Identification of Efflux Substrates Using a Riboswitch-Based Reporter in Pseudomonas aeruginosa

ABSTRACT Pseudomonas aeruginosa is intrinsically resistant to many classes of antibiotics, reflecting the restrictive nature of its outer membrane and the action of its numerous efflux systems. However, the dynamics of compound uptake, retention, and efflux in this bacterium remain incompletely understood. Here, we exploited the sensor capabilities of a Z-nucleotide-sensing riboswitch to create an experimental system able to identify physicochemical and structural properties of compounds that permeate the bacterial cell, avoid efflux, and perturb the folate cycle or de novo purine synthesis. In the first step, a collection of structurally diverse compounds enriched in antifolate drugs was screened for ZTP (5-aminoimidazole-4-carboxamide riboside 5′-triphosphate) riboswitch reporter activity in efflux-deficient P. aeruginosa, allowing us to identify compounds that entered the cell and disrupted the folate pathway. These initial hits were then rescreened using isogenic efflux-proficient bacteria, allowing us to separate efflux substrates from efflux avoiders. We confirmed this categorization by measuring intracellular levels of select compounds in the efflux-deficient and -proficient strain using high-resolution liquid chromatography-mass spectrometry (LC-MS). This simple yet powerful method, optimized for high-throughput screening, enables the discovery of numerous permeable compounds that avoid efflux and paves the way for further refinement of the physicochemical and structural rules governing efflux in this multidrug-resistant Gram-negative pathogen. IMPORTANCE Treatment of Pseudomonas aeruginosa infections has become increasingly challenging. The development of novel antibiotics against this multidrug-resistant bacterium is a priority, but many drug candidates never achieve effective concentrations in the bacterial cell due to its highly restrictive outer membrane and the action of multiple efflux pumps. Here, we develop a robust and simple reporter system in P. aeruginosa to screen chemical libraries and identify compounds that either enter the cell and remain inside or enter the cell and are exported by efflux systems. This approach enables the development of rules of compound uptake and retention in P. aeruginosa that will lead to more rational design of novel antibiotics.

The ZTP (5-aminoimidazole-4-carboxamide riboside 59-triphosphate) riboswitch senses metabolic flux through the folate cycle and the de novo purine synthesis pathway (38). The folate pathway is constituted by six enzymes that convert GTP to tetrahydrofolate (THF) (39). THF then serves as a donor of one-carbon units in metabolic pathways that synthesize methionine, glycine, thymine, and purines, which are essential components of nucleic acids and proteins (40)(41)(42). The ZTP riboswitch tightly regulates these pathways by binding either ZTP or its precursor ZMP (5-aminoimidazole-4-carboxamide ribonucleotide) and activating the expression of the enzymes from these pathways (38).
Here, we describe the implementation in P. aeruginosa of a riboswitch reporter system which couples the ZTP riboswitch sequence from Pectobacterium carovotorum (described in reference 38) to the lacZ reporter gene. The reporter construct was introduced into isogenic efflux-proficient and efflux-deficient strains of P. aeruginosa and screened against a focused library of compounds enriched in antifolate drugs. By using an efflux-deficient strain, we identified all compounds that were cell permeable and inhibited enzymes in the folate pathway. We then retested these compounds against the efflux-proficient strain, allowing us to identify compounds that are likely substrates for efflux (Fig. 1). The initial use of an efflux-deficient strain allowed the detection of active, cell-permeable compounds that would have been missed by using an effluxproficient strain alone. These observations demonstrate the potential of our approach for establishing rules of compound uptake and retention in Gram-negative bacteria such as P. aeruginosa.

RESULTS
The Pectobacterium carovotorum ZTP riboswitch reporter responds to folate cycle inhibition in Pseudomonas aeruginosa. The ZTP riboswitch regulates the expression of genes involved in the folate cycle and the de novo purine synthesis pathway (38). To detect alterations of these pathways in P. aeruginosa, we developed a reporter system consisting of a translational fusion between the ZTP riboswitch sequence from Pectobacterium carovotorum and the lacZ reporter gene. This reporter construct was integrated into the attB site of the P. aeruginosa chromosome. The reporter fusion was expressed from the PexoT promoter ( Fig. 2A), which controls the expression of the type three secretion system (T3SS) effector ExoT (43). PexoT is activated by the AraC/XylS-type transcriptional regulator ExsA (44). The ExsA antiactivator ExsD (45,46) was deleted in all reporter strains to allow for constitutive expression from the PexoT promoter (Tables 1 and 2). Strains in which exsA was deleted, as well as strains harboring the M4 mutant allele of the ZTP riboswitch reporter (defective for ligand binding [38]), served as negative controls ( Fig. 2A, Tables 1 and 2). Importantly, the M4 variant riboswitch, which no longer binds ZTP, allows compounds that directly bind and activate the substrate recognition structure of the riboswitch to be differentiated from compounds that disrupt the folate cycle.
