Type I Toxin-Antitoxin Systems: Regulating Toxin Expression via Shine-Dalgarno Sequence Sequestration and Small RNA Binding

ABSTRACT

Toxin-antitoxin (TA) systems are small genetic loci composed of two adjacent genes: a toxin and an antitoxin that prevents toxin action. Despite their wide distribution in bacterial genomes, the reasons for TA systems being on chromosomes remain enigmatic. In this review, we focus on type I TA systems, composed of a small antisense RNA that plays the role of an antitoxin to control the expression of its toxin counterpart. It does so by direct base-pairing to the toxin-encoding mRNA, thereby inhibiting its translation and/or promoting its degradation. However, in many cases, antitoxin binding is not sufficient to avoid toxicity. Several cis-encoded mRNA elements are also required for repression, acting to uncouple transcription and translation via the sequestration of the ribosome binding site. Therefore, both antisense RNA binding and compact mRNA folding are necessary to tightly control toxin synthesis and allow the presence of these toxin-encoding systems on bacterial chromosomes.

INTRODUCTION

Toxin-antitoxin (TA) systems are, by definition, simple genetic loci composed of two genes: a toxin and an antitoxin that counteracts either the toxin’s action or its expression. Usually, toxin synthesis leads to growth arrest or death of the bacterium that produces it by inhibiting essential cellular processes such as replication, translation, or cell division (1). Analogies can be made with other gene pairs with a similar bicistronic operon organization, such as bacteriocins, restriction-modification systems, and type VI secretion system effector/immunity proteins (2). For instance, they are often found within mobile genetic elements and pathogenicity islands. However, canonical TA systems generally act on the cell that produces them, and in contrast to bacteriocins, they are not involved in interbacterial competition. Thus, the presence of TA-encoded toxins in bacterial genomes strongly depends on the expression of their cognate antitoxins (3). TA systems can be classified into six different types depending on the nature and mode of action of the antitoxin (for a review, see reference 4). While the toxins are always proteins, antitoxins can be either proteins (types II, IV, V, and VI) or small RNAs (sRNAs, types I and III TA systems). Antitoxins can act by (i) sequestering the toxin (types II and III), (ii) inhibiting its expression (types I and V), or (iii) counterbalancing its activity (type IV). In this review, we will focus on type I TA systems, in which an antisense RNA plays the role of antitoxin to prevent the synthesis of its cognate toxin by directly base-pairing to the mRNA.

The hok/Sok and the par loci were the first type I TA systems discovered on the R1 and pAD1 plasmids, respectively, where they confer stability through postsegregational killing (PSK) (5–11). This complex mechanism relies on the differential stability between the stable, toxin-encoding mRNA and the labile antitoxin RNA. Consequently, upon loss of the TA locus (e.g., plasmid loss), antitoxin levels quickly drop while the stable toxin mRNA remains present, eventually leading to toxin synthesis and cell death. Thus, the toxin kills plasmid-free cells, whereas plasmid-containing cells are unaffected due to continuous de novo synthesis of the antitoxin. Although homologs of these TA systems were predicted on bacterial chromosomes (12–15), their role in this context remains enigmatic. It became even more intriguing with the later discovery of additional TA systems found exclusively on bacterial chromosomes (16–20), and with the identification of new type I TA systems during a systematic search for sRNA in bacteria (21). For instance, the Escherichia coli chromosome can contain up to 19 type I TA loci (22), but the function of only 3 of them has been reported (20, 23–25). Indeed, while toxicity could be easily assessed by toxin overexpression, single or even multiple TA deletions (in earlier studies) did not show any obvious phenotype (26, 27). The question of the function of TA systems, if any, when present on bacterial chromosomes is still elusive (Table 1).

TABLE 1.

