Repurposing of natural products for spinocerebellar ataxia type 3 using integrated network pharmacology and in silico approaches

Abstract

Spinocerebellar ataxia type 3 (SCA3) is a progressive inherited neurodegenerative disorder characterized by impaired coordination and balance due to cerebellar ataxia. Currently, there are no Food and Drug Administration (FDA)-approved drugs or curative treatments for SCA3. This study aimed to uncover the potential therapeutic mechanism of natural products (NP or NPs) against SCA3 using an integrative computational strategy combining network pharmacology, molecular docking, molecular dynamics (MD) simulation, principal component analysis (PCA), free energy landscape (FEL), dynamic cross-correlation matrix (DCCM), and Molecular Mechanics/Poisson–Boltzmann Surface Area (MM/PBSA) calculations. A total of 4,986 SCA3-related targets were retrieved from the GeneCards database, and 696 unique NP-related targets were obtained from SwissTargetPrediction and SuperPred databases. Protein–protein interaction (PPI) network analysis revealed 239 overlapping targets. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis identified the MAPK signaling pathway (hsa04010) as a key mechanism involving 33 critical proteins, including AKT1, TP53, TNF, EGFR, RELA, and MAPK1. Among the screened compounds, crocin exhibited the highest binding affinities with AKT1 (− 9.5 kcal/mol) and TP53 (− 6.6 kcal/mol) in docking studies. Further validation through MD simulation, PCA, FEL, DCCM, and MM/PBSA analysis confirmed the stability and strong interaction of crocin with these targets. This study highlights crocin as a promising candidate for SCA3 therapy and provides a foundation for future experimental validation.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-30652-8.

Introduction

Spinocerebellar ataxia type 3 (SCA3), also known as Machado-Joseph disease, is an autosomal dominant cerebellar ataxia characterized by motor incoordination, progressive external ophthalmoplegia, progressive functional disabilities, and other neurological manifestations. The worldwide prevalence of SCA3 is estimated to be 1 in 50, 000 to 100,000, with significant variation depending on geographical location and ethnic background¹. A high prevalence of SCA3 has been observed in North America (Canada and United States), South America (Brazil), Southern Europe (Portugal), Western Europe (Germany and the Netherlands), East Asia (China and Japan), and Oceania (Australia)². However, it remains challenging to accurately assess the prevalence of this disease due to low testing rates, delayed or incorrect diagnoses across all hereditary ataxias, and lack of patient registries. At present, there is no epidemiological data on SCA in Malaysia, mainly due to the lack of local genetic testing facilities. It is estimated that 1 in 100,000 Malaysians suffer from SCA, with SCA3 being the most common form, followed by SCA2 and SCA1³.

Mutations in the ataxin-3 gene (ATXN3) cause conformational changes in ataxin-3, leading to the onset of SCA3. The mutant ataxin-3 disrupts axonal transport, neuronal energy metabolism, and chromatin organization, further driving SCA3 pathology^4,5. Specifically, the expansion of cytosine-adenine-guanine (CAG) trinucleotide repeats in exon 10 of ATXN3 mapped to chromosome 14q32.1 causes improper folding of functional proteins. In unaffected individuals, the CAG repeats range from 12 to 44, whereas in patients with SCA3, the CAG repeats range from 56 to 80 or more⁶, leading to an expanded polyglutamine (PolyQ) sequence in the translated mutant ataxin-3 protein. This mutation has been shown to alter signaling proteins that modulate protein translation and decay, DNA damage repair, neuron development, and the formation of cytoskeletal proteins. Aggregates of mutant polyQ proteins can be taken up by mitochondria, resulting in diminished mitophagy and aggravated mitochondrial dysfunction, subsequently leading to the increased production of reactive oxygen species (ROS)⁷ and oxidative stress. Individuals with the mutant allele can experience gait instability, oculomotor issues, cognitive impairment, dysarthria, and pyramidal (e.g., spasticity, muscle weakness, and hyperreflexia) and extrapyramidal (e.g., involuntary movements) signs². The mean age at the onset of symptoms is generally in the third or fourth decade of life. Currently, there is no cure for this devastating, severely disabling, and life-shortening disease. Although some pharmacological interventions have shown promise for the management of SCA3, the adverse effects may become intolerable⁸.

Drug repurposing can greatly accelerate drug discovery by reducing time, cost, and risk⁹. Unlike traditional drug development, repurposed compounds already have established safety profiles^9,10. Moreover, drug candidate discovery can be improved by in silico techniques including modeling and machine learning^9,11. Such an approach improves drug approval rates, with 30% of repurposed drugs reaching the market versus 10% of new drugs¹². Natural products (NP or NPs) are central to drug discovery, contributing to advancements in modern medicines¹³. Their structural diversity and biological compatibility make them invaluable for treating cancers and multi-drug-resistant infections^14,15. Moreover, NPs targeting oxidative stress and protein aggregation show much promise for treating SCA3. Flavonoids such as chamanetin have been shown to bind to mutant ataxin-3 (− 7.2 kcal/mol) and exhibit favorable pharmacokinetics¹⁶. In the past decade, there has been much interest in traditional Chinese medicine (TCM) and its compounds due to their highly distinctive classes of phytochemicals and molecular target profiles. There is accumulating evidence that shows some TCM formulations and their representative NPs (alkaloids, bibenzyls, flavonoids, phenylpropanoids, quinones, and terpenoids) can slow the progression of neurodegenerative diseases^17–28. Therefore, repurposing NPs by leveraging their pharmacological diversity can be a cost-effective strategy^29,30. However, there have been no clinical trials that have evaluated and validated these compounds in SCA3 patients³¹.

The emergence of in silico methods, particularly screening of NPs derived from TCM has transformed drug discovery paradigms, offering cost-effective alternatives to conventional testing approaches and addressing challenges in drug development^32–34. By leveraging advanced algorithms and machine learning, in silico methods can accelerate compound identification and reduce clinical trial failures^35,36. Network pharmacology has emerged as a valuable approach in drug discovery, as it can systematically explore disease mechanisms and drug interactions^37–41. By integrating screening data from multiple databases, network-based predictions can help identify key pharmacological mechanisms from the perspective of biological network homeostasis⁴². For example, molecular docking can predict receptor-ligand interactions, whereas molecular dynamics (MD) simulation can refine drug structures, validate docking results, and assess ligand binding dynamics at atomic level⁴³. With increasing computational power, MD simulations are becoming crucial for designing safer and more effective therapies⁴⁴. Moreover, computational modeling of drug interactions and biological processes can also eliminate ethical concerns related to animal testing. With advances in computational technologies, these approaches will continue to streamline drug development and guide experimental designs, while minimizing reliance on animal models³⁶.

In this study, we collected NPs with activity against SCA3 from the literature^45–52. We then used a network pharmacology technique to design gene networks to investigate the related pathways and biological processes, which showed the candidate NPs primarily acted through key signaling pathways, specifically the hsa04010: MAPK signaling pathway. Further molecular docking studies and MD simulation with principal component analysis (PCA), free energy landscape (FEL), dynamic cross-correlation matrix (DCCM), and Molecular Mechanics/Poisson–Boltzmann Surface Area (MM/PBSA) analysis revealed the lead compound was an inhibitor of TP53 and AKT1. These findings provide significant evidence for the therapeutic potential of NPs against SCA3.

Materials and methods

Database and research process

Table 1 shows the databases used in this study and Fig. 1 shows the research approach.

Table 1.

Key details of the databases used for screening NPs against SCA3.

Database	Link
SwissTargetPrediction	http://www.swisstargetprediction.ch/
SuperPred	https://prediction.charite.de/subpages/target_prediction.php/
GeneCards	https://www.genecards.org/
Venny 2.1.0	https://bioinfogp.cnb.csic.es/tools/venny/index.html
Panther classification	http://www.pantherdb.org/
STRING	https://cn.string-db.org/
Cytoscape 3.10.2	https://cytoscape.org/download.html
DAVID	https://davidbioinformatics.nih.gov/
Bioinformatics	http://www.bioinformatics.com.cn/

DAVID: Database for Annotation, Visualization, and Integrated Discovery, STRING: Search Tool for the Retrieval of Interacting Genes/Proteins.

Fig. 1.

The detailed flowchart outlines the network pharmacology approaches used in this study. MD: molecular dynamics, MM/PBSA: Molecular Mechanics/Poisson–Boltzmann Surface Area, NP: natural product, PCA: principal component analysis, PPI: protein–protein interaction, SCA3: spinocerebellar ataxia type 3.

NP-related targets

In this study, we collected NPs with activity against SCA3 from the literature^45–52. The selected compounds (Table S1) were converted to structures using ChemDraw software and their SMILES were obtained. The SMILES representations were processed using the SwissTargetPrediction (http://www.swisstargetprediction.ch/) and SuperPred (https://prediction.charite.de/subpages/target_prediction.php/) databases to identify potential human target genes associated with the NPs. The retrieved target datasets were subsequently integrated, and duplicate entries were removed to generate a final list of predicted NP-associated targets.

