Virtual screening of phytochemicals against multiple protein targets of Alzheimer’s disease

Introduction

Alzheimer’s disease (AD) is a complex neurodegenerative disorder and the leading cause of dementia (Monteiro et al., 2023). While the etiology and pathogenesis of AD is not completely understood, it is characterized by protein aggregation of amyloid-beta (Aβ) outside neurons, and of microtube-associated protein tau (MAPT) inside neurons, leading to progressive memory and cognitive impairments (Monteiro et al., 2023; Chen Z.-L. et al., 2023). AD is broadly classified into familial (FAD), which accounts for less than 5% of cases, and sporadic (SAD), representing more than 95% of AD cases (Andrade-Guerrero et al., 2023). The leading cause of early-onset FAD is hereditary mutations in genes encoding the amyloid precursor protein (APP), or γ-secretase, leading to increased cleavage of APP, and overproduction of aggregation-prone Aβ fragments central to AD pathogenesis (Raulin et al., 2022). Sequential proteolysis of the APP is initiated by β-secretase (BACE1), and subsequent cleavage by γ-secretase subunits (PSEN1 and PSEN2) releases Aβ peptides. These peptides self-aggregate, first into soluble oligomers (protofibrils), and later into insoluble fibrils and extracellular plaques (Hampel et al., 2021; Hur, 2022). Experimental evidence shows that these soluble Aβ oligomeric protofibrils are neurotoxic, and linked to neuronal damage, including cholinergic neurons (Hampel et al., 2021; Nasb et al., 2024; Maatuk and Samson, 2013). Cholinergic neurodegeneration leads to acetylcholine deficits, crucial for learning and memory, and its deficiency weakens compromised brain areas in AD (Chen et al., 2022). Furthermore, acetylcholine hydrolysis by cholinesterase enzymes (AChE and BChE) exacerbate the condition (Zhou and Huang, 2022). In addition, Aβ binds to neuronal receptors, such as NMDA and AMPA, leading to calcium influx, and activates intracellular signaling pathways, that impair synaptic function. Aβ also induces production of reactive oxygen species (ROS) leading to oxidative stress that damages cellular components, impairs mitochondrial functions, and contributes to neuronal injury and death (Zhang et al., 2024). In parallel, Aβ toxicity induces kinase/phosphatase dysregulation and promotes tau hyperphosphorylation leading to the formation of neurofibrillary tangles (Xing et al., 2025; Zhang et al., 2023). These tangles disrupt microtubule stability and axonal transport, further impairing neuronal function and promoting neurodegeneration (Xing et al., 2025). Sporadic Alzheimer’s disease (SAD), the predominant form of AD, is driven by complex genetic and environmental factors (Andrade-Guerrero et al., 2023). The strongest risk factor is genetic variations of Apolipoprotein E4 (APOE4), that can impair clearance of insoluble Aβ, compared to APOE3/2 alleles (Allwright et al., 2023; Bassal et al., 2025). APOE4 also increases the risk of Parkinson’s disease (PD), and exacerbates α-synuclein toxic effects (Raulin et al., 2022). Moreover, APOE4 is associated with higher incidence of cardiovascular disease (Dunk et al., 2025), indicating its significant potential in multiple interconnected diseases risk factors.

Currently, there is no cure against AD, and approved drugs provide only temporary symptomatic relief without halting disease progression (Reddi Sree et al., 2025). Despite efforts in therapeutic research for AD, the conventional single target strategy has failed to provide promising treatment. Instead, multi-target therapeutic strategies have been suggested for the treatment of complex diseases like AD (Kumar et al., 2022; Pathak and Kabra, 2024; Yoo et al., 2025; Jangid et al., 2024; Takomthong et al., 2025; Iqbal et al., 2023).

Computer aided drug discovery techniques have become powerful tools in modern drug discovery. Among these, molecular docking enables rapid screening and identification of lead compounds, accelerating drug discovery and can significantly lower the cost and time required for the discovery of an effective drug (Adelusi et al., 2022; Sadybekov et al., 2023; Singh et al., 2024). In addition, ADMET (absorption, distribution, metabolism, excretion and toxicity) prediction enables to flag candidates with poor pharmacokinetics and toxicity profile, avoiding costly failure in later stages. Similarly, molecular dynamics simulation can be used to assess binding stability of docking poses over time (Adelusi et al., 2022). Notably, the availability of large chemical databases such as PubChem, ZINC, and other public libraries represents a major opportunity, enabling virtual screening over vast, structurally diverse molecules, which can improve chances of finding novel scaffolds and of discovering new hits (Korn et al., 2023). In the past, we have used these tools to identify drugs for repurposing in AD using FDA-approved drug libraries (Ben Zaken et al., 2025).

Natural products also play an important role in modern drug discovery due to their high structural diversity, and broad biological activity. These characteristics have shown higher success rates in clinical trials, compared with synthetic compounds (Butler et al., 2025; Domingo-Fernández et al., 2024). In particular, phytochemicals have shown much promise in drug discovery, because of their structural diversity, and their multitarget biological activities (Sharifi-Rad et al., 2022).

Here, we virtually screen 2,847 phytochemicals against multiple biological targets of AD using molecular docking. Then, we apply a multi-target filter followed by low heavy atoms count to identify the most promising candidates, for experimental validation in future studies.

Methods

Target identification

To identify protein targets strongly associated with Alzheimer’s disease (AD), we used a widely adopted strategy in biomedical research based on normalized PubMed association scores (Wang et al., 2020; G et al., 2022; Raj et al., 2024). Briefly, we first obtained a list of 20,594 protein coding human genes from NCBI’s gene resource (www.ncbi.nlm.nih.gov/datasets/gene/taxon/9606/?gene_type=protein-coding, retrieved in February 2025). Then, for each gene, we counted the number of citations returned in a PubMed search using the term: “gene name” [All Fields] (e.g., “amyloid beta precursor protein” [All Fields]). Then, we counted the number of co-citations returned in a combined PubMed search for the gene name, and for AD, using the term: “gene name” [All Fields] AND “Alzheimer’s Disease” [MeSH Terms] (e.g., “amyloid beta precursor protein” [All Fields] and “Alzheimer’s Disease” [MeSH Terms]). Then, the co-citation count was divided by the gene citation count to obtain normalized associations score. Normalization accounts for citation trends, such as highly cited genes, and rare mentions. In addition, the normalized score was multiplied by a penalty factor to reduce the contribution of genes with very low citation counts, to produce the adjusted associations scores. The genes were then ranked based on this Adjusted Score to highlight those with significant literature association. Importantly, an in-house python script was used to automate the citation counts for all protein-coding human genes. Normalized association and adjusted score for ranking was calculated as: where C = number of gene-AD co-citations, T = number of gene citations. Finally, the top ten protein targets most associated with AD were used for molecular docking.

