The compound library for the integrated in silico and in vitro workflow was constructed from the 14 medicinal herbs constituting JWS (Figure 1). These herbs include Acori Graminei Rhizoma, Amomi Semen, Castaneae Semen, Coicis Semen, Ephedrae Herba, Liriopis Tuber, Longanae Arillus, Massa Medicata Fermentata, Platycodi Radix, Polygalae Radix, Raphani Semen, Schisandrae Fructus, Thujae Semen, and Zizyphi Semen. Subsequently, small molecules from the constituents of each herb were systematically searched in public databases, including PubMed and TM-MC 2.0 ¹ [34]. This search process yielded 2,253 single compounds. After removing duplicate entries, the final library comprised 1,751 unique compounds. The Simplified Molecular Input Line Entry System (SMILES) strings for these compounds were obtained from PubChem ² and TM-MC 2.0 to use as input for subsequent computational analyses.

An initial in silico screening step was performed to select JWS-derived compounds with drug-like properties and predicted BBB permeability. The web-based tool SwissADME ³ was used to evaluate the 1,751 compounds curated from the JWS-derived compound library using a combined filtering approach. Two criteria were applied simultaneously: 1) compliance with Lipinski’s rule of five to assess basic oral drug-likeness and 2) predicted BBB permeability based on the BOILED-Egg model to assess CNS-targeting suitability. Compounds were selected only when they met both criteria. This filtering process excluded 1,185 compounds, leading to the selection of 566 JWS-derived compounds with potential brain-targeting properties.

First, JWS-derived 566 compounds were prepared as ligands in BIOVIA Discovery Studio Client 2025 (Dassault Systèmes, San Diego, CA, USA) according to the following protocol described in our previous study [35]. Ligand structures were imported, and ionization was performed based on pH conditions, with automatic protonation of acidic groups and deprotonation of basic groups. Tautomeric forms were generated by exhaustive enumeration up to a maximum of 10 variants, excluding carbon atoms as hydrogen donors, maintaining aromaticity patterns, and limiting tautomerization to diamide moieties. Stereoisomers were generated for unspecified stereocenters and stereobonds, excluding bridgehead atoms, retaining marked stereochemistry, and correcting valence errors. Three-dimensional (3D) coordinates were assigned, and hydrogen atoms were optimized for proper spatial orientation. As the Aβ aggregation process involves various structural forms, it is crucial to use representative structures as targets [35]. Three major structures of Aβ–1Z0Q, 2NAO, and 2BEG–were obtained from the protein data bank (PDB) for use in molecular docking and dynamics simulations. Each structure was optimized using standard protein preparation protocols: imported protein molecules underwent loop building based on SEQRES definitions for loops up to 20 residues in length, without using Looper. The chemistry at HARvard Macromolecular Mechanics (CHARMm) minimization was applied with flexible stem residues set to 0. Protonation was performed using a protein dielectric constant of 10, pH of 7.4, ionic strength of 0.145, and an energy cutoff of 0.9. In advanced settings, the CHARMm force field was used, ligands were retained, waters were excluded, and disulfide bridges were handled with the default settings. Detailed information on the Aβ structures used in the simulations is summarized in Table 1. An α-helical monomeric structure of Aβ, PDB ID: 1Z0Q, was selected to represent the monomeric state. In contrast, a β-sheet monomeric conformation, PDB ID: 2NAO, was chosen to model a state with high aggregation propensity. Additionally, the pentameric forms of Aβ derived from PDB ID: 2BEG, representing the neurotoxic oligomeric states, were included as targets for molecular docking studies. By targeting multiple structural states, we aimed to increase the probability of discovering inhibitors that can act at various stages of the aggregation process.

