FEMS Microbiology Reviews, fuv003, 39, 2015, 413–427

doi: 10.1093/femsre/fuv003Review Article

REVIEW ARTICLE

sRNAs in bacterial type I and type III toxin-antitoxinsystemsSabine Brantl∗ and Natalie Jahn

AG Bakteriengenetik, Lehrstuhl fur Genetik, Friedrich-Schiller-Universitat Jena, Philosophenweg 12,D-07743 Jena, Germany∗Corresponding author: AG Bakteriengenetik, Lehrstuhl fur Genetik, Friedrich-Schiller-Universitat Jena, Philosophenweg 12, D-07743 Jena, Germany.Tel: +49-3641-949570; E-mail: Sabine.Brantl@rz.uni-jena.deOne sentence summary: We summarize our current knowledge on small regulatory RNAs that act as antitoxins in plasmid or chromosome-encodedtype I or type III toxin–antitoxin systems with regard to their mechanisms of action, structural requirements and kinetics of interaction with theirtargets, factors that regulate their expression, RNases that degrade them and the sRNA/target RNA complexes and, finally, their biological roles.Editor: Wolfgang Hess

ABSTRACT

Toxin–antitoxin (TA) loci consist of two genes: a stable toxin whose overexpression kills the cell or causes growth stasis andan unstable antitoxin that neutralizes the toxin action. Currently, five TA systems are known. Here, we review type I andtype III systems in which the antitoxins are regulatory RNAs. Type I antitoxins act by a base-pairing mechanism on toxinmRNAs. By contrast, type III antitoxins are RNA pseudoknots that bind their cognate toxins directly in an RNA–proteininteraction. Whereas for a number of plasmid-encoded systems detailed information on structural requirements, kineticsof interaction with their targets and regulatory mechanisms employed by the antitoxin RNAs is available, the investigationof chromosomal systems is still in its infancy. Here, we summarize our current knowledge on that topic. Furthermore, wecompare factors and conditions that induce antitoxins or toxins and different mechanisms of toxin action. Finally, wediscuss biological roles for chromosome-encoded TA systems.

Keywords: small regulatory RNA; toxin–antitoxin system; RNA antitoxin; type I and type III TA systems; antisense RNA;target RNA interaction; toxic peptide

INTRODUCTION

Toxin–antitoxin (TA) loci encode two-component modules thatconsist of a stable ‘toxin’ whose ectopic overexpression eitherkills cells or confers growth stasis, and an unstable ‘antitoxin’that neutralizes the toxin action (reviewed in Brantl 2012b).Currently, five TA systems are known (summarized in Fig. 1).Whereas in type I and type III TA systems, the antitoxin is asmall RNA, and the toxin mRNA encodes a peptide, in type II,IV and V systems, both antitoxin and toxin (TA) are proteins.In type I TA systems, the RNA antitoxin interacts with the toxinmRNA, and in type III systems, it binds the toxin protein directly(see below). In type II systems, the antitoxin inhibits the toxinby a protein-protein interaction, but in type V systems, it is an

RNase that cleaves the toxin mRNA thus preventing toxin ex-pression (Wang et al. 2012). By contrast, the antitoxin in type IVTA systems interferes with binding of the toxin protein to itscellular target (Masuda et al. 2012). Biological functions of type Iand type III TA systems comprise plasmid maintenance, persis-tence against antibiotics and abortive phage infection. However,numerous chromosome-encoded systems still await their func-tional characterization.

Initially, type I TA systems have been discovered on plasmids[e.g. hok/Sok on Escherichia coli plasmid R1 (Gerdes, Rasmussenand Molin 1986); fst/RNAII on Enterococcus faecalis plasmidpAD1 (Weaver and Clewell 1989; Weaver and Tritle 1994)] wherethey act as post-segregational killing systems (PSK) to prevent

Received: 10 September 2014; Accepted: 20 January 2015C© FEMS 2015. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com

413

414 FEMS Microbiology Reviews, 2015, Vol. 39, No. 3

Figure 1. Overview of currently known types of TA systems. The black and white

bars denote the –35 and –10 boxes of promoters, respectively. Toxins are drawnin blue and antitoxins in red. The ORFs of toxins/antitoxins are represented bylight blue/red bars. Grey boxes indicate SD sequences. (A) The type I TA system

txpA/RatA from B. subtilis. The interaction between both RNAs leads to the degra-dation of the toxinmRNA. (B) The relE/relB type II TA system from E. coli. Both thetoxin and antitoxin are proteins and their interaction results in neutralizationof the toxin. (C) The toxI/N type III TA system from P. atrosepticum. The antitoxin

RNA directly binds to the toxic protein for neutralization. (D) The cbtA/yeeU typeIV TA system from E. coli. The toxin CbtA interacts with MreB and FtsZ, therebyinterfering with their polymerization and consequently with cytoskeleton as-sembly. The YeeU antitoxin counteracts the toxin action by stabilizing MreB and

FtsZ polymers. (E) The ghoT/ghoS type V TA system from E. coli. The antitoxinGhoS acts as RNase and degrades the toxin mRNA.

plasmid loss at cell division. Later on, related and unrelatedchromosome-encoded type I TA systems were found and inves-tigated in detail in Gram-negative and Gram-positive bacteria.Numerous other predicted systems still await experimental con-firmation (Fozo et al. 2010).

Type I TA systems are either arranged as overlapping, con-vergently transcribed gene pairs (as hok/Sok) or as divergentlytranscribed gene pairs located apart (as e.g. tisB/IstR1, Darfeuilleet al. 2007). In the first case, the antitoxin is a cis-encoded sRNA,and in the second, it is a trans-encoded sRNA (for reviews, seeBrantl 2009, 2012a). Only five years ago, the first type III TA sys-tems were discovered (Fineran et al. 2009; Blower et al. 2012). Inthese systems, TA genes are arranged in 5′-3′ tandemorientationand separated by a weak transcriptional terminator. Upon inter-action, a trimeric complex comprising three toxin and three an-titoxinmolecules folding into RNApseudoknots is formed (Shortet al. 2013).

What is known on toxin and antitoxin action?

The majority of type I toxins are small hydrophobic proteins(<60 aa) with transmembrane domains that are thought to actlike phage holins, i.e. they induce membrane pores, which im-pairs ATP synthesis with consequences for replication, tran-scription and translation. Other type I toxins are enzymes likeRNase SymE (Kawano, Aravind and Storz 2007) or the recentlyidentified DNase RalR (Guo et al. 2014). The few currently knowntype III toxins are endoRNases that process the antitoxin pre-cursors into unit-length molecules and cleave cellular RNAs un-specifically (Short et al. 2013).

Type I antitoxins employ several mechanisms to counteracttoxin action irreversibly: they either inhibit toxin translation bydifferent means or promote the degradation of the toxin mRNAor—as we could show recently (Jahn and Brantl 2013)—do both.Additional strategies are used to prevent premature toxin ex-pression. By contrast, type III antitoxins act in a reversible man-ner by sequestration of their toxin proteins in trimeric com-plexes (Short et al. 2013).

Whereas plasmid-encoded type I systems are required forplasmid maintenance, the biological function of chromosome-encoded type I TA systems is far less clear. Recently, in E. colitisB/IstR1, a role in persister formation has been found (Dorr,Vulic and Lewis 2010). In Bacillus subtilis, several type I TA sys-tems are located on prophages and suggested to be required fortheirmaintenance (reviewed in Durand et al. 2012). Other reportshypothesize a role in metabolic or stress response or in biofilmformation (Domka, Lee and Wood 2007). By contrast, type III TAsystems confer phage resistance by acting as abortive phage in-fection (abi) modules: they enable a phage-infected cell to com-mit altruistically suicide to protect the clonal bacterial popula-tion (Blower et al. 2011).

In this review, we discuss in detail all currently knownmech-anisms of action employed by type I and type III antitoxins,structural requirements for their interaction with the cognatetargets as well as interaction kinetics and the role of RNases andaccessory proteins in TA mRNA degradation. In these aspects,we compare RNA antitoxins with other chromosome-encodedsRNAs involved in control of metabolism, stress response andvirulence. Furthermore, we outline expression conditions fortoxin and antitoxin genes and summarize our current knowl-edge on toxins of type I and III, their functions and cellular tar-gets. Finally, we present data and hypotheses on the biologicalfunction of type I and type III TA systems.

Brantl and Jahn 415

TA SYSTEMS WITH RNA ANTITOXINS

Two examples for type I TA systems that were discovered inthe 1980s and have been studied in great detail are located onplasmids, hok/Sok in E. coli and fst/RNAII in En. faecalis. Later,a plethora of chromosome-encoded type I TA systems havebeen discovered in a broad variety of bacterial species, amongthem also hok/Sok or fst/RNAII relatives. In 2009, the first typeIII TA system has been identified, and only two type III systems,in Pectobacterium atrosepticum and in Lactococcus lactis, wereinvestigated in some detail (see below). Table 1 provides anoverview of toxin mRNAs and antitoxins of all currently known,experimentally verified/studied type I systems, their geneticorganization, mechanism of antitoxin action and regulation oftoxin expression.

ANTITOXINSType I antitoxins

Mechanism of antitoxin actionCurrently, two primary mechanisms of antitoxin action areknown: inhibition of translation or promotion of toxin mRNA

degradation. However, the first bifunctional antitoxin has beenidentified recently: B. subtilis SR4 obstructs translation of bsrGRNA and promotes its degradation (Jahn and Brantl 2013).

Inhibition of toxin mRNA translation can be accomplishedby different means. The most trivial case is E. coli symE/SymR,where binding of the SymR antitoxin at a region overlappingthe toxin RBS (ribosome binding site) inhibits ribosome bind-ing directly (Kawano, Aravind and Storz 2007; Fig. 2A). By con-trast, E. coli hok/Sok and ldrD/RdlD employ inhibition of leader-peptide translation. In hok/Sok, translation of the toxin hok (hostkilling) is difficult and requires translation of the leader-peptidemok (modulation of killing). The antitoxin Sok (suppression ofkilling) binds to the Shine Dalgarno (SD) sequence of mok andthereby indirectly prevents hok mRNA translation (Fig. 2B). Like-wise, a mok-like open reading frame (ORF) designated ldrX over-laps ldrD, as the ldrDmRNA cannot bind ribosomes directly, andthe antitoxin RdlD obstructs leader-peptide ldrX translation (re-viewed in Gerdes andWagner 2007; Kawano 2012). Alternatively,the RNA antitoxin can inhibit translation by blocking a ribo-some standby site (RSS; Darfeuille et al. 2007): in E. coli tisB/IstR1,the tisB RBS is sequestered by intramolecular base pairing. Theantitoxin IstR1 binds 100 nt upstream of the tisB translation

Table 1. Overview of currently known Type I TA systems: occurrence, characteristics, mechanism of action and regulation of expression.

