Molecular Dynamics Studies on the Inhibition of Cholinesterases by Secondary Metabolites

Abstract

The search for selective anticholinergic agents stems from varying cholinesterase levels as Alzheimer’s Disease progresses from the mid-to-late stage and from butyrylcholinesterase’s (BChE) role in β-amyloid plaque formation. While structure-based and pharmacophore-based virtual screening could search from large libraries in a short time, these methods do not consider dynamic features that result from a ligand’s inhibition of the enzyme and consequently may under- or overexaggerate enzyme selectivity of a given ligand. In this computational study, we probed the selectivity of representative secondary metabolite compounds against acetylcholinesterase and BChE through molecular dynamics simulations. The results were evaluated by analysis of the root mean squared deviation of ligand heavy atoms, the radius of gyration of each inhibited and uninhibited enzyme, root mean squared fluctuation of residues, intermolecular interaction energy, and linear interaction energy approximation of the Gibbs free energy of binding. These considerations further reveal the induced-fit characteristics contributing to ChE selectivity that are predominantly due to the greater flexibility of BChE’s active site gorge.

1. Introduction

1.1. Selective Cholinesterse Inhibitors and Alzheimer’s Disease

Acetylcholinesterase (AChE) has been the primary target for inhibitor-based Alzheimer’s Disease (AD) treatment for over 30 years [1,2,3,4,5,6]. More recently, interest in elucidating selectivity trends between AChE and another cholinesterase (ChE), termed butyrylcholinesterase (BChE), has grown [7,8,9,10,11,12,13]. The reasoning for such stems from BChE’s role as a bioscavenger for neurotoxic organophosphates that may enter the central nervous system (CNS) and the desire to produce more effective treatments in the early-to-mid and late stages of the disease, during which AChE and BChE concentrations are known to fluctuate [14,15].

Preliminary in silico investigations have proved to be modestly successful in the prediction of selective cholinergic agents. Methodologies based on the principles of structure-based drug discovery usually begin with a pharmacophore or molecular-docking based virtual screening technique for lead-type molecule generation [16,17,18,19,20]. Molecular dynamics (MD) simulations go beyond these calculations and allow for more comprehensive simulation by allowing the enzyme–ligand complex to undergo Newtonian motion and interact with solvent at varying temperatures and pressures [21,22].

In this article, we utilize the binding conformations previously found through molecular docking for a selection of secondary metabolite phytochemicals (Figure 1) as the starting conformation for MD simulations and approximate the change in the Gibbs free energy of binding (ΔG) between select ligands and the ChEs. The ligands in the following section were chosen specifically based on the unique properties they possess (vide infra) and docking-predicted selectivity. The ligand dataset from which the ligands were chosen contains thousands of secondary metabolite phytochemicals, many of which have either large deviation from the Lipinski Rule of Five or unreasonable therapeutic windows, unknown therapeutic windows, etc.

1.2. Secondary Metabolite Phytochemicals

1.2.1. Afzelechin

Afzelechin is a flavonoid compound commonly extracted from Bergenia ligulate, though an enantioselective synthesis has also been proposed [23,24]. This compound is a tetrahydroxyflavan with S-type chirality and has hydroxy groups distributed at positions 3, 5, 7, and 4’. The theoretical water–octanol partition coefficient (LogP) is predicted to be 0.7, just below the 1–3 range suitable for molecules to cross the blood–brain barrier (BBB) [25]. With four and five hydrogen bond donors and acceptors, respectively, this compound follows the Lipinski Rule of Five with respect to these metrics. Additionally, the molecular weight (MW) is about 274 Da, also in agreement with the Lipinski Rule of Five. It is further classified as a catechin, a group of flavonoids that act as antioxidants in vivo. In previous work, afzelechin has been found to inhibit KATP channel, inhibit alpha-glucosidase, alleviate sepsis associated pulmonary injury, and serves as an anti-inflammatory agent in cases of lipopolysaccharide-induced inflammatory processes [26,27,28,29].

1.2.2. Aspalathin

Aspalathin is a secondary metabolite product of Aspalathus linearis and is often consumed when the leaves of this species are brewed into tea [30]. This compound is a member of the dihydrochalcones class and a member of the catechol class. The theoretical LogP is predicted to be −0.2, which is below the range for permeation across the BBB [31]. Similarly, the compound has nine hydrogen bond donors and eleven hydrogen bond acceptors, thereby disqualifying it from adhering to the Rule of Five. Considering the MW of this compound is above 450 Da, this suggests that this compound best be considered a lead-type compound rather than a legitimate therapeutic target as it will likely be unable to reach our ChEs at the post-synaptic neuromuscular junction (PSNMJ) before undergoing reduction. Aspalathin has been found to inhibit xanthine oxidase and protect pancreatic beta-cells [32,33]. Through the latter effect, the compound can decrease the effects of hyperglycemia and has thus been proposed as a potential therapeutic agent for individuals with type-2 diabetes.

1.2.3. D-Maslinic Acid

D-maslinic acid is a natural product derived from Olea europaea and has the structure of a pentacyclic triterpenoid [34]. As far as its agreement with the Lipinski parameters, D-maslinic acid falls short on only one of the four metrics. With a MW of about 470 Da, three hydrogen bond donors, and four hydrogen bond acceptors, the compound passes these stipulations of the Rule of Five. Despite this, a predicted LogP of about 6.5 would effectively hamper this compound from entering the CNS [35]. Thus, it is reasonable to consider this compound solely as lead-type against ChEs and to highlight the structural differences between the enzymes. This compound has been found to have a healthy therapeutic range in humans and exhibits anti-inflammatory and antioxidant effects [34,36]. Maslinic acid is also currently being studied for its anticancer effects due to its ability to significantly decrease pancreatic tumor growth in rodent in vivo models [37].

1.2.4. Isoliensinine

This compound is a member of the isoquinolones class of compounds. Upon administration, it is likely that the ether connecting the two quinolones in this structure will be hydrolyzed and break into two separate quinolone compounds. The compound is a phytochemical primarily extracted from the species Nelumbo nucifera [38]. The predicted LogP value of this compound is 6.4, indicating it would have a difficult time permeating the BBB [39]. However, with hydrogen bond donor and acceptor counts of two and four, respectively, this compound falls within the guidelines of the Lipinski Rule of Five in these regards. Isoliensinine is commonly found in everyday soups and teas from southeast Asia and thus has been deemed a safe candidate for administration in vivo. A recent study has found the compound to have drug-like potential and effective in the treatment of dizziness, insomnia, and symptoms of type-2 diabetes [40]. This molecule has been evaluated against both ChEs in vitro and was found to be a dual inhibitor with selectivity favoring AChE.

