a.m. up to 4 p.m. The apparatus consisted of a white circular pool (80 cm in diameter and 50 cm deep) divided into four equal imaginary quadrants for data analysis. Water was mixed with milk powder, and the temperature was maintained between 21 and 23°C. The hidden platform is 10 cm in diameter and placed at one quadrant. Three visual stimuli hung on the curtains that surrounded the pool. Mice underwent two separate trials, a training trial, and a test trial. For the training trial, mice were trained to reach the hidden platform within 60 s; if the mouse failed to find the platform, the experimenter guided it and allowed the mouse to stand on the platform for 10 s. Each mouse had four trials per day for every four quadrants. The training trial was conducted for five consecutive days. For the test trial, on the sixth day, the platform was removed, and the mice were released at the opposite quadrant. The latency to the platform and the number of crossings over the platform area were recorded. Also, the swimming speed was calculated to confirm the standard moving capability.

The passive avoidance shuttle box test is a widely accepted rapid and straightforward method to evaluate fear memory. The apparatus consisted of two same size compartments separated by a wall with a guillotine door. One of the two chambers was illuminated, and the other was dark. Floors constructed of 3.175-mm stainless-steel rods were set 8 mm apart and connected to a shock generator. Mice freely adapted to the apparatus for 5 min After adaption for 24 h, mice underwent two separate trials, a training trial (5 min each mouse) and a test trial (10 min each mouse). For the training trial, mice were initially placed in the illuminated chamber. When mice entered the dark compartment, an electrical foot shock (0.4 mA) of 2 s duration was delivered through the stainless-steel rods. For the test trial, given 1 day after the training trial, the procedure was performed in the same manner without the electric shock and the latency of step through to the dark compartment and the error in 10 min were recorded as the reflection of the learning. The latency to enter the dark compartment recorded up to 600 s.

All the analyzed data presented as mean ± standard error of the mean (S.E.M.). We performed one-way ANOVA or two-way ANOVA followed by post hoc comparison of the mean ± S.E.M. using Tukey’s multiple comparison test and Bonferroni’s or Dunnett’s T3 methods. Values of p < 0.05 were considered as statistically significant. The graphical abstract was created with BioRender.com.

To mimic self-feeding without stress, we designed a self-feeding system to guide mice to take drugs (Figure 1A). First, mice accommodated the feeding system for 1 week. Two days before drug feeding, mice fasted 2 h earlier than the food balls provided. Driven by hunger, they took the food ball spontaneously. One hour later, we provided the normal food supply. After the training, 4-month-old APP/PS1 transgenic mice were administered with the drug and continued for 3 months. The timeline for early intervention and the schedule for behavioral experiments are shown in Figure 1B.

To assess Aβ pathology in the brains, we observed amyloid deposits and insoluble Aβ production in the cortex and hippocampus. First, we illustrated the typical feature of the whole brain by Thioflavine-S staining (Figure 2A). No positive staining was presented in WT mice, but we observed numerous amyloid plaques in the brains of APP/PS1 transgenic mice. The number of amyloid plaques in all intervention groups decreased significantly (Figures 2B,C).

Accordingly, no insoluble Aβ peptide was detected in WT mice. In contrast, APP/PS1 mice showed significantly higher levels of insoluble Aβ peptides (Figures 2D–I). In addition, NP and PNS significantly decreased the levels of Aβ40 and Aβ42 in both cortex and hippocampus areas (Figures 2D–I). GP only showed the effect on the production of Aβ42 in the cortex (Figure 2E). Furthermore, we examined the amyloid precursor protein (APP) protein level in the cortex. We found that PNS did not change the protein expression of APP, while GP significantly reduced the APP expression (Figures 2J,K).

We next observed the dendritic structure since their enrichment is impaired early than clinical pathologies and degenerated throughout AD processing. The typical structure of CA1 pyramidal cells is shown in Figure 3A. Since the length of dendrites and their branches seem enriched under drug administration, we further measured the distance of dendrites from soma and the number of intersections per 10 µm from soma to 0–40 μm, 40–160 μm, 160–210 μm, and more than 210 µm accordingly (Figure 3B). Analyzing neuron tracing, APP/PS1 mice showed the shorter length of dendrites, and only NP significantly promoted the length (Figure 3C). Then, the number of intersections per 10 µm was analyzed accordingly since the intersection number considered another sign of dendritic structure richness. The number of intersections was analyzed in four categories: 0–40 μm, 40–160 μm, 160–210 μm, and more than 210 µm. In terms of the total number of intersections, all the intervention groups were significantly higher than APP/PS1 mice (Figure 3D). By category, the positive drug Aricept showed significant richness at the near distance, 0–160 μm, while NP showed significant richness at the far distance from the soma, 40–210 µm and ≥210 µm (Figures 3F–H). PNS or GP partially contributed to the richness of intersection (Figures 3E–G). In addition, Golgi staining calculated the apical and basal spinal density. All the intervention groups significantly increased the spine number at the apical spinal density of the pyramidal cells. Only NP and GP significantly increased the basal spine density (Figures 3I–K).

Furthermore, we detected the synaptic function-related protein level by using Western blot. APP/PS1 mice had a lower protein level of PSD-95 compared with their littermate WT control. Only NP significantly increased the expression of PSD-95 in the hippocampus (Figure 3L). These results indicated that NP enriched the dendritic structure and promoted synaptic-related protein expression in the hippocampus.

We next confirmed the effectiveness of using behavioral tests. Using the MWM test, the escape latencies assessed the ability of mice to acquire and recall spatial information. In the training trial, the mean escape latency was recorded from four quadrants in successive 5 days. The results showed a significant difference between APP/PS1 mice and their WT littermates (Figure 4A, p < 0.01), and the differences among GP mice and transgenic mice were also significant (Figure 4A, p < 0.05) from the second day to the fifth day. On the final training day, Aricept- or PNS- or NP-treated mice showed significant differences compared with transgenic mice (Figure 4A, p < 0.01, p < 0.05, p < 0.01). In the testing trial, removal of the platform, APP/PS1 mice showed a significantly longer time to reach the platform area (Figure 4B, p < 0.01). Only NP significantly shortened the latency or the percentage of time on the target quadrant compared with APP/PS1 mice (Figures 4B,C, p < 0.05 C), while all the intervention groups significantly shortened the duration at the opposite quadrant (Figure 4D).

The passive avoidance test was also applied to further confirm the fear memory. Compared with WT mice, APP/PS1 mice showed a significantly shorter step-through time. NP, Aricept, and GP significantly increased the duration compared with the APP/PS1 group (Figure 4E). In addition, APP/PS1 mice significantly increased the number of passing errors, while GP and NP significantly reduced the error number (Figure 4F). Together, NP can significantly improve spatial learning and memory and fear memory in the 7-month-old APP/PS1 mice. The spatial learning seemed partially contributed by both PNS and GP, while the fear memory was mainly contributed by GP.