SCA3-related targets

To systematically identify SCA3-related targets, the keywords ‘Spinocerebellar ataxia type 3’ and ‘Machado-Joseph disease’ were searched in the GeneCards database (https://www.genecards.org/). The search was conducted under the ‘All’ category to comprehensively capture disease-associated entries, including protein-coding genes, RNA genes, genetic loci, functional elements, pseudogenes, and uncategorized gene clusters. The retrieved targets were then filtered in a stepwise manner. First, duplicate entries arising from overlapping categories were eliminated to generate a non-redundant gene set. Next, targets were prioritized based on the GeneCards relevance score, which integrates evidence from multiple sources such as genomic annotations, functional studies, and literature mining. Genes with higher relevance scores were considered to have stronger associations with SCA3 and were thus selected as key candidates for downstream analysis. To ensure biological reliability, overlapping targets that appeared under multiple categories (e.g., protein-coding and RNA-associated functions) were cross-validated and retained only if supported by multiple lines of evidence. This approach enabled the refinement of the raw GeneCards output into a curated, high-confidence set of unique potential SCA3-associated genes suitable for subsequent network pharmacology and pathway analysis.

NP-SCA3 common target screening and protein–protein interaction network construction

Venny 2.1.0 was used to determine the intersection between NP-associated targets and SCA3-related targets (https://bioinfogp.cnb.csic.es/tools/venny/index.html). Overlapping genes were identified as possible therapeutic targets of NPs against SCA3. The related proteins were categorized according to their functions using the Panther categorization system (http://www.pantherdb.org/). The PPI network was built for common targets using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database (https://cn.string-db.org/)⁵³ with the species limited as “Homo sapiens”. Within each network, all interactions had confidence scores ≥ 0.4 (medium + high confidence). After importing the results to a .tsv file, Cytoscape 3.10.2 was used to visualize and analyze the network⁵⁴. To identify core targets, nodes were ranked according to their Degree, Maximal Clique Centrality (MCC) and Betweenness values using the CytoHubba plugin in Cytoscape 3.10.2. A highest score-based node represented putative crucial targets of NPs in the PPI network. The top 10 targets were selected according to the rank value.

Gene Ontology and Kyoto Encyclopedia of Genes and Genomes enrichment analysis

The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway enrichment analysis^55,56 were performed on the potential NP targets in SCA3 using the ‘Functional Annotation’ tool available in Database for Annotation, Visualization, and Integrated Discovery (DAVID) (https://davidbioinformatics.nih.gov/). The study examined the GO activity and KEGG pathway enrichment of the 239 interacting targets that are part of the PPI network. Additionally, target proteins were found in the KEGG pathways, biological process (BP), molecular function (MF), and cellular component (CC). To control for multiple testing, the Benjamini–Hochberg method was applied, and pathways with an adjusted p-value (false discovery rate, FDR) < 0.05 were considered statistically significant⁵⁷. Furthermore, to ensure a thorough analysis, the acquired enrichment results were further methodically processed and visualized using bioinformatics software (http://www.bioinformatics.com.cn/).

Core target screening

The top 10 targets were identified using the ‘CytoHubba’ plugin in Cytoscape 3.10.2. The top three targets with the highest rank values were then chosen for the molecular docking studies.

Molecular docking

Ligand preparation

The selected compounds (Table S1) were converted to structures using ChemDraw software and saved in .sdf format for the molecular docking study.

Protein preparation

Protein preparation involves refining macromolecular structures for compatibility with computational studies⁵⁸. In this work, crystal structures of the target proteins AKT1 (PDB ID: 3CQU; 2.20 Å resolution, R-value 0.207)⁵⁹ and TP53 (PDB ID: 6GGE; 1.25 Å resolution, R-value 0.151)⁶⁰ were retrieved from the Protein Data Bank. AKT1 consists of 342 amino acids (41.02 kDa) with chains A and C, while TP53 has 219 residues (49.86 kDa) with chains A and B. Structures were visualized in Biovia Discovery Studio 4.5, where only Chain A was retained. Heteroatoms, water molecules, and metal ions were removed, and polar hydrogens were added. The cleaned and refined protein structures were then saved in PDB format for subsequent computational analysis.

Prediction of active binding sites

Targeted drug delivery is essential in pharmacotherapy, aiming to modulate disease-related proteins by binding specific ligands to their active sites. This interaction fosters effective protein–ligand binding within a favorable catalytic microenvironment, while improper binding may cause adverse effects or toxicity⁶¹. In this study, we utilized the PyMOL Visualizer Tool 3.1.1 to identify the active site of the target protein, ensuring precise docking and interaction analysis⁶².

Molecular docking process

For molecular docking, we used the PyRx virtual screening application (AutoDock Vina) to examine the binding interactions between the chosen drugs and the target proteins⁶³. Using PyRx’s default setup parameters, the binding energy with the largest negative value (kcal/mol) and root mean square deviation (RMSD) of 0 was chosen for further examination. To facilitate flexible binding, a center-orientated grid box covering the entire macromolecule was applied. For the AKT1 protein (PDB ID: 3CQU), a grid (dimensions: X = 28.1173 Å, Y = 46.3499 Å, Z = 36.9521 Å) was centered at X = 3.2429, Y = − 3.9313, Z = 24.3803. For the TP53 protein (PDB ID: 6GGE), a grid (dimensions: X = 26.3717 Å, Y = 25.2356 Å, Z = 29.9216 Å) was centered at X = 91.5353, Y = 92.6331, Z = − 43.3931. The compound-protein interactions were then visualized and analyzed using DS4.5.

Molecular dynamics simulation

MD simulation is a widely recognized computational technique for exploring the structural stability and dynamic behavior of protein–ligand complexes under physiological-like conditions⁶⁴. In this study, 100-ns MD simulation were conducted for five systems: 3CQU–crocin, 3CQU–reference compound, 6GGE–crocin, 6GGE–reference compound, and their respective apo proteins. All simulations were performed using GROMACS 2023.2 software package⁶⁵. Ligand structures were initially converted to MOL2 format using OpenBabel, with all hydrogen atoms explicitly added⁶⁶. Ligand topology and parameter files were generated using the Merck Molecular Force Field (MMFF) through the SwissParam web server, while the CHARMM36 all-atom force field was employed for protein topology preparation^67–69. Each system was solvated using the TIP3P explicit water model within a triclinic periodic boundary condition box, maintaining a minimum distance of 1.0 nm between any protein atom and the box edge⁶⁷. To neutralize the system’s net charge, appropriate numbers of sodium (Na⁺) and chloride (Cl⁻) ions were added⁷⁰ Energy minimization was then performed using the steepest descent algorithm for 50,000 steps to remove any unfavorable contacts or steric clashes⁷¹. Subsequently, the system underwent a two-phase equilibration process: an initial NVT (constant number of particles, volume, and temperature) equilibration at 300 K using the modified Berendsen thermostat, followed by an NPT (constant number of particles, pressure, and temperature) equilibration at 1 bar using the Parrinello–Rahman barostat. Both equilibration phases lasted 100 ps each⁷². The production MD simulation was then carried out for 100 ns with a 2-fs time step, and snapshots were saved every 10 ps⁷³. Long-range electrostatic interactions were treated using the Particle Mesh Ewald (PME) method, and periodic boundary conditions were applied throughout the simulations⁷⁴. Post-simulation, trajectory files were processed using the trjconv tool to remove periodicity and recenter the protein within the simulation box. To evaluate the structural stability and flexibility of each complex, several key parameters were calculated, including RMSD, root mean square fluctuation (RMSF), radius of gyration (Rg), and the number of hydrogen bond (HB or HBs) formed throughout the trajectory.

Principal component analysis and free energy landscape analysis

The PCA was conducted using GROMACS 2023.2 (gmx covar and gmx anaeig) to examine major conformational motions of the apo and ligand-bound protein complexes. A covariance matrix of Cα atomic fluctuations was generated, producing eigenvectors that reflect collective motions and eigenvalues indicating motion magnitude⁷⁵. This analysis highlights key dynamic regions and structural flexibility associated with ligand binding during molecular dynamics simulation.

The gmx anaeig tool was used to project molecular trajectories onto the first two principal components (PC1 and PC2), capturing essential motions⁶⁷. For accurate analysis, only the equilibrated portion - the final 100 ns corresponding to the RMSD-stable phase was used for both PCA and FEL calculations.

The FEL was generated using PCA data via the gmx sham module, which calculates Gibbs free energy based on principal components⁷⁶. The resulting two-dimensional FEL reveals the most thermodynamically stable conformations, with distinct energy minima representing frequently visited low-energy states. This visualization highlights conformational diversity and transitions between stable states during the simulation.

Dynamic cross-correlation matrix analysis

Dynamic cross-correlation analysis of Cα atoms was performed to assess structural dynamics, revealing how protein regions move relative to each other⁷⁷. This provides insights into the stability, interactions, and collective motions of protein–ligand complexes.

The following formula was used to calculate the cross-correlation matrix (Cij) between residues I and J:

where the ith and jth Cα atoms of two residues are indicated by the letters i and j. The two atoms traveling in the same direction are said to be in correlated motion when the value of Cij is positive and greater than zero. Two atoms moving in opposing direction are said to be in anti-correlated motion when the value of Cij is negative and less than zero. A Cij value of 0 indicates that there is no relationship between the two atoms’ motions.

Furthermore, the values of Cij greater than 0.7 were interpreted as strong correlated (positive) or anti-correlated (negative) motions. These patterns of motion provided insight into how ligand binding influences internal communication and domain movement within the protein. Only the final 100 ns of the trajectories were analyzed to ensure stability and convergence in dynamic behavior. The resulting correlation maps were visualized to highlight regions exhibiting significant dynamic interplay.

