Retraction

MICROBIOLOGYRetraction for “Stochastic induction of persister cells by HipAthrough (p)ppGpp-mediated activation of mRNA endonucleases,”by Elsa Germain, Mohammad Roghanian, Kenn Gerdes, andEtienne Maisonneuve, which was first published April 6, 2015;10.1073/pnas.1423536112 (Proc Natl Acad Sci USA 112:5171–5176).The authors wish to note the following: “In this article, we reported

that HipA-induced persistence in Escherichia coli K-12 depends on(p)ppGpp-mediated activation of the ten mRNase-encoding toxinantitoxin (TA) modules through a signaling pathway involving theinorganic polyphosphate and the protease Lon. After the publi-cation of the article, we became aware that the previously describedcontribution of polyphosphate and TA modules to bacterial per-sistence (1) were artifact results of inadvertent lysogenization withthe bacteriophage ϕ80, a notorious laboratory contaminant (2, 3).Therefore, the role and the connection of polyphosphate and TAmodules in a linear pathway in persister cells formation are nolonger supported. In this article, the following data were likely af-fected by the lysogenic bacteriophage: Fig. 3, Fig. 4 C and D, Fig. 5C andD, Fig. S2, Fig. S3, and Fig. S4 C and D. For these figures,we indeed confirmed ϕ80 lysogenization for the Δ10TA and theΔ(ppk ppx) strains. We believe that the most appropriate courseof action is to retract the paper. We offer our sincerest apologiesto the scientific community for these inadvertent errors andfor any inconvenience they may have caused.”

1. Maisonneuve E, Castro-Camargo M, Gerdes K (2013) (p) ppGpp controls bacterial per-sistence by stochastic induction of toxin-antitoxin activity. Cell 154:1140–1150, re-tracted (2018) 172:1135.

2. Harms A, Fino C, Sørensen MA, Semsey S, Gerdes K (2017) Prophages and growthdynamics confound experimental results with antibiotic-tolerant persister cells. MBio8:e01964-17.

3. Goormaghtigh F, et al. (2018) Reassessing the role of type II toxin-antitoxin systemsin formation of escherichia coli type II persister cells. MBio 9:e00640-18.

Published under the PNAS license.

Published online May 20, 2019.

www.pnas.org/cgi/doi/10.1073/pnas.1906160116

www.pnas.org PNAS | May 28, 2019 | vol. 116 | no. 22 | 11077

RETR

ACT

ION

Dow

nloa

ded

by g

uest

on

July

1, 2

021

Dow

nloa

ded

by g

uest

on

July

1, 2

021

Dow

nloa

ded

by g

uest

on

July

1, 2

021

Dow

nloa

ded

by g

uest

on

July

1, 2

021

Dow

nloa

ded

by g

uest

on

July

1, 2

021

Dow

nloa

ded

by g

uest

on

July

1, 2

021

Dow

nloa

ded

by g

uest

on

July

1, 2

021

Dow

nloa

ded

by g

uest

on

July

1, 2

021

https://www.pnas.org/site/aboutpnas/licenses.xhtmlhttps://www.pnas.org/cgi/doi/10.1073/pnas.1906160116

Stochastic induction of persister cells by HipA through(p)ppGpp-mediated activation of mRNA endonucleasesElsa Germaina,b, Mohammad Roghaniana,b, Kenn Gerdesa,b,1, and Etienne Maisonneuvea,b,1

aDepartment of Biology, University of Copenhagen, DK-2200 Copenhagen, Denmark; and bCentre for Bacterial Cell Biology, Institute for Cell and MolecularBiosciences, Newcastle University, NE2 4AX, Newcastle upon Tyne, United Kingdom

Edited by Tina M. Henkin, The Ohio State University, Columbus, OH, and approved March 13, 2015 (received for review December 9, 2014)

The model organism Escherichia coli codes for at least 11 type IItoxin–antitoxin (TA) modules, all implicated in bacterial persistence(multidrug tolerance). Ten of these encode messenger RNA endonu-cleases (mRNases) inhibiting translation by catalytic degradation ofmRNA, and the 11th module, hipBA, encodes HipA (high persisterprotein A) kinase, which inhibits glutamyl tRNA synthetase (GltX).In turn, inhibition of GltX inhibits translation and induces the strin-gent response and persistence. Previously, we presented strongsupport for a model proposing (p)ppGpp (guanosine tetra andpenta-phosphate) as the master regulator of persistence. Stochas-tic variation of [(p)ppGpp] in single cells induced TA-encodedmRNases via a pathway involving polyphosphate and Lon prote-ase. Polyphosphate activated Lon to degrade all known type IIantitoxins of E. coli. In turn, the activated mRNases induced persis-tence and multidrug tolerance. However, even though it wasknown that activation of HipA stimulated (p)ppGpp synthesis,our model did not explain how hipBA induced persistence. Herewe show that, in support of and consistent with our initial model,HipA-induced persistence depends not only on (p)ppGpp but also onthe 10 mRNase-encoding TA modules, Lon protease, and polyphos-phate. Importantly, observations with single cells convincingly showthat the high level of (p)ppGpp caused by activation of HipA doesnot induce persistence in the absence of TA-encoded mRNases.Thus, slow growth per se does not induce persistence in theabsence of TA-encoded toxins, placing these genes as central ef-fectors of bacterial persistence.

