Structure-guided cheminformatics optimization of NLRP3 inhibitors based on the NP3-253-NLRP3 complex for potential neurodegenerative disease therapeutics

Introduction

Neurodegenerative diseases are progressive disorders characterised by the deterioration of neuronal structure and function, ultimately leading to cognitive, motor, and behavioural impairments. Alzheimer's disease (AD), Parkinson's disease (PD), and other neurodegenerative disorders collectively represent an increasing global health burden, particularly with population ageing. Although these diseases differ in their clinical and pathological manifestations, accumulating evidence indicates that dysregulated interactions between the immune system and the central nervous system (CNS) are common features of disease progression [1] . The principal resident immune cells in the CNS, microglia, play a critical role in maintaining neuronal homeostasis by monitoring the neural environment, removing cellular debris, and responding to injury or infection. However, persistent microglial activation can promote chronic neuroinflammation and contribute to neuronal dysfunction and degeneration. One of the major innate immune pathways implicated in this process is the NOD-, LRR-, and pyrin domain-containing protein 3 (NLRP3) inflammasome.

NLRP3 functions as an intracellular sensor of cellular stress and danger-associated signals, responding to diverse stimuli, including damaged-cell components, microbial products, and environmental stressors [2] . NLRP3 assembles with the adaptor protein ASC and pro-caspase-1, resulting in caspase-1 activation. Caspase-1 activation promotes maturation of the pro-inflammatory cytokines interleukin-1β (IL-1β) and interleukin-18 (IL-18) and can induce pyroptosis, thereby amplifying inflammatory signalling. While this pathway is essential for host defence, sustained NLRP3 activation in microglia can produce prolonged inflammatory responses and contribute to neuronal injury. Increased NLRP3 activity has been associated with AD and PD, where it may interact with pathological protein aggregation, synaptic dysfunction, and neuronal loss. [3] Consequently, NLRP3 is likely be one of the most potential therapeutic target for neurodegenerative diseases. Smallmolecule NLRP3 inhibitors have demonstrated encouraging anti-inflammatory effects in preclinical models, although no NLRP3-targeting compound has yet achieved broad clinical approval [4] .

Structure-guided optimisation and cheminformatics approaches may therefore facilitate the identification of potent, selective, and drug-like NLRP3 inhibitors for future neurodegenerative disease therapeutics.

Methodology 2.1 Visualization and Validation of Sample

The crystal structure of human NLRP3 bound to NP3-253 (PDB ID: 9GU4) was visualized using RasMol and PyMOL to examine its overall fold, domain organization, and NACHT-domain-ADP cofactor arrangement. Structural quality was evaluated using SAVES and ERRAT, which assess stereochemical and nonbonded interaction quality. [5] Structural similarity between 9GU4 and the reference structure 7ALV was assessed using PyMOL alignment and Cα RMSD analysis.

Fetching of NLRP3 Structure

The 9GU4 structure was fetched from the Protein Data Bank in mmCIF format and converted to PDB format using Open Babel. Chain A contained 4, 024 atoms and 488 resolved residues (130-895). The similar protein sequence was extracted for subsequent similarity and conservation analyses.

Similarity and Conservation of Sequence

The chain-A sequence was compared with other NLRP3 sequences using BLAST to identify homologous and conserved regions [6] . Multiple-sequence alignment and physicochemical colour coding facilitated the identification of conserved residues that are potentially important for structural stability and function.

Preparation of Receptor and Binding-Site Identification

The receptor was converted into PDBQT format using Open Babel for AutoDock Vina docking [7] . Potential binding pockets were identified using fpocket, which evaluates cavity geometry, volume, solvent accessibility, hydrophobicity, and druggability [8,9] . Thirty-one candidate pockets were identified, with the biologically relevant NP3-253-binding region selected for docking.

Ligand Library Preparation

Twenty-three structurally related dibenzothiazepine, dibenzothiazine, and dibenzazepine derivatives were selected from ChEMBL. Canonical SMILES were converted into three-dimensional conformers using RDKit ETKDG and optimized using UFF. Ligands were subsequently converted to PDBQT format using Open Babel [10, 11] .

