Evolutionary Convergence on a Dihydrofolate Reductase Mutation Drives Trimethoprim–Sulfamethoxazole Resistance in Burkholderia thailandensis

Abstract

Trimethoprim–sulfamethoxazole (SXT) remains central to melioidosis eradication therapy, yet the genetic basis of resistance evolution in Burkholderia pseudomallei is not fully defined. Using Burkholderia thailandensis as a biosafe surrogate, we subjected populations to stepwise in vitro evolution under increasing SXT concentrations. Whole-genome sequencing revealed rapid population diversification followed by selective sweeps leading to fixation of fitter resistant variants. Early adaptation involved mutations in regulators of RND efflux systems and folate-associated genes, consistent with multifactorial resistance. Despite this heterogeneity, all evolutionary trajectories converged on a single nonsynonymous substitution, I99L, in dihydrofolate reductase (DHFR/FolA), which rose to fixation at high SXT concentrations. This substitution mirrors changes reported in SXT-resistant B. pseudomallei isolates from chronic infections. Allelic reconstruction demonstrated that Bt-DHFR­(I99L) is sufficient to confer a 4- to 16-fold increase in trimethoprim MIC. Biochemical and structural analyses showed preserved catalytic activity but reduced trimethoprim binding, consistent with an allosteric resistance mechanism. Together, these findings establish DHFR as a dominant evolutionary target under SXT pressure and support B. thailandensis as a robust model for dissecting resistance evolution in B. pseudomallei.

Introduction

Burkholderia pseudomallei is a Gram-negative bacterium that lives naturally in soil and water. It is also the cause of melioidosis, a serious disease of humans and animals. The infection is well recognized in Southeast Asia and northern Australia, where the bacterium is firmly established in the environment. However, its reach is broader than that melioidosis occurs across much of the tropics, and in many places, it is almost certainly under-diagnosed and under-reported. A modeling study published in 2015 estimated that melioidosis causes roughly 165,000 cases and 89,000 deaths each year worldwide. These numbers are sobering, and they underscore the argument that melioidosis should be recognized as a neglected tropical disease by the World Health Organization. Outbreaks often follow major weather eventsheavy rains, floods, and cyclones. This connection raises a concern for the future: As climate change drives more extreme events, we may see rising incidence in affected regions. Genomic studies have also confirmed that B. pseudomallei is endemic in areas that were not previously recognized, including the southern United States, where cases have been recently reported. Melioidosis arises through direct environmental exposure. The bacterium can enter the body through breaks in the skin, through inhalation, or by ingestion. The risk is higher in people whose immune defenses are already compromised, such as those with diabetes, and the burden of disease falls disproportionately on disadvantaged communities. Clinical presentations vary widely. Some patients develop pneumonia, while others suffer infections of the skin, soft tissues, or genitourinary tract. The bacterium can also invade deeper sites, such as bone or brain, and it is notorious for causing abscesses in organs like the liver and spleen. Melioidosis can take different courses. In many patients, it appears as an acute illness, with symptoms that strike suddenly and progress aggressively. However, there are also chronic cases, where the disease lingers for more than 2 months. Even after treatment, some patients face recurrence: Symptoms may return because the original infection was never fully cleared, or because the person has been exposed again and infected with a new strain.

Treating melioidosis is not an easy task. Like many saprophytic bacteria, B. pseudomallei carries an intrinsic resistance to a broad set of antibiotics (among them aminoglycosides, macrolides, polymyxins, penicillins, and the older cephalosporins). This built-in resistance leaves clinicians with far fewer therapeutic options than they would normally have against a bacterial infection. Therapy follows a two-step course. First comes an intensive phase, usually 14 to 30 days of intravenous ceftazidime or meropenem, aimed at halting the sepsis and bringing the infection under control. That is followed by a much longer eradication phase, at least 12 weeks of oral cotrimoxazole (SXT) to eliminate any lingering bacteria and lower the risk of relapse. Studies do not all agree on the exact dose of SXT or on how long the eradication phase should last. Reported durations range from 2 months to a full year. What is clear, though, is that treatments extending beyond 3 months are linked with fewer relapses. The length of oral eradication therapy is therefore one of the key factors in how well the infection is controlled. Relapse rates are hard to measure, but they have been estimated at about 10% in Thailand and much lower in Australia, and in both settings, the risk falls as the duration of therapy increases. Keragala et al. showed that patients often tolerate SXT poorly. In some cases, the side effects were severe enough to force a switch to the alternative, amoxicillin–clavulanic acid. In others, the adverse reactions led simply to poor compliance with the treatment. In addition, over the past 25 years, researchers have looked closely at how B. pseudomallei can resist to SXT, the key drug used in the oral eradication phase. Despite these efforts, clinically confirmed resistance to SXT remains relatively uncommon, although accurate detection can be complicated by challenges associated with susceptibility testing. − What has become clear is that testing for susceptibility is not straightforward. In earlier work, we showed that clinical isolates of B. pseudomallei can display a striking heterogeneity in their antimicrobial sensitivities. This means that resistance may sometimes be underestimated, depending on which sample is taken and how the bacterium is isolated. Two other controversial studies have highlighted just how tricky it can be for microbiologists to interpret antibiograms. , An Australian study went further, reporting the emergence of resistance to SXT, especially in patients with chronic infections such as cystic fibrosis, and describing the mechanisms underlying this resistance. Laboratory work has shown that the main drivers of multidrug resistance are active efflux systems, particularly the RND (Resistance-Nodulation-Division) family of pumps. Among these, AmrAB-OprA and BpeEF-OprC stand out as the only ones clearly implicated in resistance in clinical isolates. In many Gram-negative pathogens, resistance to trimethoprim is frequently mediated by horizontally acquired plasmid-borne dihydrofolate reductase genes, whereas in B. pseudomallei, resistance is mainly chromosomal. Resistance to SXT has been linked to overexpression of bpeEF-oprC, triggered by mutations in regulatory genes such as bpeS or bpeT, which encode LysR-type transcriptional regulators controlling efflux pump expression. Other pathways can also be involved. Viberg et al. and Podnecky et al. showed that mutations in enzymes of the folate pathway, targeted by both trimethoprim and sulfamethoxazole, play a role in resistant isolates from chronic cases of melioidosis. Mutations in folA, encoding dihydrofolate reductase (DHFR), and folM, encoding a pteridine reductase involved in folate metabolism, were both associated with resistance. , Recent work has further highlighted the importance of additional folate-pathway enzymes in Burkholderia adaptation to antifolate stress. In particular, FolE2 was shown to become conditionally essential in the presence of subinhibitory trimethoprim concentrations, underscoring the role of broader folate metabolic responses in antibiotic tolerance.

Although clinically confirmed resistance to SXT in B. pseudomallei remains relatively uncommon, experimental evolution approaches provide a powerful framework to explore potential adaptive trajectories that may arise under sustained antibiotic pressure.

In this study, we implemented a stepwise in vitro selection framework designed to model adaptive trajectories under sustained trimethoprim–sulfamethoxazole SXT selection pressure. While this system does not reproduce the precise pharmacokinetic fluctuations observed during eradication therapy, it enables controlled exploration of evolutionary responses to prolonged antibiotic exposure. Moreover, variability in drug exposure and uncertainties related to clinically relevant dosing levels may influence resistance selection dynamics. Thus, this system should be interpreted as a tractable model of adaptation to escalating antibiotic pressure rather than a direct simulation of eradication therapy conditions. Using B. thailandensis as a nonpathogenic surrogate closely related to B. pseudomallei, we subjected bacterial populations to progressively increasing concentrations of SXT. This model allowed us to explore how sustained antibiotic pressure shapes adaptive diversity within bacterial populations and to identify recurrent genetic changes associated with reduced susceptibility.

