Strains, plasmids, and culture media

All strains used in this study were constructed from the reference strain BY4741 as an initial parent strain [54]. All transformations were conducted using the standard lithium acetate transformation protocol [55]. All PCR reactions were conducted using KAPA HiFi Hotstart DNA polymerase (Sigma-Aldrich, catalog #BK1000) unless stated otherwise. Details for all strains are available in S1 Table, plasmids in S2 Table, oligonucleotides in S3 Table, media in S4 Table and PCR reactions in S5 Table.

To test if PjDHFR is functional in S. cerevisiae, we had to inactivate or delete the yeast’s endogenous DHFR, DFR1. Since it is an essential gene, to be able to grow a dfr1 deletion strain, we had to first insert another DHFR (dfrB1) which would be expressed under the control of a β-estradiol induced promoter (GAL1pr) [75]. GAL1pr allows for the expression of the downstream gene in the presence of β-estradiol (Sigma-Aldrich, catalog #E2758-5G). Construction of strain IGA130 (his3 ura3 met15 LEU2::GEM, GAL1pr::dfrB1-NATMX) was done by inserting the dfrB1 gene (previously known as R67—yeast codon optimized sequence) at the GAL1 locus in strain AKD0679 (LEU2::GEM, GAL1pr::Stuffer-NATMX) by CRISPR-Cas9-mediated integration following the protocol detailed in [56] and using pCas plasmid with a gRNA sequence targeting the Stuffer from [57]. Transformants were selected on YPD + Geneticin 200 μg/mL (G418) + Nourseothricin 100 μg/mL (NAT). Following the loss of pCAS plasmid, the insertion of dfrB1 in the genome was verified by colony PCR using oligos OP151-E5 and ADHterm_R with BioShop Taq DNA Polymerase and validated by Sanger sequencing. Construction of strain FDR0001 (his3 ura3 met15 LEU2::GEM, GAL1pr::dfrB1-NATMX dfr1Δ ::Stuffer) was done using a similar CRISPR-Cas9-mediated protocol as described above, but using a Stuffer sequence and a gRNA targeting DFR1. Oligos DFR1_KO_F and DFR1_KO_R were used to amplify the Stuffer sequence used to knockout DFR1. Transformants were selected on YPD + G418 + NAT + 100 nM β-estradiol. Following the loss of pCAS plasmid, deletions were confirmed by PCR using oligos DFR1_KO_valid_F and DFR1_KO_valid_R and Sanger sequencing (BioShop Taq DNA Polymerase). Detailed PCR protocols are available in S5 Table. Details for media composition are available in S4 Table.

All yeast experiments were conducted at 30°C for 48 h unless stated otherwise. All experiments involving bacteria were conducted at 37°C using 2YT with appropriate antibiotics unless stated otherwise. All liquid cultures were incubated with 250 rpm orbital agitation. All Sanger sequencing was performed by the Sequencing Platform of the Centre Hospitalier de l’Université Laval (Université Laval, Québec, Canada), and all next-generation sequencing was performed using the MiSeq Reagent Kit v3 on an Illumina MiSeq for 600 cycles at the Sequencing Platform of the Institut de Biologie Intégrative et des Systèmes (Université Laval, Québec, Canada).

Plasmid pAG416-GPD-PjDHFR was constructed from pAG416-GPD-ccdB and Pneumocystis jirovecii DHFR (PjDHFR) (UniPROT: A0EPZ9), codon-optimized using GenScript Biotech’s GenSmart^™ Codon Optimization tool, and ordered from Twist Bioscience (San Francisco, California, USA) as a Gene Fragment. pAG416-GPD-ccdB was digested using XbaI and HindIII-HF in CutSmart buffer (New England Biolabs, catalog #R0145 and #R0104), according to manufacturer’s instructions, as the backbone. PjDHFR was amplified by PCR using oligos Pj_to_pAG_F and Pj_to_pAG_R under parameters specified in S5 Table, as the insert. Backbone and insert were assembled using an in-house Gibson assembly mix (1:2 vector:insert ratio) and cloned into E. coli MC1061 ([araD139]B/r Δ(araA-leu)7697 ΔlacX74 galK16 galE15(GalS) λ-e14-mcrA0 relA1 rpsL150(strR) spoT1 mcrB1 hsdR2) using a standard heat shock transformation method and plated onto ampicillin-containing 2YT media (100 μg/mL). Clones were confirmed by colony PCR using oligos pAG_insert_F and pAG_insert_R, and the amplicon sequence was confirmed by Sanger sequencing. The same protocol was used to construct pAG416-GPD-ScDHFR (DFR1 insert amplified from BY4741 genomic DNA using oligos Sc_to_pAG_F and Sc_to_pAG_R) and pAG416-GPD-mDHFR (mDHFR, synthetic, L22F and F31S, insert amplified from pGEST-linker-fulllength-DHFR-HYG_B plasmid using oligos Murine_to_pAG_F and Murine_to_pAG_R).

Functional complementation assay

The complementation assay was conducted by transforming the DHFR-containing plasmids into strain FDR0001 and plating transformations on solid SC-URA+MSG+100 nM β-estradiol media to ensure growth of transformants even in the absence of complementation. Four individual colonies were harvested from each transformation and streaked on solid SC-URA+MSG+100 nM β-estradiol media, and colony PCR was conducted to confirm transformation success. For both spot-dilution assays, transformants were grown in liquid SC-URA+MSG+100 nM β-estradiol media overnight. Cultures were diluted to 0.1 OD₆₀₀ in the same media and grown to 0.3 OD₆₀₀ to ensure that cells were in exponential phase. 1.0 OD₆₀₀ unit worth of cells were harvested from the cultures, washed, and resuspended in ddH₂0, and 5 μL was spotted on MTX media (See S4 Table for detailed recipe) with and without 440 μM methotrexate (BioShop, catalog #MTX440) and with and without 100 nM β-estradiol. The following spots were done by five-fold serial dilution on the same media.

