Pathway-based network medicine identifies novel natural products for Alzheimer’s disease

Abstract

Background

Alzheimer’s disease (AD) is the leading cause of dementia, characterized by a complex pathogenesis that complicates the development of effective treatments. Natural products are promising multitarget agents because of their ability to interact with multiple molecular targets. Network-based medicine presents a robust strategy for discovering such agents, which can address the intricate mechanisms underlying AD.

Methods

In this study, we constructed an AD-related pathway-gene network via text mining and pathway database construction. This network facilitated the identification of natural products that target multiple pathways and genes associated with AD. We evaluated the safety profiles of two selected natural products in C57BL/6J mice through assessments of general behavior, body weight changes, vital organ weight and morphology, and hematological and biochemical parameters. APP/PS1 transgenic mice were subsequently treated with these natural products—either individually or in combination—to assess their therapeutic effects. Cognitive function was evaluated via behavioral tests, such as novel object recognition, Y-maze, and Morris water maze tests. Additionally, immunohistochemical staining and enzyme-linked immunosorbent assays were performed to examine Aβ-associated pathological changes. Transcriptomic analysis and quantitative real-time polymerase chain reaction (qRT-PCR) were employed to elucidate the mechanisms underlying the effects of the natural products.

Results

The constructed AD-related pathway-gene network encompassed three perspectives: (i) Most Studied Pathways (21 pathways with 5325 genes), (ii) Gene-Associated Pathways (26 pathways with 2557 genes), and (iii) Popular Pathways (24 pathways with 3435 genes). Two natural products, (-)-Vestitol and Salviolone, were selected for further validation. Their safety was confirmed in C57BL/6J mice. Notably, the combination of (-)-Vestitol and Salviolone synergistically affected cognitive function in APP/PS1 transgenic mice by reducing Aβ deposition and lowering toxic soluble Aβ levels in the brain. Transcriptomic analysis and qRT-PCR experiments revealed that their combination regulated AD-related pathways and genes more comprehensively, particularly affecting the Neuroactive ligand-receptor interaction and Calcium signaling pathway.

Conclusions

Our findings demonstrate that screening potential natural products through an AD-related pathway-gene network is a promising strategy for discovering novel therapeutics for AD. The therapeutic potential of (-)-Vestitol and Salviolone as novel candidates for AD treatment is underscored by their synergistic effects, attributed to their comprehensive regulation of AD-associated pathways and genes.

Background

Alzheimer’s disease (AD) is the most prevalent neurodegenerative disorder and is characterized by a progressive decline in learning and memory abilities. The global prevalence of AD is projected to increase from 57 million to 152 million by 2050, imposing a significant economic burden on society [1]. Despite the development of numerous drugs targeting specific pathological features of AD, such as amyloid-beta (Aβ) deposition and neurofibrillary tangles, very few have been approved in the past 20 years, underscoring the limitations of single-target therapeutic approaches. The mechanisms underlying AD are complex and involve a wide range of biological processes. Therefore, an innovative strategy to address the multifaceted nature of AD is to identify compounds capable of modulating multiple targets [2].

Signaling pathways are crucial for understanding disease mechanisms and investigating the mechanisms of action (MOA) of drugs. While the pathogenesis of AD is not yet fully understood, numerous studies have illuminated the signaling pathways and molecular targets implicated in its pathology. It has been reported that 91% of all signaling pathways and over 3,000 disease targets are related to AD [3]. Thus, identifying novel therapeutics that target multiple pathways and genes associated with AD represents a promising strategy for treatment.

Natural products, derived from various sources, exhibit high safety profiles and possess the ability to interact with multiple disease targets, making them valuable resources for drug development [4]. Compounds such as Curcumin, Resveratrol, Quercetin, Cryptotanshinone, Honokiol, and Sulforaphane have been found to mitigate AD pathology through multiple pathways and targets [5,6,7]. Network-based medicine, an advanced approach that integrates systems biology and network science, provides a comprehensive understanding of the complex interactions within biological systems. This approach has demonstrated unique advantages in addressing complex diseases such as cancer, neurodegenerative disorders, and cardiovascular diseases [8, 9]. Unlike the traditional “one drug-one target-one disease” paradigm, network-based drug discovery reveals disease mechanisms from a holistic perspective, fully considering the complexity of biological systems [10].

In this study, we aimed to establish an AD-related pathway-gene network to identify novel natural products for AD treatment. We conducted a safety evaluation of the candidate compounds, followed by an assessment of their therapeutic effects in APP/PS1 transgenic mice. Finally, transcriptomic analysis was performed to explore the underlying MOA of these natural products (Fig. 1).

Fig. 1

Study overview. In this study, we first constructed an AD-related pathway-gene network to screen natural products with greater potential for AD treatment. Combined with literature searches, (-)-Vestitol and Salviolone were selected for further experimental study as novel natural products that target the most AD-related pathways and genes classified as “Therapeutic effects for non-AD” (a). A safety evaluation experiment was conducted to assess the safety of the natural products administered to mice prior to the treatment experiment. The indices assessed included body weight trajectories and body weight change, hematological and serum biochemical analysis, organ index and histopathological characteristics of vital organs (b). A treatment experiment was subsequently conducted to explore the therapeutic effects of the two natural products, particularly their combination. A battery of behavioral tests was performed to evaluate the cognitive function of the mice, while IHC and ELISA were performed to examine Aβ pathology and toxic soluble Aβ levels (c). Finally, transcriptomic analysis was conducted to reveal the MOA of the synergistic effects of (-)-Vestitol and Salviolone cotreatment, including analyses of gene expression correlation, DEGs between groups, KEGG pathway enrichment, and clustering of DEGs in key pathways. qRT-PCR was performed to validate the expression levels of several key genes in the key pathways (d)

Methods

Network-based identification of natural products

Construction of the AD-related pathway-gene network

In this study, we aimed to screen AD-related pathways from three perspectives: Most Studied Pathways (pathways with the most literature on AD), Gene-Associated Pathways (derived from AD-associated gene enrichment analysis), and Popular Pathways (pathways showing an increasing trend in the literature). These three categories were selected to comprehensively elucidate the mechanisms of AD from different dimensions, addressing the often observed disconnect between drug development and mechanistic research. AD-related pathways were curated on the basis of a previous study that systematically evaluated the associations of Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways with AD through text mining [3]. The detailed procedure for screening the pathways is described below.

In this text-mining study, the AD-pathway_score (ranging from 1 to 5) reflects the strength of the relevance of a given pathway to AD, based on a curated review of the specific pathway-associated literature (lower scores indicate stronger associations). To identify pathways closely related to AD, we applied a filter with an AD-pathway_score ≤ 3. This threshold was chosen to prioritize pathways with at least moderate evidence of a link to AD, ensuring that our network was focused on biologically relevant pathways while still being inclusive enough to capture a broad range of potential mechanisms. Subsequently, from the previously filtered pathways, three types of pathways were then identified. Most Studied Pathways were those ranked within the top 30 based on the count of their AD-related literature (AD-specific_paper_rank ≤ 30). AD-specific_paper_rank is the rank of the pathways based on their total count of the literature related to AD. Gene-Associated Pathways were identified using the Enrichment_rank of the pathways based on their statistical significance (P values) of KEGG pathway enrichment analyses for AD-associated genes from different sources: Enrichment_rank_1 used AD-associated genes from Open Targets, and Enrichment_rank_2 used genetically associated AD genes identified in genome-wide association and familial studies. Pathways with either Enrichment_rank_1 ≤ 30 or Enrichment_rank_2 ≤ 30 were considered Gene-Associated Pathways. For both Most Studied Pathways and Gene-Associated Pathways, the threshold of rank ≤ 30 was chosen to balance sensitivity and specificity, capturing pathways with substantial evidence while minimizing the inclusion of less relevant ones. Popular Pathways were identified by analyzing trends in dementia-related publications over the past 10 years using Pathway_ranks (the rank of the pathways based on their yearly count of the literature related to AD). We conducted linear regression analysis of the Pathway_ranks over the past 10 years of each pathway, and pathways exhibiting a statistically significant upward trend (P < 0.05 and the correlation coefficient R < 0, indicating a decreasing rank number, and thus an increasing trend in popularity over time) were classified as Popular Pathways. By focusing on pathways with a trend of increasing literature on AD, we aimed to increase the likelihood of identifying pathways that are truly involved in promising novel AD pathogenesis and may represent promising novel therapeutic targets for drug development. Detailed information about each pathway’s association with AD, reflected by the filtering indices, is provided in Additional file 1: Table S1.

