Twenty 11–12 months 3 × Tg AD female mice (No: 14002A) were randomly divided in 2 groups: the 3 × Tg AD model group and the BBR group (50 mg∙kg⁻¹∙d⁻¹). Ten age-matched wild-type female C57 mice were used as controls. All animals were purchased from Beijing HFK Bioscience Co., Ltd. All animals were caged in SPF barrier-protected facilities with a temperature of (22 ± 3) °C, relative humidity of (50 ± 10) % and automatic light cycles (12 h light/dark). Food and water were freely available. This experiment started after 1 week of adaptive feeding. All experiments were carried out according to animal care guidelines and were approved by the Ethics Committee of the Xiyuan Hospital of the China Academy of Chinese Medical Sciences (No. 2022XLC020-2). Each mouse was fed 0.1 ml/10 g body weight by gavage for 3 months, and the BBR group received the corresponding drug, while the 3 × Tg AD model group and the control group received equal amounts of distilled water. The final numbers of animals in the control group, model group and BBR group were 9, 9 and 7, respectively due to mishandling.

Berberine at ≥98% purity (LDSW220209-1) was supplied by Shaanxi Langde Biotechnology Co., Ltd., and the HE staining kit was purchased from Servicebio (G1003).

The open field test (OFT) was performed by placing the mice in a box (40 cm L × 40 cm W × 40 cm H). The test area in the box was divided into 16 squares (10 × 10 cm) by the computer software and the four innermost squares were defined as the central area. During 5 min OFT test, total distance (cm), total movement (s), speed (cm / s), distance in the center (cm), and time in the center (s) were calculated.

Twenty-four hours after OFT test, a novel object recognition test (NOR) was performed. This test consisted of an exposure phase to a familiar object, followed by a phase of exposure to a novel object the next day. Each phase lasted 10 min. During the exposure phase to a familiar object, mice were placed in the same box with two identical objects in two parallel corners. 24 h later, one of the two objects was replaced by a novel one. The preference index (PI) and the recognition index (RI) were recorded. PI was defined as the time spent exploring two objects during the exposure phase to a familiar object, while RI was calculated as follows: T_novel/(T_novel + T_familiar). Notes: T_novel and T_familiar indicate the time spent with the novel and familiar objects during the exposure phase to the novel object.

Next, the Morris water maze (MWM) test was used to assess spatial memory of mice. The facility consists of a white tank (diameter 120 cm, height 50 cm), an automatic camera, and a computer analysis system. The pool was divided into four quadrants with specific markers, respectively, and a cylindrical escape platform (diameter 10 cm, height 15 cm) was placed in the first quadrant 1 cm beneath the water filled with nontoxic white dye. For training, spatial navigation experiments were performed for five consecutive days; each mouse was trained four trials per day for 90 s each using different entry points. The mice were placed in the water facing the wall of the pool, and the time between entering the water and finding the escape platform (with all limbs on the platform for 6 s; escape latency) was recorded using a video tracking system. On day 6, the escape platform was removed for spatial exploration experiments; mice were placed in the water facing the wall of the pool (at the midpoint of the third quadrant), and the number of times the mouse crossed the original platform area within 90 s was recorded, the time in the target quadrant and distance moved in the target quadrant were also recorded to examine memory of the target quadrant where the original platform was located. A quiet environment with a stable light source and a water temperature of 23 ± 1°C were maintained during experiments.

After the behavioral experiments, the brains were quickly removed. For each group, half of the brains were fixed in 4% paraformaldehyde, and the other half were snap frozen in liquid nitrogen and stored at −80° C for further analysis.

Brain tissues were embedded in paraffin and sectioned. Then sections were processed as follows: Xylene for 20 min, two times; 100% ethanol for 5 min, two times; 75% ethanol for 5 min; and tap water rinsing. Sections were subjected to hematoxylin–eosin (HE) and toluidine blue (Nissl) staining. Hematoxylin–Eosin Staining Kit (G1003; Wuhan Servicebio Technology Co., Ltd., Wuhan, China) was used for HE staining, and Toluidine blue staining solution (G1032; Wuhan Servicebio Technology Co., Ltd.) was used for Nissl staining. Sections were finally sealed with neutral gum and examined under an upright optical microscope (NIKON ECLIPSE E100) for image acquisition.