Species(plasmid) TA system

Length of AT(nt)

Length of toxinRNA

Mode of ATaction

Arrangement(complementaryregions)

Regulation/peculiarity

E. coli (R1)chromosome

hok/SokhokA,B,C,D,E,F

64 398, 361 hokB360, 330

TI Convergent (64) Locus on plasmid,homologous loci inchromosome

E. coli symE/SymR 77 454 TI Convergent (77) Toxin is RNase andSOS induced

E. coliS.typhim./typhCitrobacter

ldr/Rdl (ABCD)ldr1, ldr2/RdlldrCf/Rdl

66 372 TI Convergent (66) Homologous lociin other bacteria

E. coli tisB/IstR1 75 292/250/186 TI Divergent (21) Toxin SOS inducedE. coli ibs/Sib 110-140 ≈200 TI Convergent (110–140) Multiple lociE. coli zorO/OrzO 81 264 TI? Divergent (19)E. coli shoB/OhsC 60 320/280 TI Divergent (18)E. coli dinQ/AgrAB 84 331 TI Divergent (34) Toxin SOS-induced

tisB/IstR homologE. coli ralR/RalA 179 195 TI ? RD? Convergent (16) Toxin is DNase

Hfq requiredB. subtilis txpA/RatA 222 270 RD Convergent (118)B. subtilis bsrG/SR4 180 294 TI + RD Convergent (123) bsrG RNA

temperaturesensitive

B. subtilis bsrE/SR5 170 255-260 RD Convergent (113) bsrE RNA nutrientdependent

B. subtilis yonT/as-yonT 100 >450 ? Convergent (111)B. subtilis bsrH/as-bsrH 200 285 RD? Convergent (140) bsrH RNA degraded

by RNase J1En. faecalis(pAD)

fst/RNAII 66 214 TI, RS Convergent (8, 12, ≈35)

S. aureuschromosome/plasmids

sprA1/SprA1AS 60 208 TI RS or RD? Convergent (10, 16)/(9, 14)

fst/RNAIIhomologs

S. aureus sprF/SprG 141 439/312 TI? RD? Convergent (134) txpA/RatAhomolog multipleloci

St. mutans fst-Sm/SRSm ≈65 ≈210 TI? Convergent (12, 16, ≈32) fst/RNAII homolog

TI, translational inhibition; RD, promotion of toxin mRNA degradation; RS, RNA stabilization; ?, regulation proposed but not experimentally shown. For further details

see text.

416 FEMS Microbiology Reviews, 2015, Vol. 39, No. 3

Figure 2. Mechanisms employed by type I antitoxins. The black and white bars denote the –35 and –10 boxes of promoters, respectively. Yellow boxes indicate LexA

binding sites for genes under SOS control. Toxins are drawn in blue and antitoxins in red. The toxin ORFs are represented by light blue bars. Grey boxes denote SDsequences, and white boxes the RSS. Green arrows symbolize endoribonucleases (dark green: RNase III, light green: RNase Y) and circular segments 3′-5′ exoribonu-cleases (white: unknown, yellow: PNPase, orange: RNase II, brown: RNase R). ?, cleavage was not demonstrated experimentally. (A–C) illustrate three different modesof translation inhibition. (A) Direct blocking of SD sequence. SymR antisense RNA inhibits translation of symE mRNA by direct blocking of the SD sequence. The LexA

box indicates that the symE promoter is under SOS control. (B) Inhibition of leader-peptide translation. The hok/sok locus from E. coli plasmid R1 comprises three genes:‘host killing’ (hok) encodes the toxic peptide, ‘modulation of killing’ (mok) overlaps with hok and is required for hok translation and finally, ‘suppression of killing’ (sok)represents the antitoxin that blocks translation of mokmRNA. Since translation of hok depends onmok translation, Sok indirectly inhibits hok translation by inhibiting

mok translation. (C) Blocking of an RSS. IstR1 of E. coli binds to the RSS of tisB mRNA. Consequentially, ribosomes cannot access this RNA region and the standby-dependent translation of tisB mRNA is prevented. (D) Promotion of RNA degradation. Antitoxin RatA interacts with the txpA toxin mRNA by base pairing at their 3′

ends. RNase III cleaves the txpA/RatA duplex, followed by further degradation by RNase Y, thereby preventing toxin translation. (E) RNA degradation and translationinhibition. Antitoxin SR4 and bsrG toxin mRNA interact at their 3′ ends. Binding of SR4 to bsrG RNA induces the extension of the 4 bp to an 8 bp double-stranded

region sequestering the SD sequence that inhibits bsrGmRNA translation to a certain degree. Additionally, the SR4/bsrG RNA interaction promotes toxin mRNA degra-dation by an initial RNase III cleavage followed by further degradation by both RNases R and Y. (F) Translation inhibition and RNA stabilization. RNAI and RNAII havethree dispersed complementary segments (the direct repeats DRa, DRb and the transcription terminators). DRa and DRb are shown in black. The RNAI/RNAII complexprevents toxin translation, since the region around the toxin start codon is sequestered by the antitoxin. Complex formation results in stabilization of both RNAs.

initiation region (TIR) at an unstructured RSS, thereby restrain-ing ribosomes from binding there and sliding into the TIR, whenthe stem-loop region unfurls incidentally (Fig. 2C, reviewed inWagner and Unoson 2012). In E. coli zorO/OrzO, the antitoxin alsobinds within the 180 nt long 5′ UTR, and it is not yet clear, if thisregion comprises a small translated ORF as in hok/Sok or an RSSas in tisB/IstR1 (Wen,Won and Fozo 2014). Alternatively, cleavagesites detected in a �rnc strain within the complementary regionand downstream of it, but upstream of the zorO RBS, do not ex-clude that OrzO promotes RNA degradation by employing a sofar unidentified E. coli endoribonuclease (the existence of whichwas proposed by Opdyke et al. 2011).

In the B. subtilis type I TA systems investigated so far[txpA/RatA (Silvaggi, Perkins and Losick 2005); bsrG/SR4 (Jahnet al. 2012)], the RNA antitoxin binds at the 3′ end of the toxinmRNA and supports its degradation (Fig. 2D). Both RNAs share

a long complementary stretch of ≈120 bp. For bsrG/SR4, wehave found recently that SR4 does not only promote degrada-tion of bsrGmRNA but induces a conformational change aroundthe bsrG RBS that further obstructs ribosome binding, thus,additionally impeding toxin translation (Jahn and Brantl 2013;Fig. 2E) making SR4 the first bifunctional antitoxin. Such a con-formational change was neither observed upon RatA bindingto txpA mRNA (Durand, Gilet and Condon 2012) nor for an-titoxin SR5 in another B. subtilis type I TA system, bsrE/SR5[formerly bsrE/as-bsrE (Meißner, Jahn and Brantl unpublished)],suggesting that RatA and SR5 do solely cause toxin degrada-tion. It remains to be seen if this also holds true for yonT/as-yonT, a system, for which evidence that it is a type I TA sys-tem is so far only indirect, as deletion of as-yonT has not beeninvestigated in vivo in B. subtilis (Durand, Gilet and Condon2012).

Brantl and Jahn 417

In all the cases discussed above, TA RNA binding involvesonly one complementary region.

By contrast, En. faecalis RNAII and fst toxin mRNA (RNAI) in-teract at different regions located apart: binding starts at the 3′

end of both RNAs at the loops of a bidirectional transcription ter-minator that comprises ≈35 bp, and later, two direct repeat (DR)regions located further upstream interact yielding a partial du-plex that suffices to block fst translation (Greenfield et al. 2001;Fig. 2F). Mutations in either the terminator or simultaneously inboth DR regions abrogated the ability of the antitoxin to neu-tralize toxicity in vivo (Greenfield et al. 2000). In the RNAI/RNAIIcomplex, the toxin start codon is located in a double-strandedregion. Interestingly, TA RNA interaction does not result in fsttoxinmRNA degradation, but in accumulation of a complexwithhigher stability of both the toxin mRNA (40 instead of 9 min,measured with a mutant that lacked the 5′ stabilizing helix),and the antitoxin RNAII (16 instead of 4 min). For the recentlypublished Staphylococcus aureusTA system sprA1/SprA1AS (Sayed,Jousselin and Felden 2011; Sayed et al. 2012), a chromosomalmember of the fst/RNAII family, toeprinting and in vitro transla-tion in the presence and absence of the antitoxin SprA1AS indi-cate that it inhibits translation. However, as RNA half-lives havenot yet been determined, it cannot be excluded that translationinhibition is accompanied by RNA stabilization as in fst/RNAII orby RNA degradation as in bsrG/SR4. Similar to En. faecalis RNAII,the E. coli Sib antitoxins use two target recognition domains,TRD1 within the ibs ORF and TRD2 in the ibs RBS (Han et al. 2010;reviewed in Fozo 2012), to inhibit ibs translation. The interactionof TRD1 with its complement within the ibs ORF is critical forthe discrimination between different, highly similar target tox-ins (Han et al. 2010).

Another S. aureusTA system, sprG1/SprF1 (Pinel-Marie, Brielleand Felden 2014), bears similarity to B. subtilis txpA/RatA orbsrG/SR4, as both genes overlap at their 3′ ends in a long comple-mentary region (134 nt) which would suggest an RNA degrada-tionmechanism. However, since no influence of SprF1 on sprG1-mRNA levels has been investigated (neither in wild-type nor inRNase III knockout strains), it is not clear, if the SprF1-mediatedtranslation inhibition observed in vitro is caused by induction ofstructure alterations at the toxin RBS as in B. subtilis bsrG/SR4followed by RNA degradation in vivo.

Interaction kinetics and binding pathway of RNA antitoxinand toxin mRNAFor many cis-encoded antisense/sense RNA pairs, binding path-ways have been studied in detail and binding kinetics measured(reviewed in Brantl 2007). Apparent pairing rate constants (kapp)are usually in the range of 1 × 105 to 1 × 106 M−1 s−1. Simi-lar values have been also calculated for several type I TA pairs:for hok/Sok, 1 × 105 M−1 s−1 (Thisted et al. 1994) and in bothfst/RNAII (Greenfield et al. 2001) and bsrG/SR4 (Jahn and Brantl2013) kapp values of 106 M−1 s−1 were reported. For the two in-teraction sites between ibs RNA and Sib, 2 × 105 M−1 s−1 and5 × 105 M−1 s−1 were determined, respectively (Han et al. 2010).For E. coli tisB/IstR, 2 × 105 M−1 s−1 have been measured (F. Dar-feuille and E. G. Wagner unpublished data).

Initial contacts between sRNA and target mRNA can eitheroccur between two complementary loops (as found for a num-ber of replication control systems) or between a loop and asingle-stranded region. The latter has been found for hok/Sok,where the 5′ single-stranded tail of Sok recognizes a loop ofthe hok target RNA that tops a stem comprising the mok SD se-quence (Thisted and Gerdes 1992; Fig. 2B), and binding mostprobably follows a one-step scheme (Fig. 3A). Independent of a

Figure 3. Interaction pathway between TA mRNA. Toxin mRNAs are drawn inblue and antitoxins in red. U-turns (YUNR motives) are indicated by green linesin the target/toxin RNA. The SD sequences are demonstrated by grey boxes.Chronology of interaction is designated by 1, 2 and 3. L = loop. Whereas in B,

TA RNA interactions involve separated complementary regions and the result isa partial duplex, the interaction in A, C and D yields a complete duplex. (A) hokmRNA and Sok of E. coli. The initial contact occurs between the single-stranded5′ end of Sok and a loop of hok mRNA that tops the mok-TIR (1) and progresses

towards the 5′ end of hokmRNA leading to formation of a complete duplex. Bind-ing follows a one-step pathway. (B) RNA I (fstmRNA) and RNA II of En. faecalis. Thedirect repeats (DRa and DRb) are shown in black. The two RNAs have three dis-

persed complementary segments. Interaction between the two RNAs is initiatedat the U-turn motif present in the terminator loop of RNA I (1). Subsequently,the direct repeat sequences DRa and DRb interact subsequently (2, 3) forming apartial duplex. (C) bsrG mRNA and SR4 of B. subtilis. The initial contact between

SR4 and bsrG RNA occurs between L4 of SR4 and L3 of bsrG RNA (1), followed byhelix progression to an interaction between L3 of SR4 and the 3′ part of helix P1of bsrG (2) and reaches finally L2 of SR4 that binds terminator loop L4 of bsrG RNA(3). (D) ibsC mRNA and SibC of E. coli. SibC has two target recognition domains,

TRD1 and TRD2, which function independently (1). The target site for TRD1 islocated within the ORF of ibsC mRNA, whereas the target site for TRD2 is not.The simultaneous interactions at these two recognition sites allows for progressof base pairing into the flanking sequences eventually resulting in a complete

duplex.