We collected pathway-associated genes (downloaded in September 2022) from three well-established databases: KEGG, REACTOME, and Wiki Pathways. KEGG provides manually curated pathway maps, emphasizing molecular interactions and reactions, particularly in metabolism and signaling [11]. REACTOME is a peer-reviewed database dedicated to representing biological pathways in humans, emphasizing molecular details of signal transduction, transport, metabolism, and other cellular processes [12]. WIKI pathways is an open, community-driven platform offering a broad range of biological pathways, facilitating collaborative editing and updates [13]. These databases were chosen for their complementary strengths, differing curation strategies, and comprehensive coverage of biological pathways, ensuring a robust and reliable network for natural products screening. All genes collected from these three databases were merged to form a comprehensive gene set for each pathway. The AD-related pathway-gene network was then constructed based on these pathways and their corresponding gene sets and visualized using Cytoscape 3.10.1 software [14].

Network-based identification of natural products

Natural products with favorable absorption, distribution, metabolism, and excretion (ADME) properties that can penetrate the blood-brain barrier and exert pharmacological effects in the central nervous system were screened from the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) [15]. The screening criteria included drug likeness ≥ 0.18, oral bioavailability ≥ 30%, and blood-brain barrier ≥ 0.3. The target gene set for each selected natural product was also obtained from TCMSP (downloaded in September 2022) and subsequently mapped to the AD-related pathway-gene network. Natural products with greater potential for AD treatment were identified using Cytoscape 3.10.1 software, which applied a threshold of twofold the median degree value of all natural products [16,17,18].

Selection of natural products for experimental validation

To demonstrate the feasibility of the AD-related pathway-gene network-based method for identifying natural products for AD, we conducted a comprehensive search of PubMed, Web of Science, and Scopus (searched in December 2022) to gather information on the research status of the screened natural products. Based on this literature search, the research status were categorized as “No therapeutic effects reported”—no reported therapeutic effects for these natural products in relation to any specific disease; “Therapeutic effects for AD”—these natural products have been reported to have therapeutic effects specifically for AD; “Therapeutic effects for non-AD”—these natural products have reported therapeutic effects for diseases other than AD; and “Toxic effects reported”—these natural products have been reported to have toxicity. In line with our goal of identifying novel natural products with multi-target potential against AD, we focused on the “Therapeutic effects for non-AD” category. This category was chosen because these natural products had demonstrated biological activity in other contexts, lacked the reported toxicity, and, importantly, had not been previously explored for AD treatment, positioning them as promising candidates for novel AD therapeutics. Therefore, from this refined group, two natural products targeting the most AD-related pathways and genes were selected for further experimental study. Building on this multi-target approach, we primarily explored the therapeutic effects of a combination of the two selected natural products, hypothesizing that cotreatment would modulate a broader range of AD-related pathways compared to individual treatments, potentially leading to synergistic effects. The natural product-gene-pathway networks for these two compounds were constructed based on their corresponding pathways and genes.

Materials and reagents

(-)-Vestitol (HPLC ≥ 98%, BBP01776) was purchased from BioBioPha (Kunming, China). Salviolone (HPLC ≥ 98%, WuXiNP01246) was obtained from WuXi AppTec (Tianjin, China). Donepezil hydrochloride (LCMS ≥ 99%, HY-B0034), PEG300 (HY-Y0873), and Tween-80 (HY-Y1891) were purchased from MedChemExpress (NJ, USA). Dimethyl sulfoxide (DMSO, D2650), BCA Protein Assay Kit (BCA1), and Triton X-100 (9036-19-5) were purchased from Sigma-Aldrich (Missouri, USA). The anti-β-amyloid 1–16 antibody (6E10, 803015) was purchased from BioLegend (San Diego, USA). The secondary antibody (ab6789) and DAB Substrate Kit (ab64238) were purchased from Abcam (Cambridge, UK). Blocking goat serum (SL038) and phenylmethylsulfonyl fluoride (PMSF, P0100) were purchased from Solarbio (Beijing, China). Radioimmunoprecipitation assay (RIPA) buffer (WB3100) and protease and phosphatase inhibitor cocktail (P002) were purchased from NCM Biotech (Suzhou, China). Enzyme-linked immunosorbent assay (ELISA) Kits for Aβ₁₋₄₀ (27713), Aβ₁₋₄₂ (27711) and Aβ oligomers (27725) were purchased from Immuno-Biological Laboratories (Fujioka, Japan). The RNAsimple Total RNA Kit (DP419) was purchased from Tiangen Biotech (Beijing, China). The PrimeScript™ RT Reagent Kit with gDNA Eraser (RR047A) and SYBR^® Premix Ex Taq™ II (Tli RNaseH Plus) Kit (RR820A) were purchased from Takara Bio (Shiga, Japan).

Mice

Male APP/PS1 transgenic mice and C57BL/6J wild-type mice, aged 6 months and weighing 28–32 g, were obtained from Jiangsu Wukong Biotechnology Co., Ltd. (Jiangsu, China) and acclimated for 1 week prior to the experiments. The mice were housed in a specific pathogen-free facility at a constant temperature of 22 ± 1 ℃ and a relative humidity of 50 ± 10% and were subjected to a 12 h light/dark cycle. Standard food and water were provided ad libitum. All animal experiments conducted in this study were approved by the Institutional Animal Ethics Committee, Capital Medical University, and adhered to ethical regulations for animal research and testing. Every effort was made to minimize the number of animals used and their suffering.

Compound preparation

(-)-Vestitol, Salviolone, their mixture, and Donepezil hydrochloride were initially dissolved in 10% DMSO, followed by the sequential addition of 40% PEG300, 5% Tween-80, and 45% normal saline due to the poor water solubility of (-)-Vestitol and Salviolone [19]. This dissolution protocol incorporates PEG300 and Tween-80 as surfactants that are well tolerated by mice to facilitate the dissolution of compounds into a solution or the formation of a homogeneous suspension. To ensure complete dissolution, (-)-Vestitol, Salviolone, and their mixture were sonicated at 37 ℃ until clear and transparent after the addition of DMSO. A solvent mixture without compounds was prepared for the control group mice to ensure identical injection contents and volumes for each group. All the compound solvents were prepared on demand to maintain freshness and stability.

Safety evaluation of natural products

Given the lack of prior animal studies on (-)-Vestitol or Salviolone, evaluating their safety in mice before treatment experiments is essential. To this end, 12 C57BL/6J mice were randomly divided into 4 groups (n = 3 per group) and received intraperitoneal (i.p.) injections of different natural products once daily: (a) vehicle, (b) (-)-Vestitol at 10 mg/kg, (c) Salviolone at 10 mg/kg, and (d) a combination of (-)-Vestitol at 5 mg/kg and Salviolone at 5 mg/kg. The doses used for safety evaluation were twice the treatment doses, which were administered for 14 consecutive days as previously described with minor modifications [20, 21]. Throughout the experimental period, the general behavior and mortality of the mice were monitored. Body weight was recorded daily, with body weight change (%) calculated as (body weight on day 14 − body weight on day 0)/body weight on day 0 × 100%. At the end of the 14-day experiment, blood samples were collected for hematological and serum biochemical analyses, and the organ index and histopathological characteristics of six vital organs—heart, liver, spleen, lung, kidney, and brain—were compared between the groups. The detailed methods are provided in Additional file 2.

