Deacylated tRNA Accumulation Is a Trigger for Bacterial Antibiotic Persistence Independent of the Stringent Response

ABSTRACT Bacterial antibiotic persistence occurs when bacteria are treated with an antibiotic and the majority of the population rapidly dies off, but a small subpopulation enters into a dormant, persistent state and evades death. Diverse pathways leading to nucleoside triphosphate (NTP) depletion and restricted translation have been implicated in persistence, suggesting alternative redundant routes may exist to initiate persister formation. To investigate the molecular mechanism of one such pathway, functional variants of an essential component of translation (phenylalanyl-tRNA synthetase [PheRS]) were used to study the effects of quality control on antibiotic persistence. Upon amino acid limitation, elevated PheRS quality control led to significant decreases in aminoacylated tRNAPhe accumulation and increased antibiotic persistence. This increase in antibiotic persistence was most pronounced (65-fold higher) when the relA-encoded tRNA-dependent stringent response was inactivated. The increase in persistence with elevated quality control correlated with ∼2-fold increases in the levels of the RNase MazF and the NTPase MazG and a 3-fold reduction in cellular NTP pools. These data reveal a mechanism for persister formation independent of the stringent response where reduced translation capacity, as indicated by reduced levels of aminoacylated tRNA, is accompanied by active reduction of cellular NTP pools which in turn triggers antibiotic persistence.

the genes that encode RelA or SpoT and so does not appear to utilize the stringent response (20,24). The Chlamydia persisters form by a molecular mechanism similar to the one described previously in which their trigger for persistence seems to be a direct effect from slowing down translation via accumulation of deacylated tRNA but independent of the stringent response.
Several studies have shown that the accuracy and efficiency of aminoacyl-tRNA synthesis are critical determinants of bacterial homeostasis (25)(26)(27). To investigate how this global role of aaRSs might impact persistence, mutations in pheS and pheT, which encode the a-subunit and b-subunits of PheRS, respectively, were investigated in both E. coli MG1655 relA 1 and DrelA backgrounds. Changes were made in both the active site and editing sites of PheRS to investigate the effects of varying aminoacylation efficiency and accuracy, respectively. The PheRS editing site is used to clear misacylated Tyr-tRNA Phe and m-Tyr-tRNA Phe . m-Tyr is a product of Phe oxidation and, in the absence of PheRS editing, has been shown to be mistranslated, causing cytotoxicity and other growth defects in E. coli (28,29). PheRS quality control is also important beyond its primary role in maintaining translation accuracy, as demonstrated by the finding that when m-Tyr-tRNA Phe is not hydrolyzed by PheRS, the stringent response is suppressed (25). Changes in the active site of PheRS led to a significant decrease in the amount of aminoacylated tRNA Phe in response to amino acid stress, and this also resulted in a significant increase in antibiotic persistence but only in the strain that is unable to mount the stringent response. These data indicate that disruption of bacterial homeostasis via both reduced translation quality control and suppression of the stringent response together increases antibiotic persistence.

RESULTS
Changes in the active site of PheRS perturb discrimination against noncognate amino acids. It has been previously shown that an editing-deficient PheRS E. coli variant, pheT G318W, had significant effects on cellular homeostasis under oxidative and amino acid stress (25,29). This strain is not able to edit Tyr-tRNA Phe or m-Tyr-tRNA Phe , and consequently m-Tyr, a nonproteogenic amino acid, was shown to be incorporated into the proteome and cause cytotoxicity (28,29). This PheRS editing-deficient strain was also shown to not activate the stringent response upon amino acid stress by m-Tyr addition (25). Taken together, these previous findings provided a basis to investigate the effects of PheRS quality control on bacterial antibiotic persistence and its dependence on the stringent response.
