In this study, we utilized 4 plant-derived compounds: catalpol (CAS: 2415-24-9), puerarin (CAS: 3681-99-0), gastrodin (CAS: 62499-27-8) and borneol (CAS: 124-76-5). Catalpol, puerarin, and gastrodin were sourced from Dierg Pharmaceutical Technology Co., Ltd., Nanjing, China, while borneol was obtained from Tongrentang, Chongqing, China. Supplementary Figure S1 presents the chemical structure diagrams of these compounds. In our previous research, we discovered that a combination of catalpol, puerarin, and borneol in a 1:4:2 ratio maximized brain absorption and exhibited neuroprotection effects in an ischemic model (Qiang et al., 2016; Tang et al., 2016). Building on these findings, we incorporated gastrodin into the formulation, using a 1:4:2:2 ratio for CPGB, which yielded significantly improved results in AD models compared to other ratios (data not shown). For this optimized ratio, we prepared 15 formulations: catalpol alone (C), puerarin alone (P), gastrodin (G), borneol (B), catalpol + puerarin (CP), catalpol + gastrodin (CG), catalpol +borneol (CB), puerarin +gastrodin (PG), puerarin + borneol (PB), gastrodin + borneol (GB), catalpol + puerarin + gastrodin (CPG), puerarin + gastrodin + borneol (PGB), catalpol + gastrodin + borneol (CGB), catalpol + puerarin + borneol (CPB), catalpol + puerarin + gastrodin + borneol (CPGB). These formulations were prepared using a 35% hydroxypropyl-beta-cyclodextrin solution (Merck, Kenilworth, NJ, United States), which was commonly used to solubilize poorly soluble drugs due to its ability to form inclusion complexes with hydrophobic molecules (Wang and Brusseau, 1993). Specifically, 800 mg of puerarin and 400 mg of borneol were dissolved in 50 mL of the solution, followed by the addition of 200 mg of catalpol and 400 mg of gastrodin to create the high-dose CPGB solution (360 mg/kg). The solution was subsequently diluted to obtain mild (180 mg/kg) and low (90 mg/kg) doses. All these formulations were freshly prepared and used immediately, and no changes in the physical appearance of the solution were observed during the preparation or usage process. As a result, a stability study on these formulations was not conducted in this study. Additionally, we used donepezil hydrochloride tablets (DPN; Zein Bio, Nanjing, China) as a positive control to evaluate the therapeutic effects of these formulations.

The pheochromocytoma cell line (PC12) and the BV2 microglial cell line (BV2) used in this study were kindly provided by Prof. Hongyi Qi from the College of Pharmacy, Southwest University, Chongqing, China. These cell lines were cultured in a complete medium composed of 89% Dulbecco’s Modified Eagle Medium: Nutrient Mixture F-12 (DMEM/F12; Gibco, Waltham, MA, United States), 10% fetal bovine serum (FBS; Gibco, Waltham, MA, United States), and 1% penicillin-streptomycin (PS; Gibco, Waltham, MA, United States) maintained at 37%°C in an incubator with 5% CO₂. Once the cells reached 90% confluence, they were detached, collected, and resuspended in a complete medium for subsequent experiments.

Referencing previous studies (Gupta et al., 2018; Kamat, 2015; Li et al., 2020), we established an AD cell model by treating PC12 and BV2 cells with 0.5 mol/mL STZ. Briefly, PC12/BV2 cells were evenly seeded into 6-well plates (ThermoFisher, Shanghai, China) at a density of 2 × 10⁵ cells each well. After 24 h of proliferation, the supernatant was discarded, and a fresh complete medium containing either no additives, 0.5 mol/mL streptozotocin (STZ; Solarbio, Beijing, China) or 0.5 mol/mL STZ plus various formulations was added for further culture. Each treatment was performed with 3 replicates. Following cultivation, the supernatant was collected for subsequent enzyme-linked immunosorbent assay (ELISA) experiments, while cells from each well were collected for total RNA or total protein extraction.

Alternatively, PC12/BV2 cells were seeded onto sterile coverslips in 24-well plates (ThermoFisher, Shanghai, China) at a density of 5 × 10⁴ cells/mL, with 1 mL of cell suspension added to each well. After 24 h of proliferation, the supernatant was discarded, and a fresh complete medium containing either no additives, 0.5 mol/L STZ, or 0.5 mol/L STZ plus various formulations was added for further culture. Each treatment was performed with 3 replicates. After cultivation, the coverslips were collected for subsequent immunofluorescence analyses.