Specifically, we aimed (i) to identify the genetic determinants that enable adaptation to increasing SXT concentrations, with a particular focus on resistance mechanisms repeatedly described in clinical isolates and (ii) to characterize the biochemical and physiological consequences of a key adaptive mutation affecting dihydrofolate reductase (DHFR). The close genetic relatedness between B. thailandensis and B. pseudomallei, together with the reduced biosafety constraints associated with the former, makes this system a robust and tractable model for investigating the emergence, convergence, and functional impact of resistance-associated mutations under antibiotic pressure.

Results

Genetic Adaptations of B. thailandensis to Trimethoprim–Sulfamethoxazole (SXT) Pressure

To investigate the genetic basis of adaptation to SXT, B. thailandensis populations were sequentially exposed to increasing concentrations of the antibiotic and analyzed by whole-genome sequencing at each selection step. Across the course of the experiment, a broad diversity of mutations accumulated, affecting genes involved in multiple cellular functions (Supporting Information 1). In the present study, we focused our analysis on mutations affecting pathways previously implicated in SXT resistance in B. pseudomallei, namely, efflux regulation and the folate biosynthesis pathway ,, (Figure ).

1.

Stepwise emergence and redistribution of resistance-associated mutations in Burkholderia thailandensis populations exposed to increasing trimethoprim–sulfamethoxazole pressure. The Sankey diagram, generated in R using the ggalluvial package, illustrates the flow and relative abundance of mutations detected in genes involved in folate metabolism (red) and in the regulation of RND efflux systems (blue) across B. thailandensis populations subjected to progressively increasing concentrations of trimethoprim–sulfamethoxazole (SXT). Populations are shown on the left (WT, SXT_0.5, SXT_1, SXT_2, SXT_4, SXT_8, SXT_16, and SXT_32), corresponding to successive selection steps, while the right panel lists the proteins affected, annotated using B. thailandensis nomenclature and described as FolA/DHFR, FolM-like, BpeS-like, BpeT-like, and BpeR regulators. Mutations are grouped by functional category: folate-pathway enzymes (red), efflux regulation (blue), and absence of detected mutations (gray). Numeric values displayed next to mutation labels correspond to the relative frequency of sequencing reads (in percent) carrying each mutation within the population at the indicated selection step, as determined by breseq population analysis (−p mode). The width of each flow is proportional to these frequencies. Only mutations affecting genes previously implicated in antimicrobial resistance are shown; additional mutations detected across the genome are reported in Supporting Information 1. This representation highlights the progressive diversification of efflux-regulator mutations at early and intermediate SXT concentrations, followed by convergent selection on folate-pathway enzymes, culminating in dominance of the DHFR I99L variant at the highest levels of antibiotic pressure.

At early stages of selection (≤1 mg/L SXT), no mutations were detected in genes previously associated with SXT resistance.

Mutations in Efflux-Regulator–Associated Genes

Genome sequencing revealed multiple mutations in genes encoding transcriptional regulators of RND efflux systems as SXT concentrations increased. These mutations affected BTH_RS15410, BTH_RS10785 (encoding CeoR), and BTH_RS15655 (encoding BpeR). BTH_RS15410 encodes a LysR-type transcriptional regulator whose protein product is homologous to BpeS, a regulator of the BpeEF–OprC efflux system in B. pseudomallei.

At 2 mg/L SXT, a nucleotide substitution in BTH_RS15410 produced the S280P amino-acid change in the BpeS-like protein, detected in 29.2% of sequencing reads. Its frequency increased to 55.8% at 4 mg/L and remained at 50% at 8 mg/L, before decreasing to 10% at 16 mg/L and becoming undetectable at 32 mg/L. At 4 mg/L SXT, a mutation was also identified in BTH_RS10785, an LysR-type transcriptional regulator and homologue of BpeT from B. pseudomallei, resulting in the S280P substitution in the BpeT-like protein (6.8%). A second mutation in BTH_RS15410 generated the K267T substitution in the BpeS-like protein, detected in 19.7% of reads at 4 mg/L and increasing to 42% at 32 mg/L. At the highest SXT concentration (32 mg/L), additional mutations were detected in BTH_RS15410 and BTH_RS15655, producing the BpeS­(K265T) (14.9%) and BpeR­(L166P) (7.1%) substitutions, respectively. BTH_RS15655 encodes BpeR, a TetR-family transcriptional regulator of the BpeAB–OprB efflux system.

Overall, the detection of multiple substitutions in regulators homologous to those controlling BpeEF–OprC and BpeAB–OprB across the SXT gradient illustrates the progressive diversification of efflux-regulator variants during adaptation.

Mutations in Folate-Pathway–Associated Genes

Mutations affecting the folate biosynthesis pathway were detected at later stages of selection and increased in frequency with rising SXT concentrations. The first such mutation was observed at 2 mg/L SXT, where a 7 bp deletion was identified in BTH_RS00215 (9.6%). This gene is annotated as encoding an SDR-family oxidoreductase (Supporting Information 1). The corresponding protein shares 97% identity with its orthologue in B. pseudomallei, which has been described as FolM-related. The 7 bp deletion introduces a premature stop codon and is predicted to truncate the encoded protein.

At 8 mg/L SXT, two additional nonsynonymous mutations affecting the same FolM-like protein were detected, producing the T201P (16.8%) and D102G (15%) amino-acid substitutions. The D102G variant increased in frequency with rising SXT concentrations, reaching 87% at 16 mg/L and 100% at 32 mg/L.

At 4 mg/L SXT, a nonsynonymous point mutation was detected in BTH_RS20745 (10%), a gene annotated as encoding the dihydrofolate reductase of B. thailandensis (Bt-DHFR). The protein encoded by BTH_RS20745 shares 95.2% identity with FolA, the DHFR of B. pseudomallei. This nucleotide substitution results in the I99L amino-acid change in the DHFR enzyme. The proportion of sequencing reads carrying the mutation producing the FolA-like­(I99L) variant increased to 35.7% at 8 mg/L, 88% at 16 mg/L, and became dominant at 32 mg/L.

Overall, stepwise exposure to increasing SXT concentrations revealed a progressive diversification of adaptive mutations affecting multiple cellular functions, followed by the selection of well-characterized resistance pathways. While early adaptation involved a variety of mutations, including regulators of RND efflux systems, sustained selection ultimately favored alterations in the folate pathway, with successive mutations in FolM-like enzyme preceding the fixation of a DHFR variant, highlighting folate metabolism as the central target of SXT resistance.

Comparative Phenotypic Analysis of B. thailandensis SXT-Adapted Populations and the Bt-DHFR­(I99L) Variant

Given the progressive increase in the frequency of the mutation producing the I99L amino-acid substitution in Bt-DHFR during SXT adaptation, we next assessed the phenotypic consequences associated with this substitution. To this end, the nucleotide change generating the Bt-DHFR­(I99L) variant encoded by BTH_RS20745 was introduced into the wild-type strain by homologous recombination. We then compared antibiotic susceptibility profiles and growth characteristics of SXT-adapted populations and of the reconstructed Bt-DHFR­(I99L) strain in both rich (MH2) and minimal (M9) media.

