The search term “Lonicerae Japonicae Flos” was employed to access the active constituents within the SymMap database (oral bioavailability ≥30%) (19), the BATMAN-TCM online bioinformatics analysis tool (score cutoff = 20, adjusted p-value <0.05) (20), the Traditional Chinese Medicine Systems Pharmacology Database and Analysis Platform (TCMSP) database (oral bioavailability ≥30%, drug-likeness ≥0.18) (17), and the Encyclopedia of Traditional Chinese Medicine (ETCM) database (drug-likeness weight ≥0.5) (21).

The active components were screened for their targets using four databases, namely SymMap [false discovery rate (Benjamini–Hochberg) <0.05], BATMAN-TCM (score cutoff = 20, adjusted p-value <0.05), TCMSP (random forest score ≥0.7, support vector machine score ≥0.8), and ETCM (similarity value >0.8), with a filter for Homo sapiens species. The target names of the proteins under investigation were consistently standardized using the RCSB Protein Data Bank database.¹ Duplicate targets were eliminated, and related targets were retained for subsequent analysis.

Four disease databases included the MalaCards database (22), the GeneCards database,² the DisGeNET database (23), and the Online Mendelian Inheritance in Man (OMIM) database³ were utilized to screen potential targets related to AD using the keyword “Alzheimer’s disease.” To enhance the correlation with AD, we filtered targets with a relevance score ≥10 in the MalaCards database, score ≥10 in the GeneCards database, score ≥0.05 in the DisGeNET database, and selected all targets in the OMIN database. Redundant targets were removed, while relevant targets were preserved for subsequent analysis.

To obtain the relationship between AD-related and LJF-related targets, we conducted an intersection analysis of the respective target lists using an online interactive tool that facilitates comparison through Venn diagrams.⁴ We retained the intersection targets for subsequent analysis.

The Database for Annotation, Visualization and Integrated Discovery (DAVID)⁵ was utilized to distinguish and enrich the biological attributes, including biological processes, cellular components, molecular functions, and pathways. A false discovery rate (FDR) <0.05 significance threshold was established for enrichment analysis. The top 20 terms were selected for visualization according to their frequency count.

Intersection targets were chosen and input into the STRING database ⁶ to retrieve the protein interaction network with a confidence score of 0.4 or higher. The results were input into the Cytoscape 3.7.1 software to establish and analyze the interaction network (24). In order to identify potential core targets, the cytoHubba plugin was utilized to evaluate and prioritize nodes (proteins) based on network characteristics with seven scoring algorithms, including maximal clique centrality (MCC), density of maximum neighborhood component (DMNC), maximum neighborhood component (MNC), edge percolated component (EPC), degree, closeness, and betweenness (25). Ultimately, this study identified putative core targets that exhibited concordance across three or more topological algorithms.

The postulated core targets were validated differential expression in GEO datasets. To assess expression levels of putative core targets between AD cases and normal controls, we conducted differential expression analyses through the comprehensive AlzData database⁷ with data from the hippocampal datasets, including GSE28146 (expression data from early-stage AD), GSE29378 (expression data from late-stage AD), GSE36980 (expression data from post mortem AD brains), GSE48350 (expression data from post mortem AD brains), and GSE5281 (undistinguished stages). The data were shown as mean ± standard deviation, with a p-value of <0.05 indicating significant differences. Differentially expressed core targets were utilized for further validation.

To further validate the relativity with AD, correlation analysis was performed on putative core target genes and major causative genes, including amyloid precursor protein (APP), microtubule-associated protein tau (MAPT), and presenilin2 (PSEN2) through the gene expression profiling interactive analysis (GEPIA) (26). Putative core target genes were imported into the search words, and brain-hippocampus was selected from GTEx expression datasets. We selected genes that exhibited a significant association with major causative genes of AD for further validation.