To establish the 3D-NVU model, hippocampal neural stem cells (NSCs) and brain microvascular endothelial cells (BMECS) were harvested according to previous established methods (Liu et al., 2013; Wang et al., 2021). Briefly, NSCs were isolated from the hippocampi of 24-h-old SD rat pups. After anesthesia and decapitation, the brains were quickly removed under sterile conditions and placed in cold D-Hanks solution (Solarbio, Beijing, China). The hippocampal region was dissected, minced, and centrifuged at 800 rpm for 5 min. The resulting tissue was digested with accutase (ThermoFisher, Shanghai, China) for 5 min at 37°C. The cell suspension was filtered and centrifuged. The pellet was resuspended in DMEM/F12 medium containing 20 ng/mL epidermal growth factor (EGF), 10 ng/mL basic fibroblast growth factor (bFGF), 2% B27, and 0.0002% heparin, then cultured at 37°C with 5% CO2. Cells were passaged or used for experiments after 6-7 days. BMECs were isolated from the cortex of 10-day-old SD rats. After decapitation and removal of the brain, the cortex was dissected, and large blood vessels and meninges were removed. The tissue was homogenized, followed by centrifugation at 1,200 rpm for 5 min. The homogenate was treated with 25% BSA and centrifuged at 2,500 rpm to collect the microvascular fragments. The fragments were digested with 0.1% type II collagenase for 5 min, then centrifuged and resuspended in complete medium with 20% FBS. Cells were seeded into collagen-coated flasks and cultured at 37°C with 5% CO2. Media were changed every 2-3 days, and cells were used when 90% confluence was reached.

NSCs and BMECs were mixed with Matrigel (Corning, Corning, NY, United States) at a 1:10 density ratio. The mixture was resuspended, and 150 μL of the cell/Matrigel mixture was evenly distributed at the bottom of a laser confocal dish and incubated to solidify. After 20 min of incubation, an additional 50 μL of the cell/Matrigel mixture was spread over the surface of the solidified layer and returned to the incubator for another 20 min before adding the medium. The NSC/BMEC co-culture medium consisted of DMEM/F12 and Neurobasal (ThermoFisher, Shanghai, China) in a 1:1 ratio, supplemented with 2% B27, 20 ng/mL recombinant human EGF (rhEGF), and 10 mg/mL recombinant human basic fibroblast growth factor (rhbFGF). On the 3rd day, the B27 supplement was switched to a formulation containing vitamin A. After 7 days of culture, the 3D-NVU model was successfully replicated, as confirmed by the positive expression of β-tubulin3, glial fibrillary acidic protein (GFAP), and CD31 marker proteins (Wang et al., 2021). After confirming the successful establishment of the 3D-NVU model, a medium containing either no additives, 0.5 mol/L STZ, 0.5 mol/L STZ plus various formulations, or 0.5 mol/L STZ + various formulations + inhibitors were added for further culture. Each treatment was performed with 3 replicates. Following cultivation, the supernatant was collected for subsequent ELISA experiments, while cells from each well were collected for total RNA or total protein extraction.

Referencing a previous study (Qureshi et al., 2022), a conditioned medium of microglia was employed to induce cellular damage. Briefly, BV2 cells were evenly seeded into 6-well plates at a density of 2 × 10⁵ cells per well. After 24 h of culture, the medium in each well was replaced with a fresh complete medium containing either no additives, 0.5 mol/L STZ, or 0.5 mol/L STZ plus various formulations. Following another 24 h of culture, the medium was discarded, and the cells were washed 3 times. Subsequently, 1 mL of completing medium was added to each well for an additional 24 h of culture. The medium in each well was then collected as a conditioned medium. Additionally, PC12 cells were seeded into 96-well plates at the density of 2 × 10³ per well. After 24 h of culture, the medium was discarded, and the collected conditioned medium from BV2 cells was added to the wells for an additional 24 h of culture. The viability of the PC12 cells was then assessed using the Cell Count Kit-8 (CCK8) assay. Briefly, the supernatant was discarded, and 100 μL of fresh complete medium containing 10% CCK-8 solution (Solarbio, Beijing, China) was added. After 2 h of incubation, the optical density at 450 nm (OD450) was measured using a microplate reader (ThermoFisher, Shanghai, China).

The rats used in this study were male Sprague-Dawley rats of specific pathogen-free grade, aged 3–4 months with weights ranging from 350 to 380 g. They were purchased from Liaoning Changsheng Biotechnology Co., Ltd. (Shenyang, China). The rats were maintained at the Experimental Animal Center of the College of Pharmacy, Southwest University. They were housed in standard cages at 22°C ± 2°C, on a 12-h light/dark cycle, with a humidity level of 40%–60%. Throughout the experimental period, all rats had free access to water and food, and the bedding was refreshed every 2 days.