To further evaluate long-term safety, the general behavior and mortality of APP/PS1 mice in the treatment experiment (described in the next section) were also monitored daily for 3 months during the administration period. Body weight was recorded weekly, with body weight change (%) calculated as (body weight on week 12 − body weight on week 0)/body weight on week 0 × 100%.

Treatments of natural products

APP/PS1 mice were randomly assigned to one of five treatment groups (n = 8 per group), while 8 C57BL/6J mice served as the wild-type (WT) control group. For three months, the mice received daily i.p. injections of the following compounds: (a) WT group: vehicle; (b) control group: vehicle; (c) (-)-Vestitol group: (-)-Vestitol at 5 mg/kg; (d) Salviolone group: Salviolone at 5 mg/kg; (e) combination group: (-)-Vestitol at 2.5 mg/kg and Salviolone at 2.5 mg/kg; and (f) standard drug group: Donepezil at 1 mg/kg. We selected 5 mg/kg for the individual treatments with (-)-Vestitol or Salviolone, and 2.5 mg/kg for each compound in the combination treatment (maintaining a consistent total dose), based on previous studies demonstrating that an i.p. dose of 5 mg/kg of natural products showed efficacy in mitigating AD pathology in mouse models [5, 22,23,24]. The dose of Donepezil was determined based on previous studies [25, 26].

Behavioral tests

A battery of behavioral tests was performed to assess locomotor activity and cognitive function in the different groups. The tests included the open field (OF) test, novel object recognition (NOR) test, Y-maze test, and Morris water maze (MWM) test, which evaluate exploratory and locomotor activity, recognition memory, short-term working memory, and spatial learning and memory, respectively. All behavioral tests were conducted while the mice were in the active phase in a dark room illuminated only by red light, ensuring that the environment was quiet and isolated from the external noise. The results of the behavioral tests were recorded and analyzed using an automated video-tracking system (Smart 3.0, Panlab).

OF test

The mice were individually placed at the center of an open-field arena (a white acrylic box, 40 × 40 × 40 cm) and allowed to explore freely for 5 min. Rearing was defined as the mouse resting on its hind limbs, assuming a vertical position with its front legs in the air or leaning against the wall. The arena was thoroughly cleaned with 70% ethanol after each test to eliminate any residual odor cues. The track path, mean speed (cm/s), and rearing number were recorded. The track path and mean speed served as measures of locomotor activity, whereas the rearing number was used as an indicator of exploratory behavior.

NOR test

The test was conducted over three consecutive days, consisting of a habituation phase, a training phase, and a testing phase [27]. During the habituation phase, the mice were individually placed in the center of the open-field arena and allowed to explore for 5 min without any objects. Twenty-four hours after habituation, in the training phase, two identical objects (same in texture, color, shape, and size) were placed symmetrically in opposite quadrants of the arena. Each mouse was placed at the center of the arena, equidistant from both objects, and allowed to explore for 10 min. The testing phase occurred 24 h later. In the testing phase, one randomly selected familiar object was replaced with a novel object (different in color and shape), and the mice were reintroduced to explore for another 10 min. The arena and objects were thoroughly cleaned with 70% ethanol between the mice to eliminate residual odor cues. All the objects were of sufficient weight to prevent movement by the mice. Exploration was defined as the mouse directing its nose toward an object within 2 centimeters. The time spent exploring each object was recorded during the training and testing phases. Location preference (%) in the training phase was calculated as the exploration time of one of the identical objects (s)/total exploration time of the two identical objects (s) × 100%. The recognition index (%) in the testing phase was calculated as the exploration time of the novel object (s)/total exploration time of the novel and familiar objects (s) × 100%. A longer exploration time for the novel object and a higher recognition index indicate better recognition memory. Location preference served as an environmental control, which should be approximately 50%, to rule out the influence of object location.

Y-maze test

The Y-maze apparatus consisted of a three-arm horizontal maze (21 cm long, 7 cm wide, and 15.5 cm high) with arms symmetrically positioned at 120° angles from each other. The mice were individually introduced at the distal end of one arm, facing the terminal wall, and allowed to explore freely for 8 min [28]. Between trials, 70% ethanol was used to remove residual odors. Entrance was recorded when all four paws of the mouse entered an arm. The total number and sequence of arm entries were recorded. An alternation was defined as successive entries into all three different arms. The alteration triplet (%) was calculated as the number of alternations/(total number of arm entries − 2) × 100%. A higher alteration triplet indicated better short-term working memory.

MWM test

The experimental procedure followed a previously established protocol [29]. The apparatus consisted of a circular water pool (diameter: 120 cm, depth: 60 cm) filled with white opaque water (20 ± 2 °C) by adding nontoxic titanium dioxide. The pool was divided into four virtual quadrants (NE, SE, SW, and NW), with a transparent escape platform (diameter: 10 cm) placed in the center of the NE quadrant (target quadrant). Visual cues were placed around the maze to assist the mice in locating the platform.

The test comprised a navigation phase (5 days) followed by a probe phase (1 day). During the navigation phase, the platform was 1 cm above the water on day 1 and 1 cm below the water from days 2 to 5. The mice were individually trained to find the escape platform within 60 s across four trials per day. Each trial began with the mice being gently released into the water maze from a specific starting position that varied between trials at the pool’s border (Table 1). If a mouse failed to find the platform within 60 s, it was guided there manually. After each trial, the mice were dried with a towel and returned to their home cages for at least 30 min before the next trial. Two indices were recorded in this phase: escape latency (s) and escape distance (cm), which indicate the time and distance required to reach the platform, respectively. If a mouse did not find the platform within 60 s, the escape latency was recorded as 60 s, and the corresponding distance was noted. The escape latency and escape distance for each day were the average of the four trials.

In the probe phase (day 6), the escape platform was removed from the maze. All the mice were released individually at the starting position SW—the farthest point from the platform—and allowed to explore freely for 60 s. The following indices were recorded during this phase: time spent in platform quadrant (s), distance in platform quadrant (cm), number of platform location crosses, mean swim speed (cm/s), and swimming paths. A shorter escape latency and escape distance indicate better spatial learning ability. A greater amount of time spent in platform quadrant and greater distance in platform quadrant, and more platform location crosses indicated better spatial memory. Mean swim speed should not differ significantly among groups to rule out locomotor ability influencing spatial learning and memory comparisons.

Table 1 Starting positions in the Morris water maze test

Brain tissue preparation

After the behavioral tests were completed, the mice were anesthetized with 1% sodium pentobarbital (40 mg/kg, i.p.) and euthanized via cardiac perfusion with ice-cold phosphate-buffered saline (PBS). The brains were rapidly removed, gently rinsed in cold PBS, and immediately bisected along the midsagittal plane. One half was fixed in 4% paraformaldehyde at 4 ℃ for at least 24 h and subsequently processed in paraffin for immunohistochemistry (IHC). The other half was used to isolate the cerebral cortex and hippocampus, which were immediately stored at -80 ℃ for subsequent analysis by ELISA, transcriptomic analysis, and quantitative reverse transcription polymerase chain reaction (qRT-PCR).

IHC analysis

Paraffin-embedded brain tissues were coronally sectioned into 5 μm thick sections. After deparaffinization and rehydration, the sections were subjected to antigen retrieval using citric acid (pH 6.0) at 95 °C for 15 min, followed by a 10-min incubation with 3% hydrogen peroxide at room temperature to prevent endogenous peroxidation. The sections were blocked with 5% normal goat serum and 0.3% Triton X-100 in PBS for 1 h at room temperature. The primary antibody (6E10, 1:1000) was subsequently applied to the sections and incubated overnight at 4 °C. After rewarming for 1 h, the sections were washed thoroughly in PBS and incubated with a secondary antibody (1:500) for 1.5 h at room temperature. The sections were then visualized using a DAB Substrate Kit, followed by counterstaining with hematoxylin for 50 s. Finally, the sections were dehydrated, made transparent, and sealed before observation and imaging using an optical microscope (Olympus, Japan). The percentage area of Aβ plaque relative to the cortex or hippocampus in each image was analyzed blindly using ImageJ2 software [30].

