Deep mutational scanning of Pneumocystis jirovecii dihydrofolate reductase reveals allosteric mechanism of resistance to an antifolate

Abstract

Pneumocystis jirovecii is a fungal pathogen that causes pneumocystis pneumonia, a disease that mainly affects immunocompromised individuals. This fungus has historically been hard to study because of our inability to grow it in vitro. One of the main drug targets in P. jirovecii is its dihydrofolate reductase (PjDHFR). Here, by using functional complementation of the baker's yeast ortholog, we show that PjDHFR can be inhibited by the antifolate methotrexate in a dose-dependent manner. Using deep mutational scanning of PjDHFR, we identify mutations conferring resistance to methotrexate. Thirty-one sites spanning the protein have at least one mutation that leads to resistance, for a total of 355 high-confidence resistance mutations. Most resistance-inducing mutations are found inside the active site, and many are structurally equivalent to mutations known to lead to resistance to different antifolates in other organisms. Some sites show specific resistance mutations, where only a single substitution confers resistance, whereas others are more permissive, as several substitutions at these sites confer resistance. Surprisingly, one of the permissive sites (F199) is without direct contact to either ligand or cofactor, suggesting that it acts through an allosteric mechanism. Modeling changes in binding energy between F199 mutants and drug shows that most mutations destabilize interactions between the protein and the drug. This evidence points towards a more important role of this position in resistance than previously estimated and highlights potential unknown allosteric mechanisms of resistance to antifolate in DHFRs. Our results offer unprecedented resources for the interpretation of mutation effects in the main drug target of an uncultivable fungal pathogen.

Fig 1. Experimental setup for complementation assay using PjDHFR and dose response curves.

a) Experimental setup and strain construction. Yeast strain FDR0001 harbors the dfrB1 gene under the regulation of a β-estradiol inducible promoter (Galpr) [75], and a dfr1 deletion. dfrb1 is insensitive to MTX, therefore all strains can grow on MTX-containing media in the presence of β-estradiol. This panel was made with BioRender. b) Spot-dilution assays to test for DHFR complementation and for MTX sensitivity. PjDHFR, ScDHFR and mDHFR are from P. jirovecii (GenBank: ABB84736.1, codon optimized for S. cerevisiae), S. cerevisiae BY4741 DFR1, and M. musculus mDHFR (synthetic, L22F and F31S, from [77]), respectively. ScDHFR was used as a complementation control, and mDHFR as a MTX-resistant control. When β-estradiol is added to the media, dfrB1 is expressed, which is necessary for cell growth in a dfr1Δ S. cerevisiae background without DHFR on the vector. c) Growth curves of yeast strain IGA130 (wild-type for DFR1) each expressing a different DHFR. Shaded areas represent confidence intervals across biological triplicates. Empty vector shows sensitivity to low concentrations of MTX. Strains with vector-expressed DHFRs show increased resistance, with PjDHFR being the most sensitive, followed by ScDHFR, and mDHFR(L22F/F31S) being fully resistant. d) Growth coefficient (measure from growth rate relative to Empty without selection) of strain IGA130 with different DHFRs. R² for Hill equation fit: Empty = 0.96, PjDHFR = 0.98, ScDHFR = 0.91 and mDHFR = 0.

Fig 2. Effects of amino acid substitutions on the resistance of PjDHFR to methotrexate.

a) Comparison of selection coefficients between control condition (DMSO) and at IC75 (Spearman’s ρ = 0.1, p = 6.49e-10), b) DMSO and IC90 (Spearman’s ρ = 0.14, p = 1.02e-18), and c) IC75 and IC90 (Spearman’s ρ = 0.86, p < 1e-100). The different types of variants found in the library are marked. Nonsense mutants have premature stop codons. Synonymous are mutants that have different codon sequences but that have the same amino acid sequences as the wild-type sequence. Non-synonymous are amino acid changes relative to wild-type PjDHFR. Resistant (grey/dotted line) and very resistant (black/dashed line) thresholds are shown for both MTX conditions (See Statistical analysis of selection coefficients in methods for threshold determination, S1 Fig). d) Selection coefficient of a given amino acid (y-axis) at a given position (x-axis) at a MTX concentration corresponding to 90% of growth inhibition (IC90). Red points show the wild-type sequence of PjDHFR. e) Positions of contacting residues along PjDHFR modeled on structural alignments between PjDHFR and orthologous P. carinii DHFR (PDB: 3CD2 (MTX and NADPH) and 4CD2 (DHF)). Contact was established as amino acids with an α carbon located less than 8 Å from MTX, DHF or NADPH. Detailed heatmaps for DMSO and IC75 are available in S5 Fig.

Fig 3. Validation of pooled competition assay with individually reconstructed mutants.

a) Selection coefficient at IC75 and IC90 for selected mutants. Variants at positions showing large deviations between IC75 and IC90 are visible in the bottom right quadrant (gray lines, resistant threshold, black lines, very resistant thresholds, from gaussian mixture models, S1 Fig). Wild-type PjDHFR (blue X) is at 0.0 selection coefficient. Error bars in x represent mutant selection coefficients across synonymous codons and replicates at IC90 MTX and in y, at IC75 MTX. b) Growth rate (OD₆₀₀·h^-1) compared to selection coefficient in IC90 for validation mutants. The blue line represents the derivative growth rate for the strain with ScDHFR on the plasmid and the red line represents the growth rate for the strain with the Empty plasmid (negative control). Error bars in x represent mutant selection coefficients across synonymous codons and replicates, and error bars in y represent growth rate measurements in triplicate. Suspected outliers are outlined in green. c) Growth rate compared to selection coefficient in DMSO for selected mutants. All mutants grew at similar rates in this condition. All selection coefficient axis limits are scaled to ease comparison between panels a), b) and c). Individual growth curves are shown in S7 Fig. Statistical test in a), b) and c) is Spearman’s rank correlation. d) Spot-dilution assay for controls and some resistant mutants with their respective selection coefficients in IC90. ScDHFR and Empty are used as positive and negative controls, respectively.

Fig 4. Selection coefficients mapped onto PjDHFR structure in complex with methotrexate and NADPH.

a) Maximum selection coefficient from IC90 and b) minimum selection coefficient from IC75 observed for amino acid substitutions mapped on the PjDHFR structure by position. MTX (yellow) and NADPH (green) are visible. Nonsense mutations were not considered when identifying minimum selection coefficients. c) Structure superposition of PjDHFR (blue/red scale) and E. coli FolA (gray/black scale). FolA is colored according to predicted allostery from [67], with darker shades of gray representing increased confidence in predicted allosteric score. PjDHFR is colored by maximum selection coefficient in IC90. Orientations of superimposed structures are the same as panel A. Positions 199 and 142 are predicted to have allosteric effects on the substrate and cofactor binding pocket [67]. I10 is not predicted to have an allosteric effect. Side chains for these positions, and corresponding in E. coli., are shown and labeled. d) Rosetta FlexddG predicts that mutations at position 199 would destabilize interactions between PjDHFR and MTX, and this effect correlates with the measured selection coefficient. Spearman’s rank coefficient and Persons’ correlation coefficient were used to measure correlation. Y axis units are REU (Rosetta Energy Units).