1.2.5. Luteolin

Luteolin is a tetrahydroxy flavone that is very similar in structure to a compound previously discussed afzelechin. This compound is a flavonoid with four hydroxy groups substituted to the main structure. It is found in many natural products including celery, green pepper, and chamomile [41]. The compound has a LogP values of 1.4, making it fall slightly below the recommended value for permeating the BBB [42]. MW, hydrogen bond acceptors, and hydrogen bond donors, for this compound are reported to be around 280 Da, six, and four, respectively. These figures firmly place this molecule within the guidelines of the Lipinski Rule of Five. Luteolin has been reported to scavenge reactive oxygen species due to its antioxidant capabilities and has been shown to exhibit properties of an anti-inflammatory agent [43]. In addition, this molecule has been tested against AChE and BChE and was found have an IC₅₀ of 20.47 ± 1.10 μM for AChE and an IC₅₀ of 46.15 ± 2.20 μM for BChE [44].

1.2.6. Matricin

This compound is also derived from the chamomile flower and is classified as a sesquiterpene [45]. The compound has a MW of about 300 Da and a predicted LogP value of around 1.0 [46]. With one hydrogen bond donor and four acceptors, the ligand then falls within the Lipinski Rule of Three and Rule of Five, indicating a high degree of likelihood of entering the CNS. This compound has been tested as an inhibitor against chymase enzymes and in the treatment of non-small-cell lung cancer cells by increasing apoptosis and preventing proliferation [47]. Due to its high concentration in chamomile tea, it is considered to have a healthy therapeutic window and has thus been chosen as a candidate compound for numerous biomolecular targets, which is a potential indication of its inability to be highly selective towards ChEs relative to other macromolecule targets [48].

1.2.7. Sedanolide

This compound is a member of the benzofurans and is the smallest compound tested in this study. Extraction of this compound has been performed with numerous plant species, including Ligusticum striatum and Angelica sinensis, though it is most often extracted from celery seed oil [49,50]. While it is less than the Lipinski Rule of Three with a MW of less than 200 Da, it has a predicted LogP of around 3.3, which makes it optimal for permeation into the CNS [51]. Hydrogen bond donor and acceptor counts are at zero and two, respectively, which gives further indication of the feasibility of this compound to reach the PSNMJ, where our ChEs are located. Previous research has found that sedanolide is highly bioactive and has the ability to upregulate antioxidant genes through the activation of the KEAP1-NRF2 pathway [52]. As it is a primary component of common edible products, it is considered to have a workable therapeutic window.

1.2.8. Thebaine

Thebaine is derived from the poppy plant and is a common precursor of drugs such as oxycodone and naloxone [53]. As such, it is referred to as a morphinane alkaloid and is listed as a schedule II-controlled substance by the Drug Enforcement Administration. While this inherently limits thebaine as a legitimate candidate for an anti-cholinergic agent therapeutic, several opiates were predicted to be AChE-selective in previous in silico studies performed by our group, and the reasoning for the selectivity is studied here through MD simulations. Thebaine has a LogP value of around 2.2, a MW of about 310 Da, no hydrogen bond donors, and four hydrogen bond acceptors [54]. This means that the compound is in accordance with the Rule of Five and is likely to cross the BBB. This makes sense considering its derivatives commonly enter the CNS to bind to endogenous opiate receptors. As this compound is being studied for its trends in selectivity and not for legitimate use as an AD drug candidate, a therapeutic window discussion is excluded.

2. Results

2.1. Secondary Metabolite Phytochemicals

2.1.1. Afzelechin

Through the course of the MD simulations, the ligand migrated from its initial docking-predicted conformation more greatly when bound with BChE than AChE. This was determined through a calculation of the root mean square deviation (RMSD) of the distance between non-hydrogen atoms of ligands from their starting conformation. With that noted, the fluctuation of RMSD values followed a similar pattern between the ligand and the two complexes, except for a slight shift along the abscissa and a difference in magnitude of about 0.08 nm, or 0.8 Angstroms, along the ordinate axis (Figure 2b). In other simulations, the RMSD of afzelechin heavy atoms in the BChE complex were found to increase in magnitude up to 0.6 nm compared to heavy atoms of the ligand in the AChE complex (Figures S4a and S8a).

Short-range intermolecular Coulombic and Lennard-Jones interaction energies were also calculated between the residues of the binding pocket and afzelechin. Though energies between the enzymes were relatively similar with a difference of about 25 kJ/mol, the AChE–afzelechin complexes were predicted to have the strongest attractive interaction. For AChE complexes, the electrostatic contribution to intermolecular energy was greater than that of the Lennard-Jones interactions. The opposite trend was observed in the BChE–afzelechin complexes, with Van der Waals interactions primarily contributing to the overall energy interaction (Figure 2a, Figures S5a and S9a).

For radius of gyration considerations, the enzyme–ligand complex was compared to that of the free uninhibited enzyme. In the case of the AChE–ligand complexes, the residues underwent slight relaxation to accommodate the introduction of the ligand as compared to unbound AChE. After the full 100 ns for all three simulations, the difference between uninhibited AChE and afzelechin–AChE for this metric was less than 0.02 nm (Figure 2c, Figures S6a and S10a). In contrast, there was very little movement of the residues necessary to accommodate the ligand in the active site of BChE in the simulation pictured below (Figure 2d). Due to the reduction in R_g values compared to the uninhibited BChE structure, the afzelechin appears to have further stabilized the BChE when within the gorge. In other simulations, unfolding of the inhibited BChE was seen to occur with an increase in R_g values (Figures S6b and S10b).

The estimated Gibbs free energy between afzelechin and AChE fluctuated less greatly than that of afzelechin and BChE throughout the simulation. For the AChE–afzelechin case, the fluctuation appears nearly uniform for all three simulations (Figure 2e, Figures S7a and S11a). A relatively similar uniform fluctuation occurs for the BChE–afzelechin case, though the ΔG tends to decrease as the simulation progresses, indicating the migration of the ligand to a more energetically favorable conformation than its initial docking predicted conformation (Figure 2f, Figures S7b and S11b). Coupled with the fact that the R_g for BChE–afzelechin was found to be less in magnitude than the uninhibited BChE up until the final 20 ns of simulation time, the estimated ΔG further suggests a stabilization of BChE due to the introduction of afzelechin in this case.

2.1.2. Aspalathin

For the initial 20 ns of our simulations, the RMSD values of the heavy atoms of aspalathin were very similar between the AChE and BChE complexes with nearly identical patterns. Around 35 ns in the simulation pictured below, a large spike in RMSD values in the AChE complex was found before dropping back within the range of the BChE complex after about five additional ns (Figure 3b). This is noteworthy as most jumps in RMSD values from our enzyme–ligand simulations occur within the BChE–ligand complexes rather than with AChE. In other tested simulations, however, RMSD of heavy atoms of aspalathin in the BChE complex were found to fluctuate much greater than in the AChE case similarly to the majority of tested complexes (Figures S4b and S8b).