To examine the activity of the riboswitch reporter constructs, bacteria were exposed for 3 h to subinhibitory concentrations of trimethoprim (TMP, 200 mg/mL) before b-galactosidase activity was measured (Fig. 2B). TMP is a dihydrofolate reductase (DHFR) inhibitor (47) that disrupts the folate pathway in bacteria, which results in increasing concentrations of the purine biosynthesis intermediate ZTP within the bacterial cell (38). As observed in Fig. 2B, treatment with TMP significantly increased reporter activity in the PA14 DexsD attB::PexoT-ZTP-lacZ strain but failed to do so in the strain carrying the mutant ZTP riboswitch [PA14 DexsD attB::PexoT-ZTP(M4)-lacZ]. Reporter activity was not detected in PA14 DexsA strains following treatment with TMP, as predicted by the ExsA dependence of the PexoT promoter (44). These results demonstrate that the riboswitch reporter construct is responsive to the antifolate drug TMP when tested in P. aeruginosa and that the observed response is strictly dependent on ligand (ZTP) binding to the riboswitch aptamer region.
Absence of active efflux significantly increases ZTP riboswitch reporter sensitivity. To assess ZTP reporter behavior in an efflux-deficient background, we integrated the ZTP reporter into PA0397 (PAO1 DexsD DmexAB-oprM nfxB DmexCD-oprJ DmexJKL DmexXY DopmH362 DmexEF-oprN) (48). This mutant carries deletions in multiple efflux systems previously shown to promote antibiotic export (14,15,28,49,50) and exhibits an increased sensitivity to trimethoprim (TMP), with a MIC of 15 mg/mL, compared to the 450 mg/mL MIC observed for the PAO1 wild-type (WT) strain. We used lower concentrations of TMP (0.5 to 15 mg/mL) to test reporter activity in the efflux-deficient background (Fig. 2C). ZTP reporter activity was observed in the efflux-deficient strain in response to 0.5 mg/mL TMP, increased in a dose-dependent manner until 2 mg/mL Unpaired two-tailed t tests were used to compare the b-galactosidase values of cultures grown in M9 plus 1% CAA and cultures grown in M9 plus 1% CAA with TMP. ****, P , 0.0001. (C) b-Galactosidase activity of the PexoT-ZTP-lacZ reporter in efflux-deficient (black bars) and -proficient (teal bars) bacteria grown in M9 plus 1% CAA and exposed to TMP. Bars show the means 6 standard deviations from 3 independent experiments. Unpaired two-tailed t tests were used to compare the effluxdeficient and -proficient strains at various concentrations of TMP. **, P , 0.01; ***, P , 0.001; ****, P , 0.0001; ns, not significant.
Efflux Substrate Identification in P. aeruginosa mSphere TMP, and remained stable at higher concentrations. In contrast, reporter activity was not detected for the isogenic efflux-proficient strain at concentrations of TMP below 15 mg/mL (Fig. 2C). Similar levels of lacZ activity were measured for the efflux-proficient strain treated with 100 mg/mL TMP as for the efflux-deficient strain exposed to 2 mg/mL TMP. As TMP is reported to be a substrate for the MexAB-OprM, MexCD-OprJ, and MexEF-OprN efflux systems (28,(51)(52)(53), active efflux of TMP likely prevents folate cycle disruption and riboswitch activation at lower antibiotic concentrations, leading to the observed ;50-fold difference in reporter sensitivity between the two strain backgrounds. Development and validation of a high-throughput screening protocol. We next adapted the riboswitch-based reporter assay for high-throughput screening (HTS) applications. The tube-based b-galactosidase Miller assay (54) was transferred to a 384-well plate format in which P. aeruginosa bacterial cells were efficiently lysed using a mixture of the commercial detergent PopCulture and chicken egg white lysozyme (55,56). b-Galactosidase activity was assayed using the substrate ONPG (o-nitrophenyl-b-D-galactopyranoside), with the optical density at 420 nm (OD 420 ) measured every minute for 30 min and calculated as the enzymatic rate normalized by the number of cells present in each well (V max /OD 600 ).