Binding free energy analysis

The binding free energy (ΔG_binding) of the protein–ligand complexes was calculated using the MM/PBSA approach, implemented through the g_mmpbsa package⁷⁸. The analysis was performed on the final 10 ns segment of the MD simulation trajectory (90 to 100 ns), corresponding to the equilibrated and stable phase of the system.

A total of 100 snapshots were extracted at regular intervals (one frame every 100 ps) to ensure a statistically meaningful estimate of the binding energies. The dielectric constant was set to 2 for the solute (protein–ligand complex) and 80 for the solvent, corresponding to water. The binding free energy was computed using the following equation:

where ΔG_complex, ΔG_protein, and ΔG_ligand represent the total free energies of the complex, the unbound protein, and the free ligand, respectively. Each energy term includes contributions from molecular mechanics (van der Waals and electrostatic interactions) and solvation energies (polar and non-polar components). Entropy contributions were excluded due to limitations in normal mode accuracy for large systems. MM/PBSA analyses commonly focus on enthalpic components, which primarily drive binding free energy differences.

ADMET properties of lead compound (crocin)

ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties of the selected compounds were predicted using admetSAR 2.0 (http://lmmd.ecust.edu.cn/admetsar2/)⁷⁹, with canonical SMILES from PubChem as input. Key parameters included HIA, Caco-2 permeability, blood-brain barrier (BBB) penetration, CYP450 interactions, AMES mutagenicity, carcinogenicity, hERG inhibition, and oral toxicity. This aided in prioritizing compounds based on pharmacokinetic and safety profiles.

Results

NP-SCA3 common target screening

We retrieved 1252 NP-related targets from the SwissTargetPrediction database (Table S2) and 1538 targets from the SuperPred dataset (Table S3), resulting in 696 unique targets (Table S4). Additionally, we retrieved 5678 SCA3-related targets from the GeneCards database using the search terms ‘spinocerebellar ataxia type 3’ and ‘Machado-Joseph disease’ (Table S5), resulting in 4986 unique targets (Table S6). The intersection of the 696 NP-related targets and the 4986 SCA3-related targets revealed 239 potential targets of NPs against SCA3 (Fig. S1a and Table S7). These 239 targets were further categorized using the Panther Classification System, resulting in 17 distinct groups (Fig. S1b). The classifications included DNA metabolism proteins (2.4%), RNA metabolism proteins (1.2%), calcium-binding proteins (0.4%), cell adhesion molecules (0.4%), chaperones (2.8%), chromatin/chromatin-binding or regulatory proteins (2.0%), cytoskeletal proteins (1.6%), extracellular matrix proteins (0.4%), gene-specific transcriptional regulators (3.6%), intracellular signaling molecules (2.0%), membrane trafficking proteins (0.8%), metabolite interconversion enzymes (18.4%), protein-modifying enzymes (34.4%), protein-binding activity modulators (2.8%), translational proteins (0.8%), transmembrane signal receptors (5.2%), and transporters (8%).

Protein–protein interaction network construction

The STRING database was used to build a PPI network of the targets in SCA3, with an emphasis on the 239 overlapping targets. After excluding isolated targets, the PPI network was generated using an interaction score threshold between nodes automatically set to > 0.518 by the software (Fig. 2a and Table S8). This network had 238 nodes and 2944 edges, with an average node of 24.7. Cytoscape 3.10.2 was used to further visualize the PPI network (Fig. 2b). Nodes with values higher than the average (24.7) for Degree, MCC and Betweenness were chosen as core targets according to their topological characteristics.

Fig. 2.

The PPI plot of key targets of NPs in SCA3. (a) PPI plot using STRING and (b) PPI plot using Cytoscape 3.10.2 software. NP: natural product, PPI: protein–protein interaction, SCA3: spinocerebellar ataxia type 3, STRING: Search Tool for the Retrieval of Interacting Genes/Proteins.

Gene Ontology and Kyoto Encyclopedia of Genes and Genomes enrichment analysis

For the GO and KEGG pathway enrichment analysis, the 239 potential NP targets in SCA3 were first imported into the DAVID database. The GO enrichment results yielded 971 terms across three categories: BP (615 terms; Table S9), CC (124 terms; Table S10), and MF (232 terms; Table S11). The top 10 terms from each category were selected for further investigation (Fig. 3a). The related BPs included protein phosphorylation, chromatin remodeling, response to xenobiotic stimuli, peptidyl-serine phosphorylation, negative regulation of apoptosis, intracellular signal transduction, positive regulation of transcription by RNA polymerase II, cellular response to cocaine, neuronal apoptosis, and peptidyl-threonine phosphorylation. The BP terms identified exhibited count values ranging from 7 to 57, with chromatin remodeling showing the highest frequency, appearing in 57 instances. Additionally, fold enrichment values for these BP terms ranged from 2.86 to 56.86. Among them, the cellular response to cocaine demonstrated the highest fold enrichment value at 56.86, indicating a particularly strong overrepresentation in the analyzed dataset (Table S9). The related CCs included cytosol, cytoplasm, nucleoplasm, protein-containing complexes, membrane rafts, extracellular exosomes, nucleus, ficolin-1-rich granule lumen, perinuclear region of cytoplasm, and axon. The CC terms identified exhibited count values ranging from 23 to 154, with cytosol showing the highest frequency, appearing in 154 instances. Additionally, fold enrichment values for these CC terms ranged from 2.08 to 12.45. Among them, the ficolin-1-rich granule lumen demonstrated the highest fold enrichment value at 12.45, indicating a particularly strong overrepresentation in the analyzed dataset (Table S10). The related MF terms included protein serine/threonine kinase activity, protein kinase activity, ATP binding, protein binding, enzyme binding, histone H3T6 kinase activity, histone H3T11 kinase activity, AMP-activated protein kinase activity, and histone H3S28 kinase activity. The MF terms identified exhibited count values ranging from 26 to 224, with protein binding showing the highest frequency, appearing in 224 instances. Additionally, fold enrichment values for these CC terms ranged from 1.31 to 10.34. Among them, the protein serine kinase activity demonstrated the highest fold enrichment value at 10.34, indicating a particularly strong overrepresentation in the analyzed dataset (Table S11).

Fig. 3.

Bubble plot of enrichment analysis (a) GO enrichment analysis and (b) KEGG pathway enrichment analysis of NP targets in SCA3. GO: Gene Ontology, KEGG: Kyoto Encyclopedia of Genes and Genomes, NP: natural product, SCA3: spinocerebellar ataxia type 3.

The KEGG pathway enrichment analysis identified 173 signaling pathways (Table S12). The top 20 pathways were selected for further investigation (Fig. 3b). Notable pathways included lipid and atherosclerosis, thyroid hormone signaling, prostate cancer, MAPK signaling, PD-L1 expression and PD-1 checkpoint pathways in cancer, Kaposi sarcoma-associated herpesvirus infection, Alzheimer’s disease (AD), neurotrophin signaling, T cell receptor signaling, HIF-1 signaling, Yersinia infection, proteoglycans in cancer, hepatitis B, pancreatic cancer, cancer pathways, C-type lectin receptor signaling, measles, insulin resistance, human cytomegalovirus infection, and neurodegeneration-related signaling pathways (Fig. 4). Furthermore, the KEGG pathways identified exhibited count values ranging from 19 to 35, with hsa05010:AD showing the highest frequency, appearing in 35 instances including APP, GSK3B, PSEN2, PSEN1, HSD17B10, TNF, RELA, IKBKB, NCSTN, GRM5, CASP8, PSMB2, TUBB3, PSMB1, AKT1, MAPK1, IKBKG, NAE1, CAPN1, SNCA, MAP2K2, MME, NOS2, CSNK2A1, CHUK, ADAM10, CSNK1E, MTOR, BACE1, ERN1, CDK5, IL1B, MAPT, CALM1 and GAPDH. Alzheimer’s disease (hsa05010) is a chronic neurodegenerative disorder that progressively destroys neurons, leading to severe cognitive impairment and dementia⁸⁰. The APOE ε4 allele, a major genetic risk factor for AD, modifies SCA3 progression by enhancing language/visual memory performance while worsening speech disturbances⁸¹. Both disorders involve protein clearance dysfunction: AD is associated with amyloid-beta and tau accumulation, while SCA3 is associated with mutant ataxin-3 aggregates that disrupt autophagy by destabilizing Beclin-1^1,82. Neuroimaging reveals perivascular space enlargement in SCA3 patients, mirroring vascular pathology observed in AD and correlating with motor impairment⁸³. Additionally, fold enrichment values for these KEGG terms ranged from 3.67 to 8.79. Among them, the prostate cancer demonstrated the highest fold enrichment value at 8.79, indicating a particularly strong overrepresentation in the analyzed dataset (Table S12). On the other hand, among the 33 hub genes identified as potentially important in the hsa04010: MAPK signaling pathway of NP against SCA3, we identified CACNA1B, TNF, CACNA1H, RELA, EGFR, IKBKB, RPS6KA3, IRAK1, ERBB2, RPS6KA1, AKT1, MAPK1, IKBKG, PRKACA, MAP4K4, PRKCG, MAP4K2, HSPA8, JUN, MAP2K2, DUSP3, CHUK, PRKCA, MAPK14, VEGFA, PPM1A, TAOK1, IL1B, MAPKAPK2, MAPT, TP53, MAP3K14, and HSPA1A (Table S12). The hsa04010: MAPK signaling pathway, among other significantly enriched pathways, was selected for further investigation due to its high enrichment significance, as evidenced by the lowest adjusted p-value (9.87 × 10⁻¹³), and its well-established role in neurodegenerative processes, including the pathogenesis SCA3. This pathway also exhibited a notable fold enrichment value of 8.20, further supporting its biological relevance in the context of SCA3.

