A Chemical-Genomic Screen of Neglected Antibiotics Reveals Illicit Transport of Kasugamycin and Blasticidin S
Abstract
Fighting antibiotic resistance requires a deeper understanding of the genetic factors that determine the antibiotic susceptibility of bacteria. Here we describe a chemical-genomic screen in Escherichia coli K-12 that was designed to discover new aspects of antibiotic resistance by focusing on a set of 26 antibiotics and other stresses with poorly character- ized mode-of-action and determinants of resistance. We show that the screen identifies new resistance determinants for these antibiotics including a common signature from two antimicrobials, kasugamycin and blasticidin S, used to treat crop diseases like rice blast and fire blight. Following this signature, we further investigated the mechanistic basis for susceptibility to kasugamycin and blasticidin S in E. coli using both genetic and biochemical approaches. We provide evidence that these compounds hijack an overlapping set of pep- tide ABC-importers to enter the bacterial cell. Loss of uptake may be an underappreciated mechanism for the development of kasugamycin resistance in bacterial plant pathogens.
Introduction
The emerging threat of antibiotic resistance [1] necessitates new efforts and ideas to control bacterial pathogens. Mapping the determinants of antibiotic resistance in bacteria will be criti- cal for evaluating new antibiotics. In addition to the direct target of the antibiotic, drug efflux, drug permeability, and stress response pathways all contribute to resistance [2]. Global genetic approaches such as chemical-genomic screens, which measure the sensitivities of a large library of mutants to a set of stresses, can be a first-step in discovering resistance determinants and characterizing the mode-of-action of antibiotics. Chemical-genomic screens in the model bac- terium Escherichia coli K-12 [3–10] have already provided a critical resource for the bacterial research community and catalyzed insights into molecular systems critical for bacterial viabil- ity, stress survival, and resistance to antibiotics [11–15].Despite these successes, the chemical-genetic space of E. coli remains largely unexplored, as only slightly more than 50% of the genes in E. coli K-12 are “responsive”, defined as having a statistically significant fitness effect for at least one stress [8]. Reasoning that the remaining unresponsive genes may encode resistance determinants to previously untested antibiotics and stresses, we conducted a new chemical-genomic screen of the previously screened library [8,10] focusing on antibiotics with unique or unknown modes of action. We integrated the data from this screen with the results of Nichols et al. [8] to create an expanded chemical-genomics data- set that revealed new determinants of antibiotic resistance.From this dataset, we further investigated resistance to the antibiotics kasugamycin (Ksg) [16] and blasticidin S (BcS) [17]. Both antibiotics were discovered in the mid-20th century as antifungals effective against Magnaporthe oryzae, the causative agent of rice blast. Kasugamy- cin has a continuing use in the treatment of M. oryzae, bacterial pathogens of rice, and Erwinia amylovora, the bacterial pathogen that causes fire blight. We discovered that both antibiotics enter bacterial cells using illicit transport, the active uptake of non-physiological compounds, through two peptide ABC-importers. We suggest that loss-of-function mutations in homolo- gous peptide ABC-importers are likely to play a role in the development of kasugamycin resis- tance for E. amylovora and many other pathogens.