Aβ ELISA

The cortex and hippocampus from the control, (-)-Vestitol, Salviolone, combination, and standard drug groups (n = 5 per group, randomly selected) were homogenized in RIPA buffer (10-fold volume) containing 1% protease and phosphatase inhibitor cocktail and 1% PMSF. After incubation on ice for 30 min, the homogenized brain tissues were centrifuged at 12,000 × g for 15 min at 4 °C, and the supernatant fraction was collected. Total protein concentrations were determined using the BCA Protein Assay Kit. Toxic soluble Aβ levels, including those of Aβ₁₋₄₀, Aβ₁₋₄₂, and Aβ oligomers, were quantified by corresponding ELISA Kits according to the manufacturer’s instructions. Optical density was measured at 450 nm using a microplate reader (Thermo Fisher Scientific, Inc.).

Transcriptomic analysis

The cortex of WT, control, (-)-Vestitol, Salviolone, and combination groups (n = 5 per group, randomly selected) were used for transcriptomic analysis. RNA-seq was performed by Genesky Biotechnologies Inc. (Shanghai, China), which included total RNA extraction, mRNA enrichment and fragmentation, cDNA library construction, and mRNA sequencing on the Illumina HiSeq 2000 platform in 2 × 150 bp paired-end sequencing mode. The raw data were trimmed to remove adaptor sequences and low-quality sequences, yielding clean data for all downstream analyses.

For differential expression analysis, fragments per kilobase of transcript per million mapped reads (FPKM) were used for data normalization. The DESeq2 R package [31] was employed to analyze digital gene expression data using a negative binomial distribution model. Differentially expressed genes (DEGs) were identified at thresholds of P < 0.05 and|log2(FC)| > 0.5. KEGG pathway enrichment analysis of DEGs was performed using the clusterProfiler R package [32], with P < 0.05 considered significant. Gene expression levels were standardized using the z score method, and cluster analysis of DEGs in key pathways was performed using the pheatmap R package [33] based on standardized FPKM values.

qRT-PCR

The mRNA levels of Glp1r, Htr4, Ntsr1, Nos1, and Ntrk1 were quantified by qRT-PCR using the cortex of the WT, control, (-)-Vestitol, Salviolone, and combination groups (n = 3 per group, randomly selected). Total RNA was extracted and reverse-transcribed into cDNA using the RNAsimple Total RNA Kit and the PrimeScript™ RT Reagent Kit with gDNA Eraser. A SYBR^® Premix Ex Taq™ II (Tli RNaseH Plus) Kit was used to perform qRT-PCR in technical triplicate on a StepOnePlus™ Real-time PCR System (Applied Biosystems, USA) following the manufacturer’s instructions. The qRT-PCR protocol included preincubation at 95 °C for 30 s for 1 cycle; 40 cycles of denaturation at 95 ℃ for 5 s followed by annealing at 60 ℃ for 30 s; and melting at 95 ℃ for 15 s, 60 ℃ for 1 min, and 95 ℃ for 15 s for 1 cycle. Relative mRNA levels were analyzed using the 2^−ΔΔCt method, with glyceraldehyde-3-phosphate dehydrogenase (Gapdh) used as an internal reference gene. The oligonucleotide primers used in this study were designed and synthesized by Sangon Biotech Co., Ltd. (Shanghai, China), and their sequences are listed in Table 2.

Table 2 Sequences of oligonucleotide primers used for qRT-PCR

Statistical analysis

The values are presented as mean ± standard error of the mean (SEM). Two-way analysis of variance (ANOVA) was used to analyze the statistical significance of the differences between multiple groups in terms of the time spent exploring each object during the training phase and testing phase in the NOR test, as well as the escape latency and escape distance in the MWM test. One-way ANOVA was used to analyze differences in other variables among groups. Post hoc analysis for group differences was performed using Bonferroni’s multiple comparisons test. A value of P < 0.05 was considered statistically significant. Statistical analysis and data graphing were performed using IBM SPSS 26.0, GraphPad Prism 8.0.1, and R 4.3.2.

Results

AD-related pathway-gene network identified (-)-Vestitol and Salviolone as novel natural products for AD

In this study, we first constructed an AD-related pathway-gene network to screen for natural products targeting AD. The network comprised 59 pathways in total, including 21 pathways in the Most Studied Pathways, 26 in the Gene-Associated Pathways, and 24 in the Popular Pathways (Fig. 2, Additional file 1: Tables S2–S5, and Additional file 2: Figure S1). Among these pathways, Alzheimer disease, Serotonergic synapse, Calcium signaling pathway, Amyotrophic lateral sclerosis (ALS), Glutamatergic synapse, Cholesterol metabolism, Long-term potentiation, Phagosome, Immune system, and TNF signaling pathway belong to at least two perspectives (Fig. 2).

Fig. 2

Construction of the AD-related pathway-gene network. The network was constructed via AD-related pathways and corresponding pathway gene sets. Most Studied Pathways, Gene-Associated Pathways, and Popular Pathways were screened on the basis of a previous text-mining study (Morgan et al., 2022). Pathway gene sets were obtained from the KEGG, REACTOME, and WIKI pathways. Most Studied Pathways refer to pathways with the most literature on AD. Gene-Associated Pathways were obtained through AD-associated gene enrichment analysis. Popular Pathways refer to those with an increasing trend in the literature. Cytoscape 3.10.1 software was used to visualize the network and calculate the degree values (the number of genes in each pathway). Pathways were ranked according to their degree values within the network, with a higher degree value corresponding to a smaller number

To comprehensively include pathway-associated genes, we collected genes from the KEGG, REACTOME, and WIKI pathways. Overlaps of gene sets from different databases are shown in Additional file 2: Figure S2. We merged genes from the three databases to create a gene set for each pathway (Additional file 1: Tables S2–S4), resulting in 5325, 2557, and 3435 genes for Most Studied Pathways, Gene-Associated Pathways, and Popular Pathways, respectively. We then identified 713 natural products with favorable ADME properties and their corresponding target genes from TCMSP (Additional file 1: Table S6). A total of 177 natural products with greater potential for AD treatment were screened by mapping the target gene set of each natural product to the constructed network, yielding 177, 103, and 29 natural products from Most Studied Pathways, Gene-Associated Pathways, and Popular Pathways, respectively (Additional file 1: Table S7).

To validate the feasibility of our AD-related pathway-gene network-based method for identifying natural products for AD, we conducted a thorough search of PubMed, Web of Science, and Scopus to determine the research status of the 177 natural products. Of these, 43% were reported to have therapeutic effects for AD (“Therapeutic effects for AD”), with 35% based on basic experiments and 8% based on network pharmacology studies (Fig. 3a and Additional file 1: Table S7). This finding indicates that our method is a viable strategy for screening natural products with good potential for AD. Additionally, 6% of the natural products were reported to have therapeutic effects for diseases other than AD (“Therapeutic effects for non-AD”) (Fig. 3a). These natural products are biologically active and have no reported toxic effects or therapeutic effects associated with AD. We selected two natural products that target the most AD-related pathways and genes from this group—(-)-Vestitol and Salviolone—for subsequent experimental studies.

Fig. 3

Research status of screened natural products and natural product-gene-pathway networks of selected compounds. The research status of the natural products identified as having greater potential for AD treatment is shown (a). (-)-Vestitol and Salviolone were selected for further experimental study, and their natural product-gene-pathway networks were constructed (b). PowerPoint was used to display the research status of these natural products. Cytoscape 3.10.1 software was used to visualize the networks, which displayed only pathways with a degree value ≥ 5

We constructed natural product-gene-pathway networks for (-)-Vestitol and Salviolone on the basis of their corresponding pathways and genes. The networks showed that the major potential targeting pathways of (-)-Vestitol included Neuroactive ligand-receptor interaction pathway, Calcium signaling pathway, Metabolic pathways, Alzheimer disease, Estrogen signaling pathway, PI3K-Akt signaling pathway, and Immune system. For Salviolone, the major potential targeting pathways included Neuroactive ligand-receptor interaction pathway, Calcium signaling pathway, Cholinergic synapse, and Metabolic pathways (Fig. 3b).