To evaluate the predictive efficacy of core targets, we performed a ROC analysis, calculating the area under the receiver operating characteristic (ROC) curve (AUC) as a metric for their predictive capability. Therefore, we applied ROC curve analysis with hippocampus datasets, including GSE28146, GSE29378, GSE36980, GSE48350, and GSE5281, to evaluate these core targets’ predictive utility.

Four databases, including SymMap, BATMAN-TCM, TCMSP, and ETCM, were used to investigate the essential bioactive constituents of core targets. Duplicate ingredients were removed, while pertinent ingredients were retained. To elucidate the interaction between core targets and essential ingredients, we intersected ingredients of core targets mapped and LJF using an online tool to compare lists with Venn diagrams. Intersection ingredients with oral bioavailability (OB) ≥0.3, drug-likeness (DL) ≥0.18, and blood-brain barrier (BBB) ≥−0.3 were retained for subsequent analysis.

The AutoDock Vina software evaluated the molecular binding affinity between bioactive constituents and their corresponding core targets. The two-dimensional molecular structures of bioactive compounds in LJF were acquired from the PubChem database.⁸ The crystallographic structures of core target proteins were acquired from the RCSB Protein Data Bank and SWISS-MODEL.⁹ PyMOL ¹⁰ and AutoDock Vina were employed to structurally manipulate the proteins, involving eliminating water molecules and heteroatoms and incorporating charges and hydrogen atoms. The resulting profiles were then saved in the PDBQT format for binding studies. Subsequently, docking analyses were conducted utilizing AutoDock Vina, and the resulting docking conformations were visualized using PyMOL and Discovery Studio 2020 software. The results were presented through two-dimensional (2D) and three-dimensional (3D) graphical representations. Affinity was utilized to assess the binding efficacy of the primary active compounds with their respective targets. The affinity values of −4.25 kcal/mol, −5.0 kcal/mol, and −7.0 kcal/mol suggest varying degrees of binding strength between the compound and the target, with lower values indicating stronger binding and higher values indicating weaker binding interactions (24).

The molecular dynamics simulations of protein-component complexes derived from molecular docking were conducted using GROMACS (version 2020.6).¹¹ The receptor proteins topology files were created utilizing the AMBER99SB force field, while the ligands component topology files were generated by applying the sobtop script with the AMBER force field. The system was neutralized with NaCl counterions. Before conducting the dynamics simulation, the complex underwent minimization for 10,000 steps and equilibration by executing canonical ensemble and constant-pressure, constant-temperature simulations for 100 picoseconds. The Coulomb force intercept and van der Waals radius intercept measured 1.4 nm. The system was subsequently equilibrated with the regular and isothermal isobaric system, followed by molecular dynamics simulations conducted for 10 nanoseconds under room temperature and pressure conditions. Ultimately, the root mean square deviation (RMSD) curves of protein-component complexes were analyzed to assess binding stability.

Active constituents of LJF were identified through a comprehensive review of relevant literature from databases, including SymMap, BATMAN-TCM, TCMSP, and ETCM. A total of 132 active ingredients with an OB score ≥0.3 were identified from the SymMap database, 72 active ingredients with a score exceeding 20 from the BATMAN-TCM database, 23 active ingredients with an OB ≥0.3 and DL ≥0.18 from the TCMSP database, and 47 active constituents from the ETCM database were found to be effective (Figure 1A). Following the screening criteria, 212 species were identified after eliminating duplicates. The SymMap, BATMAN-TCM, TCMSP, and ETCM databases were utilized to identify potential targets of efficacious active compounds. The UniProt database was utilized to standardize the official names of all retrieved proteins. All duplication proteins were removed after merging to screen 566 potential targets (Figure 1B). MalaCards, GeneCards, DisGeNET, and OMIM databases were queried with the subject heading “Alzheimer’s disease” to screen potential targets. The MalaCards database yielded 112 related targets with a score greater than 10, the GeneCards database identified 338 related targets with gifts exceeding 30 and relevance surpassing 25, the DisGeNET database revealed 220 related targets with a score exceeding 0.1, and the OMIM database provided information on 545 related targets (Figure 1C). Four sources of targets were merged and duplication was removed to yield 1,020 AD-related targets. A total of 112 intersection targets were identified between LJF and AD by constructing Venn diagrams, as shown in Figure 1D.