Fig. 4.

The hsa04010: MAPK signaling pathway (red marks represent potential targets of NPs in SCA3). DNA: deoxyribonucleic acid, NP: natural product. Permission has been obtained from Kanehisa laboratories for using KEGG pathway database (https://www.kegg.jp/kegg/kegg1.html).

Core target screening

Table 2 and Table S13 show TP53 and AKT1 were the top two genes exhibiting the highest-ranking values of Degree, MCC and Betweenness methods. Additionally, the top 10 core targets identified were TP53, AKT1, GAPDH, TNF, HSP90AA1, JUN, ESR1, EGFR, SRC, and STAT3 for the Degree method (Table 2 and Fig. S2a); TP53, JUN, ESR1, GAPDH, STAT3, HIF1A, TNF, HSP90AA1, AKT1 and EP300 were identified using MCC method (Table 2 and Fig. S2b); and AKT1, TP53, GAPDH, APP, TNF, SRC, EGFR, HSP90AA1, CALM1 and GSK3B were identified using the Betweenness method (Table 2 and Fig. S2c). The interactions between these key targets suggest that the efficacy of NPs in treating SCA3 may depend on their molecular connections. To identify potential inhibitors against SCA3, we performed molecular docking and MD analysis, focusing on TP53 and AKT1 based on their ranking.

Table 2.

Rank values of the top 10 targets in Degree, MCC and Betweenness methods.

Name	Degree score	MCC Score	Betweenness score
TP53	126	2.9115018445521553E24	5284.970117
AKT1	125	2.9115012221102364E24	5869.472982
GAPDH	118	2.9115018442897264E24	5128.526684
TNF	98	2.9115018415225144E24	2164.30543
HSP90AA1	93	2.911501356659616E24	1552.673526
JUN	92	2.9115018444658737E24	–
ESR1	89	2.911501844353454E24	–
EGFR	86	–	1806.359654
SRC	83	–	1969.806131
STAT3	82	2.9115018438380574E24	–
HIF1A	–	2.9115018435094946E24	–
EP300	–	2.911427786163159E24	–
APP	–	–	2722.792853
CALM1	–	–	1503.591695
GSK3B	–	–	1374.195147

AKT1: AKT serine/threonine kinase 1, APP: amyloid beta precursor protein, CALM1: calmodulin 1, EGFR: epidermal growth factor receptor, EP300: EP300 lysine acetyltransferase, ESR1: estrogen receptor 1, GAPDH: glyceraldehyde-3-phosphate dehydrogenase, GSK3B: glycogen synthase kinase 3 beta, HIF1A: hypoxia inducible factor 1 subunit alpha, HSP90AA1: heat shock protein 90 alpha family class A member 1, JUN: Jun proto-oncogene, AP-1 transcription factor subunit, SRC: SRC proto-oncogene, non-receptor tyrosine kinase, STAT3: signal transducer and activator of transcription 3, TNF: tumor necrosis factor, TP53: tumor protein p53.

Molecular docking

Molecular docking was used to determine the optimal intermolecular interactions between the target proteins (AKT1 and TP53) and 15 selected compounds. The identification of active site residues for AKT1 and TP53 was guided by computational methods. Initially, key binding and functional residues were collected from PDB structural data and relevant literature. These residues were further validated using PyMOL, which predicts accessible surface pockets and energetically favorable ligand-binding regions through machine learning-based pocket prediction. Molecular docking of 15 natural compounds with the target proteins was performed using PyRx to evaluate binding affinities and interaction profiles. Compound–protein interactions were visualized and analyzed using DS4.5. As PyMOL provides reliable active site predictions without the need for co-crystallized ligands, additional redocking-based protocol validation was considered unnecessary. This widely accepted approach emphasizes potential ligand–protein interactions without reproducing the reference ligand pose, therefore providing sufficient confidence in the docking results and the biological relevance of the predicted interactions.

For the AKT1 protein, the active site residues included Leu156, Gly157, Phe161, Val164, Ala177, Lys179, Thr211, Glu228, Tyr229, Glu234, Arg273, Glu278, Asn279, Met281, Thr291, Asp292, Lys297, Met307, Lys308, Phe309, Cys310, Tyr326, and Phe438. The docking results showed the binding affinities of the test compounds ranged from − 6.7 to − 9.5 kcal/mol. Among them, crocin emerged as the lead compound, as it had the highest binding affinity (− 9.5 kcal/mol) and the most interactions with the active site residues, including four HBs with Glu191, Glu198, Glu228, and Thr291, and two carbon-HBs with Gly157 and Glu234. Additionally, crocin showed other interactions including pi-alkyl interactions with Phe161, which contributed to its strong binding stability (Fig. S3a). Compared to the control compound troriluzole (− 8.4 kcal/mol), licochalcone A (− 9.2 kcal/mol) and T-11 (− 9.3 kcal/mol) also showed strong interactions, whereas puerarin exhibited the lowest binding affinity (− 6.7 kcal/mol). Notably, the control compound troriluzole exhibited three HBs with Ala230, Lys276, and Asp292, and a carbon-HB with Lys158. Additional interactions included a pi-sigma bond with Met281, pi-pi stacking bond with Phe161, and several alkyl/pi-alkyl interactions with Leu156, Val164, and Ala177 (Fig. S3b).

For the TP53 protein, the active site residues included Leu145, Trp146, Val147, Thr150, Pro151, Pro152, Pro153, Gly154, Thr155, Cys220, Glu221, Pro222, Pro223, Ser227, Asp228, Cys229, and Thr230. The docking results showed the binding affinities of the test compounds ranged from − 5.2 to − 7.9 kcal/mol. Licochalcone A had the strongest binding affinity (− 7.9 kcal/mol), forming several HBs with Pro219, Ser227, and His233, as well as additional stabilizing interactions such as pi-donor HBs and pi-alkyl interactions. Crocin also exhibited a strong binding affinity (− 6.6 kcal/mol), forming strong HBs with Gln144, Leu201, Glu224, Ser227, and Cys229, as well as carbon-HBs with Glu198, Asn200, Pro223, Thr231, and His233, which stabilized its interactions within the active site (Fig. S4a). In comparison, the control compound troriluzole had a binding affinity of − 6.0 kcal/mol, forming three HBs with Gln144, Ser227, and Cys229, as well as a carbon-HB with Thr231 and a pi-cation bond with Thr230 (Fig. S4b).

The findings show that the creation of HBs with residues in the active site was the main mechanism stabilizing the compound-protein complexes, which were further supported by other stabilizing factors, such as pi-sigma, alkyl/pi-alkyl, and pi-pi stacking interactions^57,84. A thorough overview of each compound’s docking scores and interactions with the AKT1 and TP53 proteins is given in Table S14 and Table S15, respectively. According to our results, the lead molecule crocin had the highest binding affinities for both AKT1 and TP53, which makes it an ideal candidate for further research.

Molecular dynamics simulation

Root mean square deviation analysis

In MD simulation, structural aberrations and stability can be assessed by RMSD analysis, with higher RMSD values indicating conformational oscillations, and lower RMSD values indicating increased structural stability⁸⁵. In this study, the structural dynamics and conformational variations of the protein backbones and protein–ligand complexes were evaluated using RMSD analysis across the complete 100-ns MD simulation, with no limitation to only the stable regions (Fig. 5).

Fig. 5.

Molecular dynamics simulation pattern of lead compound (crocin) and control compound (troriluzole) with the core target proteins (AKT1 and TP53). (a) RMSD values of apo protein (black), crocin (green), and control compound (troriluzole) complexes with AKT1, and (b) RMSD values of apo protein (black), crocin (green), and control compound (troriluzole) complexes with TP53. (c) RMSF values of apo protein (black), crocin (green), and control compound (troriluzole) complexes with AKT1, and (d) RMSF values of apo protein (black), crocin (green), and control compound (troriluzole) complexes with TP53. (e) Rg values of apo protein (black), crocin (green), and control compound (troriluzole) complexes with AKT1, and (f) Rg values of apo protein (black), crocin (green), and control compound (troriluzole) complexes with TP53. ns: nanosecond, ps: picosecond, Rg: radius of gyration, RMSD: root mean square deviation, RMSF: root mean square fluctuation.

The average RMSD values for 3CQU-crocin, 3CQU-troriluzole, and the apo protein were 0.42, 0.38, and 0.21 nm, respectively (Fig. 5a). The 3CQU-crocin complex showed an initial RMSD increase up to 20 ns, followed by a stable conformational pattern for the remainder of the simulation. In contrast, the 3CQU-troriluzole complex showed an RMSD increase up to 10 ns, followed by a distinct trajectory up to 52 ns, and then rising sharply to 0.83 nm and then decreasing around 75 ns, before finally stabilizing. The apo protein maintained a stable RMSD profile throughout the simulation period.

