Predicting antifolate resistance in the unculturable fungal pathogen Pneumocystis jirovecii

Pneumocystis jirovecii is an opportunistic fungal pathogen responsible for Pneumocystis pneumonia (PCP) in immunocompromised patients. Antifolate drugs targeting the dihydrofolate reductase (DHFR), including trimethoprim (TMP), remain central to treatment, but studying the effects of mutations in DHFR on resistance to treatment is limited by our inability to culture this organism in vitro or in animal models. We expressed P. jirovecii DHFR (PjDHFR) in Saccharomyces cerevisiae and performed deep mutational scanning (DMS) on this protein to measure the effects of all single amino-acid substitutions on enzyme function and resistance to methotrexate (MTX), a model antifolate which shares structural features with TMP. We integrated experimental results with structural and evolutionary features from multiple biophysical modeling approaches, and by using an interpretable machine-learning framework, we trained a random forest model to classify MTX resistance-conferring mutations in PjDHFR. We then leveraged this framework as a prediction tool to model the effects of mutations on resistance to TMP, which cannot be directly assayed experimentally. Functional measurements from DMS were the strongest contributors to resistance prediction and generally outperformed purely computational features. Resistance-conferring mutations were constrained by function, revealing a functional-resistance trade-off within this essential protein. Feature contribution analyses highlighted key predictors such as distance to ligand, flexibility, stability, and functional trade-off as determinants of resistance. When extrapolated to TMP, the model identified candidate resistance mutations consistent with known biochemical constraints of DHFR. We demonstrate how experimentally measured functional landscapes can be combined with biophysical modeling to help understand and predict antifolate resistance in an unculturable fungal pathogen. Our results provide biological insight into the constraints affecting the evolution of resistance in PjDHFR, and support that resistance arises from mutations altering drug interactions while preserving function. We illustrate how DMS data can enable generalizable, mechanistically interpretable models of drug resistance across structurally related antifolates.