Evolution of SXT Susceptibility in Adapted Populations

To monitor the phenotypic evolution of B. thailandensis under increasing SXT pressure, minimum inhibitory concentrations (MICs) were determined by E-test on MH agar (Table ). The wild-type strain displayed a baseline MIC of 1 mg/L for SXT. In contrast, populations selected at increasing SXT concentrations showed a stepwise elevation of MIC values, reaching 2 mg/L for TS2 and up to 32 mg/L for TS8 and TS16.

1. Susceptibility of B. thailandensis Adapted Populations to Antibiotics from Different Classes by the E-Test Method on Mueller–Hinton Agar ^,

	MIC (mg/L)
	wt	TS0.5	TS1	TS2	TS4	TS8	TS16	TS32
SXT	1	1	1	2	3	32	32	32
NAL	16	12	12	256	256	256	256	256
CHL	8	6	8	256	256	256	256	256
CAZ	0.5	0.75	0.5	0.75	0.75	0.75	0.75	0.75
MEM	0.5	0.75	0.5	0.5	0.5	0.5	0.5	1
CIP	0.75	0.5	0.38	0.75	1.5	1.5	0.75	3
DOX	1.5	1	1.5	1.5	3	2	1.5	8

^aMinimal inhibitory concentrations (MICs) are expressed in mg/L. Measurements were performed for the wild-type strain and for populations evolved under increasing trimethoprim–sulfamethoxazole pressure. Each value represents the median of three biological replicates, obtained from three independent experimental repetitions.

^bAbbreviations: SXT, trimethoprim/sulfamethoxazole; NAL, nalidixic acid; CHL, chloramphenicol; CAZ, ceftazidime; MEM, meropenem; CIP, ciprofloxacin; DOX, doxycycline. TS0.5, TS1, TS2, TS4, TS8, and TS16 correspond to populations evolved under 0.5, 1, 2, 4, 8, and 16 mg/L of SXT, respectively.

Evolution of Cross-Resistance Profiles in SXT-Adapted Populations

In addition to reduced susceptibility to SXT, adapted populations displayed altered susceptibility profiles toward other antibiotics. MICs of nalidixic acid (NAL) and chloramphenicol (CHL) increased markedly, reaching 256 mg/L from TS2 onward. In contrast, susceptibility to ceftazidime (CAZ) and meropenem (MEM) remained unchanged across all populations tested. Ciprofloxacin (CIP) MICs varied between 0.38 and 1.5 mg/L. In contrast, doxycycline (DOX) MICs displayed a non monotonic pattern across the selection series, with a transient 2-fold increase at TS4, a return to baseline levels at TS8 and TS16, and a subsequent rise at TS32. In populations adapted at the highest SXT concentration (TS32), higher MICs for CIP and DOX were observed compared with populations selected at lower SXT levels (Table ).

Antibiotic Susceptibility Profile of the Reconstructed Bt-DHFR­(I99L) Strain

In MH2 medium, the reconstructed strain carrying the Bt-DHFR­(I99L) variant exhibited a 4-fold increase in trimethoprim (TMP) MIC compared with the wild-type strain (64 versus 16 mg/L). A similar 4-fold increase was observed for SXT (16 versus 4 mg/L). In contrast, MICs for sulfamethoxazole (SMX), ceftazidime (CAZ), nalidixic acid (NAL), doxycycline (DOX), chloramphenicol (CHL), gentamicin (GEN), ciprofloxacin (CIP), and meropenem (MEM) differed by no more than 2-fold or remained unchanged (Table -A).

2. Antibiotic Susceptibility Profiles (MICs, mg/L) of Wild-Type Burkholderia thailandensis and the Reconstructed Bt-DHFR­(I99L) Mutant Strain to Antibiotics from Different Classes, Determined According to CLSI (Clinical and Laboratory Standards Institute) Guidelines in (A) Mueller–Hinton Broth II (MH2) and (B) M9 Minimal Medium,

A
	MIC (mg/L)
	TMP	SMX	SXT	CAZ	NAL	DOX	CHL	GEN	CIP	MEM
wt	16/8	1216	4	1	64	2	8	>128	4	0.5
Bt-DHFR(I99L)	64	608	16	1	32	2	8	>128	2	1

B
	MIC (mg/L)
	TMP	SMX	STX	CAZ	NAL	DOX	CHL	GEN	CIP	MEM
wt	4	608	1	2	64	8	16	>128	4	0.5
Bt-DHFR(I99L)	64	304	4	2	64	8	16	>128	4	0.5

^aEach value represents the median of three biological replicates, obtained from three independent experimental repetition.

^bAbbreviations: MIC, minimal inhibitory concentration, TMP, trimethoprim; SMX, sulfamethoxazole; SXT, trimethoprim/sulfamethoxazole; CAZ, ceftazidime; NAL, nalidixic acid; DOX, doxycycline; CHL, chloramphenicol; GEN, gentamicin; CIP, ciprofloxacin; MEM, Meropenem.

Comparable results were obtained in minimal medium. The TMP MIC increased 16-fold (from 4 to 64 mg/L), and the SXT MIC increased 4-fold (from 1 to 4 mg/L), while no substantial changes were observed for antibiotics outside the folate-pathway inhibitors (Table B). These data indicate that the Bt-DHFR­(I99L) substitution selectively affects susceptibility to trimethoprim and cotrimoxazole.

Growth Fitness of the Reconstructed Bt-DHFR­(I99L) Strain

To evaluate the fitness impact associated with the Bt-DHFR­(I99L) substitution, growth kinetics of the reconstructed strain and the wild-type strain were compared in different culture media. In MH2 and TS broth, the Bt-DHFR­(I99L) strain exhibited a modest but statistically significant increase in generation time relative to the wild type. In minimal medium (M9), this difference was more pronounced, with the generation time increasing from 1.7 h for the wild-type strain to 2.3 h for the Bt-DHFR­(I99L) strain (Table ).

3. Generation Time in Different Growth Media .

strain	M9 (Tg, hr)	MH2 (Tg, hr)	TS (Tg, hr)
wt	1.7 ± 0.08	0.8 ± 0.01	1 ± 0.05
Bt-DHFR(I99L)	2.3 ± 0.11	0.9 ± 0.04	1.1 ± 0.02

^aGeneration times (Tg) of the wild-type and Bt-DHFR­(I99L) strains were determined in minimal medium (M9), Mueller–Hinton Broth II (MH2, cation-adjusted), and trypticase soy broth (TS). Tg values (in hours, h) were calculated from maximal growth rates (μmax) derived from growth curves obtained in 96-well microplates (see Supporting Information 1 for detailed curves) Values are expressed as mean ± SEM (standard error of the mean) and correspond to six independent measurements (n = 6), derived from three independent experiments performed in duplicate.

Overall, these results indicate that the Bt-DHFR­(I99L) substitution confers a measurable fitness cost that is condition dependent, being most apparent under nutrient-limited growth conditions.

