All the ingredients of bilberry were collected from the FooDB database (https://foodb.ca/, Version 1.0). The basic physical and chemical properties were calculated using ChemDes (28) and PyBioMed (29) toolkits, which were widely used in the drug design and enabled quick calculation of a variety of molecular descriptors. The properties used here included molecular weight (MW), number of hydrogen bond receptors (nHA), number of hydrogen bond donors (nHD), number of rotatable bonds (nRot), number of rings (nRing), and formal charge (fChar).

The ADMET prediction was conducted using ADMETlab (Version 1.0 and 2.0) (20, 30), which was the famous platform for early ADMET evaluation in drug development. The ADMET indicators used here included LogP, hERG, H-HT, Ames, ROA, Carcinogenicity, Respiratory, Non-Genotoxic_Carcinogenicity, and Genotoxic_Carcinogenicity_Mutagenicity. A detailed explanation of these indicators can be found at https://admetmesh.scbdd.com/explanation/index. By using KNIME software and Python programming, the rules for filtering ingredients are set as shown in Supplementary Figure S1. By sorting the calculated ADMET scores, the best ones will be listed at the top.

To construct a reasonable pipeline for the target prediction, we first reviewed publications and found that Aβ (1–42) (UniProt ID: P05067), AChE (P22303), and caspase-3 (UniProt ID: P42574) could be three important targets though they act as different roles in different Alzheimer's hypotheses. Then, we tried to search for target prediction tools driven by advanced artificial intelligence. SEA (31), SwissTargetPrediction (32), TargetNet (21), PPB2 (33), PharmMapper (34, 35), and SuperPred (36, 37) were employed to perform the target prediction for ingredients refined after the ADMET filtering, since this step was a rough prediction, and the scoring standards varied between different tools. At the same time, more possibilities were expected to be seen. We set a relatively loose threshold, that is, the first 20 hits were retained and then forwarded to analysis.

Ma-3-gal-Cl (cat. no. IP-0246) was purchased from Shanghai Tauto Biotechnology Co., Ltd. (Shanghai, China). Fetal bovine serum (FBS) (cat. no. 10270-106), Dulbecco's Modified Eagle Media: Nutrient Mixture F-12 (DMEM/F12) (1:1) (cat. no. C11330500BT), and 0.25% trypsin (cat. no. 25200-056) were purchased from Gibco (CA, USA). Aβ (1–42) (cat. no. 107761-42-2) for atomic force microscopy was purchased from American Peptide Company, Inc. (California, American). Active oxygen detection kit (cat. no. S0033S) and mitochondrial membrane potential detection kit 5,5′,6,6′-Tetrachloro-1,1′,3,3′-tetraethyl-imidacarbocyanine (JC-1) (cat. no. C2006) were obtained from Beyotime Biological Reagent Co., Ltd (Shanghai, China). Coumarin-3-carboxylic acid (CCA) (cat. no. C85603), ascorbic acid (AA) (cat. no. A800296), 3-(4,5-Dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) (cat. no. M2003), CuSO₄ (cat. no. 209198), 30% H₂O₂ (cat. no. 10011208), and other reagents were purchased from Sigma-Aldrich Co. (St Louis, MO, USA). Cleaved caspase-3 (1: 1000, cat. no. ARG57512) and P-p38 (1: 500, cat. no. ARG51850) were obtained from Arigo (Hsinchu City, Taiwan, China), and p38 (1: 1000, cat. no. #9212) was obtained from Cell Signaling Technology (Danvers, MA, USA). The Ma-3-gal-Cl was first dissolved in DMSO (2 mmol/L). Then, it will be diluted to different concentrations by cell culture medium of DMEM/F12 (1:1) in cell-related experiments and ultrapure water in other experiments for further use. The Aβ (1–42) powder was first dissolved in 20 mmol/L NaOH to get the concentration of 500 μmol/L and then diluted to different concentrations by DMEM/F12 in cell-related experiments and PBS (pH = 7.4) in other experiments for further use. CCA powder was dissolved in PBS (20 mmol/L, pH = 9.0) and then adjust the pH to 7.4 with KH₂PO₄ solution, and the CCA concentration was calculated as 5 mmol/L. Thioflavin (ThT) was prepared into a concentration of 5 μmol/L by PBS (10 mmol/L, pH = 7.4).