Molecular docking simulations were performed using the CDOCKER algorithm within the Discovery Studio 2025 software. CDOCKER is a CHARMm-based algorithm that employs a semi-flexible docking approach, which accounts for the conformational flexibility of the ligand. Key simulation parameters included generating 10 top docking poses per compound and a pose cluster radius of 0.5 Å. The docking was performed using 10 random conformations, 1,000 dynamics steps at a target temperature of 1,000 Kelvin (K) and included electrostatic interactions. Simulated annealing was performed using 2,000 heating steps to 700 K followed by 5,000 cooling steps to 300 K. Semi-flexible ligand docking of the selected small molecules was carried out with the CHARMm force field. These docking settings produced a range of energy scores, including CDOCKER Interaction Energy. The following energy values were used to evaluate and rank the docking performance of the candidate compounds: CDOCKER Energy, CDOCKER Interaction Energy and Ligand Strain Energy. The Ligand Strain Energy represents the internal energy change associated with ligand deformation upon binding and was calculated by subtracting the −CDOCKER Energy from the −CDOCKER Interaction Energy and taking the absolute value. The 2D-ligand interactions were visualized using the visualization module in Discovery Studio 2025.

Thioflavin T and synthetic Aβ₄₂ peptide were purchased from Tokyo Chemical Industry Co., Ltd. (Tokyo, Japan) and Bachem AG (Bubendorf, Switzerland), respectively. Morin was obtained from Sigma-Aldrich (St. Louis, MO, USA). Five compounds were purchased: 1-monopalmitin was obtained from Chemscene (Monmouth Junction, NJ, USA); dibutyl sebacate and pentadecanoic acid were purchased from Combi-Blocks (San Diego, CA, USA); dimethyl azelate and isobutyl octyl oxalate were procured from An-gene (Shanghai, China).

The inhibitory effects of the five compounds—1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, dimethyl azelate, and dibutyl sebacate—on Aβ₄₂ aggregation were assessed using a ThT fluorescence assay. Synthetic Aβ₄₂ peptides were initially dissolved in 0.1 M aqueous ammonia. For the assay, this peptide stock solution was adjusted to 25 µM using triple-distilled water and transferred into black 96-well plates. Compound stock solutions were formulated using dimethyl sulfoxide (DMSO) as the solvent. Each well was treated with the five compounds at final concentrations of 2.5, 25, and 250 μM. Morin at 100 µM and distilled water were used as the positive control and vehicle, respectively. Incubation of the ThT-assay mixtures was carried out at 37 °C for 48 h in a 50 mM glycine buffer at pH 8.9 containing 15 μM ThT. To evaluate the potential effect of DMSO on Aβ₄₂ aggregation, Aβ₄₂ was treated with DMSO at final concentrations corresponding to those present in each compound-treated group and compared with the distilled water-treated control. This experiment was performed in technical duplicate within a single experiment (Supplementary Figure S2).

To assess the dissociation effects, Aβ₄₂ fibrils were first prepared by incubating 25 μM Aβ₄₂ peptide in triple-distilled water at 37 °C for 24 h in black 96-well plates. The preformed Aβ₄₂ fibrils were then treated with each compound at final concentrations of 2.5, 25, and 250 μM at 37 °C for 48 h in black 96-well plates. Distilled water and crude JWS were used as the vehicle and positive control, respectively. Fluorescence intensity was recorded on a SpectraMax iD3 Multi-Mode Microplate Reader (Molecular Devices, San Jose, CA, USA), setting the excitation to 440 nm and emission to 484 nm. Both the aggregation and dissociation assays for the compound-treated conditions were performed in technical triplicate within a single experiment.

MD simulations were performed using BIOVIA Discovery Studio Client 2025 with the CHARMm force field. We selected the best-ranked docking poses from the molecular docking and used them as starting structures for the MD simulations. A simulation of the ligand-free Aβ dimer was also performed for comparison. Prior to dynamics, the ligand-free Aβ_17–42 dimer and each ligand-bound complex were solvated in Discovery Studio using the Explicit Periodic Boundary model with an orthorhombic cell and a minimum distance of 10 Å from the solute to the box boundary. Counterions were added, and Na⁺ and Cl⁻ were included at a salt concentration of 0.145 M. MD simulations were then conducted on the solvated systems using the standard dynamics cascade protocol. The simulation workflow consisted of four sequential steps: 1) minimization phase: initial geometry optimization via Steepest Descent algorithm for 100 iterations with an RMS gradient threshold of 1, followed by secondary minimization using the Adopted Basis Newton-Raphson method for 500 iterations at 0.1 RMS gradient, 2) heating phase: 50 picoseconds (ps) of simulation with a 2 femtosecond (fs) time step, initial temperature of 50 K ramped to a target of 300 K, velocity rescaling applied every 50 steps, and trajectory data recorded every 2 ps, 3) equilibration phase: 100 ps simulation at 300 K using 2 fs time step, with velocity adjustment every 50 steps, trajectory data recorded every 2 ps, and 4) production phase: 10,000 ps simulation with a 2 fs time step at a constant temperature of 300 K, carried out in NAMD. Production was run under isothermal–isobaric (NPT) ensemble conditions with a reference pressure of 1 atm, and trajectories were saved every 50 ps. Production MD runs were performed in triplicate for the ligand-free Aβ_17–42 dimer and for each ligand-bound complex using independent initial velocities.