Results
We tested the sensitivities of 3975 mutants of E. coli K-12 to 57 stresses, split between new and previously screened conditions. The new stresses included neglected antibacterial compounds and other noxious chemicals with poorly characterized modes of action (Table 1). We pinned the arrayed mutant library onto agar plates containing each compound, imaged the plates after suitable colony growth, and quantified colony opacity using the image analysis software Iris [10]. We assigned fitness-scores to each mutant, using an in-house software package that builtupon previous analyses [8,18] by implementing additional filtering and normalization steps to improve data quality (Methods). These fitness-scores represent the statistical significance of a change in colony size for a particular condition, with negative and positive fitness-scores repre- senting sensitivity and resistance, respectively.After reanalyzing the original images from the Nichols et al. [8] screen with our improved workflow, we integrated both datasets. Fitness-scores from stresses present in both screens were significantly correlated (Fig 1A). A responsive gene is defined as having at least one condi- tional-phenotype in the dataset. We identified more than 5,000 conditional-phenotypes for the 26 new stresses, as well as more than 500 additional responsive genes from the 57 stresses tested (14% of the library) (Fig 1B). The conditional-phenotypes that identified additional responsive genes were spread evenly throughout the current screen, ranging from 5 to 54 con- ditional-phenotypes per condition from within the set of new responsive genes. The integrated dataset also contained more than double the number of statistically significant correlations between genes than each screen considered separately (Fig 1C).This increase in the number of significant correlations came from two factors. First, the inte- grated dataset captured more conditional-phenotypes that in turn drove higher correlations for some genes. Second, smaller datasets require more stringent cutoffs for statistical significance that exclude a sizable fraction of the correlations (S1 Fig).
Because of these factors, integration of chemical genomic screen with a larger dataset was critical for extracting as much responsive gene is defined as a gene with at least one phenotype in the dataset. (C) Significant correlations between genes were determined using a cutoff for Pearson’s correlation that allowed an FDR of 5%.The pathways that sense and respond to different stresses are nearly as diverse as the types of stress that are encountered. Genes that are involved in drug permeability, drug efflux and deg- radation, stress responses, and the drug target all contribute to resistance. The chemical-geno- mic screen revealed a global picture of antibiotic resistance that reflected this diversity in mechanism (Fig 2A).Growth at 10°C resulted in multiple sensitivities from genes in pathways known to be affected by cold shock, with almost double the number of cold-sensitive mutations as the next lowest temperature (16°C). 10°C-sensitive genes were most enriched for COG categories [19] related to translation (J, p = 0.005) and DNA replication and repair (L, p = 0.01), with 20% of the sensitive genes falling into one of these two categories (Fig 2A, S1 Table). Ribosome assem- bly and function is particularly sensitive to temperatures of 10°C and below [20], which may explain part of the expansion of sensitivities at this temperature.Deletions in the trans-translation complex (ΔssrA and ΔsmpB) were sensitive to aminogly- cosides and macrolides (Fig 2B), while the deletion of the alternative ribosome rescue factor (ΔarfA) was sensitized to members of the tetracycline family and chloramphenicol (Fig 2B). Deletions in the cytoplasmic protease HslUV (ΔhslU,V) and adaptors to the inner membrane protease FtsH (ΔhflC,K) were sensitized to blasticidin S (Fig 2B). These protease deletions have been demonstrated to be sensitive to aminoglycosides, both in E. coli [5,8] and in P. aeruginosa [21,22]. This set of pathways could be directly counteracting harmful effects of the translationinhibitors, with different pathways required to respond to unique mechanisms of action. As an example, deletion of the ribosome-bound ppGpp synthase (ΔrelA) resulted in sensitivity to both tRNA synthetase inhibitors used in the screen (Fig 2B).