Dose response curves

Dose response curves were measured in triplicate using a Tecan Infinite M Nano (Tecan Life Sciences) plate reader. DHFR-containing plasmids were transformed into strain IGA130. Three individual colonies were sampled from transformation plates and grown individually in 3 mL SC-URA+MSG in V-bottom 24-well deep well plates overnight. The next morning, 1.0 OD₆₀₀ unit of cells was harvested and spun down at ~500 rcf for 4 minutes. The supernatant was removed, and the pellet washed with sterile ddH₂0, and then resuspended in 1 mL of sterile ddH₂0. 20 μL of this resuspension was added in a flat bottom Greiner 96 well plate, to 180 μL of liquid MTX media prepared the same day, for a final OD₆₀₀ of 0.1. Liquid MTX media was prepared using a ten-fold dilution of MTX media at 440 μM MTX to 44 μM, and then subsequent two-fold dilutions to attain 0.086 μM of MTX. Plates’ edges were sealed with Parafilm to limit media evaporation and placed into plate readers at 30°C for 48h without agitation. OD₆₀₀ was read every 15 minutes at four points within the well, and mean OD₆₀₀ was used as a reading.

The growth rate for strain IGA130 with a plasmid-encoded DHFR was calculated from the slope at every time point before 40h, and taking the median of the five highest values for a more robust estimate. This was done for every replicate. Growth coefficient was measured for each replicate using the following equation (Eq 1), with the derivative growth rate of strain IGA130 with plasmid pAG416-GPD-Empty (Empty) without MTX as reference for maximum growth rate: (1)

Inhibitory concentrations (ICs) were computed using the standard Hill equation, using Python’s SciPy optimize.curve_fit method (SciPy version 1.7.3).

Deep mutational scanning library preparation

A deep mutational scanning library of PjDHFR was constructed using the previously described megaprimer method [58,59]. Primers with NNK degenerate sequence at the codon of interest were ordered from Eurofins Genomics for all positions along PjDHFR, except for the start and stop codons. By using these primers with reverse primers Mega_1_R, Mega_2_R and Mega_3_R, for fragments 1, 2 and 3, respectively, megaprimers were generated by PCR using KAPA Hi-Fi polymerase and pAG416-GPD-PjDHFR as template on a per-position basis. This PCR product was diluted 1/2500 with two serial 50-fold dilutions, and 2 μL of these dilutions were used as megaprimers for the amplification of pAG416-GDP-PjDHFR using KAPA Hi-Fi polymerase. PCR products were digested using 0.2 μL of DpnI restriction enzyme (New England Biolabs, catalog #R0176L) (37°C for 2 h) to remove parental DNA. This product was then transformed in E. coli MC1061, and colonies were counted. Biomass from petri dishes with >400 colonies was harvested (>10x coverage for each codon), and plasmid libraries extracted per position using the FroggaBio Presto Mini Plasmid Kit (catalog #PDH300.) Each plasmid positional library was quantified using a ThermoScientific Nanodrop 2000 spectrophotometer and library content was assessed for diversity using the RC-PCR method from [60] with primers Lib_seq_F1_F/R, Lib_seq_F2_F/R and Lib_seq_F3_F/R for the different fragments. Positions where diversity was too low were redone using the same protocol. All oligos used for library generation are available in S3 Table.

Once diversity was confirmed at every position, plasmid libraries were transformed using the same method as above into IGA130 and plated on SC-URA+MSG media. Biomass from Petri dishes with >1000 colonies were harvested, and glycerol stock cultures were prepared for each position. Once every position was transformed in yeast and appropriately frozen as glycerol stocks, 100 μL of each glycerol stock was added to individual 5 mL culture tubes of Sc-URA+MSG liquid media and grown overnight. OD₆₀₀ was measured for each position, and an equal number of cells from each position was mixed as a master pool for each fragment. A glycerol stock was prepared for each fragment.

Pooled competition assay

Starting from these master stocks, 100 μL were added to 5 mL of SC-URA+MSG media and grown overnight in triplicate. These cultures were considered as T₀ samples. OD₆₀₀ was measured, and 1.5 OD₆₀₀ units were spun down at 224 rcf for 3 minutes and resuspended in 1 mL of ddH₂0. 14 mL of 1.07x MTX media was inoculated with 1 mL of 1.5 OD₆₀₀ culture, to yield 1.0x media at 0.1 OD₆₀₀, in triplicate. Depending on the condition, these cultures had either 0 μM (DMSO), 8.8 μM (IC 75) or 44 μM (IC90) of MTX. DMSO was used as the drugless control, as it is the solvent for MTX. Once cultures reached either 1.2 OD₆₀₀ or were grown for 72 h, cultures were diluted back to 0.1 OD₆₀₀ in a fresh batch of the same media using the same method. Once grown, these culturse were considered as T_final, and cells were harvested once cultures reached either 1.2 OD₆₀₀ or 72 h of growth in this second passage. At every passage and for every culture, triplicate glycerol stocks were made, and 5.0 OD₆₀₀ units were harvested and spun down at 224 rcf for 3 minutes. Supernatant was removed and pellets were frozen for downstream DNA extraction. DNA extractions were conducted using Zymoprep Yeast Plasmid Miniprep II (Zymo Research, catalog #D2004). The coding sequence of PjDHFR was split into three fragments of roughly equal length (F1, positions 2 to 73, F2, 74 to 145, and F3, 146 to 206) to allow Illumina MiSeq 300 bp paired-end sequencing. Extracted libraries were amplified by fragments using Lib_seq_F/R primers using a cycle described in S5 Table, purified using magnetic beads (AMPureXP from Beckman Coulter Life Sciences, catalog #A63882) and sent for sequencing for a total of 15M reads.

Next-generation sequencing read processing

Reads were analyzed using custom Python and Bash scripts. Briefly, demultiplexed R1 and R2 reads were trimmed to remove primer sequences and merged using PANDAseq [61]. Merged reads were aggregated and counted using VSEARCH [62]. Using custom Python scripts, expected codon mutations were matched with read counts, and the frequency of each mutant was assessed in each individual culture. To minimize sampling bias, we ensured that every retained mutant had high initial coverage in T₀ to minimize bias in downstream analysis and discarded all mutants that had an initial coverage of less than 15 high quality reads. Synonymous codons to wild-type PjDHFR were used as internal control and reference for normalization between fragments. The selection coefficient was calculated according to Eq 2, where the normalized log2-fold change of the mutants between T₀ and T_final was divided by the median normalized log2-fold change of the synonymous mutants and normalized by the number of mitotic generations measured for each replicate (g). The final selection coefficient value for a mutant was computed using the median selection coefficient between all synonymous mutants within each individual culture.

(2)

This normalization allowed us to generate a standardized score where 0 is the median score of synonymous mutants, which gives an accurate estimate of the wild-type fitness in the tested conditions and of the relative performance of all mutants in comparison to wild-type PjDHFR in every condition. The resulting distribution of selection coefficients was used in downstream analyses.