Normalized Score N = C T × 100

Adjusted Score A = N × 1 − 100 T + 100

Target preparation

For each target, a representative PDB structure with human sequence, a complete chain, and the best resolution was retrieved (Wang S. et al., 2022). Then, PDB files were prepared for molecular docking, by removing native ligands, non-standard residues, non-target chains, and water molecules; adding polar hydrogens, and Kollman charges; and converting the files to PDBQT format. For this purpose, visualization tools such as ChimeraX, PyMol and AutoDock Tools were used (Houston et al., 2015; Meng et al., 2023).

To explore the entire surface of protein targets, we employed blind docking. Table 1 lists the PDB IDs and grid box parameters (grid center and grid size), which covers the entire volume of the target protein. The grid box parameters were set using AutoDock Tools (Houston et al., 2015).

TABLE 1. PDB ID and AutoDock grid-box settings of targeted proteins.

In addition to the primary protein-ligand complexes used in this study, two additional experimentally characterized co-crystal structures, namely, AChE complexed with tacrine (PDB ID: 7XN1) and GSK3β complexed with ANP (also known as SB216763; PDB ID: 1PYX), were incorporated into the docking validation dataset to further evaluate the reliability and reproducibility of the docking protocol. The grid box parameters for AChE (PDB ID: 7XN1) were set to a grid center of X = 55.84, Y = −32.94, and Z = −30.23 Å, with grid dimensions of X = 126, Y = 126, and Z = 126 Å. Similarly, the grid box parameters for GSK3β (PDB ID: 1PYX) were set to a grid center of X = 28.17, Y = 7.84, and Z = 38.91 Å, with grid dimensions of X = 126, Y = 126, and Z = 126 Å.

Preparation of the ligand library

A ligand library consisting of 2,847 phytochemicals was retrieved from the PubChem database “KEGG: Phytochemicals” (retrieved from https://pubchem.ncbi.nlm.nih.gov/classification/#hid=5 in February 2025). The ligands obtained in SDF format were prepared for molecular docking by removing water, adding polar hydrogens, allowing rotatable bonds to be rotatable, adding Gasteiger charges, and converting to PDBQT format suitable for AutoDock Vina. Ligand preparation was performed using Open Babel 3.1.1, and automated using an in-house python script (Houston et al., 2015).

Molecular docking

Following target and ligand preparation, AutoDock vina_1.2.7 was utilized to dock the ligands to the target proteins (Trott and Olson, 2010). An in-house python script was used to automate the docking process. The default setting of AutoDock Vina, including the number of poses (N = 9), exhaustiveness (8), and spacing angstrom (0.375) were left unchanged. For each ligand, the pose with the lowest binding energy was picked as the best pose. Co-crystalized ligands were used to validate our docking settings.

Multitarget filter

After molecular docking, the top 100 ligands (of 2,847 phytochemical compounds) were retained for each target based on their binding energy for further screening. Top ligands that targeted three or more protein were considered multitarget ligands.

Other physicochemical descriptors, including log P, molecular weight, hydrogen bond donors/acceptors, and total polar surface area (TPSA), were also considered using relaxed criteria rather than strict drug-like cutoff thresholds, allowing compounds with potentially favorable characteristics outside conventional drug-likeness ranges to be retained. This is because adherence to strict drug-like filter cutoffs can reduce exploration of chemical space and exclude potentially effective compounds (Kattuparambil et al., 2025) and, moreover, relaxing conventional physicochemical filters have been shown to expand accessible chemical diversity and capture novel scaffolds in drug discovery (Alzyoud et al., 2022).

Interaction analysis

To understand the binding mechanism and interacting residues, interaction analysis was performed for selected ligands using Protein-Ligand Interaction Profiler (PLIP), an automated web-based tool to analyze protein-ligand interactions based on the PDB file format of protein-ligand complexes (Schake et al., 2025).

Molecular dynamics simulation

To understand the stability of the protein-ligand complex over time, molecular dynamics simulations were performed for selected ligands commonly targeting Acetylcholinesterase (AChE) and glycogen synthase kinase three beta (GSK3β). We Followed protocols from previous studies (Vander Meersche et al., 2024) and gromacs tutorial (Lemkul, 2018). Briefly, the crystal structure of the targets was processed, and missing atoms were fixed using PDBfixer. Gromacs compatible protein topology was generated using charmm36 force field and charmm-modified tip3p water model. The terminal residues were assigned their standard protonation states, with the N-terminus protonated (NH₃ ⁺) and the C-terminus deprotonated (COO⁻), to represent physiological ionization states and ensure proper electrostatic interactions during molecular dynamics simulations. Prior to processing the ligand topology, the pdbqt format of the poses was converted to mol2 format, suitable for CGenFF processing, and all hydrogens were added to make suitable for all atom MD simulations. Ligand structures were processed using ‘perl sort_mol2_bonds.pl’ script, downloaded from gromacs tutorial, to ensure consistent connectivity and compatibility with the CHARMM/CGenFF topology generation. Then, CGenFF server was used to generate gromacs compatible ligand topology and parameter files and, the ligand coordinate file was converted to its gromacs format. The protein-ligand complex was built by merging the ligand and protein coordinate files and the system topology was updated to include ligand specific parameters and atom count. The system was then placed in a dodecahedral simulation box with a 1.0 nm buffer from the protein surface and solvated with TIP3P water models and neutralized with Na⁺/Cl⁻ ions. To optimize the system’s geometry, energy minimization was performed using the steepest descent algorithm for 50,000 steps. Equilibration was then performed following the application of positional restraints to the heavy atoms of the ligand, allowing the solvent and protein to relax while maintaining the ligand in its binding site. Subsequently, we conducted equilibration in a canonical ensemble (NVT) for 100 ps with a 2fs time step using leap-frog integrator. Similarly, equilibration in an isothermal-isobaric thermodynamic ensemble (NPT) was followed by employing leap-frog integrator. The temperature was maintained at 300 K using the modified Berendsen thermostat with a constant time of 0.1 ps. During NPT equilibration, we maintained the pressure at 1 bar using the isotropic Berendsen barostat with a constant time of 2 ps. The final MD simulations were carried out with a time step of 2 fs and total simulation time of 50 ns. Covalent bonds involving hydrogen atoms were constrained using the LINCS algorithm in all the simulations. Long-range electrostatic interactions were managed using the Particle-Mesh Ewald (PME) method. FDA approved acetylcholinesterase inhibitor drugs, Donepezil (Lilienfeld, 2002; Shiget et al., 2001), and clinical inhibitors of GSK3β (Ramurthy et al., 2020) was used as a standard comparison. The obtained MD trajectories were subjected to the analysis of root mean square deviation (RMSD); root mean square fluctuations (RMSF) and radius of gyration (Rg) to assess the overall behavior of the protein and the local flexibility of its backbone.

