Computational identification of novel acetylcholinesterase inhibitors as potential treatments for Alzheimer's disease Authors: Sebastián Matallana Rincón, Fabián Orozco López, Johan F Galindo Published: 2026 ABSTRACT BackgroundAlzheimer's disease represents a major public health issue that affects millions of people worldwide. Although symptomatic treatments are available, they neither prevent nor halt disease progression; therefore, it is necessary to develop new therapeutic alternatives.ObjectiveTo identify acetylcholinesterase inhibitors with potential biological activity through a computational protocol.MethodsIn this study, a computational approach based on virtual screening, molecular docking, and molecular dynamics simulations was applied to identify new potential acetylcholinesterase inhibitors.ResultsThe results allowed the identification of three compounds with higher binding affinities than donepezil, which was used as a reference. Among them, ligand code 24771824 stood out for establishing hydrophobic and aromatic interactions that maximize dispersive contributions and promote a rigid and stable conformation within the active site. In contrast, ligand codes 151171 and 21081761 were favored by more directional polar contacts, which increased specificity but limited the overall affinity toward the enzyme.ConclusionsAltogether, the free energy, structural fluctuation, hydrogen bond occupancy, and molecular clustering analyses suggest that 24771824 exhibits the most favorable energetic and dynamic behavior, consolidating it as the best candidate for future experimental validation. Introduction Alzheimer's disease (AD) is a neurodegenerative disorder associated with functional, behavioral, and cognitive impairments, characterized by the progressive accumulation of extracellular amyloid-β (Aβ) plaques and intracellular neurofibrillary tangles (NFTs), which disrupt synaptic signaling,1,2 leading to neuronal death and resulting in inflammation and atrophy of brain tissue.2–4 Aβ plaque deposits in the brain can be generated through either reducing Aβ clearance or excessive production, while NFTs are formed as a consequence of the abnormal accumulation of hyperphosphorylated tau protein.5,6 AD causes damage to cognitive and behavioral functions, which include comprehension, memory, attention, language, reasoning, and judgment.7,8 Although, it does not directly cause death, it considerably increases the clinical and functional vulnerability of the patient, favoring the development of critical comorbidities (metabolic, cardiovascular, respiratory, infectious) that can indeed be lethal.1,9 Nowadays, several hypotheses have been proposed regarding the causes of AD; one of them is the cholinergic hypothesis, which states that AD is due to a reduction in the biosynthesis of acetylcholine (ACh), a neurotransmitter involved in sending signals to cells, controlling memory, and the action of some muscles. It is also noted that inhibition of the enzyme acetylcholinesterase (AChE), through increasing cholinergic levels, assists cognitive function and therefore serves as a therapeutic strategy against AD.10,11 Currently, symptomatic treatments for AD are limited to acetylcholinesterase inhibitors (AChEIs) and an NMDA (N-methyl-D-aspartate) receptor antagonist.7,10 Tacrine was the first reported compound against AChE in 1993, but it was withdrawn from the market due to high hepatotoxicity; 12 since then, other inhibitors such as donepezil (DNP), rivastigmine, and galantamine have been proposed, but due to gastrointestinal side effects, are administered mildly and, consequently, the overall therapeutic impact remains restricted.7,13 It is also noteworthy that these inhibitors are effective against AD symptoms, but they do not prevent or cure the disease. Furthermore, several clinical studies aimed at developing drugs against Aβ plaques and reducing tau protein accumulation have been initiated; however, the results have been modest, and certain trials have not achieved the expected endpoints. 14 For this reason, there is a need to search for new compounds that may be more efficient AChE inhibitors than those already reported. Currently, the therapeutic approach to AD has evolved from exclusively symptomatic strategies toward the development of disease-modifying therapies aimed at intervening in the biological processes responsible for its progression. These approaches include a variety of strategies, such as the inhibition of Aβ aggregation through BACE1 (beta-site amyloid precursor protein cleaving enzyme 1) inhibitors, monoclonal antibodies, and vaccines, as well as therapies targeting tau protein, immune system modulation, neuroprotection, and neuronal repair. In addition, non-pharmacological interventions such as brain stimulation, physical exercise, and cognitive training have gained importance in disease management. However, despite recent advances, particularly with anti-Aβ therapies, clinical benefits remain limited, and there are concerns related to cost, safety, and clinical applicability. In this context, although a wide range of strategies exists, most of them show limited clinical efficacy or are still under investigation, highlighting the need to continue developing new therapeutic alternatives. 15 AChE is a cholinergic enzyme located at postsynaptic neuromuscular junctions, typically in nerves and muscles. Its main function is the hydrolysis of ACh to choline, which leads to the termination of cholinergic signaling. 16 It is also involved in nervous system development and is expressed in developing neurons, particularly during the period of axonal growth.17,18 AChE is found in cholinergic neurons, where it participates in the termination of the nerve signal through the degradation of acetylcholine. It is also present in red blood cells and in cholinergic synapses, specifically in the postsynaptic membrane region, where it performs its catalytic function.19,20 The neurotransmitter ACh is the endogenous substrate of AChE, which is found in both the peripheral and central nervous systems and it is the predominant transmitter in autonomic ganglia, the neuromuscular junction, and parasympathetic nerve endings. 