Type I TA systems identified in bacteria

System	Bacteria	Present on:	Expression	Role	Copy number	Reference(s)
Gram Positive
fst/RNAII	Enterococcus faecalis	pAD1, VRSAp; SaPlbov2 pathogenicity island	Constitutive	PSK	1	5, 7, 81, 82
fst-like/RNAII	E. faecalis, Staphylococcus saprophyticus, Streptococcus mutans, and Lactobacillus casei	Chromosome	Unknown	Carbohydrate metabolism (?)	1	14, 15, 83
sprA1/SprA1ASa	Staphylococcus aureus N315	νΣαβ pathogenicity island	Constitutive	Pathogenicity (?)	1	62, 99
sprG1/SprF1b	S. aureus N315	MGEg ΦSa3 pathogenicity island	Constitutive	Unknown	1	67
txpA/RatA	Bacillus subtilis	Skin element	Constitutive	PSK-like of Skin element, sporulation (?)	Up to 6	63, 97
bsrG/SR4	B. subtilis	SPβ prophage region	ResD O2 limitation response	Unknown	1	64
bsrE/SR5	B. subtilis	Prophage-like region P6	Multistress responsive	Unknown	1	65, 100
bsrH/as-bsrH	B. subtilis	Skin element	ResD O2 limitation response	Unknown	1	19
yonT/SR6	B. subtilis	SPβ prophage region	Multistress responsive	Unknown	1	68
lpt/RNAII	Lactobacillus rhamnosus	Plasmid	Constitutive	PSK (?)	1	84
CDS2517.1/RCd8	Clostridium difficile	Chromosome CRISPR (clustered regularly interspaced short palindromic repeat) 12 region	Coregulated with CRISPR	Unknown	Up to 6	66
Gram Negative
hok/Sok	Escherichia coli	Plasmid R1	Constitutive	PSK	1	8, 10
hok/Sok	Enterobacteriac	Chromosome	Unknown	Persistence	Up to 15	12
hokB/SokB	E. coli K-12	Chromosome	ppGpp induced, Obg (?)	Persistence	1	23
srnB′/SrnC-RNA	E. coli	Plasmid F	Constitutive	PSK	1	11
pndA/PndB-RNA	E. coli	Plasmid R483	Constitutive	PSK	1	11
Ldr/Rdl	E. coli K-12, Salmonella, and Shigella	Chromosome	Unknown	Unknown	Up to 10	17
Ibs/Sib	Enterobacteriad and Helicobacter pylori	Chromosome	Unknown	Unknown	Up to 7	16, 21
shoB/OhsC	E. coli and Shigella	Chromosome	Unknown	Unknown	1	16, 79
zorO/OrzO	E. coli	Chromosome	Unknown	Unknown	2	38, 59
symE/ISymR	E. coli	Chromosome	SOS response (LexA)	RNA recycling upon SOS (?)	1	36
tisb/IstR	Enterobacteriae	Chromosome	SOS response (LexA)	Persistence	1	18, 58
dinQ/AgrAB	E. coli	Chromosome	SOS response (LexA)	Nucleoid compaction	1	20, 60
dinQ-like/AgrAB	Enterobacteriaf	Chromosome	Unknown	Unknown	1	60
ralR/RalA	E. coli	Cryptic prophage Rac	Unknown	Biofilm maintenance (?)	1	35
aapA1/IsoA1	H. pylori	Chromosome	Constitutive	Unknown	Up to 2	31
aapA/IsoA	Helicobacter and Campylobacter	Chromosome and plasmid	Unknown	Unknown	Up to 9	31

^a This system has been predicted to be a homolog of fst/RNAII, initially discovered in E. faecalis (13–15).

^b This system has been predicted to be a homolog of txpA/RatA, initially discovered in B. subtilis (22, 97).

^c Escherichia (4–15), Shigella (7–12), Enterobacter (6), Klebsiella (2), Serratia (5), Vibrio (1), Yersinia (1), Shewanella, Photobacterium, and Photorhabdus (3). Numbers in parentheses indicate copy number.

^d Escherichia (3–7), Shigella (3–6), Citrobacter (1), Salmonella (3), and H. pylori (2, 3). Numbers in parentheses indicate copy number.

^e E. coli, Salmonella (large insertion), Citrobacter, Shigella, Enterobacter, and Klebsiella.

^f E. coli, Salmonella enterica serovar Typhimurium, Citrobacter koseri, Shigella sonnei, Shigella flexneri, Shigella dysenteriae, and Shigella boydii.

^g MGE, mobile genetic element.

Many excellent reviews covering the field of type I TA systems have already been published (10, 28–30). However, recent work on the characterization of a new family of type I TA systems in Helicobacter pylori (31) pointed to the importance of cis-encoded mRNA functional elements that act together with the sRNA antitoxin to prevent toxin expression. For this reason, this review will focus on the comparison of the various modes of regulation adopted by TA loci, and, more specifically, on the role played by mRNA structure to regulate toxin expression at the translation initiation step.