A key requirement of a high-throughput screening (HTS) protocol is the ability to differentiate true active compounds (hits) from noise with confidence (57,58). Ideally, a reporter should exhibit low basal activity under noninducing conditions and high activity upon induction of the system. These traits ensure a high dynamic range of induction and minimize false-positive hits. Using TMP as a positive control, we optimized the bacterial growth phase for the 384-well format, varying both the duration of subculture (2 to 4 h) from overnight Source of inducible flp recombinase; Ap R (Cb R ) 8 3 cultures and the time of exposure to TMP (1 to 3 h) (Fig. 3). The highest fold change of induction between cultures treated with TMP versus untreated controls was observed with 3 h of subculture followed by 1 to 2 h of TMP exposure (Fig. 3). Based on these results, we subcultured bacteria for 3 h before exposing them for 1 h to test compounds in subsequent experiments. The Z9-factor screening coefficient described by Zhang and colleagues (58) assesses the ability of an HTS assay to identify true hits from a compound library. We determined the Z9-factor for our assay, using the 3-h subculture and 1-h treatment times established above, with TMP (1 mg/mL) and vehicle (1% dimethyl sulfoxide [DMSO]) serving as positive and negative controls, respectively. Three independent trials, carried out in triplicate, yielded Z9-factor values of 0.605, 0.792, and 0.695, indicating that our HTS screening approach can reliably discriminate between a positive hit and background noise (0.5 # Z9 , 1) (58).
The ZTP riboswitch reporter is responsive to trimethoprim analogs. To further validate the functionality of the PexoT-ZTP-lacZ fusion in P. aeruginosa, we exposed our reporter strains to a small library of structurally diverse compounds with known mechanisms of action and distinct biological targets (Table 3). This library was enriched with     Efflux Substrate Identification in P. aeruginosa mSphere dihydrofolate reductase and dihydropteroate synthetase inhibitors, as well as a range of other compounds currently in use in the clinic. The PAO1 Defflux strain carrying the PexoT-ZTP-lacZ reporter was treated with each compound at two concentrations, 10 and 50 mM. TMP was included as a positive control of induction (10 mM = 2.9 mg/mL and 50 mM = 14.5 mg/mL). Most tested compounds did not induce reporter expression and gave similar values of lacZ induction as the negative control (M9 1 1% Casamino Acids [CAA] DMSO = 1, denoted as a dotted line in Fig. 4). Only typical dihydrofolate reductase inhibitors, including the positive control TMP and compounds 2 to 5 and 7 to 12, induced reporter activity (Fig. 4A). Notably, for TMP and WR99210 (compound 4), reporter expression was higher at 10 mM than at 50 mM, suggesting that the higher concentration affected bacterial gene expression. Only methotrexate (compound 1), known to have poor antibiotic activity against whole cells despite being a potent inhibitor of the isolated enzyme (59,60), and compound 6, with a disubstituted benzene ring substitution pattern that is atypical for dihydrofolate reductase (DHFR) inhibitors, were inactive among the diaminopyrimidines (Fig. 4A). To confirm that reporter induction was exclusively due to ligand binding to the riboswitch aptamer, the positive hits identified in Fig. 4A were rescreened with the PAO1 D-efflux strain carrying the PexoT-ZTP(M4)-lacZ riboswitch reporter variant defective for ZTP binding. The absence of lacZ induction in this strain (Fig. 4B) confirmed that these compounds act by disrupting the folate pathway in P. aeruginosa and increasing ZTP/ZMP concentrations. All nominal dihydropteroate synthetase inhibitors, including dapsone (compound 13) and sulfamethoxazole (compound 16), were inactive, both at the screening concentrations and when subsequently tested at 250 mM and 500 mM (Fig. 4C). Lastly, all compounds that are not known folate pathway inhibitors-acetazolamide, glibenclamide, imatinib, lidocaine, phenytoin, proguanil, ribavirin, desloratadine, moroxydine, guanabenz, fludrocortisone acetate, nateglinide, telmisartan, aminoglutethimide, and floxuridine-were inactive in this assay.
The ZTP riboswitch reporter system can distinguish compounds that are subject to efflux versus those that are retained inside cells. The ZTP riboswitch reporter, expressed in efflux-deficient bacteria, identified several dihydrofolate reductase inhibitors capable of crossing the bacterial cell envelope and disrupting purine synthesis. To determine if the activity of these compounds was affected by the presence of efflux systems, we repeated our screen using the efflux-proficient strain carrying the ZTP riboswitch (Fig. 5A). In most instances, b-galactosidase expression was diminished in the efflux-proficient cells relative to their efflux-deficient counterparts, suggesting that these compounds were substrates for efflux. Three compounds, however, continued to show values of the fold change of induction of b-galactosidase expression of $3 in the effluxproficient strain: WR99210 (compound 4), desmethyltrimethoprim (compound 7), and baquiloprim (compound 11). We employed the PexoT-ZTP(M4)-lacZ mutant riboswitch construct to confirm that lacZ induction was dependent on ZTP binding to the riboswitch aptamer, allowing us to conclude that these compounds targeted the folate pathway in the efflux-proficient background (Fig. 5B).