21 During the progression of AD, many types of neurons are damaged and there is a profound loss of cholinergic neurons in the forebrain, resulting in a progressive decrease in ACh. Additionally, increased levels of AChE around Aβ plaques and NFTs are considered a common feature of AD neuropathology.22,23 Considering the central role of AChE in the nervous system and in the pathophysiology of AD, our work focuses on finding new AChE inhibitors through computational methods, with the aim of identifying compounds with higher affinity and improved pharmacokinetic properties compared to donepezil, which was used as the reference compound. First, a virtual screening approach was applied to a compound database to identify molecules with similar pharmacophoric features to the reference inhibitor DNP. The selected ligands were evaluated by molecular docking and a scoring consensus to predict their binding affinity. The compounds with the highest scoring values were filtered based on predictive pharmacokinetic studies, including validation using the BOILED-Egg model. Finally, molecular dynamics simulations were carried out to analyze the stability and structural behavior of the enzyme–ligand complexes within the AChE active site. The per-residue root mean square fluctuation (RMSF), binding free energy estimation using the MM/GBSA (Molecular Mechanics/Generalized Born Surface Area) method, per-residue energy decomposition, hydrogen bond persistence along the trajectory, and interaction analysis through molecular clustering were evaluated. The obtained results allowed the identification of compounds with greater stability and affinity than DNP, suggesting higher therapeutic potential and motivating their evaluation in future in vitro assays. Methods System and database selection The crystallographic structure of the acetylcholinesterase enzyme and co-crystallized with DNP (a clinically approved inhibitor), obtained from the Protein Data Bank (PDB ID: 7E3H, https://doi.org/10.2210/pdb7E3H/pdb), 24 was used as a reference to validate the molecular docking protocol and to compare the affinity and interactions of the candidate compounds. The database consisted of the GESACH group's chemotheque, consisting of 135 previously synthesized heterocyclic molecules, which is not publicly available and was exclusively provided for this work, and the DrugBank database, 25 which contains 9716 bioactive, clinically approved compounds and was incorporated in order to consider molecules with known pharmacological properties and to expand chemical diversity. In total, an initial dataset of 9851 compounds was obtained. The original codes from the DrugBank database were retained and used for their identification throughout the study. As exclusion criteria, a data curation process was carried out using OpenBabel 26 and RDKit, 27 with the aim of ensuring the quality of the dataset. In this stage, duplicate molecules were removed, as well as compounds containing highly reactive functional groups such as nitro groups, reactive halogens, thiols, and other fragments that may interfere with computational analyses or subsequent experimental stages. Additionally, structures containing PAINS (Pan-Assay Interference Compounds) patterns, known to produce false positives in biological assays, were excluded. As a result of this curation process, 8218 compounds were selected for the subsequent stages of the study. Pharmacophore model Starting from the 8218 compounds obtained from the curation process, a pharmacophore model was constructed using the PDB structure with code 7E3H to identify the essential geometric and chemical features involved in ligand binding to the active site of acetylcholinesterase. To validate the methodology a re-docking was performed using DOCK6, 28 validating the protocol through RMSD calculation with the DockRMSD web server, 29 establishing RMSD values below 2 Å as the validation criterion. The results were analyzed in Discovery Studio and Proteins Plus. 30 Subsequently, the DNP structure was uploaded to ZINCpharmer, 31 with the purpose of generating a pharmacophore based on the real geometry and spatial arrangement of the ligand in its bioactive conformation. This allowed the server to filter the databases and identify molecules capable of reproducing the interactions of the reference ligand DNP. As a result of this filtering step, 89 compounds were selected for subsequent analyses. Molecular docking and ADMET properties The 89 compounds that passed the previous filtering step were evaluated through molecular docking using AutoDock, 32 AutoDock Vina 33 and DOCK6. These three programs use different scoring functions and search algorithms, which allows a more complete evaluation of receptor–ligand interactions. AutoDock4 employs a semi-empirical scoring function and a genetic search algorithm. AutoDock Vina uses a hybrid scoring function, while DOCK6 uses a force-field–based scoring function, in which the energetic parameters are taken from the Amber force fields. 34 For the molecular docking studies, the crystallographic structure of human AChE (PDB ID: 7E3H) was used. Protein preparation was carried out in Chimera 35 by adding hydrogens, partial charges, and assigning the protonation states of histidine residues. In the AutoDock Vina and DOCK6 protocols, the AMBER ff14SB force field 36 was employed, whereas Kollman charges were used in AutoDock4 through the integrated PyRx tools. On the other hand, ligand preparation included the conversion of their structures into pdbqt and mol2 files using OpenBabel and AutoDock Tools. Subsequently, multiligand docking calculations were performed for all selected compounds and the reference ligand DNP, generating an affinity ranking based on the obtained binding energies. The grid in each software was adjusted according to the size of the co-crystallized ligand and the size of the grid box were 53 Å, 42 Å and 49 Å for x, y and z directions and center at −43.740 Å, 37.597 Å, and −30.039 Å, respectively. Additionally, each software was validated by calculating the RMSD (Supplemental Table 1), obtaining values below 2 Å, which guarantees the reliability of the protocol. In addition, an autoscaling consensus scoring approach was performed, in which each score value from the three programs was scaled to a value between 0 and 1 using Equation 1: 37 (1) Xscaled=(x−xmin)(xmax−xmin) where Xscaled is the autoscaled value for each ligand, x is the score value of each ligand, xmin is the smallest score value of all the ligands evaluated, and xmax is the highest score value of all the ligands tested. Finally, an average of all autoscaled values was calculated, obtaining the Average Autoscaled Score (AAS) through Equation 2: (2) AAS=XAutoDock+XAutoDockVina+XDock63 According to the consensus scoring results obtained from the three molecular docking programs, 17 compounds were selected for subsequent ADMET evaluation. The structures and names of the compounds are shown in Supplemental Table 2. To evaluate the predictive capability of the virtual screening protocol, a retrospective validation was performed using a set of 17 known AChE inhibitors reported in the literature and 810 decoy compounds. All compounds were subjected to the same computational workflow employed during the virtual screening process, including molecular docking with AutoDock4, DOCK6, AutoDock Vina, and the application of an autoscaled consensus scoring approach. The discriminative performance of the protocol was assessed through receiver operating characteristic (ROC) curve analysis, considering the known inhibitors as the positive class and the decoy compounds as the negative class. Overall performance was quantified using the area under the curve (AUC).38,39 Additionally, enrichment factors at 1%, 5%, and 10% (EF1%, EF5%, and EF10%) were calculated to evaluate the ability of the protocol to recover active compounds within the top positions of the final ranking. These metrics were determined by comparing the proportion of active compounds recovered in a given fraction of the ranked dataset with the proportion of active compounds present in the complete dataset. Subsequently, the physicochemical descriptors and ADMET properties were evaluated using SwissADME, 40 including gastrointestinal absorption, blood–brain barrier permeability, P-glycoprotein substrate prediction, CYP450 isoform inhibition, and BOILED-Egg analysis. 