Biochemical and Structural Characterization of Burkholderia thailandensis DHFR (Bt-DHFR)

Enzymatic Activity of Bt DHFRs and Inhibition by Trimethoprim

His-tagged versions of wild-type Bt-DHFR and of its I99L variant (Bt-DHFR­(I99L)) were produced in Escherichia coli and purified, before removal of the His-tag by thrombin cleavage (see Materials and Methods). In order to compare the catalytic properties of the wild-type and I99L versions of the enzyme, we used the classical spectrophotometric assay. In this assay, the simultaneous reduction of DHF and oxidation of NADPH are monitored through the lowering of A₃₄₀ (Figure A). We first measured the k _cat of wild-type Bt-DHFR and the associated K _M of DHF at a fixed concentration of the NADPH substrate. Initial rate measurements allowed us to derive a k _cat of 52 ± 5 s^–1 and a K _M of 2.7 ± 0.6 μM (Figure B). Introduction of the I99L substitution in Bt-DHFR did not significantly change the K _M of DHF (2.2 ± 0.6 μM). However, it caused a reduction of k _cat by a factor of nearly 3 (17 ± 4 s^–1). This lowering of the k _cat value likely explains the reduced fitness of the strain carrying the Bt-DHFR­(I99L) mutant protein.

2.

Activity of Bt-DHFR and of its I99L variant. (A) Scheme of the reaction catalyzed by DHFR. (B) Typical experiment for determination of the K _M of DHF and associated k _cat values in the presence of 0.4 mM NADPH for Bt-DHFR (triangles) or its I99L variant (circles). Initial rates of the reaction were determined spectrophotometrically at 25 °C (see Materials and Methods). Values determined from three independent experiments were k _cat = 52 ± 5 s^–1/K _M = 2.7 ± 0.6 μM (Bt-DHFR) and k _cat = 17 ± 4 s^–1/K _M = 2.2 ± 0.6 μM (I99L variant). (C) Inhibition of Bt-DHFR activity by TMP. Initial rates were measured at 25 °C in the presence of 0.9 nM Bt-DHFR, 0.4 mM NADPH, 2.6 μM of DHF, and the indicated concentrations of TMP. Rates are indicated as percentages of the values measured in the absence of TMP. Triangles: Bt-DHFR; circles: I99L variant. The results are from a single series of experiments. (D) Same experiment except that the DHF concentration was 100 μM.

In a second step, we studied the inhibitory effect of trimethoprim (TMP) on the wild-type and I99L enzymes. Nanomolar concentrations of TMP inhibited the activity of wild-type Bt-DHFR. k _cat and K _M (DHF) measurements made in the presence of substoichiometric amounts of TMP (0.125 or 0.25 nM in the presence of 0.9 nM Bt-DHFR) were consistent with tight (K _d < 1 nM) binding of the antibiotic that cannot be chased by concentrations up to 50 μM of DHF during the few minutes of the assay. Inhibition at TMP concentrations smaller than that of the enzyme prevented accurate determination of inhibition constants. Thus, we compared the effect of TMP on the activity of the Bt-DHFR­(I99L) with that of the wild-type enzyme by measuring the initial rates in the presence of either saturating (100 μM) or nonsaturating (2.7 μM) concentrations of DHF (Figure C,D). With the wild-type enzyme and nonsaturating DHF concentration, inhibition was essentially complete when the TMP/Bt-DHFR stoichiometry reached 0.5. This suggests that the concentration of wild-type Bt-DHFR, determined using a computed extinction coefficient, was overestimated. Therefore, the results were consistent with stoichiometric inhibition of wild-type Bt-DHFR by TMP, showing that the antibiotic binds the enzyme with a subnanomolar dissociation constant. At high DHF concentration (100 μM), inhibition was still very strong confirming the subnanomolar affinity of TMP. The I99L variant was less sensitive to TMP, at both low and high concentrations of DHF. In the presence of 100 μM DHF, a concentration of 16 nM of TMP was necessary to reduce the initial rate of the mutant enzyme by 41%. The results clearly show that the binding affinity of trimethoprim to the I99L variant is smaller than that to the wild-type Bt-DHFR.

Binding of Trimethoprim to Wild-Type Bt-DHFR and Its I99L Variant

In order to study the binding of the antibiotic to the enzymes in the absence of the substrates, we used differential scanning fluorimetry (DSF). In this method, the binding of a ligand to an enzyme is monitored through its ability to increase the melting temperature of the protein. Protein denaturation is followed thanks to an increase of the fluorescence of a dye, SYPRO orange (Invitrogen), upon its binding to denatured protein. In this assay, we used a protein concentration of 8 μM, necessary to obtain an adequate S/N ratio, and concentrations of TMP in the 0.1–64 μM range. With the wild-type enzyme, the effect of TMP was observed from substoichiometric amounts of the antibiotic (Figure A,E). This prevented us to deduce a dissociation constant but confirmed the high affinity of TMP for wild-type Bt-DHFR (in the submicromolar range). In sharp contrast, no effect of TMP (up to 64 μM) on the stability of the Bt-DHFR­(I99L) variant could be detected (Figure B,E). This suggested that the mutation strongly diminished the affinity of the antibiotic for the enzyme. Synergistic effects between TMP and NADPH bindings have already been observed for bacterial DHFRs. We therefore repeated the DSF experiments in the presence of 100 μM of NADPH. The results showed that TMP strongly stabilized the wild-type enzyme in the presence of saturating amounts of NADPH (Figure C,F). The denaturation curves in the presence of substoichiometric amounts of TMP indicated the existence of two enzyme populations, one stably bound to TMP and NADPH and the other one bound to NADPH only because no free TMP is available. This behavior is consistent with a long lifetime of the ternary complex wild-type Bt-DHFR:NADPH:TMP, and therefore with its high stability. Synergistic binding of TMP and NADPH also occurs with Bt-DHFR­(I99L) (Figure D,F). Indeed, in the presence of NADPH, binding of TMP to the mutant enzyme can be observed, but the denaturation curves were consistent with a dissociation constant in the micromolar range. Thus, the binding affinity of the antibiotic to the I99L variant was smaller than that for the wild-type enzyme. Overall, the DSF analyses were consistent with the enzymatic assays.

3.

Thermal shift assay for the binding of TMP to Bt-DHFR (A, C) or its I99L variant (B, D). Each denaturation curve is shown with a color code representing the concentration of TMP used. (C, D) Experiments were performed in the presence of 100 μM NADPH. (E) Plot of the melting temperatures as a function of TMP concentration for Bt-DHFR (open circles, experiment shown in A) and I99L variant (closed triangles, experiment shown in B). (F) Same as (E) for the experiments in the presence of 100 μM NADPH (C, D). In (F), for Bt-DHFR, the two Tm values corresponding to the two transitions observed at 1 and 2 μM TMP are plotted.