The undifferentiated human neuroblastoma cell line (SH-SY5Y) is a commonly used neuronal cell model for Alzheimer's disease (38–40). Undifferentiated SH-SY5Y was purchased from USA ATCC Company (Washington, DC, USA). SH-SY5Y cells were cultured in DMEM/F12 (1:1) basic medium containing 10% FBS and 1% penicillin and streptomycin mixture in a cell incubator under 5% CO₂ at 37°C. The well-cultured cells were transferred to a sterile 96-well plate with approximately 1 × 10⁴ cells/well. Aβ (1–42), Aβ (1–42)/Cu/AA, or Ma-3-gal-Cl solutions were mixed and incubated with the SH-SY5Y cells for 24 h. The viability of the SH-SY5Y cells exposed to each solution was determined with an MTT assay.

Coumarin-3-carboxylic acid was used as a fluorescence probe for OH· determination (41) with a Hitachi F-4600 spectrofluorometer from Hitachi High-Tech Corporation (Tokyo, Japan). CCA fluorescence was recorded at 540 nm, with an excitation wavelength of 390 nm. The widths of the entrance and exit slits were both 5 nm. The OH·amounts in the solutions of Aβ (1–42) (10 μmol/L)/Cu²⁺ (5 μmol/L)/AA (1 mmol/L) mixture in the presence or absence of different concentrations of Ma-3-gal-Cl (0.5, 4, and 10 μmol/L) were detected with CCA probe. The fluorescence ratio was calculated as follows:

Where F is the CCA fluorescence intensity in each solution, and F₀ is the CCA fluorescence intensity of the Aβ (1–42) (10 μmol/L)/Cu²⁺ (5 μmol/L)/AA (1 mmol/L) mixture.

DCFH-DA was used to detect the level of intracellular reactive oxygen species (ROS). The SH-SY5Y cells were treated with different solutions for 24 h and then treated according to the instruction of the DCFH-DA kit. The fluorescence intensity was recorded with the Hitachi F-4600 spectrofluorometer from Hitachi High-Tech Corporation (Tokyo, Japan) with the excitation and emission wavelengths at 485 nm and 525 nm. The fluorescence ratio was calculated as follows:

Where F is the DCFH-DA fluorescence intensity of Aβ (1–42) (10 μmol/L)/Cu²⁺ (5 μmol/L)/AA (1 mmol/L)/Ma-3-gal-Cl (0.5, 4, 10 μmol/L) mixtures treated SH-SY5Y cells, and F₀ is the DCFH-DA fluorescence intensity of the Aβ (1–42) (10 μmol/L)/Cu²⁺ (5 μmol/L)/AA (1 mmol/L) treated SH-SY5Y cells.

Fluorescent probe JC-1 was used for detecting the mitochondrial membrane potential (MMP) of SH-SY5Y cells. After the incubation treatment, the cells were treated with JC-1 dye following the instruction of the JC-1 kit. The fluorescence intensity was recorded with a DNM-9602 microplate reader from Plan New Technology Co., Ltd. (Beijing, China), and the MMP was calculated as the ratio of JC-1 monomer (FI 530)/JC-1 aggregate (FI 590) fluorescence intensity (42).

Proteins were extracted from pretreated SH-SY5Y cells, and protein level was measured with a BCA protein quantification assay kit [Labgic Technology Co., Ltd. (Hefei, China)]. Proteins (8–12 μg/μL) were mixed with equal volume (8–12 μL) of sodium dodecyl sulfate (SDS) buffer (0.125 mol/L Tris-HCl, pH 6.8, 2% SDS, 0.5% 2-mercaptoethanol, 1% bromophenol blue, and 19% glycerol) and boiled for 5 min. Proteins were separated by SDS-polyacrylamide gel and transferred to nitrocellulose membranes. Then, the membranes were incubated with caspase-3 (cleaved), p38, and P-p38 antibodies overnight at 4°C, probed with horseradish peroxidase-conjugated secondary antibodies at room temperature, and imaged with Gel Imager 721-BR10883 from BIO-RAD Co., Ltd. (Hercules, CA, USA).