Concerning interaction energies, Lennard-Jones interactions were found to be the largest contributors for both AChE and BChE complexes. With that said, in all three simulations, the Coulombic contribution would occasionally peak below the Lennard-Jones contribution for the AChE–ligand cases (Figure 3a, Figures S5b and S9b). The aspalathin–AChE was found to have more significant interactions than BChE, with a difference in energy value of about 40 kJ/mol for Lennard-Jones interactions and a 25 kJ/mol difference for Coulombic interactions. A much tighter grouping between Lennard-Jones and Coulombic interaction energies was observed for both ChE–aspalathin complexes, more than most other enzyme–ligand complexes in this study.

Radius of gyration calculations were found to follow a similar trend across simulations. Near the beginning of the simulations, AChE began to unfold more than BChE to account for the presence of the ligand within its docking-predicted conformation at the binding site. As the simulations reached 100 ns, the inhibited AChE structures’ R_g reduced to match more closely with the uninhibited structure (Figure 3c, Figures S6c and S10c). While the uninhibited BChE began to increase its R_g levels periodically, BChE in complex with aspalathin did not undergo the same R_g increase during this time and remained relatively constant throughout the production runs with values ranging from 2.30 to 2.34 nm (Figure 3d, Figures S6d and S10d).

With regard to the AChE–aspalathin ΔG, we note that the ΔG values primarily remain positive through the production run pictured (Figure 3e). In other AChE–aspalathin simulations, however, ΔG values were found to trend negative as the simulations progressed (Figures S7c and S11c). The BChE–aspalathin ΔG results remained consistent through the three simulations (Figure 3f, Figures S7d and S11d). An increase at 45 ns in the case below is consistent with an increase observed in the R_g plot of uninhibited BChE case at 45 ns, though the ΔG values quickly reduce to negative values, providing more context for the stabilization observed in the R_g values of the inhibited BChE case near 45 ns. Taking all simulations into account, AChE–aspalathin was found to have more reliably negative ΔG values than the BChE–aspalathin case.

2.1.3. D-Maslinic Acid

Results from the MD analysis showed that over the simulations, the RMSD values of the ligand fluctuated more greatly in the BChE complex than the AChE complex. While RMSD values in the case of AChE–D-maslinic acid generally stabilized as the simulations wore on, RMSD values in the BChE case generally required more steps to do so. With that noted, the RMSD was significantly lower in the BChE complex in two of the three simulations (Figure 4b and Figure S4c). This potentially indicates an unfavorable initial docked configuration for the ligand AChE. Contrary to other simulations, the initial RMSD values in the AChE complex pictured below began at nearly 0.35 nm greater than the starting value for BChE.

The interaction energy values for either complex were found to not significantly fluctuate as frequently as results obtained with the other ligands used in this study. With that noted, Lennard-Jones interactions predictably dominated the contribution with a nearly 150 kJ/mol difference in Coulombic and Lennard-Jones interactions for both ChE–ligand complexes across the three simulations per complex. In two of the BChE cases, the Coulombic contribution was found to be much less than the AChE case (Figure 4a and Figure S9c). The Lennard-Jones contribution was found to be approximately equal in the AChE and BChE cases as the simulations reached 100 ns.

The radius of gyration results suggests that inhibited BChE underwent more unfolding than inhibited AChE relative to their uninhibited counterparts. Though for both complexes the obtained R_g values were greater in those from the uninhibited enzyme, in the BChE case, the spike of R_g values in uninhibited BChE was almost exactly reproduced by the inhibited complex in two of the simulations, suggesting very little dissimilarity between the stability of free BChE and the BChE–D-maslinic acid complex during this part of the simulations (Figure 4d and Figure S10f). While the R_g values of inhibited AChE initially were increased from those of the uninhibited AChE, the R_g values reduced to the magnitude of uninhibited case in all simulations attempted.

The ΔG values between D-maslinic acid and AChE were found to be greater than those obtained in the BChE cases. While the ΔG between the ligand and AChE occasionally peaks below −40 kJ/mol across simulations, the ΔG primarily bottoms out around −20 kJ/mol (Figure 4e, Figures S7e and S11e). For the BChE–D-maslinic acid cases, the ΔG values consistently cross beyond −40 kJ/mol thought the simulations (Figure 4f, Figures S7f and S11f). For either ChE, the ΔG values were found to be relatively consistent through the simulation, though shifts in ΔG were more pronounced for the AChE complexes.

2.1.4. Isoliensinine

Over the course of the 100 ns simulations, the RMSD values of heavy atoms of isoliensine indicate gradual deviation from the initial starting conformation with either ChE. Though RMSD values were notable higher in the BChE complexes for two of the simulations (Figure 5b and Figure S8d), in one run the AChE case had larger reported RMSD values (Figure S4d). In the simulation pictured below, the RMSD distribution adopted a similar shape. Despite a minor shift along the abscissa, there were spikes in RMSD values at nearly the same points in time for the ligand in each complex near the beginning of the simulation. The rise in these peaks was observed to have a greater slope in the BChE–isoliensinine case than that of the compound bound to AChE.

Regarding interaction energies, the calculated Coulombic contribution was relatively even between the ligand and our two ChEs through the course of the simulations settling at a value of around 50 kJ/mol (Figure 5a, Figures S5d and S9d). Greater deviation was observed with the short-range Lennard-Jones potential, with AChE–ligand values being more favorable than the BChE–ligand complexes. With that noted, the shapes of the distributions between ChEs are very similar with a notable decrease in interaction energy observed around 30–33 ns in the case below. Both complexes ended their simulations with Lennard-Jones interactions and Coulombic interactions nearly equal in magnitude.

For radius of gyration, there was a marked difference between the enzymes. For the bound AChE complex, the radius of gyration values was shifted up uniformly by about 0.04 nm compared to the unbound enzyme in all three simulations (Figure 5c, Figures S6g and S10g). In comparison, there was less deviation in R_g between the bound and unbound BChE with this compound (Figures S6h and S10h). An exception is seen in the case pictured below, where a larger spike was observed in the bound protein in the final 10 ns (Figure 5d). With that noted, the range of R_g values for the inhibited enzyme are found to be close in value to the range of the uninhibited enzyme for either ChE complex.

The average Gibbs free energy between isoliensinine and the ChEs did not remain constant throughout the simulations, contrary to many other results observed. For the AChE cases, the ΔG values were predicted to be positive at the outset of the production runs and systematically decreased until bottoming out at around −40 kJ/mol (Figure 5e, Figures S7g and S11g). A nearly opposite trend was observed for the BChE–isoliensinine complexes, whose plots shows an initially negative ΔG followed by sustained increases in ΔG as the simulation progressed (Figure 5f, Figures S7h and S11h). By the conclusion of the production runs, the AChE–ligand complex is found to have greater Gibbs free energy of binding than that of the BChE–ligand complex.