Hit identification via Virtual screening

Multiple binding poses were generated for each ligand and ranked according to predicted binding affinity with the help of AutoDock Vina. The lowest-energy poses were retained for structural analysis and subsequent candidate prioritization [12,13] .

Cheminformatics and ADMET Analysis

RDKit was used to calculate molecular descriptors, including molecular weight, LogP, TPSA, hydrogenbonding properties, and rotatable bonds. Lipinski's Rule of Five, PAINS filtering, and synthetic accessibility were assessed. Morgan fingerprints supported similarity analysis, hierarchical clustering, and chemical-space visualization, while Bemis-Murcko scaffolds identified common structural frameworks. ADMET-AI was used to predict blood-brain barrier permeability, hERG inhibition, hepatotoxicity, and AMES mutagenicity. Finally, docking scores and pharmacokinetic, structural, and toxicity parameters were integrated into a multiparameter ranking framework for prioritizing potential NLRP3 inhibitors [14,15] .

Results

Retrieved Structure Overview

The three-dimensional structure of NLRP3 (PDB ID: 9GU4), retrieved and processed, served as the receptor model for all subsequent analyses in this study. The structural coverage is non-continuous, with several distinct gaps interrupting the sequence. These gaps correspond to regions that were not resolved in the experimental electron density, reflecting flexible loops and linker segments with high conformational mobility. Such discontinuities are common in large multidomain proteins, where highly flexible regions connecting folded domains often remain structurally undefined, even in high-quality crystallographic datasets. The overall distribution of resolved residues indicates that the structured regions of NLRP3 relevant to docking and binding-site analysis are well represented in the 9GU4 model, whereas the unresolved regions are primarily located away from the nucleotide-binding pocket examined in this study [16,17] .

Structural Quality and Validation

The 9GU4 structure returned an ERRAT score exceeding 90%, a level generally considered indicative of a reliable, well-refined structural model at the 2-3 Å resolution range applicable to this entry. To further assess stereochemical quality, the φ (phi) and ψ (psi) backbone dihedral angles were calculated for all residues in chain A and visualized as a Ramachandran plot. The majority of residues cluster within the favoured regions associated with α-helical and βsheet conformations, consistent with the mixed secondarystructure composition expected of the NLRP3 NACHT domain [18,19] .

Hydrophobicity Profile

The resulting profile shows alternating hydrophobic and hydrophilic stretches distributed along the sequence, reflecting the heterogeneous physicochemical character typical of a cytosolic, multidomain protein. Notably, no extended hydrophobic stretch characteristic of a transmembrane helix was observed anywhere in the profile, which is consistent with the known cytosolic localization of NLRP3. Instead, several shorter, localized hydrophobic regions are visible, which likely contribute to the formation of stable hydrophobic cores within the folded domains and may additionally correspond to surfaces relevant to ligand binding, a possibility examined further during binding-site identification [20,21] . In general, lower B-factor values are concentrated in the protein core, including the nucleotide-binding region, whereas higher values are enriched in surface-exposed loops, in line with the pattern typically expected for a folded, multidomain protein [22,23] .

Atomic Composition and Residue Mass Distribution

The elemental composition of the modelled protein was also examined, showing the expected predominance of carbon, followed by oxygen and nitrogen, with a comparatively small proportion of sulphur, consistent with a standard protein composed of the 20 canonical amino acids. The relatively low sulphur content reflects the limited occurrence of cysteine and methionine residues, which are typical of most protein sequences. No anomalous residue masses were detected, confirming that the modelled structure consists solely of standard proteinogenic residues suitable for the downstream chemoinformatic and docking analyses that follow.

RMSD Comparison

Molecular Docking Results

All 23 prepared ligands were docked against the prepared NLRP3 receptor using AutoDock Vina. For each ligand, multiple candidate binding poses were generated and ranked by predicted binding free energy; the lowest-energy pose was retained as the representative docked conformation.

Predicted binding affinities ranged from -6.563 kcal/mol (CHEMBL1688549) to -4.679 kcal/mol (CHEMBL1764), a spread of approximately 1.9 kcal/mol across the library. While none of the 23 compounds reached binding energies typically associated with high-affinity lead compounds (generally below -8 kcal/mol), every ligand produced a negative docking score, indicating that the shared tricyclic scaffold is broadly compatible with the geometry of the NLRP3 nucleotide-binding pocket, even where individual substituents produce weaker or stronger predicted binding.