RelA has a known role in sensing and responding to uncharged tRNAs in the A-site of the ribosome [23].Sensitivities from the screen also indicated that many of the antibiotics were subject to drug efflux and degradation (Fig 2A). Loss of AcrAB-TolC, the major efflux pump of E. coli [24], sensitized cells to gliotoxin, clindamycin, and pseudomonic acid A (Fig 2C). Other effluxpumps had more specific effects. The deletion strain ΔcrcB was sensitive to fluoride [25] while ΔmdtK was sensitive to serine hydroxamate [26]. The strain ΔyjjG was sensitive to 5-fluorouri- dine, consistent with the role of YjjG as a protective nucleotidase [27] (Fig 2C). For ΔcrcB, ΔmdtK, and ΔyjjG these chemical sensitivities were the strongest for each strain across the inte- grated dataset.Additionally, gene deletions within two peptide ABC-importers were resistant to the antibi- otics kasugamycin and blasticidin S. The ABC-importers of bacteria have been implicated in uptake of a diverse set of antibiotics, including blasticidin S, in a process termed illicit transport [28–32]. This phenotype was particularly interesting because of the agricultural importance of kasugamycin and because the uptake mechanism for these drugs has not been described in E. coli K-12. We therefore tested whether Ksg and BcS were directly imported by these peptide ABC-importers.Hierarchical clustering of the fitness scores revealed a potential connection between the trans- lation inhibitors kasugamycin (Ksg), an inositol-based aminoglycoside, blasticidin S (BcS), an aminonucleoside, (Fig 3A) and the major peptide ABC-importers of E. coli K-12; oligopeptide permease (Opp) and dipeptide permease (Dpp)[33–36] (Fig 3B). Dpp import is specific for dipeptides [37] while Opp can import peptides less than 5 amino acids in length [38].Cells harboring deletions of each component of the Dpp complex were resistant to Ksg and three clustered (ΔdppC, ΔdppD, and ΔdppF) with Ksg resistance as their major phenotype. Sim- ilarly, most deletions of Opp subunits were resistant to BcS and three clustered (ΔoppB, ΔoppD, and ΔoppF). In addition, deletions of the negative regulators of Opp and Dpp expression (ΔΔgcvB, Δhfq, and ΔgcvA) led to hypersensitivity to both Ksg and BcS.
We investigated these phenotypes further by constructing precise deletions of the ABC- importer operons (Δopp and Δdpp) and analyzing their phenotypes in MG1655, the standard wild-type background, growing in minimal media at a neutral pH. This growth condition enhances drug efficacy [16,17] and is associated with increased expression of opp and dpp [39,40]. Spot dilution tests revealed that the operon deletion strains grew equivalently to MG1655 in minimal media (Fig 3C, left), but were resistant to Ksg (Fig 3C, middle) and BcS (Fig 3C, right). Individual deletion strains Δopp and Δdpp were resistant to Ksg, but the highest level of Ksg resistance was conferred by the Δopp Δdpp double mutant. This suggested that both complexes participate in Ksg import. For BcS, Δopp alone was sufficient to confer high- level resistance.Consistent with elevated expression of the importers leading to sensitivity to Ksg and BcS, expression of the entire Opp operon from a plasmid (pOpp) was sufficient to confer sensitivity to both drugs (Fig 3D). Furthermore, the sensitivities of ΔgcvA and ΔgcvB to Ksg and BcS were completely suppressed in a Δopp Δdpp background (S2 Fig) confirming that these phenotypes were due to overexpression of Opp and Dpp.Both Ksg and BcS must transit through the inner membrane into the cytoplasm before they can bind to the ribosome and inhibit translation. The rate at which the drug enters the cyto- plasm is likely the rate-limiting step for drug action. If so, an in vivo assay quantifying the rate translation is inhibited after drug treatment is a reasonable proxy for the rate of drug import. We used the amount of 35S-methionine (35S-Met) incorporated in a 1 min pulse as a measure of translation rate, and then quantified its rate of decline after addition of antibiotic. We foundthat the rate of translation inhibition after addition of Ksg (Fig 4A) and BcS (Fig 4B) was dependent on the presence of the peptide ABC-importers, with the ABC-importer deletion strains showing a 7 to 10-fold slower rate of decrease in translation than MG1655.