During minimization, equilibration, and NAMD production, the implicit solvent model was not used (Generalized Born λ = 0), because solvation had already been treated with explicit periodic water and 0.145 M NaCl. Nonbonded electrostatics used an internal dielectric constant of 1. The implicit-solvent dielectric constant of 80 shown in the Discovery Studio dynamics interface is a default field for continuum-solvent calculations and was not applied. Nonbond lists were generated with a 14 Å cutoff; electrostatic and van der Waals cutoffs were 12 Å and 10 Å, respectively. The Leapfrog Verlet algorithm was used for numerical integration of equations of motion. Total energy was analyzed over 1,000–10,000 ps after the initial relaxation phase. For each replicate, the mean energy over this window was calculated first; values are reported as the mean ± SD of the three replicate means. Total-energy profiles were used as descriptive indicators of energetic stationarity and replicate reproducibility within each simulated system. Because the ligand-free and ligand-bound systems differed in molecular composition and total atom number, their absolute mean total energies were not considered directly comparable across systems and were not used to rank ligand–Aβ interactions or to infer preservation of the Aβ_17–42 dimer interface. These simulations were intended to assess short-timescale ligand behavior and early dimer-interface perturbation, not to compute equilibrium binding free energies.

The backbone root mean square deviation (RMSD) was used to quantify structural deviation of the Aβ_17–42 dimer during production molecular dynamics. Analyses were performed in BIOVIA Discovery Studio Client 2025 with the Analyze Trajectory protocol. For each replica, RMSD was computed relative to the first frame of the production trajectory, which corresponded to the final snapshot of the preceding equilibration. Each frame was superimposed onto this reference by least-squares fitting of backbone atoms (N, Cα, C, and O) before the RMSD was evaluated on the same atom set. Water, counter-ions, hydrogen atoms, and ligand atoms were excluded from both the fit and the peptide RMSD. Three complementary RMSD series were calculated for every trajectory. Whole-dimer RMSD was obtained by fitting and scoring the backbone atoms of both peptide chains together. Chain-specific RMSD was obtained by defining separate atom groups for chain A and chain B and repeating the fit and scoring on each chain independently, so that intra-chain conformational change could be distinguished from relative motion of the two chains. RMSD values were analyzed in the 1,000–10,000 ps window after the initial relaxation phase. For each replica, the mean RMSD over this window was calculated first; values are reported as the mean ± standard deviation (SD) of the three independent replica means. These analyses were used to assess short-timescale structural persistence of the dimer and early ligand-dependent changes at the chain and interface levels. They were not used to compute equilibrium binding free energies or to infer long-timescale fibril stability.

The drug-like properties of four JWS-derived compounds were assessed based on the absorption, distribution, metabolism, and excretion (ADME). The physicochemical characteristics of four natural small compounds were determined using the web-based program PreADMET ⁴ (bioinformatics and molecular design research center, Incheon, Republic of Korea) and pkCSM-pharmacokinetics ⁵ (Bio21 institute, the University of Melbourne, Australia) to predict the pharmacokinetic properties [36]. The PreADMET web-based software can predict several important properties of drugs, including logarithm of partition coefficient between n-octanol and water (LogP_o/w), plasma protein binding (PPB), Caco-2 permeability, human intestinal absorption (HIA), BBB penetration, as well as CYP2C19 and CYP2D6 inhibition. Furthermore, the pkCSM-pharmacokinetic software can predict the total clearance of compounds.