Crystallographic Structures of Wild-Type Bt-DHFR and of Its I99L Variant

The crystal structure of Bt-DHFR bound to NADPH and TMP was determined to 1.85 Å resolution using X-ray crystallography (Supporting Information 2, Table S1). All enzyme residues could be modeled, with the exception of the three C-terminal ones. In addition, the two ligands were clearly defined in the electron density. The enzyme displays the usual fold of bacterial DHFRs with two subdomains known as the “adenosine binding subdomain” (residues 40–93) and the “loop subdomain” (residues 1–39 and 94–164; Figures A and ). Overall, the fold is built around an eight-stranded sheet (A–H) surrounded by four helices (B, C, E, and F). The four loops, usually called Met20, CD, FG, and GH, are well-defined. The Met20 loop, known to have crucial motions during catalysis, is observed in its closed state, consistent with the binding of NADPH and TMP. The overall structure is closely similar to that of E. coli DHFR:NADPH:TMP complex. The two structures were superimposed with an rmsd of 0.798 Å for 141 C atoms compared. NADPH and TMP binding sites are shown in Figure . For the I99L variant, crystals were obtained in space group P312 with five molecules in the asymmetric unit. Models of the five DHFR molecules, refined to 2.2 Å resolution, are very similar, except that the three C-terminal residues are not visible in monomers D and E. The four monomers B–E can be superimposed on molecule A (the reference monomer hereafter) with rmsd values of 0.273 Å for all 167 C pairs (molecule B), 0.267 Å (molecule C), 0.337 Å (molecule D), and 0.268 Å (molecule E). The structure of the I99L variant is closely similar to that of the wild-type enzyme. The two structures can be superimposed with an rmsd value of 0.487 Å for all C-alpha atoms. The largest deviation concerns residues 131–136 in the FG loop with maximal main-chain displacements of 1.2 Å. It is also notable that the adenine moiety of NADPH is slightly rotated in the mutant enzyme, although this difference may be due to crystalline contacts involving the adenine ring in the wild-type enzyme crystals. Such contacts do not occur in crystals obtained with Bt-DHFR­(I99L). Importantly, the TMP binding site is not involved in crystal packing, in both the wild-type and mutant enzyme crystals. In the wild-type Bt-DHFR, I99 makes several contacts with the trimethoprim molecule. This includes van der Waals interactions as well as a hydrogen bond of the main-chain carbonyl group with the proximal amino group (N7) of the antibiotic. In addition, the side chain of I99 makes van der Waals contacts with F33, I7, and I42 that further contribute to stabilize the binding of the antibiotic (Figure B,D). Change of I99 into L causes subtle changes, with some modifications in the network of van der Waals contacts with TMP (Figure C, Supporting Information 2, Figure S1 and Tables S2 and S3). Additionally, in the mutant enzyme, the two delta carbons of L99 make new contacts with residues such as T48 and I52 in the α-helix. These modifications do not straightforwardly explain the loss of affinity for TMP, suggesting that it is rather the consequence of accumulating several suboptimal contacts. Notably, residue 99 is located at the interface between the “loop” and “adenosine binding” subdomains of the enzyme. Substitution of I99 into L has an impact on the network of interactions at this interface (Figure C,E and Supporting Information 2, Tables S2 and S3). These new interactions may therefore affect the concerted motions within the enzyme, known to be important for catalysis. ,, To further substantiate this idea, we examined the temperature factors in the Bt-DHFR structures (Supporting Information 2, Figure S2). This analysis highlighted that the CD loop (65–75) was less mobile in crystalline Bt-DHFR, as compared to any monomer in crystals of the I99L variant. Conversely, the FG loop (121–139) was more mobile in the wild-type case. Accordingly, crystal packing involved the CD loop in the wild-type case, whereas the FG loop was less implicated (Supporting Information 2, Figure S3). As noted above, the 131–136 region in the FG loop is the most displaced one when structures of Bt-DHFR and of the I99L variant are compared. The opposite situation was observed in the I99L case. In this context, mobility of the CD loop caused by the I99L substitution may explain why we failed to obtain crystals isomorphic to those of the wild-type protein, despite extensive trials. Instead, crystallization of the mutant enzyme rather involved the less mobile FG loop. Another behavior favors the idea that the I99L substitution affects enzyme allostery. In the thermal shift assay experiments (Figure C), baseline fluorescence is decreased in the presence of NADPH whereas this effect is less pronounced for the I99L variant (Figure D). This suggests that the SYPRO orange fluorophore can bind the NADPH site in wild-type Bt-DHFR but that this binding is affected by the I99L mutation.

4.

Crystal structures of wild-type and I99L DHFR. (A) Cartoon representation of wild-type Bt-DHFR complexed with TMP and NADPH. The Loop subdomain (1–39 and 94–164) is in cyan and the “Loop subdomain” (40–93) in blue. Ligands are shown in sticks. Secondary structure elements are labeled using the E. coli nomenclature. (B) Close-up view of wild-type enzyme residues interacting with TMP. Side chains are represented as sticks. Contacts of I99 with TMP are visualized with dashed lines. For (B–E), contacts were calculated in ChimeraX using default parameters (van der Waals overlap ≥−0.4 Å). (C) Close-up view of I99L enzyme residues interacting with TMP. I99L enzyme is colored with two shades of blue. Contacts of L99 with TMP are visualized with dashed lines (see also Supporting Information 2, Table S3). (D) Close-up view of wild-type enzyme residues interacting with I99 (see also Supporting Information, Table S2). Contacts are visualized with dashed lines. (E) Close-up view of I99L enzyme residues interacting with I99 (see also Supporting Information, Table S2). Contacts are visualized with dashed lines. Comparison of views D and E shows that residue 99 is located at the interface between the two DHFR subdomains. Modification of I99 into L increases the number of contacts (see also Supporting Information 2, Table S2) that might affect the dynamics of the enzyme.

5.

Structure-based sequence alignment of dihydrofolate reductase from Burkholderia thailandensis (Bt-DHFR) and its Escherichia coli homologue (Ec-DHFR; PDB code 1RX2 ). The amino-acid sequences of Bt-DHFR and Ec-DHFR were aligned based on their three-dimensional structures. Identical residues are highlighted in red, and conserved substitutions are indicated in blue. Secondary structure elements (α-helices and β-strands) are shown above or below the alignment and labeled according to the classical DHFR nomenclature (Met20 loop, CD loop, FG loop, and GH loop). TT stands for tight turn, and h indicates 3₁₀ helices. The alignment was generated using crystallographic coordinates of DHFR–NADPH–trimethoprim complexes, allowing direct comparison of conserved structural features and residues involved in substrate and inhibitor binding.

Taken together, enzymatic, biophysical, and structural analyses demonstrate that the I99L substitution in Bt-DHFR selectively reduces trimethoprim binding while preserving the overall fold of the enzyme and the K _M of DHF. This substitution is associated with a decrease in catalytic turnover and a marked reduction in inhibitor affinity, as evidenced by enzymatic assays, DSF measurements, and crystallographic structures. Rather than having large direct effects, the I99L substitution acts through subtle modifications that affect the allostery of the enzyme. These results are in line with those concerning the behaviors of E. coli DHFR variants, including the equivalent I94L one (Figure ), where extensive evolution experiments and computational studies concluded that many mutations resulting in TMP resistance caused allosteric effects.

Discussion

In this study, we used a stepwise in vitro evolution model in Burkholderia thailandensis to investigate how prolonged exposure to trimethoprim–sulfamethoxazole (SXT) drives the emergence of resistance. Our objective was 2-fold: first, to assess whether controlled antibiotic selection recapitulates adaptive trajectories previously documented in B. pseudomallei; , and second, to characterize the genetic, biochemical, and physiological consequences of key resistance-associated mutations. By combining population genomics with phenotypic and structural analyses, our work provides an integrated view of how resistance to folate-pathway inhibitors can emerge and stabilize under sustained selective pressure.

A major outcome of this study is the demonstration that progressive SXT exposure generates a genetically diverse population, characterized by the coexistence of multiple adaptive lineages. Early during selection, numerous low-frequency mutations affecting a wide range of cellular functions were detected, reflecting the broad mutational space accessible under antibiotic stress (see Supporting Information 1). This diversity progressively narrowed as selection intensified, leading to the enrichment of mutations affecting a limited number of resistance-associated pathways. Such dynamics are consistent with stepwise adaptation through the competition and replacement of subpopulations with incremental fitness advantages, a process that closely mirrors evolutionary patterns observed in chronic infections.