Thioflavin was used for detecting the aggregation process of Aβ (1–42). The ThT fluorescence intensity of Aβ (1–42) (80 μmol/L) incubated with or without Ma-3-gal-Cl (10 μmol/L) at different times (0, 2, 4, 6, 12, 24, 48, 72, and 96 h) was detected with a Hitachi F-4600 spectrofluorometer (Hitachi, Japan) with the excitation and emission wavelengths of 450 and 490 nm, respectively. The widths of the entrance and exit slits were both 10 nm. The concentration of ThT was 5 μmol/L.

Atomic force microscopy (AFM) can be used to observe the surface structure of Aβ (1–42) aggregates (43, 44). In this experiment, morphologies of Aβ (1–42) aggregates were characterized with a Nanoman vs. AFM (Bruker, Germany) with tapping mode. Aβ (1–42) was pretreated with hexafluoro-isopropyl alcohol (HFIP) and freeze-dried into powder. The freeze-dried powder was dissolved with 20 mmol/L NaOH and diluted into the experimental concentration with 10 mmol/L PBS buffer (pH 7.4) solution. Samples taken from incubated Aβ (1–42) solutions or Aβ (1–42)/Ma-3-gal-Cl mixtures were dropped on Ni²⁺-treated mica sheets, briefly washed with ultrapure water, and eventually dried under the gentle nitrogen stream.

The docking studies were performed using Molecular Operating Environment (MOE) software (version 2019). Before the docking study, we first prepared the Ma-3-gal-Cl by searching the conformation using the default parameters. Then, for proteins, water molecules, ions, and non-standard amino acid residues were detached from the proteins and the hydrogen atoms were added under an AMBERT10 force field. After the automated correction of protein structure using the “Structure Preparation” module, the binding sites were detected. For caspase-3, crystal structure 2J33 (PDB ID: 2J33) (45) was selected as the unactivated model, and the binding sites were selected according to the original ligand. For the crystal structure of MKK6 (PDB ID: 3VN9) (46), the prior site could be ATP binding site. For the Aβ crystal structure of monomer (PDB ID: 1IYT) (47) and fibril (PDB ID: 2BEG) (48), we scanned the whole surface to find better sites. The prepared Ma-3-gal-Cl was then flexibly docked into the receptor using the “Triangle Matcher” placement method and “London dG” scoring with other default parameters. Finally, ten docking poses were obtained, and the one with the best score was chosen for analysis.

The results were calculated using the SPSS l7.0 statistical analysis, and Origin was used for drawing. Each experiment and each group of data were repeated three times. The results were expressed by mean ± standard deviation (mean ± SD), and the single-sample t-test was used to compare the group differences. P < 0.05 indicated a statistical difference.

From the FooDB database, 4,143 components of bilberry were collected. From these components, glycerol and fatty acids such as triglyceride (TG), diacylglycerol (DG), and phosphatidyl ethanolamine (PE) were preliminarily deleted, and 182 ingredients were obtained. After removing 38 entries such as water, inorganic substances, and triglycerides, 144 ingredients were obtained. Next, we retained the ingredients with a molecular weight of <100 because compounds with very small molecular weights always behave without specificity. After this, 132 ingredients were retained for the next analysis.

The 132 ingredients were then fed into the pipeline for ADMET and drug-likeness screening. As the process described in the Python code (Supplementary Figure S1), this step adopted the strategy of a cumulative score. Each ingredient got an inherent score according to whether the properties meet the requirements, and finally each ingredient got a total score. After ranking the total scores, we get the top 10 ingredients. The structures of these 10 ingredients are listed in Supplementary Table S1. We found that delphinidin (49), cyanidin (50), petunidin (51), quercetin (52), gallocatechin (53), epigallocatechin (54), and myricetin (55) all have been reported to relate with their function against AD, while ingredients based on Malvidin scaffold seem no specific reports. This led to our way of thinking: can we make further study of malvidin-3-O-galactoside (Malvidin scaffold) to see whether it is a neuroprotective ingredient that will help to complete a full profile of the bilberry against AD?