Following a previous study (Moreira-Silva et al., 2018), an AD rat model was established through intracerebroventricular (ICV) injection of STZ. Briefly, after 1 week of adaptive feeding and a subsequent 12-h fasting period, the rats were anesthetized with 5% isoflurane. Their heads were then secured in a stereotaxic apparatus, and anesthesia was maintained with 2% isoflurane. The fur on the scalp was trimmed, and the skin was disinfected with iodine and alcohol. A midline incision approximately 1.5 cm in length was made along the sagittal suture to expose the skull. The periosteum was washed with 3% hydrogen peroxide (Solarbio, Beijing, China) and pushed aside to fully expose the cranial bone. Bilateral burr holes were drilled based on coordinates referenced from the bregma: −0.8 mm anteroposterior, ± 1.4 mm mediolateral, and −3.6 mm dorsoventral, according to the rat brain atlas by Paxinos and Watson (Paxinos and Watson, 2006). Small burr holes were carefully drilled using a dental drill, thinning the skull without penetrating it. A microsyringe (Hamilton, Reno, NV, United States) was then carefully inserted vertically to a depth of 3.8 mm. Afterward, both STZ and arachidonoylethanolamide (AEA; Sigma-Aldrich, St. Louis, MO, United States) were prepared in citrate buffer (Merck, Kenilworth, NJ, United States) and administered using an automated microinjection system (Insight Equipment LTDA, São Paulo, Brazil). This system included a polyethylene microtube connected to an injection needle and the inserted microsyringe coupled with an infusion pump. Each rat received 2 ICV bilateral injections of 2 μL at a rate of 1 μL/min. The first injection consisted of either citrate buffer (for sham procedure; 0.05 M) or AEA (100 ng). Five minutes later, a second injection of either citrate buffer or STZ (2 mg/kg) was administered. To prevent reflux, the injection needle was left in place for at least 2 min following each infusion. Post-surgery, the rats received an intraperitoneal injection of ceftriaxone sodium (100 mg/kg; Merck, Kenilworth, NJ, United States) once daily for 3 days to prevent infection, along with analgesic carprofen (5 mg/kg; Merck, Kenilworth, NJ, United States). Additionally, 2 mL of saline was administered subcutaneously for rehydration. The rats were then kept on thermal blankets with controlled temperature until full recovery.

On the day following surgery, rats in both the sham and model groups were orally administered a 35% hydroxypropyl β-cyclodextrin solution. Meanwhile, those in the treatment groups received DPN and CPGB at doses of 90, 180, and 360 mg/kg for low, medium, and high doses, respectively, via gavage at a rate of 1 mL/kg. Each group consisted of 16 rats. The treatment was administered daily for 42 consecutive days. The doses of CPGB were determined referencing to our previous research (Tang et al., 2016). An earlier acute toxicity study confirmed that these doses did not produce any observable toxic effects on the major organs of the rats. In summary, rats administered 5.12 g/kg of CPGB exhibited no abnormal reactions compared to normal rats. In contrast, rats given 6.4 g/kg of CPGB exhibited mild symptoms, such as reduced activity and partial paralysis, which resolved within 24 h. Furthermore, a 14-days treatment with CPGB at doses of 5.12 g/kg or lower did not result in any signs of toxicity, mortality, or significant changes in behavior, body weight, or organ function. Behavioral assessments were performed using the Novel Objective Recognition (NOR) test from days 31–33, the Elevated Plus Maze (EPM) test from days 34–35, and the Morris Water Maze (MWM) test from days 46 to 36. Upon completing the MWM test on the 42nd day, 3 rats from each group were euthanized to collect hippocampal electrical signals. The remaining rats were anesthetized and euthanized by cervical dislocation, after which their brains were collected for further experiments.

To evaluate the hippocampus-dependent non-emotional memory in rats, the NOR test was conducted from days 31–33, as referenced in a previous study (Leger et al., 2013). The experimental setup consisted of a cylindrical enclosure made of black Plexiglas plastic, measuring 50 cm in both diameter and height. To minimize anxiety due to unfamiliarity, rats were allowed to acclimate to the environment by being placed in the apparatus for 5 min, 24 h before the test. During the familiarization phase, 2 identical objects, labeled A1 and A2, were positioned at opposite corners of the enclosure, each 5 cm from the corner. Each rat was gently placed in the enclosure and allowed to explore the objects for 5 min. After a 24-h interval, object A2 was replaced with a new object for the test session, and the rat was reintroduced to the enclosure to explore the objects for another 5 min. Exploratory behavior was defined as orienting the nose towards an object at a distance of less than 2 cm or making physical contact with the object using the nose. The exploration time for each object during both the familiarization and test sessions was recorded. The Discrimination Index (DI) was calculated using the following formula: DI = Time spent in exploring the novel object/(Time spent in exploring old object + Time spent in exploring the novel object).