Along with an E. coli PheRS editing-deficient strain, additional E. coli mutant strains were made which chromosomally encode PheRS active site variants in both MG1655 relA 1 and DrelA backgrounds (29)(30)(31). The two active site mutations that were made on the E. coli chromosome encode the A294G and A294S variants in the a-subunit of PheRS. The A294G replacement in the active site of PheRS is predicted to have reduced amino acid substrate discrimination while the A294S replacement is predicted to have increased discrimination (32). These replacements were also made in recombinant E. coli PheRS to allow determination of the amino acid activation kinetics for the PheRS active site variants ( Table 1). The specificity of Phe-to-m-Tyr for wild-type (WT) PheRS is 22, and for aA294G PheRS, it is 1.8, confirming that editing is required to prevent misacylated m-Tyr-tRNA Phe accumulation and that aA294G PheRS can only minimally discriminate between Phe and m-Tyr. In contrast, the specificity of Phe-to-m-Tyr for aA294S PheRS is 370, indicating that this mutant is not able to efficiently activate m-Tyr for aminoacylation. The same trend follows for the specificity of Phe-to-Tyr: aA294G PheRS has a reduced substrate specificity compared to wild-type PheRS, and aA294S PheRS has an increased specificity. Bacterial antibiotic persistence increases when quality control is present, and the stringent response is disrupted. To investigate if differences in noncognate substrate discrimination by PheRS variants affect antibiotic persistence, minimum duration of killing (MDK) assays were performed in both MG1655 relA 1 and DrelA E. coli backgrounds (see Fig. S1 in the supplemental material). The four different PheRS strains (wild-type pheS/pheT, pheT G318W, pheS A294G, and pheS A294S) in both backgrounds were grown to early log phase and then treated with 100 mg/ml ampicillin for 3 h; each hour, an aliquot was taken out, washed with sterile phosphate-buffered saline (PBS), plated on LB plates, and incubated at 37°C overnight to calculate CFU. These assays were performed in two different types of media. The first, medium A, is the control medium which is a supplemented M9-based minimal medium that contains 40 mg/ml of all 20 proteogenic amino acids. The second, medium B, is the starvation medium which is also a supplemented M9-based minimal medium that contains 40 mg/ml of 18 proteogenic amino acids, 10 mg/ml Tyr, 40 mg/ml m-Tyr, and no Phe. The total amounts of persisters for wild-type PheRS, editing-deficient PheRS, and the two different active site PheRS mutant strains in both E. coli backgrounds were calculated at the endpoint after 3 h of exposure to ampicillin (Fig. 1). There was a significant increase in persisters for pheS A294S in the DrelA background grown in medium B compared to all the other PheRS strains that were tested. Three different variables were tested in this assay: relA either present or knocked out, different pheT and pheS mutations, and nutrient limitations. Each of the variables had an independent effect on persistence. For example, the deletion of relA caused an increase in persistence when the other two variables remained constant (compare solid blue bar and striped blue bar in WT pheS/pheT data set in Fig. 1). When the three different variables are combined, deletion of relA, PheRS quality control, and amino acid starvation, the effect on persistence is compounded and the largest amount of persister cell formation is observed. These data indicate that this pathway for bacterial antibiotic persistence is independent of the RelA-dependent stringent response. FIG 1 Antibiotic persistence increases when quality control is present, and the stringent response is disrupted. Persistence assays were done with wild-type pheS/pheT, pheT G318W, pheS A294G, and pheS A294S in both E. coli MG1655 relA 1 (solid bars) and a XDrelA -36292-8262-3643 strain (striped bars). Persistence was measured and quantified after 3 h of exposure to 100 mg/ml ampicillin and is shown on a log scale set relative to wild-type pheS/pheT in the relA 1 background grown in medium A. Medium A (blue) is a supplemented M9 minimal medium that contains 40 mg/ml of all 20 proteogenic amino acids, and medium B (green) is a supplemented M9 minimal medium with 40 mg/ml of 18 proteogenic amino acids, 10 mg/ml Tyr, 40mg/ml m-Tyr, and no Phe. Error bars represent standard deviations from 3 biological replicates. WT pheS/pheT relA 1 in medium A and medium B is significant to pheS A294G DrelA in medium B with a P value of ,0.03. All data sets are significant to pheS A294S DrelA in medium B with a P value of ,0.0001, except for pheS A294G DrelA in medium B with a P value of 0.0008. Statistical analysis was performed using two-way analysis of variance (ANOVA).