Statistical evaluations were carried out with GraphPad Prism 10.0 software. Group comparisons for the ThT assays were assessed via one-way analysis of variance (ANOVA) followed by Fisher’s LSD post hoc test. Statistical significance was defined as p < 0.05. Data from the ThT assays are presented as the mean ± standard error of the mean (SEM) of technical replicates within a single experiment. For the MD simulations, data from three independent simulation runs are presented as the mean ± standard deviation (SD), as specified in the corresponding figure legends.

A comprehensive search was conducted across public databases, including PubMed and TM-MC 2.0, to identify all 14 herbs constituting the JWS formula (Table 2). This analysis initially yielded a total of 2,253 compounds, which were reduced to 1,751 unique structures after removal of duplicate entries (Supplementary Table S1). The initial filtering step involved evaluating drug-likeness based on Lipinski’s rule of five and predicting BBB permeability using the web-based tool SwissADME. Filtering using these two criteria identified 566 candidate compounds (Supplementary Table S2). The selection process prioritized compounds exhibiting favorable oral bioavailability, including high gastrointestinal absorption predicted by the BOILED-Egg model and BBB permeability, which are key pharmacokinetic properties for CNS-targeted therapeutics. Together, these early filters identified 566 JWS-derived compounds with favorable oral bioavailability and BBB permeability.

Effective Aβ-targeting therapeutics should both inhibit the formation of Aβ aggregates and promote the dissociation of pre-formed aggregates to slow AD progression [5, 23, 35]. To evaluate the binding of small molecules to Aβ, molecular docking was performed for 566 compounds against three Aβ conformations, including [1] the α-helical monomeric Aβ structure, PDB ID 1Z0Q [2], the β-sheet monomeric Aβ structure, PDB ID 2NAO, and [3] the pentameric Aβ structure, PDB ID 2BEG (Table 1). The three Aβ structures represent distinct conformational states associated with Aβ aggregation and fibril formation. Among the successfully docked compounds, docking against the α-helical monomeric Aβ structure yielded CDOCKER energy values ranging from 39.33 to −116.36 kcal/mol among 566 compounds. In addition, the docking against β-sheet monomeric Aβ structure yielded CDOCKER energy values ranging from 34.91 to −97.28 kcal/mol, whereas docking against pentameric Aβ structure yielded values ranging from 46.80 to −927.22 kcal/mol (Supplementary Table S3). The docking results were further analyzed according to the predefined selection criteria. First, the 566 compounds were ranked according to their CDOCKER energy for each Aβ structural model, and the top 35 compounds were selected using CDOCKER energy cutoff values of −21.89 kcal/mol for 1Z0Q, −18.72 kcal/mol for 2NAO, and −27.56 kcal/mol for 2BEG. Compounds consistently ranked within the top 35 across all three Aβ structural models were retained, resulting in 13 candidates (Supplementary Table S4). These 13 compounds were subsequently evaluated based on their docking interactions with key aggregation-prone residues within the central hydrophobic region of Aβ, specifically Leu17, Val18, Phe19, Phe20, and Ala21. Compounds showing direct interactions with one or more of these residues were prioritized, resulting in the selection of five compounds—1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, dimethyl azelate, and dibutyl sebacate—for further experimental validation (Supplementary Figure S1).