Efflux Regulation as an Early and Additive Adaptive Mechanism

Among the first resistance-associated pathways to be repeatedly targeted during selection were regulatory circuits controlling RND efflux systems. Mutations affecting homologues of BpeS and BpeT, regulators of the BpeEF–OprC efflux pump, and BpeR, the repressor of BpeAB–OprB, were detected at multiple stages of the SXT gradient. These regulators have been extensively characterized in B. pseudomallei and are well-established contributors to multidrug resistance in clinical isolates. ,,

The recurrence of mutations at positions previously described by Podnecky et al. and Rhodes et al. supports the idea that efflux regulation represents a predictable and readily accessible adaptive route under antibiotic pressure. , Phenotypically, these mutations correlated with cross-resistance to antibiotics known to be influenced by efflux activity, such as nalidixic acid, chloramphenicol, doxycycline, and ciprofloxacin, while susceptibility to β-lactams remained unaffected. The progressive diversification of efflux-regulator mutations across selection steps suggests fine-tuning of efflux activity rather than reliance on a single regulatory change.

However, efflux-related mutations did not become fixed early during evolution, consistent with the known metabolic cost associated with constitutive pump overexpression. Increased transcriptional burden and energy consumption likely limit the fitness benefit of efflux activation when antibiotic pressure is moderate, positioning efflux as an additive or reinforcing mechanism rather than the primary driver of SXT resistance.

Folate-Pathway Remodeling: An Upstream Role for FolM

Beyond efflux regulation, adaptation consistently targeted the folate biosynthesis pathway, the direct target of both trimethoprim and sulfamethoxazole. Notably, mutations affecting a FolM-like pterin reductase were among the earliest alterations detected in this pathway. Loss-of-function mutations in FolM homologues have previously been reported in SXT-resistant B. pseudomallei clinical isolates , and, in E. coli, have been shown to remodel folate metabolism by increasing reliance on dihydrofolate reductase (DHFR). Recent work has also highlighted the contribution of additional folate-pathway enzymes to antifolate stress adaptation in Burkholderia. In particular, FolE2 has been shown to become conditionally essential in the presence of subinhibitory trimethoprim concentrations, suggesting that metabolic remodeling of the folate pathway can support bacterial survival under antifolate pressure.

In our model, multiple independent mutations accumulated in the FolM-like protein across increasing SXT concentrations, including a deletion predicted to abolish function and nonsynonymous substitutions that rose to high frequency. The progressive enrichment of these variants suggests that perturbation of the pterin-reduction branch of the folate pathway provides a selective advantage under combined trimethoprim–sulfamethoxazole pressure. Whether this advantage arises from altered folate flux, reduced metabolic waste, or indirect modulation of DHFR demand remains to be determined. Nevertheless, the reproducibility of FolM alterations across selection steps points to a nonrandom and potentially compensatory role during adaptation. ,

Convergent Selection on FolA/DHFR as the Dominant Resistance Mechanism

Ultimately, adaptation converged on a single dominant mechanism: the I99L substitution in FolA/DHFR. This substitution, previously identified in a B. pseudomallei clinical isolate from a cystic fibrosis patient and experimentally validated by Podnecky et al., became fixed at high SXT concentrations in our in vitro model. Its repeated emergence across independent contexts underscores the central role of DHFR as a critical evolutionary bottleneck in resistance to folate-pathway inhibitors.

Studies with recombinant wild-type and I99L enzymes reveal that the mutation primarily affects catalytic turnover rather than substrate binding, while also weakening trimethoprim affinity. Structural comparisons indicate that the mutation does not induce major rearrangements in the active site or significant changes in trimethoprim positioning. These observations suggest that resistance arises not through direct steric interference with inhibitor binding but through more subtle effects affecting enzyme dynamics. Residue I99 in Bt-DHFR corresponds to residue I94 in E. coli DHFR and is located near the interface between structural elements involved in coordinated conformational changes during catalysis. Substitutions at this position may therefore perturb long-range interaction networks within the enzyme. Similar mechanisms have been proposed for the homologous I94L mutation in E. coli DHFR, where resistance has been attributed to alterations in conformational dynamics and allosteric communication rather than to direct modification of the inhibitor-binding pocket. ,,, Together, these observations support the idea that this region of DHFR represents a recurrent evolutionary hotspot for trimethoprim resistance across bacterial species, where mutations can modulate inhibitor binding through indirect structural or dynamic effects. This region of DHFR therefore represents a recurrent hotspot for resistance evolution under trimethoprim selection.

Despite conferring a clear resistance benefit, the I99L substitution imposed a measurable fitness cost. While growth defects were modest in rich media, they became pronounced under nutrient-limited conditions, highlighting a context-dependent trade-off between resistance and metabolic efficiency. Such a cost is nevertheless insufficient to prevent fixation of the mutation under sustained antibiotic pressure, explaining its dominance at later stages of selection.

Implications for Resistance Evolution and Therapeutic Strategies

Although folA mutations are relatively rare in clinical collections of B. pseudomallei, this scarcity does not contradict the central role of DHFR in resistance. While mutations in folate-pathway genes have been reported less frequently in clinical isolates, the relatively high SXT concentrations used in our experimental evolution framework may favor the emergence of such adaptive solutions. Experimental evolution under sustained antibiotic pressure can reveal a broader range of evolutionary trajectories that may not be commonly detected in clinical settings but nonetheless represent potential resistance pathways.

As demonstrated by Oz et al., laboratory evolution under controlled conditions reveals highly predictable mutational trajectories, whereas clinical evolution is shaped by fluctuating drug exposure, host immunity, population bottlenecks, and mixed infections that can obscure intermediate resistance states. Our model thus exposes the evolutionary potential of Burkholderia under sustained selection, even when such trajectories remain under-represented in clinical sampling.

Taken together, our findings suggest that DHFR may represent a strategic and repeatedly accessible target for resistance evolution in B. pseudomallei under sustained SXT pressure. Convergence toward a conserved biochemical resistance mechanism associated with moderate fitness costs, together with the availability of high-resolution structures of both wild-type and resistant DHFR variants, supports the potential of DHFR as a candidate for structure-guided drug discovery.

Importantly, recent studies have demonstrated that resistance-aware inhibitor design targeting DHFR is feasible. Zhang et al. identified novel DHFR inhibitors active against both wild-type and trimethoprim-resistant variants by integrating structure-based modeling and experimental validation, illustrating that evolved resistance mechanisms can be accommodated computationally. Broad mutational scanning of DHFR homologues across bacterial species has further mapped the fitness and resistance landscape of this enzyme, providing a valuable evolutionary framework for rational inhibitor design.

In parallel, advances in computational and artificial-intelligence–assisted drug design are increasingly being applied to antibacterial discovery. Recent reviews highlight how machine-learning approaches can accelerate hit identification, scaffold optimization, and resistance-aware prioritization of compounds. In addition, generative deep-learning platforms have been shown to produce structurally novel antibacterial molecules with predicted high affinity and activity against resistant pathogens, establishing a proof of principle for AI-driven antibiotic design.