41 Additionally, the potential cardiotoxicity associated with hERG channel inhibition was rapidly assessed using admetSAR, 42 where the selected compounds showed no predicted hERG-related toxicity. This filter excluded compounds with good theoretical affinity but low probability of oral administration or central nervous system permeability, which are important characteristics for acetylcholinesterase inhibitors intended for neurodegenerative disease applications. Based on the predicted ADMET properties and docking performance, 3 compounds were selected to perform molecular dynamics simulations. Molecular dynamic simulations Molecular dynamics simulations were subsequently performed for the complexes formed by AChE with the 3 selected compounds and the reference ligand DNP. The system preparation was carried out in the Chimera software 35 at pH = 7.0 ± 1, adding partial charges, missing hydrogens, and removing the co-crystallized waters. In addition, H++ was used to determine the protonation states of the monomer residues. The missing crystallographic residues were added using MODELLER, 43 generating a complete and suitable protein topology for the simulation. The parametrization of the ligands was performed with the GaussView visualizer, and a geometry optimization was performed using the B3LYP/6-31G* level of theory using the Gaussian g16 program. 44 Furthermore, the point charges were developed according to the RESP procedure using the Antechamber program. 45 Five replicas of classical MD simulations were carried out, where the systems were initially minimized in 300,000 steps applying the steepest descent algorithm followed by 100,000 steps of conjugate gradients. The heating phase was carried out in an NVT ensemble with an integration step of 1 fs and Langevin coupling (γ = 2 ps−1), increasing the temperature gradually through three successive ramps of 0–100 K, 100–250 K, and 250–300 K until stabilizing the system within 500 ps. A production time of 400 ns were done for four replicas, while one replica with production time of 600 ns was performed, reaching a total simulation time of 2.2 µs for each molecule, using the AMBER force field (ff14SB version), 36 a saline concentration of 150 mM, and the OPC water model with a 10 Å rectangular box. The production stage was conducted in an NPT ensemble at 1 atm and 300 K, with a 2 fs integration step and hydrogen bond constraints using the SHAKE algorithm. The random seeds used for each replica are shown in Supplemental Table 3, to ensure the statistical independence and reproducibility of each simulation. The trajectory results were visualized in VMD (Visual Molecular Dynamics) 46 and the analyses in CPPTRAJ. 47 The analysis of the conformational changes obtained in MD was performed using the RMSD metric. Residue-wise RMSF differences were calculated with respect to the reference system and hydrogen bonds were also determined along the trajectory using a donor-acceptor distance cutoff of 3.0 Å and a donor – H – acceptor angle cutoff of 150°. Additionally, molecular clustering was performed using a hierarchical agglomerative algorithm, from which representative structures were obtained and molecular interaction diagrams were constructed, providing a more realistic view of ligand affinity and interactions within the active site. Finally, the MMPBSA.py script 48 was used to calculate binding free energy through the MMGBSA methodology. Results The database was built with 9851 compounds from DrugBank and from the GESACH group's compound library, which underwent a structural curation with RDKit that included: removal of duplicate compounds through their molecular fingerprints, elimination of compounds with PAINS (Pan–Assay Interference Compounds) patterns, and removal of undesired reactive substructures such as nitro groups, reactive halogens in certain positions, nitro oxides, thiols, and silanes, resulting in a final set of 8218 compounds. In addition, chemical cleaning was performed using OpenBabel, assigning protonation states at physiological pH, adding missing hydrogens, and generating 3D coordinates, thereby ensuring adequate preparation of the ligands. This curation guarantees rigorous cleaning and standardization of chemical data, since according to Nittinger et al. (2023), 49 there are random and systematic errors in databases, such as the incorrect translation of one-, two-, and three-dimensional structure representations, or manual entry errors in databases ranging from 0.1% to 3.4%. Other authors, such as Kelm et al. (2024), 50 mention that a large part of the drug discovery process is carried out in highly innovative but non-standardized and non-regulated research environments, which leads to low reproducibility and results in high attrition rates in preclinical and clinical stages. For this reason, by carrying out this curation process, the possibility of false positives is reduced, and the comparability and reliability of computational results are ensured. Re-docking and interaction analysis An initial interaction analysis of the crystallographic structure 7E3H was performed using DNP as reference, since it is a clinically reported inhibitor of AChE and it is used in the symptomatic treatment of AD. Therefore, by studying the interactions of DNP, we can obtain a robust model of the interactions essential for enzyme binding, which can help predict how other compounds may interact in the same active site. 51 Furthermore, it serves as a validation model to prove that the methodology responds appropriately. Figure 1 shows the interaction diagram of the AChE enzyme with DNP using Discovery Studio and Proteins Plus. In both cases, important interactions can be observed, such as hydrogen bonding with Phe-295, which is critical for ligand-receptor specificity and binding affinity. π–π stacking interactions are also found between the indanone portion of the ligand and Trp-286, as well as between the aromatic ring of DNP and Trp-86. Finally, hydrophobic interactions were identified with Tyr-341, Tyr-337, Phe-338, and Trp-86, which increase the biological activity and binding affinity of the inhibitory ligand. 52 Interaction diagram of the enzyme 7E3H with DNP using discovery studio (top) and proteins plus (bottom): hydrogen bonds (Phe 295), π–π stacking interactions (Trp 86 and Trp 286), π–alkyl interactions (Tyr 337, Tyr 72 and Phe 338), and C–H interactions (Tyr 341 and Ser 293) are indicated. The molecular docking protocol was validated using RMSD values for the re-docked and co-crystallized poses. A low RMSD (<2.0 Å) indicates that the protocol is able to accurately reproduce the orientation of the ligand within the active site of the enzyme, suggesting that the generated models are reliable for predicting this type of interactions. 