To elucidate the specific binding modes, we explored the docking poses of these five compounds against all three Aβ conformations: the monomeric α-helical, the β-sheet–rich monomeric, and the pentameric forms. First, in the α-helical monomeric Aβ structure, 1-monopalmitin exhibited the lowest CDOCKER energy of −39.3 kcal/mol. It formed hydrogen bonds with Phe19 and Asp23, in addition to hydrophobic interactions that included π-alkyl contacts with His14 and Val18 (Table 3; Figure 2A). Isobutyl octyl oxalate displayed a CDOCKER energy of −24.6 kcal/mol and hydrogen bonding was observed with Phe19 and Asp23, accompanied by a hydrophobic alkyl interaction with Val18 and a hydrophobic π-alkyl interaction with His14 (Table 3; Figure 2B). Pentadecanoic acid showed a CDOCKER energy of −27.8 kcal/mol and established a hydrogen bond with His14, together with hydrophobic alkyl interactions involving Val18 (Table 3; Figure 2C). Dimethyl azelate had a CDOCKER energy of −25.1 kcal/mol and formed multiple hydrogen bonds with His14, Gln15, Phe19, Glu22, and Asp23, supplemented by hydrophobic alkyl interactions with Val18 (Table 3; Figure 2D). Dibutyl sebacate yielded a CDOCKER energy of −34.3 kcal/mol and interacted predominantly through hydrophobic alkyl interactions with Lys28 (Table 3; Figure 2E). In summary, 1-monopalmitin demonstrated the lowest CDOCKER energy toward the α-helical monomeric Aβ structure, whereas dimethyl azelate exhibited the greatest number of hydrogen interactions with multiple residues.

Second, in the β-sheet–rich monomeric Aβ structure, 1-monopalmitin showed a CDOCKER energy of −25.7 kcal/mol. It formed hydrogen bonds with Lys16, Val18, and Phe20, along with hydrophobic alkyl interactions involving Lys16, Val18, and Ala21 (Table 3; Figure 3A). Similarly, isobutyl octyl oxalate exhibited a CDOCKER energy of −22.7 kcal/mol and formed hydrogen bonds with Lys16, Val18, and Phe20, together with hydrophobic interactions involving Ala21 (Table 3; Figure 3B). Pentadecanoic acid displayed the most favorable CDOCKER energy of −30.6 kcal/mol and formed hydrophobic alkyl interactions with Lys16, Val18, Ala21, and Ile32 (Table 3; Figure 3C). Dimethyl azelate had a CDOCKER energy of −25.2 kcal/mol and formed hydrogen bonds with Lys16, Val18, and Phe20, in addition to hydrophobic interactions involving Val18 and Ala21 (Table 3; Figure 3D). Dibutyl sebacate showed a CDOCKER energy of −33.9 kcal/mol and formed hydrogen bonds with Lys16, Val18, and Ala21, along with hydrophobic alkyl interactions with Ala21 and hydrophobic π-alkyl interactions with Phe19 (Table 3; Figure 3E). Collectively, 1-monopalmitin, isobutyl octyl oxalate, and dimethyl azelate formed hydrogen bonds with the common residues Lys16, Val18, and Phe20, whereas hydrophobic interactions were predominantly observed with Ala21 across all five compounds.

Third, in the pentameric Aβ structure, 1-monopalmitin showed a CDOCKER energy of −39.4 kcal/mol. 1-monopalmitin formed hydrophobic alkyl interactions with Leu17, Val40, and Ala42, as well as hydrophobic π-alkyl interactions with Phe19 (Table 3; Figure 4A). Isobutyl octyl oxalate exhibited a CDOCKER energy of −35.6 kcal/mol and formed a hydrogen bond with Leu17 and hydrophobic alkyl interactions with Leu17 and Val40, along with hydrophobic π-alkyl interactions with Phe19 (Table 3; Figure 4B). Pentadecanoic acid had a CDOCKER energy of −43.0 kcal/mol, forming a hydrogen bond with Leu17 and hydrophobic alkyl interactions with Leu17, Val40, and Ala42 (Table 3; Figure 4C). Dimethyl azelate showed a CDOCKER energy of −38.6 kcal/mol and formed a hydrogen bond with Leu17, along with hydrophobic alkyl interactions involving Phe19, Leu17, and Val40, and π-alkyl interactions with Phe19 (Table 3; Figure 4D). Lastly, dibutyl sebacate showed the most negative CDOCKER energy of −47.6 kcal/mol and predominantly formed both hydrophobic alkyl and π-alkyl interactions with Phe19 and Val40 (Table 3; Figure 4E). Collectively, hydrophobic interactions were predominantly observed with residues Leu17, Phe19, Val40, and Ala42. The pentameric form of Aβ exhibited fewer hydrogen bonds compared to the monomeric structures, whereas hydrophobic interactions became more prominent.