418 FEMS Microbiology Reviews, 2015, Vol. 39, No. 3

one- ormulti-step pathway, the final result of the interaction is acomplete duplex that is degraded by the double-strand-specificRNase III (see below).

Although a time-course experiment for fst/RNAII showedthat both terminator and DR regions of toxin and antitoxin RNAsmight interact simultaneously, the use of a series of loop muta-tions allowed to ascertain that interaction starts at the two ter-minator loops, afterwards involves the DRa region, reaches theDRb region and from there, progresses into the final complex,which is, however, only a partial duplex (Greenfield et al. 2001;Fig. 3B). In the case of bsrG/SR4, binding initiates between theterminator stem loop (SL4) of SR4 and loop 3 of bsrG RNA, fol-lowed by an interaction between SL3 of SR4 and a region withinthe main helix of bsrG RNA. Finally, SL2 of SR4 and the bsrG ter-minator loop interact, but this last step is not required for inhi-bition (Jahn and Brantl 2013; Fig. 3C). By contrast, in E. coli ibs/Sib,a simultaneous interaction between TRD1 and TRD2 of Sib an-titoxin and their complementary domains in ibs RNA was ob-served (Han et al. 2010; Fig. 3D). This is reminiscent of the simul-taneous interaction of two loop pairs in RNAII/RNAIII, the repli-cation control system of plasmid pIP501 (Heidrich and Brantl2007).

In many sense/antisense systems, 5′ YUNR motifs presentin either the antisense or the target mRNA were found to beimportant to provide a scaffold for rapid RNA-RNA interaction(reviewed in Franch and Gerdes 2000). Interestingly, the firstof these 5′ YUNR motifs was discovered in hok mRNA, andthis motif was experimentally shown to form a U-turn (Franchet al. 1999; Fig. 3A). Surprisingly, in the bsrG/SR4 system, twopotentially important 5′ YUNR motifs were detected. However,whereas that in loop L3 of bsrG RNA was involved in the ini-tial contact with loop L4 of SR4 (see Fig. 3C), the motif in loopL2 of SR4, which interacts with the bsrG terminator loop, wasnot required, as the entire SL2 was not needed for the inhibitoryfunction of SR4 (see above, Jahn and Brantl 2013). Preliminarydata suggest that in B. subtilis bsrE/SR5 the initial rapid interac-tion occurs between two loops each containing a U-turn motif(Meißner, Jahn and Brantl unpublished). In fst/RNAII, the fst ter-minator loop contains a 5′ YUNR motif, and, as stated above, itis involved in the primary interaction with the RNAII termina-tor loop (Fig. 3B). In the case of ibs/Sib, a 5′ YUNR motif is foundin the decisive loop within the ibs mRNA ORF (Han et al. 2010),whose interaction is important for the discrimination betweenthe different ibs/Sib systems (see Fig. 3D). In the S. aureus fst ho-molog sprA1/SprA1AS (Sayed, Jousselin and Felden 2011), the an-titoxin loop L1 carries a potential 5′ YUNR motif, whose ‘R’ was,however, mapped in the double-stranded stem region whichmakes it unlikely to form a U-turn.

However, except for hok/Sok, no experimental proof for theformation of U-turns at the 5′ YUNR positions in other TA sys-tems was provided, so far. In txpA/RatA, neither of the interact-ing loops carries a 5′ YUNR motif (see secondary structures inDurand, Gilet and Condon 2012). For the other type I TA sys-tems, no experimentally probed RNA secondary structures areavailable to ascertain, if potential 5′ YUNR motifs in either an-titoxin or toxin mRNA are present in loop regions. Recently, forzorO/OrzO and zorP/OrzP, it has been shown that the immediate5′ end of the RNA antitoxin is required for specific pairing anddiscrimination between different zor/Orz systems (Wen, Wonand Fozo 2014). However, although this end contains a 5′ YUNR,it cannot form a loop.

Many antisense RNAs mediate inhibition by forming com-plexes that involve limited numbers of base pairs with theirtargets, and full duplex formation, which would be too slow to

account for the observed biological effects, is not required forcontrol (reviewed in Wagner and Brantl 1998). Whereas e.g. inhok/Sok, symE/SymR or bsrG/SR4, toxin-antitoxin binding finallyleads to a complete duplex, other systems such as fst/RNAII,tisB/IstR, zorO/OrzO or ralR/RalA employ only short complemen-tary regions whose interaction results in a partial duplexes.

Additional modes to prevent premature toxin expressionAlthough toxin inhibition by the antitoxin is the main regula-tory principle, additional strategies are employed to ensure tightregulation of toxin expression. All these strategies use the highcapacity of RNA to fold and refold into complex structures.

One strategy is the requirement for a processing event thatconverts an inactive long toxinmRNA into a shorter translation-ally competent molecule: in E. coli hok/Sok, the 3′ end (termedfbi = fold-back inhibition element) of the 398 nt long hok mRNAinteracts with its 5′ end to block the mok TIR (Thisted, SørensenandGerdes 1995; Fig. 2B). The long hok species is slowly degradedby 3′-5′-exoribonucleases RNase II and PNPase into a shorter (361nt) species that is translationally competent (Fig. 2B). After pro-cessing, Sok binds very efficiently to the mok SD in the shorthok mRNA to inhibit its translation (Thisted et al. 1994; Thisted,Sørensen and Gerdes 1995). Rapid binding of Sok to full-lengthhok mRNA would be detrimental, as the hok RNA/Sok duplex israpidly cleaved by RNase III (Gerdes et al. 1992) and thus, the ac-tivatable pool of hok mRNA would be depleted too quickly.

Similarly, in E. coli tisB/IstR1 (Vogel et al. 2004; Darfeuilleet al. 2007), the primary tisB transcript (+1) is translationally inac-tive and has to be processed within its 5′ end into a shorter (+42)translationally competent (lowly abundant) molecule (Fig. 2C).Only the +42 transcript has an unstructured 5′ segment contain-ing an RSS to which, at the same time, the antitoxin IstR1 canbind to prevent binding of the 30S subunit. After IstR1 binding,RNase III cleaves the +42 transcript into another translationallyinactive +106 tisB RNA molecule.

In E. coli shoB/OhsC, different 5′ ends of shoB (+1, +42) hadbeen mapped (Kawano et al. 2005). The longer of two shoBmRNA species (≈320 nt) was translationally inactive, whereasthe shorter (≈280 nt) showed low activity and could be ≈2-foldinhibited by OhsC (Fozo et al. 2008), indicating another case oftoxin-mRNA processing.

Alternative approaches were found in Gram-positive bacte-ria: in B. subtilis, txpA, bsrG and bsrE sequester their SD sequencesin double-stranded regions comprising 4 or 5 GC base pairs (re-viewed in Durand et al. 2012), which suffice to obstruct ribo-some binding. For bsrG, neither toeprint nor detectable amountsof in vitro translation product could be obtained from wild-type mRNA unless mutations were introduced that opened thedouble-stranded region (Jahn and Brantl 2013). In En. faecalis, thefst SD sequence is also sequestered in a double-stranded regionof four adjacent base pairs in a stem loop termed 5′ SL (Green-field et al. 2000) and postulated to temporarily inhibit ribosomebinding until the terminator loop can be transcribed (Shokeenet al. 2008). A processing event was neither observed in fst/RNAIInor in the B. subtilis systems.

Furthermore, txpA and yonT have perfect RBS (≥11 bp com-plementarity to anti-SD in 16S rRNA), which are predicted toefficiently recruit, but slowly release ribosomes (reviewed inDurand et al. 2012).

Beyond that, in some cases—E. coli dinQ, B. subtilis yonT andEn. faecalis fst—the rare start codon GUG is used instead of AUG,which is also expected to decrease the efficiency of translation,as it is used in only 14% and 9%of all genes in E. coli and B. subtilis,respectively (Rocha, Danchin and Viari 1999).

Brantl and Jahn 419

Interestingly, for S. aureus sprA1mRNA, it was suggested thatthe compact secondary structure comprising RNA pseudoknotsflanked by stable stem loops would obstruct internal ribosomebinding (Sayed, Jousselin and Felden 2011). Recently, the samepseudoknots were proposed for the structurally similar En. fae-calis fst mRNA (Weaver 2014).

Role of HfqSurprisingly, the ralR/RalA TA system of E. coli prophage rac re-quires Hfq for antitoxin function (Guo et al. 2014). This might bedue to the rather short (16 nt) sequence complementarity be-tween ralR and RalA. At high concentrations, RalA binds Hfq,but it is still unknown if Hfq promotes ralR/RalA complex forma-tion or if it affects the stability of either RNA or perhaps transla-tion of ralR (Guo et al. 2014). The latter would resemble anothersRNA/target RNA system, SR1/ahrC, where Hfq was needed forahrC mRNA translation (Heidrich, Moll and Brantl 2007). Cur-rently, ralR/RalA is the only type I TA system, where a func-tion for Hfq has been found. Neither the other E. coli antitoxinswith relatively short (18 to 19 nt) complementary regions to theirtoxin mRNAs like IstR1 (Darfeuille et al. 2007), OrzO (Wen, Wonand Fozo 2014), OhsC or the Sibs (E. Fozo, personal communica-tion) nor the B. subtilis antitoxins like SR4 (Jahn et al. 2012) thatuse ≈120 nt of complementarity need Hfq. En. faecalis (fst/RNAIIsystem) does not even encode Hfq. However, it is not excludedthat other, still unidentified RNA binding proteins might play arole for some of the toxin-antitoxin pairswith short base-pairingregions.

Type III antitoxins

In contrast to type I antitoxins that use a base-pairing mech-anism to inhibit their corresponding toxins, type III antitoxinsact by RNA-protein binding. In this regard, they resemble regula-tory sRNAs that sequester proteins like e.g. CsrB from E. coli thatbinds CsrA, 6S RNA that interacts with RNAP or Rcd from plas-mid ColE1 that binds tryptophanase (reviewed in Brantl 2009).

The first type III TA system, toxN/ToxI, was discovered on acryptic plasmid from Erwinia carotovora subsp. atroseptica, laterrenamed P. atrosepticum subsp. atroseptica (Fineran et al. 2009).Whereas type I toxins and antitoxins are transcribed from in-dependent promoters on complementary DNA strands, type IIIantitoxins and toxins are cotranscribed from a single promoterupstream of the antitoxin gene (see Fig. 1C). Infrequent (10%)read-through of a Rho-independent transcriptional terminatorseparating both genes warrants an excess of antitoxin overtoxin. Full-length antitoxin ToxI (200 nt) consists of 5.5 sequencerepeats of 36 nt, but a single repeat suffices to repress ToxNtoxicity (Fineran et al. 2009; Blower et al. 2011). ToxI monomersare generated from the 200 nt precursor via sequence-specificprocessing by the ToxN endoRNase. Structure analysis revealedthat ToxI folds into an RNA pseudoknot. Three ToxI monomersbind three ToxN proteins in a trimeric ToxI/ToxN complex re-sulting in inhibition of ToxN function. An extensive set of RNA–protein contacts is used. ToxN functions by a reversible growth-inhibiting (bacteriostatic) mechanism (Fineran et al. 2009).

The second currently known type III TA system, abiQ/AntiQ,was reported in 2013 in L. lactis and the toxin structure solved(Samson et al. 2013). As toxN/ToxI, it is located on a plasmid andrequired for abortive phage infection. Although the primary se-quences of both type III loci show only 31% similarity, the an-titoxin gene is always located 5′ of the toxin gene, separatedfrom it by a weak terminator and comprises several (here 2.8 in-stead of 5.5) 35 nt repeats. The AbiQ toxin is also an RNase that

cleaves the antitoxin repeats, and its 3D structure is similar tothat of ToxN. AntiQ repeats also form pseudoknots, and, uponAbiQ binding, the RNase activity of the toxin is inhibited.

