Abhishek S R¹, Dr. VG Shanmuga Priya¹
¹Department of Life Sciences, School of Sciences, Garden City University, Bengaluru, India
Abstract
**Background:** The inhibition of Beta-site APP Cleaving Enzyme 1 (BACE1) remains a cornerstone therapeutic strategy for Alzheimer's disease (AD), predicated on the amyloid cascade hypothesis. However, the clinical development of BACE1 inhibitors has been beset by failures, primarily due to a combination of mechanism-based toxicities and a lack of cognitive improvement in patients, despite effective reduction of amyloid-beta (Aβ) peptides. This challenging landscape necessitates the exploration of novel chemical scaffolds with potentially more favorable biological profiles. Natural products, such as ginsenosides from *Panax ginseng*, offer a rich source of structurally diverse molecules with known neuroprotective properties.
**Methods:** This study employed a comprehensive *in silico* workflow to identify and characterize ginsenoside analogues as potential BACE1 inhibitors. A library of analogues was generated from the ChEMBL database and subjected to molecular docking against the BACE1 crystal structure (PDB: 1FKN) using AutoDock Vina. The top-ranked candidate was further evaluated for its pharmacokinetic and drug-likeness properties via ADMET (Absorption, Distribution, Metabolism, Excretion, Toxicity) profiling with SwissADME. Finally, a 1-nanosecond all-atom molecular dynamics (MD) simulation was performed to assess the dynamic stability and interaction persistence of the BACE1-ligand complex.
**Results:** Virtual screening identified several ginsenoside analogues with high predicted binding affinities. The lead candidate, CHEMBL3594353, exhibited a binding energy of -9.7 kcal/mol, superior to the clinically tested BACE1 inhibitor Verubecestat (-8.3 kcal/mol). Detailed interaction analysis revealed extensive engagement with key active site residues. The MD simulation confirmed the formation of a structurally stable complex, characterized by low root-mean-square deviation (RMSD) and a ligand-induced rigidification of the enzyme's functionally critical flap region. Despite this high predicted potency, ADMET analysis revealed significant liabilities, including a high molecular weight, multiple Lipinski's Rule violations, and poor predicted blood-brain barrier permeability.
**Conclusion:** This study successfully identifies the ginsenoside scaffold as a potent source of novel BACE1 binders, validated through both static docking and dynamic simulation. However, it also critically highlights the substantial pharmacokinetic hurdles that must be overcome for this class of compounds to be viable as central nervous system drug candidates. The findings provide a validated starting point and a clear medicinal chemistry roadmap for the lead optimization of ginsenoside-based BACE1 modulators, framing a path forward that learns from the complex history of BACE1 inhibitor development.
## 1. Introduction
### 1.1 The Alzheimer's Disease Conundrum and the Amyloid Hypothesis
Alzheimer's disease (AD) constitutes a profound and escalating global public health crisis. Characterized by a relentless and progressive deterioration of cognitive functions, memory, and behavior, it is the most common cause of dementia worldwide.[1, 1] Current estimates suggest over 50 million individuals are affected, a figure projected to surge to approximately 130 million by 2050, imposing an immense burden on patients, families, and healthcare systems.[1, 1] The therapeutic landscape for AD remains starkly limited. For decades, the primary pharmacological interventions have been symptomatic treatments, most notably acetylcholinesterase (AChE) inhibitors like Donepezil. These agents aim to increase the availability of the neurotransmitter acetylcholine, offering modest and temporary palliation of symptoms but failing to halt or reverse the underlying neurodegenerative process.[1, 2, 3]
For over two decades, the dominant paradigm guiding therapeutic development has been the "amyloid cascade hypothesis".[2] This hypothesis posits that the central pathogenic event in AD is the imbalanced production and clearance of amyloid-beta (Aβ) peptides, particularly the aggregation-prone 42-amino-acid form (Aβ42).[2, 4] According to this model, the accumulation and subsequent aggregation of Aβ into soluble oligomers and insoluble extracellular plaques trigger a complex neurotoxic cascade, leading to synaptic dysfunction, neuroinflammation, the formation of neurofibrillary tangles (composed of hyperphosphorylated tau protein), and ultimately, widespread neuronal death and cognitive decline.[1, 1, 5]
### 1.2 BACE1: The Prime Target with a Perilous History
The amyloid cascade hypothesis logically identifies the enzymes responsible for Aβ production as prime therapeutic targets for disease-modifying therapies. Aβ peptides are generated through the sequential proteolytic cleavage of the large transmembrane Amyloid Precursor Protein (APP). The first and rate-limiting step in this amyloidogenic pathway is carried out by the Beta-site APP Cleaving Enzyme 1 (BACE1), an aspartic protease primarily expressed in neurons.[1, 3, 4, 6] The subsequent cleavage by γ-secretase releases the Aβ peptide. Therefore, inhibiting BACE1 presents a direct and mechanistically compelling strategy to reduce Aβ production and, theoretically, halt the progression of AD at its origin.[7]