The average RMSD values for 6GGE-crocin, 6GGE-troriluzole, and the apo protein were 0.98, 2.0, and 0.22 nm, respectively (Fig. 5b). The 6GGE-crocin complex showed an RMSD increase up to 62 ns, followed by a distinct stable pattern. In contrast, the 6GGE-troriluzole complex showed an RMSD increase up to 50 ns, followed by a sharp decline to 60 ns, before increasing again and then stabilizing into a unique trajectory for the remainder of the simulation. The apo protein maintained a stable RMSD profile throughout the simulation period.

Root mean square fluctuation analysis

In this study, the structural dynamics and conformational variations of the protein backbones and protein–ligand complexes were evaluated using RMSF analysis across the complete 100-ns MD simulation, with no limitation to only the stable regions (Fig. 5). The conformational stability of certain residues is essential for protein–ligand interactions. In MD simulation, the flexibility of protein residues can be evaluated by RMSF, a critical measure. Lower RMSF values indicate more stiff and stable areas, whereas higher RMSF values imply more residue flexibility, which increases the chance of interacting with ligand molecules⁸⁶. According to Bornot et al.⁸⁷, residues with large RMSF variations are frequently implicated in ligand binding, whereas those with smaller RMSF variations have a lower interaction potential. Here, RMSF values were computed to examine the effect of ligands on protein backbone flexibility (Fig. 5).

In the 3CQU system, the average RMSF values for 3CQU-crocin, 3CQU-troriluzole, and the apo protein were 0.17 nm, 0.27 nm, and 0.13 nm, respectively. The maximum fluctuation in RMSF for the 3CQU-crocin complex has occurred at the residues of Gly1, Arg2, Arg206, Lys301, Asp302 and Ala303 (Fig. 5c), whereas 3CQU-troriluzole complex has occurred at the residues of Gly1, Leu210, Thr211 and Lys214 (Fig. 5c). Furthermore, the maximum RMSF fluctuations in the apo form have occurred at the residues of Arg2 and Val167 (Fig. 5c). Among the residues, Gly1 residue fluctuates with the maximum RMSF value for both the complexes and Arg2 fluctuates with the maximum RMSF value for the apo form (Fig. 5c). These findings indicate that 3CQU-crocin complex with lower RMSF values had greater structural stability than the 3CQU-troriluzole complex, which had reduced fluctuations and greater structural rigidity.

Similarly, in the 6GGE system, the average RMSF values for 6GGE-crocin, 6GGE-troriluzole, and the apo protein were 0.53 nm, 1.33 nm, and 0.27 nm, respectively. The average RMSF values are almost similar for 6GGE-crocin and 6GGE-troriluzole complexes in comparison with the apo form except for few fluctuations. The maximum fluctuation in RMSF for the 6GGE-crocin complex has occurred at the residues of Ser96, Val97, Arg209, Asn210 and Lys292 (Fig. 5d), whereas 6GGE-troriluzole complex has occurred at the residues of Ser96, Arg209, Asn210 and Lys292 (Fig. 5d). Furthermore, the maximum RMSF fluctuations in the apo form have occurred at the residues of Ser96, Arg209 and Lys292 (Fig. 5d). Among the residues, Ser96, Arg209 and Lys292 residues fluctuate with the maximum RMSF value for both the complexes as well as apo form (Fig. 5d). These findings indicate that 6GGE-crocin complex with lower RMSF values had greater structural stability than the 6GGE-troriluzole complex, which had increased flexibility.

Radius of gyration analysis

The Rg represents the spatial distribution of the atomic positions relative to the molecular axis and is a critical parameter for assessing the structural compactness and stability of protein–ligand complexes. Monitoring Rg during MD simulation can provide insights on conformational changes and allow the evaluation of the overall structural integrity and flexibility of macromolecular systems. The Rg profiles of the 3CQU-crocin complex, the 3CQU-troriluzole complex, and the apo protein during the 100-ns MD simulation are shown in Fig. 5e. The average Rg values for 3CQU-crocin, 3CQU-troriluzole, and apo protein were 2.06 nm, 2.04 nm, and 2.01 nm, respectively. Notably, the 3CQU-troriluzole complex exhibited the lowest Rg value, indicating a more compact structure compared to the 3CQU-crocin complex. The reduced compactness suggests that there is a possibility of conformational rearrangement, which can influence protein–ligand interactions. The Rg profiles of the 6GGE-crocin complex, the 6GGE-troriluzole complex, and the apo protein are shown in Fig. 5f. The average Rg values for 6GGE-crocin, 6GGE-troriluzole, and apo protein were 1.70 nm, 1.73 nm, and 1.68 nm, respectively. These findings indicate that 6GGE-crocin complex had a more condensed structure than the 6GGE-troriluzole complex, suggesting a greater degree of structural stability.

Hydrogen bond analysis

Protein–ligand interactions are stabilized by hydrogen bonding, which affects complex stability and binding affinity. In MD simulation, the quantity and durability of HB between a ligand and its target protein are important markers of complex stability. The calculated HBs and graphical depictions of HB interactions during a 100-ns simulation period are shown in Fig. 6. In the 3CQU system, the 3CQU-crocin complex was observed to have 15 HBs (Fig. 6a), whereas the 3CQU-troriluzole complex had only six HBs (Fig. 6b). These findings indicate that crocin establishes a more extensive HB network than troriluzole, contributing to its enhanced stability. Similarly, in the 6GGE system, the 6GGE-crocin complex was observed to have 12 HBs (Fig. 6c), whereas the 6GGE-troriluzole complex had only five HBs (Fig. 6d).

Fig. 6.

The HB of (a) crocin-AKT1 complex, (b) control compound (troriluzole)-AKT1 complex, (c) crocin-TP53 complex, and (d) control compound (troriluzole)-TP53 complex during the 100-ns MD simulation. HB: hydrogen bond, MD: molecular dynamics, ns: nanosecond.

A greater frequency of HBs and prolonged duration are typically associated with improved ligand binding strength. The higher number of HBs in crocin-bound complexes suggests that it has stronger and more stable interactions with the respective target proteins. Moreover, analyzing HB occupancy can help pinpoint key residues involved in hydrogen bonding, providing deeper insights into ligand recognition and molecular interaction mechanisms. These findings highlight the potential of crocin as a robust binding partner and reinforces its therapeutic significance in targeting 3CQU and 6GGE proteins.

Principal component analysis

The PCA is a widely used nonparametric method for reducing dimensionality. It can be used to facilitate the interpretation of high-dimensional MD simulation data by transforming it into a lower-dimensional space⁸⁸. Such an approach can allow the characterization of complex collective motions that influence protein stability and function. By adjusting key parameters, PCA can simplify the molecular motion to provide insights into essential biological processes, for example, protein folding and the opening and closing of ion channels^89,90.

The PCA results for the 3CQU complexes are illustrated in Fig. 7a and b, which revealed the top 20 principal components (PCs) accounted for 90.9% and 73.9% of the total variance in 3CQU-crocin and 3CQU-troriluzole complexes, respectively. This suggests that 3CQU-troriluzole complex exhibited a more constrained phase space and lower conformational flexibility compared to the 3CQU-crocin complex. Among the principal components, PC1 contributed the highest variance at 37.8% and 28.8%, PC2 contributed 20.91% and 14.12%, and PC3 contributed the lowest variance at only 8.9% and 7.77%, in the 3CQU-crocin and 3CQU-troriluzole complexes, respectively. The relatively lower variance of PC3 in the 3CQU-troriluzole complex suggests a stable binding interaction and compact structural arrangement.

Fig. 7.

The PCA of (a) crocin-AKT1 complex, (b) control compound (troriluzole)-AKT1 complex, (c) crocin-TP53 complex, and (d) control compound (troriluzole)-TP53 complex during the 100-ns MD simulation. ns: nanosecond, PC: principal component, PCA: principal component analysis.

Similarly, the PCA results for the 6GGE complexes are illustrated in Fig. 7c and d, which revealed the top 20 PCs accounted for 90.9% and 82.1% of the total variance in 6GGE-crocin and 6GGE-troriluzole complexes, respectively. This suggests that 6GGE-troriluzole complex exhibited a more restricted phase space and reduced conformational flexibility compared to the 6GGE-crocin complex. Among the principal components, PC1 contributed the highest variance at 37.8% and 28.08%, PC2 contributed 20.91% and 13.46%, and PC3 contributed the lowest variance at only 8.9% and 6.94%, in the 6GGE-crocin and 6GGE-troriluzole complexes, respectively. The low variance observed in PC3 in the 6GGE-troriluzole complex further supports the notion of a stable binding interaction and compact structural arrangement.

Free energy landscape analysis

The FEL provides valuable insights into protein folding mechanisms by showing how proteins transition between different conformational states when returning to their native forms. This approach, facilitated by MD simulation, can allow the analysis of protein and protein–ligand complex stability within a solvent environment⁹¹. Here, we reconstructed the conformational landscape and identified the energy minima to analyze the stability of 3CQU-crocin, 3CQU-troriluzole, 6GGE-crocin, and 6GGE-troriluzole complexes based on their principal components.

