Allosteric pockets in the measles and Nipah virus polymerases: Mechanobiological insights and AI-driven drug discovery opportunities

Abstract

Nonsegmented negative-sense RNA viruses (nsNSVs)—including highly pathogenic pathogens such as measles virus (MeV), Nipah virus (NiV), Hendra virus (HeV), Ebola virus (EBOV), and others—pose major global health threats, yet most lack approved antiviral therapeutics. In the recent study, high-resolution cryo-electron microscopy (cryo-EM) revealed previously unrecognized allosteric pockets in the large (L) polymerase proteins of MeV and NiV, spatially distinct from the catalytic nucleotide-binding site. We further demonstrated that the non-nucleoside inhibitor ERDRP-0519 engages these pockets to allosterically ‘lock’ the polymerase in a mechanically inactive state. These findings reveal an allosteric mechanism of inhibition rooted in the conformational mechanics of the enzyme and highlight opportunities for integrating artificial intelligence (AI)-aided drug discovery (AIDD) into rational drug design.

Graphical abstract

1. Introduction

nsNSVs encompasses diverse human and animal pathogens, such as morbilliviruses (e.g., MeV, rinderpest virus), henipaviruses (e.g., NiV, HeV), and filoviruses (e.g., EBOV, Marburg virus).¹ Their genome replication and transcription are carried out by a multifunctional RNA-dependent RNA polymerase (RdRp) complex, consisting of the catalytic Large (L) protein and its essential cofactor, the phosphoprotein (P).^1,2 This complex has long been considered a highly attractive antiviral target.

Historically, therapeutic development has centered on nucleoside analogues that compete with natural nucleoside triphosphates (NTPs) at the catalytic site.^1,3 While effective in some viral systems, such inhibitors may also interact with host polymerases, increasing the risk of off-target toxicity.³ In contrast, allosteric inhibition—modulating activity by binding to distal, non-catalytic sites—offers an alternative route that can enable greater selectivity and novel mechanisms of action.

2. A newly identified allosteric pocket in MeV and NiV polymerases

The L protein of nsNSVs is a large, multi-domain enzyme encompassing an N-terminal domain (NTD), an RdRp domain, a GDP polyribonucleotidyltransferase (PRNTase) domain, a methyltransferase (MTase) domain, and a C-terminal domain (CTD).¹ Several non-nucleoside inhibitors have been developed against related polymerases. For example, AVG-233 is thought to block domain rearrangements in Respiratory Syncytial Virus (RSV); JNJ-8003 disrupts initiation by stabilizing inhibitory conformations in RSV and Human Metapneumovirus (HMPV); and GHP-88309 interferes with the conformational transitions needed for RNA synthesis in morbilliviruses.⁴

Prior to our work, inhibitors such as ERDRP-0519 and AS-136A were hypothesized to bind within the PRNTase domain of the MeV L protein,^5,6 but no direct structural evidence was available. In contrast to these predictions, our recent cryo-EM structures of MeV and NiV L–P complexes bound to ERDRP-0519 and AS-136A, respectively, revealed a distinct allosteric pocket located adjacent to the catalytic core.⁷

Binding of ERDRP-0519 to the newly identified allosteric pocket induces a pronounced conformational change in the L protein that underpins its inhibitory effect. Specifically, the piperidine ring of the inhibitor likely distorts the catalytic loop through close contact, displacing it from the pocket and misaligning residues essential for nucleotide incorporation.⁷ This conformational misalignment effectively “locks” the polymerase in a mechanically inactive state, preventing the dynamic transitions normally required for RNA chain elongation (Fig. 1A). From a mechanobiological perspective, the inhibition arises not from direct blockade of the catalytic site but from allosteric modulation of the enzyme's conformational mechanics, illustrating how small-molecule inhibitors can reprogram the structural energy landscape of viral polymerases. Such insights underscore the importance of mechanobiology in antiviral drug design, as targeting conformational dynamics can yield selective and potent inhibitors distinct from classical active-site competitors.

Fig. 1.

Structural analyses of allosteric pockets in MeV and NiV L–P complexes and their potential for AI-aided drug design. (A) Structural superposition of the catalytic loops from MeV L–P complexes in the apo state (gray) and ERDRP-0519–bound state (salmon). ERDRP-0519 is shown in stick representation. (B) Structural comparison of the ERDRP-0519–bound MeV L–P complex with the apo EBOV L–P complex (PDB: 7YER) and the apo MuV L–P complex (PDB: 8IZM). Interacting residues from the ERDRP-0519–bound MeV L–P complex, together with the corresponding residues from apo EBOV and MuV L–P complexes, are depicted as sticks and color-coded. (C) Illustration of the identified allosteric pockets in MeV and NiV polymerases and a proposed workflow for inhibitor discovery and development.