To investigate anxiety-related behavior in rats, the EMP test was conducted, referencing a previous study (Walf and Frye, 2007). The maze consisted of 1 open arm and 1 closed arm. On the 34th after surgery, each rat was placed facing outwards on the open arm for free exploration. The time taken for the rat transition from the open arm to its initial entry into the closed arm was recorded by a camera and noted as the initial escape latency (EL1). If this time exceeded 90 s, it was recorded as 90 s. On the 35th day, the same test method was used to obtain the ultimate escape latency (EL2). The relative escape latency (REL) was calculated using the following formula: REL = EL2/EL1 × 100%.

The spatial learning and memory abilities of rats were assessed using the Morris Water Maze (MWM) experiment (Vorhees and Williams, 2006). The procedure involved a series of acquisition trials conducted over 5 consecutive days. Each day, the rats underwent 4 trials with a minimum inter-trial interval of 15 s. In each trial, rats were released from a randomly selected starting point and given up to 120 s to locate the platform. Upon finding the platform, they were allowed to remain on it for 5 s. If a rat failed to find the platform independently, it was manually guided to the platform and held there for 15 s. The latency times for all trials were recorded. On the 6th day, a probe trial was conducted to evaluate spatial memory without the platform. During this trial, the frequency of platform crossings, as well as the duration and distance traveled within the target quadrant, were measured. The EthoVisionXT 8.5 software (Noldus, Wageningen, Netherlands) was used to analyze the frequency of each rat passing through the location of the original platform, the swimming distance within the quadrant, and the time spent in the platform area. More crossings and longer routes indicated better learning and memory retention, as did longer durations spent in the quadrant containing the original platform.

After the MWM test, 3 randomly selected rats were anesthesia with 5% isoflurane, and their heads were secured on a brain stereotaxic instrument (Insight Equipment LTDA, São Paulo, Brazil). Recordings were made using a NeurLynx 32-channel electrophysiological recorder. An 8-channel recording electrode was implanted with coordinates relative to the brain Bregma point as the zero point: dorsoventral at −3.6 mm and lateral ventricle at 2.5 mm. Following a 30-min baseline recording, the spike peak potential was recorded for an additional 30-min.

ELISA was employed to determine the concentrations of Aβ1-42, TNF-α, IL-1β, and IL-6 in the hippocampal tissue homogenate or collected supernatant using commercialized ELISA kits (Nanjing Jiancheng, Nanjing, China) according to their instructions. In brief, 100 μL of supernatant was added to each well of the reaction plate, which was then sealed with a membrane and incubated at room temperature for 2 h. Subsequently, biotinylated antibody, horseradish peroxidase-labeled streptavidin, and tetramethylbenzidine solution were added to each well. The plate was incubated overnight at room temperature in the dark. Finally, 50 μL of stop solution was added to terminate the reaction, and the OD450 was measured using a microplate reader.

Total RNA from collected hippocampal tissue and cells was extracted using Trizol Reagent (TransGen, Beijing, China) following the manufacturer’s protocol. The concentration and purity of extracted RNA were measured using a NanoDrop ND-2000 spectrophotometer (Thermo Fisher Scientific, Wilmington, NC, United States), and its integrity was confirmed through 1% agarose gel electrophoresis. Following this, the total RNA samples were reverse transcribed to complementary DNA (cDNA) using the RevertAid First Strand cDNA Synthesis Kit (ThermoFisher, Shanghai, China) following the manufacturer’s instructions. RT-qPCR was then conducted under the following conditions: initial denaturation at 95°C for 30 s, followed by 40 cycles of denaturation at 95°C for 5 s, and annealing and extension at 60°C for 30 s. The primer sequences used in this study are listed in Supplementary Table S1, and the data was analyzed using the 2^ΔΔC method in Excel.

Proteins were extracted from collected hippocampal tissue and cells using RIPA lysis buffer (Solarbio, Beijing). The protein concentrations were determined with a BCA kit (Beyond, Shanghai, China). Subsequently, the proteins were denatured at 100°C for 10 min. And 30 μg of the extracted total protein from each sample were loaded onto a 12% sodium dodecyl sulfate-polyacrylamide for electrophoresis. After electrophoresis, the proteins were transferred onto a polyvinylidene fluoride membrane. The membrane was blocked with 5% skim milk in TBST for 2 h. Following blocking, the membranes were incubated overnight at 4°C with primary antibodies. On the following day, the membranes were treated with a secondary antibody and enhanced chemiluminescent reagents to visualize the protein bands. The antibodies used in this study are listed in Supplementary Table S2. The intensity of the protein bands was analyzed using ImageJ software.

For Golgi staining, a Golgi staining kit (Servicebio, Chengdu, China) was employed. Briefly, the brain tissues collected were first washed and fixed with 4% paraformaldehyde (Life iLab Bio, Shanghai, China) at −20°C for 24 h. Following fixation, the tissues were immersed in a dark A + B mixture at room temperature for 14 days, then soaked in C solution for an additional 5 days. The tissues were subsequently sectioned using a freezing microtome set to a thickness of 180 μm. These sections were stained with D solution and observed under a microscope for imaging.