In contrast, the rate of translation inhibition by kanamycin and spectinomycin was similar between MG1655 and the operon deletion strains (S3 Fig). This indicated that, like the antibiotic sensi- tivities, the effect of deleting these importers on antibiotic uptake was specific to Ksg and Bls. Finally, although less striking in magnitude, the rate of translation inhibition by kasugamycinwas significantly faster when the opp operon was overexpressed (p<0.02 Paired Student’s t-test) (Fig 4C). These results provided evidence that the observed antibiotic sensitivities were due to altered drug uptake.To further test the hypothesis that Ksg was being directly imported by Opp and Dpp, we used an in vivo substrate competition assay (Fig 5A) to test whether the presence of high affin- ity substrates of Opp (Pro-Phe-Lys; PFK) and Dpp (Ala-Ala; AA) [41,42] competed with Ksg for uptake. Indeed, when Ksg was co-administered with these substrates, the rate of translation inhibition by Ksg was slowed dramatically, approximating that of the Δopp Δdpp double mutant strain. In contrast, the inhibition rate of a Δopp Δdpp strain was virtually insensitive to the addition of competitors.We next used an in vitro binding assay to test for a direct interaction between Ksg and OppA. We expressed and purified His-tagged OppA and tested for Ksg binding using intrinsic tryptophan fluorescence. Addition of 1 mM Ksg markedly increased the apparent KD of OppAfor its high-affinity substrate PFK [43,44,42] (Fig 5B) but did not appreciably shift OppA fluo- rescence when added alone. A similar phenomenon has been shown for DppA, for which the relative change of fluorescence differs markedly even between natural high affinity peptidesubstrates [41]. While we cannot exclude the possibility that Ksg only indirectly alters OppA affinity for PFK in our in vitro assays, the increased apparent KD of OppA for PFK in the pres- ence of Ksg is consistent with competitive binding between Ksg and PFK. The solute binding protein (SBP) of each peptide ABC-importer freely diffuses in the peri- plasm, binding its substrate and delivering it to the pore of the complex (Fig 6A). In some cases, a single SBP has been shown to interact with the pores of multiple importers [45–47]. In particular, the SBP MppA has been shown to interact with both the Opp [46] and Dpp [47] pore in E. coli, depending on its substrate. Furthermore, DppA moonlights as the chemorecep- tor for the peptide chemotaxis system of E. coli [37]. We refer to this phenomenon as crosstalk. To test whether crosstalk between importers or to other systems contributed to the import pathway for Ksg or BcS, we measured genetic interactions between components of the two complexes using the gold-standard assay for drug efficacy: MIC changes determined from a liq- uid 2-fold dilution series (see Methods).We first validated that the MIC changes of the peptide importer mutants were consistent with observations from the spot test assay. The single mutant Δopp was sufficient to confer an increase in the MIC for blasticidin S by more than an order of magnitude. For kasugamycin, Δopp conferred a 2-fold MIC increase, and the double mutant Δopp Δdpp exhibiting a a 4-fold MIC (Table 2). Δdpp did not confer greater than a two-fold increase in MIC in this assay. Over- all, liquid MIC measurements were consistent with the original observation that BcS was imported through Opp while Ksg was imported through both Opp and Dpp.We then tested for crosstalk during illicit transport. We asked whether deleting the SBP had the same quantitative effect on drug efficacy as deleting the pore. If crosstalk occurred, deleting a single SBP would provide less resistance than deleting the pore because all SBPs that deliver that substrate to the same pore must be removed to completely eliminate transport. For Opp, deleting the SBP (ΔoppA), the pore (ΔoppB), or the entire complex (Δopp) all resulted in the same extent of resistance to both Ksg and BcS (Table 2). This is consistent with OppA deliver- ing Ksg and BcS to the Opp pore without crosstalk. Import of Ksg through Dpp also depended on DppA alone. Deleting both SBPs (ΔoppA ΔdppA) or both pores (ΔoppB ΔdppB) conferred the same degree of resistance to Ksg as deleting both operons (Δopp Δdpp), excluding a contri- bution from any additional unidentified SBPs in Ksg import (Table 2). We suggest a straight- forward model in E. coli K-12 in which Opp directly