(-)-Vestitol, Salviolone, and their combination showed no toxicity in the safety evaluation experiment

To assess the safety of the natural products, we recorded the general behavior, mortality, and body weight of the mice. During the 14-day experiment, no signs of toxicity or mortality were observed, and there were no significant changes in body weight trajectories among the different groups (Fig. 4a). Compared with the control group, there was no significant difference in body weight change (%) among the groups that were administered different natural products (Fig. 4a).

Fig. 4

(-)-Vestitol, Salviolone, and their combination showed no toxicity in the safety evaluation experiment. C57BL/6J mice were i.p. administered different natural products or vehicle for 14 consecutive days at doses twice those used in the treatment experiment, as described in the Methods section. Compared with those in the control group, there were no significant differences in body weight trajectories, body weight change (%) (a), organ index (b), or histopathological characteristics (c) among the groups treated with different natural products. Body weight change (%) was calculated as (body weight on day 14 − body weight on day 0)/body weight on day 0 × 100%. The organ index (mg/g) was calculated as organ weight (mg)/body weight (g). Sections of the organs were stained with H&E. The error bars represent mean ± SEM (n = 3). Statistical analysis for (a, b) was performed via one-way ANOVA with Bonferroni’s multiple comparisons test. A value of P < 0.05 was considered statistically significant. Scale bar = 100 μm in (c). Abbreviations: Ves: (-)-Vestitol; Sal: Salviolone

After 14 days of administration of different natural products or vehicle, blood samples were collected for hematological and serum biochemical analyses. No significant differences in hematological or serum biochemical parameters were observed between any treatment group and the control group (Additional file 2: Tables S8–S9). Additionally, 6 vital organs (heart, liver, spleen, lung, kidney, and brain) were isolated to calculate the organ index and observe histopathological changes. Compared with those of the control group, there were no significant differences in the organ index for any of the organs (Fig. 4b). Representative images of the organs via hematoxylin and eosin (H&E) staining showed no obvious signs of injury or toxicity across the different groups (Fig. 4c).

Long-term safety evaluation in APP/PS1 mice during the 3-month treatment also revealed no toxicity. Specifically, no signs of toxicity or mortality were observed. Body weight trajectories and final body weight change (%) were comparable between all groups (Additional file 2: Figure S3).

The combination of (-)-Vestitol and Salviolone rescued cognitive deficits in APP/PS1 transgenic mice

To investigate the therapeutic effects of (-)-Vestitol and Salviolone, particularly their combination, we administered different compounds to 6-month-old APP/PS1 transgenic mice for three months. Following the treatment, a battery of behavioral tests was conducted to evaluate locomotor activity and cognitive function.

First, we performed the OF test to exclude any influence of exploratory or locomotor activity on cognitive function evaluation. During the 5-min exploration in the open-field arena, there were no significant differences in the mean speed (P = 0.3225; Fig. 5a), rearing number (P = 0.9997; Fig. 5b) or track paths (Fig. 5c) among the groups, indicating that there were no significant differences in exploratory or locomotor activity.

Fig. 5

The combination of (-)-Vestitol and Salviolone rescued cognitive deficits in APP/PS1 transgenic mice. The mean speed (a), rearing number (b), and representative track paths (c) in the OF test showed no significant differences in exploratory or locomotor activity among the groups. The time spent exploring left and right objects during the training phase (d) and novel and familiar objects during the testing phase (e), the recognition index (f), and location preference (g) in the NOR test indicated that the combination rescued recognition memory. Alteration triplet (h) in the Y-maze test showed that (-)-Vestitol, Salviolone and their combination rescued short-term working memory. The escape latency (i), time spent in platform quadrant (j), distance in platform quadrant (k), number of platform location crosses (l), mean swim speed (m), and representative swimming paths (n) in the MWM test demonstrated that the combination rescued spatial learning and memory. The recognition index (%) was calculated as the exploration time of the novel object (s)/total exploration time of the novel and familiar objects (s) × 100%. Location preference (%) was calculated as the exploration time of one of the identical objects (s)/total exploration time of the two identical objects (s) × 100%. The alteration triplet (%) was calculated as the number of alternations/(total number of arm entries − 2) × 100%. The error bars represent mean ± SEM (n = 8). Statistical analysis was performed via one-way ANOVA for (a, b, f, g, h, j, k, l, and m), and two-way ANOVA for (d, e, and i), followed by Bonferroni’s multiple comparisons test. Significance is indicated as *P < 0.05 and **P < 0.01 in (e). Significance is indicated as *P < 0.05, **P < 0.01, and ***P < 0.001 compared with the WT group; ^#P < 0.05 and ^###P < 0.001 compared with the control group in (f, h, i, j, and k). Abbreviations: Ves: (-)-Vestitol; Sal: Salviolone; Don: Donepezil

Next, we evaluated cognitive function using the NOR test, Y-maze test, and MWM test. In the NOR test, there was no difference in the time spent exploring the left and right objects during the training phase among the groups (Fig. 5d). However, during the testing phase, the time spent exploring the novel object was significantly longer than that spent exploring the familiar object in the WT group (P = 0.0025; Fig. 5e), (-)-Vestitol group (P = 0.0282; Fig. 5e), Salviolone group (P = 0.0151; Fig. 5e), combination group (P = 0.0327; Fig. 5e), and standard drug group (P = 0.0211; Fig. 5e). These findings indicate impaired recognition memory in APP/PS1 transgenic mice, whereas the administration of (-)-Vestitol, Salviolone, and their combination improved recognition memory. Furthermore, compared with that in the WT group, the recognition index significantly decreased in the control group (P = 0.0390; Fig. 5f). In contrast, compared with that in the control group, the recognition index significantly increased in both the combination group (P = 0.0198; Fig. 5f) and the standard drug group (P = 0.0220; Fig. 5f), indicating that the combination of (-)-Vestitol and Salviolone rescued recognition memory in APP/PS1 transgenic mice. There was no significant difference in location preference among the different groups in the NOR test (P = 0.6100; Fig. 5g), suggesting that object location did not influence the results.

In the Y-maze test, the control group presented significantly fewer altered triplets than the WT group did (P = 0.0001; Fig. 5h), indicating impaired short-term working memory in APP/PS1 transgenic mice. In contrast, the (-)-Vestitol group (P = 0.0121; Fig. 5h), Salviolone group (P = 0.0183; Fig. 5h), combination group (P = 0.0001; Fig. 5h), and standard drug group (P = 0.0001; Fig. 5h) presented significantly increased number of altered triplets compared with the control group, suggesting that (-)-Vestitol, Salviolone and their combination rescued short-term working memory.

During the 5-day navigation phase of the MWM test, all the mice tended to have shorter escape time and distance. However, on day 5, the control group exhibited significantly longer escape latency (P = 0.0116; Fig. 5i) and distance (P = 0.0029; Additional file 2: Figure S4) than the WT group did, indicating impaired spatial learning ability in APP/PS1 transgenic mice. Notably, both the combination group (P = 0.0337; Fig. 5i) and the standard drug group (P = 0.0261; Fig. 5i) presented significantly shorter escape latencies than did the control group. Similarly, both groups exhibited significantly shorter escape distances than did the control group (P = 0.0259 for the combination group and P = 0.0291 for the standard drug group; Additional file 2: Figure S4), suggesting that the combination of (-)-Vestitol and Salviolone rescued the spatial learning ability of APP/PS1 transgenic mice.