We hypothesized that the measured differences in lacZ induction between efflux-deficient versus efflux-proficient bacteria reflected the relative retention of folate cycle-disrupting compounds in these otherwise isogenic bacterial strains. To test this directly, we extracted the cell-internalized compounds from efflux-proficient and efflux-deficient P. aeruginosa strains treated with 50 mM ormetoprim (compound 5) or desmethyltrimethoprim (compound 7) and measured their relative abundance by high-resolution liquid chromatography-mass spectrometry (LC-MS) (Fig. 6). The internal abundance of ormetoprim was greater in the efflux-deficient strain relative to the efflux-proficient strain, in good agreement with the ability of this compound to induce lacZ expression solely in the efflux-deficient reporter strain. In contrast, the internal abundance of desmethyltrimethoprim, which induces lacZ expression to a similar extent in both efflux-proficient and efflux-deficient reporter strains, did not differ significantly between these two backgrounds. These results suggest that the ability of a compound to induce lacZ expression in the ZTP riboswitch reporter system reflects its ability to both cross the cell envelope and avoid active efflux.

DISCUSSION
We present an experimental approach to identify folate pathway inhibitors that cross the bacterial envelope of P. aeruginosa and to classify them as substrates for  (Table 3). The dotted line represents the basal level of b-galactosidase activity of cultures grown in M9 plus 1% CAA and DMSO (control). Values are averages and standard deviations from 3 independent experiments. One-way ANOVA coupled with Dunnett's multiple comparison test was employed to compare the b-galactosidase activity values of cultures of efflux-deficient (black bars) and efflux-proficient strains (teal bars) exposed to different concentrations of each compound versus the control (M9 plus 1% CAA and 1% DMSO, dotted line). *, P , 0.05; **, P , 0.01; the absence of asterisks means that the difference was not significant.  (Table 3). For the efflux-deficient strain (black bars), most compounds were used at a final concentration of 50 mM, except for compounds 4 and TMP, which were used at a final concentration of 10 mM. The efflux-proficient strain (magenta and teal bars) was treated with both 50 and 100 mM each compound. Values are averages and standard deviations from 3 independent experiments. One-way ANOVA coupled with Dunnett's multiple-comparison test was employed to compare the b-galactosidase activity values of cultures treated with each of the compounds (black bars for the efflux-deficient strain; magenta and teal bars for the effluxproficient strain) with the control (M9 plus 1% CAA and 1% DMSO, labeled M9 DMSO). *, P , 0.05; **, P , 0.01; ****, P , 0.0001; ns, not significant. (B) b-Galactosidase activity of PAO1 DexsD PexoT-ZTP-lacZ (black and teal bars), and PAO1 DexsD PexoT-ZTP(M4)-lacZ (magenta and violet bars) exposed to the hits identified in panel A at final concentrations of 50 mM and 100 mM. The dotted line represents the basal level of b-galactosidase activity of cultures grown in M9 plus 1% CAA and 1% DMSO (control). Values are averages and standard deviations from 3 independent experiments. Unpaired two-tailed t tests were used to compare the b-galactosidase values of PAO1 DexsD PexoT-ZTP-lacZ with PAO1 DexsD PexoT-ZTP(M4)-lacZ cultures exposed to the compounds from the library in final concentrations of 50 or 100 mM. ****, P , 0.0001. Efflux Substrate Identification in P. aeruginosa mSphere efflux systems. This approach is robust enough to undertake a high-throughput screening campaign to elucidate the dynamics of compound uptake, retention, and efflux in this Gram-negative bacterium. The screening system employs a previously reported ZTP riboswitch-based reporter (32, 38) adapted to P. aeruginosa, with the reporter inserted in a single copy at a silent chromosomal location (attB) (61) and expressed constitutively and uniformly from a type three secretion system promoter (43) (Fig. 2A). The strength of our approach lies in the use of this reporter system in two isogenic backgrounds: an efflux-proficient PAO1 wild-type strain and an isogenic effluxdeficient mutant lacking multiple efflux systems, i.e., MexAB-OprM, MexCD-OprJ, MexJK, MexXY, and MexEF-oprN (48). These efflux systems belong to the resistance nodulation division (RND) family and are important contributors to the efflux of noxious compounds (28,(62)(63)(64). In the efflux-deficient strain, the reporter is activated by the control drug trimethoprim (TMP) at a concentration ca. 50 times less than that required to observe a similar effect in the efflux-proficient strain (Fig. 2C). These results are consistent with previous reports that identified TMP as substrate of the MexAB-OprM, MexCD-OprJ, and MexEF-OprN efflux systems in P. aeruginosa (reviewed in references 8, 28 and 65). Most importantly, this observation suggests that this reporter system can (i) identify compounds that cross the bacterial envelope and (ii) distinguish between compounds that remain inside the cell versus those that are efflux substrates. We subsequently adapted this reporter system for high-throughput screening (HTS) of compounds in a 384-well plate format. The method requires few hands-on manipulations, is straightforward to automate, and allows thousands of compounds to be screened per day. Although HTS assays employing lacZ-based reporters often rely on fluorogenic substrates such as 4-methylumbelliferyl-b-D-galactopyranoside (4-MUG) (31,32), this