2.1.5. Luteolin

When considering the structural similarities of this compound with afzelechin, results from MD production runs suggest some differences between the two compounds. For RMSD, a greater deviation between BChE and AChE complexes was observed in two of the simulations with a separation of about 0.6–0.8 nm at the conclusion of the 100 ns run (Figure 6b and Figure S4e). While the general shapes of the distributions appear similar in the case below (Figure 6b), other simulations do not show this pattern (Figures S4e and S8e). During most simulations, the BChE complexes’ slope of the peaks between minimum and maximum values was greater in magnitude than in the AChE complexes.

In the beginning of the simulations, Coulombic interactions were predicted to be more significant than Lennard-Jones interactions for both enzymes. As the simulations progressed, Coulombic energies in both species periodically fluctuated above and below the Van der Waals interactions. There was greater deviation in the BChE complexes values for Coulombic interaction than the AChE complexes, while for Lennard-Jones, both ChEs values remained relatively constant at a similar magnitude through the 100 ns production run (Figure 6a, Figures S5e and S9e).

The radius of gyration results showed that the bound and unbound AChE were close in magnitude for the full extent of the simulation in the below case and tended towards the unbound AChE values in other simulations (Figure 6c, Figures S6i and S10i). While a similar trend was observed for BChE, multiple peaks in R_g values are the distinguishing factor (Figure 6d, Figures S6j and S10j). This greater fluctuation was observed in all simulations and indicates BChE unfolds more frequently with luteolin than the uninhibited BChE, uninhibited AChE, and inhibited AChE.

Near 30 ns in the simulation pictured below of the AChE–luteolin complex, the ΔG values were found to dramatically decrease and sustain a new average around 0 kJ/mol before increasing again (Figure 6e). Though a similar decrease was observed around 25 ns in the BChE case below, the ΔG then increased to sustain the average it had up until that point (Figure 6f). In other simulations, ΔG values in the BChE case were found to increase over time (Figures S7j and S11j), while ΔG values remained relatively consistent in the AChE case (Figures S7i and S11i).

2.1.6. Matricin

The RMSD values of non-hydrogen atoms of the ligand between ChEs show great deviation and minimal similarity in the overall shape of their distributions. An initial large increase in RMSD around 10–20 ns was observed in the BChE complexes across all three simulations, which continued to gradually increase to a value up to 1.75 nm (Figure 7b, Figures S4f and S8f). In contrast, the RMSD values from the AChE complexes remained relatively constant in two of the three simulations, primarily sustaining a value below 0.25 nm with minimal fluctuation or notable increases/decreases in magnitude.

Interaction energies too differed between ChEs. While the AChE complex held interaction energies relatively constant through the whole production runs, the BChE complex’s interacting energies oscillated through the 100 ns (Figure 7a, Figures S5f and S9f). In the BChE complex pictured below, increases/decreases in Coulombic potential perfectly aligned with increases/decreases in the values for the Lennard-Jones energy, a trend unlike any other tested compounds. An explanation may lie in the tight grouping of functional groups that would induce both electrostatic and Van der Waals interactions present in matricin’s structure.

For both ChE complexes, the bound state yielded R_g values that were higher than the unbound enzymes for the majority of the runs. With that noted, the difference in R_g values between bound and unbound states began to diminish after about 40 ns of simulation in AChE cases (Figure 7c, Figures S6k and S10k). In two of the three BChE complex simulations (Figure 7d and Figure S10l), however, the uninhibited proteins’ R_g values never reduced in magnitude to that of the uninhibited BChE, indicating unfolding of BChE as a result of its binding with matricin.

The resulting Gibbs free energy of matricin and BChE is an interesting outlier with respect to all other simulations ran for this study. Firstly, the ΔG values in the BChE–matricin complex deviate greatly from those in the AChE case, with an approximate average difference of 30 kJ/mol (Figure 7e,f, Figures S7k,l and S11k,l). While the Linear Interaction Energy (LIE) method of estimating ΔG is expected to overexaggerate the binding energy for the BChE case on account of that enzyme’s greater flexibility (vide infra), the difference in average ΔG values between complexes is the greatest of any other ligands tested in this study across triplicates.

2.1.7. Sedanolide

Unlike the majority of our tested compounds, the RMSD values were found to be higher in the AChE complex than BChE at the end of two of the three 100 ns production runs (Figure 8b, Figures S4g and S8g). While the RMSD values in the AChE complexes began at a lower value than the BChE complexes, they experienced greater fluctuation as the simulation progressed with a notable peak around 20–40 ns. In comparison, while increases were noted in the BChE complex, the RMSD values did not deviate as greatly.

The Coulombic and Lennard-Jones interaction energies were found to be very close in magnitude with both ChE complexes, showing greater interactions at differ stages of the production runs. For either enzyme–ligand complex, Coulombic interactions were found to fluctuate more significantly than Lennard-Jones interactions, potentially on account of sedanolide’s multiple hydroxyl groups (Figure 8a, Figures S5g and S9g). Coulombic interactions were found to be significantly weaker than Van der Waals interactions, a trend consistent with the majority of ligands tested.

Radius of gyration analysis from two of the simulations showed that the AChE enzyme began to unfold to account for the presence of sedanolide relative to the unbound AChE (Figure 8c and Figure S6m). The BChE complex was not required to undergo such energy stabilizing conformational changes in two of the simulations, and the bound and unbound complexes showed very similar R_g values (Figure 4d and Figure S10n). From 30 ns until the end of the run in these cases, the bound BChE complex R_g values were found to be below that of the bound state, indicating a more stable enzymatic conformation than the unbound BChE.

Unlike many other results, ΔG values appear similar in magnitude between the AChE and BChE complexes. In the AChE cases, the estimated ΔG values initially were found to be positive and remained as such as the simulation progressed (Figure 8e, Figures S7m and S11m). The ΔG values in the BChE cases also began the run at positive values and sustained near their initial magnitude for the extent of the simulation (Figure 8f, Figures S7n and S11n). Additionally, the ΔG values in the AChE cases were found to have a narrower range than those from the BChE cases.

2.1.8. Thebaine

Results for RMSD analysis of non-hydrogen atoms show that much greater ligand migration was observed in the BChE complexes than the AChE complexes. The RMSD values for the AChE complexes primarily stayed below 0.3 nm, while the BChE complex had values that increased up to 1.1 nm. Notably, a peak in RMSD values near 10 ns in the simulation in the BChE case is absent in the below AChE case (Figure 9b), though these peaks are found in the other AChE complex simulations (Figures S4h and S8h).