Within this context, integrating the structural data generated here for wild-type and I99L Bt-DHFR into structure-based or AI-assisted pipelines could help prioritize antifolate candidates predicted to retain activity against recurrent resistance mutations. Continued investigation of compensatory pathways, such as FolM, and of efflux regulation will further refine our understanding of how resistance emerges in B. pseudomallei, stabilizes, and might ultimately be mitigated.

Materials and Methods

Bacterial Strain and Plasmids Used in the Study

Bacterial strains and plasmids used in this study are listed in Supporting Information 2, Tables S4 and S5, respectively. Manipulations of B. thailandensis were performed in a biosafety level 2 laboratory.

Experimental Adaptation of Burkholderia thailandensis under SXT Pressure

The E264 strain (wild-type) was streaked onto trypticase soy agar (TS, bioMérieux) and incubated overnight at 37 °C. A loopful of biomass was used to inoculate 5 mL of TS broth, which was incubated overnight at 37 °C with shaking (180 rpm). The resulting culture was then plated onto TS agar supplemented with 0.5 mg/L SXT (Sigma) (in a 1:19 ratio, of trimethoprim:sulfamethoxazole respectively). After overnight incubation, a portion of the bacterial growth was transferred with a sterile loop into 5 mL of TS broth containing the same SXT concentration. This alternating solid–liquid transfer procedure was repeated while progressively increasing SXT concentrations to 1, 2, 4, 8, 16, and 32 mg/L. At each selection step, an aliquot of the liquid culture was stored at −80 °C in 25% glycerol.

Genomic changes arising during adaptation were investigated by whole-genome sequencing of the evolved populations, with variant identification performed using the breseq pipeline; full sequencing and bioinformatic details are provided in Supporting Information 2.

Construction of the Bt-DHFR­(I99L) Strain through Directed Mutagenesis

The B. thailandensis strain carrying the I99L substitution in the dihydrofolate reductase–encoding gene (BTH_RS20745) was constructed by allelic exchange using the pDM4 suicide vector, following a strategy adapted from Chapartegui-González et al. The genomic region encompassing the target mutation was amplified from the population adapted to 32 mg/L trimethoprim–sulfamethoxazole using a high-fidelity DNA polymerase. The PCR fragment spanned approximately 2 kb and included 1 kb of homologous sequence on each side of the mutation. Primers were designed to introduce flanking regions homologous to the SacI and SmaI restriction sites of pDM4 (Supporting Information 2, Table S6).

The purified PCR product was assembled into the SacI/SmaI-digested pDM4 vector using using the NEBuilder HiFi DNA Assembly kit (NEB). The resulting construct was transformed into E. coli SM10 λpir and selected on chloramphenicol-supplemented agar (50 μg/mL). Plasmids recovered from transformants were verified by Sanger sequencing to confirm the presence of the I99L-encoding nucleotide substitution and the integrity of the cloned region.

The validated pDM4::Bt-DHFR­(I99L) construct was transferred into B. thailandensis E264 by biparental mating. Donor and recipient strains were mixed, deposited onto a 0.22 μm cellulose filter placed on nonselective agar, and incubated at 37 °C for 4 h to allow conjugation. Bacteria were then recovered and plated onto selective agar containing polymyxin B (50 μg/mL) and chloramphenicol (125 μg/mL) to select for B. thailandensis integrants resulting from a first homologous recombination event. Polymyxin B was used for counter-selection, as B. thailandensis is intrinsically resistant to polymyxin B whereas the E. coli donor strain is susceptible.

Chloramphenicol-resistant colonies were grown under nonselective conditions to permit a second recombination event leading to excision of the pDM4 backbone. Counter-selection was performed on 15% sucrose-containing medium, exploiting the presence of the sacB gene in pDM4. To enrich for clones retaining the I99L substitution, selection was carried out in the presence of trimethoprim (16 μg/mL). Sucrose-resistant colonies were screened for loss of chloramphenicol resistance and analyzed by PCR amplification of the BTH_RS20745 locus, followed by Sanger sequencing to confirm allelic replacement.

The reconstructed Bt-DHFR­(I99L) strain was further characterized phenotypically to verify the expected trimethoprim–sulfamethoxazole resistance profile. Whole-genome sequencing was finally performed to ensure that the engineered strain carried only the intended point mutation and did not harbor additional genomic alterations introduced during the allelic exchange procedure.

Antibiotic Susceptibility Tests

For bacterial populations evolved under SXT pressure, susceptibility to antibiotics was assessed using E-tests (bioMérieux) following the manufacturer’s recommendations. Because evolved cultures represent heterogeneous populations rather than clonal isolates, E-test was used as a comparative population-level readout during stepwise selection. MIC values obtained by E-test therefore reflect population-level susceptibility and should not be interpreted as clonal MIC values. The antibiotics tested included trimethoprim (TMP), sulfamethoxazole (SMX), trimethoprim/sulfamethoxazole (SXT, 1:19), ceftazidime (CAZ), nalidixic acid (NAL), doxycycline (DOX), chloramphenicol (CHL), gentamicin (GEN), ciprofloxacin (CIP), and meropenem (MEM).

For the wild-type and the reconstructed strain carrying the Bt-DHFR­(I99L) variant, minimum inhibitory concentrations (MICs) were determined using the microdilution method in accordance with CLSI recommendations. MIC assays were determined in cation-adjusted Mueller–Hinton broth (MH2) according to CLSI recommendations. In addition, susceptibility assays were performed in M9 minimal medium supplemented with glucose (20 mM) and MgSO₄ (2 mM). Because minimal medium lacks exogenous folate or related intermediates, susceptibility testing in M9 enables direct assessment of resistance phenotypes dependent on endogenous folate biosynthesis. These values are not intended for clinical interpretation but for mechanistic comparison. MIC values were read after 24 h of incubation in MH2 and after 48 h in minimal medium due to the slower growth rate observed in nutrient-limited conditions.

Maximal Growth Rate Determination (μ_max)

The maximal growth rate (μ_max) for the wild-type strain and the strain carrying the Bt-DHFR­(I99L) variant was quantified in three different media (MH2 (BD), TS (BD), and M9 supplemented with glucose 20 mM and MgSO₄ 2 mM (Sigma)). Growth curves were generated in a 96-well microplate using the serial dilution method. An initial inoculum of 10⁷ CFU/mL of bacteria was prepared in each medium and subjected to successive 2-fold dilutions using a multichannel pipet, resulting in six dilutions. Subsequently, 100 μL of each dilution was aliquoted in duplicate into a microplate. The microplate was then incubated at 37 °C with agitation at 216 rpm in a spectrophotometer (Tecan, Spark), and optical density at 600 nm was recorded every 5 min for 24 h. Three independent experiments were performed for each condition.