Statistical analysis of selection coefficients and false-discovery rate

To assess resistance thresholds using statistical means, we used the sklearn mixture package (version 1.0.2) [63] to estimate the fraction of neutral mutations, resistant and very resistant mutations by fitting the distribution of selection coefficients under MTX selection to a mixture model of underlying Gaussian distributions. Using custom scripts, we estimated how many parameters recapitulated the underlying function the best by minimizing Akaike information criterion (AIC) and Bayesian information criterion (BIC) values for the gaussian mixture models using an expectation-maximization model. AIC and BIC are metrics to minimize to favor the simplest model to maximize fitting. By using the best fit model, we separated our distributions of selection coefficients into their underlying Gaussian mixture models, and set our thresholds based on parameters from the two rightmost distributions (most resistant mutants and second-to most resistant mutants). To minimize overlap effects between underlying Gaussian distributions, we chose the median values of the second-to-highest underlying distribution as the resistant threshold, and the intersection second-to-highest and highest as the very resistant threshold in each condition where the drug was present (S1 Fig).

To correct for experimental noise and false discovery rate, we conducted a one sided Welch’s t-test (scipy.stats version 1.7.3) [64]. We generated a matrix for each mutation at each position, which regrouped values for all replicates and synonymous mutations for a specific amino acid substitution and compared it to the distribution of silent mutations to assess if the selection coefficient of mutants with a substitution was significantly greater (i.e. more resistant) than the distribution of silent mutations. We then used the Benjamini-Hochberg method to control for the false discovery rate of significantly resistant mutants at 5% (statsmodels version 0.13.5) [65]. This allowed us to identify a subset of mutants that have selection coefficients significantly greater than the wild-type, which we classify as being resistant, and corrected for false discovery rate. All custom code is available at https://github.com/Landrylab/Rouleau_et_al_2023.git.

Construction of validation mutants

To validate results from the pooled competition assay, we selected different mutants based on their scores, patterns of resistance, and reproducibility across replicates and conditions, and reconstructed them individually (S6 Table). To generate these mutants, identified by their specific reconstructed mutation in S2 Table, we used fusion PCR followed by Gibson assembly. First, by using oligonucleotides containing the mutation of interest that are a reverse complement to each other, we generated the two fusion PCR fragments using Valid_master_F and any R_valid oligo, and Valid_master_R and any F_valid oligo. pAG416-GPD-PjDHFR was used as the template vector. PCR products were mixed and diluted 1/2500, and used as a template with Valid_master_F and Valid_master_R oligos for the generation of the final product. The fusion PCR product was then used in Gibson cloning, using the same method and backbone as described above, and clones were validated using Sanger sequencing. Constructions were transformed into the relevant strain (IGA130 or FDR0001) and plated on SC-URA+MSG +/- 100 nM β-estradiol depending on the strain. Individual growth curves and growth rates were measured as described above. Subsequent spot-dilution assays were conducted as described above, but with 44 μM of MTX instead of 440 μM.

Western blot and flow cytometry analysis

For western blot analysis, different variants of PjDHFR and ScDHFR were tagged using either 3xFLAG tag in C-terminal (3’) or 1xFLAG tag in N-terminal (5’). For N-terminal constructions, pAG416-GPD-PjDHFR (wild-type, validation mutants pAG416-GPD-PjDHFR-M33P and pAG416-GPD-PjDHFR-F199P) or pAG416-GPD-ScDHFR were digested using ClaI enzyme in rCutSmart buffer (New England Biolabs, catalog #R0197) according to manufacturer’s instructions, and purified using magnetic beads as described above. 3xFLAG cassettes were amplified using oligos C3xFLAG_F and _R from p3-3xFLAG-KanMX. The PCR product was purified using magnetic beads. Gibson assembly was conducted as described above and confirmed by Sanger sequencing using pAG_insert_F. For C-terminal tagging, 1xFLAG cassette was amplified from p3-3xFLAG-KanMX using N1xFLAG_F1_F and N1xFLAG_F1_R_nomet or _met, to remove or keep the PjDHFR start codon, respectively. Then, PjDHFR was amplified from pAG416-GPD-PjDHFR using N1xFLAG_F2_F_nomet or _met and N1xFLAG_F2_R, to remove or keep the PjDHFR start codon, respectively. These fragments were assembled using fusion PCR, as described above, and assembled into a digested (HindIII, rCutSmart, New England Biolabs, catalog #R0104) pAG416-GPD-ccdB using in-house Gibson assembly. All constructions were confirmed by Sanger sequencing using pAG_insert_F. All constructions were transformed in IGA130 and transformants were grown overnight in SC-URA+MSG+100 nM β-estradiol in triplicate. A schematic for all constructions is available in S2A Fig. The mEGFP tagged PjDHFR plasmids were constructed using the same procedure.

Yeast proteins were extracted by resuspending 12.5 OD₆₀₀ units in 250 μL of lysis buffer (Complete Mini Protease inhibitor cocktail, Millipore-Sigma 11836153001). Glass beads were added to the cell suspension and the tubes were vortexed for 5 min on a Turbomix. Following the vortexing step, 25 μL of SDS 10% were added to each tube before boiling them for 10 min. Once the tubes were boiled, they were centrifuged for 5 min at 16,000 rcf to clear the supernatant. Samples for migration were prepared by mixing 17.5 μL of cell extract with 2.5 μL of DTT 1 M and 5 μL of loading buffer 5X (250 mM Tris-Cl pH 6.8, 10% SDS, 0.5% Bromophenol Blue, 50% glycerol). Samples were migrated on a 15% SDS-PAGE acrylamide gel for 55 min, then transferred on a nitrocellulose membrane using a semi-dry system (TE 77 PWR, Amersham Bioscience) for 1h at 0.8 mA/cm². Following the transfer, membranes were soaked in a 1 mg/mL Ponceau red solution for 5 min, destained quickly with water and imaged. To remove remaining Ponceau staining, the membranes were shaken in PBS (10 mM sodium phosphate dibasic, 135 mM NaCl, 2 mM KCl, 1.5 mM monopotassium phosphate) solution for around 45 min. Membranes were then blocked overnight with agitation in blocking solution (Intercept (PBS) Blocking Buffer, Mandel Scientific, catalog #927–70003) at room temperature. The following morning, membranes were incubated with a primary antibody 0.15 μg/mL (Anti-Flag M2, F3165-1MG, Millipore-Sigma) in the blocking solution + 0.02% Tween 20 for 30 min. Membranes were washed 3 times for 5 min with PBS-T (PBS with 0.1% Tween 20). The secondary antibody was then added at 0.075 μg/mL (Anti-mouse 800, catalog #LIC-926-32210, Mandel Scientific) in the blocking solution + 0.02% Tween 20 for 30 min. Membranes were washed again 3 times for 5 min with PBS+T, followed by imaging with an Odyssey Fc instrument (Licor, Mandel Scientific) in channels 700 (30s) and 800 (120s).