bacterial persistence | (p)ppGpp | toxin–antitoxin | HipA | single-cellanalysis

Bacterial persistence (multidrug tolerance) is caused by rarecells of a clonal bacterial population surviving the lethal actionof antibiotics. As opposed to resistance, persistence is a non-inherited, metastable phenomenon in which the cells have tran-siently entered a physiological state that renders them tolerant toantibiotics. Since their original discovery by Joseph Bigger workingwith Staphylococcus aureus (1), persister cells have been observedwith all bacteria tested, including major pathogens such asMycobacterium tuberculosis, Pseudomonas aeruginosa, and Salmo-nella enterica. That several antibiotics were particularly active againstgrowing cells originally suggested that persisters are cells that haveentered a state of low metabolic activity, referred to as dormant orslow-growing cells (2).Bacterial persistence has been implicated in recurrent and

chronic infections (3–6), and it is therefore important to under-stand the underlying molecular mechanisms. Previous analysisof a high persistence (hip) mutant of Escherichia coli indicatedthat persisters are generated by antibiotic-independent stochasticswitching from a rapidly growing, drug-sensitive state to a slow-growing, insensitive state (7). More recently, we established thatpersisters also arise stochastically in wild-type (WT) E. coli cellcultures (8). Cytological analyses confirmed that indeed slow-growing persister cells formed stochastically within rapidly growingbacterial cultures (7, 8). These observations are consistent with theview that persistence is a genetically evolved bet-hedging strategyenabling bacteria to survive unpredictable stress (9–11). Importantly,

persisters also form stochastically in stationary bacterial cultures andwithin biofilms and with much higher frequencies (8, 12).The first hipmutation of E. coli, called hipA7, defined the hipA

(high persister protein A) gene (13). Later analyses showed thathipA encodes the toxin of the type II hipBA toxin–antitoxin (TA)locus (14) (Fig. 1A). The hipA7 allele resulted in two amino acidchanges in HipA and triggered a dramatic 100–1,000-fold increasein persistence. Interestingly, HipA exhibits similarity to eukaryoticserine/threonine kinases and efficiently inhibited cell growth andthereby provoked a bacteriostatic, drug-tolerant condition thatcould be reversed by HipB antitoxin (15). The direct inhibition ofHipA by HipB and the weakened interaction between HipB andHipA7 readily explained the hipA7 phenotype, as it would lead tohyperactivation of HipA and thereby increase the persister celllevel (16, 17).In addition to hipBA, E. coli K-12 has 10 type II TA genes

(Fig. 1A), and numerous observations indicated that these TAgenes function as inducers of persistence (18–23). The 10 additionalTA modules all encode mRNases (also called mRNA interferases)that inhibit translation and induce a static, drug-tolerant conditionfrom which the cells could be resuscitated by the subsequent in-duction of cognate antitoxin-encoding genes (20, 22, 24–26). Im-portantly, sequential deletion of the 10 TA modules was associatedwith a corresponding gradual decline in the persister cell level, re-vealing a strong positive correlation between the number of type IITA genes and the persistence level (20).It has been of considerable interest to understand the mo-

lecular mechanism behind HipA-mediated persistence, not only

Significance

Bacteria produce persister cells that are tolerant to multipleantibiotics because they are hibernating in a dormant state inwhich the antibiotics cannot eradicate them. Persisters can thussurvive after drug treatment of infections and cause relapse ofdisease. The formation of persister cells depends on the ubiqui-tous bacterial regulatory nucleotides tetra and penta-guanosinephosphate [(p)ppGpp] that activate inhibitors of cell growth.One such inhibitor, HipA (high persister protein A), is an enzymethat halts translation by inhibiting glutamyl tRNA synthetase, anessential tRNA charging enzyme. Here we show that, surpris-ingly, HipA-induced persistence depends on (p)ppGpp-mediatedactivation of yet other inhibitors of translation that catalyticallydegrade messenger RNA. This discovery expands our mechanis-tic insight into the common and important phenomenon ofbacterial multidrug tolerance.

Author contributions: E.G., M.R., K.G., and E.M. designed research; E.G., M.R., and E.M.performed research; E.G., M.R., K.G., and E.M. analyzed data; and E.G., M.R., K.G., andE.M. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence may be addressed. Email: kgerdes@bio.ku.dk or e.maisonneuve@bio.ku.dk.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1423536112 PNAS | April 21, 2015 | vol. 112 | no. 16 | 5171–5176

MICRO

BIOLO

GY

See Retraction Published May 20, 2019

http://crossmark.crossref.org/dialog/?doi=10.1073/pnas.1423536112&domain=pdfmailto:kgerdes@bio.ku.dkmailto:e.maisonneuve@bio.ku.dkmailto:e.maisonneuve@bio.ku.dkhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplementalhttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplementalwww.pnas.org/cgi/doi/10.1073/pnas.1423536112

because hipA was the first “persister” gene discovered but alsobecause of the strong phenotype of the hipA7 allele. We andothers recently discovered that HipA inactivates glutamyl tRNAsynthetase (GltX) by phosphorylation (27, 28). The resultingincreased concentration of uncharged tRNAGlu activated (p)ppGpp(guanosine tetra and penta-phosphate) synthesis by RelA [(p)ppGpp synthetase I] (27, 28). In turn, the high level of (p)ppGpp dramatically increased the persistence level (27).

Previously, we presented strong evidence for a model linking(p)ppGpp to persistence (8, 29). The model, shown in Fig. 1B(left part), proposed that stochastic, single-cell variation of thelevel of (p)ppGpp induced slow growth, drug tolerance, andpersistence. We further revealed that (p)ppGpp activated thetoxins encoded by type II TA genes and that this activationresulted in inhibition of translation and slow growth. The signalfrom (p)ppGpp to the TA genes was conveyed by Lon and poly-phosphate [Poly(P)]: (p)ppGpp competitively inhibited exopoly-phosphatase (PPX), the cellular enzyme that degrades Poly(P)(30). The resulting accumulation of Poly(P) activated Lon pro-tease to degrade type II antitoxins of E. coli K-12. In turn, thesubsequent activation of the corresponding toxins inhibited cellgrowth and induced persistence.Here we investigate the molecular mechanism underlying HipA-

induced persistence. We show that persistence triggered by ectopicproduction of HipA depends on (p)ppGpp, whereas persistencetriggered by ectopic production of mRNases occurs indepen-dently of (p)ppGpp. Surprisingly, we find that the (p)ppGpplevel caused by HipA expression cross-activates the TA-encodedmRNases. Consistent with our previously proposed model (8),we show that HipA-mediated transactivation of mRNases dependshierarchically on (p)ppGpp, Poly(P), and Lon protease. Inter-estingly, activation of HipA halted cell growth of both a WT strainand a strain lacking the 10 TA-encoded mRNases but inducedpersistence only in the former strain. These results show that slowgrowth per se is not sufficient to induce persistence and that TA-encoded mRNases function as essential inducers of persistence.Finally, we show that a strain carrying the hipA7 allele producespersisters by stochastic induction of high levels of (p)ppGpp withan increased frequency, thereby lending further support to thepreviously proposed stochastic mechanism underlying bacterialpersistence.