On day 6, after the escape platform was removed, spatial memory was evaluated. Compared with the WT group, the control group spent significantly less time (P = 0.0076; Fig. 5j) and traveled shorter distance in the platform quadrant (P = 0.0034; Fig. 5k), indicating impaired spatial memory in APP/PS1 transgenic mice. In contrast, both the combination group (P = 0.0438; Fig. 5j) and the standard drug group (P = 0.0327; Fig. 5j) spent significantly more time in the platform quadrant than did the control group. Similarly, both groups traveled longer distance in the platform quadrant than the controls did (P = 0.0467 for the combination group and P = 0.0191 for the standard drug group; Fig. 5k), suggesting that the combination of (-)-Vestitol and Salviolone rescued spatial memory. Additionally, while both groups presented more platform location crosses than the control group did, this difference did not reach statistical significance (Fig. 5l). Importantly, there was no significant difference in the mean swim speed among the groups (Fig. 5m), ruling out the influence of locomotor ability on spatial learning and memory comparisons. Finally, representative swimming paths indicated that, compared with those in the control group, the spatial memory of the mice treated with the combination of (-)-Vestitol and Salviolone improved (Fig. 5n).

(-)-Vestitol, Salviolone, and their combination reduced the Aβ plaque burden and toxic soluble Aβ levels in APP/PS1 transgenic mice

We examined Aβ deposition in the cortex and hippocampus of the mice. As shown in Fig. 6a, the mice in the control group presented numerous Aβ plaques stained by 6E10, whereas treatment with (-)-Vestitol, Salviolone, their combination, and Donepezil reduced the Aβ plaque burden to varying extents. Compared with that in the control group, the percentage area of Aβ plaques in the cortex was significantly reduced in the (-)-Vestitol group (P = 0.0126; Fig. 6b), Salviolone group (P = 0.0299; Fig. 6b), combination group (P < 0.0001; Fig. 6b), and standard drug group (P < 0.0001; Fig. 6b). Similarly, the percentage area of Aβ plaques in the hippocampus was significantly reduced in the (-)-Vestitol group (P < 0.0001; Fig. 6c), Salviolone group (P = 0.0092; Fig. 6c), combination group (P < 0.0001; Fig. 6c), and standard drug group (P < 0.0001; Fig. 6c) than in the control group.

Fig. 6

(-)-Vestitol, Salviolone, and their combination reduced Aβ pathology in APP/PS1 transgenic mice. Representative images of 6E10 staining in the cortex and hippocampus (a) show a significant decrease in the area of Aβ plaque in the cortex (b) and hippocampus (c) after the administration of (-)-Vestitol, Salviolone, and their combination. Toxic soluble Aβ levels, including Aβ₁₋₄₀ (d), Aβ₁₋₄₂ (e), and Aβ oligomer (f), in the cortex and Aβ₁₋₄₀ (g), Aβ₁₋₄₂ (h), and Aβ oligomer (i) in the hippocampus were also decreased after administration. The combination exhibited synergistic effects on reducing the Aβ plaque burden and toxic soluble Aβ levels. The percentage area of Aβ plaque stained was analyzed blindly using ImageJ2 software. Error bars represent mean ± SEM (n = 8 in b, c and n = 5 in d ~ i). Statistical analysis was performed via one-way ANOVA with Bonferroni’s multiple comparisons test. Significance is indicated as *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001 compared with the control group; ^#P < 0.05, ^##P < 0.01, ^###P < 0.001, and ^####P < 0.0001 compared with the combination group. Abbreviations: Ves: (-)-Vestitol; Sal: Salviolone; Don: Donepezil

ELISAs were performed to examine the levels of toxic soluble Aβ, including soluble Aβ₁₋₄₀, Aβ₁₋₄₂, and Aβ oligomer in the cortex and hippocampus. In the cortex, the levels of soluble Aβ₁₋₄₀, Aβ₁₋₄₂, and Aβ oligomer were significantly lower in the (-)-Vestitol group (P = 0.0041, 0.0088, and 0.0049, respectively), Salviolone group (P = 0.0361, 0.0365, and 0.0158, respectively), combination group (P < 0.0001 for all), and standard drug group (P < 0.0001, P < 0.0001, and P = 0.0001, respectively) than in the control group (Fig. 6d ~ f). Similarly, in the hippocampus, the levels of soluble Aβ₁₋₄₀, Aβ₁₋₄₂, and Aβ oligomer were significantly lower in the (-)-Vestitol group (P = 0.0062, 0.0077, and 0.0098, respectively), Salviolone group (P = 0.0063, 0.0109, and 0.0137, respectively), combination group (P < 0.0001 for all), and standard drug group (P < 0.0001, P = 0.0002, and P < 0.0001, respectively) than in the control group (Fig. 6g ~ i). These findings indicated that (-)-Vestitol, Salviolone, and their combination reduced both the Aβ plaque burden and toxic soluble Aβ levels in the cortex and hippocampus of APP/PS1 transgenic mice.

Notably, the combination of (-)-Vestitol and Salviolone exhibited synergistic effects on reducing the Aβ plaque burden and toxic soluble Aβ levels. In the cortex, both the percentage area of Aβ plaques and the levels of soluble Aβ₁₋₄₀, Aβ₁₋₄₂, and Aβ oligomer were significantly reduced in the combination group than in the (-)-Vestitol group (P = 0.0391, 0.0041, 0.0471, and 0.0425, respectively) and the Salviolone group (P = 0.0166, 0.0005, 0.0115, and 0.0132, respectively) (Fig. 6b, d ~ f). Similarly, in the hippocampus, both the percentage area of Aβ plaques and the levels of soluble Aβ₁₋₄₀, Aβ₁₋₄₂, and Aβ oligomer were significantly reduced in the combination group than in the (-)-Vestitol group (P = 0.0491, 0.0206, 0.0400, and 0.0476, respectively) and the Salviolone group (P < 0.0001, P = 0.0202, 0.0282, and 0.0344, respectively) (Fig. 6c, g ~ i). Furthermore, when comparing the combination therapy to the standard treatment, Donepezil, the reduction in Aβ plaque burden in both the cortex and hippocampus was similar between the groups, while the combination group showed a trend towards lower levels of soluble Aβ. These findings indicated that the combination of (-)-Vestitol and Salviolone had synergistic effects on reducing both the Aβ plaque burden and toxic soluble Aβ levels in the cortex and hippocampus of APP/PS1 transgenic mice.

The combination of (-)-Vestitol and Salviolone synergistically regulated more genes and pathways related to AD

To elucidate the underlying MOA for the synergistic effects of the combination of (-)-Vestitol and Salviolone, we performed RNA-seq analysis on the cortex of the WT, control, (-)-Vestitol, Salviolone, and combination groups. Initially, we analyzed the correlation of gene expression between each pair of samples using the Spearman correlation coefficient. As shown in Fig. 7a, the gene expression pattern of the combination group was more similar to that of the WT group and more distinct from that of the control group than those of the single-natural product groups.