The remaining freshly collected brain tissues were fixed in 4% paraformaldehyde for 48 h. After fixation, the tissues were rinsed several times with tap water, dehydrated through a gradient alcohol series, soaked in paraffin, and then embedded. For Haematoxylin and Eosin (HE) staining analysis, the paraffin sections (5 μm thick) were incubated at 80°C for 20 min, de-paraffinized, and then soaked in haematoxylin solution (Solarbio, Beijing, China) for 15 min. The sections were rinsed with tap water for re-bluing, stained in eosin, hydrated through a gradient alcohol series, cleared with xylene, and mounted with neutral resin before being air-dried, observed, and photographed under a microscope.

For Nissl staining, a Nissl staining kit (Beyotime, Shanghai, China) was employed. Briefly, brain sections were dewaxed, stained with methylene blue solution for 10 min, differentiated for 1 min, treated with ammonium molybdate solution for 5 min, and sealed using a neutral gum. The Nissl bodies were then visualized and imaged using a fluorescence microscope (Olympus, Tokyo, Japan) and quantified with ImageJ software.

For Immunohistochemical staining, the sections were dewaxed, washed 3 times, immersed with 3% H₂O₂ at room temperature for 10 min, and incubated with antigen retrieval solution at 37°C for 20 min. The sections were blocked with 5% bovine serum albumin (ThermoFisher, Shanghai, China) at room temperature for 2 h. Following blocking, sections were incubated with primary antibody at 4°C for 16 h, followed by 3 times washes. The sections were then incubated with a secondary antibody at 37°C for 30 min. DAB coloration was applied, and the nucleus was counterstained with hematoxylin. After dehydration, sections were sealed with neutral resin. The antibodies used in this study are listed in Supplementary Table S2. The Aβ-16 was visualized and imaged using a fluorescence microscope and quantified with ImageJ software.

The collected brain tissues and cell slices were dewaxed and fixed with 4% paraformaldehyde at −20°C for 10 min. Following fixation, the tissues and cells were permeabilized with 0.1% Triton X-100 (Solarbio, Beijing, China) at room temperature for 10 min. They were then blocked with 10 g/L bovine serum albumin (Solarbio, Beijing, China) for another 10 min and incubated overnight at 4°C with primary antibodies. After 3 washing, the secondary antibody was applied and incubated at room temperature for 4 h. The antibodies used in this study are listed in Supplementary Table S2. The cell nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI; Solarbio, Beijing, China) for 5 min at room temperature. Finally, the brain tissues and cells were observed and digitally captured using a fluorescence microscope.

Total RNA was extracted from the collected hippocampus tissues using Trizol Reagent, following the standard phenol/chloroform phase separation method. The RNA concentration and integrity were measured with a NanoDrop ND-2000 spectrophotometer and an Agilent Bioanalyzer 2100 (Agilent Technologies; Santa Clara, CA, United States). Only samples with an RNA integrity number (RIN) above 0.8 were selected for subsequent sequencing. Complementary DNA libraries were constructed using the Illumina TruSeq RNA Sample Prep Kit (Illumina; San Diego, CA, United States), resulting in an average fragment size of 150 bp, excluding adaptors. The quality and integrity of these libraries were assessed with an Agilent 2100 Bioanalyzer and an ABI StepOnePlus Real-Time PCR System (Thermo Fisher Scientific; Waltham, MA, United States). The RNA-Seq FASTQ files were aligned to the rat genome using the Hisat2 algorithm (Kim et al., 2019), referencing data from the Ensembl Rat Genome Database (https://ftp.ensembl.org/pub/release-112/fasta/rattus_norvegicus/). Cufflinks (Trapnell et al., 2012) were used to process the resulting binary alignment/map (BAM) files to evaluate transcript abundance and identify potential mRNA isoforms. Additionally, StringTie (v1.3.3b) (Pertea et al., 2015) was used to assemble the mapped reads from each sample using a reference-based approach. Transcript expression levels were quantified in terms of fragments per kilobase million (FPKM), and Principal Component Analysis (PCA) was conducted based on these FPKM values. The Kyoto Encyclopedia of Genes and Genomes (KEGG) and gene ontology (GO) enrichment analyses of differentially expressed genes (DEGs) were performed using DESeq2 software (Love et al., 2014) and the ClusterProfiler R package (Wu et al., 2021). The Benjamini-Hochberg method was applied to control the false discovery rate.