The five top-ranked compounds, CHEMBL1688549, CHEMBL422, CHEMBL1688551, CHEMBL11, and CHEMBL570, span both dibenzothiazine and dibenzazepine scaffolds, suggesting that the binding site geometry can accommodate more than one closely related chemotype rather than favouring a single rigid scaffold shape. Notably, no clear correlation was observed between predicted docking score and experimental IC₅₀. For example, CHEMBL1688548 (IC₅₀ = 700 nM, among the most potent compounds experimentally) ranked only ninth by docking score, while CHEMBL11 (IC₅₀ = 22, 000 nM, comparatively weak experimentally) ranked fourth. This discrepancy is expected rather than concerning: the IC₅₀ values were originally measured against a different biological assay target, whereas docking here was performed specifically against the NLRP3 nucleotidebinding pocket. From a structure-activity perspective, several trifluoromethyl-substituted dibenzothiazine derivatives appear among the better-ranked compounds, suggesting this substitution pattern may offer favourable complementarity with the NLRP3 pocket and represents a reasonable starting point for future optimization.

Structures of the Top-Ranked Candidates

Among the top-ranked ligands, two notable outliers stand apart from the dominant scaffold: CHEMBL11, which carries a dibenzazepine core rather than a dibenzothiazine core, and CHEMBL418995, which carries a carboxylic-acid substituent. Both nonetheless achieved docking scores competitive with those of the dibenzothiazine-dominated set, demonstrating that, although the phenothiazine-like core clearly dominates the highest-scoring compounds, alternative chemotypes within the same broader chemical space can still be accommodated by the binding pocket. Among the ten top-ranked docking candidates specifically, seven compounds showed no Lipinski violations, while three compounds, CHEMBL1688549, CHEMBL570, and CHEMBL479, each showed a single violation, attributable to LogP values marginally exceeding the threshold of 5. No compound violated more than one rule, indicating generally good physicochemical suitability for oral bioavailability across the top-ranked set (Table 2). These results suggest that the screened chemical space is largely consistent with drug-like property norms, plausibly reflecting the historical optimization of this tricyclic chemotype toward CNS-active small molecules with balanced permeability and lipophilicity.

Representative Ligand Structure

To illustrate the structural features underlying the screened library, a representative compound, CHEMBL1688548, was visualised in 3D using py3Dmol. The structure displays a tricyclic phenothiazine scaffold composed of two fused

benzene rings connected by a central heterocyclic ring containing sulfur and nitrogen atoms. This aromatic tricyclic system provides a rigid, hydrophobic core capable of favourable interactions within the NLRP3 nucleotidebinding pocket, with structural variation across the library arising primarily from substituent modifications rather than changes to this core framework. CHEMBL1688548 itself achieved a docking score of -5.622 kcal/mol and has an experimentally reported IC₅₀ of 700 nM, making it a reasonably representative member of the screened series for the purpose of this structural illustration. On the basis of its predicted binding affinity of-5.622 kcal/mol, CHEMBL1688548 was retained, alongside the remaining 22 screened compounds, for the chemoinformatic and ADMET analyses presented in the sections that follow.

Synthetic Accessibility (SA) Score

Synthetic accessibility for each compound was estimated using the RDKit implementation of the SA score (Ertl & Schuffenhauer, 2009), which ranges from approximately 1 (very easy to synthesize) to 10 (highly complex). As shown in Figure 14, the distribution of SA scores across the full 23compound library is tightly clustered between approximately 1.9 and 2.95, with most compounds concentrated in the 2.0-2.4 range

Bemis-Murcko Scaffold Analysis

Bemis-Murcko scaffolds (Bemis & Murcko, 1996), representing the core ring systems of each compound after removal of peripheral side chains, were extracted for all 23 compounds. This analysis confirmed that the library is dominated by two related scaffold families: a tricyclic dibenzo[b, f] [1,4] thiazepine/phenothiazine-type scaffold, present in the majority of compounds, and a dibenzazepine (iminodibenzyl) scaffold, represented by CHEMBL11 and CHEMBL418995 (Table 3). This confirms that the dataset represents a focused, scaffold-conserved chemical series rather than a structurally diverse screening library, an observation consistent with the shared-core structureactivity trends and one that makes the library well suited to SAR-driven optimization in future work.