A recent bioinformatics approach predicted a variety of typeIII TA systems in the chromosomes and plasmids of differentbacterial species. In addition to toxN/ToxI, two novel type III TAfamilies, cptN/CptI and tenpN/TenpI, with slightly longer anti-toxin repeats (40–48 and 39–57, respectively) and toxins of ap-proximately the same size as ToxN (148 to 160 aa) were found.Among them, cptIN is only chromosome encoded and—so far—not detected on plasmids (Blower et al. 2012).

Role of RNase III and other RNases involvedin antitoxin/toxin-mRNA degradation

Already 25 years ago, it had been shown for many cis-encodedregulatory sRNAs ( = antisense RNAs) that the ubiquitousdouble-strand-specific RNase III cleaves the antisense/senseRNA duplex thereby initiating degradation of the target mRNAand, consequently, inhibiting target gene expression. The firstcases analysed in detail were plasmid-, transposon- or phage-encoded antisense RNAs like e.g. CopA regulating replicationof E. coli plasmid R1 (Blomberg, Wagner and Nordstrom 1990),RNA-OUT controlling transposition of IS10 (Case, Simons andSimons 1990) and phage λ OOP RNA regulating cII mRNA, anactivator of λ cI transcription (Krinke and Wulff 1987, 1990).In type I TA systems, a role of RNase III was also found forhok/Sok of plasmid R1 (Gerdes et al. 1992) and 20 years later forbsrG/SR4 (Jahn et al. 2012) and txpA/RatA (Durand, Gilet and Con-don 2012) from the B. subtilis chromosome. However, in somecases—λ OOP, RatA, as-yonT (Durand, Gilet and Condon 2012)and OrzO (Wen, Won and Fozo 2014)—RNase III is essential fortarget inhibition, whereas it is not required by CopA, RNA-OUTand the type I antitoxins Sok (reviewed in Brantl 2007), SymR(Kawano, Aravind and Storz 2007), IstR1 (Wagner and Unoson2012), Sib (Fozo 2012), B. subtilis SR4 (Jahn et al. 2012) and mostprobably, En. faecalis fst (Weaver 2014). RNase III seems to beneeded only in cases, where target control is exerted exclu-sively via mRNA degradation, whereas in those cases, wheresRNAs/antitoxins regulate steps preceding degradation, e.g.translation (Sok, SymR, IstR1), RNase III is dispensable andmerely employed to cleave subsequently the sRNA/target RNAduplex. Interestingly, for zorO/OrzO, so far no mechanism of an-titoxin action has been elucidated, and it is tempting to spec-ulate that this antitoxin that binds to a region 73 to 91 nt up-stream of the zorO start codon within the 5′ UTR might induce aprocessing event to trigger toxin mRNA degradation. Indeed, inthe presence of RNase III, 19 and 20 nt downstream from the endof the zorO/Orz complementary region, two clevage sites, mostprobably due to action of a single-strand-specific endoribonucle-ase, have been mapped (Wen, Won and Fozo 2014).

The above hypothesis is corroborated by a comparison be-tween bsrG/SR4 and txpA/RatA from B. subtilis: SR4 is a bifunc-tional antitoxin that does not only promote degradation but alsoinhibits translation of bsrG mRNA (Jahn and Brantl 2013). Thefact that neither cell lysis nor mutations in the bsrG ORF wereobserved in a �rnc strain (Jahn et al. 2012) confirms that RNaseIII is not essential for toxin inhibition. RNase III cleaves bsrGmRNA at position 185 i.e. 14 nt downstream of the 3′ end ofcomplementarity with SR4. Half-lives of both SR4 and bsrG RNAwere ≈2.5-fold increased in the absence of RNase III, whereasdegradation of SR4 or bsrG RNA alone were not affected (Jahnet al. 2012). By contrast, a recent analysis of txpA/RatA revealedthat RNase III cleavage of txpA RNA in the txpA/RatA duplex

420 FEMS Microbiology Reviews, 2015, Vol. 39, No. 3

prevents toxin expression and is essential (Durand, Gilet andCondon 2012). RatA acts only on target mRNA stability and doesnot induce a txpA mRNA conformation that further inhibitstranslation (shown by in vitro secondary structure probing inDurand, Gilet and Condon 2012). In a �rnc strain, txpA (toxin)RNA levels were 6-fold increased, whereas those of RatA wereunaffected. Instead, RatA degradation is initiated by the ma-jor B. subtilis endonuclease RNase Y reflected by higher RatAlevels in an rny strain (see Fig. 2D). After initial cleavage byRNase Y, the 5′ segment of RatA alone is degraded by 3′-5′-exoribonuclease PnpA and the 3′ segment by 5′-3′ exoribonu-clase J1 (Even et al. 2005). The RNase III cleavage products of txpAmRNA are further processed by the 3′-5′ exoribonuclease PnpA(Durand, Gilet and Condon 2012).

The role of RNase III for antitoxin stability differs betweenthe various type I TA systems analysed so far: two other B. sub-tilis type I antitoxins, as-yonT (Durand, Gilet and Condon 2012)and SR5 that regulates bsrE RNA (Ring, Jahn and Brantl unpub-lished data), behave like RatA in that their half-lives are not af-fected by RNase III. Most probably, the relative amount of TAmRNA is decisive: both RatA (Durand, Gilet and Condon 2012)and SR5 (Ring, Jahn and Brantl unpublished) are in excess overtheir target RNAs, whereas for SR4, an excess over bsrG RNA isonly apparent under certain growth conditions (Jahn and Brantlunpublished).

Surprisingly, in B. subtilis bsrH/as-bsrH, the levels of toxinmRNA are neither influenced by RNase III nor RNase Y, but byRNase J1 and degradation is proposed to be rather exoribonucle-olytically (Durand, Gilet and Condon 2012). This is reminiscentof some targets of trans-encoded sRNAs in E. coli, whose degra-dation is not initiated by RNase III but by endoribonuclease E asa consequence of sRNA binding (e.g. RyhB/sodB, Masse, Escorciaand Gottesman 2003; Prevost et al. 2011). It remains to be seen ifother type I toxinmRNAs are substrates for degradation by alter-native RNases and not by RNase III. Recently, we also observedan effect of RNase J1 on the stability of B. subtilis antitoxin SR5that regulates bsrE (Ring, Jahn and Brantl unpublished data).

For bsrG/SR4, an about 2-fold influence of RNase Y on thedegradation of both RNAs was found. Presumably, RNase Ycleaves bsrG RNA to yield ≈213 nt and ≈256 nt long species (Jahnet al. 2012). Themain 3′-5′ exoribonuclease involved in the degra-dation of both bsrG mRNA and SR4 is RNase R, whereas RNasesJ1, RpH and YhaM did not influence the stability of either RNA.The 3′-5′ exonuclease PnpA trims three SR4 precursors fromtheir 3′ ends to the mature 180 nt SR4, but only marginally con-tributes to SR4 or bsrG RNA degradation. In hok/Sok from E. coli,in addition to the role of RNase III for hok mRNA cleavage, con-tributions of the main endonuclease RNase E (the E. coli equiv-alent to RNase Y) and polyA polymerase PAPI have been foundfor Sok degradation (Mikkelsen and Gerdes 1997). By contrast,degradation of full-length hok mRNA (398 nt) is initiated by the3′ exoribonucleases RNase II and PnpA yielding the translation-ally active 361 nt long species (reviewed in Gerdes and Wagner2007).

Although in the case of fst/RNAII, no RNases have been dis-covered so far that influence the stability of antitoxin or toxinmRNA; an upstream helix designated 5′ UH has been analysedthat comprises the immediate 5′ end and nt 133–139 in the cen-tral part of fst RNA (Shokeen et al. 2009). 5′ UH protects the 5′ partof fst RNA from RNases and is at least partially responsible forthe greater stability of fst RNA compared to its antitoxin, RNAII.As described above, TA RNA interaction results in accumulationof a complex with higher stability of the toxin mRNA, namely 40instead of 9min, and of antitoxin RNAII from4 to 16min (Weaveret al. 2004).

In summary, in all but one (bsrH) cases analysed to date, toxinmRNAs are, when in duplexwith the antitoxin, cleaved by RNaseIII, but this is only essential for their inhibition in cases, whereantitoxins exclusively promote RNAdegradation. By contrast, ef-fects of RNase III on RNA antitoxins are only evident when theyare in large excess over their target toxin RNAs. A combinationof other endo- and exoribonucleases—so far only investigatedin detail for E. coli hok/Sok and some B. subtilis TA systems—isinvolved in regulating TA mRNA stability.

Regulation of toxin and antitoxinexpression—threshold-based mechanisms

Toxin expression has to be tightly controlled. A number of type Itoxin genes from E. coli are under SOS control: symE, tisB and dinQhave Lex boxes. By contrast, B. subtilis bsrG is temperature sen-sitive: bsrG RNA has a 3- to 4-fold shorter half-life at high tem-peratures (Jahn et al. 2012). This might be explained by refoldingof bsrG mRNA at 55◦C that makes it accessible to single-strand-specific RNases (Jahn and Brantl unpublished). In other cases,transcriptional regulation depending onmetabolic or stress con-ditions has beenproposed but not yet experimentally confirmed:for instance, upstream of bsrG, bsrH and bsrE, hypothetical ResDsites have been found. In the case of bsrG, however, first anal-yses have shown that this toxin is not regulated by ResD (Jahnand Brantl unpublished). Staphylococcus aureus sprA1 expressiondecreases 2-fold at lower pH, while it increases 3-fold under ox-idative stress (Sayed, Jousselin and Felden 2011). For these alter-ations, responsible factors are not yet known. Interestingly, E. coliralR was found to be higher expressed in later stages of biofilmdevelopment (Domka, Lee and Wood 2007). In the case of SymE,an additional post-translational regulation by Lon protease wasdiscovered.

Much less is known on regulation of antitoxin expression.Usually, antitoxins are expressed constitutively. For instance,E. coli Sib RNAs (Fozo et al. 2008; Fozo, Hemm and Storz 2008),B. subtilis SR4 (Jahn et al. 2012) are expressed equally well in bothrich and minimal media. Exceptions are En. faecalis RNAII and S.aureus SprA1AS (Sayed, Jousselin and Felden 2011) that peak atmid-log phase.

For the type III TA system toxI/ToxN, an autoregulation of thetoxIN promoter by the trimeric toxI/ToxN complex was observed(Blower et al. 2009).

The balance between the levels of TAmRNA is critical for theconditions under which the toxin is expressed. Genetically iden-tical bacterial populations exhibit heterogeneity, with e.g. bothactively growing and dormant cells. This is a great advantageupon rapid changes of environmental conditions. Under condi-tions where antitoxin levels are higher than toxin RNA levels,toxin expression is successfully prevented and the cell survives.By contrast, when toxin RNA levels significantly exceed anti-toxin levels, the toxin escapes repression and causes cell deathor growth arrest. However, when toxin and antitoxin levels areclose, some cells will have an imbalance in the toxin-antitoxinratio and might become persisters (see below). A detailed math-ematical analysis of persister formation was recently published(Rotem et al. 2010).