ResultsHipA but Not RelE, MazF, or YafO Triggers (p)ppGpp Synthesis. Pre-viously, we and others established that persistence mediated byHipA correlates with the activation of the stringent response (27,28, 31). Indeed, (p)ppGpp sharply increased by over 10-fold afterhipA induction (Fig. 2A and Fig. S1A, lanes 5–8). Concomitantlywe observed a 250-fold increase in persister cell fraction (Fig.2B). However, hipA expression failed to induce persistence in aΔrelA strain (Fig. 2B) (27). Moreover, HipA also failed to induce(p)ppGpp synthesis in a ΔrelA strain (31), consistent with the factthat HipA-mediated persistence indeed depends on the activa-tion of the stringent response in E. coli (27, 28, 31).When expressed ectopically, all TA-encoded mRNases in-

duce persistence (19–22). To test if mRNase-induced persistencedepended on the stringent response, we monitored the (p)ppGpplevels after expression of three representative mRNases: RelE,YafO, and MazF. RelE and YafO cleave mRNA positioned atthe ribosomal A site (26, 32), whereas MazF cleaves RNA site-specifically and independently of the ribosomes (33). As shown inFig. 2A, none of these toxins triggered (p)ppGpp synthesis (Fig.S1A, lanes 9–20). Moreover, the persistence levels conferred byectopic expression of relE, mazF, and yafO were not significantlyaffected by deletion of relA (Fig. 2B). These results showed thatHipA specifically required (p)ppGpp to induce persistenceand, importantly, raised the possibility that HipA induced per-sistence by cross-activating TA modules known to be activatedby (p)ppGpp (8).These observations also raised the possibility of DksA being

involved in HipA-mediated persistence because (p)ppGpp andDksA affected RNA polymerase activity synergistically (34–36).However, HipA still supported high persistence in a ΔdksA strain(Fig. S1B). Consistently, deletion of dksA did not reduce thepersistence level of a strain containing the hipA7 allele (Fig. S2A).

Fig. 1. TA loci and molecular model integrating HipA into (p)ppGpp-mediatedpersistence. (A) Chromosomal locations of 11 type II TA loci of E. coli K-12. Ri-bosome-dependent and -independent mRNase-encoding TA modules are shownin green and blue, respectively. The serine/threonine kinase-encoding hipBA lo-cus is shown in red. In the model (B), free HipA phosphorylates GltX and theresulting inhibition of Glu-tRNAGlu increases the rate of uncharged tRNAGlu

loading at the A site of the ribosome and triggers RelA-dependent (p)ppGppproduction. Then (p)ppGpp competitively inhibits PPX, the cellular enzyme thatdegrades Poly(P). In turn, Poly(P) is synthesized by polyphosphate kinase (PPK)and stimulates Lon to degrade the 11 type II antitoxins (including HipB), therebyactivating the mRNases that inhibit translation and cell growth and inducepersistence. Thus, activation of HipA generates a feed-forward loop that theo-retically should lock the cells in a state with a high level of (p)ppGpp. However,such a locked state has never been observed experimentally and may be over-come by the ability of HipA to inactivate itself by intermolecular phosphorylation(40), as indicated in B by a line that points back to HipA. The arrow emergingfrom RelA and pointing back to RelA indicates a positive feedback loop in which(p)ppGpp activates further (p)ppGpp synthesis (47). This positive feedback maycontribute to the observed stochasticity in the variation of (p)ppGpp in singlecells (8) Broken arrows indicate the possible signaling pathway that by positivefeed forward would lock the cells in a state of a permanent high level of (p)ppGpp and therefore constantly activate HipA and TA-encoded mRNases,whereas unbroken arrows indicate the signaling pathway leading to per-sistence as described previously.

5172 | www.pnas.org/cgi/doi/10.1073/pnas.1423536112 Germain et al.

See Retraction Published May 20, 2019

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2www.pnas.org/cgi/doi/10.1073/pnas.1423536112

These results show that HipA-mediated persistence does notdepend on DksA.

HipA-Mediated Persistence Depends on TA Modules, Lon, andPolyphosphate. The finding that (p)ppGpp is the master regulatorof persistence predicted that HipA-induced persistence should de-pend also on Poly(P), Lon, and TAs. To test this, we overexpressedHipA in a strain lacking the 10 TA modules coding for mRNases(20). As shown in Fig. 3, overexpression of HipA failed to inducehigh persistence in the Δ10TA strain. A similar result was obtainedwhen the hipA7 allele was introduced into the Δ10TA strain (Fig.S2A). In contrast, RelE, MazF, YafO, YoeB, MqsR, ChpB, HicA,and YhaV toxins all induced high persistence when overexpressedin the Δ10TA strain at a level similar to that of the WT strain (Fig.3 and Fig. S2B). These results strongly support that HipA-mediatedhigh persistence depends on cross-activation of the other type II TAloci via the (p)ppGpp-controlled signaling pathway.To be consistent with our model, the HipA-induced (p)ppGpp

synthesis (Fig. 2A) and persistence (Fig. 2B) should depend onPoly(P) and Lon (Fig. 1B, left part). To test this inference, HipA

was ectopically expressed in Δ(ppk ppx) and Δlon strains. In-deed, HipA did not induce persistence in these strains (Fig. 3).Moreover, the hipA7 allele also failed to induce high persistencein these two genetic backgrounds (Fig. S2A). In contrast, RelE,MazF, and YafO increased persistence in the absence of Poly(P)or Lon (Fig. 3). Taken together, these results show that HipA-mediated persistence depends hierarchically on (p)ppGpp, Poly(P),Lon, and TAs.