The generation time (Tg) of each growth condition was calculated in h^–1 using the standard formula:

$$\mathrm{T}(\mathrm{g})=\ln (2)/{\mu }_{\max }(\mathrm{hr})$$

Expression and Purification of Bt-DHFR

The Bt-DHFR- and Bt-DHFR­(I99L)-associated genes were cloned into NdeI and NotI sites of pET15blpa. This expression plasmid contains a T7 promoter and an N-Terminal His-Tag sequence followed by a thrombin site and a cloning multisite where additional restriction sites have been added as compared to pET15b (Supporting Information 2, Table S5). Cloning was performed by amplifying the gene coding the DHFR on the wild-type strain and the strain carrying the Bt-DHFR­(I99L) variant with primers containing 25 bases complementary to the plasmid, using the NEBuilder Hifi DNA Assembly Kit (Neb). NEB 10-beta Competent E. coli were transformed by the reaction and cultivated in the presence of 100 μM of ampicillin (Sigma). The sequence of the insets was verified by Sanger sequencing. Protein production was carried out from this plasmid using BL21-Rosetta cells (Novagen). Cells were grown in 250 mL of TBAI autoinducible medium (Formedium, England) at 37 °C until the midexponential phase and then transferred overnight at 18 °C. The pellet was resuspended in 40 mL of buffer A (10 mM Hepes–NaOH pH 7.5, 500 mM NaCl, 3 mM 2-mercaptoethanol, 0.1 mM PMSF, 0.1 mM benzamidine) and disrupted by sonication. The soluble fraction was then loaded onto a 4 mL column of Talon affinity resin (Clontech) and washed with buffer A supplemented with 10 mM of imidazole. Protein was then eluted with buffer A supplemented with 125 mM of imidazole. The eluted fraction was then supplemented with thrombin (Sigma, 0.5 unit per mg of eluted protein) and dialyzed overnight against 10 mM Hepes–NaOH pH7.5, 300 mM NaCl, 2.5 mM CaCl₂. The remaining tagged protein was then removed by flowing through Talon affinity resin. The preparation was finally polished by size-exclusion chromatography using a Superdex 75 HR (Cytiva) column in a buffer containing 10 mM Hepes–NaOH pH 7.5, 300 mM NaCl, and 3 mM 2-mercaptoethanol. This procedure yielded about 10 mg of Bt-DHFR. The Bt-DHFR concentration was determined by measuring the absorbance at 280 nm using a computed extinction coefficient of 15,470 M^–1 cm^–1. The procedure was identical for the Bt-DHFR­(I99L) variant, except that 5u of thrombin per mg was used for His-Tag removal. The yield was about twice smaller.

Enzymatic Assays of DHFR

The activity of DHFR was assayed spectroscopically by following the loss of absorbance at 340 nm due to the reduction of DHF and the oxidation of NADPH. A molar extinction coefficient of 11800 M^–1 cm^–1 was used to convert the A _340 nm decrease into substrate consumption. Reactions (300 μL) were assembled in buffer A (50 mM Hepes–HCl pH7.0 and 150 mM NaCl) directly in a spectrophotometer cuvette. In standard conditions, the concentrations of NADPH and DHF were 100 μM each. Reactions were started by adding the DHF component and incubated at 25 °C in a Cary 100 UV–vis spectrophotometer (Agilent Technologies). Absorbance at 340 nm was monitored as a function of time, and the initial rate of the reaction was deduced from the linear portion of the curve. The following extinction coefficients were used to determine the concentrations of the stock solutions: NADPH, 6220 M^–1 cm^–1 (340 nm); DHF, 7750 M^–1 cm^–1 (340 nm). k _cat and K_M values are reported as mean ± sd values from three independent experiments. Inhibition experiments were performed in the same conditions in the presence of either 2.7 or 100 μM TMP. Because of their semiquantitative character, the results of a single series of experiments are shown.

Thermal Shift Assays (Differential Scanning Fluorometry)

Experiments were done as described in 96-well plates using a Bio-Rad CFX-96 Real-Time PCR system. Reaction mixtures (25 μL) were assembled in 50 mM Hepes–HCl pH 7.0, 150 mM NaCl and contained Bt-DHFR (8 μM), 5× SYPRO orange (Invitrogen), and various concentrations of TMP (0.1 to 6.4 mM), with or without 100 μM NADPH. Temperature was increased by 0.5 °C every 60 s and fluorescence curves recorded. Data were processed using the Bio-Rad CFX Manager software. Tm values were determined as minima of the derivatives of the fluorescence vs temperature curves.

Crystallization and Structure Determination

Bt-DHFR (6.6 mg/mL) was supplemented with dithiothreitol (4 mM) and 1 mM of each dihydrofolate and trimethoprime. Crystallization attempts were performed at 6 °C using commercial screens (Hampton Research, Aliso Viejo, USA). Crystals were obtained in the presence of 30% PEG 8000 plus 0.2 M ammonium sulfate. For cryoprotection, the crystallization solution was supplemented with 0.15 volume of glycerol. Final concentrations in the cryoprotecting solution were 26.1% PEG 8000, 0.17 M ammonium sulfate, and 13% glycerol.

Crystals of Bt-DHFR belonged to the C2 space group (Supporting Information 2, Table S1) and diffracted to 1.85 Å resolution. The structure was solved by molecular replacement using Phaser and an AlphaFold 2 model of Bt-DHFR. Data were processed with XDS and programs of the CCP4 suite (CAD, CTRUNCATE, and FREERFLAG). The model of the Bt-DHFR:DHF:TMP complex was obtained through multiple cycles of refinement in Phenix and manual building in Coot. The asymmetric unit contained a single complex. Final electron density was of very good quality (Supporting Information 2, Figure S1).

We failed to obtain Bt-DHFR­(I99L) crystals in the same conditions as the wild-type enzyme. Crystals of the mutant enzyme were obtained at 18 mg/mL in the presence of dithiothreitol (4 mM) and 1 mM of each dihydrofolate and trimethoprime. The crystallization solution was 24% PEG 8000 plus 0.2 M ammonium sulfate and 0.1 M sodium cacodylate (pH 6.5). For cryoprotection, the crystallization solution was supplemented with 0.15 volume of glycerol. Final concentrations in the cryoprotecting solutions were 20.9% PEG 8000, 0.17 M ammonium sulfate, 0.087 M sodium cacodylate, and 13% glycerol. Crystals diffracted to 2.2 Å resolution. Reflection data could be indexed either in C222 (with cell parameters 118.5, 205.5, and 146.4 Å, at 90, 90, and 90°, respectively) or in P622 (with cell parameters 118.3, 146.1, and 146.1 Å, 90, 90, and 120°, respectively). In C222, a molecular replacement solution with five molecules in the asymmetric unit was found using Phenix.phaser, but refinement was stuck at R and free-R values greater than 32%. In P622, a molecular replacement solution was also found with two molecules in the asymmetric unit, but again, refinement could not be achieved to R and free-R values better than 0.32. L-test analysis of diffraction data in CTRUNCATE showed some deviation from what is expected for untwinned crystals. We therefore processed the data in P321 and P312, two space groups that can give rise to apparent P622 symmetry in case of merohedral twinning. Molecular replacement solutions could be obtained with five molecules in the asymmetric unit. Initial twinned refinement attempts gave better statistics in P312 (R = 0.261, free-R = 0.273) as compared to P321 (R = 0.276, free-R = 0.283). Refinement was therefore continued in P312 with merohedral twinning (twin law -h-k,k,-l with a fraction close to 0.5). The asymmetric unit contained five Bt-DHFR­(I99L):DHF:TMP complexes. Final electron density was of very good quality (Supporting Information 2, Figure S1). Data collection and refinement. Data collection and refinement statistics for the two structures are shown in Supporting Information 2, Table S1. Figure was generated using ChimeraX, and Figure was drawn with Espript.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsinfecdis.5c01132.

• Mutation data sets; growth curves and generation time data (XLSX)

• Crystallographic data collection and refinement statistics, structural contact analyses for DHFR and trimethoprim, and electron density maps, B-factor analyses, and crystal packing, bacterial strains, plasmids, and primers, whole-genome sequencing analysis (PDF)

The authors declare no competing financial interest.