Overall, the selected compounds showed stronger binding to the pentameric Aβ structure than to the monomeric forms, with hydrophobic interactions predominantly observed at aggregation-associated residues. Hydrophobic interactions were predominantly observed in the pentameric structure, whereas hydrogen bonding was more frequently observed in the monomeric structures.

To validate the inhibitory effects of the five selected JWS-derived compounds on Aβ₄₂ aggregation predicted by in silico screening, ThT fluorescence assays were performed. Aβ₄₂ was incubated in the presence or absence of 1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, dimethyl azelate, and dibutyl sebacate at concentrations of 2.5, 25, and 250 μM, with morin at 100 μM as a positive control (Figure 5A). Among the five selected JWS-derived compounds, four compounds—1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, and dimethyl azelate—inhibited Aβ₄₂ aggregation, with 1-monopalmitin, isobutyl octyl oxalate, and dimethyl azelate showing concentration-dependent inhibition and pentadecanoic acid exhibiting significant inhibition only at 25 μM; dibutyl sebacate showed no significant inhibitory effect at any concentration tested. Morin reduced ThT fluorescence to 51.61% of the vehicle. Among the four selected compounds, 1-monopalmitin exhibited the strongest inhibitory activity against Aβ₄₂ aggregation, reducing ThT fluorescence to 67.08%, 69.63%, and 86.58% at 250, 25, and 2.5 μM, respectively. Treatment with isobutyl octyl oxalate reduced ThT fluorescence to 73.03%, 81.81%, and 88.40% at 250, 25, and 2.5 μM, respectively. Meanwhile, pentadecanoic acid significantly inhibited Aβ₄₂ aggregation only at 25 μM, reducing ThT fluorescence to 85.63%. Dimethyl azelate significantly reduced ThT fluorescence to 80.68% and 85.40% at 250 and 25 μM, respectively, without a significant effect at 2.5 μM. In addition, the crude JWS extract exhibited more potent inhibitory effects on Aβ₄₂ aggregation than any individual compound, reducing ThT fluorescence to 40.14% and 33.07% at 400 and 600 μg/mL, respectively. These results suggest that multiple constituents within the JWS extract may collectively contribute to its stronger inhibitory activity against Aβ₄₂ aggregation.

To further examine whether the selected compounds could dissociate preformed Aβ₄₂ aggregates, pre-aggregated Aβ₄₂ was separately incubated with each compound at 2.5, 25, and 250 μM (Figure 5B). Among the five compounds tested, only isobutyl octyl oxalate and pentadecanoic acid significantly dissociated preformed Aβ₄₂ aggregates: isobutyl octyl oxalate reduced ThT fluorescence to 60.10% of the vehicle at 25 μM, while pentadecanoic acid showed a stronger dissociation effect at 250 μM, reducing ThT fluorescence to 25.59%. The crude JWS extract also reduced ThT fluorescence of preformed Aβ₄₂ aggregates to 46.16% at 600 μg/mL. In contrast, 1-monopalmitin, dimethyl azelate, and dibutyl sebacate did not significantly dissociate preformed Aβ₄₂ aggregates at any tested concentration. Collectively, these results identify isobutyl octyl oxalate and pentadecanoic acid as JWS-derived compounds with Aβ-disaggregating activity, with pentadecanoic acid exhibiting particularly potent dissociation activity that exceeded that of the JWS extract at the highest concentration tested.