A pheT G318W/pheS A294G double mutant was made in both the E. coli MG1655 relA 1 and DrelA backgrounds. These strains would be able to synthesize, but not edit, m-Tyr-tRNA Phe and Tyr-tRNA Phe and therefore would generate high levels of misacylated tRNA Phe in the cell. The pheT G318W/pheS A294G double mutation allowed normal growth compared to the wild-type PheRS in medium A in both the relA 1 and DrelA backgrounds. However, we were not able to perform persister assays on the pheT G318W/pheS A294G/DrelA mutant because the cells showed a substantial growth defect in medium B (Fig. S2).
Levels of deacylated tRNA Phe correlate with the levels of persisters for the active site mutants of PheRS. To investigate if there is a correlation between the levels of persistence and the levels of deacylated tRNA Phe , assays were performed to quantify tRNA Phe aminoacylation in all four PheRS strains in both relA 1 and DrelA E. coli backgrounds grown in either medium A or medium B (Fig. 2). A representative Northern blot that was probed with a 32 P-59-end-labeled oligonucleotide that is specific for tRNA Phe is shown ( Fig. 2A), and quantification from biological triplicates was performed (Fig. 2B). Overall, there was no difference in the levels of aminoacylated-tRNA Phe across all 4 different PheRS strains in the relA 1 background grown in either medium A or medium B. The only instances in which significantly decreased levels of aminoacylated-tRNA Phe were observed were for the DrelA pheS A294G and DrelA pheS A294S mutant strains grown in medium B, the same conditions that also gave rise to the highest observed levels of persistence in this study.
Intracellular nucleotide concentrations indicate a reduction in metabolic activity when persistence increases. The intracellular nucleotide concentrations were determined by letting the wild-type pheS/pheT and all three mutants in both the relA 1 and DrelA backgrounds grow in medium A and medium B to early log phase. Metabolites were extracted and then analyzed by liquid chromatography-mass spectrometry (LC-MS). There was a Cultures were grown to late log phase in either medium A (solid bars), which is a supplemented M9 minimal medium that contains 40 mg/ml of all 20 proteogenic amino acids, or medium B (striped bars), which is a supplemented M9 minimal medium with 40 mg/ml of 18 proteogenic amino acids, 10 mg/ml Tyr, 40 mg/ml m-Tyr, and no Phe. Error bars represent standard deviations from 3 biological replicates. *, P value , 0.04; statistical analysis was performed using multiple t tests.
Deacylated tRNA Is a Trigger for Persistence ® general trend of having a lower intracellular nucleotide concentration in the strains that showed elevated levels of persistence (Table 2), indicating a lower level of metabolic activity in these cells. These strains all have quality control present, but their stringent response has been disrupted. Bacterial antibiotic persister cells are able to evade killing by antibiotics because they have a low metabolic state and antibiotics mostly target actively growing cells. Also, the strains that resulted in very low levels of persistence had a higher intracellular concentration of ppGpp. The stringent response is activated by RelA in response to amino acid stress by synthesizing the alarmone ppGpp, so these cells were able to respond to the nutritional stress and did not enter into a persistent state. The highest levels of antibiotic persistence that were observed were in the strains that have quality control present and are in the DrelA background. This would indicate that these persister cells are independent of the production of ppGpp. The editing-deficient mutant, pheT G318W, in the DrelA background grown in medium B had elevated intracellular levels of AMP and GMP, indicative of ATP and GTP hydrolysis compared to when this strain was grown in medium A. These data are broadly inversely correlated with the changes in levels of persistence ( Fig. 1), indicating that when quality control is absent, the ability to enter into persistence is significantly reduced.
The active site mutant, pheS A294S, in the DrelA background grown in medium B had the highest level of antibiotic persistence and also had a consistently lower nucleotide pool than the other pheS A294S strains. It was also observed that when the stringent response is suppressed, which correlated with higher levels of persistence, there is a lower concentration of the secondary messenger cyclic di-GMP. These data support the idea that the strains that enter into antibiotic persistence have an overall lower intracellular metabolite pool.