Unless specified otherwise, all data were expressed as mean ± standard error and analyzed statistically using SPSS and RStudio software. Initially, continuous variables were tested for normal distribution and variance homogeneity. If the data satisfied both conditions, an independent samples analysis of variance (ANOVA) was employed for preliminary hypothesis testing, followed by the LSD method for post hoc analysis. For data not meeting these criteria, the Kruskal-Wallis test was used, with the Tukey method for post hoc testing. For repeated-measures data (e.g., escape latency), a repeated measures ANOVA method was applied to assess differences across groups and time points. A P value less than 0.05 was considered indicative of statistical significance in all hypothesis tests. In transcriptomics analysis, the similarity of gene expression profiles among different groups was evaluated using the permutational multivariate analysis of variance (PERMANOVA) method. In differential expression analysis, genes with |Log2 (FoldChange)| ≥ 1 and p-value < 0.05 were deemed significantly differentially expressed genes (DEGs). In KEGG and GO enrichment analysis signaling pathways with p values < 0.05 were considered significantly enriched. Data plots were created using OriginPro software (v2024), while all schematic diagrams and microscope images were illustrated or enhanced using Adobe Illustrator and Adobe Photoshop unless otherwise noted.

To investigate the therapeutic effects of the 15 formulations on AD, we first measured concentrations of proinflammatory cytokines in the supernatant of PC12 cells treated with STZ and various formulations. The results indicated that STZ treatment significantly elevated the levels of TNF-α, IL-1β, IL-6, and Aβ1-42 (p < 0.001; Figure 1A; Supplementary Figures S2A–C) in the supernatant, as well as the expression of p-Tau231 (p < 0.001; Figure 1B; Supplementary Figure S2D) in PC12 cells. These increases were significantly mitigated by the positive drug DPN and 4 formulations: C, G, CPG, and CPGB (p < 0.01), while other treatments did not show significant effects (p > 0.05) (Figures 1A, B; Supplementary Figures S2A–D).

Next, we examined the therapeutic effects of these 4 formulations in an AD rat model. Rats in each group exhibited similar probe times and exploration speeds in the test phase of the NOR test and swimming speeds in the MWM test (p > 0.05; Supplementary Figures S2E–G). However, AD rats demonstrated significant reductions in exploratory behavior counts and discrimination index in the NOR test, as well as in the ratios of distance and duration of swimming in the target quadrant, and platform crossing counts in the MWM test (p < 0.001; Figures 1C, D, G–I). They also showed significant increases in relative escape latency in the EMP test and latency in the MWM test (p < 0.001; Figures 1E, F). These changes were significantly ameliorated by CPGB treatment (p < 0.05), but not by C or G treatments (p > 0.05; Figures 1C–I). Although CPG treatment exhibited some mitigating effects, its therapeutic impact was notably less than that of CPGB treatment (p < 0.05; Figures 1C–I). Hence, we chose CPGB as our formulation in the following experiments.

After confirming that CPGB exhibits greater therapeutic effects on AD rats, we further explored the impacts of low (90 mg/kg), mild (180 mg/kg), and high (360 mg/kg) doses of CPGB on the cognitive behaviors of AD rats. In the familiarization phase of the NOR test and MWM test, rats across all groups displayed similar exploratory behavior counts, probe times, and swimming speeds (p > 0.05; Supplementary Figures S1H–J). However, during the test phase of the NOR test, DPN and all 3 doses of CPGB significantly improved the STZ-induced reductions in exploratory behavior counts and discrimination index, independent of dose (p < 0.01; Figures 1J, K). STZ treatment significantly increased the relative escape latency and latency in EMP and MWM tests (p < 0.001; Figures 1L, M). Both DPN and CPGB treatments significantly reduced these increases (p < 0.01), with the mild dose of CPGB resulting in significantly lower relative escape latency and latency compared to the low or high doses (p < 0.05; Figures 1L, M). In the MWM test, DPN and CPGB treatments also significantly mitigated the STZ-induced reductions in the ratios of distance and duration of swimming in the target quadrant, as well as in platform crossing counts (p < 0.05; Figures 1N–P). Rats treated with the mild dose of CPGB showed significantly higher ratios of distance and duration of swimming in the target quadrant and platform crossing counts compared to those treated with low or high doses (p < 0.05; Figures 1N–P). Detailed trajectories of rats in the MWM test are visualized in Figure 1Q.

