A New Fragment‐Based Pharmacophore Virtual Screening Workflow Identifies Potent Inhibitors of SARS‐CoV‐2 NSP13 Helicase

Jordi Doijen, Jiexiong Xie, Simone Marsili, Trpta Bains, Mandeep Kaur Mann, Pravien Abeywickrema, Nick Van den Broeck, Christian Permann, Thierry Langer, Gökhan Ibis, Charles‐Alexandre Mattelaer, Jeremy Harvey, Sebastiaan van Raalte, Roberto Fino, Vineet Pande, Danielle Peeters, Aaron Patrick, Ellen Van Damme, Herman van Vlijmen, Marnix Van Loock, Edgar JacobyJournal of Computational Chemistry 2025 · OpenReview

TL;DR

Researchers have developed a novel virtual screening workflow called "FragmentScout" that leverages X-ray crystallography data from fragment screening to identify potential inhibitors of the SARS-CoV-2 NSP13 helicase. This new method successfully identified 13 potent micromolar inhibitors, demonstrating its effectiveness in accelerating the discovery of new antiviral agents.

What Problem Does This Paper Address?

The development of effective antiviral drugs often faces a significant bottleneck: translating initial "hits" from fragment-based drug discovery (FBDD) into potent lead candidates. FBDD typically identifies weak binders (millimolar potency) through techniques like X-ray crystallography screening (e.g., XChem). While these fragments provide valuable structural information about how they interact with a target protein, optimizing them into potent inhibitors (micromolar or better) is a complex and time-consuming process. Traditional virtual screening methods, while useful, can struggle to efficiently explore the vast chemical space and leverage the detailed structural insights gained from FBDD. This paper addresses the need for a more integrated and efficient approach to bridge the gap between fragment hits and potent drug candidates, specifically for the SARS-CoV-2 NSP13 helicase, a crucial enzyme for viral replication.

Methods

The core innovation presented is the FragmentScout workflow. This method combines several key steps:

  1. Fragment Clustering and Binding Site Identification: The workflow begins by analyzing existing X-ray crystallographic data from XChem fragment screening of the SARS-CoV-2 NSP13 helicase. Fragments that bind to similar regions are clustered, and key binding sites (e.g., nucleotide binding site, 5'-RNA binding site, stalk-zinc domain) are identified.
  2. Joint Pharmacophore Generation: For each identified binding site cluster, FragmentScout generates a "joint pharmacophore query." This is a crucial step where the pharmacophore features (e.g., hydrogen bond donors/acceptors, hydrophobic regions) from all experimental fragment poses within that cluster are aggregated. This creates a more comprehensive representation of the binding site's requirements than a single fragment pose would provide.
  3. Virtual Screening with LigandScout XT: The generated joint pharmacophore queries are then used to screen large 3D conformational databases using the LigandScout XT software. This pharmacophore-based virtual screening approach allows for the identification of compounds that match the aggregated features of multiple fragments, potentially leading to more potent binders.

For comparison, the researchers also employed Glide SP docking, a more traditional structure-based virtual screening method, for the nucleotide and 5'-RNA binding sites.

Datasets

The study utilized the following datasets:

  • SARS-CoV-2 NSP13 XChem Crystallographic Data: Fifty-one XChem PanDDA NSP13 fragment screening crystallographic coordinate files were downloaded from the RCSB Protein Data Bank (PDB IDs listed in the paper).
  • SARS-CoV-2 Cryo-EM Structure: The 6XEZ cryo-EM structure of the SARS-CoV-2 replication-transcription complex was also included.
  • J&J Internal Screening Compound Collection: A proprietary library of approximately 1.5 million small-molecule compounds was used for virtual screening.

Key Insights and Ideas

  • Leveraging Fragment Synergy: The key insight is that combining the pharmacophore information from multiple fragment hits binding to the same site can create a more robust and predictive query than relying on a single fragment. This "synergy" of fragments helps to better define the optimal binding interactions.
  • Bridging the FBDD Gap: FragmentScout effectively bridges the gap between millimolar fragment hits and micromolar lead candidates by using the detailed structural information from FBDD to guide a more focused virtual screening effort.
  • Importance of Excluded Volumes: The authors noted that defining and handling protein "excluded volumes" (regions where ligands cannot fit) is critical for accurate pharmacophore modeling and can significantly impact screening performance.
  • FragmentScout Outperforms Glide SP: The FragmentScout workflow demonstrated superior performance compared to Glide SP docking for both the nucleotide and 5'-RNA binding pockets, highlighting the value of its fragment-based pharmacophore approach.

Key Results

  • Discovery of 13 Potent Inhibitors: The FragmentScout workflow successfully identified 13 novel micromolar potent inhibitors of the SARS-CoV-2 NSP13 helicase.
  • Validation in Cellular and Biophysical Assays: These 13 compounds were validated in cellular antiviral assays (measuring inhibition of viral infection) and biophysical ThermoFluor assays (measuring protein stabilization).
  • Broad-Spectrum Activity: Of the 13 validated hits, most showed antiviral activity against SARS-CoV and HCoV-229E, suggesting potential for broad-spectrum coronavirus inhibitors.
  • Hit Rate Comparison: FragmentScout achieved higher hit rates for both the nucleotide and 5'-RNA binding sites compared to Glide SP docking. For instance, FragmentScout identified 1000 hits from ~10,969 structures for the 5'-RNA site, while Glide SP identified 1275 hits from ~26,111 structures.

Limitations and Future Work

  • Manual Steps in Workflow: While progress has been made, some steps in the FragmentScout workflow, such as supervised clustering and the inspection/elimination of blocking excluded volume features, remain interactive and could benefit from further automation.
  • Protein Conformer Consistency: The authors recommend using consistent protein conformers for building joint pharmacophore models, especially when fragments might induce different conformational states in the target. This is an area for further refinement.
  • QM/MM for Fragment Poses: Preliminary investigations suggest that QM/MM computations could improve the accuracy of fragment pose prediction, which in turn could enhance pharmacophore models. This is a promising avenue for future work.
  • Broader Target Application: The success of FragmentScout on NSP13 suggests its potential applicability to other targets where XChem fragment screening data is available, enabling more efficient drug discovery campaigns.
  • Further Validation: While 13 compounds were validated, the authors suggest further biophysical assays and structure determination experiments for the remaining cell-based hits to uncover additional starting points.

Summary updated

BibTeX
@article{doijen2025new,
  author = {Jordi Doijen and Jiexiong Xie and Simone Marsili and Trpta Bains and Mandeep Kaur Mann and Pravien Abeywickrema and Nick Van den Broeck and Christian Permann and Thierry Langer and Gökhan Ibis and Charles‐Alexandre Mattelaer and Jeremy Harvey and Sebastiaan van Raalte and Roberto Fino and Vineet Pande and Danielle Peeters and Aaron Patrick and Ellen Van Damme and Herman van Vlijmen and Marnix Van Loock and Edgar Jacoby},
  title = {{A New Fragment‐Based Pharmacophore Virtual Screening Workflow Identifies Potent Inhibitors of SARS‐CoV‐2 NSP13 Helicase}},
  journal = {Journal of Computational Chemistry},
  year = {2025},
  url = {https://openreview.net/forum?id=JlnfiETUob},
}
Figure