Results

Protein Targets strongly Associated with Alzheimer’s Disease (AD). To identify top genes associated with AD, we used a widely adopted strategy in biomedical research based on normalized PubMed association scores (Wang et al., 2020; Grissa et al., 2022; Raj et al., 2024). Our literature-mining approach identified the top target proteins associated with AD. PubMed literature count for the top genes and their cooccurrence with AD is presented in Table 2. This table presents the number of publications mentioning each gene, and their co-occurrence with AD.

TABLE 2. Protein Targets strongly Associated with Alzheimer’s Disease.

The most frequently associated genes include γ secretase subunits, Presenilin 1 and 2 (PSEN1 and PSEN2), Amyloid precursor protein (APP), and β secretase enzymes (BACE 1) with a normalized PubMed association of 71%, 65%, 48% respectively. These top genes collectively represent a cascade of amyloidogenic pathway in AD. Mutations in PSEN1/PSEN2 alters cleavage site preference of γ-secretases, resulting to the increased production of longer and aggregation prone Aβ peptides (Willumsen et al., 2023; Kurth et al., 2023; Cai et al., 2015; Liu et al., 2024). Mutations in PSEN1 are the predominant cause of familial Alzheimer’s disease (fAD), whereas PSEN2 mutations are much rarer (Xiao et al., 2021). APP is the substrate for β and γ-secretases and central to Aβ production in AD (Sirisi et al., 2024). The Swedish mutation in APP makes its β-site to be more accessible to β-secretase enzymes (BACE1) and increases its cleavage efficiency, associated to increased production of Aβ (Zhang et al., 2017). Collectively, these top genes are major drivers of familial Alzheimer’s disease (FAD). Following, are microtubule associated protein tau (MAPT), apolipoprotein E4 (APOE4), acetylcholinesterase (AChE) and butyrylcholinesterase (BChE), with normalized association of 32%, 24%, 19% and 16% respectively. Hyperphosphorylation of tau and its aggregation into neurofibrillary tangles is a hallmark feature of AD (Giovinazzo et al., 2021). Glycogen synthase kinase three beta (GSK3β) and cyclin dependent kinase 5 (CDK5) are the major kinase proteins responsible for hyperphosphorylation of MAPT (Giovinazzo et al., 2021; Jayapalan and Natarajan, 2013). Hyperphosphorylation of MAPT decreases its binding affinity to microtubules resulting in its destabilization and cellular dysfunction (Man et al., 2023).

Apolipoprotein E4 (APOE4, 24.4%) is the strongest genetic risk factor in AD. It reduces clearance of Aβ and promotes its accumulation and aggregation (Raulin et al., 2022), enhance tau phosphorylation and aggregation (Koutsodendris et al., 2023) and disrupt blood brain barrier (BBB), increasing permeability and enabling entry of neurotoxic agents (Reas et al., 2024; Kirchner et al., 2023). The cholinesterase enzymes (AChE and BChE) are responsible for hydrolyzing acetylcholine, a neurotransmitter critical for learning and memory, and cause cholinergic deficits contributing to cognitive dysfunction in AD (Chen et al., 2022; Sridhar and Gumpeny, 2024). BChE activity increases in later stage AD and compensates for lost AChE, further lowering acetylcholine levels. Glycogen synthase kinase three beta (GSK3β, 7.4%) and Cyclin dependent kinase (CDK 5, 12.4%) are the most common kinases driving tau hyperphosphorylation in AD (Rekha et al., 2025; Fu et al., 2025). GSK3β phosphorylates tau at multiple sites associated with reduced microtubule binding, increased aggregation, and formation of neurofibrillary tangles. Brain derived neurotrophic factor is responsible for promoting neuron survival and plasticity and its depletion is associated with cognitive decline (Ismail et al., 2020; Gao et al., 2022). Overall, our PubMed co-occurrence analysis have identified significantly AD associated genes, and this result is in agreement with previous studies (Grissa et al., 2022). The PubMed-normalized frequencies reflect not only literature prevalence but also the centrality of these genes in known AD molecular mechanisms.

Molecular docking validation

To validate our molecular docking setup, we predicted the binding affinity, of the native ligands with their respective target, namely, Donepezil with AChE (PDB ID: 4EY7, K_d = 8 nM) (Cheung et al., 2012), KQP with APOE4 (PDB ID: 6NCO, K_d = 30 μM) (Petros et al., 2019), Verubecestat with BACE1 (PDB ID: 5HU1, K_i = 2.2 nM) (Scott et al., 2016), 65L with CDK5 (PDB ID: 7VDP, IC₅₀ = 2.3 nM) (Daniels et al., 2022), QQA with GSK3β (PDB ID: 6V6L, EC₅₀ = 130 nM) (Ramurthy et al., 2020), and Osugacestat (a.k.a. BMS906024) with PSEN1 (PDB ID: 8KCS, IC₅₀ = 21.7 nM) (Guo et al., 2025), Tacrine with AChE (PDB ID: 7XN1, IC₅₀ = 44.6 nM) (Alrouji et al., 2024)) and ANP (a.k.a SB216763) with GSK3β (PDB ID: 1PYX, IC50 = 34.3 nM) (Hopkins et al., 2014). The redocked co-crystalized ligand’s predicted binding affinities were compared with experimental binding data. Redocking accuracy was further assessed by calculating the RMSD between the predicted and crystallographic ligand poses. Redocking of co-crystallized ligand into their corresponding receptor yielded RMSD values ≤ 2.1Å (Table 3), indicating that the docking protocol reproduced the experimentally observed binding poses (Gupta et al., 2019). The correlation between the experimental and predicted ΔG values shown in Table 3 was moderately strong (R = 0.61). This correlation is acceptable, particularly since K_d values were unavailable for some ligands, and K_i, IC₅₀, and EC₅₀ values were used to calculate ΔG. As such, the correlation attests to the precision molecular docking.