Proteome homeostasis is disrupted in strains that have a higher rate of antibiotic persistence. Total proteome analyses were performed by using wild-type pheS/pheT and pheS A294S strains in both the relA 1 and DrelA backgrounds and grown in medium B to mid-log phase. After the proteins were digested, the smaller peptides were injected for LC-MS/MS analysis. There were a total of 2,351 proteins quantified across all samples including PheS, PheT, and PheA. These 3 proteins either had no change or were slightly overexpressed in the strains tested, which indicates that the expression of these proteins has not been affected by our strain construction or growth conditions. Overall, proteome homeostasis was disrupted for all strains compared to the wild-type pheS/pheT in the relA 1 background (see Fig. S3A to E and Data Set S1 in the supplemental materials). Upon further investigation, we choose to focus the data analyses on the wild-type pheS/ pheT in the relA 1 and DrelA background and pheS A294S in the DrelA background. The differentially expressed proteins for wild-type pheS/pheT between relA 1 and DrelA ( Table S1) and pheS A294S mutant in the DrelA background versus the wild-type pheS/ pheT in the relA 1 background (Table S2) were used to develop a protein-protein interaction (PPI) network (Fig. 3).
The PPI network showed several proteins that were differentially expressed in pathways for cellular amino acid biosynthesis, drug metabolism, and nucleotide FIG 3 Protein homeostasis is disrupted during bacterial antibiotic persistence. Differential protein expression protein-protein interaction (PPI) network was performed for wild-type pheS/pheT and pheS A294S in relA 1 and DrelA backgrounds grown in medium B. The differentially expressed proteins are represented by their fold change for either DrelA wild-type pheS/pheT versus relA 1 wild-type pheS/pheT (inner ring) or DrelA pheS A294S versus relA 1 wildtype pheS/pheT (outer ring). The E. coli gene name is shown in the node, and cellular processes for amino acid biosynthesis (green), drug metabolism (yellow), and nucleotide biosynthesis (pink) are highlighted. Edges are derived from previously determined protein-protein interactions within StringDB. Deacylated tRNA Is a Trigger for Persistence ® biosynthesis (represented by the green, yellow, and pink nodes, respectively, in Fig. 3). In general, the proteins involved in amino acid biosynthesis were downregulated in the pheS A294S strains (represented by the outer ring in Fig. 3) and were either slightly downregulated or had no change in the wild-type pheS/pheT (represented by the inner ring in Fig. 3). These results are consistent with pheS A294S having a higher fraction of persisters than the wild-type pheS/pheT in the DrelA background compared to the relA 1 background (Fig. 1). Strikingly, MazF and MazG were both enriched in all strains compared to the wild-type pheS/pheT in the relA 1 background. MazF is an RNase toxin that comprises part of the MazEF toxin-antitoxin module which has been previously studied in persister formation in E. coli (33,34). MazG is a broad-specificity nucleoside triphosphatase (NTPase) that modulates MazEF and is involved in regulating cell survival under amino acid starvation conditions (35,36).

DISCUSSION
Aminoacyl-tRNA synthetase quality control is a determinant for antibiotic persistence. Aminoacyl-tRNA synthetase quality control is important for accurate synthesis of the proteome. When quality control is absent, misacylation of tRNAs can occur, which was once thought to always be detrimental to the cell, but it is now understood that in certain cases a low level of mistranslation can be beneficial for the cell to adapt to a new or challenging environment (27,37). It was previously observed that when PheRS quality control is abolished by the use of a PheRS editing-deficient E. coli strain, pheT G318W, the cells were not able to activate the stringent response (25). The stringent response is critical for bacteria to withstand amino acid stress, and it is activated when deacylated-tRNA enters the A-site of the ribosome which signals for RelA to begin synthesizing the alarmone (p)ppGpp (38). When the PheRS editing-deficient strain is grown under starvation conditions and in the presence of m-Tyr, a nonproteogenic amino acid that is produced from the oxidation of Phe, misacylated m-Tyr-tRNA Phe falsely inhibits the stringent response from being triggered because it senses starvation via deacylated-tRNA and not misacylated-tRNA (25). The stringent response has been thought to be a determinant for antibiotic persistence; however, there have been a few studies where persistence has been shown to be stringent response independent (12,13,20).