Comparison between antitoxins and sRNAs involved incontrol of metabolism, stress response and virulence

So far, only differentmodes of translational inhibition or promo-tion of RNA degradation have been found as regulatory mecha-nisms used by type I antitoxins. The only exception is fst/RNAII,where translation inhibition is accompanied by stabilization

Brantl and Jahn 421

of both interacting RNAs. By contrast, cis- and trans-encodedsRNAs from plasmids or chromosomes that act by base pair-ing on their target RNAs can employ a much broader variety ofregulatorymechanisms: someplasmid-encoded antisense RNAsexert target control by transcriptional attenuation (plasmidspT181 and pIP501; reviewed in Brantl 2014), translational attenu-ation (plasmid pSK41, Kwong, Skurray and Firth 2006) or inhibi-tion of pseudoknot formation (In1α/IncB plasmids, Asano andMizobuchi 1998). Other chromosome-encoded sRNAs employtranscriptional interference (Clostridium acetobutylicum ubiGoperon, Andre et al. 2008) or mRNA stabilization owing to a pro-cessing event (E. coliGadY/gadWX,Opdyke et al. 2011). Therefore,it remains to be seen in the future if type I TA systems will bediscovered that use other, so far unprecedented mechanisms ofantitoxin action.

For many chromosomally trans-encoded sRNAs, in particularfrom Gram-negative bacteria, the abundant RNA chaperone Hfqis required for either stabilization of sRNA or target mRNA or topromote complex formation between them. By contrast, only forone type I TA system, ralR/RalA, a role of Hfq has been reported(Guo et al. 2014). However, it is not excluded that other—perhapsstill unknown—RNA binding proteins are involved in the stabi-lization of the rather short complementary regions between TAmRNA in vivo. Candidates could be the small RNA binding pro-tein FbpA in B. subtilis,which is required for the regulation of thelutABC operon by the sRNA FsrA (Smaldone et al. 2012; reviewedin Brantl and Bruckner, 2014) or E. coli CsrA, which was recentlyfound to bind, additionally to Hfq, to the sRNA McaS (Jørgensenet al. 2013).

The majority of type I antitoxins are transcribed constitu-tively, except SprA1AS from S. aureus that peaks at mid-log phase(Sayed, Jousselin and Felden 2011) and the chromosomal En. fae-calis fst homolog, encoded by parEF0409 (Weaver 2014). By con-trast, the majority of chromosome-encoded sRNAs are only ex-pressed under certain environmental or stress conditions (e.g.OxyS under oxidative stress, Altuvia et al. 1997). Consequently,their amount varies significantly. It is in the nature of type Iantitoxins to be short-lived, with half-lives between 30 s (Sok,Gerdes and Wagner 2007), 3–4 min (SR4, Jahn et al. 2012) and8–9 min (RNAII, Weaver et al. 2004). By contrast, half-lives ofchromomosome-encoded sRNAs display much broader varia-tions from 2 to 32 min (Vogel et al. 2003) or even 55 min (Preiset al. 2009).

Many trans-encoded sRNAs have more than one target (e.g.S. aureus RNAIII or E. coli RybB, reviewed in Brantl 2009), and fre-quently use different mechanisms to regulate them. Some tar-gets are repressed, while others are activated by the same sRNA(e.g. S. aureus RNAIII). In some cases, only translation is affected,in others both translation and RNA degradation (reviewed inBrantl 2009). It is tempting to speculate if some of the type I an-titoxins might, as they are usually higher expressed than theirtoxin mRNAs, moonlight to regulate additional target RNAs.

A comparison between type III antitoxins and sRNAs thatact by protein binding like E. coli CsrB/C, Ps. fluorescens RsmX/Y(reviewed in Brantl 2009) or Ps. aeruginosa CrcZ reveals differ-ent strategies. Whereas type III antitoxins form small pseu-doknots composed of 35–36 nt that sequester their cognatetoxins in trimeric complexes using a number of RNA–proteininteractions, CsrB and CsrC present minimal 5′ GGA bindingmotifs—often embedded in an ANGGA context—in the loops ofnumerous (9 to 22) short stem-loop regions that bind small (61aa) translational regulator proteins like CsrA or RsmE. Both CsrAand Crc proteins are homodimers of intertwined β-strands withprotruding α helices that use specific residues in the β-strands

to bind two RNA sites simultaneously (CsrA, Schubert et al. 2007;Crc, Wei et al. 2013). However, in both cases, the proteinsare bound in a sequence-specific and not a structure-specificmanner.

TYPE I AND TYPE III TOXINS AND THEIRMODE OF ACTION

The classical characterized type I toxins are—with the excep-tion of SymE and RalR—very small hydrophobic peptides of <60aa (summarized in Table 2). All currently known hydrophobictype I toxins have an identical predicted or experimentally con-firmed secondary structure containing an α-helical transmem-brane domain and are supposed to be localized in the mem-brane (e.g. Fozo et al. 2008; Gobl et al. 2010; Sayed et al. 2012).It is proposed that these toxins can—similar to phage holins—interact with each other to create pore-like structures that de-stroy the membrane potential and, consequently, inhibit ATPsynthesis (e.g. Fozo et al. 2008; Unoson and Wagner 2008).Nevertheless, information about their mode of action and theircellular targets is limited, as in most systems, deletion of theantitoxin does not result in phenotypic alterations making itdifficult to analyse toxin effects at cellular levels. Much currentinformation is based on overexpression experiments, in whichmost of the toxins cause cell death. Several studies have demon-strated that overexpression of hok, relF, srnB, pndA, fst, ibsC, shoB,tisB and dinQ leads to the destruction of membrane potentialor membrane itself (Gerdes et al. 1986; Ono et al. 1986; Weaveret al. 2003; Fozo et al. 2008; Unoson and Wagner 2008; Weel-Sneve et al. 2013). Due to the hydrophobic nature of the toxinpeptides and their similarity to both phage holins (Wang, Smithand Young 2000) and antimicrobial peptides (Henriques, Meloand Castanho 2006), these effects on the membrane are not sur-prising (Fozo, Hemmand Storz 2008). However, interactionswithother cellular targets cannot be excluded.

The hok/sokmodule on E. coli plasmid R1 (Gerdes, Larsen andMolin 1985) acts as PSK system (Gerdes, Rasmussen and Molin1986b). By contrast, most of the Hok-like toxins encoded in thechromosome of E. coli K12 seem to be inactive (Pedersen andGerdes 1999). It has been demonstrated that overexpression ofhok leads to formation of ‘ghost’ cells, which are characterizedby condensed cell poles and a centrally located clearing result-ing in cell death (Gerdes et al. 1986). For this reason, Hok pro-teins in general are supposed to kill cells by causing irreversibledamage to the cell membrane. In accordance with this, hok over-expression leads to the collapse of membrane potential, respi-ratory arrest, efflux of small molecules including Mg2+ as wellas ATP and influx of extracellular molecules as e.g. ONPG (sum-marized in Gerdes et al. 1997). The overproduction of the hok ho-molog, RelF from the E. coli relB operon (Bech et al. 1985), resultsin the same phenomena: collapse of membrane potential, arrestof respiration, change in cell morphology and cell death (Gerdeset al. 1986).

Overexpression of the pAD1-encoded Fst toxin from En. fea-calis (Weaver et al. 1996) is not only toxic in its original host butalso in several bacilli, including B. subtilis and in S. aureus (Weaveret al. 2003, 2009; Patel and Weaver 2006). Toxicity can also be de-tected in E. coli, when the 5′ stem-loop of fst mRNA is deleted(Shokeen et al. 2008). These results indicate that Fst does nothave a host-specific target. The primary effect observed in allfour species is the condensation of the nucleoid. Whereas inboth B. subtilis and E. coli, an elongation of cells could be detecteddue to the inhibition of cell division at an early time point, the

422 FEMS Microbiology Reviews, 2015, Vol. 39, No. 3

Table 2. Overview of type I and type III toxins.

Toxin Peptide length (aa) Organism plasmid Localization Effects

Hok∗ 52 E. coli plasmid R1 Inner membrane Loss of membrane potential → arrest ofrespiration, efflux/influx of small molecules,ghost cells

RelF∗ 51 E. coli Inner membrane Loss of membrane potential → arrest ofrespiration, ghost cells

PndA∗ 50 E. coli plasmid R483 Inner membrane? Increase of membrane permeabilitySrnB∗ 68 E. coli plasmid F Inner membrane? Increase membrane permeabilityIbsC 18-19 E. coli Inner membrane? Loss of membrane potentialShoB 26 E. coli Inner membrane? Loss of membrane potentialLdrD 35 E. coli Inner membrane? Nucleoid condensation, inhibition of

translation, loss of cell viability, physiologicalalteration (purin metabolism, cellular cAMPlevels)

TisB 29 E. coli Inner membrane Loss of membrane potential/disruption bypore → decrease in intracellular ATP levels→ inhibition of macromolecular synthesis;single copy: persister formation

DinQ 27 E. coli Inner membrane Loss of membrane potential → decrease inintracellular ATP levels, modulatesrecombination and nucleoid compaction

SymE 113 E. coli Cytoplasm? Degradation of selected RNAsRalR 64 E. coli Cytoplasm? Non-specific endoDNaseFst 33 En. faecalis plasmid pAD1 Membrane Cell anomalies: nucleoid condensation,

division and segregation defects, secondaryeffect: increase of cell permeability→ inhibition of macromolecular synthesis

Fst-Sm 32 St. mutans Membrane? Mild overexpression of Fst-Sm/srSm system:decrease of persister levels

SprA1 30–33 S. aureus Membrane? Antimicrobial activity against Gram-negativeand Gram-positive bacteria;cytolytic for human cells

SprF1 31, 44 S. aureus Membrane Two peptides are secreted in small amounts;antimicrobial activity against Gram-negativeand Gram-positive bacteria;cytolytic for human cells

ToxN 171 P. atrosepticum plasmid pECA1039 Cytoplasm? EndoRNase activity; abi systemAbiQ 172 L. lactis plasmid pSRQ900 Cytoplasm? EndoRNase activity; abi system

∗hok gene family; ?, suggested, but not experimentally corroborated.

division septa in S. aureus can be formed, but not fully com-pleted. Instead, Fst overproduction in En. faecalis leads to mis-placed septa resulting in daughter cells with little or withoutDNA, indicating a segregation defect. The additionally observedincrease ofmembrane permeability turned out to be a secondaryeffect, probably caused by ongoing division and/or segregationdefects (Patel and Weaver 2006). Recently, it was proposed thatFst induces upregulation of energy-requiring membrane trans-porters (pumps), as an rpoC mutant that was resistant againstFst did not induce these pumps, and chemical inhibition ofthe pumps rescued the cells from the effects of Fst (Brinkmanet al. 2013).

Overproduction of the chromosomally encoded E. coliLdrD, an Fst-homolog from a Gram-negative bacterium (Fozoet al. 2010), evokes a rapid chromosomal condensation as well,and inhibits global translation (Kawano et al. 2002; Kawano 2012).Furthermore, microarray analysis revealed that overexpressionresults in physiological alterations in the cell, including upregu-lation of purine metabolism and decrease of intracellular cAMPlevels. Recently, it was shown that Ldrmay inhibit ATP synthesis,possibly due to its localization in the cell membrane (Yamaguchiet al. 2014).