Fig. 7

The combination of (-)-Vestitol and Salviolone synergistically regulated more genes and pathways related to AD. The gene expression correlation graph was generated via the Spearman correlation coefficient between each pair of samples (a). The results showed that the gene expression pattern of the combination group was more similar to that of the WT group and more distinct from that of the control group than those of the single drug groups. A volcano plot of the DEGs between the control group and the WT group, and between treatment groups and the control group showed that the combination regulated more genes than any individual natural product did (b). The upset plot based on all the DEGs showed a greater intersection between the combination group and the control group than between the single drug groups (c). The comet plots of the results of the KEGG pathway enrichment analysis of the DEGs between the control and WT groups (d), the (-)-Vestitol and control groups (e), the Salviolone and control groups (f), and the combination and control groups (g) showed that the combination group regulated more pathways, with greater overlap with those in the control group than the single drug groups did. In heatmaps of DEGs in Neuroactive ligand-receptor interaction pathway (h) and Calcium signaling pathway (i), red-marked genes were upregulated DEGs with increased expression levels, whereas blue-marked genes were downregulated DEGs with decreased expression levels compared with those in the single drug groups. Most DEGs in these pathways were synergistically regulated by the combination of (-)-Vestitol and Salviolone. Note: Four upregulated DEGs with excessively high log2(FC) values are not shown in (b). The figures in (d ~ g) represent the number of DEGs enriched in the corresponding pathways; only pathways with P < 0.01 in the control group and the combination group are shown because of the large number of enriched pathways. Abbreviations: Ves: (-)-Vestitol; Sal: Salviolone

Next, we identified DEGs between the control group and the WT group, the (-)-Vestitol group and the control group, the Salviolone group and the control group, and the combination group and the control group. The results revealed that the combination of (-)-Vestitol and Salviolone regulated a greater number of genes in APP/PS1 transgenic mice than either natural product alone (Fig. 7b). Further analysis of the DEGs showed that the combination group had a significantly greater overlap with the control group than the single-drug groups did (Fig. 7c). Moreover, intersections between upregulated DEGs in the control group and downregulated DEGs in the treatment groups, as well as between downregulated DEGs in the control group and upregulated DEGs in the treatment groups, also demonstrated that the combination group had a greater overlap with the control group than either single-drug group did (Additional file 2: Figure S5). These findings indicated that the combination of (-)-Vestitol and Salviolone synergistically regulated more genes dysregulated in AD than either natural product alone did.

KEGG pathway enrichment analysis of the DEGs from the control group (Fig. 7d), (-)-Vestitol group (Fig. 7e), Salviolone group (Fig. 7f), and combination group (Fig. 7g) revealed that the combination regulated a greater number of pathways than either natural product alone did. Notably, pathways in the combination group presented a greater degree of overlap with those in the control group than those in the single-natural product groups did, indicating that this combination regulated more AD-related pathways. Among the enriched pathways, Neuroactive ligand-receptor interaction and Calcium signaling pathway were prominently represented across all groups. Since these two pathways were previously identified as crucial in the AD-related pathway-gene network, we performed cluster analysis on the union of DEGs from all four groups involved in either Neuroactive ligand-receptor interaction or Calcium signaling pathway. The results showed that, compared with those in response to either natural product alone, most DEGs in both pathways were synergistically regulated by the combination (Fig. 7h and i).

Following a literature search, we selected the Glp1r, Htr4, Ntsr1, Nos1, and Ntrk1 genes from the Neuroactive ligand-receptor interaction or Calcium signaling pathway for further validation. Comprehensive transcriptomic analysis revealed relatively low expression levels in both the control and single-natural product groups; however, significantly higher expression levels were detected in the combination group (Fig. 8a). Notably, the P values for these genes were substantially more significant, and the fold changes in their differential expression levels were greater in the combination group than those in the single-natural product groups (Fig. 8a). These results suggested that the Glp1r, Htr4, Ntsr1, Nos1, and Ntrk1 genes were novel DEGs in the combination group, whose expression levels did not reach the DEG status in either the (-)-Vestitol group or the Salviolone group.

Fig. 8

The combination of (-)-Vestitol and Salviolone synergistically regulated the expression of five pathway genes. The FPKM values of each gene were averaged across 5 samples per group to obtain average gene expression levels, which were then standardized via the z score method. Transcriptomic analysis comprised a heatmap on the left plotted using standardized gene expression levels and a bubble plot on the right based on the P values and FC values, with redder bubbles indicating smaller P values and larger bubbles indicating higher FC values (a). Quantification of Glp1r, Htr4, Ntsr1, Nos1, and Ntrk1 gene expression via qRT-PCR (b). The results showed that the combination of (-)-Vestitol and Salviolone synergistically regulated the expression of these genes. Relative mRNA levels were analyzed using the 2^−ΔΔCt method, with Gapdh used as an internal reference gene. The error bars represent mean ± SEM (n = 3). Statistical analysis was performed via one-way ANOVA with Bonferroni’s multiple comparisons test. Significance is indicated by **P < 0.01 and ***P < 0.001 compared with the WT group; ^##P < 0.01 compared with the control group. Abbreviations: Ves: (-)-Vestitol; Sal: Salviolone

To validate these findings, we quantified the mRNA levels of these genes using qRT-PCR. Compared with those in the WT group, the expression levels of these five genes were lower in the control group (P = 0.1413, 0.0098, 0.0006, 0.0007, and 0.3512 for Glp1r, Htr4, Ntsr1, Nos1, and Ntrk1, respectively), whereas the expression levels were significantly greater in the combination group than in the control group (P = 0.0013, 0.0017, 0.0043, 0.0023, and 0.0016 for Glp1r, Htr4, Ntsr1, Nos1, and Ntrk1, respectively) (Fig. 8b). These results from both transcriptomic analysis and qRT-PCR indicated that the combination of (-)-Vestitol and Salviolone synergistically regulated the expression of the Glp1r, Htr4, Ntsr1, Nos1, and Ntrk1 genes within the Neuroactive ligand-receptor interaction and Calcium signaling pathway.

Discussion

In this study, we constructed an AD-related pathway-gene network by integrating text-mining information and pathway-associated databases. A previous study on the association between KEGG pathways and AD revealed that the pathways most closely related to AD pathology-associated genes or targets of approved AD drugs were not necessarily the most extensively studied, highlighting a disconnect between drug development and mechanism research [3]. To address this, we incorporated three distinct types of pathways into our AD-related pathway-gene network, providing a comprehensive understanding of AD mechanisms from different perspectives.

The Most Studied Pathways, being the most extensively researched and containing a larger number of identified pathway genes, along with Gene-Associated Pathways, which are closely related to AD pathology-associated genes and often focused on drug development, are both more likely to yield natural products with multiple gene targets. In contrast, although Popular Pathways have gained increasing attention in recent years, less comprehensive research has been conducted, resulting in a limited number of genes on which natural products can act and consequently fewer identified natural products.

Among the screened natural products, we selected (-)-Vestitol and Salviolone, two compounds with reported therapeutic effects for diseases other than AD, for further experimental investigation. (-)-Vestitol, a right-handed chiral molecule from the isoflavonoid family, is derived primarily from Pterocarpus soyauxii and Dalbergia sissoo. This compound exhibits various physiological activities in vitro, including antibacterial, anti-inflammatory, antioxidative, and antitumor effects [34,35,36,37,38]. Furthermore, (-)-Vestitol has been identified as an α-glucosidase inhibitor and can bind to the PPAR-γ receptor, suggesting potential therapeutic effects for diabetes mellitus [39, 40]. Salviolone, a bisnorditerpene compound derived from Salvia miltiorrhiza and Salvia paramiltiorrhiza, has been shown to have inhibitory effects on melanoma cells by inhibiting matrix metalloproteinase 2 activity and the STAT3 pathway [41]. Additionally, Salviolone has been shown to inhibit cervical cancer cells by inducing autophagy and inhibiting the NF-κB/mTOR/PI3K/AKT pathway [42]. It has also been found to inhibit ovarian cancer by lowering cyclooxygenase-2 mRNA and prostaglandin E2 levels [43]. In addition to its antitumor effects, Salviolone has anti-inflammatory and antioxidative stress effects and has been shown to have antifibrotic effects by reducing the levels of fibrosis-related markers such as fibronectin and collagen IV [44, 45].

