Using Network Pharmacology to Find Traditional Chinese Medicine Extracts That May Help Alzheimer's by Reducing Brain Inflammation and Improving Gut-Brain Communication

Bioactive compounds from VT, PM, AVF, and EF were screened using the TCMSP database based on OB and DL inclusion criteria. A total of 27, 10, 6, and 3 active components were identified from VT, PM, AVF, and EF, respectively. Notably, vitetrifolin C in VT, procyanidin B1 in AVF, baicalin and stigmasterol in PM, and quercetin and kaempferol in EF exhibited high OB (>30%) and DL (0.24–0.80) (Table S2). From 14,423 AD-related targets, the top 250 compounds were selected by relevance and intersected with drug targets, yielding 25 overlapping potential AD-related targets (Figure 1A).

A network depicting "active ingredient–disease–target" interactions was constructed (Figure 1B). The 25 shared target genes (blue ovals) link AD (red triangle) with bioactive compounds (orange rectangles) derived from VT, PM, AVF, and EF (green arrows). The bioactive compounds contributing most to AD-related targets were 10 in VT, 5 in PM, 2 in AVF, and 2 in EF. PPI analysis revealed 25 target proteins with 195 interactions, highlighting AKT1, IL1B, and IL6 as central nodes (Figure 1C,D).

GO enrichment (top 20 terms, p < 0.01) indicated involvement in cytokine activity (GO:0005125), receptor–ligand interactions (GO:0048018), and immune signaling pathways (Figure 1E). KEGG pathway enrichment identified 88 pathways, with the top 20 significantly enriched (p < 0.00005). These included lipid and atherosclerosis (hsa05417), advanced glycation end-product (AGE)-receptor for AGE (RAGE) signaling in diabetic complications (hsa04933), and fluid shear stress–atherosclerosis pathways (hsa05418) (Figure 1F). These pathway enrichment results provided strong predictive pathways for our subsequent experimental validation. Specifically, we focused on experimentally testing the modulation of the AGE-RAGE and the phosphoinositide 3-kinase (PI3K)-Akt signaling pathway, given their central roles in neuroinflammation, apoptosis, and the cross-talk between peripheral and central inflammation, which aligns with our gut–brain axis investigation.

The binding energy between bioactive compounds (baicalein, kaempferol, luteolin, quercetin) and core AD targets (AKT1, IL1B, IL6) showed the binding affinity between them. The binding energy indicated favorable binding (≤−9.0 kcal/mol), particularly for AKT1, which exhibited the lowest energies: −9.8 (baicalein), −9.7 (kaempferol), and −9.8 kcal/mol (quercetin) (Figure 2A). Hydrogen bonding analysis showed interactions between Akt1 and baicalein (THR-211, SER-205) (Figure 2B), kaempferol (THR-211, LYS-268) (Figure 2C), and quercetin (THR-211, LYS-268, ILE-290) (Figure 2D), supporting their potential as AD-target modulators.

Crude extract yields after freeze-drying were 2.08%, 10.21%, 13.61%, and 11.55% in VT, PM, AVF, and EF, respectively. Polysaccharides and polyphenols were highest in EF (28.5 and 27.6 mg/g dried herb) and followed by AVF (24.0 and 22.8 mg/g dried herb) (). Flavonoids were highest in VT (2.5 mg/g dried herb). Using HPLC-DAD, the key bioactive compounds were confirmed: Kaempferol in VT, AVF, and EF 5.66 mg/g, 0.136 mg/g, and 0.042 mg/g; baicalein and quercetin in PM 1.217 mg/g and 0.146 mg/g; luteolin in EF 0.641 mg/g; and quercetin in VT 8.75 mg/g. Their presence was consistent. Their presence was consistent with network pharmacology results (). Figure S2 Figure S3

Treatment with baicalein, kaempferol, luteolin, and quercetin (2 µM and 5 µM) improved viability, with 2 µM luteolin and quercetin showing the strongest effects in PC12 cells administered with 10 µM Aβ and LPS (p < 0.05). VT, PM, and AVF extract treatment at 20 µg/mL also restored viability close to control levels (Figure 3A,B; p < 0.05), while at 10 µg/mL and 50 µg/mL, AVF increased cell viability more than the normal control without Aβ and LPS administration, but VT, PM, and EF increased compared to the control with Aβ and LPS administration (Figure S4C). In LPS-treated Caco-2 cells, baicalein and quercetin (2 µM) and kaempferol; luteolin; and quercetin (5 µM) significantly improved cell survival (Figure 3B and Figure S4E; p < 0.05). Positive control, VT, PM, AVF, and EF showed protective effects at both doses (Figure 3A; p < 0.05).