Transcription of tisB, dinQ and symE RNA is induced by SOSstress response indicating that the toxin proteins might playa role under these stress conditions. TisB targets the innermembrane of E. coli, and its overexpression leads to the de-crease of membrane potential resulting in reduced intracellu-lar ATP levels, shut down of macromolecular synthesis includ-ing replication, transcription and translation and, consequently,in cell death in a fraction of the population (Unoson andWagner 2008; Gurnev et al. 2012; Fig. 4A). Synthetic TisBmonomers bind rapidly to membranes, and antiparallel dimerswere postulated to assemble as an electrostatic charge zippervia a ladder of salt bridges that enables protons to cross thehydrophobic membrane (Steinbrecher et al. 2012). The depolar-ization of the membrane is probably caused by the formationof small, anion selective pores, which could make the mem-brane permeable for hydroxyl anions (Gurnev et al. 2012). How-ever, all these in vivo effects were only observed upon tisB over-expression, and are, therefore, not conclusive for the biologicalrelevance of TisB during SOS response. Particularly, cell killingis not observed with a single chromosomal tisB copy. Instead,the 1000-fold induced expression of tisB under SOS condi-tions results in dormant persisters that are highly tolerant to

Brantl and Jahn 423

Figure 4. Mode of action of toxins. A Gram-negative bacterium with inner andouter membrane, chromosomal DNA and, if present, plasmids are shown. Blackand white bars denote –35 and –10 boxes of promoters, respectively. The LexAbinding site is represented by a yellow box. Toxins are drawn in blue and an-

titoxin RNA in red. Toxin ORFs are shown as light blue bars. Grey boxes indi-cate SD sequences. (A) TisB (chromosomal type I TA system tisB/IstR1). TisB is anSOS-induced, hydrophobic toxic peptide that arrests cell growth by decreasingthe membrane potential, which results in decreasing ATP levels and shut-off of

macromolecule synthesis. (B) SymE (chromosomal type I TA system symE/SymR).SymE is activated by the SOS response system (LexA) and acts as endoribonu-clease that degrades mRNAs and non-coding RNAs, but not SymR. The SymE

toxin itself can be degraded by Lon protease. (C) ToxN (plasmid-encoded typeIII TA system toxN/ToxI). Transcriptional read through from the toxI promoterinto toxN allows the synthesis of toxIN mRNA and, consequently, the expres-sion of the toxin ToxN. ToxN acts as a ribonuclease directed against both gen-

eral cellular targets and its own antitoxin ToxI. (D) RalR (chromosomal type ITA system ralR/RalA). The expression of ralR is induced under unknown condi-tions. RalR functions as a non-specific endodeoxyribonuclease that cleaves bothmethylated and unmethylated DNA resulting in cell death.

unrelated antibiotics (Unoson and Wagner 2008; Dorr, Vulic andLewis 2010). These effects can be brought into line with the ef-fects of TisB on membrane integrity and reduced intracellularATP levels (Unoson and Wagner 2008), since the growth arrestwould explain the antibiotic-resistant phenotype of persisters(Unoson and Wagner 2008; Dorr, Vulic and Lewis 2010).

The recently identified chromosomally encoded dinQ/Agrmodule of E. coli shares many features with the tisB/istR1 lo-cus. DinQ is also localized in the inner membrane and itsoverexpression leads to membrane depolarization and, conse-quently, to the decrease of intracellular ATP levels (Weel-Sneveet al. 2013). Furthermore, it has been shown that it affectsmembrane-dependent activities like nucleoid compaction andrecombination.

A number of type I toxins evoke other effects in the cell asidefrom disruption of membrane potential or membrane damage:The SOS-induced SymE from E. coliwith a length of 113 aa is nothydrophobic, but is an RNase suggested to play a role in recy-cling of damaged RNAs under SOS conditions (Kawano, Aravindand Storz 2007). This protein does not show functional homol-ogy to other type I toxin proteins, but instead has homology tothe AbrB-fold superfamily proteins, which act as transcriptionalfactors and antitoxins in different type II TA systems like mazEF(Kawano, Aravind and Storz 2007). On the other hand, SymEshares properties with type II toxins such as MazF that cleavesRNAs ribosome independent, since it enhances the degradationof bothmRNAs and non-coding RNAs but not its antitoxin SymR(Kawano, Aravind and Storz 2007; Fig 4B).

This mode of action was also found in type III toxins(Blower et al. 2011; Samson et al. 2013). The two type III toxinsToxN and AbiQ known so far have sequence-specific, ribosome-independent endoRNase activities which, on the one hand, pro-cess their antitoxins from a precursor and, on the other hand,degrade phage RNA acting as abi (abortive phage infection) sys-tems (Blower et al. 2011; Samson et al. 2013). In the case ofToxN, it was demonstrated that it does not only protect its hostagainst multiple phages but also cleaves within host RNA (Shortet al. 2013; Fig. 4C). In both ToxN and AbiQ, a serine residuepresent at position 51 or 52, respectively, is involved in cataly-sis.

The recently discovered type I toxin RalR of the cryptic E. coliprophage rac is the first known toxin that acts as a non-specificendoDNase (Guo et al. 2014; Fig. 4D). It is probably located in thecytoplasm and has been shown to cleave methylated and un-methylated DNA equally well.

For years it was speculated about extracellular effectsof type I toxins, similar to those of antimicrobial peptides.S. aureus toxin SprA1 that belongs to the fst family (Fozoet al. 2010) was reported to behave not only as a type I toxin instaphylococci but also as a toxin that kills bacterial rivals in amixed-cell population and as a cytolysin that lyses human ery-throcytes (Sayed, Jousselin and Felden 2011). This would indi-cate that it might outcompetes other S. aureus strains or otherspecies like E. coli or B. subtilis and that it is—upon infection—released from themembrane and could act on human host cells.All these properties would be contradictory to a toxin of a bonafide TA system, which, by definition, is only directed against itsown cell. A previous study with a chemically synthesized En. fae-calis Fst suggested that it can neither lyse human erythrocytesnor could it act against B. subtilis or E. coli as potential bacterialrivals (Gobl et al. 2010). Likewise, it was demonstrated that theexogenous application of Hok does not significantly kill otherbacterial strains (Pecota et al. 2003). A comparison of the aasequences of Fst and SprA1 reveals only structural, but no

424 FEMS Microbiology Reviews, 2015, Vol. 39, No. 3

sequence similarity. In particular, SprA1 contains a central cys-teine residue that is lacking in Fst. NMR studies revealed thatthis cysteine is located at a flexible hinge, and the authors hy-pothesize that it might trigger dimerization (Sayed et al. 2012).Therefore, it cannot be excluded that differences in the primarysequences might be responsible for the diverging properties ofboth toxins.

It will be exciting to see which other modes of action of typeI toxins will be identified in the future. The most common tar-get of these toxins is the membrane, with two exceptions: SymEand RalR. In contrast to type I toxins, type II toxins typically actas endoRNases, either as free enzymes (Jørgensen et al. 2009) orribosome-associated ones (Neubauer et al. 2009). Other mech-anisms elucidated for type II toxins are inhibition of DNA gy-rase, of protease or phosphotransferase activity (Jiang et al. 2002;Meinhart et al. 2003; Yamamoto et al. 2009).

BIOLOGICAL ROLES OF TYPE I AND TYPE IIITA SYSTEMS

The role of plasmid-encoded type I TA systems is evident: theyare PSK systems, i.e. in plasmid-free cells, the short-lived an-titoxin is rapidly degraded and thus, cannot prevent toxin ex-pression anymore, whereas the stable toxinmRNA is translated,eventually causing cell death. This scenario has been verifiedfirst for E. coli hok/Sok (reviewed in Gerdes andWagner 2007) andlater for En. faecalis fst/RNAII (reviewed in Weaver 2012, 2014).

By contrast, the function of chromosome-encoded type I TAsystems, among them homologs of hok/Sok (Gerdes et al. 1997;Pedersen and Gerdes 1999) or fst/RNAII (Weaver et al. 2009) islargely unknown. For chromosome-encoded type II TA systems,a role in persister formation has been well established: whereasindividual deletions of the 10 known type II TA systems compris-ing mRNase toxins from the E. coli genome had only minor ef-fects, the simultaneous deletion of all 10 currently known type IIloci dramatically reduced persistence (Maisonneuve et al. 2011).Persister cells are a subset of a bacterial population that hasstochastically entered a dormant state and thus became re-fractory to the action of antibiotics, whereas their isogenic,rapidly growing siblings remain sensitive. After long periods oftime, persisters may be resuscitated and resume growth, re-establishing the population of antibiotics-susceptible cells. Themechanism of persister formation is still unknown. So far, onlyfor one type I TA system, a role in persister formation has clearlydemonstrated: in E. coli tisB/IstR1, SOS-dependent induction oftisB expression increased the level of persister cells resistantto the antibiotic ciprofloxacin—a gyrase inhibitor—significantly,whereas deletion of the entire tisB/IstR locus caused a sharp de-crease of persisters (Dorr, Lewis and Vulic 2009 and Dorr, Vulicand Lewis 2010). This finding can be reconciled with the effectof TisB on membrane integrity (see above).

Recently, fst-Sm/SrSm, a chromosomal fst/RNAII homolog inStreptococcus mutans (Tables 1 and 2) has been also linked to per-sister formation (Koyanagi and Levesque 2013): however, mildoverproduction of the entire fst-Sm/srSm locus from amulti-copyplasmid surprisingly decreased the levels of persister cells tol-erant to cell-wall synthesis inhibitors like vancomycin. Perhaps,placing toxin and antitoxin genes out of their chromosomal con-text on a plasmid resulted in an imbalance between toxin andantitoxin levels.

Although E. coli dinQ/AgrB is similar to tisB/IstR1, persister for-mation has not been studied, but this TA system was found tobe important for chromosome stability: a �agrB strain showed

2-fold elevated DinQ levels and displayed a 400-fold reduced re-combination frequency, suggesting that the toxin interfereswithhomologous recombination (Weel-Sneve et al. 2013). No effect ongeneral DNA uptake during conjugation was found.

The recently discovered ralR/RalA TA system from E. coli wasalso shown to be beneficial for resistance against a cell-wall in-hibiting antibiotic, fosfomycin (Guo et al. 2014). However, persis-ter formation has not been investigated so far.

The RNA-cleaving E. coli toxin SymEhas been suggested to re-cycle damaged RNAs produced under SOS stress conditions or toprevent infection with RNA phages (reviewed in Kawano 2012).

On the contrary, many B. subtilis toxin genes (txpA, yonT,bsrG, bsrE, bsrH) are located on prophages and prophage-likeregions and were proposed to be required for maintenance ofthese elements, reminiscent of the role of plasmid-encoded PSKsystems. For instance, txpA/RatA was suggested to sustain theskin element, which is excised by a DNA rearrangement dur-ing sporulation from the B. subtilis genome (reviewed in Durandet al. 2012). This excision ensures reconstitution of a functionalsigK gene encoding a transcription factor that controls mother-cell-specific gene expression late in sporulation. All TA modulesin B. subtilis are under control of σA promoters allowing rapidtoxin expression under defined conditions. Therefore, the iden-tification of transcriptional activators thatmight perturb the bal-ance between toxin and antitoxin expression could provide uswith clues about the function of the individual toxins. Althoughputative resD sites have been predicted upstream of bsrG, bsrEand bsrH, it remains to be seen if increased expression underoxygen-limiting conditions as suggested (Nicolas et al. 2012) isindeed due to ResD binding to its cognate site.

Some TA systems are induced under certain environmentalconditions like metabolic stress (e.g. glucose starvation) or oxy-gen stress, and the function of the corresponding toxins could beto cause bacteriostasis to limit nutrient consumption or oxygenexhaustion (see Durand et al. 2012), i.e. these systems would beinvolved in metabolic or stress adaptation. In accordance withthis hypothesis, txpA mRNA expression seems to be phosphatedependent, and txpA, bsrG, bsrH and yonT RNA levels were pro-posed to increase when glucose is exhausted, linking these tox-ins to themetabolic state of the cell (based onNicolas et al. 2012).Interestingly, bsrG could play a role in response to tempera-ture changes, as its half-life is temperature dependent, and themRNA is rapidly degraded upon temperature shock at 48 or 55◦C(Jahn et al. 2012, see above).

The two known type III TA systems, toxN/ToxI andabiQ/AntiQ, are located on plasmids, but they are abi mod-ules, i.e. they are not required for prophage maintenance, butfor defence against phage invasion (e.g. Samson et al. 2013). Therole of chromosome-encoded type III systems is as elusive asthat of type I systems (Blower et al. 2012).