To estimate the Gibbs FEL, the first two principal components (PC1 and PC2) were analyzed as in Fig. 8. The energy distribution for each system is shown using a color-coded contour map, with lower energy regions represented by darker blue shades. Notably, the 3CQU-crocin (Fig. 8a) and 6GGE-crocin (Fig. 8c) complexes predominantly exhibited a single energy minimum, as indicated by the uniform blue regions, which suggests a stable conformational state throughout the simulation. In contrast, the 3CQU-troriluzole (Fig. 8b) and 6GGE-troriluzole (Fig. 8d) complexes displayed multiple conformational states, characterized by the presence of multiple energy minima, which suggests a greater degree of structural variability and dynamic transitions between the different conformations.

Fig. 8.

The FEL of (a) crocin-AKT1 complex, (b) control compound (troriluzole)-AKT1 complex, (c) crocin-TP53 complex, and (d) control compound (troriluzole)-TP53 complex during the 100-ns MD simulation. FEL: free energy landscape, ns: nanosecond, PC: principal component.

Dynamic cross-correlation matrix analysis

In the DCCM study, all Cα atoms were analyzed to understand the conformational dynamics of the AKT1 and TP53 target proteins⁹². The correlated motion of the residues during the whole MD simulation are depicted in the two-dimensional DCCM diagrams (Fig. 9). The colors represent the varying degrees of correlation, with deeper hues signifying greater correlation and lighter shades suggesting weaker correlation. The correlation values vary from − 1.0 to 1.0, with residues having a negative correlation (− 1 to 0) when traveling in opposing directions and a positive correlation (0 to 1) when traveling in the same direction.

Fig. 9.

The DCCM of (a) crocin-AKT1 complex, (b) control compound (troriluzole)-AKT1 complex, (c) crocin-TP53 complex, and (d) control compound (troriluzole)-TP53 complex during the 100-ns MD simulation. DCCM: dynamic cross-correlation matrix, MD: molecular dynamics, ns: nanosecond.

The comparative analysis of the DCCM diagrams between the 3CQU-crocin complex (Fig. 9a) and 3CQU-troriluzole complex (Fig. 9b) revealed notable differences in their correlation patterns. The 3CQU-crocin complex showed substantially more positively correlated motions compared to the 3CQU-troriluzole complex, as highlighted by the black dashed boxes. Similarly, comparing between the 6GGE-crocin complex (Fig. 9c) and 6GGE-troriluzole complex (Fig. 9d), the 6GGE-crocin system had a higher degree of positively correlated motions.

Binding free energy analysis

The binding free energy (ΔG_bind) can be used to determine the stability and affinity of protein–ligand complexes. This method takes into consideration a variety of contact forces, such as van der Waals and electrostatic interactions. Stronger ligand–protein binding affinity and improved stability are indicated by lower ΔG_bind values.

The binding free energy of the 3CQU-crocin, 3CQU-troriluzole, 6GGE-crocin, and 6GGE-troriluzole complexes were calculated using the MM/PBSA technique. To ensure accurate predictions, the binding free energy (ΔG_bind) computations were carried out during the last 10 ns of the simulation trajectory.

The computed binding free energy values for each complex are presented in Table 3 and Fig. S5. The MM/PBSA analysis revealed that the 3CQU-crocin (Fig. S5a) and 6GGE-crocin (Fig. S5c) complexes exhibited lower binding free energy compared to the 3CQU-troriluzole (Fig. S5b) and 6GGE-troriluzole (Fig. S5d) complexes, indicating greater stability and stronger interactions in the crocin-bound systems. Furthermore, the consistency between the MM/PBSA-derived binding free energy values and the molecular docking results validate the accuracy of the computational approach (Table 3). These findings highlight the superior binding affinity and stability of the tested crocin complexes compared to the troriluzole complexes, showing their potential in future drug discovery and molecular interaction studies.

Table 3.

Binding free energy of crocin and troriluzole with AKT1 (PDB ID: 3CQU) as well as crocin and troriluzole with TP53 (PDB ID: 6GGE).

Name	∆Evdw (kcal/mol)	∆Eelec (kcal/mol)	ΔEGB (kcal/mol)	∆EDISP (kcal/mol)	∆Egas (kcal/mol)	ΔEsolv (kcal/mol)	ΔG binding free energy (kcal/mol)
3CQU-crocin	− 65.1	− 156.91	149.44	00	− 222.01	138.09	− 83.91
3CQU-troriluzole	− 42.74	− 29.06	53.89	00	− 71.79	47.78	− 24.01
6GGE-crocin	− 38.09	− 30.85	49.81	00	− 68.93	44.78	− 24.16
6GGE-troriluzole	− 38.41	− 5.78	29.47	00	− 44.19	25.88	− 18.31

∆E_DISP: dispersion energy, ∆E_elec: electrostatic energy, ΔG: binding free energy, ∆E_gas: gas-phase energy, ΔE_GB: polar solvation energy in Generalized-Born method, ΔE_solv: solvation energy, ∆E_vdw: van der Waals energy.

Furthermore, we investigated the ligand binding positions across a 100-ns MD trajectory, capturing snapshots at 20 ns intervals (Fig. S6–S9). In the 3CQU-crocin complex, crocin consistently remained stably bound within the active site, exhibiting only minor conformational shifts throughout the simulation. At 0 ns, it formed strong HBs with Glu419 and Phe418, and by 20 ns, repositioned slightly to interact with Ala429 and Lys431. At 40 ns, it maintained hydrogen bonding with Gly417 and Thr418 and π–π stacking with Phe418. Between 60 and 80 ns, crocin embedded deeper in the pocket, forming stable interactions with Leu413 and Ala429. By 100 ns, crocin preserved robust binding, confirming a stable and adaptive interaction profile (Fig. S6). In comparison, the 3CQU-troriluzole complex displayed more variable binding behavior. Initially, it engaged Phe429, Glu431, and Met438 (0 ns), then shifted toward Asp431, Met438, and Ala434 by 20 ns. As the simulation progressed, new interactions emerged with Asn438, Val421, and Lys433 (40 ns), and later with Phe418, Glu448, and Thr452 (60 to 80 ns). At 100 ns, troriluzole formed interactions with Glu448, Met438, and Gly414, reflecting stable yet less consistent binding compared to crocin (Fig. S7).

Similarly, in the 6GGE-crocin complex, crocin exhibited stable and flexible binding throughout the trajectory. It initially interacted with Glu419, Gly417, and Asp421 (0 ns), then maintained contacts with Ala418 and Gly417 (20 ns). By 40 ns, crocin formed deeper HBs with Pro419 and Thr418. From 60 to 100 ns, it sustained key interactions with Gln423, Gly429, and Glu419, demonstrating a well-conserved and adaptable binding mode (Fig. S8). In contrast, the 6GGE-troriluzole complex showed more fluctuation. At 0 ns, it engaged GLU A:221, GLU A:224, and THR A:231. These interactions weakened by 20 to 40 ns. By 60 ns, new interactions appeared with THR A:150 and various PRO residues. At 80 ns, it adopted a broader contact profile with residues like CYS A:120 and ARG A:156, which remained consistent at 100 ns, suggesting late-stage stabilization (Fig. S9). Overall, crocin demonstrated more consistent, stable, and adaptive binding across both 3CQU and 6GGE complexes compared to troriluzole, highlighting its superior binding performance throughout the MD simulations.

ADMET properties of lead compound (crocin)

The ADMET profiles of crocin and troriluzole were evaluated using the admetSAR platform to assess their drug-likeness, safety, and pharmacokinetic behavior (Table S16). Crocin exhibited moderate aqueous solubility (LogS = –1.76) and poor Caco-2 permeability, intestinal absorption (HIA–), and BBB– permeability, indicating limited oral bioavailability and CNS accessibility. It is localized to the mitochondria and showed non-substrate behavior for major CYP450 isoforms (2C9, 2D6, 3A4), along with no inhibitory effects against CYP1A2, 2C9, 2D6, 2C19, and 3A4, suggesting a low risk of metabolic interactions. Crocin also demonstrated low CYP inhibitory promiscuity and was predicted as a non-AMES toxic, non-carcinogenic, and non-inhibitor of the hERG gene, with an acute oral toxicity class III, indicating relatively safe toxicity parameters.

In contrast, troriluzole showed lower aqueous solubility (LogS = –3.43), but better predicted intestinal absorption (HIA+) and BBB+ permeability, suggesting promising CNS activity. It is localized to the lysosome and was identified as a substrate for CYP3A4 and an inhibitor of CYP2C9 and CYP2C19, implying possible risks of drug–drug interaction. While it exhibited similar safety parameters to crocin (non-AMES toxic, non-carcinogenic, weak hERG inhibition), its overall CYP inhibitory promiscuity was lower, though it showed some inhibitory behavior, which may necessitate further metabolic profiling. Both compounds were classified under acute oral toxicity class III.

Discussion

The SCA3 is the most prevalent form of autosomal dominant spinocerebellar ataxia worldwide¹⁶. It is a progressive neurodegenerative disorder characterized by cerebellar ataxia, pyramidal signs, dystonia, and peripheral neuropathy, with symptoms typically appearing in mid-adulthood¹. The misfolding and aggregation of mutant ataxin-3 protein contribute to the neuronal dysfunction and degeneration, primarily affecting the cerebellum, brainstem, and basal ganglia^93,94. Currently, no disease-modifying therapies exist for SCA3, with treatments limited to managing motor and non-motor symptoms through physiotherapy, pharmacological approaches, and speech therapy⁹⁵.