Interaction energies tell a similar story, in that the AChE complex was predicted to have more energy-stabilizing contacts than the BChE complex. With respect to the Coulombic contributions, the magnitude between energy values was found to be smaller than the difference in the Lennard-Jones energies between ChEs. More specifically, about a 75 kJ/mol difference exists with the Lennard-Jones energy between ChEs as compared to a 25 kJ/mol difference in electrostatic interactions across all simulations (Figure 9a, Figures S5h and S9h). The difference between Coulombic and Lennard-Jones interactions, about 100 kJ/mol and 125 kJ/mol for AChE and BChE complexes, respectively, was found to reliably be one of the greatest amongst the ligands tested.

Radius of gyration values paint a different picture that the previous two analyses. In two of the AChE complex simulations, the bound state was predicted to be higher in R_g values than the unbound state for the full extent of the 100 ns simulation, indicating unfolding of the protein was required to account for thebaine (Figures S6o and S10o). For bound AChE, R_g values were found to trend closer to the unbound AChE near the tail end of the run. Bound BChE was found to reduce the R_g relative to the uninhibited enzyme throughout the production runs. With that noted, the R_g were found to fluctuate more frequently than AChE in two of the runs, potentially indicating unfolding of BChE as well due to the presence of thebaine (Figure 9d and Figure S10p).

When considering ΔG, we note that on average, the values for the BChE complexes are more consistently negative than those obtained from the AChE complexes. The range of ΔG values was found to be narrower in the AChE cases than the BChE cases. Though greater fluctuation in estimated ΔG was greater in the BChE–thebaine complex, as the simulations progressed, ΔG values trended consistently toward more positive values (Figure 9f, Figures S7p and S11p). AChE complexes too trended towards positive values on average toward the end of the simulation, but ΔG values were found to marginally decrease near the middle of the production runs (Figure 9e, Figures S7o and S11o).

2.2. Binding Modes and RMSF Results

2.2.1. W6R and 4BDS in Pure and Crystal State

The crystal structures of the 1W6R and 4BDS enzymes from RCSB Protein Data Bank include water molecules, small molecules, and ions. The pure state only keeps the enzyme itself, which is also used for ligand docking. The pure state and original state show very similar RMSF values for both 1W6R and 4BDS. It is interesting the residues with large RMSF values are mainly located on the surface of enzyme (red tubes in Figure 10d,f), and the largest RMSF values appear at residues 378 and 379, which are originally actually missing from 4BDS crystal structure.

2.2.2. Afzelechin and Aspalathin

The RMSF plots of the ligands afzelechin and aspalathin resemble the RMSF plot of the enzyme very well, but the confirmations of afzelechin and aspalathin show significant differences. Aspalathin extends its glucose ring out of the gorge in both 1W6R and 4BDS (Figure 11c–f). It is also clear that afzelechin has more space to maneuver in 4BDS than in 1W6R (Figure 11c,e).

2.2.3. D-Maslinic Acid

The docking with different searching boxes produces significantly different conformers for D-Maslinic acid in 1W6R but gives almost identical conformers in 4BDS. To use a small searching box in docking locates a conformer of D-Maslinic acid to be inside the gorge of 1W6R and the RMSF plot around residues 105, 285, 434, and 485. To enlarge the searching box results in the out-of-gorge location of D-Maslinic acid and the improved RMSF plot in agreement with pure enzyme 1W6R (Figure 12). Both small and large search boxes produce the same conformer of D-Maslinic acid in 4BDS, and the conformer could stay in 4BDS without causing larger RMSF values on certain residues of 4BDS.

2.2.4. Isoliensinine

Docking produces two different preferred conformers in 1W6R and one conformer in 4BDS using two different search boxes. However, the RMSF plots show similarity between all conformer complexes and the pure enzyme (Figure 13). The results indicate 1W6R can accommodate the double-fold isoline at the bottom of gorge and one extended conformer through the entrance to better fit into the gorge of enzymes, while 4BDS could accommodate a double fold conformer around the entrance of the gorge. The flexibility of isoliensinine does not cause significant RMSF changes in either enzyme, though isoliensinine docked using the search box of 22 × 24 × 28 ų does increase the RMSF relative to the extended conformation associated with the of 40 × 40 × 40 ų search box.

2.2.5. Luteolin and Matricin

The Docking produces similar conformers for luteolin and matricin in both 1W6R and 4BDS, but the orientations in the gorge are different. Luteolin is docked deeper within 1W6R than matricin, while both ligands are intertwined in 4BDS, indicating a more flexible gorge in 4BDS. The RMSF plots indicate strong similarity between pure enzyme and enzyme complexes for these ligands (Figure 14).

2.2.6. Sedanolide and Thebaine

Sedanolide and thebaine can fit into the bottom of the gorge in 1W6R, while sedanolide moves slightly outside the gorge relative to thebaine in 4BDS. In terms of RMSF plots, the ligand–enzyme complexes follow the same pattern as pure enzymes. Thebaine has slightly stronger RMSF values than pure enzyme in 1W6R than in 4BDS, while sedanolide roughly keeps the same level as the pure enzyme (Figure 15).

3. Discussion

3.1. Previous Results and Docking

We studied molecular docking predicted ligand selectivity against AChE/BChE in a previous article [55]. Following docking, we had chosen four representative ligands, ambrisentan, ergotamine, ZINC_253700110, and caffeine, to run MD simulations. The results obtained for these ligands are in agreement with those discussed above. The ligand ZINC_253700110, with its relatively large MW (934 Da), and ergotamine, with an MW of 581 Da and extended structure in one dimension, demonstrated a typical RMSD spike with BChE and greater fluctuation of Coulombic intermolecular energy in the BChE case (Figures S3, S5, S7, and S9 from [55]). Results for ambrisentan, with its slightly smaller MW (380 Da) and comparable 3-dimensional expansion, demonstrate the rigidity of AChE and flexibility of BChE (Figures S4, S8, S14 and S15 from [55]). Caffeine, with its small size (194 Da), can move relatively freely in AChE and BChE with no clear preference (Figures S2, S6, S10 and S11 from [55]). Caffeine was experimentally found to inhibit AChE but not BChE [56], possibly due to its blockage of the gorge entrance in AChE or trapping within the miniature gorge distinct to AChE that surrounds the catalytic triad within the main gorge, potentially maximizing the frequency of π-π interactions between caffeine and the residues of said triad. This difference in the tertiary structure between ChEs may be an important factor in ChE selectivity as previous frequency analysis of docking results has shown that nearly the exact same binding site residues are involved in interactions when binding a large dataset of ligands to each ChE [55].

The binding affinities of the most stable pose of ligands (

Table 1. The binding affinity (kcal/mol) of the most stable pose ¹.