The DAVID database was utilized to perform GO and KEGG enrichment analyses on 112 intersection targets. With FDR <0.05 as screening conditions, 923 GO entries (biological process 694 entries, cell composition 103 entries, and molecular function 126 entries) and 175 KEGG pathways entries were acquired. According to counts and the FDR value, the top 20 items in the three biological attributes were selected to draw bubble diagrams, as shown in Figures 2A–C. In terms of biological processes, intersection targets are primarily involved in positive regulation of gene expression (GO: 0010628), positive regulation of transcription from RNA polymerase II promoter (GO: 0045944), signal transduction (GO: 0007165), negative regulation of apoptotic process (GO: 0043066), positive regulation of apoptotic process (GO: 0043065), etc. In terms of cellular components, intersection targets are primarily involved in the cytoplasm (GO: 0005737), cytosol (GO: 0005829), plasma membrane (GO: 0005886), nucleus (GO: 0005634), membrane (GO: 0016020), etc. In terms of molecular functions, intersection targets are primarily involved in protein binding (GO: 0005515), identical protein binding (GO: 0042802), enzyme binding (GO: 0019899), ATP binding (GO: 0005524), protein homodimerization activity (GO: 0042803), etc. For the KEGG pathway enrichment, the re-screening condition is that the number of enriched targets is ≥20. The bubble chart of the top 20 pathways is ranked according to the counts, and FDR value is shown in Figure 2D. Intersection targets are primarily involved in pathways in cancer (hsa05200), Alzheimer disease (hsa05010), pathways of neurodegeneration—multiple diseases (hsa05022), lipid and atherosclerosis (hsa05417), fluid shear stress and atherosclerosis (hsa05418), etc.

To improve comprehension of targets interactions, the STRING online tool was utilized to analyze the protein–protein interaction (PPI) network involving intersection targets. The Cytoscape software presented 112 nodes (targets) and 2,069 edges (interactions), as shown in Figure 3A. We used the MCC, MNC, DMNC, EPC, degree, closeness, and betweenness algorithms in the cytoHubba plug-in of the Cytoscape software to select potential core targets and the top 30 targets of each algorithm for intersections, as shown in Figure 3B. As shown in Figure 3C, 32 putative core targets (TP53, TNF, TLR4, TGFB1, PTGS2, PTEN, PPARG, NFKB1, MTOR, MMP9, MAPK8, INS, IL6, IL1B, IL10, IFNG, ESR1, EGFR, CXCL8, CTNNB1, CCL2, CASP3, BCL2, ALB, AKT1, GSK3A, ERBB2, IL1A, ICAM1, NFE2L2, CAV1, and APP) shared more than three topological methods were obtained.

To verify the clinical implications of 32 putative core targets, the brain transcriptome datasets of GSE28146, GSE29378, GSE36980, GSE48350, and GSE5281, which examined the expression of human hippocampus in normal control (N = 76) and AD (N = 74) patients, were used to assess the expression changes in brain hippocampus tissue of AD patients. The result showed that MAPK8 significantly down-regulated, CTNNB1, NFKB1, EGFR, CXCL8, CCL2, BCL2, and NFE2L2 were significantly up-regulated in AD patients compared to controls (Figure 4). However, there was no significant difference in TP53, TNF, TLR4, TGFB1, PTGS2, PTEN, PPARG, MTOR, MMP9, INS, IL6, IL1B, IL10, IFNG, ESR1, CASP3, ALB, AKT1, GSK3A, ERBB2, IL1A, ICAM1, CAV1, and APP levels between control and AD groups in the hippocampus (data not shown). These results showed that only core targets of MAPK8, CTNNB1, NFKB1, EGFR, CXCL8, CCL2, BCL2, and NFE2L2 might play a significant role in AD progression.