MOL001454 (CAS, 2086-83-1) was found by searching for the chemical name “berberine” in the traditional Chinese medicine systems pharmacology database (TCMSP),¹ and absorption, distribution, metabolism, and excretion (ADME) parameters for BBR were also obtained. The molecular structure of BBR was downloaded from the PubChem database² and dealed with PyMOL2.4.0 software.

Potential BBR targets were identified and collected using the PharmMapper platform,³ which is an integrated pharmacophore matching platform including targets extracted from TargetBank, DrugBank, BindingDB, and PDTD database and over 7,000 receptor-based pharmacophore models (Liu et al., 2010; Wang et al., 2017). All predicted targets were imported in EXCEL to establish the BBR target database.

The comparative toxicogenomics database (CTD),⁴ DisGeNet⁵ (Piñero et al., 2015, 2017, 2020, 2021) and the human gene database (GeneCards)⁶ were used to identify AD targets with the keyword ‘Alzheimer’s disease’. Then, BBR-associated targets of AD were generated by intersecting BBR targets with the above targets of Alzheimer’s disease.

BBR-associated targets of AD were imported into STRING,⁷ and protein–protein interaction (PPI) data were exported under the condition that the minimum required interaction score was 0.700. We then built a network diagram using the Cytoscape 3.7.2 software. In addition, the MCODE and cytoHubba plug-ins were used to screen important PPI network modules.

The biological process (BP), molecular function (MF), and cellular component (CC) are three important parts of gene ontology (GO). GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis were used for functional enrichment analysis. The R 3.6.3 software was used for statistical analysis and visualization; the ggplot2 package 3.3.3 and the clusterProfiler package 3.14 were used. Significant terms were identified under the condition that false discovery rate (FDR) was <0.05. The top terms identified were visualized in a diagram.

The molecular structure of BBR was downloaded from TCMSP, while the crystal structures of target proteins were obtained from the Protein Data Bank (PDB) database.⁸ PyMOL 2.4.0 software,⁹ Auto Dock Tools 1.5.7 software were used for pre-docking molecular processing to obtain PDBQT file. The Auto Dock Vina software 1.1.2¹⁰ was then used for molecular docking and calculation of the affinity score. A lower affinity score indicates stronger binding. The results of molecular docking were visualized by PyMOL2.4.0 software. We calculated the RMSD (Root Mean Square Deviation) using PyMOL2.4.0 to verify the reliability, and RMSD <2A was considered reliable.

An RNA extraction kit (Servicebio, G3640-50 T) was used to extract total RNA from brain tissue. ReverTra Ace qPCR RT Master Mix (TOYOBO, FSQ-201) was used for reverse transcription to generate cDNA templates and then real-time PCR was performed under the following conditions: 95° C for 10 min, followed by 40 cycles of 95 ° C for 15 s and 60 ° C for 60 s. Real-time PCR was performed using Applied Biosystem 7,500 Real-Time PCR System, and the relative expression of mRNA was calculated using the 2^−ΔΔCt method. The primer sequences of all target genes are shown in Table 1.

Brain tissues were lysed on ice in lysis buffer and then centrifuged at 12,000 rpm for 20 min at 4°C. The protein content was determined using the bicinchoninic acid (BCA) method (G2026, Servicebio, Wuhan, China). The extracted proteins were subjected to SDS-PAGE electrophoresis, transferred to 0.45 μm polyvinylidene difluoride membranes (Millipore, Bedford, MA), and then blocked with 5% BSA in TBST for 60 min. The membranes were subsequently incubated with primary antibodie. The primary antibodies were listed as follows: Erk 1/2 (diluted 1:2000, rabbit, Cell Signaling Technology, Cat# 4695), p-Erk 1/2 (diluted 1:2000, rabbit, Cell Signaling Technology, Cat# 4370), Akt (diluted 1:5000, rabbit, proteintech, Cat No. 10176-2-AP), p-Akt (diluted 1:5000, Mouse, proteintech, Cat No: 66444-1-Ig), or β–actin (diluted 1:10000, Mouse, proteintech, Cat No: 66009-1-Ig) overnight at 4°C. The membrane was incubated with the HRP conjugated secondary antibodies, goat anti-mouse (diluted 1:10000, Abbkine, China) or goat anti-rabbit IgG (diluted 1:10000, Abbkine, China) for 90 min after the membrane had been washed with TBST 4 times for 10 min each. The blots of proteins of interest were visualized using sensitive ECL western HRP substrate (# 17046, ZENBIO, Chengdu, China). Quantitative analysis of protein bands were performed using ImageJ software.