Ligands 2	AChE 3	BChE 3
afzelechin (274.3)	−9.7 (−9.7)	−8.7 (−8.7)
aspalathin (452.4)	−8.9 (−9.2)	−9.2 (−9.5)
D-maslinic acid (472.7)	−1.1 (−9.0)	−11.0 (−11.0)
isoliensinine (610.7)	−1.5 (−10.9)	−11.2 (−11.3)
luteolin (286.2)	−10.6 (−10.6)	−9.1(−9.1)
matricin (306.4)	−8.8 (−8.7)	−8.6 (−8.6)
sedanolide (194.3)	−7.8 (−7.9)	−6.6 (−6.8)
thebaine (311.4)	−9.5 (−9.7)	−8.9 (−8.9)

¹ The search box size was 22 × 24 × 28 ų. ² MW in Da in parenthesis. ³ The values in parentheses are from the search box size of 40 × 40 × 40 ų.

) indicate docking-predicted BChE selectivity for D-maslinic acid and isoliensinine when the search box is restricted to the area of the gorge surrounding the catalytic triad. To dock with larger search box (40 × 40 × 40 ų) allows isoliensinine to bind further towards the gorge entrance of AChE, increasing selectivity toward AChE, but the rigidity of D-maslinic acid still makes it 2.0 kcal/mol more favorable for BChE than AChE. The ligand flexibility and medium size (<300 Da) aid in allowing afzelechin and luteolin to be predicted as the most AChE-selective via molecular docking. The rigidity of matricin and thebaine and the large size of aspalathin reduces their relative AChE selectivity against BChE. With that said, our previous study showed that molecular docking is often unable to account for the known selectivity between ChEs for many ligands. Most notably is caffeine, which only has a difference in docking-predicted binding affinity of a less than two kcal/mol, regardless of box size [55]. The interaction energies and docking binding affinity for sedanolide seem to support AChE selectivity, but the magnitudes are smaller than other ligands’ results obtained in this study, and MD simulations do not show obvious preferences for AChE and BChE otherwise, similar to results previously obtained for caffeine. Our kinetic assay measurements confirmed that caffeine was AChE-selective through a non-competitive inhibition mechanism.

3.2. MW and Enzyme–Ligand Rigidity Relationship

A relationship between MW and enzyme–ligand rigidity was observed for our representative set of ligands. Isoliensinine, a flexible ligand with the largest MW (610.7 Da) tested, shows RMSD spikes in both enzyme–ligand complexes (Figure 5). D-maslinic acid is the second largest ligand in terms of MW (472.70 Da) and demonstrates RMSD peaks right away with BChE from 0 to 40 ns and after 20 ns with AChE (Figure 4b), which might be explained by the rigidity of D-maslinic acid’s structure and the rigidity of AChE gorge. The R_g plots for AChE (Figure 4c and Figure 5c) and for BChE (Figure 4d and Figure 5d) may also explain the necessity for ligand to migrate in AChE. Aspalathin also has a relatively high MW (452.4 Da) with respect to the Lipinski Rule of Five, but it has rotatable bonds between rings, potentially allowing its functional groups to interact with AChE’s triad. While the average ΔG is lesser for the AChE complex than BChE, we note that within the 100 ns, the lowest sampled ΔG value was found in the AChE case. To summarize, both the large size and rigidity play a role for how these three larger ligands spatially fit into both AChE and BChE (Figure 3, Figure 4 and Figure 5). Furthermore, the larger MWs contribute to the greater reported Lennard-Jones interactions (Figure 3a, Figure 4a and Figure 5a) for these three ligands.

Afzelechin (274.3 Da) and luteolin (286.2 Da) are comparable in size and number of functional groups with four hydroxyl groups each. According to the R_g plots, both AChE and BChE could accommodate these ligands and result in stabilization of the ChEs relative to those of the uninhibited ChEs. This is especially true for BChE, since the R_g values for the inhibited complexes were less than those of the uninhibited complexes (Figure 2c,d and Figure 6c,d). Alternatively, the RMSD values for these two ligands both have a distinct peak for each ChE, where the peak for BChE is notably more pronounced. In other words, both afzelechin and luteolin fit into the AChE’s gorge better than the BChE’s gorge given the willing migration from the docking-optimized binding conformation. Intermolecular interaction energies for these ligands and the ChEs were also similar. Coulombic interactions were predicted to be closer in magnitude to Lennard-Jones interactions than most other ligands. This trend is likely due to the hydroxyl groups substituted to these ligands structure, additionally leading to large fluctuation of the Coulombic interactions relative to the Lennard-Jones interactions. The ΔG values between these ligands and the ChEs were similar in their average magnitudes, though in the luteolin case, the ΔG values were found to decrease more significantly with both ChEs than those observed with afzelechin bound to the ChEs.

Matricin (306.4 Da) and thebaine (311.4 Da) are roughly comparable in MW and they can be considered relatively more symmetrical in both 2-dimensions and 3-dimensions than the other ligands tested in this study. Furthermore, they are relatively more rigid than other ligands on account of their lack of torsionable bonds. Though matricin nor thebaine demonstrates large RMSD fluctuations for the majority of the steps in their respective runs, both matricin and thebaine show large RMSD spikes at about 10 ns during the simulations (Figure 7b and Figure 9b). Furthermore, the R_g plots (Figure 7c,d and Figure 9c,d) indicate both matricin and thebaine induce adjustments in the ChEs to eventually stabilize AChE and BChE comparable to the uninhibited ChEs. This notion has further evidence in ΔG plots (Figure 7e,f and Figure 9e,f) where the AChE case tends to trend towards more negative ΔG values as ΔG values trend toward positive values in the BChE case. Short-range intermolecular interactions for these ligands shared a similar trend as well. The average Coulombic contributions for these ligands were close in magnitude regardless of whether AChE or BChE was the enzyme. However, for the Lennard-Jones contributions between these ligands and the ChEs, both ligands were found to have more favorable interactions with AChE than BChE.

The smallest ligand sedanolide (194.3 Da) had the weakest reported intermolecular interactions amongst the studied ligands. Furthermore, the Lennard-Jones contributions were found to be approximately equal in magnitude throughout the production run. A variance between ChEs arose with respect to the Coulombic interactions, which were found to be decrease in the AChE case at the beginning and end of the run. The RMSD and R_g plots (Figure 8) show sedanolide may not be restricted to the gorge but migrate out of it due to its small size and less than ideal contacts with the residues of the catalytic triad (Figure S3). R_g results for this molecule and AChE demonstrate how the small size of sedanolide and its migration around the gorge as highlighted by RMSD results prevent the AChE from stabilizing relative to the uninhibited enzyme on account of AChE’s relative inflexibility. This behavior is also reflected in ΔG plot which shows larger fluctuation in the BChE case than AChE. Also reflected in ΔG values for this ligand is that in the later stages of the production run, negative ΔG values were more prevalent in the AChE case rather than the BChE case, agreeing with docking-predicted affinity. These trends in ΔG values are similar to that seen in the case of caffeine (Figure S2b,c).