TABLE 3. Predicted and experimental free-energy values (ΔG) of ligand/receptor pairs.

Molecular docking and multitarget screening

Molecular docking is an efficient technique to identify potential binding compounds to the biological target of interest (Alrouji et al., 2024). Here, a total of 2,847 phytochemicals retrieved from PubChem database were subjected to molecular docking using AutoDock Vina. The top 100 ligands of each target were retained based on docking binding affinity scores, resulting 362 unique phytochemical compounds. The fact that only 362, and not close to 1,000, different ligands bind to the top 10 protein targets, indicates that only a portion of the ligands bind to multiple targets, at the same time. Next, we screened based on multitarget binding and retained ligands that target three or more AD targets, reducing the dataset to 96 unique phytochemicals (Supplementary Table S1).

Following, we excluded ligands with more than 33 heavy atoms. Ligands with less heavy atoms are associated with better ligand efficiency metrics (Hopkins et al., 2014). In addition, ligands with heavy atoms counts in the range 10–33, are associated with better prediction scores of BBB penetration (>0.6), using the cubic polynomial function F(x)=(0.0000443x³−0.004556x²+0.12775x−0.463)/0.624231) developed by Weaver and coworkers (Gupta et al., 2019). This reduced the final dataset to 36 ligands with 33 heavy atoms, or less. Figure 1 describes the general procedure followed in our virtual screening.

Molecular docking results of the top hits of all phytochemicals to the AD targets are presented in Figure 2. To visualize the binding site, multitarget ligands were superimposed against their respective targets and visualized using PyMol.

Our docking result showed that top ligands of cholinesterase enzymes, AChE and BChE (Figures 2B,E respectively), potentially bind very strongly at the entrance of the gorge leading to the catalytic site occluding access to the active site (Žužek, 2024). Ligands may either effectively blocking the entrance peripheral anionic site (PAS) or occupying the catalytic active site (CAS), thus potentially preventing acetylcholine from reaching the active site, and from undergoing catalytic breakdown (Žužek, 2024). Our top ranked AChE ligands exhibited favorable binding energies (ΔG between - 11.0 and −9.0 kcal/mol) comparable to the co-crystallized AChE inhibitor, Donepezil (ΔG = −11.8 kcal/mol) approved by the FDA, supporting their promising potential as lead compounds for experimental validation of AChE inhibitors.

The top ligands of CDK5 and GSK3β preferentially docked strongly (ΔG between −10.0 to −8.4 kcal/mol and −10.3 to −8.6 kcal/mol respectively) inside the canonical ATP-binding pocket (Figures 2G,H), a well-characterized active site where many of the classical CDK5 and GSK3β inhibitors bind to prevent phosphorylation of downstream substrates (Gao et al., 2021; Zeb et al., 2019; Balboni et al., 2023; Shri et al., 2023). As such, our top and multitarget ligands may act similarly by occluding ATP-binding competitively thus potentially reducing CDK5 and GSK3β mediated phosphorylation. Notably, recent studies, have also identified highly potent CDK5 and GSK3β inhibitors at the same binding site, by employing structure guided design and cell-based assays (Balboni et al., 2023; Azm et al., 2025; Alhassan, 2025; Pandya et al., 2024; Daniels et al., 2022). These inhibitors, which were co-crystallized in the native PDB structures of CDK5 (Daniels et al., 2022) (PDB ID 7VDP, ligand ID: 65L) and GSK3β (Ramurthy et al., 2020) (PDB ID: 6V6L, ligand ID: QQA) were redocked and used as a standard internal reference of our molecular docking. The co-crystallized inhibitors were also predicted to bind in their native ATP binding-site with binding energies similar to those our top multitarget ligands. As such, our top ranked ligands exhibited favorable binding energies to CDK5 and GSK3β comparable, or better, to the co-crystallized inhibitors 65L (ΔG = −8.4 kcal/mol), and QQA (ΔG = −8.4 kcal/mol), supporting their promising potential as lead compounds for experimental validation.

The top ligands of PSEN1 (Figure 2I) potentially bind strongly (ΔG between −10.8 to −8.7 kcal/mol) at the active site cavity like many other γ-secretase inhibitors (GSIs), including the former drug candidates Semagacestat and Avagacestat, that have been shown to occlude the active site of PSEN1, and inhibit formation of Aβ peptides (Yang et al., 2021). Moreover, the top ligands predicted binding energies were comparable, or higher, than those of Osugacestat (ΔG -9.2 kcal/mol), the co-crystallized inhibitor of PSEN1 (PDB ID 8KCS, ligand ID: Q70) (Guo et al., 2025). As such, our top ligands are promising candidates for experimental validation of γ-secretase inhibitors.

The top ligands of BACE1 (Figure 2D) potentially binds at the active site determined by high-resolution X-ray crystallography, like other BACE1 inhibitors (Pettus et al., 2020; Rueeger et al., 2021). The predicted binding energies of our top ligands of BACE1 (ΔG between −9.3 to −8.1 kcal/mol) were comparable or better to those of Verubecestat (ΔG −8.3 kcal/mol), the co-crystalized ligand of BACE1 PDB structure (PDB ID 5HU1, ligand ID: 66F) (Scott et al., 2016). Like Verubecestat our top ligands may competitively block APP substrate access to the catalytic site, thereby disrupting amyloid-β generation via BACE1 cleavage, pending experimental verification.

The top ligands of APOE4 potentially binds strongly (ΔG between −9.6 and −7.4 kcal/mol), with binding poses and affinities comparable with the previously identified APOE4 stabilizer, co-crystallized in the native PDB structure (PDB ID 6NCO, ligand ID: KQP) (Petros et al., 2019), (Figure 2C). Like KQP, our top ligands may also contribute to stabilizing the APOE4 pending experimental validation.

Collectively, the higher binding affinity to these targets can be attributed to the presence of well-defined pocket, or groove-like architectures, on the protein structures, which provide shape complementarity and persistent interaction hotspots (Cucuzza et al., 2023; Spassov, 2024).