Aβ and LPS increased lipid peroxidation by about 5.5-fold and AChE activity in PC12 cells. Pre-treatment with AVF and EF (10 µg/mL) only significantly reduced lipid peroxidation levels. Pre-treatment with positive control, VT, PM, AVF, and EF (20/50 µg/mL) significantly reduced lipid peroxidation and AChE levels as much as the control (Table 1 and Table S3A; p < 0.05). Baicalein, kaempferol, and luteolin were similarly effective against lipid peroxidation, while quercetin showed non-significant reductions in Aβ and LPS administered dual cell model (Table 1 and Table S3B). All four bioactive compounds reduced LPS-induced peroxidation significantly and inhibited AChE activity at 2 µM (Table 1; p < 0.05), suggesting neuroprotective potential via improved cholinergic function.

Gene expression analysis revealed that AVF (50 µg/mL) significantly downregulated TNF-α, IL-1β, IL-6, and Tau and upregulated BDNF in PC12 cells (Figure S5A,B; p < 0.05). VT and EF also showed significant anti-inflammatory and neurotrophic effects at 20 µg/mL. In addition, the positive control group reduced TNF-α and Tau at 20 µg/mL, while upregulating BDNF. (Figure 4A,B; p < 0.05). Individual bioactive compounds reduced TNF-α and enhanced BDNF expression, with baicalein and luteolin being especially effective (Figure 4C,D). In Caco-2 cells, all extracts reduced TNF-α, IL-1β, and IL-6 mRNA expression (p < 0.05), with VT, PM, and AVF showing effects comparable to the controls (Figure 4A; p < 0.05). These results confirm that VT, PM, and AVF had anti-inflammatory and neurotrophic actions. The significant downregulation of these key pro-inflammatory cytokines (TNF-α, IL-1β, IL-6) directly validates the predicted enrichment of the AGE-RAGE signaling pathway and inflammatory-related GO terms. This demonstrates that the extracts effectively suppress the neuroinflammatory and peripheral inflammatory responses predicted by our network pharmacology approach.

Western blot analysis showed that Aβ-induced GSK-3β activation and LPS-induced apoptosis were modulated via the AKT/AMPK/GSK-3β pathway. AVF (20 µg/mL) phosphorylated AKT and reduced apoptosis significantly (p < 0.05). PM, AVF, and EF (20 µg/mL) increased AKT phosphorylation, while the positive control, VT, PM, AVF, and EF (20 µg/mL) enhanced AMPK phosphorylation, and VT, PM, AVF, and EF upregulated the phosphorylation of GSK-3β (Figure 5A,B), suggesting these extracts regulate key neuroinflammatory and apoptotic pathways.

Morphological analysis using H&E staining showed that AVF and EF (20/50 µg/mL) restored LPS-induced cell integrity loss in Caco-2 cells (Figure 6A,B and Figure S6A,B; p < 0.05). Toluidine blue staining revealed that positive control, PM, and AVF significantly reduced mast cell numbers (p < 0.05), while VT and EF had weaker effects at 20 µg/mL (Figure 6A,B; p < 0.05). At 50 µg/mL, all extracts normalized mast cell counts (Figure S6A,B; p < 0.05), indicating anti-inflammatory properties. Mitochondrial membrane potential (MMP) assessment showed that Aβ and LPS decreased MMP in co-cultured PC12 and Caco-2 cells, as indicated by reduced JC-1 aggregation. Treatment with positive control, VT, PM, AVF, EF, and bioactive compounds restored MMP in a concentration-dependent manner, reducing JC-1 monomers and increasing aggregate formation (Figure 6C and Figure S6C), suggesting mitochondrial protection and anti-apoptotic effects.

First, while our Caco-2/PC12 co-culture model provides valuable mechanistic insights into gut–brain axis interactions and has been validated in previous studies [,], it cannot fully recapitulate in vivo complexity, including gut microbiome interactions and blood–brain barrier function. Future research should investigate how these extracts affect microbial communities and their ability to cross the blood–brain barrier in animal models. Second, our study lacks overall pharmacokinetic characterization and systematic comparison between individual compounds and whole extracts. While our concentrations were based on cytotoxicity screening and literature evidence, optimal therapeutic dosing, bioavailability profiles, and potential synergistic or antagonistic interactions within complex herbal formulations remain undetermined. Future studies should employ combination analysis approaches and detailed pharmacokinetic profiling to optimize therapeutic formulations and establish consistent bioactivity relationships. Finally, our findings represent foundational mechanistic research requiring extensive in vivo validation before clinical translation. The pharmacokinetics, bioavailability, safety profiles, optimal dosing regimens, and potential drug interactions of these herbal extracts must be systematically characterized in animal models before any clinical application can be considered. ^{34 35}