3.3. Trends in MD Results of ChE–Ligand Complexes

When using the Vina-predicted lowest-energy binding configurations as the starting coordinates for an enzyme–ligand molecular dynamics run, interesting patterns emerged when plotting the results of the same ligand against both ChEs. Perhaps most noticeable is the seemingly similar RMSD plots of certain ligands in complex with AChE and BChE. For example, consider Figure 5b and Figure 6b. It appears as if the RMSD spikes in the BChE–ligand complexes occur around a similar frame that the RMSD spikes in the AChE–ligand complexes. The difference lies in that the RMSD spikes in the BChE–ligand complex are much greater than those in the AChE–ligand complex. This is likely due to the higher flexibility of BChE and the ability for the ligand to move freely around the active site gorge, as the pocket surrounding the catalytic triad in AChE is more enclosed than that of BChE. Overall, RMSD values of ligand heavy atoms were found to deviate more greatly in BChE complexes than AChE complexes. Notable exceptions to this trend are results for sedanolide and D-maslinic acid.

Similarities were also observed when short-range interaction energies were evaluated with the most notable being Figure 5a and Figure 7a. Isoliensinine exhibits both trends, implying it may be a likely dual inhibitor that is slightly selective towards AChE on account of Figure 5a. This agrees with previous research, where isoliensinine was found to be an inhibitor of both ChEs with and IC₅₀ of 6.82 μM and 15.51 μM against AChE and BChE, respectively [40]. Though the Coulombic interactions appear similar in magnitude for thebaine, there is a near 50 kJ/mol gap in Lennard-Jones interaction, suggesting AChE selectivity and aligning with trends observed with other opiates during docking. With that noted, not all enzyme–ligand complexes intermolecular energies give a clear indication of selectivity one way or another and require further explanation. Some such examples are caffeine and sedanolide, whose Lennard-Jones and Coulombic interaction energies remained nearly constant across the simulation. Luteolin and sedanolide will be able to form π-π interactions with the residues of the catalytic triad like caffeine and the similarity in the two compounds interaction energy trends and existence of a pocket near the triad in AChE may trap the small molecule sedanolide in a similar way to caffeine and lead sedanolide to be selective towards AChE over BChE.

The radius of gyration results across tested ligands supports experimental findings that suggest BChE is more flexible than AChE. For most compounds considered, the R_g of uninhibited BChE was close in magnitude to the inhibited BChE values, while the R_g of uninhibited AChE was markedly lower in magnitude than the R_g of inhibited AChE until the end of the 100 ns simulations. As the active site gorge of BChE lacks the rigid pocket around the catalytic triad that AChE has, our R_g results suggest BChE does not need to necessarily distort its conformation and unfold to allow relatively large ligands into its active site in the same way as AChE. In the case of luteolin, however, the inhibited ChEs’ R_g values do not appear to vary greatly in magnitude from either of the uninhibited ChEs. Due to its similarity in RMSD values as seen in isoliensinine, large fluctuations in intermolecular energy in the BChE complex, and the AChE catalytic triad pocket, results indicate luteolin is selective towards AChE as well, in agreement with experiment [44].

To begin to consider the estimated ΔG values, it is important to note that the LIE approximation that was used will be slightly biased towards BChE than AChE. Up to this point, our MD results have found that BChE is significantly more flexible than AChE when accommodating larger ligands. The flexibility in the BChE enzyme and the discussed ability for our set of ligands to match or reduce the radius of gyration values of uninhibited BChE indicate that BChE is less perturbed by the introduction of a ligand to it’s binding site than AChE is. As such, the LIE method is prone to over- and underestimating the ΔG values for the BChE case, which is reflected in the fact the range for BChE–ligand ΔG values was found to be broader than the AChE–ligand cases. In any event, the results from our tested enzyme–ligand complexes confirm experimentally determined greater flexibility of BChE relative to AChE. Moreover, for the ligands sedanolide and aspalathin, the ΔG of the BChE–ligand complex is shown to noticeably increase near the end of the simulations, while AChE–ligand ΔG values are found to decrease. These results match those obtained for caffeine (Figure S2c,d) and isoliensinine (Figure 5e,f), who are both known to be selective toward AChE, indicating these two ligands potential selectivity towards AChE over BChE as well.

3.4. BChE Selectivity in the Literature

While most ligands in Table 1 have been previously found to be AChE-selective through in vitro experimentations or are predicted to be favorable toward AChE following analysis of docking and MD results, several BChE-selective ligands have been reported by others in the literature. In particular, bis(n)-lophine analogs were found to be strongly selective to BChE, for none of them inhibit AChE [7]. While these compounds have been found to be less toxic to liver cells than tacrine, an infamous and defunct AChE inhibitor, no further toxicity results have been established. Notably, these compounds bear both inflexible structures and large MWs, preventing them from easily inhibiting the inflexible AChE over BChE. The ligand N-isobutyl-N-((2-(p-tolyloxymethyl)thiazol-4-yl)methyl)benzo[d][1,3]dioxole-5-carboxamide interacts with several residues in active site gorge of BChE, which contributes to its BChE selectivity [8]. It is likely that another contributing factor of this highly substituted compound’s selectivity for BChE is due to its low number of molecular torsions, which prevents the compound from snaking one of its functional groups into the enclosed space surrounding the AChE catalytic triad and thus prevents effectively binding in that key region of the enzyme. Additionally, δ-sultone-fused pyrazoles were optimized to be BChE inhibitors by mimicking and slightly modifying the hydrophobic interactions of donepezil, a commonly prescribed and FDA-approved AChE inhibitor also known to inhibit BChE, at the PAS [9,10,11]. N-benzylpyridinium moieties linked to arylisoxazole derivatives were found to be selective BChE inhibitors [12]. These compounds generally carry a high MW and rigid structure with a low number of torsions, benefitting from the flexibility of BChE relative to AChE in its selectivity of the former.

While there are far more selective BChE inhibitors than those noted here, there are currently no FDA-approved BChE inhibitors for the mitigation of AD nor have any such inhibitors entered clinical trials to this point. One source of hesitation may arise from the large presence of BChE in serum, the lungs, and other bodily tissues, while AChE is primarily located at the PSNMJ and is present in lesser concentration overall in vivo [57]. Selective inhibition of BChE therefore would undoubtably affect these other systems in ways which have yet to be comprehensively studied. Despite such challenges, the fact that BChE concentration levels are shown to rise as AChE levels decrease in late-stage AD patients and their ability to affect the proliferation of amyloid plaques drives the inquest into whether these inhibitors may offer refuge to patients when in the final stage of the disease [14,15].