To further validate the therapeutic effects of CPGB on AD in rats, we evaluated the pathological changes in the lateral ventricle and hippocampus, as well as the electrophysiological dysfunction of hippocampal neurons. HE staining results showed that STZ treatment significantly enlarged the lateral ventricles in rats (p < 0.01). This enlargement was significantly reduced by treatments with mild and high doses of CPGB, as well as DPN (p < 0.05), but not by a low dose of CPGB (p > 0.05) (Figures 2A, B). Additionally, most hippocampal neurons in the Model group were degenerated, displaying blurred cell contours, ruptured cell bodies, and irregular arrangements compared to the Sham group. In contrast, neurons in rats treated with DPN or CPGB were intact and well-organized with clear outlines (Figure 2C). Nissl staining and immunofluorescence results indicated that the nuclear structure in the hippocampus of the Model group was unclear, with indistinct surrounding Nissl bodies. The number of neurons with intact membranes and NeuN⁺ cells in the hippocampus was significantly decreased in the Model group compared to the Sham group (p < 0.05). Rats treated with DPN and CPGB exhibited distinct Nissl bodies around the nucleus and a significantly increased number of neurons with intact membrane and NeuN⁺ cells compared to the Model group (p < 0.05) (Figures 2D–G). Golgi staining results demonstrated that the density, branch number, and total length of dendritic spines in hippocampal neurons were significantly lower in the Model group compared to the Sham group (p < 0.001). Treatments with DPN and CPGB significantly mitigated these decreases (p < 0.01), with a mild dose of CPGB exhibiting a better mitigative effect than the low and high doses (p < 0.05) (Figures 2H, I). Immunoblotting results showed that DPN and CPGB treatments significantly reversed the STZ-induced downregulation of Syp protein in hippocampal neurons (p < 0.05) (Figures 2J, K). Finally, the electrophysiological test identified a significantly decreased number of neurons with firing peak potentials in the hippocampal of the Model group (p < 0.01), which was significantly lower than in those treated with DPN and CPGB (p < 0.05) (Figure 2L).

In the meanwhile, we also investigated the effects of CPGB on AD-related markers in rats, including Aβ protein deposition, Tau protein hyperphosphorylation, hippocampal glial cell activation, and pro-inflammatory cytokine production. Immunohistochemistry results demonstrated a significantly higher level of Aβ1-16 in the Model group compared to the Sham group (p < 0.001). This increase was significantly inhibited by treatments with DPN and CPGB (p < 0.001), with the mild dose of CPGB resulting in a lower Aβ1-16 level compared to the low and high doses (p < 0.05) (Figures 3A, B). ELISA results indicated that treatments with DPN and CPGB significantly reversed the STZ-induced increases of Aβ1-42, TNF-α, IL-1β, and IL-6 concentrations in the Model group (p < 0.05; Figures 3C–F). Immunoblotting analysis revealed a significantly higher relative expression of amyloid precursor protein (App) and a higher pTau396 to Tau ratio (p < 0.01), both of which were significantly inhibited by CPGB treatments (p < 0.05) (Figures 3G–I). Finally, immunofluorescence analysis showed a significant increase in the number of microglia expressing ionized calcium-binding adaptor molecule 1 (Iba-1) and astrocytes expressing GFAP in the Model group compared to the Sham group (p < 0.001). These increases were also significantly ameliorated by treatments with DPN and CPGB (p < 0.001) (Figures 3J–L).

To uncover the underlying mechanisms of CPGB’s therapeutic effects on AD rats, we randomly collected hippocampus samples from the 3 groups of rats: Sham (n = 4), Model (n = 4), and Mild (n = 4) groups. Then, we analyzed their mRNA expression profiles using transcriptome sequencing. This process generated a total of 141.93 Gbp of raw bases and 946.16 million raw reads, averaging 11.83 Gbp raw bases and 78.85 million raw reads per sample (Supplementary Table S3). After quality control and assembly, we annotated and quantified 20057 genes across the 12 hippocampal samples (Supplementary Table S4). PCA and PERMANOVA demonstrated no significant differences in gene expression profiles among groups (p = 0.264; Figure 4A). In the model group, compared to the Sham group, we identified 75 down-regulated and 143 up-regulated DEGs (Figure 4B; Supplementary Table S5). Similarly, in the CPGB group, compared to the Model group, there were 46 down-regulated and 86 up-regulated DEGs (Figure 4C; Supplementary Table S5). We further identified 35 DEGs common to both comparisons, including Pkib, Hif3a, SNORD115, AABR07049688.1, Clec3a, LOC690082, AABR07036007.1, Ccr3, Ros1, Fndc9, AABR07053635.2, Wnt10b, Nod2, AABR07037487.1, U4, Tmem71, LOC100363125, Cdk1, Phex, PVR, AABR07047576.1, Ramp3, Cyp11b2, Spats1, Clec12b, AABR07043510.1, Aspg, AABR07019663.1, C7, Ccl5, Fam110d, Mcm3, AABR07028036.1, Vsig8, and Metazoa_SRP (Figure 4D). These genes were enriched in 40 GO terms, primarily related to immune-inflammatory response regulation (Supplementary Table S6), and in 10 KEGG signaling pathways, including Viral protein interaction with cytokine and cytokine receptor, TNF signaling pathway, Cell cycle, Cushing syndrome, Chemokine signaling pathway, Toll-like receptor signaling pathway, Viral carcinogenesis, Human cytomegalovirus infection, Cytokine-cytokine receptor interaction, and DNA replication (Supplementary Table S6; Figure 4E).