In conclusion, our integrated approach demonstrates that VT, PM, AVF, and EF extracts mitigate Aβ-induced neurotoxicity and LPS-induced intestinal inflammation in vitro through multi-target mechanisms involving suppression of neuroinflammatory pathways. Furthermore, this study provides novel evidence that VT, PM, AVF, and EF can simultaneously address both neuronal and intestinal components of AD pathology through gut–brain axis modulation. Our innovative co-culture approach reveals therapeutic mechanisms not previously described in TCM research and demonstrates that network pharmacology predictions can be successfully validated through targeted experimental investigations. The dual protective effects observed in our study suggest that integrated gut–brain axis interventions may offer superior therapeutic outcomes compared to traditional single-target approaches, representing a significant advancement in the development of therapeutic strategies for AD.

Screening of Bioactive Compounds in the Herbs: Active ingredients from VT, PM, AVF, and EF were screened using the Traditional Chinese Medicine Systems Pharmacology (TCMSP) database (https://www.tcmsp-e.com/#/home↗, 23 July 2024). Compounds with oral bioavailability (OB) ≥ 30% and drug-likeness (DL) ≥0.18 were selected as bioactive candidates [36]. Target Identification and Network Construction: Compound-related targets were identified from the TCMSP database. Target names were standardized using the UniProt database. AD-associated targets were obtained from the GeneCards^® database using "Alzheimer's disease" as the keyword. Common targets between the active ingredients and AD were identified using the VENNY 2.1 drawing tool and visualized with the Cytoscape 3.8.2 application to construct a drug-component-disease-target network. Protein–Protein Interaction (PPI) Network: PPI networks were generated using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database with Homo sapiens as the organism and a confidence threshold of ≥0.40. The network was analyzed with Cytoscape 3.8.2, and key targets with degree values higher than average were selected for further analysis. Enrichment Analysis: Drug-disease overlapping targets were analyzed using the Kyoto Encyclopedia of Genes and Genomes (KEGG) database to identify associated canonical pathways. Gene Ontology (GO) enrichment analysis was performed using the Bioconductor package to explore biological processes, cellular components, and molecular functions related to the identified targets [37]. Significance was set at p < 0.05.

The chemical structures of key bioactive compounds (baicalein, kaempferol, luteolin, and quercetin) were downloaded from PubChem and prepared using Chem3D 17.0. Target proteins serine/threonine kinase AKT (AKT1, PDB ID: 3o96), interleukin-6 (IL-6, PDB ID: 1ALU), and IL-1β (PDB ID: 1hib) were retrieved from the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank. Molecular docking was performed using AutoDockTools 1.5.6 with semi-flexible docking parameters. The grid was set to 126 Å × 126 Å × 126 Å centered on the ligand, with 1000 docking runs [38]. The grid box was made of the active site of the protein, corresponding to the website (https://proteins.plus/↗, 30 July 2024). Binding energies were calculated to evaluate ligand–target interactions, and the lowest-energy conformations were selected for interpretation.

VT, PM, AVF, and EF were purchased from the Korean Oriental Medicine Cooperative (Seoul, Republic of Korea). Each of them was decocted in distilled water (1:10 ratio) at 95 °C for 6 h, filtered, and the process was repeated. Each filtrate was concentrated using rotary vacuum evaporation (Buchi R-215 Rotavapor) at 55 °C until the volume was reduced to approximately 1/10th of the original. They were centrifuged at 11,300× g at 4 °C for 20 min, and the supernatant was filtered and freeze-dried. Bioactive compounds (baicalein, kaempferol, luteolin, and quercetin) purchased from the Avention Research Center were used as standards.

Lyophilized herbal extracts were dissolved in methanol (0.1 g/mL) and filtered with a 0.45 μm microporous membrane. Standards such as baicalein, kaempferol, luteolin, and quercetin were also dissolved in methanol (50 μg/mL). High-performance liquid chromatography-diode-array detection (HPLC-DAD) equipped with a SunFire C18 column (4.6 mm × 75 mm, 5 µm) was used to identify the amounts of the index bioactive compounds in the herbal extracts with a flow rate of 1 mL/min, a column temperature of 30 °C, and an injection volume of 10 µL. The chromatographic conditions were optimized. The mobile phase used for analyte separation comprised a 0.1% phosphoric acid solution (A) and acetonitrile (B). The run-time was 4 min, and elution was performed in the isocratic mode with A: 40% and B: 60%. Detection was carried out at 370 nm.