53 A RMSD value of 0.379 Å was obtained for the re-docking of the receptor with donepezil (Supplemental Figure 1), which suggests a good reproducibility of the docking and validates the methodology used. The interaction analysis identified structural features of DNP that are important and contribute to its affinity for the receptor; therefore, they were considered for the creation of the pharmacophore model. These include the hydrogen bond acceptor in the indanone portion, the π-π stacking interactions present in the aromatic rings at the ends of DNP, and the hydrophobic interactions formed around them. Subsequently, the pharmacophore model was generated with the respective features, as shown in Figure 2, visualized as spheres with distinct colors for different interactions. Therefore, ligands with these characteristics, have a high probability of success in terms of specificity and affinity for the receptor, making it possible to develop more effective compounds with fewer side effects. 54 In addition, it has been reported that pharmacophore models, when complemented with molecular docking, increase the efficiency of virtual screening by reducing false positives and provide better results in the identification of lead compounds.54,55 After applying the filter to the previously created database, 89 compounds were selected for the molecular docking study. Pharmacophore model. The wireframe spheres represent the aromatic and hydrophobic features, and the sphere with an arrow the hydrogen bond acceptor. Molecular docking and ADMET properties Molecular docking was performed with the remaining 89 compounds in order to identify those with the best affinity toward AChE. Supplemental Figure 2 shows docking scores obtained with AutoDock4, AutoDock Vina, and DOCK6 as scatter plots, allowing visual identification of relationships between ligand affinities and comparison with the reference compound, in this case, DNP. Each software employs different algorithms and scoring functions, which lead to distinct values and varying ways of discriminating ligands. For example, in AutoDock4, the scores range from −6.15 to −13.24 kcal/mol, with most ligands showing scores similar to or better than DNP. This may be due to the fact that AutoDock4 uses a rigid receptor model, which limits conformational space and generates similar affinities, since the receptor cannot adjust to small conformational changes during binding. On the other hand, AutoDock Vina produced more dispersed scores because it is optimized for a faster conformational search. 56 Finally, DOCK6 applies a force field–based scoring function (vdW, electrostatics, and solvation). This higher accuracy in describing interactions can penalize certain conformations, which results in a more noticeable dispersion in the scores, as shown in Supplemental Figure 2C. When using different docking programs, incorrect ligand conformations in the enzyme may be selected because each one applies distinct scoring functions and sampling algorithms. For this reason, a consensus scoring approach was applied, integrating the scoring functions and sampling algorithms of each program, thereby compensating for the deficiencies of individual methods and improving ligand enrichment as well as the identification of correct conformations. 37 In this study, the autoscaling method described in equations (1) and (2) of the methodology was used. The complete docking results and AAS values for the 89 compounds are provided in Supplemental Tables 4 and 5. From the analysis of AAS values, 17 out of the 89 evaluated compounds outperformed DNP, suggesting a better theoretical affinity profile toward AChE. On the other hand, in order to validate the performance of the molecular docking protocol, a retrospective validation was carried out using a benchmark set composed of 17 known AChE inhibitors and 810 decoy compounds. The discriminative ability of the protocol was evaluated using ROC curves and enrichment factors, which are commonly employed to assess the effectiveness of virtual screening strategies.38,39 The ROC curve obtained yielded an AUC value of 0.650 (Supplemental Figure 3), indicating a moderate ability to distinguish between known inhibitors and decoy compounds. This result suggests that the combination of the three molecular docking programs through consensus scoring allows active compounds to be ranked more efficiently than would be expected by random selection. Although the AUC value does not indicate excellent discrimination, it demonstrates that the protocol is capable of identifying favorable trends toward the recovery of active compounds. Additionally, the enrichment factor analysis showed favorable performance in the top-ranked positions of the screening results. The obtained EF1%, EF5%, and EF10% values were 6.08, 3.56, and 2.97 respectively. These results indicate that active compounds are concentrated in the highest-ranking positions at a frequency considerably greater than that expected by chance. In particular, the EF1% value reveals that compounds classified within the top 1% of the ranking contain approximately six times more active inhibitors than would be expected from a random selection. This enrichment capability observed in the top-ranked positions represents a key characteristic of virtual screening studies, as it significantly reduces the number of compounds requiring further evaluation. 38 Therefore, although the overall discriminative performance of the protocol was moderate, the enrichment factors obtained demonstrate that the protocol is effective in prioritizing active compounds within the most relevant positions of the ranking. To complement the docking studies, the pharmacokinetic properties of these 17 compounds were evaluated using the SwissADME web server, which calculated Lipinski's rule of five and predicted gastrointestinal absorption and blood–brain barrier permeability of small molecules using the BOILED-Egg model. Full results are shown in Supplemental Table 6 and Supplemental Figure 4. Most compounds showed at least one violation of Lipinski's rules, while only two (21081761 and 24771824) fully complied with all established criteria, both were also located in the yellow region of the Boiled-Egg model, indicating a high probability of blood–brain barrier penetration. In total, four compounds showed potential for brain penetration, leading to the exclusion of the remaining 13. Although molecule 118666864 was able to cross the blood–brain barrier, it was excluded due to its molecular weight of 611.79 g/mol, which implies poor intestinal absorption and oral bioavailability. Additionally, its docking performance was similar to DNP, limiting its viability as a candidate. On the other hand, compound 151171 showed favorable results in docking and consensus scoring. Although it presented an mLogP value of 4.44, which is slightly above the threshold of 4.15, it is known that values within this range can still be consistent with good bioavailability