This rationale fueled over twenty years of intensive drug discovery efforts by pharmaceutical companies, resulting in the development of numerous potent, brain-penetrant small-molecule BACE1 inhibitors.[8, 9] However, the journey from preclinical promise to clinical success has been a story of profound and consistent failure. Early candidates like Eli Lilly's LY2811376, while demonstrating target engagement by lowering Aβ levels in human subjects, were discontinued due to off-target toxicity, in this case, damage to the eye's pigment epithelium in animal models.[10]
More devastatingly, a series of highly advanced inhibitors failed in late-stage Phase III clinical trials. Merck's Verubecestat, despite robustly reducing Aβ levels in the central nervous system (CNS) of AD patients, was halted first in mild-to-moderate AD and later in prodromal AD because it showed no benefit in slowing cognitive decline.[9, 11] This was followed by similar failures from Janssen's Atabecestat, which was stopped due to liver safety concerns [12], and AstraZeneca/Eli Lilly's Lanabecestat, which was terminated for futility.[11] This string of high-profile failures created a paradox that has shaken the foundations of AD research: the drugs successfully engaged their target and modulated the biomarker (Aβ), yet failed to produce any meaningful clinical improvement.[1, 1, 13] This disconnect has forced a critical re-evaluation of the amyloid hypothesis, suggesting that by the time clinical symptoms of AD are present, the neurodegenerative cascade may have become independent of Aβ production, rendering its inhibition insufficient to alter the disease course.[14]
### 1.3 Mechanism-Based Toxicity and the Need for New Strategies
The challenges with BACE1 inhibition extend beyond the lack of efficacy. Emerging evidence from the failed trials and preclinical studies has highlighted the risk of mechanism-based toxicity.[8] BACE1 is not merely an "Aβ factory"; it is a pleiotropic enzyme with numerous physiological substrates involved in crucial neuronal functions, including myelination, synaptic plasticity, and neuronal connectivity.[5, 7] For instance, BACE1 is the exclusive protease for Seizure protein 6 (Sez6), a protein required for maintaining dendritic spine density and plasticity. Prolonged and potent inhibition of BACE1 diminishes Sez6 processing, which has been linked to the cognitive worsening observed as a side effect in some clinical trials.[6]
This realization has led to a paradigm shift in the field. The goal is no longer simply to find the most potent BACE1 inhibitor, but to develop a "smarter" modulator. This has given rise to new therapeutic hypotheses, such as the need for partial, rather than complete, BACE1 inhibition to strike a balance between reducing Aβ and preserving essential physiological functions.[5, 15] Another promising avenue is the development of substrate-selective inhibitors that could preferentially block the cleavage of APP while sparing other vital substrates like Sez6.[8, 13] The clinical failures have thus provided an inadvertent but valuable roadmap, defining the stringent criteria for a successful next-generation BACE1-targeting therapy: high selectivity for BACE1 over its homologue BACE2 and other proteases, and a carefully titrated level of inhibition to avoid disrupting neuronal homeostasis.
### 1.4 Rationale for Exploring Natural Products: The Case for Ginsenosides
In the search for novel chemical scaffolds that might possess these more nuanced biological properties, natural products offer a compelling starting point. Historically, they have been a rich source of therapeutic agents due to their vast structural diversity and biological relevance. *Panax ginseng*, a cornerstone of traditional Chinese medicine, produces a class of secondary metabolites known as ginsenosides.[1, 1] These triterpene saponins are renowned for a wide array of pharmacological activities, including anti-inflammatory, antioxidant, and cardiovascular effects.[1, 1] More recently, their neuroregulatory and neuroprotective properties have garnered significant attention, with studies suggesting potential benefits in models of Parkinson's disease, epilepsy, and AD.[1, 1]
The unique and complex structures of ginsenosides present an opportunity to explore a chemical space distinct from the synthetic compounds that have so far failed in the clinic. The present study was therefore designed to leverage a multi-step *in silico* workflow to systematically investigate ginsenoside analogues as potential BACE1 inhibitors. By combining virtual screening, molecular docking, ADMET profiling, and molecular dynamics simulations, this research aims to identify promising candidates, characterize their binding mechanisms at an atomic level, and critically evaluate their potential as starting points for a new class of AD therapeutics. This work represents a foundational step in exploring whether this class of natural products can offer a solution to the complex challenge of safely and effectively modulating BACE1 activity.