EXTENSIVE REGULATION OF TYPE I TA SYSTEMS

A major consequence of toxins being lethal to the cell that produces them is that unwanted expression can induce bacterial cell death. Therefore, this lethality applies a strong selection pressure to control the expression of toxin-encoding genes in bacterial genomes. Except for type I, TA systems have an operon organization in which both the toxin and the antitoxin are usually produced from the same promoter, the antitoxin being always encoded upstream from the toxic gene (1). This ensures the availability of an antitoxin molecule ready to counteract toxin activity as soon as it is produced, via either sequestration or an indirect mechanism (antagonism) (Fig. 1A and B). Although this operon organization appears to be sufficient to control toxicity, several additional layers control toxin expression at both the transcriptional and posttranscriptional levels. For instance, in type II TA loci, transcriptional autoregulation by the TA complex is combined with translational coupling of the upstream antitoxin, ensuring that for each toxin molecule produced, an antitoxin molecule has been synthesized as well (Fig. 1A). A reverse gene order has been described for some type II TA systems (32), but it correlates with the use of a noncanonical start codon (GUG) or a short leader region of the toxin-encoding mRNA, probably leading to poor translation efficiency (33, 34).

FIGURE 1.

Various modes of antitoxin-mediated regulation. The three main types of toxicity regulation by antitoxins. (A) Direct sequestration. In types II and VI TA systems, toxin inactivation involves a direct interaction between the toxin (T) and the antitoxin (A). The formation of an inactive TA heterocomplex (T-A) can, in its turn (for type II), lead to the transcriptional repression of the operon (red star). In type VI, the formation of the inactive TA complex favors toxin degradation by cellular proteases (yellow circle). In type III, the antitoxin is an RNA that directly binds to the toxin to prevent its toxic activity. (B) Antagonism. Both toxin and antitoxin compete for binding to the same target. The interaction can additionally have opposite functional (antagonistic) effects. (C) Control of expression. In types I and V, regulation occurs at the posttranscriptional level. In type I, antitoxins are antisense RNA molecules that base-pair to the toxin-encoded mRNA to alter its expression by either inhibiting translation initiation or promoting its degradation. In type V, the antitoxin is an RNase that cleaves the toxin-encoded mRNA.

In type I TA systems, on the contrary, the lack of direct interaction between the toxic peptide and the RNA antitoxin implies that toxicity can only be controlled at the level of toxin expression and not toxin activity (Fig. 1C). This regulation is achieved via the base-pairing of an untranslated RNA with the toxin-encoding mRNA, which inhibits toxin translation and/or promotes its degradation (Fig. 1C) (10, 28). Both RNAs are transcribed from their own promoters. Another shared feature of many type I TA systems is that toxins are small proteins (peptide size, <60 amino acids). With the exceptions of RalR and SymE toxins (Table 1) (35, 36), they target the bacterial membrane (28), leading to membrane depolarization and ATP loss (20, 27, 37, 38) or membrane invagination (39). However, some families of toxins, such as Fst/Ldr, have been shown to lead to nucleoid condensation as the most immediate effect upon production (17, 40). Remarkably, in the case of the type V ghoT/GhoS TA locus, the toxin is also predicted to be a small membrane protein (57 amino acids), but in this case, the antitoxin GhoS is an RNase that specifically cleaves the toxin mRNA (Fig. 1C) (41). Hence, similarly to type I TA systems, GhoS regulates toxicity indirectly at the level expression.

Shine-Dalgarno Sequestration Is Required To Uncouple Transcription and Translation

The absence of nuclear compartmentalization is one of the key features distinguishing prokaryotic from eukaryotic cells. One consequence is that it allows the coupling, in time and space, of two major cellular processes: transcription and translation (42) (Fig. 2A). Although it does not mean a complete absence of physical compartmentalization in bacteria, as some specific RNases (RNase E and RNase Y) have been observed to be at the membrane (43, 44), it is clear that a physical link does exist between transcription and translation, at least during the first round of protein synthesis (45). This physical interaction between both machineries is thought to occur not only at early stages of translation to allow cellular colocalization for efficient translation initiation (46) but also during translation elongation to prevent RNA polymerase from backtracking (42). This coupling between transcription and translation of nascent mRNAs is important for some regulatory mechanisms. It can be very useful to prevent the accumulation of nonfunctional transcripts via nonsense polarity (47). However, when it comes to toxin-encoding mRNAs, transcription/translation coupling can be lethal and make difficult the regulation by an antisense RNA.