The hipA7 Allele Increases the Frequency of (p)ppGpp “ON” Cells.Previously, we established an rpoS::mcherry translational fusionas a reliable proxy of the (p)ppGpp level in single cells andshowed that the RpoS-mCherry ON cells were persisters (8). Todirectly test the hypothesis at the single cell level that HipA in-duces persistence by increasing [(p)ppGpp], we induced hipA ina strain carrying an rpoS::mcherry translational fusion and ana-lyzed the cells by fluorescence microscopy. We observed a veryheterogeneous population with only a few fluorescent cells. Webelieve that the strong inhibition of translation resulting fromectopic induction of hipA prevented sufficient synthesis ofRpoS-mCherry.In a different approach, we turned to the hipA7 allele that has

a 100–1,000-fold increase in persistence (13) (Fig. S2A) due to areduced interaction between HipB and HipA7 (16, 17). Theabove results predicted that the hipA7 allele increased persis-tence by increasing the frequency of cells with a high level of(p)ppGpp. To test this inference, we measured the frequency ofRpoS-mCherry ON cells in a hipA7 strain carrying the rpoS::mcherry fusion. Indeed, statistical analysis of more than 100,000cells showed that the frequency of ON cells was 20-fold elevatedin the hipA7 strain (8.9 × 10−3 in hipA7 vs. 3.8 × 10−4 in WT)(Fig. 4A and Fig. S3A vs. Fig. 4B and Fig. S3B). These resultssuggest that once activated, HipA promotes [(p)ppGpp] varia-tion at a single cell level.That HipA failed to induce persistence in a strain lacking the

10 TA modules encoding mRNases raised the possibility that(p)ppGpp was not produced when HipA was activated in thisstrain. To test this possibility we analyzed the fluorescence levels

Fig. 2. HipA-induced persistence depends specifically on RelA. (A) Accu-mulation of (p)ppGpp following ectopic expression of toxins. MG1655 car-rying either hipA, relE, mazF, or yafO on pBAD33 was grown exponentiallyin low phosphate Mops minimal medium (Materials and Methods). Sampleswere collected before and after toxin gene induction (0.2% arabinose) andseparated by TLC. Here, we present the quantification of (p)ppGpp levelafter toxin induction. A representative autoradiograph of the TLC plates isshown in Fig. S1A, and growth arrest after toxin induction was monitored bygrowth curve (Fig. S1C). (B) Exponentially growing cells of MG1655 (graybars) and MG1655 ΔrelA (black bars) carrying either hipA, relE, mazF, oryafO on pBAD33 induced for 30 min were exposed to 2 μg/mL of cipro-floxacin (for details, seeMaterials and Methods). Percentage of survival after4 h of antibiotic treatment was compared with that of the control strainscarrying the pBAD33 vector plasmid (log scale). Error bars indicate the SDs ofaverages of at least three independent experiments.

Fig. 3. HipA-mediated persistence depends on Poly(P), Lon, and the othertype II TA loci. Exponentially growing cells of MG1655 (gray bars) and isogenicdeletion strains Δ10TA (black bars), Δ(ppk ppx) (white bars), and Δlon (dashedbars), overexpressing hipA, relE, mazF, or yafO from plasmid pBAD33, wereexposed to 2 μg/mL of ciprofloxacin (for details, see Materials and Methods).Percentage of survival after 4 h of antibiotic treatment was compared withthat of the control strains carrying the pBAD33 vector plasmid (log scale). Errorbars indicate the SDs of averages of at least three independent experiments.

Germain et al. PNAS | April 21, 2015 | vol. 112 | no. 16 | 5173

MICRO

BIOLO

GY

See Retraction Published May 20, 2019

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1

of the rpoS::mcherry reporter in the Δ10TA strain and a Δ10TAharboring the hipA7 allele. However, introducing the hipA7 al-lele in the Δ10TA increases the frequency of (p)ppGpp ON cellsby more than 16 times (4.52 × 10−4 vs. 7.26 × 10−3) (Fig. 4C andFig. S3C vs. Fig. 4D and Fig. S3D). This increase is very similar tothat observed when hipA7 was introduced in the WT strain, thussuggesting that HipA still enhanced (p)ppGpp synthesis in theabsence of TA modules. In addition, the Δ10TA strain produceda frequency of (p)ppGpp ON cells similar to that of the WT strain(4.52 × 10−4 and 3.84 × 10−4, respectively) (Fig. 4A and Fig. S3Avs. Fig. 4C and Fig. S3C), showing that the stochastic variation of[(p)ppGpp] was entirely unaffected by deletion of the TA genes.To further substantiate that (p)ppGpp synthesis was not im-paired in the Δ10TA strain, we measured the intracellular level of(p)ppGpp after HipA overproduction. As expected, we found thatthe level and the time needed to turn on (p)ppGpp synthesis in theΔ10TA strain was very similar to that of the WT strain (Fig. S3E,lanes 5–8 vs. Fig. S1, lanes 5–8). Taken together these results es-tablish that activation of HipA influences the stochastic variation of(p)ppGpp at the single cell level independently of TA genes.