Tanimoto Similarity Heatmap

The resulting matrix exhibits a clear block-diagonal pattern, indicating the presence of structurally coherent clusters rather than a uniformly diverse library. High-similarity clusters (Tanimoto > 0.6-0.7) are observed among compounds CHEMBL1688548, CHEMBL1688549, CHEMBL824, and CHEMBL1688551, corresponding to halogenated and non-halogenated dibenzothiazine derivatives bearing primary amine side chains, while a second closely related pair is observed between CHEMBL1323355 and CHEMBL1764, reflecting nearidentical two-dimensional structural features. In contrast, CHEMBL418995, a dibenzazepine carboxylic-acid derivative, shows consistently low similarity (Tanimoto < 0.15) to nearly all other compounds, confirming its position as the most structurally distinct member of the library.

Hierarchical Clustering

Ward-linkage clustering of Morgan fingerprints identified three major structural clusters, with CHEMBL728 emerging as a distinct outlier.

Similarity Network Analysis

A molecular similarity network was constructed by connecting compound pairs with a Tanimoto similarity greater than 0.60. The resulting graph forms a single large, densely connected component comprising 19 of the 23 compounds, reflecting the shared dibenzothiazine-based scaffold and substantial similarity in substituents across the dataset. Within this main cluster, CHEMBL11 and CHEMBL1620412 occupy more peripheral positions, connected primarily through intermediate nodes such as CHEMBL564, suggesting that these dibenzazepine and diethylamino variants represent structurally extended members of the same chemical space rather than core scaffold representatives.

Physicochemical-Docking Relationship

To further examine whether physicochemical properties relate to predicted binding affinity, a scatter plot of molecular weight against calculated LogP was generated for the ten highest-ranked docking candidates, with point colour representing docking score and point size representing the number of hydrogen-bond donors. The best-performing ligand, CHEMBL1688549 (-6.563 kcal/mol), exhibits a moderate molecular weight (338 Da) and a LogP value close to the Lipinski threshold (5.05), positioning it near the upper boundary of favourable lipophilicity rather than at a physicochemical extreme.

ADMET Prediction and Final Candidate Selection

ADMET properties were predicted for the full ligand library using the ADMET-AI platform, which provides DrugBankreferenced percentile rankings for pharmacokinetic and toxicity-related endpoints, including solubility, lipophilicity, plasma-protein binding, volume of distribution (VDss), halflife, and multiple toxicity predictors. Predicted VDss values were moderate to high, suggesting effective tissue distribution. The top candidates showed moderate-to-high predicted lipophilicity and relatively low aqueous solubility, a pattern typical of tricyclic aromatic scaffolds that may favor membrane permeability and intracellular target engagement, though formulation optimization may be required for the least soluble compounds. This composite ranking identified CHEMBL11, CHEMBL288110, CHEMBL1688549, CHEMBL422, and CHEMBL1688551 among the highest-priority compounds. Notably, CHEMBL11 achieved the best overall composite score despite not being the single strongest docking hit, reflecting a favourable balance of binding affinity, molecular size, lipophilicity, and synthetic accessibility rather than dominance in any one criterion. Overall, the convergence of docking performance, drug-likeness, synthetic accessibility, BBB permeability, and PAINS filtering supports the prioritization of this final candidate set for future experimental validation. Overall, this docked complex supports the structural plausibility of CHEMBL1688549 as a binder of the NLRP3 NACHT-domain nucleotide pocket. The observed binding mode is consistent with the cavity definition used throughout docking and corroborates the earlier fpocket and docking results, reinforcing that the selected pocket represents a chemically and geometrically reasonable binding site for this compound class, and providing a structurally grounded conclusion to the results presented in this chapter.