To elucidate the underlying MOA of the synergistic effects of the combination of (-)-Vestitol and Salviolone on AD treatment from a pathway and genetic perspective, we conducted transcriptomic analysis. The results revealed that Neuroactive ligand-receptor interaction and Calcium signaling pathway were the primary mechanisms through which this combination exerted synergistic effects. Furthermore, these two pathways are vital components of both the AD-related pathway-gene network and the natural product-gene-pathway networks. Numerous studies have established a close link between these pathways and the pathogenesis of AD [46,47,48]. The Neuroactive ligand-receptor interaction pathway plays a crucial role in neuronal communication, involving neurotransmitters like glutamate, 5-hydroxytryptamine (5-HT), acetylcholine, and histamine, all of which are vital for cognitive and memory processes [49]. Our findings indicate that the combination of (-)-Vestitol and Salviolone significantly modulates this pathway, likely through a multifaceted effect on various neurotransmitter systems. Specifically, we observed an upregulation of Glp1r (encoding glucagon-like peptide-1 receptor, GLP-1R), Htr4 (encoding 5-HT4 receptor), and Ntsr1 (encoding neurotensin receptor 1, NTR1). GLP-1R, widely distributed in the central nervous system, plays a crucial role in regulating various brain pathological processes, including neuroinflammation, nutrient metabolism, mitochondrial function, and synaptic function [50,51,52]. GLP-1R agonists have demonstrated the ability to improve cognitive function, reduce hippocampal synaptic loss, and decrease Aβ plaque deposition and Aβ oligomer levels in APP/PS1 transgenic mice [53, 54]. Additionally, GLP-1R agonists inhibit tau protein hyperphosphorylation, promote the release of brain-derived neurotrophic factor, and reduce neuroinflammatory responses by inhibiting microglial and astrocyte activation [55,56,57]. The 5-HT4 receptor, specifically expressed in the prefrontal cortex and limbic system (including the hippocampus and amygdala), regulates learning, memory, and cognitive function [58, 59]. Studies have found that Htr4 gene expression is suppressed in the hippocampus of AD patients, and activating 5-HT4 receptors can improve cognitive function [60, 61]. Furthermore, 5-HT4 receptors can induce non-amyloidogenic cleavage of amyloid precursor protein (APP), reducing Aβ production [62, 63]. NTR1 levels are significantly decreased in the hippocampus of cognitively impaired rats and AD patients [64, 65]. NTR1 activation enhances neuronal excitability and improves spatial learning ability and memory in APP/PS1 transgenic mice [66]. NTR1 agonists can exert neuroprotective effects and improve cognitive function by alleviating excessive microglial activation and inhibiting oxidative stress, cell apoptosis, and synaptic loss in the hippocampus [67, 68].

The Calcium signaling pathway also plays a critical role in AD pathogenesis. Calcium homeostasis dysregulation impacts Aβ aggregation, tau hyperphosphorylation, synaptic damage, neuronal apoptosis, mitochondrial dysfunction, oxidative stress, and neuroinflammation [69,70,71,72,73,74,75]. Aβ aggregation and tau hyperphosphorylation can disrupt the calcium balance in neurons, leading to mitochondrial calcium overload, increased reactive oxygen species production, reduced ATP generation, and ultimately, neuronal apoptosis or necrosis [76, 77]. We also observed an upregulation of Nos1 (encoding nitric oxide synthase, NOS1) and Ntrk1 (encoding neurotrophic tyrosine kinase receptor 1, TrkA) in this pathway. NOS1 catalyzes the production of nitric oxide (NO), a crucial neurotransmitter involved in brain development, synaptic plasticity, learning, and memory processes [78, 79]. NOS1 is highly expressed in the hippocampus, and selective inhibition of NOS1 can block NO production and inhibit long-term potentiation, affecting cognitive function, particularly working memory [80,81,82]. Disruption of NOS1 dimerization can lead to abnormal CDK5 phosphorylation and activation, which impede synaptic plasticity and promote AD development [83, 84]. TrkA is a receptor for nerve growth factor (NGF). NGF binding to TrkA promotes APP to undergo non-amyloidogenic cleavage, reducing Aβ production [85, 86]. NGF/TrkA also enhances synaptic plasticity and inhibits Aβ-induced inflammatory responses [87]. Early impairment of NGF/TrkA signaling in AD can lead to reduced synaptic vesicle secretion and decreased presynaptic protein expression, resulting in presynaptic functional impairment of cholinergic primary neurons [88].

The synergistic effects of (-)-Vestitol and Salviolone on both the Neuroactive ligand-receptor interaction and Calcium signaling pathway could potentially lead to the observed improvements in cognitive function and the reduction in Aβ pathology seen in our in vivo experiments. These findings align with previous research demonstrating the therapeutic potential of natural products like Cinnamaldehyde, Sarsasapogenin-AA13, Ginkgo biloba extracts, and active components of Andrographis paniculata in targeting these pathways for AD treatment [49, 89,90,91,92]. The multifaceted actions of (-)-Vestitol and Salviolone, particularly their synergistic effects, highlight their potential as a novel therapeutic strategy for AD.

In this study, we used donepezil as a positive control to evaluate the efficacy of our combination therapy. As a selective, reversible acetylcholinesterase inhibitor, Donepezil primarily enhances cholinergic neurotransmission, improving cognitive function in AD [93]. Evidence also suggests that it may delay Aβ plaque deposition, though its overall disease-modifying effects remain modest [94,95,96,97]. Our combination therapy of (-)-Vestitol and Salviolone, identified via network medicine, targets multiple AD-related pathways, including Neuroactive ligand-receptor interaction and Calcium signaling pathway, potentially offering a more comprehensive, disease-modifying approach. Notably, despite mechanistic differences, our combination therapy’s efficacy in mitigating cognitive deficits and reducing Aβ pathology in APP/PS1 mice was comparable to Donepezil’s. This suggests that (-)-Vestitol and Salviolone may provide an alternative or complementary therapeutic strategy, potentially with enhanced efficacy or reduced side effects, owing to its multi-target and possible disease-modifying nature. However, translating the promising preclinical results of (-)-Vestitol and Salviolone into clinical success for AD treatment faces several significant hurdles. A key challenge is optimizing the bioavailability and pharmacokinetics of these compounds for oral administration in humans. The bioavailability of natural products varies widely based on their chemical structure, which influences absorption and metabolism [98, 99]. Therefore, further research is crucial to understand the bioavailability of these two compounds in humans. Another critical hurdle is demonstrating clinical efficacy in humans. The inherent limitations of animal models in fully capturing the heterogeneity and complexity of AD in human necessitate carefully designed clinical trials with well-defined patient populations and outcome measures. Furthermore, establishing a comprehensive safety and tolerability profile, particularly regarding long-term use and potential drug-drug interactions in the elderly AD population who often take multiple medications, is crucial.

Limitations and future directions

First, this study focused solely on male APP/PS1 mice, so sex-specific effects were not assessed. Future studies should evaluate the efficacy of the natural products in female mice to gain a more comprehensive understanding of their therapeutic potential across both sexes. Second, long-term safety is a critical consideration for any potential AD treatment, and while our short-term safety evaluation did not reveal any significant adverse effects associated with (-)-Vestitol and Salviolone administration, it is crucial to acknowledge that these findings do not guarantee long-term safety when used chronically for AD treatment. Given their potential to regulate multiple pathways and genes, (-)-Vestitol and Salviolone may exert beneficial or unforeseen effects on various physiological systems over extended periods. Furthermore, the combined use of (-)-Vestitol and Salviolone introduces the possibility of interactions beyond the observed synergistic effects for AD. The full spectrum of these interactions, and whether they pose any long-term risks, remains to be determined. These concerns necessitate further research, including dedicated long-term safety studies in animal models, specifically designed to assess potential risks associated with higher doses and prolonged exposure. Moreover, careful and comprehensive safety monitoring will be essential in any future clinical trials evaluating (-)-Vestitol, Salviolone, and especially their combination for AD treatment.

Conclusions

In summary, our study constructed an AD-related pathway-gene network and identified (-)-Vestitol and Salviolone as novel candidate natural products for AD treatment. The combination of (-)-Vestitol and Salviolone had synergistic effects on ameliorating cognitive deficits and reducing the Aβ plaque burden and toxic soluble Aβ levels in APP/PS1 transgenic mice. The synergistic effects were achieved by regulating AD-related pathways and genes more broadly and profoundly.