In this study, we have investigated the effects of PheRS quality control on bacterial antibiotic persistence in E. coli. Along with the editing-deficient PheRS, we chose to study two different active site PheRS mutants, aA294G and aA294S. It is assumed that the aA294G mutation would have decreased substrate discrimination and the aA294S mutation would have increased discrimination (32). The aA294G PheRS is unable to discriminate between Phe and m-Tyr and has a 12.7-fold reduction in discrimination for Tyr compared to wild-type PheRS. However, aA294S PheRS has a 16.8-fold increase and a 2.6-fold increase of discrimination for m-Tyr and Tyr, respectively, compared to wild-type PheRS (Table 1). With these kinetic data, it can be hypothesized that aA294G PheRS will be able to efficiently synthesize m-Tyr-tRNA Phe ; however, because editing is intact, the misacylated m-Tyr-tRNA Phe will be moved into the editing site of PheRS where it will be hydrolyzed (see Fig. S4A in the supplemental material). aA294G PheRS also has decreased efficiency for phenylalanine adenylation, and so, in addition to quality control, this mutant also fails to efficiently synthesize Phe-tRNA Phe , and this would also contribute to increased levels of deacylated tRNA Phe . With the increased discrimination of aA294S PheRS, it will not be able to synthesize m-Tyr-tRNA Phe (Fig. S4B). In both of these scenarios, the amount of the intracellular deacylated tRNA Phe would increase, and under starvation conditions, it would activate the stringent response.
Three different PheRS-encoding mutations, pheT G318W, pheS A294G, and pheS A294S, were made on the chromosome of E. coli MG1655 and the corresponding DrelA strain. Intriguingly, there was a reduction in persistence in the pheT G318W DrelA strain grown under amino acid limitation ( Fig. 1 and Fig. S1). In this strain, the misacylated m-Tyr-tRNA Phe is not edited and so the stringent response is not triggered since the amount of intracellular deacylated tRNA Phe does not change. This correlates with our previous finding that when pheT G318W is unable to edit m-Tyr-tRNA Phe this resulted in the stringent response being unable to be activated (25). These results seem to indicate that when the intracellular concentration of deacylated tRNA Phe increases and a stringent response is able to be mounted, the cell responds appropriately; however, when RelA is not present, and the stringent response is disrupted, the cell can directly enter into a persistent state.
Deacylated tRNA triggers antibiotic persistence independent of the stringent response. Taken together, the kinetic data and the antibiotic persistence data suggest that the intracellular level of deacylated tRNA may ultimately be a trigger for persistence. To investigate this further, Northern blot analysis was performed on all 4 PheRS strains in both the relA 1 and DrelA E. coli background, which did confirm that the amount of aminoacylated tRNA Phe significantly decreased in both pheS A294G and pheS A294S in the DrelA background when grown in medium B (Fig. 2). This decrease in aminoacylated tRNA Phe correlates with these two strains displaying the greatest persistence. Our data support that the increase in persistence in the DrelA E. coli background is consistent with an increase in intracellular deacylated tRNA Phe concentrations and is independent of the stringent response. We attempted to further investigate persistence and the connection to deacylated tRNA accumulation using a pheS A294G/pheT G318W double mutant, but this strain had a severe growth defect when grown in medium B, and this growth defect was even more pronounced in the DrelA background (Fig. S2).
Other studies have proposed that when aaRS activity is restricted or inhibited, the amount of deacylated tRNA in the cell would increase and this may be the trigger for antibiotic persistence, consistent with our data shown here (16,(20)(21)(22). In several previous studies, it was shown that GluRS was inactivated by phosphorylation via the toxin HipA, leading to limited tRNA Glu aminoacylation and increased levels of persistence. When this occurs, the amount of deacylated tRNA Glu increases and it enters the A-site of the ribosome, triggering the stringent response via RelA; thus, this mechanism of persistence is RelA dependent (16,22). Interestingly, a few of these studies were conducted in bacteria that lack RelA and have a disrupted stringent response. One study was conducted using Chlamydia, a Trp auxotroph, and the authors used indolmycin to inhibit TrpRS, which led to a decrease in aminoacylated tRNA Trp that reduced translation rates and increased persistence. Since Chlamydia does not encode RelA or SpoT, this mechanism of persistence is stringent response independent (20). A few studies have been conducted in Caulobacter crescentus, which relies solely on SpoT to regulate the stringent response which requires both amino acid limitation and carbon or nitrogen starvation. In one study, HipA was able to phosphorylate both GluRS and TrpRS, which inhibited their aminoacylation activity, and upon a further carbon or nitrogen stress, SpoT would synthesize (p)ppGpp and the cells would enter into a persister state (21). In another study, the HipA phosphorylation of TrpRS was further investigated, and it was found that as the levels of Trp increase in the cell because it is not being used to aminoacylate tRNA Trp , this led to the inhibition of GlnE, which reduced the amount of glutamine production, creating a nitrogen shortage. This imbalance in amino acids ultimately resulted in an increase in persister cells that is SpoT dependent (39).