The target prediction results are shown in Figure 1. The simple network displaying the interaction between the top 10 compounds including Ma-3-gal-Cl and the three targets was constructed as shown in Figure 1B. The top 10 compounds all process the possibility that could be interacted with the targets, which was somehow confirmed by the publications mentioned above. For Ma-3-gal-Cl (number C_26), Aβ (1–42) (P05067), Caspase-3 (P42574), and AChE (P22303) were predicted as its targets. To enhance the reliability, we compared the prediction results of different tools. From Figures 1C,D, we observe that TargetNet and SwissTargetPrediction voted the Ma-3-gal-Cl to interact with Aβ (1–42), and TargetNet, SwissTargetPrediction, and PPB2 voted the Ma-3-gal-Cl to interact with AChE. Although the unified prediction criteria and degree of different tools cannot be implemented, the voting results will provide evidence of interaction from multiple perspectives (e.g., similarity, pharmacophore, and chemical genomics). This increased our confidence in exploring its mechanisms against AD related to these two targets.

In AD patients' brains, Aβ was complexed with excess metal (such as copper, iron, and aluminum) ions in senile plaques. It was reported that the Aβ-metal complex, especially the Aβ-Cu²⁺ complex, can facilitate the generation of H₂O₂ by reacting with cellular species such as AA and O₂ (56, 57). In the cellular milieu, any rogue Cu²⁺ that is not readily complexed by Aβ will also react with H₂O₂ to produce hydroxyl radicals via the Harber–Weiss reaction and induce oxidative stress to neurons. Meanwhile, the toxic Aβ aggregates are also one of the culprits for neuron apoptosis. According to the cheminformatics results, Ma-3-gal-Cl could interact with Aβ, which suggested that Ma-3-gal-Cl may influence Aβ-Cu²⁺ induced oxidative stress to cells and the formation of the toxic Aβ aggregates. Thus, in vitro experiments were performed.

H₂O₂ is an important reactive oxygen species (ROS), and it is also a precursor for other ROS (e.g., HO). In order to explore the protective effect of Ma-3-gal-Cl on oxidative damage of H₂O₂ to SH-SY5Y cells, SH-SY5Y cells were incubated with different concentrations of Ma-3-gal-Cl and H₂O₂ for 24 h, and the cell viability of SH-SY5Y cells was detected by MTT assay. The results showed that Ma-3-gal-Cl could effectively improve the cell viability of SH-SY5Y cells treated with H₂O_2. When the concentration of Ma-3-gal-Cl increased from 0 to 10 μmol/L, the cell viability was increased from 48.2 to 87.3% (P < 0.01) (Figure 2B). Additionally, when the concentration of Ma-3-gal-Cl increased to 20 μmol/L, the cell viability was decreased to 77.6%. These results indicated that Ma-3-gal-Cl with a lower concentration (below 10 μmol/L) could effectively inhibit the H₂O₂₋induced cell apoptosis. But at the high concentration of Ma-3-gal-Cl (20 μmol/L), due to the cell toxicity of Ma-3-gal-Cl, lower cell viability was observed (Supplementary Figure S2).

In the senile plaques of patients with AD, large amounts of redox-active metal ions such as Cu²⁺ and Fe²⁺ have been found to coexist with the aggregates of amyloid-beta (Aβ) peptides (58). These metal ions can strongly bind Aβ peptides (58, 59), and the resultant complexes can facilitate the generation of H₂O₂ by reacting with cellular species such as AA and O₂ (59). Furthermore, redox metal ions (e.g., Cu²⁺ and Fe²⁺) can generate hydroxyl radicals (OH·) by reacting with H₂O₂ through the Harber–Weiss and Fenton reactions (8, 60). Thus, in this study, the Aβ (1–42)/Cu²⁺/AA system was used as the OH· production model to simulate the OH· producing process in senile plaques of patients with AD.

