Testing and predicting the effects of mutations on the dihydrofolate reductase of an unculturable fungal pathogen

ABSTRACT

Investigating how protein evolution is shaped by selective pressures is critical for understanding the mechanisms underlying antifungal resistance. Here, we explore the molecular evolution and functional landscape of dihydrofolate reductase (DHFR), one of the main drug targets in *Pneumocystis jirovecii*, an opportunistic fungal pathogen responsible for *Pneumocystis* pneumonia (PCP). *P. jirovecii* cannot be grown in the lab or in animal models, being an obligate commensal of human lungs. In recent years, resistance to treatment has become more prevalent in clinical settings, but the mechanistic basis of resistance in this unculturable organism remains poorly investigated and understood. To address this important gap in our knowledge, we use a combination of deep mutational scanning (DMS), structural modelling, and evolutionary analyses to systematically test the effects of all possible single amino acid mutants in *P. jirovecii's* DHFR on protein function and resistance to the antifolate methotrexate (MTX). Then, by using machine learning (ML) approaches, we investigate the biochemical features most contributing to MTX resistance and extrapolate these results to make predictions on the effect of mutations on trimethoprim (TMP) resistance, another related antifolate. In chapter one, we describe the use of a heterologous expression system in *Saccharomyces cerevisiae* to functionally complement its endogenous DHFR with *P. jirovecii* DHFR. Using DMS and growth-based selection in the presence of MTX, we quantify the effects of over 4000 amino acid variants on MTX resistance. This comprehensive fitness landscape revealed that resistance-conferring mutations cluster in active site regions and uncovers evidence of an allosteric mechanism of MTX resistance that had not been previously described. These findings highlight the role of biophysical constraints in shaping available evolutionary paths to antifolate resistance in this protein. In chapter two, by standing on the insights gained from chapter one, we develop a framework that integrates experimental fitness data, structural metrics, and ML to predict resistance to TMP, a clinically relevant antifolate, as *in vivo* testing in *P. jirovecii* is impossible. We demonstrate that the predictive model accurately identifies known resistance mutations and uncovers previously uncharacterized variants with the potential to lead to resistance based on important features in MTX resistance. To validate model predictions, we sequenced the largest to date dataset of PjDHFR from clinical samples isolated in eastern Canada. Together, these chapters provide a quantitative map of the functional and evolutionary constraints acting on the DHFR of *P. jirovecii* and demonstrate how DMS and ML can help uncover principles driving antifungal resistance in this unculturable pathogen. This work contributes to a deeper understanding of the molecular evolution of antifolate resistance and provides a framework for predictive models to infer resistance trajectories in clinically relevant pathogens.