FIGURE 2.

Illustration of the consequences of transcription/translation uncoupling for regulation of type I TA expression. (A) The lack of nuclear compartmentalization in bacteria leads to the coupling in time and space of transcription and translation processes. (B) Transcription/translation coupling of type I toxin-encoding mRNAs would be lethal. TA systems can be conserved in bacterial genomes thanks to the decoupling of such processes through the sequestration of the SD sequence (red) during transcription (1). This SD sequestration is conserved in primary transcripts, making them unable to interact with both ribosomes and antitoxin sRNAs (2). In the cases where SD sequence sequestration involves a 5′-3′ LDI, the formation of successive metastable structures ensuring SD inaccessibility to both ribosomes and antitoxin during transcription is essential to prevent premature toxin expression and mRNA degradation, respectively. Translational activation is achieved by the enzymatic processing (of the 5′ or 3′ mRNA end, depending on the TA system) of the primary transcript followed by a structural rearrangement (3) that renders the mRNA able to interact with both ribosomes (4A) and the antitoxin (4B). Ribosome binding to the accessible SD sequence (green) leads to toxin production, inducing either growth arrest or cell death (5A). On the opposite, antitoxin binding efficiently inhibits toxin translation and promoter mRNA degradation, allowing cell survival (5B).

In bacteria, initiation is usually the rate-limiting step of translation. The canonical mechanism of translation initiation involves the binding of the 30S ribosomal subunit to the translation initiation region (TIR) of an mRNA. The TIR comprises all the elements required for translation initiation to occur (48). The translation initiation of many, though not all, mRNAs depends on the interaction between a sequence element upstream from the start codon (the Shine-Dalgarno sequence [SD]) and a complementary sequence at the 3′ end of the 16S rRNA (anti-SD sequence [aSD]) (49, 50). Additionally, a downstream AUG (or GUG/UUG) start codon, optimally spaced (at around 4 to 9 nucleotides) from the SD sequence (51), binds fMet-tRNA^fMet and GTP-bound initiation factor IF2 to set the reading frame. Besides these two regulatory elements acting at the primary sequence level, mRNAs having a highly structured TIR can also possess so-called translational enhancer sequences (often AC-, AU-, or U-rich), proposed to facilitate translation initiation by protein factors such as the ribosomal protein S1 (52–54). In the absence of an SD sequence, alternative modes of translation initiation have been described. For instance, translation could initiate via direct 70S ribosome binding or, alternatively, for polycistrons, through a 70S-scanning initiation mechanism (55). This latter mode could be used by type II TA systems to translate their toxins that lack a proper SD sequence (Fig. 1A).

Usually, the efficiency of translation initiation has been linked to ribosome accessibility and mRNA folding around the TIR (54). However, in the case of toxic genes, ribosome accessibility must be tightly controlled. Several studies have highlighted the presence of one common feature of the toxin-encoding mRNAs: the primary transcript is usually translationally inert due to the intramolecular sequestration of the SD sequence by a partially or totally complementary aSD sequence. This has been observed for many TA systems in both Gram-negative and Gram-positive bacteria (see some examples in Fig. 3). Experimental evidence of this sequestration has been shown for Hok (56, 57), TisB (58), ZorO (59), DinQ (60), and AapA1 (31) mRNAs in Gram-negative bacteria, and for Fst (61), SprA1 (62), TxpA (63), BsrE, and BrsG (64, 65) mRNAs in Gram-positive bacteria. More importantly, if we look carefully at the mRNA sequence of some TA systems for which no SD sequestration has been reported yet, we can easily find an aSD sequence located a few nucleotides upstream from the SD sequence for many of them. This is the case for the newly identified type I TA systems in Clostridium difficile, for which a 5′-CCUCCC-3′ sequence can be found 11 nucleotides upstream from the SD sequence of the toxin mRNA (66). We also identified a complementary sequence just upstream from the SD in the mRNA of the SprG1 toxin in Staphylococcus aureus (67). Remarkably, the aSD sequence is always encoded upstream from the SD sequence to immediately mask the SD after its synthesis, uncoupling transcription from translation (Fig. 2B and 3). It would be interesting to reverse this order to see if this is lethal or not.

FIGURE 3.