Data were statistically analyzed using SPSS 22.0 software (IBM Corp., Armonk, NY, USA). Quantitative data were expressed as mean with standard error of mean (SEM). Comparison between two groups were performed using t test. The MWM latency data were analyzed using a repeated-measures analysis of variance. All tests were two-sided with an α of 0.05 and statistical significance threshold of p < 0.05.

The pharmacological and molecular properties of BBR were obtained from TCMSP (Ru et al., 2014) and are as follows: molecular weight (MW) = 336.39, oral bioavailability (OB) = 36.86%, drug-likeness (DL) = 0.78, blood brain barrier (BBB) = 0.57, half-life (HL) = 6.57. BBR half-life (t_1/2) is in the mid-elimination group. The molecular structure of BBR (C₂₀H₁₈NO₄) is shown in Figure 2. BBR is believed to be more druggable in the 180–500 Dalton MW range. It is a promising therapeutic agent that can act in the central nervous system and is characterized by high OR (≥20%) (Xu et al., 2012), high DL (≥0.18) (Tao et al., 2013), and strong penetration (BBB > 0.3)(Tattersall et al., 1975).

The sequence of behavioral experiments is shown in Figure 3A. During the OFT test, the BBR group differed significantly from the model group in speed (t = 3.59, p = 0.003), total move time (t = 3.07, p = 0.008), time in the center (t = 2.27, p = 0.039), total distance (t = 3.59, p = 0.003), and distance in the center (t = 2.46, p = 0.0273), indicating differences in overall activity or anxiety-like behavior (Figures 3B–F). During the novel object recognition test, neither the control group nor the BBR group differed significantly from the model group in the preference index (PI) (p > 0.05, Figure 3G). After 3 months of BBR administration, mice in the BBR group spent significantly more time with the novel object in the NOR test as indicated by a significantly higher recognition index (t = −2.47, p = 0.028) in the NOR test (Figure 3H).

The MWM test consists of a navigation test and an exploratory experiment, which can be used to evaluate the learning memory ability of mice. In the navigation test, as shown in Figure 3I, the escape latency of all mice gradually decreased with the increasing number of training sessions, indicating that their ability to locate the platform was enhanced. The escape latency of mice in the BBR group was diminished compared to the model group and exhibited a significant difference on day 1 (p = 0.003), day 2 (p = 0.011) and day 5 (p = 0.007). There were no significant differences in the speed of swimming among the groups (p > 0.05, Figure 3J), and effects on locomotor ability could be excluded. In addition, compared to the model group, mice in the BBR group had a significantly higher number of platform crossings in the exploratory experiment on day 6 (p = 0.039, Figure 3K).

These results suggest that BBR alleviated the tense, manic, and anxious behaviors of AD mice and improves recognition and memory performance, although more studies are necessary to clarify the mechanisms involved. See details in Supplementary data S1.

The size, density and arrangement of neurons in the hippocampus may indicate neuronal damage. Using HE staining, chromatin in the nucleus and ribosomes in the cytoplasm were stained blue-violet, while components in the cytoplasm and extracellular matrix were stained red, and the degree of staining may reflect the functional properties of hippocampal neurons. Under normal conditions, the cytoplasm is only slightly stained, but it becomes overstained when cellular senescence or neurodegenerative alterations occur. The hippocampal neurons of control mice were dense and neatly organized, with full nuclei and clear boundaries, whereas the neurons of model mice were overstained and loosely arranged, with cytoplasmic deformation and swelling. In the model mice, BBR treatment was able to reverse this phenotype and the neuronal morphology resembled that of controls (Figure 4A).

Neurons in the model group showed relatively lighter blue staining, with a reduced number of Nissl bodies. Besides, in the BBR groups, there was stronger cytoplasmic staining and neurons were densely arranged (Figure 4B). Quantitatively, in the CA1 region of the hippocampus, neuron numbers were significantly higher in the BBR (t = −4.07, p = 0.015) group compared to the model group (Figure 4C), suggesting that BBR can protect hippocampal neurons to some extent and prevent degenerative necrosis.

Using the keyword ‘Alzheimer’s disease’, 215 potential BBR targets were obtained from the PM database, 24,282 AD targets from the CTD database, 3,397 AD targets from the DisGeNet database and 11,301 AD targets from GeneCards. 117 BBR-associated AD targets were generated from the intersection of these three databases using a Venn plot (Figure 5A). The names of these targets are shown in Figure 5B. See detailed gene information in Supplementary data S2.