Increased Persistence by hipA7 Depends Stochastically on the(p)ppGpp–TA Pathway. We then tested directly if the high fre-quency of (p)ppGpp ON cells exhibited by the hipA7 strain was the

basis for the increased persistence. To that end, exponentiallygrowing cells of the hipA7 rpoS::mcherry strain were introducedin a microfluidic device and subjected to growth in rich mediumover a period of 90 min. Interestingly, cells with a high level ofRpoS-mCherry (indicating high [(p)ppGpp]) were not growing(Fig. 5A and Movie S1). Next, we directly recorded the responseof these dormant cells toward a high dose of the bacteriolyticantibiotic ampicillin. Remarkably, the cell with the highest levelof RpoS-mCherry did not lyse even after 90 min of ampicillintreatment (Fig. 5B and Movie S2). Moreover, when ampicillinwas removed, this cell elongated and gave rise to progeny cellsafter a latency period of ∼1 h (Fig. 5B and Movie S2). Similaranalysis of 12 cells having the highest fluorescence level revealedthat 100% of them were not lysed by ampicillin (Movie S3).As described above, the high frequency of the RpoS-mCherry

ON cells in the hipA7 strain was unaffected by deletion of the 10mRNase-encoding TAmodules (Fig. 4D and Fig. S3D) even thoughthe high persistence phenotype was eliminated (Fig. S2A). Theseobservations predicted that the (p)ppGpp ON cells produced by thehipA7 Δ10TA rpoS::mcherry strain should be ampicillin-sensitive.Indeed, we observed that such cells with a high level of RpoS-mCherry were not growing (Fig. 5C and Movie S4). However, andin stark contrast to hipA7 cells, hipA7 Δ10TA cells exhibiting the(p)ppGpp ON phenotype lysed upon ampicillin treatment (Fig.5D and Movie S5). Statistical analysis on 12 cells having thehighest RpoS-mCherry level revealed that 11 of the 12 (91.6%)lysed upon ampicillin treatment (Movie S6). Taken together, theseresults indicate that, in single cells, HipA-mediated persistencedepends on (p)ppGpp and the other TA modules of the cell.

DiscussionPreviously, we established that (p)ppGpp controls persister cellformation by stochastically switching to a high level in singlecells. The high (p)ppGpp level activated TA-encoded inhibitors oftranslation that, in turn, induced persistence (8). We also unraveledthe signaling pathway leading to TA activation (Fig. 1B). The factthat ectopic production of HipA strongly stimulated (p)ppGppsynthesis (Fig. 2A) and persistence (Fig. 2B) therefore raised thepossibility that the HipA-induced persistence would depend onthe other TA genes of E. coli and on the pathway leading to theiractivation. Indeed, we found that ectopic production of HipA inthe Δ10TA, Δlon, or Δ(ppk ppx) strains failed to increase per-sistence (Fig. 3). We conclude that, surprisingly, HipA-inducedpersistence depends on polyphosphate, Lon, and the TA genesthat encode mRNases. In contrast, ectopic production of mRNases(RelE, MazF, and YafO) induced persistence (Fig. 2B) withoutleading to accumulation of (p)ppGpp (Fig. 2A). Thus, the molec-ular mechanisms by which HipA and the other type II toxinsinduce persistence are entirely different.Our model predicted that hipBA contributes to persistence by

increasing the (p)ppGpp level and activating TA-encoded mRNasesin a subpopulation of cells of a growing population (Fig. 1B, rightpart). To test this inference, we used a strain producing a dramat-ically increased level of persisters because it carried the hipBA7allele (14–16). Our analyses showed that indeed the hipA7mutationconferred a dramatic increase in the frequency of cells having ahigh level of (p)ppGpp and that, consistent with our model, thisfrequency was unaffected by the deletion of the 10 mRNase-encoding TA loci (Fig. 4). Crucially, our model predicted thatWT cells with a high level of (p)ppGpp should be persisters,whereas cells lacking the 10 mRNase-encoding TA loci shouldpredominantly be antibiotic-sensitive. We tested this importantprediction by two independent approaches. First, at the pop-ulation level, induction of HipA induced drug tolerance in WTbut not in the multiple TA knockout strain (Fig. 3). Second,careful accomplishment of time lapses of single cells showedthat, surprisingly, cells that were in a nongrowing state due to ahigh level of (p)ppGpp were antibiotic-tolerant if they carried

Fig. 4. The hipA7 allele increases the frequency of (p)ppGpp ON cells inde-pendently of the 10 mRNase-encoding TA loci. (A) Snapshot of exponentiallygrowing cells of MG1655 (A), MG1655 hipA7 (B), Δ10TA (C), or Δ10TA hipA7(D), carrying an rpoS::mcherry translational fusion. (i) Phase contrast. (ii) RpoS-mCherry fluorescence. (iii) Overlay of i and ii. Statistical analysis of these rarefluorescent cells is provided in Fig. S3 A–D. (Scale bar, 4 μm.)

5174 | www.pnas.org/cgi/doi/10.1073/pnas.1423536112 Germain et al.

See Retraction Published May 20, 2019

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-1http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-2http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-2http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-4http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-5http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-6http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF3www.pnas.org/cgi/doi/10.1073/pnas.1423536112

the mRNase-encoding TA modules but sensitive if they lackedthe TA modules (Fig. 5 and Movies S3 and S6). This is a uniqueand highly important observation because it has generally beenassumed that slow growth (or dormancy) per se is sufficient toinduce persistence (37, 38). Results from cell fractions enrichedfor persisters obtained by cell sorting support the view that slowcell growth by itself does not ensure drug tolerance (39).The right part of Fig. 1B integrates our findings in the model