This study used integrated in silico approaches to explore NPs as potential drug candidates against SCA3. Several NPs have been shown to be promising inhibitors of SCA3, with potential therapeutic benefits against the toxic effects of mutant ataxin-3 protein. Specific phytochemicals, such as flavonoids, diterpenoids, and alkaloids, have also been extensively investigated for their ability to interact with ataxin-3 protein^16,23.

The GO enrichment analysis revealed that the tested NPs primarily function through protein phosphorylation, chromatin remodeling, response to xenobiotic stimuli, peptidyl-serine phosphorylation, negative regulation of apoptosis, and various cellular compartments, including the cytosol, cytoplasm, nucleoplasm, protein-containing complexes, membrane rafts, and extracellular exosomes (Fig. 3a). Additionally, other functional activities were identified, including protein serine/threonine kinase activity, protein kinase activity, ATP binding, protein binding, and enzyme binding.

The KEGG pathway enrichment analysis indicated the enrichment of multiple disease pathways, many of which were not directly related to SCA3, but likely to share molecular targets across different disorders (Fig. 3b). Therefore, we focused on pathways that are highly relevant to SCA3, which identified the MAPK signaling pathway as a crucial mechanism of the therapeutic effects of NPs. Given its role in neuronal survival and apoptosis, the MAPK pathway has become a key area of interest in SCA3 drug discovery. Expanded ataxin-3 in SCA3 perturbs cellular signaling by hyperactivating the ERK branch of the MAPK cascade while inhibiting pro-apoptotic pathways, producing an imbalanced survival-death response that accelerates neurodegeneration⁹⁶. While MAPK-focused studies in SCA3 are limited, evidence from SCA1 indicates that downregulating the RAS–MAPK–MSK1 axis reduces mutant ataxin-1 accumulation and alleviates neurotoxicity in fly and mouse models, supporting MAPK modulation as a therapeutic avenue in polyglutamine ataxias⁹⁷. Given MAPKs’ central role in apoptosis, transcriptional regulation, and stress adaptation, rebalancing dysregulated MAPK signaling in SCA3 could enhance mutant protein clearance and serve as a mechanistic basis for disease-modifying drug discovery. Polyglutamine expansion in ataxin-3 contributes to neurodegeneration, yet the precise molecular mechanisms remain unclear. Studies using models of mutant ataxin-3 showed significant changes in the phosphorylation patterns of the PI3K/AKT/GSK3/mTOR pathway, leading to altered ataxin-3 signaling, dysregulated activation of the pro-survival ERK pathway, and diminished pro-apoptotic responses, all of which may contribute to disease progression⁹⁶. Beyond protein homeostasis, ATXN3 also plays a role in transcriptional regulation of disease-associated pathways. One key finding is that it suppresses Efna3 expression within the Ephrin signaling pathway, which is vital for neuronal communication and synaptic plasticity⁹⁸. Dysregulation of these pathways may further exacerbate neurodegeneration in SCA3. Identifying genetic modulators of ATXN3 expression is a priority in developing targeted therapeutics. Studies have already identified regulatory genes that influence ataxin-3 levels⁹⁹, which represent new avenues for interventions that mitigate SCA3 pathology.

Intersection analysis of the NP targets and SCA3-related genes revealed 33 potential therapeutic targets within the MAPK signaling pathway, including CACNA1B, TNF, CACNA1H, RELA, EGFR, IKBKB, RPS6KA3, IRAK1, ERBB2, RPS6KA1, AKT1, MAPK1, IKBKG, PRKACA, MAP4K4, PRKCG, MAP4K2, HSPA8, JUN, MAP2K2, DUSP3, CHUK, PRKCA, MAPK14, VEGFA, PPM1A, TAOK1, IL1B, MAPKAPK2, MAPT, TP53, MAP3K14, and HSPA1A (Table S12). Further analysis identified core targets, including TP53, AKT1, GAPDH, TNF, HSP90AA1, JUN, ESR1, EGFR, SRC, and STAT3 (Table 2, Fig. S2). Based on ranking, AKT1 and TP53 were selected for in-depth study. Dysregulation of MAPK signaling is mechanistically linked to SCA3 progression through abnormal phosphorylation and altered activity of proteins governing neuronal survival, apoptosis, and transcriptional regulation. Evidence from in vitro and in vivo models of SCA3 highlights significant disturbances in this pathway, particularly the aberrant stimulation of ERK, a member of the MAPK family. Although ERK activation is generally associated with pro-survival signaling, it occurs alongside impaired apoptotic and transcriptional responses in SCA3, creating an imbalance that undermines neuronal fate determination.

Furthermore, abnormal phosphorylation of AKT, PTEN, and mTOR disrupts mTOR activation, leading to impaired autophagy, a process critical for the clearance of toxic mutant ataxin-3 aggregates⁹⁶. Modulation of AKT1 and TP53 signaling by crocin may counteract these abnormalities and provide neuroprotective benefits. The central role of AKT kinases in regulating survival, proliferation, and protein synthesis has been shown to rescue neuronal degeneration. In contrast, TP53, regulates apoptosis and oxidative stress responses but is often dysregulated in neurodegeneration. Crocin has been shown to upregulate TP53 and modulate Bax/Bcl-2 ratios, while enhancing AKT-related pro-survival signaling, thereby restoring balance in disrupted MAPK/AKT/TP53 networks. This dual action suggests crocin could mitigate neuronal dysfunction and improve motor and cognitive outcomes in SCA3^96,100–102.

The TP53 is a newly identified substrate of ataxin-3 and has been a focus of drug development for SCA3^103,104. Ataxin-3 binds to both native and polyubiquitinated TP53, and deubiquitinates and stabilizes it by blocking its degradation via the ubiquitin–proteasome pathway¹⁰³. Studies have shown that ATXN3 deletion destabilizes TP53, reducing its activity, whereas ectopic ATXN3 expression induces TP53-dependent apoptosis in neuronal models¹⁰³. Moreover, polyQ-expanded ataxin-3 was found to exacerbate TP53-dependent neurotoxicity, while maintaining strong binding and deubiquitination activity¹⁰³. This underscores direct involvement of TP53 in SCA3 pathogenesis. Meanwhile, AKT1, a critical component of the insulin pathway, has also emerged as a promising therapeutic target for SCA3 due to its role in neurodegeneration. Activation of AKT1 by SC79 was shown to enhance neuronal survival in disease models, and improve cognitive function in AD, highlighting its potential in SCA3 drug discovery¹⁰⁵. Furthermore, mutant ataxin-3 with an abnormally expanded polyglutamine chain disrupts AKT1-mediated pro-survival signaling pathways, directly contributing to SCA3 pathogenesis. The mutant protein reduces the level of serine 473-phosphorylated AKT1 (p-AKT S473), the active form, which is critical for neuronal survival, causing diminished pro-survival signals in SCA3 models. This impairment alters downstream AKT1 effectors including GSK3, FOXO, and mTOR, which are implicated in increased susceptibility to oxidative stress and neuronal death. The dysregulated AKT1 signaling cascade represents a key mechanism by which mutant ataxin-3 drives neurodegeneration in SCA3⁹⁰.

The molecular docking simulations indicated that NPs can spontaneously bind to the target proteins, with binding energies ranging from − 6.7 to − 9.5 kcal/mol for AKT1 and − 5.2 to − 7.9 kcal/mol for TP53 (Table S14 and Table S15). In this study, troriluzole was used as the control compound, as it represents a promising therapeutic candidate for SCA3. In the phase 3 trial (NCT03701399), the primary endpoint was not met across the overall population; however, a subgroup of ambulatory patients demonstrated significant clinical benefit, with least squares mean difference of –0.71 on the functional Scale for the Assessment and Rating of Ataxia (f-SARA) versus placebo (nominal P = 0.031) after 48 weeks (https://www.neurologylive.com/view/troriluzole-shows-promise-spinocerebellar-ataxia-type-3-subgroup-despite-failing-achieve-primary-end-point)¹⁰⁶. Troriluzole also reduced patient-reported falls by 58%, improving safety and quality of life. Long-term extension studies further indicated reduced disease progression by 50-70%, corresponding to a delay in functional decline of 1.5-2.2 years. Mechanistically, it reduces glutamate excitotoxicity, supporting its potential as a first-in-class disease-modifying therapy^107–109.