To verify the correlation between core targets screened by differential expression and the main causative genes APP, MAPT and PSEN2, correlation analysis was performed using GEPIA. The results displayed that only core targets of MAPK8, CTNNB1, NFKB1, EGFR, BCL2, and NFE2L2 were significantly related to APP, MAPT, and PSEN2 (Figures 5–7). It is well-established that six core targets of MAPK8, CTNNB1, NFKB1, EGFR, BCL2, and NFE2L2 were the most prominent factors in treating LJF against AD.

We further utilize gene expression datasets of the human hippocampus (GSE28146, GSE29378, GSE36980, GSE48350, and GSE5281) to establish ROC curves to validate six core targets of MAPK8, CTNNB1, NFKB1, EGFR, BCL2, and NFE2L2 related to AD in diagnosis, as shown in Figure 8. The area under the ROC curve, commonly denoted as AUC, serves as a metric for evaluating the overall effectiveness of a predictive model. As the AUC score approaches 1, the diagnostic performance is better. The results showed that core targets of MAPK8 (AUC = 0.614), CTNNB1 (AUC = 0.606), NFKB1 (AUC = 0.635), EGFR (AUC = 0.649), BCL2 (AUC = 0.663), and NFE2L2 (AUC = 0.628) have good predictive ability. These results indicated that six core targets of MAPK8, CTNNB1, NFKB1, EGFR, BCL2, and NFE2L2 are associated with AD, indicating good specificity and sensitivity for diagnosing patients, as shown in Table 1. Accordingly, we speculated that MAPK8, CTNNB1, NFKB1, EGFR, BCL2, and NFE2L2 mediated the therapeutic effects of LJF against AD. But still we need to admit that the ROC analysis results for the predictive ability of the identified core targets show moderate values of AUC, it is necessary to further validate in the external and cross datasets.

Which chemicals in LJF can target the identified core targets? To address this question, SymMap, BATMAN-TCM, TCMSP, and ETCM databases were utilized to screen for potential effective compounds of core targets of MAPK8, CTNNB1, NFKB1, EGFR, BCL2, and NFE2L2 (Figure 9A). In order to clarify the target-active compounds mapping relationship, we intersected the 212 LJF components and 122 core target components with an online interactive tool for comparing lists with Venn’s diagrams (Figure 9B). A total of 22 putative components were obtained (Table 2). After filtration with OB ≥0.3, DL ≥0.18, and BBB≥−0.3, only six effective active ingredients, including lignan (OB = 43.32, BBB = −0.16, DL = 0.65), β-carotene (OB = 37.18, BBB = 1.52, DL = 0.58), β-sitosterol (OB = 36.91, BBB = 0.99, DL = 0.75), hederagenin (OB = 36.91, BBB = 0.96, DL = 0.75), berberine (OB = 36.86, BBB = 0.57, DL = 0.78), and baicalein (OB = 33.52, BBB = −0.05, DL = 0.21), were screened, and the structures of which were shown in Figure 9C. Accordingly, we speculated that lignan, β-carotene, β-sitosterol, hederagenin, berberine, and baicalein mediated the preventive and therapeutic properties of LJF against AD. Intriguingly, the core targets of MAPK8 and NFE2L2 were not mapped by any effective active ingredients (Figure 9D).