we previously proposed to explain (p)ppGpp-controlled persistercell formation (8) and raises mechanistic issues important for adeeper understanding of the phenomenon. The vast majorityof cells of a rapidly growing bacterial population have a low levelof (p)ppGpp and are drug-sensitive. However, by stochastic var-iation in the enzymatic activities of RelA and/or SpoT, the(p)ppGpp level increases and induces persistence via Lon/Poly(P)-mediated activation of the mRNases. Because HipB is also de-graded via the Lon/Poly(P) pathway, at least some cells will ex-perience activation of HipA that, by inference, should lead to afurther increase in (p)ppGpp levels. According to the model, sucha positive feedback loop would lock the cells in the persistent statedue to a stable high level of (p)ppGpp that would permanentlyactivate TA-encoded mRNases. Obviously this is not what weobserve because at least some cells, namely those surviving an-tibiotic treatment, are able to normalize the (p)ppGpp level,quench toxin activities, and resuscitate. Because activation of themRNases does not induce (p)ppGpp synthesis (Fig. 1A and Fig.S1A, lanes 9–20), the problem of the return to a lower level of[(p)ppGpp] is particularly important for cells experiencing activa-tion of HipA. These considerations suggest the existence of amechanism that breaks the positive feedback loop mediated byHipA-stimulated (p)ppGpp synthesis. Remarkably, HipA inhibitsits own kinase activity by intermolecular autophosphorylation of aserine residue close to the P loop required for ATP binding (24,40, 41) and was proposed to function either in controlling theduration of the persistent state and/or in the resuscitation of per-sisters (40). The latter suggestion is consistent with experimentaland theoretical single-cell analysis establishing the existence of athreshold level above which HipA induces growth arrest (16).

It has been proposed that (p)ppGpp–SpoT can be consideredanalogous to a TA pair. In this view, (p)ppGpp plays the role of atoxin that inhibits cell growth and induces persistence, whereasthe hydrolytic activity of SpoT acts analogously to the antitoxinthat degrades and inactivates the toxin (42). These authors foundthat diauxic growth induced persistence in the carbon-sourcetransition period, both during planktonic growth and in biofilm(43). Interestingly, the increased persistence was dependent notonly on (p)ppGpp but also on DksA. DksA binds to RNA poly-merases and sensitizes it to (p)ppGpp and thereby either repressesor activates initiation of transcription (34–36, 44, 45). Therefore,we tested if DksA was involved in HipA-mediated persistence.However, consistent with our model (Fig. 1B), deletion of dksAdid not reduce HipA-mediated persistence (Figs. S1 and S2A).One possible explanation may be that even though both diauxieand HipA-induced persistence depend on (p)ppGpp, DksA isrequired in the former case only. We are currently investigatingthis possibility.

Materials and MethodsBacterial Strains and Plasmids. Bacterial strains and plasmids are described inSI Materials and Methods and listed in Table S1. DNA oligonucleotides arelisted in Table S2.

Persistence Assay. Persistence was determined by measuring the number ofcolony-forming units (CFUs) per mL upon exposure to 1 μg/mL ciprofloxacin.Overnight cultures were diluted 100-fold in 10 mL of fresh LB medium andincubated for 2 h at 37 °C with shaking (typically reaching ∼2 × 108 CFUs/mL).Then aliquots of 5 mL were transferred to 28 × 114 mm Sarstedt conicalpolypropylene tubes, and the antibiotic was added. Tubes were placed in 45°inclination with shaking at 37 °C for 4 h. For determination of CFUs, 1 mLaliquots were removed and the cells harvested, resuspended in fresh me-dium, serially diluted, and plated on solid LB medium. Persisters were cal-culated as the surviving fraction, by dividing the number of CFUs/mL in theculture after 4 h of incubation with the antibiotic by the number of CFUs/mLin the culture before adding the antibiotic.

Toxin Overexpression and Persistence. To determine the number of persistersformed by cells expressing toxins (HipA, RelE, MazF, or YafO) from plasmidpBAD33, cells were grown in rich medium for 1.5 h (OD600 ∼0.25) and toxinproduction was induced by addition of 0.2% of arabinose for 30 min. Then

Fig. 5. HipA7-mediated high levels of (p)ppGpp donot induce persistence in the absence of mRNase-encoding TA modules. Exponentially growing cellsof MG1655 rpoS::mcherry were introduced into amicrofluidic device and subjected to growth medium.(A) Time-lapse images showing growth upon richmedium injection of MG1655 hipA7 cells carryingrpoS::mcherry; the arrow indicates a growth-arrestedcell with a high level of RpoS-mCherry (from MovieS1). (B) Time-lapse images showing persistence uponampicillin injection of cells of MG1655 hipA7 rpoS::mcherry expressing a high level of RpoS-mCherry(from Movie S2). (C) Time-lapse images showinggrowth upon rich medium injection of Δ10TA hipA7cells carrying rpoS::mcherry; the arrow indicatesa growth-arrested cell with a high level of RpoS-mCherry (from Movie S4). (D) Time-lapse imagesshowing the behavior of Δ10TA hipA7 expressing ahigh level of RpoS-mCherry upon ampicillin treatment(from Movie S5).The first images of each panel weregenerated by overlay of phase contrast and fluores-cence images (separate phase contrast and fluores-cence images are shown in Fig. S4 A–D, respectively).The arrow indicates a cell with a high level of RpoS-mCherry. (Scale bar, 4 μm.) t, time in minutes.

Germain et al. PNAS | April 21, 2015 | vol. 112 | no. 16 | 5175

MICRO

BIOLO

GY

See Retraction Published May 20, 2019

http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-3http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-6http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF2http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=STXThttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=ST1http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=ST2http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-1http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-1http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-2http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-4http://movie-usa.glencoesoftware.com/video/10.1073/pnas.1423536112/video-5http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF4http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF4http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=SF4

0.2% of glucose was added to repress the pBAD promoter. Aliquots weresubjected to 2 μg/mL ciprofloxacin antibiotic, and persister cell formationwas determined as described above, except that the cells were plated onsolid medium containing 0.2% glucose (without antibiotics).