imports both Ksg and BcS, Dpp directly imports Ksg, and there is no crosstalk for either complex (Fig 6A). Discussion We report a chemical-genomic screen in E. coli K-12 focused on antibiotics with poorly charac- terized modes of action and determinants of resistance. We expect that this chemical-genomic dataset will function as a valuable community resource for generating new hypotheses based on these 26 stresses. Integrating the smaller chemical-genomic screen with a larger resource was critical for extracting more information from the new stresses, and this success sets a pre- cedent for continuing to add to the dataset to characterize new antibiotics and find leads for investigating gene function. Characterizing the mechanisms of resistance to new antibiotics will be valuable for future development by identifying those compounds least likely to face already prevalent antibiotic resistance. The ability to integrate smaller-scale screens with larger resources, which we demonstrate here, will facilitate the economical application of chemical- genomics in drug discovery pipelines. Starting with one feature in this screen, we describe the illicit transport of blasticidin S (BcS) and kasugamycin (Ksg) through the peptide ABC-importers oligopeptide permease (Opp) and dipeptide permease (Dpp). The peptide ABC-importer family is central to the uptake of multi- ple antibiotics in bacteria [28–32]. The flexible binding mechanisms used to accommodate the varied side chains of different peptides [49–51] could explain the susceptibility of these import- ers to being hijacked by so many illicit substrates. However, biophysical details of substrate binding by OppA and DppA are currently limited to natural substrates, so the nature of the interaction between these importers and their illicit substrates remains unclear. Purified OppA appears to have a weak affinity for Ksg, but Opp rapidly imports this compound in vivo none- theless. Further characterization of the binding determinants of OppA and DppA for Ksg could be highly informative for defining the minimal requirements of binding and import by these promiscuous complexes. The shared uptake pathway for Ksg and BcS is likely to be conserved among many of the pathogens against which these antibiotics are regularly used [52–54], although mechanistic details may differ from species to species. Both operon structure and sequence of the SBPs are highly conserved between E. coli and E. amylovora, but E. amylovora carries a duplication of oppA (EAMY_RS26335 and EAMY_RS26330) (Fig 6B) which may complicate the uptake mechanism. In Pseudomonas aeruginosa, there are five dppA homologues (dppA1-5) scattered across the genome and no clear homologue of the opp system. Despite this divergence, it was recently shown that both Dpp and a “third” peptide ABC-importer, Npp, contribute to BcS sensitivity (Fig 6C). The exact SBPs that participate in BcS uptake in P. aeruginosa are currently unknown [31]. There are no sequence homologues of either oppA or dppA in Magnaporthe ory- zae. However, strains isolated for resistance to either Ksg (B1-100-4 kas-3) [48] or BcS (Bu7) [55] can display cross-resistance. Given their distinct chemical structures, unique binding sites on the ribosome [56], and demonstration of a shared uptake mechanism in E. coli, this cross- resistance is most likely due to a simultaneous loss of uptake of both compounds. Indeed, a functionally equivalent importer that remains to be identified could be responsible for uptake of both Ksg and BcS in this fungal pathogen (Fig 6D). Cross-resistance from Ksg treated fields [57] suggests that loss of uptake is a common resis- tance mechanism in M. oryzae. However, naturally occurring Ksg-resistant ABC-importer mutants have yet to be isolated from any bacterial pathogens. Redundancy between import complexes may reduce the occurrence of this type of mechanism. The fitness impacts of delet- ing any of the opp or dpp genes are negligible within the integrated screen, but fitness effects have not been measured in the context of infection and these mutants may be quickly outcom- peted in this environment. In addition, there are a number of alternative resistance mecha- nisms for Ksg including efflux [58], chemical modification of the antibiotics [53], and alteration of the binding site [52,59]. Further research in these organisms will be critical for predicting the contribution that loss of uptake will have