Flow cytometry assay

Mutants of PjDHFR-mEGFP were transformed in to strain IGA130 following the protocol described above. Flow cytometry measurements were conducted in triplicate for each mutant and control strain, with 5000 events per replicate per mutant. This was done using Guava’s easyCyte BG HT with a 488nm excitation laser and a 525/530 nm filter. Fluorescence intensity was corrected for front scatter, a measure of cell size, to normalize fluorescence intensity per cell, and transformed into log10 scale for visualization.

Structural analysis, alignments, and allostery analysis

The structure of PjDHFR was recovered from AlphaFoldDB (UniProt: A0EPZ9). The crystal structures of DHFR ortholog from P. carinii, in complex with MTX/NADPH (PDB: 3CD2) and DHF/NADPH (PDB: 4CD2), were recovered from UniProt. Using ChimeraX’s (version 1.6) [66] matchmaker algorithm, the structures were aligned with each other, and ligands from crystal structures were extracted and added to PjDHFR AlphaFold structure. The resulting complexed structures were then relaxed using the FoldX (version 5.0) optimize command to remove van der Waals clashes. Using the MutateX protocol on the PjDHFR complex with MTX/NADPH and DHF/NADPH, changes in protein free energy were computed [86, 87].

For visual representation of selection coefficient on protein structure, chains were colored using the ChimeraX color byattribute command, and attributes were manually defined within custom.defattr files. Data for coloring E. coli DHFR according to allostery was recovered from S1 Table from [67], with darker colors corresponding to more stringent confidence intervals for allostery. Alignments between E. coli DHFR and PjDHFR were done using ChimeraX’s matchmaker.

To measure distance between protein residues and ligands, a custom script was used, based on tools from BIO.PDB (version 1.78) [68]. Briefly, by using a PDB file coordinate parser, distance was estimated between α carbons from all residues in PjDHFR and all atoms from the ligand of interest, except hydrogen atoms. Then, the minimal distance between the residue α carbon and atoms in the ligand of interest was considered for establishing contact. Any distance shorter than 8 Å was considered a contact.

Modeling of binding affinity changes

To model changes in binding affinity to ligands caused by mutations, we used the Rosetta FlexddG method [69]. Briefly, binding molecules were parameterized using Rosetta’s molfile_to_params.py script. In silico saturation mutagenesis was conducted using a modified version of the provided script from the FlexddG GitHub tutorial, and the generated params files as extra param files [69]. 100 structures with 35000 backrub steps were used as parameters, based on recommendations from [70], and all individual resulting structures were analyzed using a modified version of the provided script from the FlexddG GitHub tutorial. This resulted in a distribution of ddG values for 100 individual replicates. To minimize extreme effects caused by potential clashes or other artifacts, the median values from these distributions were used as the final ddG for downstream analyses.

PjDHFR can functionally complement the deletion of native DHFR, DFR1, in S. cerevisiae and is sensitive to methotrexate

In the budding yeast S. cerevisiae, the DHFR is encoded by the gene DFR1 (ScDHFR). DFR1 is an essential gene [71], which means that complementation assays require the use of an engineered strain. Some strains with DFR1 deletion or downregulation have been constructed, but they require multiple mutations, such as inactivation of TUP1, and specific growth media to be handled [12,72,73]. Here, we performed the full deletion of DFR1 in S. cerevisiae for functional complementation with PjDHFR. We engineered a strain in which the function of DFR1 is complemented by introducing another DHFR-coding gene (dfrB1) from bacteria, that is nonhomologous to DFR1 and that is insensitive to antifolates [74]. Upon induction of a β-estradiol inducible promoter, expression of this dfrB1 masks the deletion of DFR1, and allows for growth in the presence of MTX (Fig 1A and 1B) [75]. This strain, FDR0001 (dfr1Δ), is therefore non-viable in the absence of β-estradiol, but viable and MTX-resistant in the presence of β-estradiol.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pgen.1011252.g001

By transforming FDR0001 with a plasmid harboring PjDHFR under the regulation of the constitutive GPD promoter and by growing it in the absence of β-estradiol, we found that PjDHFR functionally complements the deletion of DFR1 in S. cerevisiae and rescues growth at a near wild-type level (Fig 1B). In the absence of β-estradiol, the strain without DHFR on the plasmid (Empty, negative control) was unable to grow, confirming that the phenotype observed depends on the presence of a complementing DHFR on the plasmid. Using media supplemented with MTX at a concentration known to fully inhibit DFR1 at its endogenous expression level (440 μM) [76], we used this same system to assess PjDHFR sensitivity to this antifolate. The MTX-resistant mouse DHFR (mDHFR) showed complementation and resistance to MTX, as expected [77]. Complementation of the inhibition of DFR1 by this mDHFR in S. cerevisiae and its associated resistance to MTX have been reported in several studies [76,77]. Interestingly, when using a plasmid harboring ScDHFR under the regulation of the GPD promoter as a positive control for complementation, we observed resistance to 440 μM of MTX on solid media. Since DFR1, at its native expression level and without vector-expressed DHFR (Empty) is sensitive to MTX, this experiment shows that high expression of this protein is sufficient to achieve resistance to MTX at these concentrations. Resistance caused by overexpression has previously been reported in E. coli for the antifolate TMP [78]. Despite being under the regulation of the same promoter, the PjDHFR-harboring strain was sensitive to MTX. Western blot results (S2B Fig), alongside complementation assays, show that PjDHFR is expressed and functional in S. cerevisiae, and that PjDHFR is more sensitive to this antifolate than wild-type DFR1, at least in this context. This highlights the potential for PjDHFR to evolve toward resistance to MTX and the usefulness of our system to identify resistance mutations.