PERSPECTIVES/CONCLUSION

Despite all data accumulated over the past years, our knowledgeon type I TA systems is still limited. It seems that all bacteria thatdo not live as intracellular parasites encode a variety of TAmod-ules in their genomes. Notwithstanding the variation betweenthe type I systems discovered and studied so far, they seem toconstitute a conserved familywith similar regulatory properties.Although some of these systems have taught us new modes bywhich a small regulatory RNA can inhibit expression of its tar-get mRNA (e.g. IstR1—blockage of an RSS), we can expect thatother still unprecedented regulatory mechanisms will come to

Brantl and Jahn 425

our attention in which both RNA antitoxins and other sRNAs areinvolved.

Furthermore, it is tempting to speculate if, after the discoveryof two classes of RNA antitoxins that either bind toxin mRNA(type I) or toxin peptide directly (type III), another class of RNAantitoxinsmight exist that is able to bind both the correspondingtoxin mRNA and the toxin peptide.

As it is possible that stochastic expression of toxins, and con-sequently, effects on cells of only a subpopulation, might ex-plain why such modules are broadly distributed in genomes ofall bacterial species that do not live as parasites, future researchwill have to focus on two aspects. Firstly, single-cell techniquesand reporter gene fusions should be employed to detect en-dogenously expressed toxin proteins under various conditions.Thiswould allow for a better insight into population heterogene-ity. Secondly, the investigation of differences between wild-typestrains and strains deleted for all known type I TA systems un-der stress conditions like phage infection, depletion of nutrients,harmful environments as presence of antibiotics etc. would helpto elucidate the still enigmatic biological role of these numeroussystems.

ACKNOWLEDGEMENTS

The authors are grateful to Henrik Strahl and Leendert Hamoen,CBCB Newcastle for collaboration on the BsrG toxin and toChristinMeißner and Christiane Ring, AG Bakteriengenetik Jena,for unpublished data on the new TA system bsrE/SR5.

FUNDING

Thisworkwas supported by grant BR1552/7-2 from the DeutscheForschungsgemeinschaft to S. B.

Conflict of interest. None declared.

REFERENCES

Andre G, Even S, Putzer H, et al. S-box and T-box riboswitchesand antisense RNA control a sulphur metabolic operon ofClostridium acetobutylicum. Nucleic Acids Res 2008;36:5955–69.

Altuvia S,Weinstein-Fischer D, ZhangA, et al.A small stable RNAinduced by oxidative stress: role as a pleotropic regulator andantimutator. Cell 1997;90:43–53.

Asano K, Mizobuchi K. Copy number control of Inc1α plas-mid ColIb-P9 by competition between pseudoknot formationand antisense RNA binding at a specific RNA site. EMBO J1998;17:5201–13.

Bech FW, Jørgensen ST, Diderichsen B, et al. Sequence of the relBtranscription unit from Escherichia coli and identification ofthe relB gene. EMBO J 1985;4:1059–66.

Blomberg P, Wagner EGH, Nordstrom K. Control of replication ofplasmid R1: the duplex between the antisense RNA, CopA,and its target, CopT, is processed specifically in vivo and invitro by RNase III. EMBO J 1990;9:23421–40.

Blower TR, Fineran PC, Johnson MJ, et al. Mutagenesis and func-tional characterization of the RNA and protein componentsof the toxIN abortive infection and toxin-antitoxin locus ofErwinia. J Bacteriol 2009;191:6029–39.

Blower TR, Pei XY, Short FL, et al. A processed noncoding RNAregulates an altruistic bacterial antiviral system. Nat StructMol Biol 2011;18:185–90.

Blower TR, Short FL, Rao F, et al. Identification and classifica-tion of bacterial type III toxin-antitoxin systems encoded

in chromosomal and plasmid genomes. Nucleic Acids Res2012;40:6158–73.

Brantl S. Regulatory mechanisms employed by cis-encoded an-tisense RNAs. Curr Opin Microbiol 2007;10:102–9.

Brantl S. Bacterial chromosome-encoded regulatory RNAs. Fu-ture Microbiol 2009;4:85–103.

Brantl S. Acting antisense: plasmid- and chromosome-encodedsRNAs from Gram-positive bacteria. Future Microbiol2012a;7:853–71.

Brantl S. Bacterial type I toxin-antitoxin systems. RNA Biol2012b;9:1488–90.

Brantl S. Plasmid replication control by antisense RNAs.MicrobiolSpectrum 2014;2:PLAS-0001-2013.

Brantl S, Bruckner R. Small regulatory RNAs from low-GC Gram-positive bacteria. RNA Biol 2014;11:443–56.

Brinkman CL, Bumgarner R, Kittichotirat W, et al. Characteriza-tion of the effects of an rpoC mutation that confers resis-tance to the Fst peptide toxin-antitoxin system. J Bacteriol2013;195:156–66.

Case CC, Simons EL, Simons RW. The IS10 transposase mRNAis destabilized during antisense RNA control. EMBO J1990;9:1259–66.

Darfeuille F, Unoson C, Vogel J, et al. An antisense RNA inhibitstranslation by competing with standby ribosomes. Mol Cell2007;26:381–92.

Domka J, Lee J,Wood TK. Temporal gene expression in E. coliK-12biofilms. Environ Microbiol 2007;9:332–46.

Dorr T, Lewis K, Vulic M. SOS response induces persis-tence to fluoroquinolones in Escherichia coli. PLoS Genet2009;5:e1000760.

Dorr T, Vulic M, Lewis K. Ciprofloxacin causes persister forma-tion by inducing the TisB toxin in Escherichia coli. PLoS Biol2010;8:e1000317.

Durand S, Gilet L, Condon C. The essential function of B. sub-tilis RNase III is to silence foreign toxic genes. PLoS Genet2012;8:e1003181.

Durand S, Jahn N, Condon C, et al. Type I toxin-antitoxin systemsin Bacillus subtilis. RNA Biol 2012;9:1491–7.

Even S, Pellegrini O, Zig L, et al. Ribonucleases J1 and J2: two novelendoribonucleases in B. subtilis with functional homology toE. coli RNase E. Nucleic Acids Res 2005;33:2141–52.

Fineran PC, Blower TR, Foulds IJ, et al. The phage abortive in-fection system, ToxIN, functions as a protein-RNA toxin-antitoxin pair. P Natl Acad Sci USA 2009;106:894–9.

Fozo EM. New type I toxin-antitoxin families from ‘wild’ and lab-oratory strains of E. coli. Ibs-Sib, ShoB-OhsC and Zor-Orz. RNABiol 2012;9:1504–12.

Fozo EM, Hemm MR, Storz G. Small toxic proteins and theantisense RNAs that repress them. Microbiol Mol Biol R2008;72:579–89.

Fozo EM, Kawano M, Fontaine F, et al. Repression of small toxicprotein synthesis by the Sib and OhsC small RNAs. Mol Mi-crobiol 2008;70:1076–93.

Fozo EM, Makarova KS, Shabalina SA, et al. Abundance of type Itoxin-antitoxin systems in bacteria: searches for new can-didates and discovery of novel families. Nucleic Acids Res2010;38:3743–59.

Franch T, Gerdes K. U-turns and regulatory RNAs. Curr Opin Mi-crobiol 2000;3:159–64.

Franch T, Petersen M, Wagner EGH, et al. Antisense RNA regula-tion in prokaryotes: rapid RNA/RNA interaction facilitated bya general U-turn loop structure. J Mol Biol 1999;294:1115–25.

Gerdes K, Bech FW, Jørgensen ST, et al. Mechanism of postseg-regational killing by the hok gene product of the parB system

426 FEMS Microbiology Reviews, 2015, Vol. 39, No. 3

of plasmid R1 and its homology with the relF gene product ofthe E. coli relB operon. EMBO J 1986;5:2023–9.

Gerdes K, Gultyaev AP, Franch T, et al. Antisense RNA-regulatedprogrammed cell death. Annu Rev Genet 1997;31:1–31.

Gerdes K, Larsen JE, Molin S. Stable inheritance of plasmid R1requires two different loci. J Bacteriol 1985;161:292–8.

Gerdes K, Nielsen A, Thorsted P, et al. Mechanism of killer geneactivation. Antisense RNA-dependent RNase III cleavage en-sures rapid turn-over of the stable hok, srnB and pndA effectormessenger RNAs. J Mol Biol 1992;226:637–49.

Gerdes K, Rasmussen PB, Molin S. Unique type of plasmid main-tenance function: postsegregational killing of plasmid-freecells. P Natl Acad Sci USA 1986;83:3116–20.

Gerdes K, Wagner EGH. RNA antitoxins. Curr Opin Microbiol2007;10:117–24.

Gobl C, Kosol S, Stockner T, et al. Solution structure and mem-brane binding of the toxin Fst of the par addiction module.Biochemistry 2010;49:6567–75.

Greenfield TJ, Ehli E, Kirshenmann T, et al. The antisense RNA ofthe par locus of pAD1 regulates the expression of a 33-amino-acid toxic peptide by an unusual mechanism. Mol Microbiol2000;37:652–60.

Greenfield TJ, Franch T, Gerdes K, et al.Antisense RNA regulationof the par post-segregational killing system: structural anal-ysis and mechanism of binding of the antisense RNA, RNAIIand its target, RNAI. Mol Microbiol 2001;42:527–37.

Guo Y, Quiroga C, Chen Q, et al. RalR (a DNase) and RalA (a smallRNA) form a type I toxin-antitoxin system in Escherichia coli.Nucleic Acids Res 2014;42:6448–62.

Gurnev PA, Ortenberg R, Dorr T, et al. Persister-promoting bacte-rial toxin TisB produces anion-selective pores in planar lipidbilayers. FEBS Lett 2012;586:2529–34.

Han K, Kim K-sun, Bak G, et al. Recognition and discriminationof target mRNAs by Sib RNAs, a cis-encoded sRNA family.Nucleic Acids Res 2010;38:5851–66.

Heidrich N, Brantl S. Antisense RNA-mediated transcriptionalattenuation in plasmid pIP501: the simultaneous interac-tion between two complementary loop pairs is requiredfor efficient inhibition by the antisense RNA. Microbiology2007;153:420–7.

Heidrich N, Moll I, Brantl S. In vitro analysis of the interaction be-tween the small RNA SR1 and its primary target ahrCmRNA.Nucleic Acids Res 2007;35:4331–46.

Henriques ST,MeloMN, CastanhoMA. Cell-penetrating peptidesand antimicrobial peptides: How different are they? BiochemJ 2006;399:1–7.

Jahn N, Brantl S. One antitoxin—two functions: SR4 con-trols toxin mRNA decay and translation. Nucleic Acids Res2013;41:9870–80.

Jahn N, Preis P, Wiedemann C, et al. BsrG/SR4 from Bacil-lus subtilis—the first temperature-dependent type I toxin-antitoxin system. Mol Microbiol 2012;83:579–98.

Jiang Y, Pogliano J, Helinski DR, et al. ParE toxin encoded by thebroad-host-range plasmid RK2 is an inhibitor of Escherichiacoli gyrase. Mol Microbiol 2002;44:971–9.

Jørgensen MG, Pandey DP, Jaskolska M, et al. HicA of Escherichiacoli defines a novel family of translation-independentmRNA interferases in bacteria and archaea. J Bacteriol2009;191:1191–9.

Jørgensen MG, Thomason MK, Havelund J, et al. Dual function ofthe McaS small RNA in controlling biofilm formation. GeneDev 2013;27:1132–45.

Kawano M. Divergently overlapping cis-encoded antisense RNAregulating toxin-antitoxin systems from E. coli: hok/sok,ldr/rdl, symE/symR. RNA Biol 2012;9:1520–7.