## 2. Computational Methods
### 2.1 Ligand Selection and Preparation
The initial phase of the study involved the construction of a focused library of test compounds. The ChEMBL database, a comprehensive repository of bioactive molecules, was utilized for this purpose.[1, 1] Ginsenoside Rg1 (ChEMBL ID: CHEMBL501637) was selected as the query structure due to its representative ginsenoside scaffold. A similarity search was performed, applying a filter to retrieve analogues exhibiting 80-90% structural similarity to the query. This process yielded an initial set of 27 compounds, from which 9 analogues were randomly selected for detailed computational analysis to ensure a diverse representation of the chemical space.[1, 1]
For comparative analysis and validation of the docking protocol, two reference compounds were chosen. Verubecestat (PubChem ID: 51352361), a potent BACE1 inhibitor that reached Phase III clinical trials, was selected as a positive control for BACE1 binding.[1, 1] Donepezil (ZINC ID: 597013), a widely prescribed AChE inhibitor for symptomatic AD treatment, was included as a relevant clinical comparator, although its primary mechanism is unrelated to BACE1.[1, 1] All ligand structures were downloaded and prepared for docking. This preparation was conducted using BIOVIA Discovery Studio (v4.5), and involved a standardized protocol of adding hydrogen atoms, neutralizing charges, generating appropriate ionization states at physiological pH, and performing geometry cleaning to obtain energetically favorable conformations. The final prepared structures were saved in the MOL2 format.[1, 1]
### 2.2 Receptor Preparation
The three-dimensional coordinates of the BACE1 enzyme were retrieved from the Protein Data Bank (PDB). The specific structure chosen was PDB ID: 1FKN, a human BACE1 crystal structure resolved at 1.9 Å.[1, 1] While this structure provides a high-resolution view of the enzyme, it is important to note that it represents an early apo-form (unliganded). The decision to use this structure, rather than a more recent one co-crystallized with a known inhibitor like Verubecestat (e.g., PDB ID: 5HU1), represents a methodological limitation. Using an apo-structure for docking can sometimes be less accurate than using a ligand-bound (holo) structure, which represents an induced-fit conformation. This choice could potentially impact the accuracy of the predicted binding poses, particularly for the reference inhibitor Verubecestat.
The receptor was prepared for docking using a combination of UCSF Chimera (v1.13) and BIOVIA Discovery Studio (v4.5).[1, 1] The protocol involved removing all non-essential molecules, including water molecules and any co-crystallized ligands or ions. Hydrogen atoms were then added to the protein structure, and partial charges were assigned to all atoms to accurately model electrostatic interactions during the docking process. The prepared receptor was then ready for the docking simulations.
### 2.3 Molecular Docking Protocol
Molecular docking simulations were performed to predict the binding conformations and affinities of the selected ligands within the BACE1 active site. The study utilized AutoDock Vina (v4.2.1), a widely used and robust docking program, interfaced through the Pyrx software platform, which provides a graphical user interface for virtual screening workflows.[1, 1] Prior to docking, each ligand underwent an energy minimization step using the universal force field (UFF) and a conjugate gradients algorithm, as implemented in the Open Babel toolkit integrated within Pyrx.[1, 1]
For the docking simulation, a grid box with dimensions of 25 Å × 25 Å × 25 Å was defined and centered on the known catalytic active site of BACE1. This grid box size was deemed sufficient to encompass the entire binding pocket and allow for full rotational and translational flexibility of the ligands.[1, 1] AutoDock Vina's scoring function was used to evaluate the binding poses. For each ligand, the software was configured to generate the top eight binding conformations. These conformations were ranked based on their calculated free energy of binding ($\Delta G$), expressed in kcal/mol, with more negative values indicating a more favorable predicted interaction. The final ranked list of poses provided the basis for subsequent interaction analysis.[1, 1]
### 2.4 Post-Docking Analysis and Visualization
Following the completion of the docking simulations, the resulting protein-ligand complexes, saved in PDBQT format, were subjected to detailed analysis. The molecular visualization software PyMOL (v2.3.2) was used to inspect the predicted binding poses, assess their steric and chemical complementarity with the active site, and generate high-quality images of the interactions.[1, 1]
For a more quantitative analysis of the intermolecular forces stabilizing the complexes, BIOVIA Discovery Studio (v4.5) was employed. This software was used to identify and map the specific non-covalent interactions between the highest-ranked pose of each ligand and the amino acid residues of the BACE1 active site. These interactions included conventional hydrogen bonds, carbon-hydrogen bonds, van der Waals forces, and various π-system interactions (e.g., π-alkyl, π-sigma, π-π T-shaped), providing a comprehensive picture of the molecular basis for the predicted binding affinity.[1, 1]
### 2.5 *In Silico* ADMET and Physicochemical Profiling
To assess the potential of the lead compounds as viable drug candidates, a critical evaluation of their pharmacokinetic and drug-likeness properties was performed. This *in silico* analysis was conducted using the SwissADME web server, a freely accessible tool for predicting Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) properties.[1, 1] The SMILES (Simplified Molecular Input Line Entry System) strings for the top ginsenoside analogue and the two reference drugs were submitted to the server.
