Two recent studies from David Horn’s lab at the School of Life Sciences, University of Dundee,in colaboration with Drug Discovery Unit, University of Dundee published in Nature Communications (https://www.nature.com/articles/s41467-026-69187-5) and PLOS Pathogens (https://doi.org/10.1371/journal.ppat.1013764), explore exactly that question in Trypanosoma brucei, the parasite responsible for sleeping sickness.
Mapping Resistance Space Experimentally
Using multiplex oligo targeting - a genome editing approach pioneered and developed in David Horn’s lab - they introduced hundreds of possible mutations directly into native drug target genes. This method avoids overexpression systems and preserves physiological regulation, allowing resistance to be assessed in a biologically meaningful context.
In the proteasome study (Nature Communications), they made saturation-edited residues surrounding a drug-binding pocket and quantified both drug resistance and fitness effects. Although mutatio-s were theoretically possible, only a limited subset both disrupted drug binding and preserved proteasome function.
In the CPSF3 study (PLOS Pathogens), they investigated the molecular target of acoziborole, a promising single-dose oral treatment for sleeping sickness. Here, resistance proved even more constrained. Only specific combinatorial mutations conferred strong resistance, highlighting tight structural and functional limits around the binding site.
These findings reinforce an important principle:
Resistance is not random - it is constrained by fitness and structural physics.
The Computational Perspective
In the studies mentioned above, our team at DDU integrated multiple computational layers to understand resistance mechanisms in depth: Molecular dynamics (MD) simulations to characterise ligand–protein interactions, quantitative estimation of mutation-induced changes in binding affinity, analysis at atomic level to understand the impact on ligand-protein interactions, DNA language models to estimate mutation-associated fitness costs.
By integrating these approaches, we were able to mechanistically explain resistance hotspots identified experimentally.
For instance, some mutations introduced increased steric bulk near the ligand-binding site, destabilising binding. Others disrupted key hydrogen-bond networks or altered hydrophobic packing, ultimately reducing ligand affinity without compromising overall protein stability
Importantly, our computational predictions closely aligned with in-cellulo mutational profiling data. Mutations predicted to significantly reduce ligand binding while preserving structural integrity corresponded to experimentally observed resistance variants.
This synergy between experiment and computation is particularly powerful. High-resolution mutational datasets provide essential ground truth for validating models. In turn, computational modelling helps uncover the underlying molecular mechanisms and anticipate which resistance mutations are most likely to emerge within accessible evolutionary space.
Why This Matters for Drug Discovery
Antimicrobial resistance is often treated as an unpredictable outcome. But these studies suggest that resistance landscapes can be mapped systematically and predicted mechanistically.
By integrating, precision genome editing, deep mutational scanning, structural modelling, physics-based binding calculations and AI-driven sequence modelling, we can move toward a predictive framework for drug durability.
Rather than reacting to resistance after clinical failure, we can proactively design compounds that target regions of highly constrained fitness space - areas where resistance mutations are either structurally impossible or biologically costly.
Looking Forward
The future of drug discovery lies at the intersection of experimental biology and computational modelling. As structural data, mutational datasets, and machine learning models become more integrated, we will increasingly be able to anticipate resistance pathways before they emerge. Also, we can integrate more accurate calculations like FEP as we did in a previous work (https://www.nature.com/articles/s42004-025-01869-5) and where Julien Michel's lab is working actively (https://pubs.acs.org/doi/full/10.1021/acs.jctc.5c01648) in order to map more accurately the impact of mutations in binding.
For me, working at this interface - where computational modelling meets evolutionary constraint and experimental validation - highlights how molecular modelling can meaningfully guide therapeutic strategy.
The goal is not just to make effective drugs - but durable ones.
#DrugDiscovery, #ComputationalBiology, #AntimicrobialResistance, #MolecularDynamics, #AIinScience #StructuralBiology
Acknowledgments
David Horn, Simone Altmann, PhD , Melanie Ridgway, Markéta Novotná, Michael Thomas , Manu De Rycker,Michele Tinti , Jagmohan Saini, PhD , Gabriele Dalla Torre, Jed Hawes, Graeme Sloan, Julien Michel, Audrius Kalpokas, Peter E. G. F. Ibrahim,Salomé Llabrés Prat , Arun Gupta, Michael Bodkin