To investigate the inhibition effect of Ma-3-gal-Cl to Aβ (1–42)/Cu²⁺/AA induced cell apoptosis, the cell viability of SH-SY5Y cells incubated with different concentrations of Ma-3-gal-Cl/Aβ (1–42)/Cu²⁺/AA mixture was detected. As shown in Supplementary Figure S3, the low concentrations of Cu²⁺ (5 μmol/L) or AA (1 mmol/L) alone showed no cell toxicity to SH-SY5Y cells. Aβ (1–42) (10 μmol/L) reduced the cell viability to 92.3% (Supplementary Figure S3). However, when the same concentration of Aβ (1–42), Cu²⁺, and AA was mixed, the cell viability was decreased to 55.0%. This may be due to the oxidative damage from OH· produced by Aβ (1–42)/Cu²⁺/AA system, and the cell toxicity from Aβ (1–42) toxic aggregates. As shown in Figure 2C, when the concentration of Ma-3-gal-Cl increased from 0 to 10 μmol/L, the cell viability was increased from 55.0% to 90.9%. When the concentration of Ma-3-gal-Cl increased to 20 μmol/L, the cell viability was decreased to 82.73%. These results indicated that when the concentration of Ma-3-gal-Cl is below 10 μmol/L, it could inhibit the Aβ (1–42)/Cu²⁺/AA system so as to induce cell apoptosis.

To investigate the mechanism of the protective effect of Ma-3-gal-Cl to Aβ (1–42)/Cu²⁺/AA mixture treated cells, the scavenging effect of Ma-3-gal-Cl on OH· produced by Aβ (1–42)/Cu²⁺/AA mixture was detected with the CCA method (61). With the increase in Ma-3-gal-Cl concentration (0, 0.5, 4, and 10 μmol/L), the amount of the OH· decreased from 100 to 20% (Figure 3). These results indicate that Ma-3-gal-Cl could inhibit the production of OH· by Aβ (1–42)/Cu²⁺/AA mixture, and the inhibition effect is concentration-dependent.

Intracellular ROS level is an important reflection of oxidative damage, and it can be detected with DCFH-DA (62). As shown in Figure 4A, the amount of intracellular ROS in SH-SY5Y cells incubated by Aβ (1–42)/Cu²⁺/AA increased to 145%. With the increase in Ma-3-gal concentration (0–10 μmol/L), the amount of intracellular ROS was reduced (from 145.0 to 106.7%, compared with the control group). These results indicated that Ma-3-gal-Cl can reduce the intracellular ROS production of SH-SY5Y cells induced by Aβ (1–42)/Cu²⁺/AA mixture.

Oxidative stress can lead to the decline of the MMP of cells and result in cell apoptosis (63, 64). To study the protective effect of Ma-3-gal-Cl on the Aβ (1–42)/Cu²⁺/AA treated SH-SY5Y cells, the MMP was determined with JC-1 staining (42). As shown in Figure 4B, the MMP could be increased dramatically by Aβ (1–42)/Cu²⁺/AA from 0.092 to 0.464. With the addition of Ma-3-gal-Cl, the MMP values were decreased. When the concentration of Ma-3-gal-Cl was at 4 μmol/L, the MMP value decreased to 0.356 and further increased the Ma-3-gal-Cl concentration to 10 μmol/L, and the MMP value only decreased to 0.326. These results indicated that Ma-3-gal-Cl could reduce the mitochondrial damage caused by oxidative stress from Aβ (1–42)/Cu²⁺/AA.