Supplementary Table S1 lists 96 multitarget ligands that potentially bind to 3, or more, biological targets, including Aβ, AChE, APOE4, BACE1, BChE, BDNF, CDK5, GSK3β, PSEN1 and PSEN2. The multitarget ligands that potentially bind to more than a single target, are promiscuous, and have been reported to play an important role in polypharmacology (Haupt et al., 2013). The multitarget ligands are often considered for drug repurposing strategies, and have shown increasing success rates (Haupt et al., 2013; Comajuncosa-Creus et al., 2024). The multitarget ligands exhibit remarkable adaptation to multiple binding sites and have been known to expand rational design of polypharmacologic drugs (Duran-Frigola et al., 2017). Polypharmacology is a new paradigm shift in drug discovery for complex diseases where multiple proteins and pathways are involved in the onset and development of the disease, due to cumulative efficacy at all of its individual targets (Kabir and Muth, 2022). Hence, multitarget (promiscuous) ligands are considered promising in polypharmacology to address several key disease drivers simultaneously (Kabir and Muth, 2022; Abdelsayed, 2025) and, due to this, current research supports the concept of polypharmacology and intentionally incorporating promiscuity in molecules to make them polypharmacologic (Kabir and Muth, 2022). However, the promiscuity of multitarget ligands also potentially increase adverse effects due to non-specific pharmacological activity (Abdelsayed, 2025). Highly promiscuous ligands may exert unintended off-target effects and may pose toxicity. Thus, moderate multitarget ligands, binding some, but not all targets, may be advantageous, and balance therapeutic efficacy versus off-target toxicity. Therefore, not all multitarget phytochemicals may be beneficial for therapeutic use in AD, and moderately promiscuous ligand may be preferred, as seen later in our discussion.

Supplementary Table S1 also lists the physicochemical descriptors that broadly influence the ADMET (absorption, distribution, metabolism, excretion and toxicity) properties of small molecules and commonly used to define drug-likeness (Leeson et al., 2021; Rawat et al., 2024). For comparison, the physicochemical properties of our top hits, including with co-crystallized approved AChE inhibitor (Donepezil) and clinical inhibitors of APOE4 (KQP), BACE1 (Verubecestat), CDK5 (65L), GSK3β (QQA) and PSEN1 (Osugacestat) are presented in Supplementary Table S1. Multiple studies consistently shown that successful CNS drugs exhibit narrower physicochemical space than general drugs mainly to cross the blood brain barrier (BBB). Analysis of CNS drugs tend to exhibit lower molecular weight, topological polar surface area, moderate lipophilicity and smaller number of hydrogen bond donors and acceptors to facilitate passive diffusion across the BBB and enhance membrane permeability (Dehnbostel et al., 2024; Kenny, 2022; Markowicz-Piasecka et al., 2022; Odi et al., 2020). Finally, lower heavy atom counts are often associated with improved BBB permeability because they correlate with lower molecular weight and lower polarity (Gupta et al., 2019). While the strict rules are for oral bioavailability, it also affects the general ADMET properties of molecules. Here, we propose to focus on the smallest ligands, with less than 33 heavy atoms. This measure considers ligand efficiency (LE) metrics per heavy atom, where smaller compounds with good potency are preferred. Figure 3 shows 36 multitarget ligands and contain 33 or less heavy atoms, and that bind three, or more, AD protein targets. These 36 multitarget ligands emerge as promising polypharmacologic agents for experimental validation.

Notably, of our smallest 36 top multitarget ligands, 11 fulfill Lipinski’s rule of five for oral availability, 11 have no violations (Rutecarpine, Sanguinarine, Strychnine, Cyclopamine, Lactucopicrin, Hecogenin, Jervine, Ruscogenin, Verticine, Verticinone, Obacunone (Lipinski et al., 2001) and 25 have one violation (logP). Of the remaining 60 top multitarget ligands, for example, Amentoflavone, Robustaflavone, Cassiamin C and Ginkgetin have more than one violation (Lipinski et al., 2001). In comparison, the FDA approved Donepezil and other co-crystalized inhibitors also fulfill Lipinski’s rule of five. Rutecarpine, which potentially binds to AChE, CDK5 and GSK3β, is the smallest molecule, with narrow physicochemical properties space, making it an attractive CNS relevant molecule (Xu et al., 2022). Here, we focus on the 36 top ligands, with the lowest number of heavy atoms.

Interaction analysis

To better understand the binding patterns of the complexes, we performed interaction analysis for selected molecules utilizing Protein-Ligand Interaction Profiler (PLIP), an automated web-based tool to analyze protein-ligand interactions based on the PDB file format of protein-ligand complexes (Schake et al., 2025). The protein ligand complex was first formed from the AutoDock Vina docking results using PyMol and then provided to PLIP for interaction analysis. Molecules commonly targeting AChE and GSK3β (Lactucopicrin, Oleanolic acid and Rutecarpine) were used for interaction analysis and their interactions were compared with the co-crystalized ligands. The hydrogen bond interacting residues identified by PLIP were highlighted using PyMol in the protein’s structure (Figure 4).

Accordingly, Lactucopicrin formed hydrogen bonds with TYR72, TYR124 and SER293 of AChE (Figure 4A). TYR72 and TYR124 are part of the peripheral anionic site located at the entrance of the active site gorge known to modulate substrate passage and inhibitor binding (de Almeida et al., 2023). Molecules binding to these regions such as Kaempferol, induce conformational changes and inhibit catalytic activity of AChE (Shi et al., 2023). Oleanolic Acid formed hydrogen bond with ARG247 of AChE (Figure 4B), a residue identified as part of the acyl loop door and binding to this site can participate in a non-competitive inhibition mechanism of the enzyme’s activity (Moghadam et al., 2021). Rutecarpine formed π-π stacking interactions with residues TRP286 and TYR341 of the active site of AChE (Figure 4C), similarly as the native ligand, Donepezil (Gondal et al., 2023). Notably, TRP286 also formed π-π stacking interactions with Rutecarpine, lactucopicrin and donepezil, indicating its hotspot recurrent interacting residues. In comparison, the native AChE ligand Donepezil formed hydrogen bonding with PHE295 (Figure 4D), similar to previous reports (Gupta et al., 2024), validating the setup of our molecular docking experiments, and of our other prediction.