Examples of secondary structures sequestering the SD sequence of toxin-encoding mRNAs. (A) Stem-loop sequences in which the SD sequence is totally or partially sequestered by an upstream aSD sequence. These secondary structures have been experimentally validated in vitro (58, 61–64) or predicted (indicated by *) (105). For BsrG, the stem-loop sequestering the SD is shown in absence of SR4 antitoxin (**). Indeed, when the antitoxin binds to the mRNA, the stem-loop sequestering the SD is extended by 4 additional base pairs (64). (B) Stem-loop structures sequestering the SD sequence of the Mok leader peptide and the Hok toxin. In this case, the formation of the Mok SD-sequestering stem-loop is dependent on a 5′-3′ LDI (69, 106). (C) Examples of SD sequestration achieved by a stable LDI between both mRNA ends, creating a cloverleaf structure (31, 70). The start codon (AUG, shown in green) can additionally be partially or totally sequestered in one of the cloverleaf structures. Dotted gray lines indicate the presence of unrepresented structures/lengths. Full gray lines schematically represent structures and base pairs. SD sequences are shown in red; aSD sequences are shown in black. Positions are relative to the transcription start site of the mRNAs (+1). 5′ and 3′ indicate orientation of the mRNA. Toxin mRNA names are indicated under each structure. The small index next to each name indicates the host organisms: Ec, E. coli; Bs, B. subtilis; Ef, E. faecalis; Sa, S. aureus; and Hp, H. pylori.

An exception to this paradigm does exist. In the case of YonT toxin-encoding mRNA, the SD sequence is surprisingly not sequestered (63). In this particular case, the start codon is GUG instead of the canonical AUG (63, 68). This could reduce the translation initiation rate by altering the binding affinity of the fMet-tRNA^fMet to the start codon. Another strategy to decrease translation is to display a strong complementarity between the toxin SD sequence and the 16S rRNA aSD sequence. This has been described for TxpA and YonT toxin mRNAs, which have 11- and 12-nucleotide-long SD sequences, respectively. A strong SD sequence increases ribosome pausing, leading to a reduction of the translation initiation rate and an increase of the time window the sRNA antitoxin has to bind to the 3′ end of the toxin mRNA, preventing its translation (63, 68).

In some cases, the intramolecular sequestration of the SD sequence in primary transcripts involves a long-distance interaction (LDI) between the 5′ and 3′ ends of the toxin mRNA (as described for AapA1 and Hok, and Ibs mRNAs in H. pylori and E. coli, respectively) (Fig. 3) (31, 69, 70). In this case, considering the context of transcription/translation coupling, a new question arises. What prevents premature toxin translation of nascent transcripts in which the most energetically stable mRNA conformation sequestering the SD sequence has not been synthesized yet?

Previous studies on the MS2 phage maturation protein suggested a kinetic model of translational control mediated by delayed RNA folding (71). Indeed, this mRNA is untranslatable because the leader adopts a well-defined cloverleaf structure in which the SD sequence of the maturation gene is taken up in a long-distance base-pairing with an upstream complementary sequence. To allow the synthesis of the maturation protein, a transient RNA hairpin is formed before the sequence that will create the inhibitory structure is transcribed (71). In fact, such transient RNA structures have been shown to form during Hok mRNA transcription to mask the SD sequence temporally, ensuring the inaccessibility of ribosomes to the nascent mRNA (72). More recently, by using the Kinefold stochastic simulations that predict cotranscriptional folding paths of mRNAs (73), we predicted the formation of two successive metastable hairpins in the AapA1 toxin mRNA. These transient structures sequester the SD sequence during transcription and before the RNA polymerase reaches the end of the transcript (31). We recently confirmed the existence of such structures in vivo for the AapA3 toxin mRNA (from the same family as AapA1) by the use of genetic approaches (S. Masachis, N.J. Tourasse, C. Lays, M. Faucher, S. Chabas, I. Iost, and F. Darfeuille, unpublished data).