An in vitro gut–brain axis model was established using a co-culture system of PC12 neuronal cells and Caco-2 intestinal epithelial cells (KCLB 21721, Korean Cell Line Bank, Seoul, Republic of Korea), as illustrated in Figure S1. PC12 cells were cultured in poly-D-lysine-coated flasks with high-glucose Dulbecco's Modified Eagle Medium (DMEM, 4.5 g/L; ThermoFisher, Waltham, MA, USA) supplemented with 10% horse serum and 5% fetal bovine serum (FBS). Caco-2 cells were maintained in high-glucose DMEM containing 10% FBS, 0.5% penicillin/streptomycin, and 1% GlutaMAX. All cultures were maintained at 37 °C in a humidified 5% CO₂ atmosphere with media replacement every 48 h.

Cell Differentiation Protocol: PC12 cells were seeded at 4 × 10⁴ cells/well in poly-D-lysine-coated 96-well plates and differentiated over 7 days using serum-reduced DMEM (1% horse serum, 1% penicillin/streptomycin) supplemented with 50 ng/mL nerve growth factor (NGF; ThermoFisher), with media refreshed every 48 h. Caco-2 cells were seeded at 2.5 × 10⁴ cells/well onto 0.4 μm polyethylene terephthalate (PET) semipermeable membrane inserts (SPLInsert™, Pocheon, South Korea) and cultured for 21 days to establish a polarized intestinal epithelial monolayer with developed microvilli and tight junctions, confirmed by transepithelial electrical resistance (TEER) measurements >300 Ω·cm².

Co-Culture and Treatment Protocol: The experimental timeline followed a standardized 23-day protocol: Days 0–3: Cell expansion and maintenance in 75 cm² culture flasks; Day 3: Caco-2 cell seeding onto collagen type-I precoated inserts (2.5 × 10⁴ cells/well); Days 3–21: Caco-2 differentiation with medium changes every 48 h; Days 5–10: PC12 neuronal differentiation with 50 ng/mL NGF treatment; Day 10: Differentiated PC12 cell seeding onto culture plates; Day 21: Co-culture establishment with 24 h equilibration period; Day 22: Sequential treatment application—first, inflammatory challenge with 10 μM Aβ_25–35 (Sigma-Aldrich, St. Louise, MO, USA) added to the basolateral chamber (PC12 side) and 10 μM LPS (E. coli O111:B4; Sigma-Aldrich) added to the apical chamber (Caco-2 side) for 24 h, immediately followed by treatment intervention with herbal extracts (10, 20, or 50 μg/mL) or bioactive compounds (2 or 5 μM); Day 23: Sample collection and analysis.

Quality Control and Validation: Cell confluence was monitored daily using phase-contrast microscopy. Caco-2 cells were used when reaching >90% confluence with visible tight junction formation. PC12 differentiation was confirmed by neurite outgrowth assessment, with cells showing neurites ≥2 times the cell body diameter considered adequately differentiated. TEER was measured for Caco-2 monolayers to confirm barrier integrity (>300 Ω·cm²) by TEER voltohmmeter (Merck, Darmstadt, Germany), before co-culture establishment. Mycoplasma contamination was routinely tested using PCR-based detection.

Experimental Groups and Controls: Each experiment included the following systematically controlled groups: (1) Normal Control: Vehicle-treated co-cultures (PBS for extracts, DMSO ≤ 0.1% v/v for compounds); (2) Control: Aβ_25–35 + LPS + vehicle; (3) Treatment Groups: Aβ_25–35 + LPS + test compounds at specified concentrations; (4) Positive Control: Aβ_25–35 + LPS + donepezil (10 μM; Sigma-Aldrich), which was selected as a positive control due to its clinical relevance as a first-line treatment for AD and its specific mechanism of action as a potent AChE inhibitor [39]. It serves as a gold-standard reference for evaluating the efficacy of our extracts against a clinically established benchmark. Preliminary validation confirmed that Aβ_25–35 and LPS administration (10 μM each) significantly reduced PC12 and Caco-2 cell viability to <50% compared to untreated controls (Figure S4A,B).