There has also been a recent study showing that defects in tRNA, such as a lack in methylation, seem to have a link to antibiotic resistance and persistence (40,41). In the present study, the highest levels of persistence were associated with increased levels of deacylated tRNA Phe and when RelA was deleted. SpoT is encoded in our DrelA strains, accounting for our ability to detect ppGpp. In all the DrelA strains, ppGpp levels remained steady regardless of amino acid limitation, suggesting that SpoT activity was not significantly altered when persistence was increased.
Changes in metabolism reveal a cellular reprogramming in antibiotic persister cells. Metabolomics and proteomics were performed in order to further understand the metabolic state of the different strains that were used in this study. The metabolomics data showed an overall trend of decreased intracellular concentrations of nucleotides compared to the wild-type strain ( Table 2). For example, the strain with the highest level of persistence, DrelA pheS A294S grown in medium B, had a reduced concentration of the nucleotides tested except for cyclic di-GMP and ppGpp compared to the other pheS A294S strains. This is consistent with antibiotic persister cells having a low biochemical and metabolic state because they are in a dormant, nongrowing form. There was also a general trend of AMP and GMP having a higher intracellular concentration compared to their di-and triphosphate purine nucleotides, which is also consistent with persistent cells being in a low energetic state.
The proteomics data showed that cellular homeostasis was disrupted for wild-type pheS/pheT and pheS A294S in the DrelA background compared to wild-type pheS/pheT in the relA 1 background when grown in medium B (Fig. 3). From the PPI network, it was observed that most of the proteins that are involved in amino acid biosynthesis were underrepresented compared to wild-type pheS/pheT in the relA 1 background which is consistent with the cells entering into a dormant state. Furthermore, from the proteomics data PheA was overexpressed in pheS A294S in the DrelA background. Expression of PheA is required for the biosynthesis of phenylalanine and is regulated by transcription attenuation by the synthesis of the leader peptide PheL. Under normal conditions, PheRS maintains the level of Phe-tRNA Phe for the attenuation of pheA transcription (25,42). However, pheS A294S in the DrelA background led to an increased amount of deacylated tRNA Phe when grown in medium B, and this also led to the increased expression of PheA, which further supports our model. Furthermore, PheA had normal levels of expression when comparing the proteomics data of wild-type pheS/pheT in the DrelA background compared to the relA 1 background, indicating that deletion of relA does not affect the transcription of pheA.
The toxin MazF was significantly enriched in both wild-type pheS/pheT and pheS A294S in the DrelA background, and the antitoxin MazE was not detected above the limit of detection. MazEF is a type II toxin-antitoxin module in which when the labile antitoxin, MazE, is degraded in the cell MazF can cause toxicity by cleaving mRNA, which may occur independent of translation (43). The regulator of the MazEF module, MazG, was also upregulated. MazG is an NTP hydrolase that can cleave (p)ppGpp produced by RelA or SpoT and also hydrolyze the NTP substrates for the synthesis of (p) ppGpp (35,36). Since the upregulated MazG was observed in the DrelA background, it could be hypothesized that the (p)ppGpp that MazG would be degrading is being produced from SpoT. It is also possible that MazG might be degrading NTPs to prevent the synthesis of (p)ppGpp, since it was observed from the metabolomics that ATP and GTP concentrations were reduced in these strains. Furthermore, to ensure that the increase in the expression levels of MazF and MazG was not an artifact of transcriptional read-through from the disruption of relA using a kanamycin cassette, sequencing of this operon was performed (data not shown). We confirmed that both the 59 untranslated region (UTR) and 39 end of relA are intact and that the 59 region of mazEF, which contains both of its promoters, was not disrupted by the kanamycin insertion into the relA gene. Both of these promoters are required for the autoregulation of mazEF, and when MazE is degraded, MazF is released as a toxin (44,45). The promoter for mazG was also not disrupted, which indicates that the increased levels of expression for both MazF and MazG are due to the cells having entered into an antibiotic persister state. There were also changes in the levels of production for numerous proteins involved in fatty acid synthesis, purine biosynthesis, and the tricarboxylic acid cycle, which have previously been shown to be disrupted when bacterial persistence is triggered (46)(47)(48). It should also be noted that for both the metabolomics and proteomics experiments, the entire bacterial population was used for the metabolite and protein extractions, so the quantifications are an average of the population. Since bacterial persisters represent only a small fraction of the entire population (;1 to 10% in this study), these subtle changes in the metabolite concentration and protein expression may be more pronounced in the persister cell. Naturally occurring or spontaneous persister cells occur at a frequency of ;1 Â 10 26 , and mistranslation occurs every ;1 Â 10 24 codons (27,49). As these small frequencies in error begin to accumulate, so the rate of bacterial persister formation increases.
