Mapping the 3D Space of drug resistance variants

This thesis aimed to advance the investigation of the impact of protein-coding variants on protein structures. Using 3D representations of proteins, we identify the respective mutation site and from the structural environment we deduce parameters to describe the effect a specific amino acid might have in this chemical environment. The interpretation of the variant is complemented by the aggregation of conservation-based features, physicochemical features, functional annotations, solvent accessibility calculations and the estimation of the free energy difference upon mutation. By combining these features with the visual inspection of the protein structure environment of the site, a well-grounded hypothesis can be constructed on the impact the variant might have on its immediate structural environment. This process was streamlined and upscaled by the creation of the Var3D variant analysis engine.
This analysis workflow was applied to facilitate the structure-based variant interpretation of antibiotic resistance variants in MTB. The release of a resistance variant catalogue by the WHO and the publication of the revolutionary structure prediction method AlphaFold2 made it possible to obtain protein structure models for all antibiotic resistance target proteins for a comprehensive view of the MTB resistome. The mutation catalogue showed that 90% of the variants were still annotated as “uncertain”, demonstrating the need for computational tools to aid in their further characterization. Combining Var3D with the SWISS-MODEL technology stack and a manually curated structure data set of resistance targets, we were able to create the TBvar3D web server. The server facilitates the inspection of variants in a data-rich 3D context which would otherwise require manual time-consuming structure modelling steps, data integration and visualisation. The results are displayed on the web server and do not require the specific computational expertise of the user. This makes TBvar3D a valuable tool to assist researchers worldwide to form compelling hypotheses on the impact of variants for MTB.
The research goal of the second project in this thesis is the identification of naturally occurring human polymorphisms in the interfaces of antibody therapeutics and their respective antigen targets which may impact antibody binding. Individual cases of natural variants preventing epitope recognition were reported in the literature, but a comprehensive investigation was never performed before. With more than 100 antibody therapeutics approved and far more therapeutics undergoing clinical trials, the investigation of the potential impact this phenomenon might have on the efficacy of antibody drugs becomes pertinent. Through the use of Var3D, it was possible to map and annotate around 25’000 human variants on over 100 structure models of drug target complexes representing 98 therapeutics. We identified around 1’400 naturally occurring polymorphisms distributed across every single epitope, showing the phenomenon of natural polymorphisms occurring on clinically relevant epitopes to be quite common. Among this variant data set, we are planning to experimentally characterise the impact variants located on 10 prolific antibody therapeutics.