To examine the short-timescale interactions of the four selected compounds with Aβ, we performed 10,000-ps MD simulations of the Aβ_17-42 dimer (PDB ID: 2BEG) in triplicate. The simulations were performed on the ligand-free Aβ_17-42 dimer as the control and in complex with 1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, and dimethyl azelate. Mean total energies from three independent simulations over 1,000–10,000 ps are summarized in Figure 6A, and the time-resolved triplicate total-energy data are provided in Supplementary Table S5. Time-resolved energy profiles are shown in Figures 6B–F, Supplementary Figures S4A–E, S5A–E. Representative structures at 50–10,000 ps are shown in Figures 6G–K, Supplementary Figures S4F–J, S5F–J. Across all five systems, total-energy profiles fluctuated around relatively stationary mean values without substantial progressive drift during the analyzed interval. Because the ligand-free and ligand-bound systems differed in molecular composition and total atom number, absolute mean total energies were not compared across systems and were not interpreted as measures of ligand-binding affinity or Aβ_17–42 dimer stability. Accordingly, total-energy analysis was used only as a descriptive indicator of energetic stationarity and replicate reproducibility within each system. Accordingly, differences in mean total energy among the systems were not used to rank ligand–Aβ interactions or to validate the molecular docking results. Instead, the MD trajectories were examined qualitatively to assess short-timescale ligand behavior and ligand-associated structural changes at the Aβ_17–42 dimer interface.

MD simulations were performed to investigate the conformational dynamics of the dimeric Aβ peptide (PDB ID: 2BEG) over a 10,000 ps simulation in the presence and absence of four small-molecule compounds derived from JWS (Figure 6G). Representative structural snapshots of Aβ_17-42 dimer were captured at 50 ps, 2,000 ps, 4,000 ps, 6,000 ps, 8,000 ps, and 10,000 ps to illustrate time-dependent structural evolution. In the ligand-free Aβ_17-42 dimer, the two peptide chains remained closely associated throughout the representative snapshots, although conformational rearrangements were observed during the simulation (Figure 6G). The Aβ_17-42 dimer—1-monopalmitin complex showed progressive separation at the dimer interface throughout the simulation, during which 1-monopalmitin frequently changed its position and orientation, with clear interchain separation observed after 8,000 ps (Figure 6H). The Aβ_17-42 dimer complexed with isobutyl octyl oxalate displayed more pronounced structural rearrangements, including ligand-induced bending or twisting of chains, expansion of coil/turn segments, dissociation of Aβ dimers from 2,000 ps onward, and straightening of both Aβ peptides, particularly at 10,000 ps (Figure 6I). Similarly, the Aβ_17-42 dimer complexed with pentadecanoic acid showed conformational rearrangements, including α-helix formation in residues Ala30, Ile31, and Ile32 at 2,000 ps time point and partial separation of the two Aβ peptides starting at 6,000 ps onward (Figure 6J). Finally, the Aβ_17-42 complex with dimethyl azelate exhibited structural rearrangements, including the opening of the Aβ dimers at 4,000 ps, followed by partial separations between the two peptide chains at 6,000 ps and 10,000 ps (Figure 6K). Collectively, the ligand-containing trajectories showed varying degrees of dimer-interface rearrangement, partial chain separation, and changes in ligand position and orientation. These observations suggest transient ligand–Aβ interactions associated with early perturbation of the Aβ_17–42 dimer interface rather than persistent stabilization of ligand-bound complexes. Given the limited 10,000-ps simulation period, these observations should be interpreted as short-timescale trajectory behavior and not as evidence of equilibrium binding or a complete fibril-disassembly pathway.