Global mRNA folding may itself play key roles in the regulation of type I TA systems. A compact mRNA secondary structure has often been described as a key feature of toxin-encoding mRNAs (e.g., Hok, TisB, AapA1, Ibs, Fst, SprA, and BsrG). For instance, the AapA1 and Ibs mRNAs adopt, as in the case of MS2, a well-defined cloverleaf structure in which only a few nucleotides are in single-stranded regions (Fig. 3C) (31, 70). This comes together with the co- and posttranscriptional SD sequestration, but it can go way beyond it. Indeed, in the absence of transcription/translation coupling, the nascent mRNA can base-pair to the DNA template and form so-called R-loops. It is well known that highly structured RNAs are less prone to R-loop formation. Hence, for TA systems, avoiding R-loop formation may prevent premature transcription termination and enable the completion of transcription until the translationally inactive full-length mRNA is made. Further, considering that R-loops can often lead to an increased mutation rate during RNA synthesis (74), their prevention through the conservation of highly structured mRNAs may represent an evolutionary advantage for the conservation of type I TA systems.

In addition, mRNA folding is a key determinant in controlling mRNA stability. As described above, most primary transcripts of type I TAs are translationally inactive. It is well known that the absence of translating ribosomes causes mRNA destabilization in E. coli (75). Surprisingly, these untranslatable primary transcripts are particularly highly stable compared with other cellular mRNAs, displaying an extremely long RNA half-life (31, 76). A compact folding of most toxin-encoding mRNAs is probably very efficient in protecting these mRNAs from degradation by cellular RNases. Whether some specific RNA-binding proteins play a role has not been reported yet (77).

In summary, all these regulatory threats (premature toxin synthesis, R-loop formation, and RNA degradation) likely represent a significant selective pressure to preserve a compact mRNA structure in toxin-encoding mRNAs and may lead to a possible domestication process ensuring the conservation of TA loci in bacterial genomes.

Toxin-Encoding mRNAs Require an Activation Step To Be Translated

During the study of the aapA1/IsoA1 TA locus in H. pylori (31), we realized that the most abundant mRNA transcript was untranslatable. In contrast to classical mRNAs, the “active” toxin-encoding mRNA species could not be detected under normal growth conditions. This translatable transcript was revealed during rifampin assays. This antibiotic is classically used to block transcription in bacteria in order to determine RNA half-life. One hour after rifampin addition, a shorter transcript was accumulating while the pool of RNA antitoxin had been completely degraded. We further showed that this transcript underwent a two-step 3′-end nucleolytic activation process involving the 3′-5′ exonucleolytic activity of the polynucleotide phosphorylase (PNPase) (S. Masachis, H. Arnion, S. Chabas, F. Boissier, and F. Darfeuille, unpublished). This result is reminiscent of the 3′-end mRNA activation described more than 20 years ago for the hok/Sok system in E. coli (76). Indeed, in both cases, the primary transcript cannot be translated due to the stable sequestration of the SD sequence by an LDI between the 5′ and 3′ ends, and requires an activation step via 3′ processing (Fig. 4). Although the hok/Sok and aapA1/IsoA1 TA systems show no sequence homology, it was really striking to see that an activation process involving the PNPase was conserved in two evolutionarily distant Gram-negative bacteria. However, although this 3′-end trimming induces a strong structural rearrangement of the 5′ untranslated region (5′ UTR) of both AapA1 and Hok mRNAs, translational activation of the Hok mRNA requires the translational coupling of an upstream open reading frame (ORF) encoding the Mok peptide (Fig. 3B and 4) (78).

FIGURE 4.

Type I operon organization in Gram-negative bacteria and mechanistic consequences of its regulation. Type I antitoxin sRNAs in Gram-negative bacteria can be encoded in two main fashions: (i) overlapping the 5′ end of the toxin mRNA, the ORF, or a leader ORF (left panel) or (ii) not overlapping (right panel). In both cases, transcription/translation coupling forces the sequestration of the SD sequence by partially or totally complementary sequences called anti-SD (aSD). This sequestration starts during transcription but is maintained upon transcription termination, leading to the generation of a translationally inert and sRNA-inaccessible primary transcript (full-length mRNA). Location of the aSD sequence will determine whether the sequestration occurs via 5′-3′ LDI (5′-overlapping TA loci) or in a stem-loop (nonoverlapping TA loci). In both cases, an enzymatic activation step is required for the generation of the truncated (active) mRNA. When the SD sequestration involves 5′-3′ interaction, this activation step often occurs via 3′ trimming by 3′-5′ exonucleases (RNaseII, PNPase). In contrast, when SD is sequestered in a stem-loop, activation occurs via 5′-end processing by endonucleases. In either case, a light or strong structural rearrangement (refolding) is needed upon processing to render the SD accessible to both ribosomes and sRNA binding. Next, in noninduced conditions, antitoxin sRNAs outcompete the ribosomes for binding to the 5′ end of the toxin mRNA and render it translationally inert. This inactivation step can occur via direct sequestration of the SD sequence or the leader ORF SD sequence (5′-overlapping TA loci), or indirectly via the sequestration of the ribosome standby site (stand-by) or the stabilization of an SD-trapped structure (nonoverlapping TA loci). In most cases, sRNA binding to the toxin mRNA leads to RNase III-mediated toxin mRNA decay and cell survival.