Given the elevated pro-inflammatory responses in the hippocampus and the critical roles of the TLR and NF-κB pathways in neural inflammatory regulation (Squillace and Salvemini, 2022), we hypothesized that CPGB exerts its therapeutic effects on AD in rats by modulating the TLR4/Myd88/NF-κB signaling pathway. To test this hypothesis, we measured the relative expression levels of the Tlr4, Myd88, and Nfkb1 genes compared to the housekeeping gene β-actin using the RT-qPCR method. The results showed significantly higher relative expression levels of Tlr4, Myd88, and Nfkb1 in the Model group compared to the Sham group (p < 0.01). These increases were significantly reduced by treatments with DPN and CPGB (p < 0.05) (Figures 4F–H). Similarly, immunoblotting analysis also revealed significant increases in the relative express levels of Tlr4 and Myd88 proteins, as well as an increased ratio of p-p65 to p65 (p < 0.05). These increases were also significantly mitigated by DPN and CPGB treatments (p < 0.05) (Figures 4I–L).

To confirm the involvement of the TLR4/Myd88/NF-κB signaling pathway in the mitigating effects of CPGB on AD, we assessed pro-inflammatory responses and the activity of this signaling pathway in BV2 cells treated with STZ and 90 μg/mL (Low), 180 μg/mL (Mild), and 360 μg/mL (High) doses of CPGB. It was observed that the STZ challenge significantly increased the concentrations of pro-inflammatory cytokines TNF-α, IL-1β, and IL-6 in the supernatant of BV2 cells (p < 0.05; Figures 5A–C), as well as the expression levels Tlr4, Myd88 (p < 0.01; Figures 5D–F), and Iba-1 proteins, and the ratio of p-p65 to p65 (p < 0.01; Figures 5G, H) in BV cells. All these increases were significantly reduced by treatments with DPN and CPGB (p < 0.05; Figures 5A–H).

Additionally, PC12 cells treated with conditioned medium from BV2 cells in the STZ group demonstrated significantly lower viability compared to those treated with conditioned medium from BV2 in the Control group (p < 0.05; Figure 5I). In contrast, PC12 cells treated with conditioned medium from BV2 in the DPN and CPGB groups demonstrated significantly higher viability compared to those treated with conditioned medium from BV2 cells in the STZ group (p < 0.05; Figure 5I).

We also developed a 3D-NVU AD model, representing the core structure of AD pathological damage (Hongjin et al., 2020), and treated this model STZ and varying doses of CPGB: 90 μg/mL (Low), 180 μg/mL (Mild), and 360 μg/mL (High). We then assessed indicators related to AD and inflammatory response. As shown in Figures 6A–D, STZ treatment significantly increases the concentrations of TNF-α, IL-1β, IL-6, and Aβ1-42 in the supernatant (p < 0.01). These increases were significantly inhibited by treatments with DPN and CPGB (p < 0.05). Immunoblotting analysis confirmed that STZ treatment significantly elevated the relative expression level of APP and the ratio of p-Tau396 to Tau (p < 0.01). These elevations were also significantly reduced by treatments with DPN and CPGB (p < 0.05; Figures 6E–G). Additionally, immunofluorescence analysis showed that DPN and CPGB treatments reduced the expression level of GFAP in the model (Figure 6H). As expected, STZ treatment significantly increased the relative expression levels of Tlr4 and Myd88 proteins and the ratio of p-p65 to p65 (p < 0.01). These increases were significantly mitigated by treatments with DPN and CPGB (p < 0.05; Figures 6I–L).

To confirm the necessity of the TLR4/Myd88/NF-κB signaling pathway, we interfered with this pathway by inhibiting the Tlr4 protein using Tak242 (MilloporeSigma, Burlington, MA, United States) and activating the NF-κB complex with phorbol 12-myristate 13-acetate (PMA; MilloporeSigma, Burlington, MA, United States) in the 3D-NVU model treated with STZ and 180 mg/mL of CPGB. As shown in Figures 6M–O, Tak242 treatment did not enhance the inhibitory effects of CPGB on the STZ-induced increases in the relative expression level of App or the ratio of pTau396 to Tau (p > 0.05). In contrast, PMA treatment significantly blocked these increases (p < 0.05). Similarly, Tak242 treatment did not enhance the inhibitory effects of CPGB on the STZ-induced increases in the concentrations of TNF-α, IL-1β, IL-6, and Aβ1-42 (p > 0.05), whereas PMA treatment significantly blocked these increases (p < 0.05) (Figures 6P–S).