In Vivo (p)ppGpp Measurement. To determine the (p)ppGpp content formedby cells expressing TA-encoded toxins using pBbS2k as the vector plasmid,overnight cultures were diluted 100-fold in 10 mL of Mops glucose minimalmedium supplemented with all amino acid as previously described (46) andincubated at 37 °C with shaking. At OD600 0.05, cells were labeled withH3

32PO4, (100 μCi/mL). After 2–3 generations (OD600 0.2–0.3), 50 ng/mL an-hydrous tetracycline (aTc) was added to induce toxins. Samples were with-drawn 0, 10, 30, and 60 min after addition of aTc. Reactions were stopped bythe addition of 10 μL of 2 M formic acid. Aliquots (10 μL) of each reactionwere loaded and separated on PEI Cellulose TLC plates (purchased from GEHealthcare). Plates were revealed by PhosphoImaging (GE Healthcare) andanalyzed using ImageQuant software (GE Healthcare).

Microscopy. For phase contrast and fluorescencemicroscopy, cells were grownin LB to midexponential phase at 37 °C, centrifuged, resuspended in 10× dilutedLB in M9 medium, and mounted on prewarmed microscope slides covered witha thin film of 1.2% (wt/vol) agarose (in a 10× diluted LB in M9 medium). Imageswere acquired with a Cool-Snap HQ CCD camera (Roper Scientific) attached to a

Nikon Eclipse Ti-U microscope. The images were acquired and analyzed withMetamorph 6. Final image preparation was performed in ImageJ.

Microfluidic Assay and Device. Cells were grown in LB to midexponentialphase at 37 °C, centrifuged, resuspended in a 10× diluted LB in M9 medium,and mounted on a prewarmed agarose-based microfludic device as pre-viously described (8) (for more information, see SI Materials and Methods).The microfluidic device was assembled in a heating chamber (37 °C) attachedto the microscope. Medium (M9 medium supplemented with a 10 fold di-luted LB with or without 500 μg/mL of ampicillin) perfusion was ac-complished by gravity at a constant rate of ∼5–10 μL/s. During mediumtransition, medium was injected manually by supplying 2× the volume of thechamber with a 2 mL syringe to minimize the time required for the drug (i.e.,ampicillin) to reach the cells. For details, see SI Materials and Methods.

ACKNOWLEDGMENTS. We thank Manuela Castro-Camargo for technicalassistance; Mike Cashel, Emmanuelle Bouveret, and Aurelia Battesti fortheir helpful advice on the protocol for (p)ppGpp measurements; and themembers of the K.G. group, of the Centre for Bacterial Cell Biology at New-castle University, and Kim Sneppen, Namiko Mitarai, and Szabolcs Semsey ofthe Centre for Models of Life at the Niels Bohr Institute for stimulatingdiscussions. This work was supported by European Research Council Ad-vanced Investigator Grant 294517 and by a Laureate Research Grant Awardfrom the Novo Nordisk Foundation (to K.G.).

1. Bigger JW (1944) Treatment of staphyloccal infections with penicillin by intermittentsterilisation. Lancet 244(6320):497–500.

2. Levin BR, Rozen DE (2006) Non-inherited antibiotic resistance. Nat Rev Microbiol 4(7):556–562.

3. Allison KR, Brynildsen MP, Collins JJ (2011) Metabolite-enabled eradication of bac-terial persisters by aminoglycosides. Nature 473(7346):216–220.

4. Lafleur MD, Qi Q, Lewis K (2010) Patients with long-term oral carriage harbor high-persister mutants of Candida albicans. Antimicrob Agents Chemother 54(1):39–44.

5. Lewis K (2007) Persister cells, dormancy and infectious disease. Nat Rev Microbiol 5(1):48–56.

6. Mulcahy LR, Burns JL, Lory S, Lewis K (2010) Emergence of Pseudomonas aeruginosastrains producing high levels of persister cells in patients with cystic fibrosis. J Bacteriol192(23):6191–6199.

7. Balaban NQ, Merrin J, Chait R, Kowalik L, Leibler S (2004) Bacterial persistence as aphenotypic switch. Science 305(5690):1622–1625.

8. Maisonneuve E, Castro-Camargo M, Gerdes K (2013) (p)ppGpp controls bacterialpersistence by stochastic induction of toxin-antitoxin activity. Cell 154(5):1140–1150.

9. Losick R, Desplan C (2008) Stochasticity and cell fate. Science 320(5872):65–68.10. Veening JW, Smits WK, Kuipers OP (2008) Bistability, epigenetics, and bet-hedging in

bacteria. Annu Rev Microbiol 62:193–210.11. Kussell E, Leibler S (2005) Phenotypic diversity, population growth, and information in

fluctuating environments. Science 309(5743):2075–2078.12. Keren I, Kaldalu N, Spoering A, Wang Y, Lewis K (2004) Persister cells and tolerance to

antimicrobials. FEMS Microbiol Lett 230(1):13–18.13. Moyed HS, Bertrand KP (1983) hipA, a newly recognized gene of Escherichia coli K-12

that affects frequency of persistence after inhibition of murein synthesis. J Bacteriol155(2):768–775.

14. Korch SB, Henderson TA, Hill TM (2003) Characterization of the hipA7 allele of Es-cherichia coli and evidence that high persistence is governed by (p)ppGpp synthesis.Mol Microbiol 50(4):1199–1213.

15. Korch SB, Hill TM (2006) Ectopic overexpression of wild-type and mutant hipA genesin Escherichia coli: Effects on macromolecular synthesis and persister formation.J Bacteriol 188(11):3826–3836.

16. Rotem E, et al. (2010) Regulation of phenotypic variability by a threshold-basedmechanism underlies bacterial persistence. Proc Natl Acad Sci USA 107(28):12541–12546.

17. Schumacher MA, et al. (2009) Molecular mechanisms of HipA-mediated multidrugtolerance and its neutralization by HipB. Science 323(5912):396–401.

18. Dörr T, Vuli�c M, Lewis K (2010) Ciprofloxacin causes persister formation by inducingthe TisB toxin in Escherichia coli. PLoS Biol 8(2):e1000317.

19. Keren I, Shah D, Spoering A, Kaldalu N, Lewis K (2004) Specialized persister cells andthe mechanism of multidrug tolerance in Escherichia coli. J Bacteriol 186(24):8172–8180.