molecule does not cross the envelope of P. aeruginosa and requires cell lysis, while the high intrinsic fluorescence of P. aeruginosa PAO1 markedly reduces the S/N ratio of 4-MUG. We devoted substantial effort to developing an efficient P. aeruginosa lysis method compatible with polystyrene 384-well plates and requiring minimal Efflux Substrate Identification in P. aeruginosa mSphere handling. Of the different approaches tested, which included freeze/thawing cycles and lysozyme treatment (66), we found treatment with the proprietary detergentbased PopCulture reagent coupled with lysozyme-mediated lysis to be most efficient and most reproducible (Fig. 3). The assay performed well with a modest period of outgrowth (3 h of subculture) and gave a good response with only 1 h of compound treatment (Fig. 3). Under these conditions, the Z9-factor (58) ranged between 0.60 and 0.79, indicating a clear separation between the signals from the negative and positive controls (58). A restrictive bacterial envelope and the action of numerous efflux systems are key determinants for P. aeruginosa's increased basal resistance to many antimicrobials (8,(26)(27)(28)(29)67). One advantage of our screening approach is the utilization of a P. aeruginosa strain that lacks the main efflux systems responsible for antimicrobial export. As a consequence of these genetic modifications, this strain exhibits levels of antibiotic resistance that are more similar to the ones reported for Escherichia coli. For instance, the PAO1 D-efflux strain exhibits a MIC of TMP of 15 mg/mL, a value closer to the reported 0.12 to 0.5 mg/mL of E. coli (68,69) than to the 450 mg/mL observed for PAO1. The decrease in MIC is caused by the lack of efflux systems that can rapidly export TMP before it targets the folate pathway and inhibits bacterial growth. By integrating the ZTP riboswitch-based reporter into this efflux-deficient strain of P. aeruginosa, we could detect folate pathway disruption at much lower concentrations of TMP (Fig. 2C) and potentially identify compounds that would appear inactive against an efflux-proficient strain. Further, the sensitivity of the riboswitch to perturbations of folate cycle homeostasis allowed compounds to be detected at concentrations far below those that affected viability-in the case of TMP, ;30-fold below the MIC in both wild-type and efflux-deficient backgrounds (Fig. 2C). This provides a significant increase in sensitivity over traditional bacterial viability assays and the ability to identify permeating compounds even at the fixed (and relatively low) concentrations employed in highthroughput screens.
By employing a small library of bioactive small molecules enriched in folate pathway inhibitors, we could determine if, in P. aeruginosa, the ZTP riboswitch reporter is responsive to other antifolate drugs. Dihydropteroate synthetase (DHPS) inhibitors, predominantly sulfonamides, impair the incorporation of p-aminobenzoic acid into dihydropteroic acid, and dihydrofolate reductase (DHFR) inhibitors, predominantly diaminopyrimidines, impair the reduction of dihydrofolic acid into tetrahydrofolic acid, both compounds targeting key steps in the biosynthesis of tetrahydrofolic acid (70,71). Although the ZTP reporter in E. coli responds to both DHPS and DHFR inhibitors (32), the reporter in P. aeruginosa showed increased expression only when exposed to DHFR inhibitors (Fig. 4A). No induction was observed when P. aeruginosa was exposed to the DHPS inhibitors in the library, which included most class members currently in use in the clinic. Efflux-mediated intrinsic resistance of P. aeruginosa to both sulfonamide and diaminopyrimidine drugs has been previously reported (52). In contrast with this previous report, our data suggest that in the case of sulfonamides, such resistance is mostly mediated by poor permeation rather than by the action of active efflux, as efflux-deficient and -proficient strains showed similar expression levels of the reporter when challenged with sulfonamides, even at high concentrations (Fig. 4C).
Most compounds that strongly induced the ZTP riboswitch reporter in the effluxdeficient strain had a minimal effect on the efflux-proficient strain (Fig. 5A), suggesting that these molecules were substrates for active efflux. This included DHFR inhibitors which see concurrent clinical use, albeit typically not for P. aeruginosa infection, such as trimethoprim (TMP), pyrimethamine (compound 2), diaveridine (compound 3), ormetoprim (compound 5), and metoprine (compound 10). In contrast, three compounds in our library appeared to be exported less efficiently than the others: WR99210 (compound 4), desmethyltrimethoprim (compound 7), and baquiloprim (compound 11) (Fig. 7). This riboswitch-based observation was validated by LC-MS experiments that measured the relative intracellular abundance of a compound that our riboswitch assay categorized as an efflux substrate, ormetoprim (compound 5), or nonefflux substrate, desmethyltrimethoprim (compound 7). While it remains possible that WR99210, desmethyltrimethoprim, or baquiloprim are simply poor upregulators of efflux, we think this less likely given the brief period of compound exposure (1 h) in our assay. DHFR inhibitors have not been reported to induce efflux system expression in P. aeruginosa, and only two efflux pumps have been identified to be induced by antimicrobial agents: MexXY by ribosome-binding inhibitors such as tetracycline and chloramphenicol (72,73), and MexCD-OprJ by membrane damaging agents such as chlorhexidine and polymyxin B (74).