Among the selected NPs, crocin exhibited the strongest binding affinities (− 9.5 kcal/mol for AKT1 and − 6.6 kcal/mol for TP53) compared to the control compound troriluzole (− 8.4 kcal/mol for AKT1 and − 6.0 kcal/mol for TP53). Furthermore, for AKT1, residues such as Gly157, Phe161, Val164, Ala177, Lys179, and Thr211 are located within or adjacent to the ATP-binding pocket, which is part of the kinase domain (residues ~ 148–408) and includes the conserved Gly-rich loop (residues 157–164) and activation segment (residues 290–315). Notably, crocin formed HBs with Glu228 and Thr291, and interacted with Phe161 via pi-alkyl interaction, therefore highlighting engagement with both the ATP-binding loop and activation segment which is essential for regulation of kinase activity¹¹⁰. On the other hand, for TP53, the interacting residues Cys220, Glu221, Ser227, Asp228, and Cys229 are within the DNA-binding domain (residues 102–292), specifically the L3 loop (residues ~ 220–230), which is vital for structural stability and DNA interaction^111,112. This loop coordinates zinc binding and forms part of the DNA minor groove interface. Ligand binding at this site may impair DNA-binding ability and transcriptional regulation of TP53^111,112, indicating that interactions in this region could modulate the tumor suppressor function of the protein. This suggests that crocin and other selected NPs can modulate SCA3-related targets. Crocin exhibits potent neuroprotective properties through its antioxidant, anti-apoptotic, and anti-inflammatory mechanisms, which may offer therapeutic benefits for neurodegenerative disorders, including SCA3. In vitro studies have shown that crocin can prevent oxidative stress-induced apoptosis in PC-12 cells by reducing lipid peroxidation (85% cell viability at 10 μM), inhibiting caspase-8 activation, and increasing glutathione synthesis via modulation of γ-glutamylcysteine synthetase¹¹³. These effects are linked to improved mitochondrial function and suppression of ceramide-induced apoptotic signaling^113,114. Although studies specifically targeting SCA3 are limited, crocin’s inhibition of ceramide pathways and enhancement of the glutathione system suggest potential efficacy against polyglutamine-related neuronal degeneration. In vivo research supports crocin’s ability to reduce hippocampal neuronal loss in epilepsy¹¹⁵ and to mitigate AD’s pathology by decreasing Aβ1-42 deposition and improving cognition¹¹⁴. Furthermore, crocin significantly decreases infarct volume and neuronal damage in ischemia–reperfusion models¹¹⁶, counteracts paraquat-induced oxidative brain injury¹¹⁷ and enhances cognitive function via CREB-BDNF and PI3K/AKT pathways^118,119. These findings collectively highlight crocin’s broad neuroprotective potential.

MD simulations were conducted to evaluate crocin complex stability, which confirmed the stable binding of crocin to both AKT1 and TP53, as indicated by its low RMSD, RMSF, and Rg values (Fig. 5). Additionally, crocin exhibited the highest number of HBs in both systems (Fig. 6), which further stabilized the complex. Conformational motion analysis by PCA, FEL, and DCCM further validated crocin’s strong interactions with AKT1 and TP53 (Figs. 7, 8 and 9). The MM/PBSA calculations showed crocin had binding free energies of − 83.91 kcal/mol for AKT1 and − 24.16 kcal/mol for TP53, surpassing that of troriluzole (− 24.01 kcal/mol for AKT1 and − 18.31 kcal/mol for TP53) (Table 3). These findings highlight crocin’s superior stability and binding efficiency, making it a strong candidate for further experimental validation as a therapeutic for SCA3. Moreover, in the 100 ns MD simulations, crocin demonstrated superior binding stability and adaptability compared to troriluzole in both 3CQU and 6GGE complexes. Crocin consistently maintained key HBs and hydrophobic interactions with active site residues, showing minimal displacement over time. In contrast, troriluzole exhibited more fluctuations in binding orientation and contact residues. These findings suggest that crocin forms a more stable and persistent interaction network, supporting its potential as a stronger candidate for therapeutic targeting. Further ADMET sensitivity analysis highlighted crocin’s favorable pharmacological and safety profile compared to the reference compound, troriluzole. Although crocin showed poor absorption and limited BBB permeability, it demonstrated higher aqueous solubility and was neither a substrate nor inhibitor of major CYP450 enzymes, indicating a low risk of metabolic interactions. It also exhibited low CYP inhibitory promiscuity and was predicted to be non-AMES toxic, non-carcinogenic, and non-inhibitor of the hERG gene. In contrast, troriluzole showed better absorption but presented metabolic liabilities. Overall, crocin’s safety, metabolic stability, and low potential of interaction support its selection as a promising lead compound for SCA3 therapy.

Neuroprotective effects of crocin have been demonstrated through antioxidant, anti-apoptotic, and protective mechanisms against mitochondrial damage in neuronal cell models exposed to oxidative stress. For instance, crocin attenuated mitochondrial apoptosis by modulating the balance of Bcl-2 family proteins, reducing oxidative injury, and suppressing caspase activation in neuronal cells^114,120. Further, crocin has been shown to confer neuroprotection in in vivo models of cerebral ischemia, traumatic brain injury, and spinal cord injury by decreasing oxidative stress, inflammation, and neuronal apoptosis, supporting its potential relevance in neurodegenerative conditions^114,120. To test the efficacy of crocin in SCA3, adeno-associated virus (AAV)-based mouse model of SCA3 can be adopted to allow evaluation of neuroprotective and anti-aggregation effects of natural compounds including crocin¹²¹. The model expresses full-length human ATXN3 with 84 glutamine repeats throughout the brain, capable of reproducing clinical features of SCA3 including locomotor deficits, cerebellar neuronal loss and progressive cerebellar degeneration. Further, the therapeutic potential of crocin can be validated through investigations in in vitro models of primary fibroblasts and induced pluripotent stem cells (iPSCs) derived from SCA3 patients, and in vivo models of MJD84.2 transgenic mice harboring a YAC transgene that expresses a human ATXN3 modified with SCA3-associated 84 CAG repeat expansion, and AT3q130 Caenorhabditis elegans expressing full-length ataxin-3 proteins with 130 glutamines. While the epidemiological features of SCA3 are undeniable, the feasibility of conducting clinical trials is being addressed through a combination of strategies. These include improvements in clinical outcome measures, the identification of imaging and fluid biomarkers, and innovations in clinical trial design through established collaborative consortia throughout North America and Europe.

Benchmarking is an important step in the improvement, assessment, and comparison of the performance of drug discovery platforms and technologies. Ideally, placing the therapeutic potential of crocin into perspective requires comparison with well-established neuroprotective agents. For instance, riluzole, the Food and Drug Administration (FDA)-approved drug for amyotrophic lateral sclerosis (ALS), is known to mitigate glutamate-mediated excitotoxicity and extend motor function¹²². Likewise, edaravone, also licensed for ALS, acts as a free radical scavenger that reduces oxidative stress and lipid peroxidation, thereby delaying neuronal damage¹²³. In AD, memantine functions as the N-methyl-D-aspartate (NMDA) receptor antagonist to protect against excitotoxicity, while donepezil enhances cholinergic signaling and provides symptomatic relief¹²⁴. Crocin exhibits mechanistic similarities with these drugs, particularly through its antioxidant capacity, regulation of apoptotic signaling, and maintenance of mitochondrial stability^120,125. However, as a natural compound with multi-target activity, crocin may offer unique advantages in addressing the multifactorial drivers of SCA3 pathology. Comparative evaluation of crocin against these drugs in preclinical trials could clarify its relative neuroprotective efficacy and strengthen its translational potential.

Limitations

While this study utilized a comprehensive computational approach to identify potential therapeutic effects of crocin against SCA3, several limitations must be acknowledged. The research was conducted entirely through in silico methods, including network pharmacology, molecular docking, MD simulation, PCA, FEL, DCCM, and MM/PBSA calculations. Although these tools are valuable for early-stage drug discovery and provide insights into molecular interactions, they cannot fully replicate the biological complexity of living systems. Therefore, the binding affinities and stability of crocin with AKT1 and TP53, while promising, need to be validated through experimental studies. Additionally, the NP target predictions relied on databases such as SwissTargetPrediction and SuperPred, which use chemical similarity and machine learning algorithms. These tools may introduce false positives or overlook key targets due to dataset limitations. The disease-related targets retrieved from GeneCards also vary in relevance and supporting evidence, which may influence the reliability of target selection. The study emphasized the MAPK signaling pathway based on pathway enrichment analysis, but the multifactorial nature of SCA3 suggests involvement of other signaling cascades that were not explored in depth. Moreover, molecular docking and MD simulation focused on a limited number of protein targets, which might not capture the broader polypharmacological profile of crocin. These are essential considerations for the development of drug candidates targeting SCA3.

Conclusions

In this study, we employed an integrated in silico strategy combining network pharmacology and sophisticated database mining technologies to determine the main targets of NPs against SCA3. To clarify the molecular processes behind SCA3 and the therapeutic potential of NPs, we adopted a methodical approach to build the protein interaction network. Our findings revealed that the activity of NPs involved the hsa04010: MAPK signaling pathway, which has also been implicated in the pathogenesis of SCA3. Additionally, we identified key targets of NPs, including AKT1 and TP53, which could have potential for neuroprotection and disease modification. The combination of network pharmacology, molecular docking, and MD simulation provide compelling evidence for NPs as promising therapeutic agents for SCA3. However, further clinical research and experimental validation are required to develop these compounds into novel pharmacological interventions for SCA3.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, K.H.W.; Methodology and Resources, M.R., M.N.U., and S.M.I.H.; Formal analysis, M.R., and N.S.M.H.; Writing-original draft preparation, M.R., and S.M.I.H.; Writing-review and editing, L.W.L., and K.H.W.; Supervision, L.W.L., and K.H.W.; Project administration, K.H.W.; Funding acquisition, K.H.W. All authors have read and agreed to the published version of the manuscript.

Data availability

The authors declare that the data supporting the findings of this study are available within the article and its Supplementary Designation. Raw data that support the findings of this study are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.