To distinguish intra-chain conformational change from relative motion of the two peptide chains, backbone RMSD was recalculated after least-squares superposition onto the first production frame of each replica (Figure 7). Whole-dimer RMSD and chain-specific RMSD were obtained separately for the ligand-free Aβ_17–42 dimer and for complexes with 1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, and dimethyl azelate. In all five systems, fitted RMSD increased from 0 Å during the first ∼1,000 ps and then fluctuated without returning to the starting 2BEG geometry (Figures 7B–D). Mean backbone RMSD over the 1,000–10,000 ps window is summarized in Figure 7A. The ligand-free dimer yielded a mean dimer RMSD of 11.91 ± 2.16 Å, with chain A and chain B values of 10.24 ± 1.95 Å and 8.32 ± 2.29 Å, respectively. Mean dimer RMSD was 9.87 ± 2.05 Å for 1-monopalmitin, 12.39 ± 1.87 Å for isobutyl octyl oxalate, 9.81 ± 1.73 Å for pentadecanoic acid, and 8.98 ± 1.45 Å for dimethyl azelate. Per-chain means remained in a similar range and were generally comparable to or lower than the corresponding dimer values, except for dimethyl azelate, in which one replica produced higher chain-specific RMSD and therefore larger SDs for chains A and B (10.16 ± 4.82 Å and 10.60 ± 3.50 Å). Differences among the four ligand-bound complexes and the ligand-free dimer were smaller than replica-to-replica variation. Accordingly, the RMSD profiles were not interpreted as compound-specific stabilization of a 2BEG-like dimer, nor as evidence that the ligands remain in a single docked pose throughout the trajectory. Instead, the fitted RMSD data indicate that the Aβ_17–42 dimer undergoes conformational rearrangement on the 10,000 ps timescale in both the absence and the presence of the candidate compounds. Mean dimer RMSD values for the four complexes (8.98–12.39 Å) were similar to that of the ligand-free dimer (11.91 ± 2.16 Å) and did not indicate a larger deviation from the starting geometry in the ligand-bound systems. Representative snapshots showed interface rearrangement and partial chain separation in some ligand-bound trajectories (Figures 6H–K); these visual changes were not mirrored by a systematic increase in fitted RMSD relative to the ligand-free dimer control (Figure 7).

ADME analysis plays a critical role in drug discovery and development for AD, as early-stage evaluation of pharmacokinetic properties is essential to reduce the risk of late-stage failure, particularly for compounds requiring effective brain exposure [37]. In this regard, in silico tools, such as PreADMET and pkCSM-pharmacokinetics, provide a rapid and reliable alternative to experimental approaches [36, 38]. Thus, we performed detailed ADME evaluation for the four JWS-derived candidate compounds using PreADMET. Additionally, their BBB permeability was verified using pkCSM-pharmacokinetics. To predict the pharmacokinetic properties for 1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, and dimethyl azelate, we assessed key parameters including logP_o/w, PPB, Caco-2 permeability, HIA, inhibition of CYP2C19 and CYP2D6 enzymes, total clearance, and BBB permeability (Table 4). The logP_o/w value was 4.36 for 1-monopalmitin, 3.08 for isobutyl octyl oxalate, 5.16 for pentadecanoic acid, and 2.06 for dimethyl azelate, all of which indicate lipophilic characteristics favorable for membrane permeability. In addition, all compounds exhibited a PPB value of 100%, suggesting strong plasma protein binding. A PPB value of 100% indicates reduced free drug fraction in circulation, which may influence BBB permeability and pharmacological efficacy [39]. In terms of intestinal absorption-related properties, the predicted values of Caco-2 permeability were 29.86 nm/s for 1-monopalmitin, 40.92 nm/s for isobutyl octyl oxalate, 25.04 nm/s for pentadecanoic acid, and 21.08 nm/s for dimethyl azelate. The HIA values were 91.41% for 1-monopalmitin, 96.06% for isobutyl octyl oxalate, 98.11% for pentadecanoic acid, and 93.42% for dimethyl azelate, indicating efficient intestinal absorption and suitability for oral administration. Regarding metabolic interactions, 1-monopalmitin, isobutyl octyl oxalate, pentadecanoic acid, and dimethyl azelate were predicted to inhibit CYP2C19, the whereas four JWS-derived candidate compounds were not predicted to inhibit CYP2D6. The predicted inhibition of CYP2C19 may indicate a potential for drug–drug interactions, which warrants further experimental validation. The predicted total clearance was 1.787 log mL/min/kg for 1-monopalmitin, 1.463 log mL/min/kg for isobutyl octyl oxalate, 1.528 log mL/min/kg for pentadecanoic acid, and 1.464 log mL/min/kg for dimethyl azelate. Moreover, the values of BBB penetration were 6.48 for 1-monopalmitin, 0.28 for isobutyl octyl oxalate, 6.64 for pentadecanoic acid, and 0.99 for dimethyl azelate. Taken together, these in silico predictions suggest that four JWS-derived compounds exhibit suitable absorption and metabolic profiles, whereas BBB permeability differs among them, with higher values observed for 1-monopalmitin and pentadecanoic acid compared with isobutyl octyl oxalate and dimethyl azelate.