To quantitatively assess the extent of inhibition and calculate inhibitory concentrations (IC) of PjDHFR in S. cerevisiae, we measured growth curves in the presence of decreasing concentrations of MTX. By comparing growth rates measured in the presence of different concentrations of the drugs to the growth from the Empty vector in the absence of the drug as growth coefficient, we estimated the sensitivity of strain IGA130, which still has wild-type DFR1 at the genomic locus. Here, we switched from strain FDR0001 (dfr1Δ) to the parent strain IGA130 (wild-type for DFR1) for downstream screening because FDR0001 had insufficient transformation efficiency for the construction of DMS libraries. Indeed, despite media supplementation with β-estradiol, strain FDR0001 shows growth defects when compared to strain IGA130, which reduces transformation efficiency. We believe that this growth defect creates a selection for fully functional PjDHFRs mutants, and that non-functional or less-functional mutants could be lost at this step. As we focus on mutations that confer resistance to MTX, we wanted to limit pre-screening selection on PjDHFR mutants.

In strain IGA130, mDHFR and Empty vector were used as positive and negative controls for MTX resistance, respectively (Fig 1C). None of the tested concentrations had a detectable effect on the growth rate of strains harboring MTX-resistant mDHFR. Empty plasmid showed the highest sensitivity, as only the natively expressed DFR1 was present in this strain. Using a Hill-equation curve fit, we calculated ICs for these different DHFRs (Fig 1D). IC50, when expressing no exogenous DHFR, was over 10 times lower than when expressing PjDHFR (0.14 μM and 2.01 μM, respectively), highlighting the sensitivity of the genomic DFR1 and the relatively limited impact of genomic DFR1 on resistance to MTX. The conditions that were selected for downstream competition assays correspond to roughly 75% and 90% inhibition for PjDHFR (IC75 = 8.8 μM MTX and IC90 = 44 μM MTX).

To ensure that the use of strain IGA130 (wild-type for DFR1) did not introduce a bias in selection strength, we conducted identical dose response curves in strain FDR0001 (dfr1Δ) and estimated its IC50 as well (S3A and S3B Fig). IC50 measured for strain FDR0001 is of 1.95 μM, validating results from strain IGA130 on the minimal effect of genomic DFR1 on MTX resistance. We believe that by using this system, we were able to identify a maximum number of mutations in PjDHFR that confer resistance to MTX, without introducing downstream bias for PjDHFR activity.

Pooled competition assays of PjDHFR mutant library identify mutations conferring resistance to methotrexate

Deep mutational scanning (DMS) was conducted on PjDHFR using the megaprimer method [59]. We generated all possible single amino acid changes of PjDHFR on the plasmid described above. To assess the fitness effect of mutations, we measured the change in frequency for every mutant between the initial culture (T₀) and the second passage in the presence of MTX (T_final). The pooled competition assay in DMSO was included to identify any mutations that could have dominant negative effects on growth and that could thus confound resistant or sensitive phenotypes (Fig 2A). IC75 and IC90 (Fig 2B and 2C) selection coefficients were used to identify mutations that cause resistance to MTX. The final count of mutants covered 97.2% of all possible amino acid replacements (4166/4284). We calculated a selection coefficient for each mutant in each condition, which quantifies the effect on growth of an amino acid substitution relative to the wild-type PjDHFR, where a positive score represents a growth advantage, and a negative score a growth disadvantage. Replicates within conditions were well correlated (S4A, S4B and S4C Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pgen.1011252.g002

By using results from the competition assay with MTX, we identified mutations in PjDHFR that confer resistance to this antifolate. To classify mutants as being resistant or not, we first used a Gaussian mixture model to separate the distribution of fitness effects into the number of Gaussian components that best recapitulate the overlying distribution of observed fitness effects at IC75 (five components) and IC90 (four components) (S1 Fig). By using the estimated parameters from the underlying functions, we set the thresholds for resistance using quantiles of the distributions centered on a selection coefficient of 0. For IC75, the thresholds for a resistant mutant and a very resistant mutant were set at selection coefficients of over 0.220 (404/4284 mutants, 9.43%) and 0.432 (129/4284, 3.01%), respectively. For IC90, 0.243 (437/4284, 10.20%) and 0.550 (134/4284, 3.13%), were set for resistant and very resistant, respectively. Because of the difference in the number of modelled Gaussian function components in the two conditions, potentially caused by the difference in strength of the selection pressures, slight differences in the number of mutants that pass the different thresholds are expected.

Overall, 31 positions, or about 15% of the positions along the protein, had at least one mutant that passed the very resistant threshold in both concentrations of MTX. These cutoffs do not consider the various numbers of replicates per amino acid in these experiments, so mutations above these thresholds do not have the same levels of confidence. We therefore also conducted a one-sided Welch’s t-test and corrected for false discovery rates using the Benjamini–Hochberg procedure to control for the false discovery rate of significantly resistant mutants at 5%. We tested whether selection coefficients for amino acid variants were significantly greater than that of synonymous mutants of the wild-type PjDHFR. For IC75, 836 mutants with selection coefficients >0 were considered as significant, and 694 for IC90, with an intersection between the two concentrations of 603 mutations (S1C and S1D Fig, green line, and S6 Fig). Finally, analyzing intersecting datasets from thresholds and statistical significance, we identified 355 mutants as being resistant with a high level of confidence, and 116 mutants that we consider very resistant with a high level of confidence.

Individual growth assays validate results from the pooled competition assay

To validate the results of the pooled competition assays, we manually reconstructed 44 mutations in PjDHFR (Fig 3A, 3B and 3C). These mutations were chosen because they had various selection coefficients along the gradient of values observed. We also included positions that showed discrepancies between the IC75 and IC90 conditions, such as G58Q, D110W, S111E, I196W and I196F (Fig 3A, highlighted green). These mutants exhibited high resistance at IC90 but behaved differently at IC75, not passing the threshold for very resistant. Because the results strongly correlate between these two conditions, we reasoned that these could be experimental outliers, as we had observed increased noise between replicates at IC90 compared to IC75 (S4B and S4C Fig). These outlier mutants are enriched for amino acids encoded by a single codon in our experiment, which could partly explain their higher noise. Validation mutants were tested using growth curves and spot-dilution assays at IC90 in the same conditions as the competition assay, as well as in DMSO. Results of the growth assays confirmed that these outliers were indeed sensitive to MTX at IC90 (Fig 3B). Growth rate for all other mutants individually tested correlated strongly with their selection coefficient from the competition assay based on Spearman’s rank correlation (Fig 3B and 3D). All mutants grew similarly in DMSO (Fig 3C and 3D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

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.

https://doi.org/10.1371/journal.pgen.1011252.g003

To further validate our results and assess the presence of tradeoff between function and resistance, we reintroduced some of the reconstructed validation mutants into strain FDR0001 to assess the mutant protein’s potential for functional complementation. In the absence of β-estradiol, all mutants identified as resistant could complement the deletion of DFR1 in this strain, as expected. Interestingly, most mutants which were identified as being sensitive lost their ability to complement the deletion, implying that these mutations strongly negatively impacted protein function (S8A Fig). In the presence of β-estradiol, all mutants grew similarly in the presence or absence of MTX (S8B Fig). Growth experiments in strain FDR0001 exhibited higher levels of noise than similar experiments in IGA130 (Fig 3A vs S8C and S8D Fig error bars).