Cell viability was assessed using the 3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) assay with systematic dose-response screening. Cells were seeded in 96-well plates (n = 4 technical replicates per condition) and allowed to adhere for 24 h before treatment. For systematic preliminary dose-response screening, herbal extracts were tested at concentrations of 10, 20, 50, 100, 150, and 200 µg/mL, while bioactive compounds were tested at 0.1, 0.2, 0.5, 1, 2, and 5 µM to establish the maximum tolerated concentrations and identify the optimal therapeutic window. After 24 h of treatment, cells were incubated with MTT solution (2 mg/mL in PBS; Sigma-Aldrich) for 3 h at 37 °C in a 5% CO₂ atmosphere. Formazan crystals solubilized with DMSO were measured by absorbance at 590 nm using a microplate spectrophotometer. Cell viability was calculated as a percentage relative to vehicle-treated controls.

Final treatment concentrations of herbal extracts (10, 20, and 50 μg/mL) and bioactive compounds (2 and 5 μM) were selected using three criteria: (1) maintained cell viability ≥85% relative to vehicle controls, indicating minimal cytotoxicity; (2) demonstrated protective efficacy against Aβ23-35/LPS-induced cellular damage in preliminary studies; and (3) physiological relevance based on reported achievable plasma concentrations and established dosing ranges in phytochemical literature [40]. This systematic approach ensured assessment of dose-dependent biological responses within the non-toxic, therapeutically active window.

To assess oxidative stress and cholinergic activity, TBARS and AChE activity were measured post-treatment (n = 4). Lipid peroxidation was quantified using a TBARS assay kit (DoGenBio, Seoul, Republic of Korea), and AChE activity was determined in PC12 cells using a colorimetric AChE assay kit, following the manufacturer's protocols.

Total RNA was extracted from co-cultured cells using the TRIzol reagent (Thermo Scientific Inc., Santa Clara, CA, USA) (n = 4). RNA concentration was determined spectrophotometrically (260/280 nm). One microgram of RNA was reverse-transcribed into cDNA using the RevertAid RT Kit (Thermo Scientific Inc). The diluted cDNA was mixed with primers for the genes of interest (Table S1) and SYBR™ Green Master Mix (Affymetrix, Santa Clara, CA, USA). The mixture was run in real-time polymerase chain reaction (PCR). mRNA expression of TNF-α, IL-1β, IL-6, BDNF, and Tau was analyzed using the ΔΔCT method.

After drug treatment, proteins were extracted from co-cultured PC12/Caco-2 cells using a lysis buffer. Protein concentration was determined using the Bio-Rad Protein Assay Kit. Lysates were centrifuged in the presence of protease/phosphatase inhibitors, and 50 µg protein was resolved via sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), transferred to polyvinylidene fluoride (PVDF) membranes, and blocked with 3% bovine serum albumin (BSA). Membranes were incubated overnight at 4 °C with primary antibodies (β-actin, protein kinase B (Akt), phosphorylated (p)-Akt (Ser473), 5′ AMP-activated protein kinase (AMPKα), p-AMPKα (Thr172), glycogen synthase kinase (GSK)-3β, p-GSK-3β (Ser9)), followed by Alexa Fluor 700-conjugated secondary antibodies (1:20,000; Cell Signaling Technology, Danvers, MA, USA) [41]. Detection was performed using the Immun-Star™ WesternC™ Chemiluminescence Kit.

Caco-2 cells grown on Transwell inserts were fixed with 4% formalin after treatment, then stained using hematoxylin–eosin (H&E) and toluidine blue (Sigma-Aldrich, St. Louis, MO, USA) (n = 4), following a modified protocol based on the method elucidated by Yoshimoto et al. [42]. Stained cells were examined under an optical microscope at 100× magnification. Mitochondrial membrane potential (MMP) was assessed using the JC-1 dye (2 µM; MMP Assay Kit), following the manufacturer's instructions and the Meng et al. method [40]. After staining for 30 min in a CO₂ incubator, cells were washed and visualized using an inverted fluorescence microscope. Red and green fluorescence intensities represented polarized and depolarized mitochondria, respectively.

All experiments were performed with n = 4 technical replicates per experimental condition and repeated across multiple independent experiments to ensure reproducibility. Data are presented as mean ± standard deviation (SD). Statistical significance was determined using one-way ANOVA followed by Tukey's post hoc multiple comparison test, conducted using SPSS 16.0 (IBM) software. For multi-factorial experimental designs, two-way ANOVA was employed where appropriate. Different lowercase letters (a, b, c, d) above bars indicate statistically significant differences between groups (p < 0.05). All statistical analyses included appropriate control groups, and p-values were adjusted for multiple comparisons when necessary.