Our data reveal a mechanism for persister formation where reduced translation capacity is accompanied by accelerated depletion of cellular NTP pools, which in turn triggers antibiotic persistence. Our data further show that the frequency of persister formation via this mechanism is significantly influenced by the specificity and efficiency with which substrates for translation are synthesized. While in our study changes in PheRS specificity and efficiency were achieved via genetic manipulation, nutrient depletion can also have comparable effects on aminoacyl-tRNA synthesis in wild-type cells (50). This mechanism of bacterial persister formation is ppGpp independent. Our study and several other studies have recently been challenging the role of RelA and ppGpp in persister formation. Instead, changes in amino acid concentrations, inhibition of aaRSs, and slowed translation are all mechanisms now known to be able to induce bacterial antibiotic persistence. Taken together, these and previous findings suggest that small amino acid imbalances, while not significantly impacting the population as a whole, could lead to heterogenous quality control outcomes that affect a small number of cells, thereby triggering the formation of a subpopulation of antibiotic persisters.
ATP/PP i exchange. Detailed description of the methods for ATP/PP i exchange to determine steadystate kinetics of amino acid activation are in Text S1 in the supplemental material.
Growth medium. Lysogeny broth (LB) was made with 0.5% NaCl, 0.5% yeast extract, and 1% tryptone. A supplemented M9 minimal medium that was used for persister assays was developed based on a previous protocol (14). A 2Â supplemented M9 minimal medium was prepared as follows: 2Â M9 salts, 0.8% glucose, 4 mM MgSO 4 , 0.2 mM CaCl 2 , 2mg/ml thiamine, 1 mg/ml FeSO 4 solution, 80 mg/ml all proteogenic amino acids except for phenylalanine and tyrosine. For the persister assays, two different media were used, medium A and medium B. Medium A contained 1Â supplemented M9 minimal medium, 40 mg/ml phenylalanine, and 40 mg/ml tyrosine. Medium B contained 1Â supplemented M9 minimal medium, 10 mg/ml tyrosine, and 40 mg/ml m-tyrosine (no phenylalanine was used in this medium).
Minimum duration of killing (MDK) persister assays. MDK persister assays were performed as described in several studies (3,4,14,54). Overnight cultures were grown in LB and were then diluted 1:100 into 6 ml of either medium A or medium B in culture tubes. The cultures were grown at 37°C while shaking at 250 rpm until they reached early exponential phase (OD 600 = 0.2 to 0.3). At this point 1 ml of culture was taken out, serially diluted, and plated on LB plates to count CFU; 100 mg/ml ampicillin was added to the remaining culture; and growth continued at 37°C with shaking. Every hour for 3 h, 1 ml of culture was removed, washed in sterile phosphate-buffered saline (PBS) twice to remove the antibiotic, Deacylated tRNA Is a Trigger for Persistence ® serially diluted, and plated on LB plates. To calculate persisters, the CFU of survivors after ampicillin treatment was divided by the CFU of the culture before ampicillin treatment.
Quantification of aminoacylated and deacylated tRNA. Purification of total tRNA, acid/urea gel electrophoresis, and Northern blotting methods are described in detail in Text S1.