Similarly, the same ligands form multiple interactions with active site residues of GSK3β. Lactucopicrin forms hydrogen bonds with ASP200 and LYS85 (Figure 4E) and oleanolic acid formed salt bridges with ARG141 and ARG144 (Figure 4F). The native ligand QQA formed hydrogen bond with ASP133 and VAL135 (Figure 4G). The identified residues are part of the ATP binding pocket and previously reported to be consistent interacting residues in small molecule inhibitors of GSK3β (Alhassan, 2025; Lu et al., 2018).

Unlike hydrophobic interactions, which are the predominant interactions in protein-ligand complexes and largely non-directional and entropy driven, hydrogen bonds are highly directional, with defined donor acceptor distances and angles that govern ligand specificity in biological recognition (Ferreira de Freitas and Schapira, 2017; Bissantz et al., 2010). As such, ligands forming hydrogen bond at the functional site of the receptors may be effective in orienting the ligand in fixed functional positions.

Molecular dynamics simulations

To better evaluate ligand binding stability, molecular dynamics (MD) simulations were performed for the selected phytochemicals targeting both AChE and GSK3β (Lactucopicrin, Oleanolic Acid, and Rutecarpine). Importantly, the AChE backbone remained globally stable in complex with these ligands, exhibiting maximum backbone RMSD values of approximately 2.5 Å (Figure 5A), consistent with structurally stable protein-ligand complexes (Weng et al., 2021; Verma et al., 2022). Notably, Rutecarpine derivatives have been shown to inhibit AChE (Wang et al., 2010). Likewise, Oleanolic acid was found to inhibit AChE (Stępnik et al., 2023). Also, lactucopicrin have been shown to inhibit AChE (Venkatesan et al., 2017). In GSK3β, ligand binding displayed lower backbone RMSD compared to the apo (unbound) form (Figure 5B), indicating reduced conformational fluctuations upon ligand binding. Interestingly, Rutecarpine has been shown to ameliorate spatial memory deficits and enhance synaptic plasticity in AD mice by modulating the GSK-3β signaling pathway (Li et al., 2023). Likewise, Oleanolic acid alleviates cerebral ischemia/reperfusion injury via regulation of the GSK-3β/HO-1 signaling pathway (Lin et al., 2021). Notably, throughout the simulation Lactucopicrin demonstrated greater structural stability than the co-crystallized control ligand, QQA, in GSK3β (Figures 5B,D,F). RMSF analysis revealed residue-level fluctuations generally below 1.0 Å for both AChE and GSK3β complexes (Figures 5C,D), supporting overall structural stability. Consistently, radius of gyration (Rg) analysis demonstrated preserved global compactness of both protein structures throughout the simulations (Figures 5E,F), further confirming the structural integrity of the ligand-bound systems.

Collectively, our MD simulation results indicate that Lactucopicrin, Oleanolic Acid, and Rutecarpine exhibit similar, or even improved dynamic stability, compared to both apo proteins and co-crystallized reference inhibitors, supporting the potential of our phytochemicals as promising candidates for experimental evaluation in AD therapy.

Discussion

The 36 top multitarget phytochemicals are classified into: Alkaloids (Rutecarpine, Sanguinarine, Strychnine), Steroidal alkaloids (Tomatidine, Cyclopamine, Jervine, Verticine, Verticinone), Steroidal sapogenins (Diosgenin, Hecogenin, Yamogenin, Tigogenin, Smilagenin, Sarsasapogenin, Neogitogenin, Ruscogenin), Triterpenoids/sterols (Alpha-spinasterin, Fucosterol, Stigmasterol, Beta-sitosterol, Gamma-sitosterol, Isofucosterol, Germanicol, Lupeol, Beta-amyrin, Alpha-amyrin, Taraxerol, Friedelin, Tetrahymanol, Tingenone, Parkeol, Methylenecycloartanol, Ursolic acid, Oleanolic acid, Obacunone), and Sesquiterpene (Lactucopicrin).

The list of top 36 final phytochemicals along with their chemical structure are presented in Table 4.

TABLE 4. Classification of chemical structures of 36 multitarget phytochemicals.

Alkaloids (Rutecarpine, Sanguinarine, Strychnine)

Alkaloids constitute a major class of nitrogen-containing natural products characterized by extensive structural diversity and a wide range of pharmacological as well as toxicological activities (Kita et al., 2025). Due to their ability to interact with multiple cellular signaling pathways and molecular targets, many alkaloids have demonstrated significant neuroprotective, anti-inflammatory, antioxidant, and neuromodulatory effects relevant to the pathogenesis of AD (Liu et al., 2023). Among the compounds investigated in this study, Rutecarpine has emerged as a particularly promising candidate with multimodal neuroprotective potential. Rutecarpine has been reported to ameliorate spatial memory deficits and prevent tau protein hyperphosphorylation through inhibition of glycogen synthase kinase-3β (GSK3β). In addition, it has been shown to improve cognitive impairment by restoring mitochondrial function via activation of the AMPK/PGC1α signaling pathway (Gong and Jia, 2023). Consistent with these findings, Rutecarpine may exert beneficial effects in AD by suppressing platelet activation, reducing oxidative stress, and attenuating neuroinflammatory signaling while concurrently inhibiting the PI3K/Akt/GSK3β pathway implicated in tau hyperphosphorylation, thereby potentially alleviating cerebrovascular dysfunction and broader AD-associated neurodegenerative processes (Cao et al., 2024). Importantly, our molecular screening further supports these mechanistic insights, as Rutecarpine demonstrated a strong binding affinity toward GSK3β with a docking score of −9.4 kcal/mol, in agreement with previous experimental and computational studies (Huang et al., 2021).

In contrast, other alkaloids in this group are primarily associated with toxicity rather than therapeutic potential. Strychnine, a well-known neurotoxic compound, has been studied in the context of strychnine-sensitive glycine receptors, which are implicated in inhibitory neurotransmission and may be indirectly relevant to neurophysiological dysfunction in AD (Bukanova et al., 2016). However, due to its potent cytotoxic, vasoconstrictive and hypertensive effects, Strychnine is more likely to exacerbate neurovascular damage and contribute to disease pathology than to provide therapeutic benefit (Vezikov and Simpson, 2026). Similarly, Sanguinarine has been shown to induce apoptosis through reactive oxygen species generation in parasitic models (Li et al., 2022), reflecting its strong cytotoxic potential. Taken together, while certain alkaloids such as Rutecarpine demonstrate promising multitarget neuroprotective mechanisms relevant to AD, many compounds within this chemical class exhibit pronounced toxicity, limiting their suitability for direct therapeutic repurposing in AD.