Conclusion

This study set out to characterize the NLRP3 inflammasome structurally and to identify potential small-molecule inhibitors relevant to neurodegenerative disease therapeutics, using an integrated computational workflow spanning structural bioinformatics, molecular docking, and chemoinformatic profiling. The retrieved NLRP3 structure (PDB: 9GU4) was validated through multiple independent measures, including a high ERRAT score [24,25] , a favorable Ramachandran distribution, and near-complete structural identity with a closely related homolog (7ALV, RMSD = 0.494 Å), confirming its reliability as a target for downstream analysis. Binding-site identification using fpocket correctly recovered the biologically relevant nucleotide-binding cleft of the NACHT domain, the native site of the co-crystallised ADP cofactor, providing a structurally grounded docking target rather than an arbitrarily selected cavity. Molecular docking [26] of twentythree tricyclic candidate compounds against this pocket identified a consistent structure-activity pattern centred on a dibenzothiazine/phenothiazine core, with docking scores ranging from -6.563 to -4.679 kcal/mol. Subsequent chemoinformatic and ADMET profiling narrowed this library to ten candidates that simultaneously met Lipinski, PAINS, blood-brain barrier permeability, and synthetic accessibility criteria, with CHEMBL422, CHEMBL1688551, and CHEMBL11 emerging as the most promising overall candidates when binding affinity and drug-likeness were considered together. Taken as a whole, this study demonstrates that a structure-guided computational pipeline, combining validated structural data, biologically informed binding-site selection, docking, and multi-parameter chemoinformatic filtering, can meaningfully prioritize a small, experimentally tractable set of NLRP3 inhibitor candidates from a larger starting library. While the absence of experimental validation and the inherent limitations of rigid-receptor docking mean these findings represent a starting point rather than a conclusive result, the convergence of favourable binding, drug-likeness, and predicted pharmacokinetic properties across the final candidate set offers a reasonable, defensible basis for future experimental follow-up. In doing so, this work contributes to a reproducible computational framework that may support future efforts toward NLRP3-targeted therapeutics for Alzheimer's disease, Parkinson's disease, and related neurodegenerative conditions.