20. Maisonneuve E, Shakespeare LJ, Jørgensen MG, Gerdes K (2011) Bacterial persistenceby RNA endonucleases. Proc Natl Acad Sci USA 108(32):13206–13211.

21. Shah D, et al. (2006) Persisters: A distinct physiological state of E. coli. BMC Microbiol6:53.

22. Vázquez-Laslop N, Lee H, Neyfakh AA (2006) Increased persistence in Escherichia colicaused by controlled expression of toxins or other unrelated proteins. J Bacteriol188(10):3494–3497.

23. Kim Y, Wood TK (2010) Toxins Hha and CspD and small RNA regulator Hfq are in-volved in persister cell formation through MqsR in Escherichia coli. Biochem BiophysRes Commun 391(1):209–213.

24. Correia FF, et al. (2006) Kinase activity of overexpressed HipA is required for growtharrest and multidrug tolerance in Escherichia coli. J Bacteriol 188(24):8360–8367.

25. Pedersen K, Christensen SK, Gerdes K (2002) Rapid induction and reversal of a bac-teriostatic condition by controlled expression of toxins and antitoxins. Mol Microbiol45(2):501–510.

26. Christensen-Dalsgaard M, Jørgensen MG, Gerdes K (2010) Three new RelE-homolo-gous mRNA interferases of Escherichia coli differentially induced by environmentalstresses. Mol Microbiol 75(2):333–348.

27. Germain E, Castro-Roa D, Zenkin N, Gerdes K (2013) Molecular mechanism of bacterialpersistence by HipA. Mol Cell 52(2):248–254.

28. Kaspy I, et al. (2013) HipA-mediated antibiotic persistence via phosphorylation of theglutamyl-tRNA-synthetase. Nat Commun 4:3001.

29. Maisonneuve E, Gerdes K (2014) Molecular mechanisms underlying bacterial per-sisters. Cell 157(3):539–548.

30. Kuroda A, Murphy H, Cashel M, Kornberg A (1997) Guanosine tetra- and penta-phosphate promote accumulation of inorganic polyphosphate in Escherichia coli.J Biol Chem 272(34):21240–21243.

31. Bokinsky G, et al. (2013) HipA-triggered growth arrest and β-lactam tolerance in Es-cherichia coli are mediated by RelA-dependent ppGpp synthesis. J Bacteriol 195(14):3173–3182.

32. Christensen SK, Gerdes K (2003) RelE toxins from bacteria and Archaea cleavemRNAs on translating ribosomes, which are rescued by tmRNA. Mol Microbiol 48(5):1389–1400.

33. Zhang Y, et al. (2003) MazF cleaves cellular mRNAs specifically at ACA to block proteinsynthesis in Escherichia coli. Mol Cell 12(4):913–923.

34. Paul BJ, et al. (2004) DksA: A critical component of the transcription initiation ma-chinery that potentiates the regulation of rRNA promoters by ppGpp and the initi-ating NTP. Cell 118(3):311–322.

35. Lemke JJ, Durfee T, Gourse RL (2009) DksA and ppGpp directly regulate transcriptionof the Escherichia coli flagellar cascade. Mol Microbiol 74(6):1368–1379.

36. Lemke JJ, et al. (2011) Direct regulation of Escherichia coli ribosomal protein pro-moters by the transcription factors ppGpp and DksA. Proc Natl Acad Sci USA 108(14):5712–5717.

37. Lewis K (2010) Persister cells. Annu Rev Microbiol 64:357–372.38. Tuomanen E, Cozens R, Tosch W, Zak O, Tomasz A (1986) The rate of killing of Es-

cherichia coli by beta-lactam antibiotics is strictly proportional to the rate of bacterialgrowth. J Gen Microbiol 132(5):1297–1304.

39. OrmanMA, Brynildsen MP (2013) Establishment of a method to rapidly assay bacterialpersister metabolism. Antimicrob Agents Chemother 57(9):4398–4409.

40. Schumacher MA, et al. (2012) Role of unusual P loop ejection and autophosphor-ylation in HipA-mediated persistence and multidrug tolerance. Cell Reports 2(3):518–525.

41. Wen Y, et al. (2014) The bacterial antitoxin HipB establishes a ternary complex withoperator DNA and phosphorylated toxin HipA to regulate bacterial persistence. Nu-cleic Acids Res 42(15):10134–10147.

42. Amato SM, Orman MA, Brynildsen MP (2013) Metabolic control of persister formationin Escherichia coli. Mol Cell 50(4):475–487.

43. Amato SM, Brynildsen MP (2014) Nutrient transitions are a source of persisters inEscherichia coli biofilms. PLoS ONE 9(3):e93110.

44. Gummesson B, Lovmar M, Nyström T (2013) A proximal promoter element requiredfor positive transcriptional control by guanosine tetraphosphate and DksA proteinduring the stringent response. J Biol Chem 288(29):21055–21064.

45. Kajitani M, Ishihama A (1984) Promoter selectivity of Escherichia coli RNA polymerase.Differential stringent control of the multiple promoters from ribosomal RNA andprotein operons. J Biol Chem 259(3):1951–1957.

46. Cashel M (1994) Detection of (p)ppGpp accumulation patterns in Escherichia colimutants. Methods in Molecular Genetics 3:341–356.

47. Shyp V, et al. (2012) Positive allosteric feedback regulation of the stringent responseenzyme RelA by its product. EMBO Rep 13(9):835–839.

5176 | www.pnas.org/cgi/doi/10.1073/pnas.1423536112 Germain et al.

See Retraction Published May 20, 2019

http://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=STXThttp://www.pnas.org/lookup/suppl/doi:10.1073/pnas.1423536112/-/DCSupplemental/pnas.201423536SI.pdf?targetid=nameddest=STXTwww.pnas.org/cgi/doi/10.1073/pnas.1423536112