To further validate the binding affinity of effective components (lignan, β-carotene, β-sitosterol, hederagenin, berberine, and baicalein) with core targets (CTNNB1, NFKB1, EGFR, and BCL2), molecular docking analysis was conducted utilizing AutoDock Vina, followed by visualization of the results using a visualization tool. The lower binding energy observed suggests a more favorable docking result. The binding energy (kcal/mol) and amino acid residues associated with the interaction between active ingredients and core targets were displayed in Table 3. BCL2 exhibited binding affinities of −8.1, −7.57, −7.14, and −6.13 kcal/mol towards berberine, β-sitosterol, β-carotene, and baicalein, respectively. EGFR demonstrated binding affinities of −6.59, −6.23, and −4.13 kcal/mol towards baicalein, berberine, and lignan, respectively. NFKB1 displayed binding affinities of −5.43 and −5.32 kcal/mol towards hederagenin and berberine, respectively. CTNNB1 exhibited a binding affinity of −5.85 kcal/mol towards β-carotene. Overall, lignan, β-carotene, β-sitosterol, hederagenin, berberine, and baicalein exhibit binding solid affinity towards their targets. The conformation with the lowest absolute binding energy values for each target was visualized using PyMoL and Discovery Studio 2020 software, as shown in Figure 10. Generally, lower free energy indicates a more stable structure. If a higher free energy conformation has been demonstrated to be effective binding for a ligand-receptor complex, representative conformations of lower free energies are not shown.

Molecular dynamics simulation offers more valuable insights into the stability of target-compound complexes. Molecular dynamics simulations were performed to assess binding affinities of BCL2/baicalein, EGFR/lignan, NFKB1/baicalein, and CTNNB1/β-carotene complexes. Generally, RMSD analysis is crucial for assessing the stability of targets and compounds, and lower RMSD values suggest more excellent stability in the target-compound complex. Figure 11 illustrates the fluctuation and stabilization of RMSD values for various protein-ligand complexes over time. Specifically, the RMSD of the BCL2/baicalein complex stabilized after 1,500 ps, while the EGFR/lignan complex reached stability at 5,000 ps. The NFKB1/baicalein complex showed stabilization between 4,000–8,000 ps, and the CTNNB1/β-carotene complex exhibited fluctuation before stabilizing at 7,000 ps. The RMSD values suggest that the active pockets of small molecules and proteins are stable. These findings indicate that the protein’s shape remains stable when the small molecule ligand binds to it, suggesting a strong interaction. These findings provide strong evidence in support of the accuracy of the molecular docking results.

Molecular docking was performed to assess interactions between core targets (MAPK8 and NFE2L2) and active compounds, including lignan, β-carotene, β-sitosterol, hederagenin, berberine, and baicalein. The docking simulations showed strong binding affinities, with free energies ranging from −4.33 to −7.84 kcal/mol for MAPK8 and −3.40 to −7.11 kcal/mol for NFE2L2 (Figure 12A). The interactions between compounds and targets, characterized by binding energy scores <−7 kcal/mol (strong binding), were further examined using PyMoL and Discovery Studio 2020 software visualization techniques. Binding affinities were attributed to conventional hydrogen bond interactions with Met-111 and Gln-117 residues, Pi-donor hydrogen bond interactions with Asp-112 residue, Pi-Sigma interaction with Val-158 residue, as well as Pi-alky/alky interactions with Ile-32, Val-40, Ala-53, Lys-55, Met-108, Leu-110, Val-158, and Leu-168 residues of MAPK8 (Figures 12B,C). Binding affinities were attributed to Pi-sigma interaction with Phe-37 residue, Pi-alky/alky interactions with Val-32, Val-36, Arg-42, Lys-516, Leu-519, and Val-523 residues of NFE2L2 (Figure 12D). Molecular dynamics simulation provided insight into that the RMSD curves of the MAPK8/β-carotene complex stabilized after 1,500 ps, the RMSD curves of the MAPK8/berberine complex stabilized after 1,000 ps, and the RMSD curves of NFE2L2/hederagenin complex stabilized after 7,000 ps (Figure 12E). The findings of this study suggest that β-carotene and berberine exhibit significant binding affinity with MAPK8, while hederagenin demonstrates a binding solid affinity with NFE2L.