We then used these results to correlate growth rate in IGA130 and in FDR0001 to validate how the results from the high-throughput assay could be transposed to strain FDR0001 (dfr1Δ). In DMSO, all mutants in strain IGA130 grew similarity, but non-functional PjDHFR mutants in strain FDR0001 were not able to grow. In the presence of MTX, the phenotypes of resistant mutants are very well correlated between the two strains, highlighting the robustness of this assay (S8C and S8D Fig).

Premature stop codons reduce gene dosage toxicity

In the absence of MTX, the distribution of selection coefficients showed little variance around 0. As the endogenous DFR1 was present in strain IGA130, the effect of PjDHFR mutants on growth without MTX was expected to be minimal. However, we observed an enrichment of positive selection coefficients for nonsense mutations (S5A Fig, row marked with *, S9 Fig) in this condition. Gene dosage toxicity has previously been reported for DHFRs at expression levels higher than wild-type, suggesting that this enrichment is the result of a fitness increase caused by the interruption of functional PjDHFR production [77,79]. Because early stop codons are more likely to prevent the formation of a functional protein than downstream ones, this effect is mainly observable at the beginning of the sequence and is lost by the end of the protein, with selection coefficients becoming slightly negative at around position 160 (S8 Fig). Other mutations, especially before position 50, with similar fitness effects are observable across the protein, which could hint at mutations that affect the function or folding of PjDHFR, preventing gene dosage toxicity. However, the effect of most such mutations (mean across all positive mutations in DMSO = 0.023, median = 0.016) is generally smaller than the mean of stop codons before position 160 (0.060, median = 0.059), with only 0.26% (11/4284) of mutations scoring significatively higher than this mean across the entire protein. When using FoldX to predict the effect of mutations on the stability of PjDHFR, we did not find a significant correlation between the effect of mutations and the selection coefficient in DMSO (S10 Fig). We expected highly destabilizing mutations to be enriched in DMSO, as they would have potentially affected protein folding and thus catalytic activity. The absence of clear positive correlation implies that a higher selection coefficient in DMSO cannot be explained solely by changes in the protein’s free energy. While the observed effects were small, the detection of the effect of stop codons highlighted the sensitivity of this type of competition assay. Indeed, when manually validating mutants from the competition assay using growth curves and spot-dilution assays, we did not detect this phenotype, although the strain with the Empty plasmid had slightly higher growth rate than any other wild-type DHFRs in this assay (in DMSO, Empty has a growth rate of 0.308 OD₆₀₀·h^-1, and ScDHFR of 0.289 OD₆₀₀·h^-1, Fig 3C red and blue horizontal lines, respectively). Most importantly, the experiments in the absence of MTX show that there are no major dominant negative effects of amino acid substitutions that would confound the measures of sensitivity to MTX.

Resistance mutations most often appear at position within ligand binding pocket

Since there is no available crystal structure of PjDHFR, the predicted protein structure was recovered from AlphaFoldDB (UniProt: A0EPZ9). Crystal structures of the DHFR ortholog from P. carinii (61% identity, 75% positives), in complex with MTX/NADPH (PDB: 3CD2) and DHF/NADPH (PDB: 4CD2), were recovered from the Protein Data Bank (PDB) and aligned to generate the holoprotein structure of PjDHFR with MTX/NADPH and DHF/NADPH. Despite sequence identity of only 61% between the orthologs, the structures were highly conserved (root-mean-square deviation (RMSD) of 0.644 Å across the protein backbone heavy atoms, 2.094 Å across all heavy atom pairs), allowing for high-confidence structural alignments of ligands without the use of ab initio docking. We then mapped the selection coefficients on the structure of PjDHFR (Fig 4A and 4B).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

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).

https://doi.org/10.1371/journal.pgen.1011252.g004

Most resistance positions are located within the PjDHFR substrate binding pocket, or are in contact with NADPH, as we would expect for a competitive inhibitor such as MTX. These sites overlap with residues that bind MTX. Not all mutations at these positions lead to resistance. Indeed, resistance implies that the enzyme is no longer inhibited by MTX but maintains proper interactions with the substrate so that it can still perform its catalytic reaction that is required for growth. This leaves very few sites that can be modified. At some of these key positions, such as M33 (4.8 Å from MTX), resistance appears to also correlate with amino acid properties, where hydrophobic amino acids, such as the wild-type methionine, have generally lower selection coefficients in comparison to other polar residues, such as hydrophobic M33F (0.369 in IC90, p-value = 1.5e-8, from one-sided >0 Welch t-test), M33Y (0.243, p-value = 0.01) and M33V (0.441, p-value = 1.1e-5) compared to charged M33R (1.287, p-value = 8.8e-25), M33K (1.088, p-value = 4.7e-7) and short M33G (1.262, p-value = 1.3e-13). At other positions, fewer amino acids had positive selection coefficients. At I142 (7.1 Å from MTX), only two amino acid substitutions, I142W (1.428, p-value = 4.7e-4) and I142Y (0.668, p-value = 2.5e-3), caused resistance, implying that other position-dependent properties, such as side chain bulkiness, are at play. I142W, as well as a sensitive mutation at this same position, I142G (-0.167, p-value = 0.97), were manually reconstructed as part of the validations by growth curves and spot-dilution assays.