The caspase-3 and p38 are important proteins in the apoptosis pathway, and their activations can trigger a series of apoptosis reactions and lead to cell apoptosis. To investigate the molecular mechanism of the inhibition effect of Ma-3-gal-Cl to SH-SY5Y apoptosis induced by Aβ (1–42)/Cu²⁺/AA, different concentrations (0.5, 4, and 10 μmol/L) of Ma-3-gal-Cl were selected, and the expression level of target protein was detected by Western blot (WB) assay. As shown in Figures 5A,B, the expression level of the caspase-3 (cleaved) increased in the Aβ (1–42)/Cu²⁺/AA treated SH-SY5Y cells (from 100 to 155%), and the one for Ma-3-gal-Cl/Aβ (1–42)/Cu²⁺/AA mixture treated cells was decreased (from 155 to 111%). As shown in Figures 5A,C, compared with the control group, the Aβ (1–42)/Cu²⁺/AA mixture and low concentration (below 4 μmol/L) of Ma-3-gal-Cl almost have no influence on the expression of p38 in cells (Figure 5C). However, Aβ (1–42)/Cu²⁺/AA incubation could increase the expression of P-p38 in SH-SY5Y cells. Additionally, Ma-3-gal-Cl (0–10 μmol/L) could dramatically decrease the P-p38 expression (from 300 to 108%) (Figure 5D). These results indicate that Ma-3-gal-Cl could not influence the expression of p-38 in Aβ (1–42)/Cu²⁺/AA treated SH-SY5Y cells, but it could inhibit the expression of cleaved caspase-3 and P-p38 in Aβ (1–42)/Cu²⁺/AA treated SH-SY5Y cells.

The aggregates (oligomers, protofibrils, and fibrils) of Aβ were considered toxic species to neuron cells (65, 66). To investigate the inhibition effect of Ma-3-gal-Cl on Aβ aggregation, a ThT assay was employed. As shown in Figure 6, line A, in the first 6 h, the fluorescent intensity of the Aβ (1–42) solution increased slowly. When the incubation time is at the range of 6–48 h, the fluorescent intensity increased sharply, and the one for the longer incubation time (48–96 h) increased gradually again. These results are consistent with other reports (67–69). The lag phase (ca. 6 h) is indicative of the nucleation phase for Aβ (1–42) aggregation, whereas the fluorescent intensity plateau (48–96 h) corresponds to the formation of well-ordered, β-sheet-rich Aβ (1–42) fibrils. When the Ma-3-gal-Cl was incubated with Aβ (1–42), the fluorescent intensity of ThT increased gradually at the first 6 h and decreased at later times (Figure 6, line B). These results indicated that Ma-3-gal-Cl could inhibit the aggregation of Aβ (1–42) at the long incubation time, but could not completely inhibit Aβ (1–42) aggregation in the early hours of the incubation.

To investigate the morphology of products from the Aβ (1–42) and Ma-3-gal-Cl/Aβ (1–42) mixture, the atomic force microscope (AFM) was employed. In the Aβ (1–42) solutions, globular aggregates and short protofibrils were observed in the 3-h-incubation sample (cf. images juxtaposed in row A of Figure 7), which is consistent with the lag phase in Figure 6. After 6 h, more protofibrils were produced, and after 24 h, mature Aβ (1–42) fibrils were dominated. In the presence of Ma-3-gal-Cl, at the first 6 h, the morphology of the Aβ (1–42) aggregates (Figure 7, row B) was similar to those in Aβ (1–42) alone solution (Figure 7, row A), but at 12 h, the density of Aβ (1–42) fibrils was decreased and after 24 h, the mature fibrils could not be observed, and instead with short, small aggregates. At 96 h, the amount of the small aggregates dramatically decreased. These results were consistent with ThT fluorescence results (Figure 6). This observation suggested that Ma-3-gal-Cl could bind with the Aβ (1–42) monomer and large oligomers to prevent Aβ (1–42) oligomers and protofibrils from further growing into mature fibrils.