Consideration of the physical properties of the compounds that are not exported reveals no obvious trends relative to those that are. WR99210, initially developed as a potential antimalarial (75) but seeing no clinical use, has a calculated LogP (cLogP; 2.47) and a total polar surface area (TPSA) (95 Å 2 ) well within the ranges of those exported (cLogP, 0.98 to 3.07; TPSA, 77 to 104 Å 2 ). Similarly, baquiloprim, used as a veterinary anti-infective (76), has a cLogP of 2.64 and TPSA of 92 Å 2 . While desmethyltrimethoprim, recently noted as impeding antibiotic resistance evolution (77), is more polar (cLogP, 0.65; TPSA, 115 Å 2 ), the difference from trimethoprim (cLogP, 0.98; TPSA, 104 Å 2 ) is relatively small. It is therefore likely that differentiation between exported and not-exported compounds occurs at the protein-ligand interaction level.
We predict that screening thousands of compounds with this approach will lead to the discovery of numerous permeable compounds that are minimally exported by efflux systems in this Gram-negative bacterium, thus allowing investigators to identify pharmacophores and physicochemical properties that determine if a given compound will be exported by P. aeruginosa's most important efflux systems or if it will remain inside the cell. Identification of such structural properties can inform the design of drugs able to bypass two of the most efficient defenses against antimicrobial compounds in Gram-negative bacteria: the restrictive nature of the bacterial envelope and the action of efflux pumps (reviewed in reference 8, 9, 11, 12, 14, 63, 78, and 79).

MATERIALS AND METHODS
Bacterial strains, media, culture conditions, and chemicals. The bacterial strains and plasmids used in this study are listed in Table 1. Lysogeny broth (LB) was routinely used as the liquid medium for both E. coli and P. aeruginosa. M9 minimal medium supplemented with 1% Casamino Acids (M9 plus 1% CAA) was used in specific cases. LB plates contained agar at a 15-g/L final concentration. Vogel-Bonner minimal medium (VBM) (80) plates (15 g/L agar) were used to select P. aeruginosa in specific cases. Cultures were grown at 37°C with shaking at 250 rpm in a New Brunswick Innova 44/44R shaker. When required, antibiotics were added at the following concentrations: for E. coli, 15 mg/mL gentamicin, 10 mg/mL tetracycline; for P. aeruginosa, gentamicin (30 mg/L for the efflux-deficient strain or 100 mg/mL for the efflux-proficient strain), tetracycline (20 mg/mL for the efflux-deficient strain or 200 mg/mL for the efflux-proficient strain), and carbenicillin (75 mg/mL for the efflux-deficient strain or 200 mg/mL for the efflux-proficient strain). All bacterial strains were stored at 280°C as 15% (vol/vol) glycerol stocks. Trimethoprim (TMP), used as a positive control for the ZTP riboswitch reporter, was purchased from MP Biochemicals and dissolved in DMSO. The complete library of compounds selected for this study was purchased from Cayman Chemical, MedChemExpress, Ambeed, Toronto Research Chemicals, Ambinter, and Sigma-Aldrich. They are listed in Table 3 with molecular weight and function; these compounds were also dissolved in DMSO. A 20% sodium dodecyl sulfate (SDS) solution was purchased from AmericanBIO. The chromogenic substrate ortho-nitrophenyl b-D-galactopyranoside (ONPG) and chicken egg white lysozyme were purchased from Sigma-Aldrich. PopCulture reagent was purchased from Millipore.
Strain construction. The PCR primers employed for strain construction and verification were synthesized by the Keck facility (Yale University) and are listed in Table 2. Unmarked, in-frame deletion of exsD was carried out in PAO1 and its isogenic strain PA0397 (48) via allelic exchange as described in reference 81. Briefly, primers were designed to target the genomic regions upstream (Up-F and Up-R) and downstream (Down-F and Down-R) of the desired deletion (see Table 2). attB1 and attB2 sites were added to the 59 ends of the Up-F and Down-R oligos, respectively. Up-R and Down-F oligos were designed to be complements of one another. Phusion polymerase (NEB) was used to amplify the regions flanking the exsD gene, which were then spliced together by overlap extension PCR (SOE-PCR), generating a linear DNA fragment containing the desired deletion. This DNA fragment was cloned into the Gateway suicide vector pDONRX (82) and verified by sequencing using primers attL1-F and attL2-R ( Table 2). The resulting plasmid was transformed into E. coli S17-1 and mobilized into P. aeruginosa by mating. Pseudomonas aeruginosa merodiploids were selected on VBM plates with either 30 or 100 mg/mL gentamicin, depending on whether the recipient strain was efflux deficient (PA0397) or efflux proficient (WT). A second recombination event, loss of vector backbone, was selected by streaking merodiploids on VBM plus sucrose (10%). DexsD candidates that were both gentamicin and sucrose sensitive were isolated and screened by PCR using primers external to the deleted region (denoted E1 and E2, see Table 2) and confirmed by Sanger sequencing.