Another region where strong effects were observed was the glycine triplet at positions 124 to 126, which is conserved across species (3/3 in humans, S. cerevisiae and E. coli, 2/3 (positions 124 and 125) in mice). All three glycines from this triplet are in contact either with both MTX/DHF and NADPH (124 and 125) or with NADPH only (126), which could imply a role in protein activity and/or ligand binding. Interestingly, at IC75 and IC90, mutants at position 124 with a positive selection coefficient generally have short chain amino acids (G124S/A/P 0.865/0.736/0.244 in IC75, p-values = 8.5e-13/4.8e-8/3.6e-7). Position 126 is less constrained, with 12/20 substitutions having a selection coefficient above the resistance threshold. However, at IC75, position 125 was one of the least permissive positions observed in the competition assay, with the lowest median score by position (-0.735), and no substitution with a selection coefficient higher than 0, implying that the glycine at this position is under strong purifying selection in this protein.

Mechanisms of DHFR resistance to MTX have previously been investigated in eukaryotes, mainly in baker’s yeast, mice, and humans [77,80,81]. We examined how conserved the mechanisms were PjDHFR (S7 Table). Positions with mutations that are most often reported as conferring resistance to MTX, namely M35 in yeast, F31, F34 and Q35 in mice and in humans, are structurally equivalent to positions M33 (maximum selection coefficient at this position in IC90 = 1.485, p-value = 2.8e-11), F36 (maximum = 1.202, p-value = 2.1e-5) and S37 (maximum = 0.760, p-value = 3.1e-8) in PjDHFR, all being <5 Å from MTX. In PjDHFR, some mutations have been reported as conferring resistance to TMP in clinical isolates, with position F36 being reported as conferring the highest resistance to this drug [82]. This reveals that resistance mechanisms are highly conserved among DHFRs. However, while some mechanisms are conserved, and cross resistance is observed, other sites do not share these properties. One such position is position L65, which was reported as conferring high resistance to TMP in clinical isolates, and no mutations at this site were identified as resistant in this screen [38]. Other mutations that were reported or predicted to confer TMP resistance, such as mutations in the Q172-V181 loop (compiled in S7 Table), did not confer resistance to MTX in this study [38]. As antifolates are competitive inhibitors, we expect most mutations that confer resistance to come with a tradeoff in protein activity. In mice and in humans, biochemical properties were measured for some mutants at position F31, structurally equivalent to position M33 in PjDHFR. Resistance was associated with an increased K_m for DHF, which translates to a less active enzyme, up to four-fold for some mutants at this position, consistent with a tradeoff between resistance to MTX and activity [77,81,83].

Potential mechanism of resistance through allostery

By mapping the maximum selection coefficient values at IC90 (Fig 4A) and minimum values at IC75 (Fig 4B) by position on the structure, we can visualize structural determinants of resistance. Most positions where at least one mutant is highly resistant are proximal to MTX/DHF and NADPH, or close to other positions where resistance is observed (Fig 4A). Inversely, almost every position (202/204) has at least one mutant that performed worse than the wild-type PjDHFR at IC75, except for positions M33 and F199. This is consistent with the expectation that most positions along a protein can have a deleterious effect if mutated, especially in the presence of strong selection like MTX [74].

While some mutations were consistent with the literature and structural properties of eukaryotic DHFRs, other resistant mutations were unexpected and may thus reveal new mechanisms of resistance. F199, a position located in a distal β-strand, and which has no direct contacts with drug, substrate, or NADPH, showed high levels of resistance to MTX for many different substitutions, and consistently so across both conditions. The maximum selection coefficients are 1.11 at IC75 and 1.64 at IC90, the second and first highest in their respective assays. Two mutants from this position, F199E (0.585 at IC90) and F199P (1.018 at IC90), were individually reconstructed by site-directed mutagenesis, and were validated for resistance to MTX (S7 Fig). While this position had not been reported in eukaryotic resistance studies, a structurally equivalent position (F153) had been predicted in E. coli as being part of the allosteric “sector” of FolA (PDB: 1RX2), which could hint at a yet unreported mechanism of resistance to MTX [67]. One mutation at this position, F153S, has also been identified in E. coli as slightly increasing resistance to TMP [78]. Interestingly, positions I10 (I5 in E. coli, I11 in baker’s yeast, I7 in mouse and humans) and I142 (Y111 in E. coli, L139 in yeast, F134 in mouse and human) are two other positions where high levels of resistance were observed. Both positions are part of the same β-sheet and physically close to F199, forming a “line/groove” towards the α-helix containing M33, another position where several mutants had increased resistance to MTX, and which is in direct contact with MTX and DHF. Y111 in FolA, the structurally equivalent position to I142, has also been identified as being part of the allosteric “sector” of FolA [67]. All three positions, I10, I142 and F199, are relatively buried, and are hydrophobic amino acids, a property that is conserved across E. coli, baker’s yeast, mouse, and humans. In the case of F199, amino acids with similar properties have lower selection coefficients in MTX, and polar/longer amino acids have higher selection coefficients, with F199R and F199K being respectively the first and fourth mutants with the highest selection coefficients at IC90. Position F199 is also in close proximity to the side chain of position F35, and could be forming π-stacking interactions with this other phenylalanine, which could then affect the α-helix where position M33 is located [84]. Position 199 also displays unique properties in DMSO, being the only position where a mutation, F199P (0.102, p-value = 0.22), is enriched, but the stop codon is not, implying that F199P is potentially more disruptive for protein function or folding than a nonsense mutation at this position. However, this is based on a single datapoint with three replicates and could therefore be due to experimental noise. Overall, this evidence points towards a more important role of this position in DHFR function and resistance than previously estimated.

To investigate the specific effects of mutations at position 199, we conducted an in silico mutagenesis at this position using Rosetta’s FlexddG method and predicted changes in binding energy between PjDHFR, MTX and DHF (Fig 4D) [69]. The changes in binding energy predicted by Rosetta FlexddG between PjDHFR and MTX show a slight positive correlation (ρ = 0.42, Spearman’s p-value = 0.07, r = 0.58, Pearson’s p-value = 0.0098) with the selection coefficients measured at IC90 for each corresponding mutant, implying that mutations that destabilize interactions with MTX are more likely to be highly resistant (Fig 4D). Interestingly, FlexddG predicted that mutations at position 199 would generally be more destabilizing between PjDHFR and MTX than between PjDHFR and DHF, which would translate to a more resistant but still functional enzyme (S11 Fig). Most mutations that are more destabilizing correlate with measured selection coefficient. In E. coli, biochemical properties of mutant F153S were investigated. It was found to have no significant effect on the maximum rate of catalysis (k_cat) of FolA, but did increase K_m to TMP, which is consistent with the decrease in binding affinity modeled by Rosetta [78]. This evidence points towards position 199 having distal effect on binding affinity, but not necessarily on catalytic rate, suggesting that mutants at this position could lead to resistance without strongly compromising enzyme function, making this position a good candidate for the evolution of resistance mutations.