Proteins interact with each other to participate in various aspects of life processes such as biological signaling, regulation of gene expression, energy and material metabolism, and cell cycle regulation. By detecting the interrelationship between target proteins, the possible pathways of action of drugs can be predicted and provide directions for in-depth research. In this study, PPI analysis between BBR-associated AD targets was performed to explore underlying mechanisms using STRING and the results are shown in Supplementary data S3. The PPI network is visualized in Figure 6A. The results of MCODE (Figure 6B) and cytoHubba are shown in Supplementary data S4. Combining the data obtained from the above algorithms, we obtained 43 genes for downstream functional enrichment analysis. In total, 2,230 BP terms, 67 CC terms, 243 MF terms (p < 0.01 and FDR<0.01) and 118 KEGG terms (p < 0.05 and FDR<0.01) (Supplementary data S5) were obtained. The top terms for each category are shown in Figure 6C: regulation of cell death, cell death, regulation of programmed cell death, cellular response to chemical stimulus, programmed cell death, regulation of apoptotic process, apoptotic process, cellular response to organic substance, response to oxygen-containing compound, response to chemical in BP enrichment analysis Furthermore, Pathways in cancer, Proteoglycans in cancer, PI3K-AKT signaling pathway, MAPK signaling pathway, Prostate cancer, EGFR tyrosine kinase inhibitor resistance, Endocrine resistance, AGE-RAGE signaling pathway in diabetic complications, Estrogen signaling pathway, PD-L1 expression and PD-1 checkpoint pathway in cancer involved in BBR treatment for AD based on KEGG pathway analysis (Figure 6D). Importantly, ALB, EGFR, CASP3 and five members of the PI3K-Akt signaling pathway including AKT1, HSP90AA1, SRC, HRA, and IGF1 were identified as core BBR targets in AD for further study.

To validate the core targets of BBR in AD, semi-flexible molecular docking was performed using Auto Dock Vina software. A lower affinity score indicates a stronger binding force. The crystal structures of the target proteins are listed below: AKT1 (PDB: 4ejn), HSP90AA1 (PDB: 3t0h), SRC (PDB: 1fmk), HRAS (PDB: 2ce2), IGF1 (PDB: 1wqj), ALB (PDB: 6hn1), CASP3 (PDB: 2dko), EGFR (PDB: 8a27). After molecular docking and interaction analysis, BBR was found to bind in a stable conformation to its targets as indicated by the low affinity scores: −10.2 kcal/mol (AKT1), −6.5 kcal/mol (HSP90AA1), −9.6 kcal/mol (SRC), −5.2 kcal/mol (HRAS), −7.8 kcal/mol (IGF1), −8.4 kcal/mol (ALB), −6.6 kcal/mol (CASP3) and − 9.5 kcal/mol (EGFR). All RMSD values were less than 2A, suggesting the molecular docking results are reliable (Table 2). Docking of core targets and BBR is shown in Figure 7.

Relative expression changes of the core targets were detected by RT-qPCR (Figure 8A). The level of Akt1 mRNA was significantly reduced in AD mice (t = 2.96, p = 0.025). A similar trend was observed for Hsp90aa1, Hras, Igf1, and Src mRNA levels, but none of them was statistically significant (p > 0.05). Among the core targets, Akt1 (t = −5.01, p = 0.002), Hsp90aa1 (t = −3.66, p = 0.011), Hras (t = −2.99, p = 0.024) and Igf1 (t = 3.75, p = 0.019) mRNA levels were significantly increased after BBR treatment, while Src, Egfr mRNA levels showed a similar trend but lacked statistical significance (p > 0.05). Alb, Casp3 mRNA levels showed the opposite trend. These results suggest that Akt1, Hsp90aa1, Hras, Igf1 could play important roles in BBR treatment of AD.

Then, we explored whether BBR affects protein phosphorylation of AKT as well as ERK by western blotting at the protein assay level. As shown in Figures 8B,C, AKT (t = 4.489, p = 0.004) and ERK (t = 5.645, p = 0.001) phosphorylation decreased in the model group, and BBR significantly increased AKT (t = −3.721, p = 0.010) and ERK (t’ = −4.126, p = 0.014) phosphorylation levels.