and permeability, since the remaining ADMET properties were favorable. Consequently, compounds 24771824, 151171, and 21081761 (Figure 3) were considered the most promising candidates for further molecular dynamics studies. It is important to mention that these compounds correspond to previously reported molecules registered in DrugBank. In particular, compound 24771824 corresponds to PF-04457845, which has been previously investigated for the treatment of Tourette syndrome and cannabis dependence. Compound 151171 corresponds to Conivaptan, a clinically approved drug for the treatment of hyponatremia. On the other hand, compound 21081761 also corresponds to a previously reported chemical structure registered in DrugBank. These findings support the pharmacological and chemical relevance of the selected candidate compounds. Chemical structures of the candidate compounds selected for molecular dynamics studies. Molecular dynamics Stability analysis The conformational stability of the enzyme–ligand complex along the molecular dynamics trajectory was evaluated using RMSD calculations, which were performed considering the enzyme Cα atoms and using the first frame of the production stage as the reference, corresponding to the initial configuration of the system. Figure 4 shows the results of the five replicas performed for each of the studied systems. In the case of DNP (Figure 4A), the reference compound, each replica shows an increase in RMSD during the first 20 ns, which is expected since this corresponds to the initial structural accommodation phase of the system. Subsequently, all replicas exhibited minor conformational changes along the trajectory, with similar RMSD values (between 4 Å and 5 Å). Additionally, replica 3 maintains a lower RMSD (∼3 Å) compared to the others, indicating that it reached structural stability in a conformation close to the initial one. This conformational variability is normal when independent replicas are used, indicating that the system explores a diverse yet stable conformational space. 57 For the other systems (24771824, 21081761, and 151171), similar behaviors to AChE-DNP complex were observed, showing an initial phase with minor conformational changes that eventually stabilize at the end of each trajectory. In the case of compound 21081761, replica 3 (Figure 4C) was excluded from the analysis because the ligand detached from the active site around 200 ns of simulation, preventing its inclusion in the structural stability assessment. These results confirm the inherent stochastic nature of MD simulations, as some replicas may exhibit differences. Moreover, this ensures the reproducibility and reliability of the observed structural stability. 58 RMSD of the 5 molecular dynamics replicas for the AChE enzyme with: (A) DNP, (B) 24771824, (C) 21081761, and (D) 151171. Additionally, ligand RMSD was analyzed to evaluate their positional stability within the active site during molecular dynamics simulations (Supplemental Figure 5). In general terms, DNP and 24771824 showed relatively stable RMSD profiles across replicas, with moderate fluctuations compatible with conformational accommodation processes within the binding site. For compound 151171, greater variability among replicas was observed, suggesting increased conformational flexibility of the ligand within the active site, without clear evidence of dissociation. For compound 151171, greater variability was observed among the replicas. In particular, replica 2 exhibited a progressive increase in RMSD, reaching values close to 12 Å, which was associated with the displacement of the ligand from its initial binding site. However, it is important to note that the compound remained associated with the protein surface and established new interactions with nearby residues, including Val-73, Asp-74, Thr-75, Trp-286, Leu-281, Glu-250, Leu-251, Tyr-72, Val-280, Val-282, and Asn-283. This behavior suggests a relocation of the ligand toward an alternative interaction region rather than a complete dissociation of the protein–ligand complex. In the case of 21081761, most replicas showed moderate fluctuations consistent with an accommodation process within the binding site; however, one replica exhibits a pronounced increase in RMSD starting at approximately 150 ns, reaching values greater than 40 Å and remaining elevated throughout the rest of the trajectory. This behavior is consistent with ligand dissociation from the catalytic gorge and supports the exclusion of this replica from subsequent analyses. MMGBSA binding free energies Subsequently, the scores from the three molecular docking programs, the consensus scoring, and the binding free energies calculated by MMGBSA for the four studied systems were analyzed. The results are shown in Table 1. Table 1.Score values (kcal/mol) obtained in the 3 molecular docking programs, consensus scoring, and MM-GBSA for the candidate compounds and DNP.LigandAutoDockAutodock vinaDOCK6AASΔG_MMGBSADNP−10.66−12.00−64.810.67−36 ± 6 151171−13.24−14.10−60.150.76−32 ± 821081761−11.42−13.70−67.020.77−33 ± 624771824−11.51−13.50−66.750.76−46 ± 6 Score values (kcal/mol) obtained in the 3 molecular docking programs, consensus scoring, and MM-GBSA for the candidate compounds and DNP. Compared with DNP (reference ligand), compound 24771824 shows a better score even under rigid scoring environments. In the consensus scoring results, it achieved a value of 0.76, higher than DNP's (0.67), indicating that, when combining different scoring functions, its performance is more robust and consistent. However, it is important to note that molecular docking programs usually do not account for conformational rearrangements, solvation, and the dynamic flexibility of the receptor, which can lead to inaccurate affinity estimations. 59 This limitation is addressed by the MMGBSA method, where, once again, 24771824 shows its greatest advantage, with a ΔG of −46 ± 6 kcal/mol, surpassing not only DNP but also the other two candidates. This difference of 10–14 kcal/mol, compared to the other ligands, makes this compound the best candidate from a thermodynamic standpoint, while also maintaining stable interactions despite the dynamic effects of the protein environment. Later, with the aim of studying the nature of these interactions, Table 2 shows the binding free energy and the decomposition of energy terms for each ligand using the MMGBSA method. As previously mentioned, ligand 24771824 has the most favorable binding free energy compared to the other compounds. According to the energy decomposition, this value is mainly due to van der Waals interactions and, to a lesser extent, Coulombic interactions, suggesting a good fit within the active site and acceptable electrostatic complementarity between the ligand and the receptor. 