Alternative mechanisms of MTX resistance in PjDHFR

Alternative mechanisms of resistance were also investigated. Indeed, antifolate resistance caused by increased DHFR expression or protein abundance has been documented. To verify this, we constructed PjDHFR-mEGFP fusion proteins for wild-type PjDHFR and some validation mutants of interest. Fluorescence intensity was measured by flow cytometry as a proxy for protein abundance (S12A Fig). As all mutants are under the regulation of the same promoter, variation in abundance should depend on changes in protein stability caused by the mutations. We did not observe a correlation between resistance and protein abundance (S12B Fig). Mutants I10H and I10R (selection coefficients of 1.154 and 0.014, respectively) both had reduced abundance (0.71- and 0.82-fold change in relative fluorescence intensity per cell, respectively), where the most resistant mutant of the two is the least abundant. As for position 199 specifically, we had only reconstructed two validation mutants (F199P and F199E, selection coefficients of 1.018 and 0.585, respectively, both resistant since no mutations at position F199 leads to a negative selection coefficient). These two mutants showed slightly increased expression levels, but the measured increase in expression is not proportional with the measured selection coefficient (F199P and F199E, with increase of 1.05 and 1.02-fold-change in relative fluorescence intensity per cell, respectively). Based on these observations, we believe that MTX resistance cannot be solely attributed to increase in protein abundance, but it is possible that it is a contributing factor.

As the screen was conducted in strain IGA130 (wild-type for genomic DFR1), it was also possible that the expression of PjDHFR mutants from the plasmid caused fluctuations in the expression of genomic DFR1, affecting measured DMS scores. To test this, we constructed strain FDR0003, which is the same as strain IGA130, but where genomic DFR1 is tagged with 3xFLAG-tag in C-terminal. We measured changes in genomic DFR1 expression by anti-FLAG western blot in the presence of several resistant and sensitive mutants, and did not observe any changes in DFR1 expression in the presence of PjDHFR mutants (S12C Fig).

Functional complementation and DMS are powerful tools to investigate hard-to-study organisms

The study of P. jirovecii has always been difficult because of our current inability to cultivate it in vitro or in animal models. As resistance is increasing in this fungal pathogen, there is a pressing need to understand the mechanisms behind its emergence. By employing functional complementation in S. cerevisiae and a DMS library of PjDHFR, we investigated resistance to an antifolate using an exhaustive library of mutants. While this system does not allow the direct investigation of TMP resistance, we believe that in the light of the clinical and evolutionary link between MTX, TMP and P. jirovecii infection, this study provides relevant insights in the evolution of antifolate resistance, and on the biophysics of MTX resistance in DHFRs.

We demonstrate that under the regulation of a constitutive promoter, PjDHFR is sensitive to MTX in a quantitative manner, even in the presence of wild-type genomic DFR1. From this data, we targeted two selection pressures to conduct DMS followed by a competition assay to assess the effect of mutations compared to the wild-type PjDHFR. This allowed us to identify a broad range of fitness effects, both positive and negative, in the presence of MTX. We also observed mutations that affect protein function and gene dosage toxicity. Results from the competition assay are consistent with known resistance mutations to MTX in eukaryotes, implying that some resistance mechanisms are conserved across distant lineages. While individual validations in strain FDR0001 (dfr1Δ) allowed to identify resistance/function tradeoffs, the high-throughput screening in strain IGA130 (wild-type for DFR1) did not allow systematic investigation of this relationship. Further experiments will be necessary to understand the functional tradeoff associated with MTX resistance in PjDHFR.

By using the DMS library from PjDHFR to conduct competition assays in MTX at two different concentrations, we were able to measure different ranges of effect on resistance. Using higher selection pressure, such as IC90, allowed mutations that caused higher-resistance mutations to be more enriched in-regards-to wild-type-like mutations and showed a broader range of effect for highly resistant mutations. At a weaker concentration (IC75), mutations that performed worse than wild-type were more easily identifiable than at IC90, which is indicative of either mutants less resistant than the wild-type, or non-functional enzymes. Finally, by conducting the experiment in the absence of MTX selection, we were able to assess the absence of dominant negative effect, as well as to identify the presence of gene dosage toxicity through the effects of nonsense mutations. Combining the results of both datasets in the presence of MTX also helped identify mutants with inconsistent effects and guided the subsequent validation work. Validations in strain FDR0001 allowed to show a correlation between measured growth rates and scores with validations and high-throughput screening in strain IGA130 and highlighted the robustness of the system.

Resistance to methotrexate occurs through different mechanisms

Most mutations that confer resistance to MTX were within the substrate or cofactor binding pocket, and most often, several different amino acid substitutions allowed for resistance at a given position. In the case of M33, shorter chain, hydrophobic amino acids (wild-type-like) were the substitutions that caused the lowest levels of resistance, implying that changes in local amino acid properties are key for resistance, instead of the specific amino acid present at this position. While M33 is in direct contact with MTX, F199 is not, but both displayed this pattern of “property-based” resistance. Other positions where resistance was observed, such as I10, displayed no such clear pattern. While protein abundance measurements showed no clear correlation with resistance in this study, increase in protein stability has been well documented as an antifolate resistance mechanism, and at least two of the tested resistant mutants showed increase in protein abundance [77]. This implies that several mutational pathways are available for evolution towards resistance to MTX, and possibly other antifolates. This raises major concern about the future of resistance to antifolates in P. jirovecii. However, as MTX is not used to treat P. jirovecii infections in clinical settings, and because of the conflicting reports of the effects of MTX on the evolution TMP resistance and on P. jirovecii infection, it is important to be critical about the nature of these results. We believe, however, that this is a step in the right direction, and that this data will help future studies focusing on more clinically relevant antifolates.

Structural analysis and binding affinity modeling of the observed fitness effects allowed us to identify putatively allosteric mechanisms that cause resistance to MTX, which is supported by a previous study into the allosteric regulation of DHFR. These results hint at more complex internal protein dynamics for function and resistance within DHFRs than previously estimated [67,78,85].