4. Materials and Methods

4.1. Molecular File Preparation

The molecular files for Torpedo californica AChE (PDB: 1W6R) [58] and Human BChE (PDB: 4BDS) [59] were sourced from the RSCB database. These species were chosen to match those in our laboratory at the time, where human was used as substitute for equine. MD preparation was carried out using GROMACS 2022.2 open-source software [22]. The chosen ligand’s lowest energy binding poses as obtained from our previous study were taken to be the starting coordinates of the ligand for the simulation [54]. After removing the structures of all other binding modes from the files, we used the CGenFF server to generate topology files of our ligands [60]. The CHARMM forcefield has found success as the forcefield for proteins and will be chosen as the forcefield in this aspect of the experimental setup as well. The ligand and enzyme files’ coordinates are combined into complexes. The complexes were then solvated with water molecules. Ions of sodium and chlorine were added to the complexes to mimic the ionic environment of the PSNMJ and balance any charges in the ChEs. Two complexes of uninhibited AChE and uninhibited BChE were also prepared for which to compare RMSD and radius of gyration values to inhibited complexes.

4.2. Dynamics Parameters

For each of the complexes, an energy minimization step was performed for 100 ps with an energy threshold set at 10 kJ/mol. Subsequent NVT equilibration was performed using a reference temperature of 310.15 K. NPT equilibration steps were carried out with the Berendsen thermostat set for 310.15 K and the Parrinello-Rahman barostat was used for this equilibration step. The production runs were carried out for 100 ns per ligand using the GROMACS modules installed on Bridges 2 at the Pittsburgh Supercomputer Center. Three sets of simulations, including equilibration steps, were performed per enzyme–ligand complex (Figures S4–S11).

4.3. Post-Production Run Analysis

Post-simulation analysis was performed with a few programs. To generate the .xvg files necessary to plot the interaction energies, RMSD, and R_g values for each complex, GROMACS command line tools were utilized. For interaction energies, SR-LJ and SR-Coul were obtained, and RMSD of the heavy atoms of ligands were obtained. For R_g, the average value is provided. A series of ad hoc Python scripts were developed to generate plots from the .xvg files using the matplotlib library. Plots were constructed to contain multiple complexes to aid in the identification of notable differences between complexes through the course of the 100 ns simulations.

4.4. LIE Approximation of Gibbs Free Energy of Binding

We perform a free energy estimate using the LIE approximation as implemented in GROMACS. In addition, we constructed an equivalent LIE program in Python 3.11 which uses terms from a GROMACS interaction energy calculation and allows one to specify which energy terms are included in the approximation. Alpha and beta values were kept at default values 0.181 and 0.5, respectively. For the Lennard-Jones interaction between ligand and solvent (E_lj) and the Coulomb interaction between ligand and solvent (E_qq) we solvated the ligands and performed NVT and NPT equilibration using the same atom restraints and parameters as those used for the enzyme–ligand complex simulations. A 10 ns production run was carried out for each ligand and the E_lj and E_qq values necessary for the LIE calculation were taken to be those values found at the conclusion of the run. Values obtained for E_lj and E_qq are listed in the SI (Table S1).

4.5. Ligand Docking

The open-source software AutoDock Vina 1.2 was employed to carry out the molecular docking [18]. MGLTools 1.5.7 software was used to prepare .pdbqt files for both receptors and ligands [17]. During the preparation process, the missing or incomplete residues were repaired, and the polar hydrogens were added. In addition, the water molecules and substrate small molecules in the PDB files were removed, and Koopmans’ charges were added to all the atoms of the respective enzymes. The search box was centered at the center of ligands in the original PDB files (galantamine in 1W6R and tacrine in 4BDS), initially with 22 × 24 × 28 ų grid box then with 40 × 40 × 40 ų grid box. Nine conformers were sought in the docking process, and the exhaustiveness value was set to be 8. Only the most favorable mode was further analyzed. Note the docking parameters here were not those used to generate the initial ligand configurations for MD [55].

4.6. Root Mean Squared Fluctuation

Molecular dynamics simulations for RMSF studies were conducted using a web-based protein structure modeling tool CABS-flex 2.0 [61,62]. The water molecules and substrates in the PDB files for 1W6R and 4BDS were removed to get pure enzyme PDB files, and the coordinate and charges of the most favorable docking pose were appended to pure enzyme PDB files to get complex PDB files. Only these PDB files were needed for the CABS-flex simulation. The simulation involved 50 cycles, and the A chain was selected for analysis. The RMSF plots were prepared with Igor Pro 9.0 program [63], and the graphical representations were obtained using VMD 1.9.3 software [64].

5. Conclusions

The short-range intermolecular interactions, RMSD, RMSF, radius of gyration of inhibited and uninhibited proteins, and estimated Gibbs free energy of binding are found for eight representative secondary metabolite compounds following molecular dynamics simulations. The results confirm BChE is more flexible than AChE and show AChE has a more rigid catalytic gorge near the catalytic triad as opposed to that of BChE. As such, a large MW, expansion, and rigidity of ligands were found to induce significant changes in collected data for AChE and BChE from the MD production runs. In general, it was found that larger size and increased rigidity of ligands contribute to BChE selectivity, while better fitting and flexibility of ligands contribute to AChE selectivity. The flexibility of ligands and ChE was seen to affect LIE estimates of the ΔG, in accordance with flexibility expectations of ChE–ligand complexes and the constraints of the LIE method. The BChE enzyme unfolded less frequently through the addition of the representative ligands when compared to uninhibited BChE, while for AChE, the inhibited enzymes were able to match the radius of gyration of uninhibited ChE by the conclusion of the production runs. For smaller ligands, large deviation from docking-predicted binding locations and conformations was observed for both ChEs, suggesting readily migration of these ligands when solvent and dynamics are allowed and contributing to under/overestimations in ChE selectivity obtained though docking procedures for smaller ligands. The RMSF plots using free enzymes and favorable docking poses of ligand–enzyme complexes indicate that D-maslinic acid, out of eight studied ligands, induced the most noticeable changes in enzyme structure.

The MD results suggesting isoliensinine as AChE-selective agree with the experimental values [40]. We recall that molecular docking restricted to the area of the active site gorge containing the catalytic triad results in a BChE-selective prediction for isoliensinine, while expanding the docking search area to the gorge opening results in an AChE-selective prediction. This result and those for the smaller ligands considered in this study suggest the ability to bind nearer to the binding site entrance plays a role when it comes to selectivity between ChEs for certain ligands. Two of our compounds, luteolin and matricin, are both present in relatively large quantities in chamomile and are suggested to be selective towards AChE from docking and MD simulations. This agrees with previously performed in vitro kinetics studies for luteolin [44], though the IC₅₀ values for matricin for both ChEs have yet to be established. Chamomile is often brewed into tea and is known to have many therapeutic effects, such as managing inflammation, diabetes, certain cancers, and acid reflux. A host of other compounds derived from chamomile have also been shown to inhibit the ChEs, with varying selectivity [44]. Though other phytochemicals derived from plants whose components are commonly brewed into teas were evaluated in silico, there is a lack of comprehensive kinetics results for these ligands which exist in the literature at this time.