60 On the other hand, DNP also exhibits good electrostatic interactions, but these are offset by a high solvation penalty, which explains its lower ΔG. Compounds 21081761 and 151171 show less favorable binding energies, with ΔG values of −33 ± 6 and −32 ± 8 kcal/mol, respectively. The lack of significant contributions from vdW and electrostatic terms limits their competitive potential compared to 24771824. Table 2.Binding free energy (kcal/mol) calculated using the MMGBSA method.LigandΔE_vdWΔE_electΔG_solv (GB)ΔG_surfΔG_MMGBSADNP−47 ± 5−130 ± 21141 ± 20−6.0 ± 0.6−36 ± 6151171−47 ± 8−18 ± 1134 ± 10−5.4 ± 1.0−32 ± 821081761−48 ± 6−28 ± 1343 ± 12−6.1 ± 0.7−33 ± 724771824−60 ± 5−15 ± 930 ± 7−7.3 ± 0.5−46 ± 7 Binding free energy (kcal/mol) calculated using the MMGBSA method. Discussion Residue flexibility analysis Although 24771824 stands out as the most promising candidate, it is important to assess whether this behavior is maintained in terms of local flexibility. For this purpose, an analysis of the RMSF differences was performed for residues located within 5 Å of the active site, along with the structural representation of those residues in the presence of 24771824 (Figure 5), with the goal of studying the local molecular dynamics changes induced by each candidate compound in comparison with the inhibitory ligand DNP. Supplemental Figure 6 shows the corresponding structural representations for the complexes with DNP, 151171, and 21081761, where residues exhibiting notable local fluctuations throughout the simulation are identified. In addition, a threshold of 2 Å was established to identify significant increases in flexibility. RMSF differences of residues within 5 Å of the active site relative to the reference ligand DNP. (Top) ΔRMSF plot comparing the candidate ligands with DNP. The beige-shaded areas correspond to the active-site residues, while the shaded areas indicate amino acids showing variations in flexibility. The dot line and triangles corresponds to ligand 24771824, the continuous line and circle to ligand 151171, and the dashed line and square to ligand 21081761. (Bottom) Structural representation of the 7E3H enzyme in complex with ligand 24771824, highlighting residues with local flexibility variations. For 151171, it was observed that Trp-86 displayed an increase in ΔRMSF above 2 Å, indicating a local perturbation of the aromatic network, which normally generates π interactions and stabilizes the binding of the inhibitor to the substrate.24,61 This suggests rearrangements that could reduce the residence time of the ligand in the active site. Conversely, 24771824 maintained a ΔRMSF of −0.8 Å for that amino acid compared with DNP; this decrease in fluctuation may be due to a stabilization of the aromatic environment, which is associated with a better fit within the active site. This result is consistent with the MMGBSA findings, which showed that the main contribution of 24771824 arises from van der Waals interactions, supporting the idea of ligand binding through hydrophobic contacts with residues in the active site. Another amino acid that exhibited opposite behavior was Asn-283, which showed an increase in flexibility of approximately 0.4 Å for 24771824, while in 151171 and 21081761 it displayed a decrease of −0.4 Å. Although no significant contacts with the ligands are formed, this contrast indicates that the residue acts as a sensitive point in local dynamics. On the other hand, residues Gly-335 and Ser-336 showed moderate variations, with ΔRMSF increases of 0.3 Å and 0.8 Å, respectively, for 24771824, whereas 151171 and 21081761 exhibited smaller changes (−0.4/0.5 Å for Gly-335 and 0/-0.6 Å for Ser-336). These increases in 24771824 remain below the 2 Å threshold and can be interpreted as rearrangements in the C-terminal region of the helix to better accommodate the ligand without compromising the stability of the active site. A similar behavior was observed for Arg-433, a residue located in a peripheral region but dynamically connected to the catalytic site, which showed an increase in flexibility of 0.9 Å for 24771824 and 0.6 Å for 21081761, while for 151171 the ΔRMSF was −0.7 Å. These results are consistent with previous molecular dynamics studies reporting fluctuations in some residues located in distant regions as part of the usual behavior of AChE, associated with allosteric mechanisms that modulate ligand entry and exit without significantly affecting the global stability of the enzyme. 62 Hydrogen bonds during the trajectory Hydrogen bond analysis was performed between each candidate compound and the residues of the AChE active site, as these represent non-covalent interactions that play a key role in the stability and affinity of the complex. Supplemental Figure 7 shows the distribution of hydrogen bonds expressed as the fraction of frames throughout the trajectory, while Supplemental Table 7 lists the identified hydrogen bonds and their occupancy percentages. According to the criteria described by Sangande et al. (2020), 63 hydrogen bonds with occupancy greater than 10% were considered moderately stable, whereas those with occupancy above 50% were classified as stable. However, in the present study, no interactions were found within this latter range. In the case of 24771824, numerous low-occupancy interactions were observed, with percentages below 1%. The most relevant hydrogen bond was established with Glu-292, showing an occupancy of 5.05%. This behavior is also reflected in the graph, where this compound exhibited the highest fraction of frames without hydrogen bonds (∼0.79), supporting the hypothesis that its interactions with the active site are mainly hydrophobic. DNP also showed a high fraction of frames without hydrogen bonds (∼0.68); however, it maintained a dominant contact with Asp-74, with an occupancy of 22.77%, suggesting that although most of its interactions are hydrophobic, it preserves a stable polar contact throughout the trajectory, which may contribute to the complex's stability and specificity. Meanwhile, 151171 exhibited more persistent hydrogen bonds over time with Ser-336 (∼20.5%) and Lys-332 (∼11.57%), whereas 21081761 established interactions with Arg-433 (∼16.38%), Glu-334 (∼15.18%), and Glu-81 (∼14.61%). These results are consistent, as shown in the hydrogen bond fraction distribution graph, where these two compounds displayed the highest values in hydrogen bond formation, further emphasizing the polar character of their interactions throughout the dynamics. This analysis highlights the importance of evaluating polar interactions over time and complementing them with thermodynamic information from the MMGBSA calculations, as this approach allows distinguishing whether the overall complex affinity arises from the ligand's effective hydrophobic packing or from persistent polar contacts maintained during the simulation. In this context, ligands that displace high-energy water molecules from the active site upon binding can enhance complex affinity by reducing solvation penalties and contributing to system stability; thus, strong hydrophobic packing can compensate for the absence of persistent hydrogen bonds. 