The analysis focused on several key parameters. Physicochemical properties such as molecular weight (MW), lipophilicity (Consensus Log $P_{o/w}$), and topological polar surface area (TPSA) were calculated. Drug-likeness was evaluated based on compliance with established filters, including Lipinski's Rule of Five, Ghose, Veber, and Egan rules. Pharmacokinetic properties, specifically passive gastrointestinal (GI) absorption and blood-brain barrier (BBB) permeability, were predicted and visualized using the intuitive BOILED-Egg plot. The potential for the compounds to be substrates of the P-glycoprotein (P-gp) efflux pump and to inhibit major cytochrome P450 (CYP) enzymes (1A2, 2C19, 2C9, 2D6, 3A4) was also assessed to predict potential drug-drug interactions and clearance mechanisms.[1, 1]
### 2.6 Molecular Dynamics (MD) Simulation
While molecular docking provides a static snapshot of ligand binding, proteins are dynamic entities that fluctuate in a physiological environment. To validate the stability of the predicted binding pose and investigate the dynamic behavior of the protein-ligand complex, an all-atom Molecular Dynamics (MD) simulation was performed.[1, 1] The simulation was run for the highest-ranked ginsenoside analogue (CHEMBL3594353) in complex with BACE1 for a duration of 1 nanosecond (ns).
Before the production run, the system was carefully equilibrated to ensure thermodynamic stability. This involved sequential equilibration steps under NVT (constant Number of particles, Volume, and Temperature) and NPT (constant Number of particles, Pressure, and Temperature) ensembles. The stability of key parameters like temperature, pressure, and potential energy was monitored to confirm that the system had reached equilibrium.[1, 1] The 1 ns production trajectory was then analyzed using several key metrics to assess structural stability and dynamics. The Root-Mean-Square Deviation (RMSD) of the protein's C-alpha (Cα) atoms was calculated to monitor global conformational changes. The Root-Mean-Square Fluctuation (RMSF) of each residue was computed to identify regions of high and low flexibility. Finally, the Radius of Gyration (Rg) was measured over time to assess the overall compactness of the protein structure.[1, 1]
## 3. Results
### 3.1 Virtual Screening Identifies Ginsenoside Analogues with High Predicted Binding Affinity for BACE1
The virtual screening of nine selected ginsenoside analogues against the BACE1 active site identified several compounds with promising predicted binding affinities. Of these, six analogues exhibited binding energies more favorable than -7.5 kcal/mol, a commonly used threshold for identifying potential hits in early-stage screening. The docking scores for the top-performing ginsenoside analogues, along with the two reference drugs, are summarized in Table 1.
The results clearly indicate that the ginsenoside scaffold can achieve strong predicted binding to the BACE1 enzyme. Notably, the top-ranked analogue, CHEMBL3594353, achieved a binding energy of -9.7 kcal/mol. This value is significantly more favorable than those calculated for the two reference compounds, Verubecestat and Donepezil, which both docked with a predicted binding energy of -8.3 kcal/mol. Furthermore, two other analogues, CHEMBL466844 (-9.1 kcal/mol) and CHEMBL454530 (-8.8 kcal/mol), also demonstrated predicted affinities superior to the reference drugs. This primary finding suggests that specific structural variants within the ginsenoside class possess the potential to bind to BACE1 with high affinity, warranting further investigation into their specific interaction modes.