A DNA fragment consisting of the 112-bp promoter region of exoT (44), the 83-bp ZTP riboswitch sequence from Pectobacterium carovotorum subsp. carovotorum PC1 plus the first 17 nucleotides (nt0 of the rhtB gene-coding sequence (38), and the complete lacZ gene (3,078 bp) was synthesized (Genewiz, Inc.) and cloned into mini-CTX2, allowing for integration into the attB site of the P. aeruginosa chromosome (61). This mCTX2-PexoT-ZTP-lacZ plasmid was transformed into E. coli S17-1 and mobilized by mating into P. aeruginosa. Integrants were selected on VBM plates with 20 or 200 mg/mL tetracycline, depending on whether the recipient strain was efflux deficient (PA0397) or efflux proficient (PAO1 or PA14 WT). The mCTX2 vector backbone was excised by mating with E. coli SM10 carrying the pFLP2 vector, which expresses Flp recombinase (83). Successful loss of vector backbone and pFLP2 was confirmed by PCR using attB-SER-F and neolacZ primers (Table 2) for candidates that were both tetracycline and carbenicillin sensitive.
A previously described mutated version of the ZTP riboswitch (M4) (38) was constructed by PCR amplification of the mCTX2-PexoT-ZTP-lacZ plasmid using the three GA primer pairs (Table 2) followed by NEBuilder assembly (NEB). The resulting PexoT-ZTP(M4)-lacZ reporter fusion was integrated in the P. aeruginosa chromosome as described above and confirmed by PCR and Sanger sequencing.
b-galactosidase assays. Overnight cultures of P. aeruginosa carrying the riboswitch reporter fusion were diluted 1:100 into 2.5 mL of M9 plus 1% CAA medium and incubated in 14-mL tubes with aeration at 37°C until the early exponential phase (2 h, OD 600 , ;0.06 to 0.1). At this time, cultures were treated with TMP over a range of concentrations (0.5 to 300 mg/mL). The final concentration of DMSO was 1%. Colorimetric b-galactosidase assays were performed as previously described (84). Briefly, two 200-mL aliquots of each culture were collected: one was employed to determine the OD 600 , while the second aliquot was permeabilized in 1.5-mL microcentrifuge tubes containing 600 mL of Z-buffer (60 mM Na 2 HPO 4 Á 7H 2 O, 40 mM NaH 2 PO 4 Á H 2 O, 10 mM KCl, 1 mM MgSO 4 Á 7H 2 O, pH 7), 15 mL of 0.01% SDS, and 30 mL of chloroform. The b-galactosidase enzymatic reaction was initiated by adding 200 mL of ONPG 4 mg/mL as substrate; to stop the reaction, 500 mL of 1 M Na 2 CO 3 was added. Then, 200 mL of the reaction supernatant was taken to measure the OD 420 and OD 550 . Enzymatic activity was determined using the following formula: LacZ activity = [(OD 420 Â 1000) -(1.75 Â OD 550 )]/[OD 600 Â reaction time (min) Â culture volume (mL)] (84).
Determination of MICs of trimethoprim. P. aeruginosa was grown with aeration in M9 plus 1% CAA to the mid-log phase, and then samples containing ;1 Â 10 5 CFU were transferred to 14-mL tubes containing known concentrations of trimethoprim. After 16 h of incubation with aeration at 37°C, growth was scored (68).