on resistance in an agricultural or med- ical setting. The mechanism we describe is an example of one discovery from our chemical- genomic screen with broad implications for antibiotic resistance in different species and we expect many more discoveries to be made from this large-scale dataset.Chemical sensitivity screens used LB Lennox agar plates (1% (w/v) tryptone, 0.5% (w/v) yeast extract, 90 mM sodium chloride, 2% (w/v) bacto agar) unless otherwise specified. M9 minimal plates used in the screen contained M9 salts, 0.2% (w/v) glucose, and 2% (w/v) bacto agar.Media for kasugamycin and blasticidin S sensitivity was M9 minimal supplemented with metal cations and buffered at pH 7.5 (M9 salts, 0.4% (w/v) glucose, 100 μM magnesium sulfate,100 μM calcium chloride, 5 μM iron(III) chloride, 20 mM Tris-HCl, pH 7.5). To promote hightranslation rates, media used for 35S-methionine incorporation was MOPS EZ rich (-Met), 0.4% (w/v) glucose (Teknova M2101, M2102, M2103, M2109, G0520).Ordered libraries grown during the chemical-genomics screens were incubated at 37°C until the majority of colonies reached a defined size (~8 hours) then a photograph was taken of the plate. Spot-dilution plates were grown roughly 24 hours at 37°C before a photograph was taken. For MIC determination, cultures were grown in deep 96-well plates for 24 hours in anInfors-HT shaker, 900 r.p.m., 37°C. For 35S-methionine incorporation, cultures were grown at 37°C in baffled flasks in a gyrotory1 water bath shaker at 350 r.p.m.Strains used in this study are listed (S2 Table). The KEIO deletion library is derived fromBW25113 (F- λ- Δ(araD-araB)657 ΔlacZ4787(::rrnB-3) rph-1 Δ(rhaD-rhaB)568 hsdR514) [60]. All experiments subsequent to the chemical-genomics screen were conducted using strains of MG1655 (F- λ- ilvG- rfb-50 rph-1) [61]. Operon deletions were generated using λ-red recombi- neering to replace the operon with a kanamycin resistance cassette amplified from pKD4. The resistance cassette was subsequently excised to generate deletions with a single FRT site as a scar. Plasmids and oligonucleotides used in the study are listed (S2 Table).The chemical genomics screen was conducted using the same methodology as reported previ- ously [8] with few modifications. Ordered libraries were arrayed on rectangular agar plates, grown in the presence of antibiotic or other stress until the colonies reached a defined average size, and then imaged. One condition, 4°C survival, involved growth of colonies on LB plates at 37°C for 6 hours, and transfer of the colony array to 4°C for 5 weeks. Colonies were then pinned onto a fresh plate and surviving cells were allowed to grow for 6 hours at 37°C before being imaged.Plate photographs were taken with a Canon Powershot G10, using an in-house assembly to control plate illumination. Images were analyzed using Iris to measure the total intensity of pix- els within the colony to calculate an opacity metric. Data filtering, normalization, and calcula- tion of the fitness-score was conducted using an in-house analysis pipeline. Code for the analysis is available online (https://github.com/AnthonyShiverMicrobes/fitness_score.git).Steps added to the original analysis pipeline [18] include simultaneous input and analysis of colony size, opacity, and circularity from Iris (read_data.m), manual removal of data based on plate position (to eliminate false positives from minor pinning problems) (filter_data.m), higher-order surface normalization (incorporating a quartic smoothing function that better describes the systematic errors due to pinning effects for E. coli) (smooth_data.m), power- transformation of the data (to reduce variability of extreme values) (transform_data.m), and variance normalization of the data (to improve reproducibility of measurements between plates) (RC4_scale.m). The raw data and metadata files for this analysis are available online (http://dx.doi.org/10.5061/dryad.f3kc0).Unreliable measurements were removed from the dataset at multiple points in the analysis and each condition had a different number of measurements (fitness-scores) that passed analysis. Before data integration and clustering were performed, conditions and strains with less than 75% reliable measurements were removed from analysis. Hierarchical clustering was per- formed using the Cluster 3.0 [62] command line interface, using Pearson’s correlation and average-value linkage. Data was visualized using Java