In order to explore the specific structural basis and molecular mechanism of antioxidant and anti-aggregation effects of Ma-3-gal-Cl to Aβ (1–42), the molecular docking method was employed for further study. In the in vitro tests, the expressions of activated caspase-3 and P-p38 of Ma-3-gal-Cl/Aβ (1–42)/Cu²⁺/AA mixture treated cells were decreased (Figure 5). The Aβ (1–42) aggregates were also inhibited by Ma-3-gal-Cl (Figures 6, 7). Thus, the unactivated caspase-3 (crystal structure of 2J33), mitogen-activated protein kinase kinases-6 (MKK6) (PDB ID: 3VN9) related with the active process of p-38, Aβ monomer (PDB ID: 1IYT), and fibril (PDB ID: 2BEG) were selected as receptors for the molecular docking experiments.

The structure-based docking of unactivated caspase-3, MKK6, Aβ (1–42) monomer, and fibril with Ma-3-gal-Cl gave the best score of −7.12, −7.68, −5.51, and −6.52 kCal/mol, respectively. Their interactions were visualized in 2D and 3D diagrams, and the hydrogen bond interactions between Ma-3-gal-Cl and the active site residues were observed (Figure 8). When processing the docking caspase-3 with Ma-3-gal-Cl, we first made a self-docking using the ligand in the crystal structure. The RMSD between the best pose and the original ligand is 1.41 Å, which indicated a set of proper docking parameters for this system. By using this set of parameters, we found that the residues, namely Cys163 (bond length: 3.60 Å), Ser120 (bond length: 2.92 Å), Arg207 (bond length: 2.77 Å), Arg64 (bond length: 2.96 Å), and Arg207 (bond length: 3.00 Å), were observed to form five hydrogen bonds with Ma-3-gal-Cl of the caspase-3 receptor (Figures 8A,E,I). For MKK6, Ma-3-gal-Cl was docked to the ATP binding site which indicated an ATP-site-directed inhibitor (70, 71). The residues, namely Ala63 (bond length: 3.20 Å), Asn184 (bond length: 2.95 Å), Asp197 (bond length: 3.03 Å), and Lys82 (bond length: 2.88 Å), were observed to form four hydrogen bonds (Figures 8B,F,J). Ma-3-gal-Cl could stay at the site with a good pose. For Aβ (1–42) monomer, there were no obvious pockets with good geometry and hydrophobicity because of no complex secondary structures. However, we found that most of the retained 10 poses tended to interact with N-terminus residues (3–17), while the importance of the N-terminal residues has been indicated in oligomerization and neurotoxicity (72, 73). In addition, these regions are the essential components of the binding site for glycosaminoglycans, which affects the change in Aβ (1–42) secondary structure from a soluble α-helix conformation to a stable β-sheet one. In the best pose, a hydrogen bond (Glu11) and a pi–H interaction (Glu3) (Figures 8C,G,K) were observed. For Aβ (1–42) fibril, the NMR structure gives important information about the identification of interaction regions, which might be targeted by inhibitor compounds. It suggested specific structural properties to interact and destabilize Aβ (1–42) self-assembly, including (a) the hydrophilic region caused by electrostatic interaction between Asp23-Lys28, (b) the Glu22 ladder formed by the Glu22 residue side chains of adjacent β-sheets, (c) the central cleft in the interior of the U-shaped turn, and (d) the hydrophobic regions caused by Leu17-Ala21 and Ala30-Val36 residues, respectively. The best pose in our study occupied the central cleft in the interior of the U-shaped turn and formed two pi–H interactions (Ala21 and Glu22), which was in line with the proposed features (Figures 8D,H,L). In addition, three poses from the 10 retained ones tended to form pi–H interactions with Glu22, Phe19, and Phe20. This can be corroborated from the side by a recent study, which indicated the π-π interactions of inhibitor agents with Phe19 and Phe20 residues can stabilize the fibrils by a reduction in the total energy of the conformation (74).

The docking results indicate that Ma-3-gal-Cl could interact with unactivated caspase-3, MKK6, Aβ (1–42) monomer, and fibril. Although there were differences in scores and interactions, there was a relatively high degree of agreement with the evidence provided by published studies. Furthermore, the calculated results accord with the experiments, which reflected our reasonable hypothesis that Ma-3-gal-Cl could reduce neuronal apoptosis by combining antioxidation and anti-aggregation of Aβ (1–42).