384-well plate-based b-galactosidase assay (development of a high-throughput screening protocol). Overnight cultures of the PAO1 DexsD Defflux attB::PexoT-ZTP-lacZ strain were diluted 1:100 in 20 mL of fresh M9 medium plus 1% CAA and incubated in 125-mL flasks with aeration at 37°C for 2, 3, or 4 h (OD 600 , ;0.06, ;0.2, or ;0.6, respectively). At this point, the cultures were divided in two. One fraction was treated with 1 mg/mL (or 3.4 mM) TMP dissolved in DMSO. The vehicle DMSO was added to the second fraction (negative control). Then, 30 mL of each of the two fractions was aliquoted in 384-well plates and incubated at 37°C with aeration for 1, 2, or 3 h. After each exposure time, the OD 600 was recorded using a Tecan Infinite 200 Pro plate reader. Subsequently, bacteria were lysed by adding a mixture of 4.5 mL of PopCulture reagent and 0.5 mL of lysozyme 4 U/mL in each well; the mixture was incubated at room temperature for 20 min with 750 rpm shaking to guarantee more efficient lysis. Following bacterial lysis, 40 mL of 1.46 mg/mL ONPG substrate dissolved in Z-buffer was added to each well. Immediately after this, the plate was placed in the plate reader prewarmed at 28°C, and OD 420 reads were taken each minute for 30 min. A 5-s shaking step (1 mm amplitude linear shaking) was set between readings.
LacZ activity was determined following a previously described protocol (66), with some modifications. Briefly, the increase of OD 420 in time (minutes) was plotted for each well, and the reaction rate (V max ) was determined by extrapolation of the slope from the most linear part of the curve. These values were then normalized to the OD 600 recorded for each well before bacterial lysis. LacZ activity values were expressed in V max /OD 600 units. To estimate the values of the fold change of induction, the absolute LacZ activity values from the wells treated with TMP were divided by the values from the wells treated with DMSO. Z9-factor calculation was done as described by Zhang and colleagues (58).
Screening of a focused library of compounds. To screen the focused collection of compounds, overnight cultures of P. aeruginosa carrying the riboswitch reporter fusion were diluted 1:100 in 20 mL fresh M9 plus 1% CAA medium and incubated with aeration at 37°C for 3 h. At this point, 30 mL of the cultures was aliquoted into 384-well plates containing aliquots of each compound to a final concentration of 10 and 50 mM (for efflux-deficient strains), and 50 and 100 mM (for efflux-proficient strains). Subsequently, the plates were incubated at 37°C with aeration for 1 h. After incubation with each of the compounds, wells were treated as described above, with the exception that the efflux-proficient strain was lysed adding 9 mL PopCulture and 1 mL lysozyme 4 U/mL in each well (that is, twice the volume employed for the efflux-deficient strain), and the lysis incubation time was 60 min. b-Galactosidase activity quantification of each well was calculated as before.
Determination of intracellular levels of ormetoprim and desmethyltrimethoprim. Overnight cultures of the PAO1 DexsD Defflux attB::PexoT-ZTP-lacZ and the PAO1 DexsD attB::PexoT-ZTP-lacZ strains were diluted 1:100 in 60 mL of fresh M9 medium plus 1% CAA and incubated in 250-mL flasks with aeration at 37°C until the early exponential phase (3 h, OD 600 , ;0.2). At this point, the cultures were divided into three 20-mL fractions. These three fractions were treated with 50 mM solution of ormetoprim, 50 mM solution of desmethyltrimethoprim, or vehicle (1% DMSO), respectively. After 1 h of incubation at 37°C with aeration, the cultures were pelleted and washed twice with 20 mL 1Â phosphate-buffered saline (PBS). After the second wash, the pellets were dried in vacuo, resuspended in 1 mL of an acetonitrile:methanol:water (2:2:1) solution, and sonicated for 10 min. The mixtures were centrifuged, and the supernatants were transferred to glass vials and dried in vacuo. The dried samples were resuspended in 200 mL of a water:methanol (1:1) solution, and 3 mL of each sample was subjected to high-resolution liquid chromatography-mass spectrometry (LC-MS) analysis (Agilent iFunnel 6550 quadrupole time-of-flight [QTOF], positive-mode electrospray ionization). Chromatography was performed on a Kinetex 5-m C 18 100 Å column (250 by 4.6 mm) with a water: acetonitrile gradient containing 0.1% formic acid at 0.7 mL/min: 0 to 30 min, 5 to 50% acetonitrile. Using the Agilent MassHunter quantitative analysis software, extracted ion chromatograms were generated with 10 ppm mass windows around the calculated exact masses of protonated ormetoprim and desmethyltrimethoprim (m/z 275.1503 and m/z 277.1295, respectively). The areas of the peaks corresponding to the compounds were integrated and normalized to the dry weight of each sample.
Statistical analysis. GraphPad Prism software version 9.3.0 was used for statistical analysis. Twotailed Student's unpaired t tests were used to compare means between treatments (TMP or selected library compounds versus the vehicle solution DMSO) and between the efflux-deficient and -proficient strains treated with TMP. To compare means between strains treated with each of the library compounds and strains treated with DMSO, one-way analysis of variance (ANOVA) coupled with Dunnett's multiple-comparison test was employed. Unpaired t tests were also used to compare the relative abundance of the compounds ormetoprim and desmethyltrimethoprim between efflux-proficient and effluxdeficient strains. Z9-factor calculation was done as described in Zhang et al. (58).