Steroidal alkaloids (Tomatidine, Cyclopamine, Jervine, Verticine, Verticinone)

Steroidal alkaloids are major class of nitrogen containing natural products with structural diversity and broad pharmacological and toxicological activities (Daley and Cordell, 2021). Steroidal alkaloids identified among our top-ranked compounds have been increasingly associated with modulation of key pathological processes involved in AD, including tau hyperphosphorylation, amyloid processing, neuroinflammation, oxidative stress, and neuronal survival, highlighting their potential as multifunctional therapeutic.

Tomatidine, a steroidal alkaloid abundant in unripe tomatoes, has demonstrated neuroprotective effects, ameliorated cognitive deficits and reduced tau hyperphosphorylation (Zhang et al., 2025). In addition, Tomatidine significantly attenuated neuroinflammation and prevented microglial activation (Kita et al., 2025) both of which are key contributors to AD progression. These findings support its role as a multitarget neuroprotective agent capable of modulating multiple disease-relevant pathways. Likewise, Cyclopamine modulates γ-secretase-mediated cleavage of amyloid precursor protein by altering its subcellular trafficking and lysosomal degradation (Vorobyeva et al., 2014). Likewise, Cyclopamine rescues structural and behavioral deficits in a Drosophila model of AD (Nguyen et al., 2017).

On the other hand, Jervine, another steroidal alkaloid, is a potent dual-binding site inhibitors of both acetylcholinesterase and butyrylcholinesterase (Zheng et al., 2019). Jervine Verticine and Verticinone have been shown to inhibit both AChE and BChE and prevent Aβ aggregation (Zhu et al., 2026). Verticine and Verticinone also exhibit anti-inflammation, hepatoprotective, antitussive, and anticancer effects (Yin et al., 2019). However, Jervine, Verticine, and Verticinone have also shown cytotoxic and anti-tumor effects, potentially limiting their application in AD (Zheng et al., 2019).

Steroidal alkaloids are particularly noteworthy because their steroid-like hydrophobic scaffolds may facilitate interactions with membrane-associated proteins and signaling pathways implicated in neuronal survival and neuroinflammation. In addition, the presence of nitrogen-containing functional groups may enhance their capacity to interact with catalytically active residues within disease-associated protein targets and intercalate into DNA double-strands to modulate gene expression. Such structural features may contribute to multitarget pharmacological behavior increasingly considered advantageous for AD therapeutics. (Passos et al., 2022).

Steroidal sapogenins (Diosgenin, Hecogenin, Yamogenin, Tigogenin, Smilagenin, Sarsasapogenin, Neogitogenin, Ruscogenin)

Steroidal sapogenins are a structurally related class of steroid-derived natural products characterized by a hydrophobic steroidal core and varying oxygen-containing substituents that confer distinct biological activities. In this study, the steroidal sapogenins potentially bind to multiple AD protein target, in agreement with their neuroprotective, anti-inflammatory, antioxidant, and multitarget pharmacological properties relevant to the pathogenesis of Alzheimer’s disease (AD). In general, steroidal sapogenins have been shown to suppress neuroinflammatory signaling, while simultaneously attenuating oxidative stress and apoptotic pathways (Passos et al., 2022). In addition, steroidal sapogenins have demonstrated neuroprotective activity in both cellular and animal models of neurodegeneration (Sun et al., 2015). Among these compounds, Diosgenin reduces amyloid-β (Aβ) accumulation, suppresses oxidative stress and neuroinflammatory responses, and significantly improves spatial learning and memory performance (Som et al., 2022). Diosgenin also enhances cognitive function in healthy adults (Tohda et al., 2017). Ruscogenin also reduces neuroinflammation in rodents and attenuates ischemia-related BBB dysfunction via NF-κB and anti-inflammatory mechanisms (Cao et al., 2016). Moreover, Ruscogenin reduces oxidative stress and pro-inflammatory cytokine production in microglial cells in Parkinson’s disease mouse models, resulting in significant neuroprotective effects (Wu et al., 2023). In our study, Diosgenin strongly binds to amyloidogenic pathway associated proteins such as BACE1 and PSEN1, tau pathology associated proteins such as CDK5 and with the strong genetic factor protein APOE4 associated to altered clearance of Aβ. Notably, Hecogenin has also been computationally predicted as a potential BACE1 inhibitors (Padmanabhan et al., 2023). Yamogenin exhibits antidepressant effects and inhibits endoplasmic reticulum stress and microglial activation in lipopolysaccharide-induced mice (Chen P.-L. et al., 2023). Sarsasapogenin, another steroidal sapogenin, has shown to significantly inhibit key enzymes involved in pathogenesis of AD such as AChE, BChE, BACE1, and significantly inhibits Aβ42 fibrillization (Kashyap et al., 2020). Smilagenin is structural isomer of Sarsasapogenin and is likely to induce similar effects. Sarsasapogenin-derived timosaponin has also been proposed as novel Aβ-lowering modulators of APP processing (Sy et al., 2016). Collectively, these observations suggest that other top-ranked steroidal sapogenins identified in our study may exert therapeutic effects through multiple complementary mechanisms, including inhibition of amyloidogenic processing, modulation of tau-associated signaling pathways, suppression of neuroinflammatory cytokine production, attenuation of oxidative stress, and preservation of BBB integrity. Such multitarget activity is particularly advantageous in AD, a highly multifactorial neurodegenerative disorder involving interconnected pathological processes rather than a single dominant molecular driver. Consequently, steroidal sapogenins represent a promising class of naturally derived compounds for further investigation, and experimental validation, as multifunctional therapeutic candidates for AD.