3. The possibility of cross-reactivity of ERDRP-0519

In the MeV L-P-ERDRP-0519 structure, the inhibitor occupies a hydrophobic cavity formed by the ‘fingers’ and ‘palm’ subdomains of the RdRp.⁷ Cryo-EM structures of the EBOV⁸ and MuV L-P⁹ complexes provide an opportunity for direct comparison, and structural alignments reveal that the equivalent residues in these polymerases do not present steric clashes with ERDRP-0519 (Fig. 1B). Together with our finding that ERDRP-0519 can bind to and inhibit the NiV polymerase,⁷ this observation suggests a structural basis for cross-reactivity and raises the possibility that ERDRP-0519, or optimized derivatives, could be developed as broad-spectrum inhibitors against diverse nsNSVs. Nonetheless, this hypothesis remains to be experimentally validated. Progress has been hindered by low expression yields and technical challenges associated with capturing inhibitor-bound L-P complexes from EBOV and MuV. Moving forward, advances in expression systems, cryo-EM methodologies, and complementary biochemical assays will be critical to establish whether ERDRP-0519 can indeed inhibit these viral polymerases.

4. Leveraging AI-aided drug discovery

The atomic-resolution cryo-EM structures of the MeV and NiV polymerases bound to non-nucleoside inhibitors provide an excellent structural foundation for rational drug design. As noted above, these inhibitors induce conformational rearrangements in the polymerase; therefore, allosteric pockets stabilized in the “locked” conformation represent more suitable targets for drug design. Beyond optimizing the existing non-nucleoside inhibitor ERDRP-0519 to improve stability, specificity, and oral bioavailability, the most promising opportunity lies in applying AI-aided drug discovery (AIDD) to design entirely new classes of inhibitor guided by our structural data. Recent advances in artificial intelligence have transformed the landscape of active compound discovery, offering substantial improvement in efficiency, precision, and scalability compared to conventional methods.¹⁰ A key barrier, however, has been the limited availability of high-resolution target structures and the lack of structurally-defined binding pockets.¹⁰ Our identification of previously unrecognized allosteric pockets overcomes this limitation, revealing druggable surfaces that are distinct in sequence, shape, and chemical environment from the highly conserved catalytic core—thus offering ideal opportunities for selective inhibitor development (Fig. 1C).

We envision that modern AIDD approaches will accelerate the exploitation of these findings in several ways (Fig. 1C).

• 1.Pocket-specific virtual screening – Structure-based deep learning models can rapidly screen ultra-large chemical libraries (billions of compounds) against the allosteric pocket, prioritizing candidates with favorable binding poses, hydrophobic complementarity, and hydrogen-bond networks.
• 2.Generative molecule design – AI-driven generative chemistry platforms (e.g., diffusion models, graph neural networks) can create novel scaffolds tailored to the shape and electrostatics of the allosteric site, enabling the discovery of chemical matter inaccessible through traditional screening.
• 3.Multi-parameter optimization – Machine learning models can iteratively refine lead compounds to balance potency, selectivity against host polymerases, pharmacokinetic stability, and low toxicity, using both in silico predictions and experimental feedback loops.
• 4.Cross-viral prediction – By integrating structural comparisons of allosteric pockets across nsNSVs, AI algorithms can predict which viral polymerases are most likely to be susceptible to a given scaffold, informing broad-spectrum or virus-specific drug design strategies.
• 5.Resistance forecasting – AI models trained on mutational landscapes can anticipate resistance-conferring substitutions in the allosteric pocket, enabling the design of inhibitors less prone to viral escape.

After identifying inhibitors that target the allosteric pockets in the MeV and NiV L-P complexes, subsequent validation will require in vitro biochemical assays, in vivo studies in animal models, preclinical evaluations, and ultimately clinical trials to establish their potential as antiviral drugs (Fig. 1C). By merging high-resolution structural biology with advanced AI modeling, it should be possible to rapidly progress from structural discovery to clinical candidate selection, with the potential to shorten the drug development timeline from years to months. In the context of nsNSVs, where outbreaks can be sudden and severe, such integration could be transformative for pandemic preparedness.

CRediT authorship contribution statement

Yiru Wang: Methodology, Writing – original draft. Lixia Zhao: Methodology, Writing – original draft. Heqiao Zhang: Conceptualization, Formal analysis, Investigation, Methodology, Project administration, Resources, Validation, Visualization, Writing – original draft, Writing – review & editing.

Ethical approval

This study does not contain any studies with human or animal subjects performed by any of the authors.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.