Treeview [63].To predict reliable phenotypes and pairwise correlations, we used the method described by Nichols et al. [8] to determine the false-discovery rate (FDR), defined as the fraction of positive test results that are expected to be due to type I error, and set a cut-off for the fitness-score thatreflected an FDR of 5%. Using this method, 95% of the cutoff values for negative (sensitization) fitness-scores fell in the range (-2.0,-1.2) while 95% of the cutoff values for positive (resistance) fitness-scores fell in the range (+1.2,+2.1).To identify gene sets enriched for cold-sensitivity at 10°C, we used Gene Set Enrichment Analysis (GSEA) [64,65] of the fitness-scores for 10°C. We grouped genes according to the functional categories of their COG assignments. Input files for this analysis are available online at (http://dx.doi.org/10.5061/dryad.f3kc0).A liquid-broth dilution method was used to determine MIC values for the antibiotics. Fresh colonies were picked, resuspended in minimal media, and diluted to a final O.D.450 of 5x10-4. Antibiotics were added in a 1:2 dilution series spanning a 64-fold dilution range. After 24 hours of growth, 150uL of culture was transferred to a 96-well spectrophotometer plate, and the O.D.450 was measured using a Varioskan spectrophotometer (Thermo electron corpora- tion). After blank subtraction, culture densities were normalized to a no drug control, and the concentration of the first drug dilution step at which the normalized culture density fell below 10% was defined as the MIC of the drug. This quantitative measure corresponded well with a qualitative metric based on pelleting the cells and visually inspecting the size of cell pellet.Overnight cultures of relevant strains were grown inoculated into MOPS rich defined methio- nine dropout media (MOPS RDM-Met) at a starting O.D.450 of 5x10-3 and grown to an O.D.450 of 0.2. To quantify translation rate, 900 μL of culture was added to 30 μL of labeling mix (10 μCi L-35S-methionine, 50 μM cold L-methionine, in MOPS RDM-Met) (EasytagL-35S-Methionine, Perkin Elmer Corp.), incubated for 1 min in the water bath, then quenched with 100 μL of 50%(w/v) trichloroacetate (TCA) and stored on ice. A 100 μL aliquot ofthe quenched reaction mixture was deposited on a glass fiber filter (Merck Millipore Ltd.APFC02500) and washed with 10% (w/v) TCA followed by 95% (v/v) ethanol. 35S-methionine incorporation into the TCA-insoluble fraction was quantified on a scintillation counter (Beck- man Coulter LS6500 multipurpose scintillation counter). To follow translation inhibition, satu- rating concentrations of antibiotics were added to growing cultures. Translation rate was quantified in intervals following addition of drug and normalized to a timepoint taken two minutes before drug treatment. Data was fit to a double exponential decay function (kasugamy- cin treatment) or single exponential decay function (blasticidin S, kanamycin, spectinomycin) using QtiPlot [66].The oppA gene from E. coli MG1655 was amplified and cloned into pBAD22 [67] using NcoI and HindIII restriction enzymes. E. coli BW25113 was then transformed with the resulting plasmid, and over-production was induced in an exponentially growing culture at O.D.600 of0.6by addition of 0.05% (w/v) arabinose. The purification, including partial unfolding of the protein to remove bound substrates, as well as the binding assays followed the protocol in Klepsch et al. [42]. The changes made were clearing of the lysate for 1 hour at 140,000 g (Beck- man L8-M ultracentrifuge), replacement of guanidinium hydrochloride with urea for partial unfolding of OppA, and adjusting the 4-morpholineethanesulfonic acid (MES) buffer to a pH of 6.7 instead of 6.0 during purification to increase stability. OppA was highly concentrated and free of visible protein contaminations after Ni-IMAC as judged by SDS-PAGE and staining with Coomassie Blue, and was thus used in binding assays after extensive buffer exchange toremove the imidazole (Milipore Amicon Ultra centrifugation filters, 10,000 MWCO). The spectral characteristics of purified OppA were identical to those reported previously. All Fluo- rescence assays (Fluoromax-3, Jobin Yvon Horiba) were performed in 20 mM MES pH 6.0 and 150 mM sodium chloride, as described [42]. Data for the normalized change in intrinsic fluo- rescence was fit to a quadratic function that models ligand depletion Blasticidin S [68] in QtiPlot [66].