Targeted metabolomic analysis and quantification. Metabolites were extracted from exponentially growing E. coli cells as described previously (55,56). Detailed methods for the metabolite extraction are described in Text S1.
LC-MS analysis of the metabolites was performed as described previously (57). The dried metabolites were suspended in 50 ml of MilliQ water and centrifuged at 16,000 Â g for 30 min at 4°C, and the supernatant was transferred to an LC vial. Separation of metabolites was performed on a Thermo Scientific UltiMate 3000 ultra-high performance liquid chromatography system equipped with a zwitter-ionicphosphorylcholine hydrophilic interaction liquid chromatography, 150-by 2.1-mm, 3-mm column with a flow rate of 0.15 ml/min and a column temperature of 37°C; mobile phase A was 10 mM ammonium acetate, pH 6, and mobile phase B was 10% 10 mM ammonium acetate, pH 6, 90% acetonitrile. Chromatography gradient was as follows: isocratic 100% mobile phase B for 3 min, linear gradient to 20% B for 22 min, linear gradient to 100% B for 1 min, isocratic 100% B for 1 min, linear gradient to 20% B for 8 min, linear gradient to 100% B for 5 min, and isocratic 100% B for 15 min. Standard curves with a mixture of ATP, ADP, AMP, GTP, GDP, GMP, cyclic di-AMP, cyclic di-GMP, and ppGpp ranging from 0 to 5 ppm spiked with 10 ppm internal standard mix were done before and after each experimental set to ensure column integrity. The MS was performed on a Thermo Scientific TSQ Quantiva (triple-stage quadrupole MS) with a spray voltage of 3,500 V in the negative ion mode, ion vaporizer at 50°C, and ion transfer tube at 350°C. Analysis and quantification were performed using Xcalibur data acquisition and interpretation software. Intracellular concentrations were calculated assuming an OD 600 of 1.0 equals 8 Â 10 8 cells and that the volume of one E. coli cell growing in exponential phase equals 1 Â 10 215 liter (58,59). This was then normalized by the extraction efficiency for [ 13 C]ATP and [ 13 C]GTP.
Protein digestion and mass spectrometry. (i) Digestion of intact E. coli for shotgun proteomics. Twenty-milliliter cultures were inoculated to a starting OD at 600 nm of 0.01 in either medium A or medium B using an overnight culture to stationary phase. After reaching mid-log, cells were chilled on ice and pelleted by centrifugation for 2 min at 8,000 rpm. The resulting pellet was frozen at 280°C for downstream processing. For cell lysis and protein digest, cell pellets were thawed on ice and 2 ml of cell pellet was transferred to a microcentrifuge tube containing 40 ml of lysis buffer (10 mM Tris-HCl, pH 8.6, 10 mM dithiothreitol [DTT], 1 mM EDTA, and 0.5% antilymphocyte serum [ALS]). Cells were lysed by vortex for 30 s, and disulfide bonds were reduced by incubating the reaction mixture for 30 min at 55°C. The reaction was briefly quenched on ice, and 16 ml of a 60 mM iodoacetamide solution was added. Alkylation of cysteines proceeded for 30 min in the dark. Excess iodoacetamide was quenched with 14 ml of a 25 mM DTT solution, and the sample was then diluted with 330 ml of 183 mM Tris-HCl buffer (pH 8.0) supplemented with 2 mM CaCl 2 . Proteins were digested overnight using 12 mg sequencing-grade trypsin. Following digestion, the reaction was then quenched with 12.5 ml of a 20% trifluoroacetic acid (TFA) solution, resulting in a sample pH of ,3. Remaining ALS reagent was cleaved for 15 min at room temperature. The sample (;30 mg protein) was desalted by reverse-phase cleanup using C 18 UltraMicroSpin columns. The desalted peptides were dried at room temperature in a rotary vacuum centrifuge and reconstituted in 20 ml 70% formic acid-0.1% TFA (3:8 [vol/vol]) for peptide quantitation by UV 280 . The sample was diluted to a final concentration of 0.4 mg/ml, and 5 ml (2mg) was injected for LC-MS/MS analysis.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. TEXT S1, DOCX file, 0.02 MB.