64 This explains why 24771824 exhibited a low solvation energy (ΔG_solv = 29.9511 kcal/mol) and stood out in van der Waals interactions (see Table 2), indicating that its affinity is mainly driven by hydrophobic contacts, which in turn confer greater stability compared to more polar compounds. Molecular clustering and key residues in binding To identify the most relevant and recurrent interactions during the molecular dynamics simulations, a molecular clustering analysis was performed on the enzyme–ligand complex trajectories. For each system, 10 clusters were obtained using the agglomerative hierarchical clustering algorithm, and the representative structure for each group was selected. Interaction diagrams for the five most populated clusters were generated using Discovery Studio and Proteins Plus (PoseView), and the interaction results are summarized in Supplemental Table 8. Subsequently, a comparison of the identified contacts across the different clusters was performed, and the amino acid residues important for the stability of the complex were identified. Table 3 shows the residues and types of interactions that recurrently appear in the representative structures and are considered relevant for binding stability and specificity. Table 3.Recurrent and relevant interactions of the representative structures of the four ligands.Type of interactionDNP2477182421081761151171Hydrogen bondVal-331, Leu-441 Ser-125Tyr-124, Lys-332Lys-332, Val-331, Trp-286, Val-280, Tyr-124C–H bondTrp-442, Val-280, Leu-441, Glu-334Gly-121, Gly-126, Tyr-72, Ser-125, Lys-332, Gln-71, Trp-442Lys-332, Phe-80Ser-336, Lys-332π interactionsTrp-86, Trp-442, Gly-335Trp-442, Trp-86, Tyr-72, Tyr-124, Phe-80, val-280, Lys-332, Val-331, Gly-121Tyr-124, Phe-80, Tyr-72, Pro-290, Val-280, Lys-332Trp-286, Tyr-124, Lys-332, Val-280 Halogen interaction (fluorine)—Gly-120Tyr-72—Hydrophobic interactions/alkyl interactionLeu-76, Val-331, Val-280, Glu-334, Phe-80, Trp-86, Met-443, Val-331, Pro-440, Trp-442Trp-442, Tyr-124, Lys-332, Val-331, Trp-86, Tyr-72, Val-280Val-280, Lys-332, Tyr-124, Tyr-72, Phe-80Val-280, Ser-336, Trp-286, Lys-332, Tyr-124 Recurrent and relevant interactions of the representative structures of the four ligands. The molecular clustering results for the DNP complex reveal several important interactions. To begin with, contacts with Trp-86 and Trp-442 stand out, forming π–π stacking interactions that enhance ligand specificity and anchoring within the enzyme gorge, thereby reducing local mobility and promoting a more rigid arrangement. Other significant interactions include alkyl/van der Waals contacts with Val-331, Val-280, Met-443, Pro-440, Leu-76, and Phe-80, which increase the nonpolar contact surface and reinforce ligand stability within the active site. Additionally, polar contacts with Val-331 and Leu-441, together with the hydrogen bond identified with Asp-74, define a mixed binding mode in which hydrophobic stabilization predominates, accompanied by a transient polar component that contributes to ligand orientation and specificity. This heterogeneous binding mode is consistent with AChE studies reporting inhibitors that rely on hydrophobic interactions to achieve strong affinity, complemented by localized polar contacts that enhance binding specificity. 65 Meanwhile, Figure 6 shows that the affinity of 24771824 is mainly driven by aromatic stacking (π–π) interactions and hydrophobic packing, which better explains its superior performance compared to the other compounds. Across several clusters, Trp-86 and Trp-442 recurrently appear forming stacking interactions with the aromatic regions of the ligand, establishing a stable orientation within the active site. Multiple aliphatic positions and hydrophobic contacts are also observed with Val-331, Val-280, Tyr-124, Tyr-72, and Lys-332, creating a closed hydrophobic environment that restricts ligand mobility inside the pocket. In addition, localized polar contacts with Ser-125—although weak and transient—could help stabilize the ligand during protein fluctuations. Furthermore, the presence of a halogen interaction with Gly-120 may contribute a secondary directional component to ligand anchoring. This hydrophobic network is consistent with the energetic decomposition analysis and explains why 24771824 exhibits a strong ΔE_vdW contribution: the tight hydrophobic fit not only enhances favorable dispersion interactions but also promotes the displacement of water molecules from the binding pocket, thereby reducing the overall solvation penalty. In summary, the van der Waals and aromatic contacts identified for this ligand account for its more favorable ΔG and stable behavior in the RMSF analysis, positioning it as the best candidate in terms of both stability and affinity. The prevalence of hydrophobic interactions in high-affinity ligands is well documented in protein–ligand interaction analyses, where efficient ligands tend to exhibit a higher number of hydrophobic contacts compared to polar interactions. 66 Main interactions of 24771824 with active-site residues during MD. Compound 151171 exhibited a binding profile characterized by polar interactions and moderate aromatic stacking, in contrast to the predominantly hydrophobic profile observed for 24771824. Across several clusters, hydrogen bonds were formed with Lys-331, Lys-332, and Trp-286, which in some cases also participated in π–cation and π–π stacking interactions. Simultaneously, residues Val-280, Tyr-124, and Val-332 formed a combination of hydrogen bonds and π–alkyl/π–π hydrophobic contacts, reflecting a set of cooperative interactions rather than a single anchoring point. The amino acids Ser-336 and Lys-332, identified in the most populated clusters, correlate with the previously calculated hydrogen-bond occupancies (Ser-336 ∼20.50% and Lys-332 ∼11.57%), consistent with the observation that nearby residues such as Ser-336 exhibit lower ΔRMSF values when 151171 is present. This indicates that the compound tends to locally stabilize certain regions through hydrogen bonds and hydrophobic contacts, although it induces structural rearrangements in other areas, as evidenced by the increase in ΔRMSF at Trp-86. Compound 21081761 also maintains a network of polar and hydrophobic contacts, along with π-type interactions that contribute to the stability of the complex. Hydrogen bonds were identified with Tyr-124, and especially with Lys-332, which serves as a key point for ligand orientation and anchoring. These findings are consistent with the hydrogen-bond occupancy data, where the ligand forms polar contacts with Arg-433 (∼16.38%), Glu-334 (∼15.18%), and Glu-81 (∼14.61%) and, may be relevant for stability within certain conformational microstates. In the same way, aromatic interactions and hydrophobic packing also play important roles in the stability of the complex. 