Kawano M, Aravind L, Storz G. An antisense RNA controls syn-thesis of an SOS-induced toxin evolved from an antitoxin.Mol Microbiol 2007;64:738–54.

Kawano M, Oshima T, Kasai H, et al. Molecular characteriza-tion of long direct repeat (LDR) sequences expressing a stablemRNA encoding for a 35-amino-acid cell-killing peptide anda cis-encoded small antisense RNA in Escherichia coli. Mol Mi-crobiol 2002;45:333–49.

Kawano M, Reynolds AA, Miranda-Rios J, et al. Detection of5′- and 3′-UTR-derived small RNAs and cis-encoded an-tisense RNAs in Escherichia coli. Nucleic Acids Res 2005;33:1040–50.

Koyanagi S, Levesque CM. Characterization of a Streptococcusmutans intergenic region containing a small toxin peptideand its cis-encoded antisense small RNA antitoxin. PLoS One2013;8:e54291.

Krinke L, Wulff DL. OOP RNA, produced from multicopy plas-mids, inhibits λ cII gene expression through an RNase III-dependent mechanism. Gene Dev 1987;1:1005–13.

Krinke L, Wulff DL. RNase III-dependent hydrolysis of λ cII-Ogene mRNA mediated by λ OOP antisense RNA. Gene Dev1990;4:2223–33.

Kwong SM, Skurray RA, Firth N. Replication control of staphy-lococcal multiresistance plasmid pSK41: an antisense RNAmediates dual-level regulation of Rep expression. J Bacteriol2006;188:4404–12.

Maisonneuve E, Shakespeare LJ, Jørgensen MG, et al. Bacte-rial persistence by RNA endonucleases. P Natl Acad Sci USA2011;108:13206–11.

Masse E, Escorcia FE, Gottesman S. Coupled degradation of asmall regulatory RNA and its mRNA targets in Escherichia coli.Gene Dev 2003;17:2374–83.

Masuda H, Tan Q, Awano N, et al. YeeU enhances the bundlingof cytoskeletal polymers of MreB and FtsZ, antagonizingthe CbtA (YeeV) toxicity in Escherichia coli. Mol Microbiol2012;84:979–89.

Meinhart A, Alonso JC, Strater N, et al. Crystal structure of theplasmid maintenance system ε/ζ : functional mechanism oftoxin ζ and inactivation by ε2ζ2 complex formation. P NatlAcad Sci USA 2003;100:1661–6.

Mikkelsen ND, Gerdes K. Sok antisense RNA from plasmid R1 isfunctionally inactivated by RNase E and polyadenylated bypoly(A) polymerase. Mol Microbiol 1997;26:311–20.

Neubauer C, Gao YG, Andersen KR, et al. The structural basis formRNA recognition and cleavage by the ribosome-dependentendonuclease RelE. Cell 2009;139:1084–95.

Nicolas P, Mader U, Dervyn E, et al. Condition-dependent tran-scriptome reveals high-level regulatory architecture in Bacil-lus subtilis. Science 2012;335:1103–6.

Opdyke JA, Fozo EM, Hemm MR, et al. RNase III participates inGadY-dependent cleavage of the gadX-gadWmRNA. J Mol Biol2011;406:29–43.

Ono T, Akimoto S, Ono K, et al. Plasmid genes increase mem-brane permeability in Escherichia coli. Biochim Biophys Acta1986;867:81–8.

Patel S, Weaver KE. Addiction toxin Fst has unique effects onchromosome segregation and cell division in Enterococcus fea-calis and Bacillus subtilis. J Bacteriol 2006;188:5374–84.

Pecota DC, Osapay G, Selsted ME, et al. Antimicrobial propertiesof the Escherichia coli R1 plasmid host killing peptide. J Biotech-nol 2003;100:1–12.

Pedersen K, Gerdes K. Multiple hok genes on the chromosome ofEscherichia coli. Mol Microbiol 1999;32:1090–102.

Pinel-Marie ML, Brielle R, Felden B. Dual toxic-peptide-codingStaphylococcus aureus RNA under antisense regulation

Brantl and Jahn 427

targets host cells and bacterial rivals unequally. Cell Rep2014;7:424–35.

Preis H, Eckart RA, Gudipati RK, et al. CodY activates tran-scription of a small RNA in Bacillus subtilis. J Bacteriol2009;191:5446–57.

Prevost K, Desnoyers G, Jacques JF, et al. Small RNA-inducedmRNA degradation achieved through both translation blockand activated cleavage. Gene Dev 2011;25:385–96.

Rocha EPC, Danchin A, Viari A. Translation in Bacillus subtilis:roles and trends of initiation and termination, insights froma genome analysis. Nucleic Acids Res 1999;27:3567–76.

Rotem E., Loinger A, Ronin I, et al. Regulation of phenotypic vari-ability by a threshold-based mechanism underlies bacterialpersistence. P Natl Acad Sci USA 2010;207:12541–6.

Samson JE, Spinelli S, Cambillau C, et al. Structure and activity ofAbiQ, a lactococcal endoribonuclease belonging to the typeIII toxin-antitoxin system. Mol Microbiol 2013;87:756–68.

Sayed N, Jousselin A, Felden B. A cis-antisense RNA acts in transin Staphylococcus aureus to control translation of a human cy-tolytic peptide. Nat Struct Mol Biol 2011;19:105–12.

Sayed N, Nonin-Lecomte S, Rety S, et al. Functional and struc-tural insights of a Staphylococcus aureus apoptotic-like mem-brane peptide from a toxin-antitoxin module. J Biol Chem2012;287:43353–463.

Schubert M, Lapouge K, Duss O, et al. Molecular basis of mes-senger RNA recognition by the specific bacterial repressingclamp RsmA/CsrA. Nat Struct Mol Biol 2007;14:807–13.

Silvaggi JM, Perkins JB, Losick R. Small untranslated RNA anti-toxin in Bacillus subtilis. J Bacteriol 2005;187:6641–50.

Shokeen S, Greenfield TJ, Ehli EA, et al. An intramolecular up-stream helix ensures the stability of a toxin-encoding RNAin Enterococcus faecalis. J Bacteriol 2009;191:1528–36.

Shokeen S, Patel S, Greenfield TJ, et al. Translational regulationby an intramolecular stem-loop is required for intermolec-ular RNA regulation of the par addiction module. J Bacteriol2008;190:6076–83.

Short FL, Pei XY, Blower TR, et al. Selectivity and self-assemblyin the control of a bacterial toxin by an antitoxic noncodingRNA pseudoknot. P Natl Acad Sci USA 2013;110:E241–9.

Smaldone GT, Antelmann H, Gaballa A, et al. The FsrA sRNA andFbpB protein mediate the iron-dependent induction of theBacillus subtilis lutABC iron-sulfur-containing oxidases. J Bac-teriol 2012;194:2586–93.

Steinbrecher T, Prock S, Reichert J, et al. Peptide-lipid interac-tions of the stress-response peptide TisB that induces bacte-rial persistence. Biophys J 2012;103:1460–9.

Thisted T, Gerdes K. Mechanism of post-segregational killing bythe hok/sok system of plasmid R1. Sok antisense RNA regu-lates hok gene expression indirectly through the overlappingmok gene. J Mol Biol 1992;223:41–54.

Thisted T, Sørensen NS, Gerdes K. Mechanism of post-segregational killing: secondary structure analysis of the en-tire hok mRNA from plasmid R1 suggests a fold-back struc-ture that prevents translation and antisense RNA binding. JMol Biol 1995;247:859–73.

Thisted T, Sørensen NS, Wagner EGH, et al. Mechanism of post-segregational killing: Sok antisense RNA interacts with HokmRNA via its 5′-end single-stranded leader and competeswith the 3′-end of Hok mRNA for binding to the mok transla-tional initiation region. EMBO J 1994;13:1960–8.

Unoson C, Wagner EGH. A small SOS-induced toxin is targetedagainst the inner membrane in Escherichia coli. Mol Microbiol2008;70:258–70.

Vogel J, Argaman L, Wagner EGH, et al. The small RNA IstR in-hibits synthesis of an SOS-induced toxic peptide. Curr Biol2004;14:2271–6.

Vogel J, Bartels V, Tank TH, et al. RNomics in Escherichia coli de-tects new sRNA species and indicates parallel transcriptionaloutput in bacteria. Nucleic Acids Res 2003;31:6435–43.

Wagner EGH, Brantl S. Kissing and stability in the control of plas-mid replication. Trends Biochem Sci 1998;23:451–4.

Wagner EGH, Unoson C. The toxin-antitoxin system tisB-istR1:expression, regulation, and biological role in persister phe-notypes. RNA Biol 2012;9:1513–9.

Wang IN, Smith DL, Young R. Holins: the protein clocks of bacte-riophage infections. Annu Rev Microbiol 2000;54:799–825.

Wang X, Lord DM, Cheng H-Y, et al. A new type V toxin-antitoxinsystem where mRNA for toxin GhoT is cleaved by antitoxinGhoS. Nat Chem Biol 2012;8:855–61.

Weaver KE. The par toxin-antitoxin system from Enterococcus fae-calis plasmid pAD1 and its chromosomal homologs. RNA Biol2012;9:1498–503.

Weaver KE. The type I toxin-antitoxin par locus fromEnterococcus faecalis plasmid pAD1: RNA regulationby both cis- and trans-acting elements. Plasmid 2014doi:10.1016/j.Plasmid.2014.10.001.

Weaver KE, Clewell DB. Construction of Enterococcus fae-calis pAD1 miniplasmids: identification of a minimalpheromone response regulatory region and evaluation ofa novel pheromone-dependent growth inhibition. Plasmid1989;22:106–19.

Weaver KE, Ehli EA, Nelson JS, et al. Antisense RNA regulation bystable complex formation in the Enterococcus faecalis plasmidpAD1 par addiction system. J Bacteriol 2004;186:6400–8.

Weaver KE, Jensen KD, Colwell A, et al. Functional analysis ofthe Enterococcus faecalis plasmid pAD1-encoded stability de-terminant par. Mol Microbiol 1996;20:53–63.

Weaver KE, Redy SG, Brinkman CL, et al. Identification and char-acterization of a family of toxin-antitoxin systems related tothe Enterococcus faecalis plasmid pAD1 par addiction module.Microbiology 2009;155:2930–40.

Weaver KE, Tritle DJ. Identification and characterization of an En-terococcus faecalis plasmid pAD1-encoded stability determi-nantwhich produces two small RNAmolecules necessary forits function. Plasmid 1994;32:168–81.

Weaver KE, Weaver DM, Wells CL, et al. Enterococcus faecalis plas-mid pAD1-encoded Fst toxin affects membrane permeabil-ity and alters cellular responses to lantibiotics. J Bacteriol2003;185:2169–77.

Weel-Sneve R, Kristiansen KI, Odsbu I, et al. Single transmem-brane peptide DinQ modulates membrane-dependent activ-ities. PLoS Genet 2013;8:e1003260.

Wei Y, Zhan H, Gao ZQ, et al. Structure analysis of the globalmetabolic regulator Crc from Pseudonomas aeruginosa. IUBMBLlife 2013;65:50–7

Wen J, Won D, Fozo EM. The ZorO-OrzO type I toxin-antitoxinlocus: repression by the OrzO antitoxin. Nucleic Acids Res2014;42:1930–46.

Yamaguchi Y, Tokunaga N, Inouye M, et al. Characterization ofLdrA (long direct repeat A) protein of Escherichia coli. J Mol Mi-crob Biotech 2014;24:91–7.

Yamamoto S, Kiyokawa K, Tanaka K, et al. Novel toxin-antitoxinsystem composed of serine protease and AAA-ATPase ho-mologues determines the high level of stability and incom-patibility of the tumor-inducing plasmid pTiC58. J Bacteriol2009;191:4656–66.