References

1. Mamuladze T, Kipnis J. Type 2 immunity in the brain and brain borders. 2023. doi:10.1038/s41423-023-01043-.
2. Swanson KV, Deng M, Ting JPY. The NLRP3 inflammasome: molecular activation and regulation to therapeutics. Nature Reviews Immunology. 2019;19(8):477-489.
3. Smith AP, Creagh EM. Caspase-4 and-5 biology in the pathogenesis of inflammatory bowel disease. Frontiers in Pharmacology. 2022. doi:10.3389/fphar.2022.919567.
4. Heneka MT, McManus RM, Latz E. Inflammasome signalling in brain function and neurodegenerative disease. Nature Reviews Neuroscience. 2018;19(10):610-621.
5. Mangan MSJ, Olhava EJ, Roush WR, Seidel HM, Glick GD, Latz E. Targeting the NLRP3 inflammasome in inflammatory diseases. Nature Reviews Drug Discovery. 2018;17(9):588-606.
6. Heneka MT, MP, Stutz A, Delekate A, Schwartz S, Latz E, et al. NLRP3 is activated in Alzheimer's disease and contributes to pathology in APP/PS1 mice. Nature. 2013;493(7434):674-678.
7. Gordon R, Albornoz EA, Christie DC, Langley MR, Kumar V, Mantovani S, et al. Inflammasome inhibition prevents α-synuclein pathology and dopaminergic neurodegeneration in mice. Science Translational Medicine. 2018;10(465).
8. Voet S, Srinivasan S, Lamkanfi M, van Loo G. Inflammasomes in neuroinflammatory and neurodegenerative diseases. EMBO Molecular Medicine. 2019;11(6):e10248.
9. Sharif H, Wang L, Wang WL, Magupalli VG, Andreeva L, Qiao Q, et al. Structural mechanism for NEK7-licensed activation of NLRP3 inflammasome. Nature. 2019;570(7761):338-343.
10. Coll RC, Robertson AAB, Chae JJ, Higgins SC, Muñoz-Planillo R, Inserra MC, et al. A small-molecule inhibitor of the NLRP3 inflammasome for the treatment of inflammatory diseases. Nature Medicine. 2015;21(3):248-255.
11. Bandbe T, Johri V, Kumari U. Structure-guided genome-wide association analysis of ALK variants with GWAS data using R. Computational Biology and Bioinformatics. 2025;13(2):72-85. doi:10.11648/j.cbb.20251302.13.
12. Johri V, Kumari U. Sustainable neuroprotection via proteomics and integrative NGS-CADD analysis of 6CVP variant to improve future neurodegenerative health outcomes. IOSR Journal of Pharmacy and Biological Sciences. 2026;21(3):1-10. doi:10.9790/3008-2103010110.
13. Johri V, Kumari U. NGS-based polyglutamine expansion pattern detection using machine learning models to predict the conformation states of Huntingtin. IOSR Journal of Pharmacy and Biological Sciences. 2025;20(5):19-27. doi:10.9790/3008-2005031927.
14. Kumari U, Gunjan, Basak S. NGS contribution in genetic diagnosis for brain disorder. Journal of Clinical Otorhinolaryngology, Head, and Neck Surgery. 2023;27(1):5525-5536.
15. Kumari U, Mukherjee S, Pradhan M, Chakrabarti S. NGS analysis approach for neurodegenerative disease with Biopython. Journal of Emerging Technologies and Innovative Research. 2023;10(9):1-?. doi:10.1729/Journal.36043.
16. Kumari U, Pacholee K. In-silico drug discovery-based approach to treat impairments in patients of Alzheimer's disease. Journal of Emerging Technologies and Innovative Research. 2023;10(12):B236-B245. doi:10.1729/Journal.37001.
17. Kumari U, Gunjan, Basak S. NGS contribution in genetic diagnosis for brain disorder. Journal of Clinical Otorhinolaryngology, Head, and Neck Surgery. 2023;27(1):5525-5536.
18. Nagpal G, Kumari U. Next generation sequence analysis of amyloid precursor-like protein 2 (APLP2) E2 domain in Alzheimer's disease. Journal of Emerging Technologies and Innovative Research. 2025;12(1):ppd72-ppd79.
19. Uniyal U, Kumari U. Next generation sequencing data analysis of A25T transthyretin structure associated with Parkinson's disease. Journal of Emerging Technologies and Innovative Research. 2025;12(3):B612-B622. doi:10.1729/Journal.44079.
20. Kokad M, Kumari U. Next-generation sequencing data analysis in DYRK1A associated with neurodegenerative disease. IOSR Journal of Biotechnology and Biochemistry. 2025;11(6):40-48. doi:10.9790/264X-1106014048.
21. Mehrotra K, Kumari U, Pande S. Integrative workflow of Biopython and molecular docking to explore novel therapeutics targeting PARK7-DJ1. IOSR Journal of Biotechnology and Biochemistry. 2025;11(6):1-10. doi:10.9790/264X-1106010110.
22. Singh N, Kumari U. Multi-omics-based drug discovery pipelines incorporating machine learning, AI, and Biopython in Parkinson's disease: a case study of 5A polymorph of alpha-synuclein. Journal of Emerging Technologies and Innovative Research. 2026;13(2). doi:10.56975/jetir.v13i2.575492.
23. Mandra D, Kumari U, Singh N, Marouthu A. Integrated multi-omics and Biopython approaches for target identification of α-synuclein polymorph in Parkinson's disease. IOSR Journal of Biotechnology and Biochemistry. 2026;12(2):12-21.
24. Singh N, Kumari U, Mandra D, Marouthu A. Integrated machine learning-based toxicity prediction with molecular docking for safer drug candidate screening in Parkinson's disease. Computational Biology and Bioinformatics. 2026;14(1):41-53. doi:10.11648/j.cbb.20261401.14.
25. Johri V, Ramakrishnan S, Tripathi A, Kumari U, Jayaraman A. Structural insights into the interaction of TAP01 antibody with amyloid beta peptide (Aβ pE314) for Alzheimer's disease therapeutics. International Journal of Drug Delivery Technology. 2026;16(50s):1007-1017. doi:10.25258/ijddt.16.50s.105.
26. Kumari U, Johri V, Jha T, Dhopate S. Structure-based drug designing for the prediction of epitope for targeting malignant brain tumor. Journal of Emerging Technologies and Innovative Research. 2024;11(7):B689-B697. doi:10.1729/Journal.40433.