Triterpenoids/sterols (Alpha-spinasterin, Fucosterol, Stigmasterol, Beta-sitosterol, Gamma-sitosterol, Isofucosterol, Germanicol, Lupeol, Beta-amyrin, Alpha-amyrin, Taraxerol, Friedelin, Tetrahymanol, Tingenone, Parkeol, Methylenecycloartanol, Ursolic Acid, Oleanolic Acid, Obacunone)

Triterpenoids represent a diverse class of naturally occurring terpenoid-derived compounds that have attracted increasing attention because of their broad pharmacological activities, particularly their antioxidant, anti-inflammatory, anti-apoptotic, and cytoprotective effects (Ling et al., 2022). These biological effects are highly relevant to the pathogenesis of Alzheimer’s disease (AD), in which oxidative stress, chronic neuroinflammation, synaptic dysfunction, and progressive neuronal degeneration collectively contribute to cognitive decline. Notably, many of our top ranked and multitarget triterpenoids identified in our study are supported by previous experimental evidence demonstrating neuroprotective activity in neurodegenerative disease models. Among these compounds, Ursolic acid has shown to improve spatial and recognition memory, to enhance adult hippocampal neurogenesis, and to restore synaptic marker expression in an Aβ injected mouse models (Mirza et al., 2021). Additional studies further showed that Ursolic acid attenuates Aβ-induced behavioral abnormalities, including aggression-like phenotypes, and promotes the clearance of toxic Aβ species from the brain (Wang N. et al., 2022). Oleanolic acid, a structural isomer of Ursolic acid, has similarly shown neuroprotective effect through inhibition of tau hyperphosphorylation and neuroinflammation, leading to improved cognitive function (Liu et al., 2025). Stigmasterol ameliorates neuroinflammation in an AD mouse model and attenuates inflammatory response of microglia via NF-κB, and NLRP3 signaling by AMPK activation (Jie et al., 2022). Likewise, β-Sitosterol ameliorates the cognitive deficits and neuropathological hallmarks in an AD mouse model (Ye et al., 2020). Fucosterol provides protection against amyloid β-induced neurotoxicity, reduces intracellular levels of amyloid β in a cellular model of AD (Gan et al., 2019).

These collective observations reinforce the concept that triterpenoid scaffolds possess multifunctional neuroprotective properties that extend across several neurodegenerative disease mechanisms.

Importantly, AD is a highly multifactorial disorder involving simultaneous dysregulation of amyloid processing, tau phosphorylation, oxidative stress, mitochondrial dysfunction, synaptic degeneration, and chronic neuroinflammation. Consequently, compounds capable of modulating multiple pathological pathways are increasingly recognized as promising therapeutic candidates. In this context, the multitarget interactions observed for the triterpenoids and sterol-like compounds identified in our study suggest that these molecules may attenuate AD pathology through coordinated modulation of several disease-associated protein targets. Their combined antioxidant, anti-inflammatory, and neuroprotective properties therefore support their potential utility as multifunctional therapeutic candidates for AD.

Sesquiterpene (Lactucopicrin)

Lactucopicrin is a sesquiterpene lactone that has been reported to exert strong anti-inflammatory and neuroactive properties relevant to the pathogenesis of Alzheimer’s disease (AD). Lactucopricrin has been identified as a potent inhibitor of the NF-κB inflammatory pathway, a central regulator of neuroinflammatory responses, suggesting anti-inflammatory potentialy (Ávila-Gálvez et al., 2025). In addition, it has been shown to inhibit AChE, which is typically impaired in AD (Jaśkiewicz et al., 2022).

Consistent with these findings, our molecular docking results demonstrate that Lactucopicrin binds strongly to multiple AD-relevant targets, including AChE, with a binding energy of −9.5 kcal/mol. This strong binding affinity supports previously reported mechanistic evidence and further suggests that Lactucopicrin may exert neuroprotective effects through a combination of anti-inflammatory and cholinesterase inhibitory mechanisms, both of which are highly relevant to AD pathology.

Generally, the appearance of many phytochemicals in our top and multitarget hits with experimental evidence neuroprotection in AD and other neurodegeneration disease models shows the robustness of our screening approach and, as such, our other top and multitarget phytochemicals may potentially be relevant for the treatment of Alzheimer’s disease.

Overall, several of the identified phytochemicals, including Rutecarpine, Diosgenin, Oleanolic acid, Ursolic acid, Lactucopicrin, Hecogenin, Sarsasapogenin, Fucosterol, β-Sitosterol, and Stigmasterol, have previously been reported to exhibit neuroprotective or AD-relevant activities through diverse mechanisms. Our multitarget virtual screening therefore provides additional mechanistic hypotheses by identifying molecular targets through which these compounds may exert their experimentally observed pharmacological effects.

Limitations

As a potential limitation to this study, it is crucial to note that our findings are based on computational work. For example, molecular docking is inherently limited by the accuracy of scoring functions. Therefore, the beneficial or deleterious effect of these phytochemicals in AD treatment remains to be validated experimentally and clinically. In addition, clinical validation is required to evaluate the overall effect of promising candidates for repurposing.

Another possible limitation to this study is that the literature-mining method used in this study is based on normalized PubMed association scores and is thus subject to the current state of the published literature. As a result, well-studied genes may receive higher association scores than lesser studied genes with emerging roles in AD. Although this strategy provides an objective and reproducible way to rank heavily studied targets, it may be biased against new or poorly studied proteins.

As another potential limitation to this study, multitarget phytochemicals do not necessarily indicate a better chance at AD treatment. In fact, multitarget phytochemicals could indicate less discriminate selectivity, and more potential toxicity. Thus, a balance between too much selectivity, and too little selectivity can be required to identify promising candidates.

In addition, molecular dynamics simulations were performed for 50 ns in a single run for each protein-ligand complex. Although the results consistently supported complex stability, replicate and extended simulation could provide additional confidence in the observed dynamic behavior.

Finally, as a limitation of this paper, the potential candidates for phytochemical treatment of AD include the top 96 ligands, but not automatically any other members of their drug families.

Conclusion

In conclusion, our integrated computational analysis highlights that multiple natural product families including alkaloids, steroidal alkaloids, steroidal sapogenins, triterpenoids/sterols, and sesquiterpenes exhibit consistent multitarget interactions with key proteins involved in Alzheimer’s disease pathology, including Aβ, AChE, BChE, APOE4, BACE1, BDNF, CDK5, GSK3β, PSEN1 and PSEN2. Some of our claims are supported by the literature, which shows that phytochemicals can inhibit amyloid aggregation, tau phosphorylation, neuroinflammation, and cholinergic dysfunction. Finally, the view that some of our claims are supported by literature lends credibility to our unsupported findings, and several of the identified structurally diverse phytochemicals represent a valuable source of multitarget-directed lead compounds for further development in AD therapeutics.

Generative AI statement

The author(s) declared that generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fddsv.2026.1922953/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Marco A. Loza-Mejía, Universidad La Salle, Mexico

Reviewed by: Adam Dawria, King Khalid University, Saudi Arabia

Anguraj Moulishankar, Acharya and BM Reddy College of Pharmacy, India