21081761 forms π–π stacking interactions with Phe-80 and Tyr-72, as well as π–alkyl and hydrophobic contacts with Val-280, Lys-332, Tyr-124, Tyr-72, Phe-80, and Pro-290, which reinforce ligand packing within the active site. However, the hydrophobic contributions of 21081761 are not sufficient to surpass 24771824 in its ΔG_MMGBSA, as the electrostatic gains are partially offset by solvation penalties. Per-residue energy decomposition Finally, the per-residue energy decomposition allowed the identification of the amino acids that contribute most significantly to the stabilization of the four complexes studied, as well as the differences in affinity among them (Supplemental Figure 8). Compound 24771824 is characterized by a substantial contribution from aromatic residues within the catalytic gorge of the enzyme, with Tyr-124 standing out (−2.79 kcal/mol) and notable contributions from Tyr-72 (−2.08 kcal/mol) and Trp-86 (−1.78 kcal/mol), which exhibit strongly negative values. This suggests a binding mode dominated by van der Waals interactions and favorable hydrophobic packing, specifically π–π stacking and π-alkyl interactions, which is consistent with the typical profile of acetylcholinesterase inhibitors.67,68 In contrast, ligand 151171 exhibited a different behavior in agreement with the results obtained from molecular clustering, showing relevant contributions from both nonpolar and polar residues. Notably, Trp-286 (−1.93 kcal/mol) and Val-280 (−1.50 kcal/mol) contributed through hydrophobic and aromatic contacts; however, polar interactions were also observed, as discussed in the previous analysis. The most significant contribution involved Ser-336 (−1.70 kcal/mol), whose interaction suggests the presence of directed electrostatic interactions, possibly a transient hydrogen bond formed during the simulation, as previously confirmed by the hydrogen-bond analysis, which showed an occupancy of approximately 20.50%. Similarly, ligand 21081761 displayed relevant energetic contributions with Tyr-72 (−1.35 kcal/mol) and Lys-332 (−1.34 kcal/mol), as well as with Tyr-124 (−1.08 kcal/mol) and Val-280 (−1.05 kcal/mol), suggesting an intermediate binding mode stabilized by hydrophobic, aromatic and polar contacts, but with a lower overall binding energy compared to 24771824. The reference ligand DNP showed a more distributed energetic stabilization across several residues within the active site, with notable contributions from Tyr-124 (−1.34 kcal/mol), Val-331 (−1.20 kcal/mol), Thr-83 (−1.06 kcal/mol), and Trp-86 (−0.97 kcal/mol). This behavior reflects a binding mode stabilized by multiple moderate interactions, combining electrostatic and hydrophobic contacts, which is consistent with its clinical effectiveness as an AChE inhibitor. 69 Nevertheless, DNP does not exhibit highly concentrated energetic contributions at specific residues, resulting in a lower global binding energy compared to ligands such as 24771824. The differences among the ligands’ results are important because they imply that each ligand explores different microenvironments within the active site; this is a normal behavior considering the structural and dynamic flexibility of AChE and its ability to accommodate compounds with different polarity and size. 70 In this context, the predominance of aromatic and hydrophobic residues involved in the binding of 24771824, in contrast to the more polar interactions observed for 151171, helps explain the superior affinity of 24771824. Dispersive interactions confer greater stability against structural fluctuations of the complex and are more favorable in hydrophobic environments such as the AChE active site. Conversely, ligand 151171 relies to a greater extent on polar interactions, which are partially offset by desolvation effects. 71 Thus, per-residue energy decomposition provides insight into selective binding modes and enables the identification of key residues acting as energetic “hotspots” that not only stabilize the complex but may also serve as targets for the design of analogues with improved affinity or specificity. In particular, Tyr-124, Tyr-72, Val-280, and Lys-332 stand out for exhibiting recurrent favorable contributions across all studied complexes, suggesting that they constitute a conserved interaction core within the AChE active site. It is worth noting that several amino acids identified through per-residue energy decomposition coincide with key residues described in the crystallographic structure of AChE (PDB ID: 7E3H) as part of the active site and the catalytic gorge. In particular, Trp-86, a highly conserved residue associated with the anchoring of aromatic inhibitors, showed significant energetic contributions in the complex with 24771824 and to a lesser extent with DNP, reinforcing its role in π-stacking interactions. On the other hand, Trp-286, which is associated with the peripheral anionic site of AChE, exhibited a relevant energy contribution exclusively with 151171, suggesting an alternative binding mode involving peripheral regions of the active site. Finally, Tyr-72, identified as a key residue in the catalytic gorge, showed favorable energetic contributions in the complexes with 24771824 and 21081761, and to a lesser extent with 151171 and DNP, highlighting its role in stabilizing ligands with different structural properties. Overall, these findings demonstrate a strong agreement between experimental evidence and computational approaches, supporting the biological relevance of the proposed binding modes. Conclusions The results obtained support the effectiveness of the computational approach employed for the identification of new acetylcholinesterase inhibitors. The pharmacophore model enabled the selection of compounds with structural features similar to DNP and compatible with the enzyme's active site, as reflected in the molecular docking results, which identified compounds with higher binding affinities than the reference compound, supporting the consistency of the applied protocol. ADMET property analysis suggested favorable pharmacokinetic profiles for the most promising candidates. Molecular dynamics simulations and their corresponding analyses suggested that ligands 151171 and 21081761 exhibit binding affinities dominated by polar and directional contacts that provide stability without achieving maximal global stabilization. In contrast, ligand 24771824 showed a binding mode centered on hydrophobic and aromatic interactions, which may reduce solvation penalties and contribute to stability within the active site. The identification of conserved key residues such as Tyr-124, Tyr-72, Val-280, and Lys-332, which act as recurrent energetic hotspots across the studied complexes, further supports the internal consistency of the results. Finally, compound 24771824 emerged as the most promising candidate for future experimental validation. In this context, future studies are expected to include experimental assays to corroborate the inhibitory activity against AChE and the biological potential of the most promising compounds selected in this study. Supplemental Material sj-docx-1-alz-10.1177_13872877261483260 - Supplemental material for Computational identification of novel acetylcholinesterase inhibitors as potential treatments for Alzheimer's disease Supplemental material, sj-docx-1-alz-10.1177_13872877261483260 for Computational identification of novel acetylcholinesterase inhibitors as potential treatments for Alzheimer's disease by Sebastián Matallana Rincón, Fabián Orozco López and Johan F Galindo in Journal of Alzheimer's Disease