More surprisingly, for TA systems whose antitoxin is encoded in a nonoverlapping fashion (Fig. 4, right panel), translation activation happens at the 5′ end of the toxin mRNA. Multiple mRNA 5′ ends have been mapped for the tisB/IstR (58), dinQ/AgrAB (20), shoB/OhsC (79), and zorO/OrzO (59) TA systems. However, the enzyme(s) responsible for this processing is still unknown. Irrespective of how the activation step occurs, it always triggers a structural rearrangement that opens the possibility to interact with both ribosomes and antitoxin, entering into an essential decision step: toxin translation or mRNA inactivation (Fig. 2B). For the dinQ/AgrAB system, this cleavage leads to a strong rearrangement between various complementary sequences that renders the SD site directly accessible to ribosomes (60). In this case, after 5′-end processing, the mRNA becomes translatable.

In the case of Gram-positive bacteria (Fig. 5), it is so far unclear how the activation step occurs, or, eventually, if it is required at all. Indeed, similarly to what has been observed in Gram negative-bacteria, SD sequestration has also been reported for several toxin-encoding mRNAs, such as BsrG (80), BsrE (65), TxpA (63), Fst (61), and SprA1 (62). Although extensive studies have been carried out on the par locus in Enterococcus faecalis, no processing could be detected for the Fst mRNA (5, 7, 61, 81, 82). Nevertheless, the accumulation of a truncated transcript after rifampin addition was observed during the characterization of Fst-like TA systems in Streptococcus mutans and Lactobacillus rhamnosus (83, 84). However, although this observation is really intriguing, it will require further characterization. Indeed, while the sprA1/asSprA1 TA locus has been shown to be a member of the Fst family, the stem-loop masking the SD sequence in the SprA1 mRNA seems very different than the one in the Fst mRNA (Fig. 3). So, while some TA systems share strong homologies, they may undergo different pathways of activation.

FIGURE 5.

Mechanistic regulatory consequences of the 3′-overlapping antitoxin sRNAs in Gram-positive bacteria. In all type I TAs described so far in Gram-positive bacteria, antitoxin sRNAs are encoded in 3′-overlapping fashion to the toxin mRNAs. As in Gram-negative bacteria, transcription/translation coupling forces the sequestration of the SD sequence during transcription, the nascent transcript being accessible neither for the ribosome nor the antitoxin. Upon transcription termination, the full-length mRNA becomes targeted by the antitoxin RNA that binds to the 3′ end of the toxin mRNA. The question marks represent how mRNA activation could occur in Gram-positive bacteria (little is known about the mRNA activation, processing, or refolding steps compared to Gram-negative bacteria). In some cases (TxpA and YonT), sRNA binding is not sufficient to impede toxin translation, and thus, mRNA degradation by RNase III is essential to avoid toxicity. In some cases, sRNA binding leads to a structural rearrangement that enhances the sequestration of the SD sequence. This interaction can in its turn lead to mRNA degradation by RNase III or on the contrary to the stabilization of the translationally inert complex. In all cases, antitoxin binding to the toxin mRNA hampers its expression and allows cell survival.

In summary, Fig. 2B illustrates the main regulatory mechanisms occurring in Gram-negative bacteria, which could probably be extended to Gram-positive bacteria: during the course of transcription, translation of the nascent transcript, as well as antitoxin binding, is impaired by the formation of successive RNA secondary structures sequestering the SD sequence. Upon transcription termination, the full-length mRNA is translationally inactive due to the stable sequestration of the ribosome binding site (RBS). Finally, the mRNA undergoes a nucleolytic activation process followed by a structural refolding that renders it able to interact with the ribosomes or the antitoxin RNA. As a result, either ribosomes bind and the toxic peptide is produced, leading to growth arrest or cell death, or the sRNA antitoxin binds, primarily inhibiting translation initiation and leading to the irreversible degradation of the toxin mRNA (Fig. 2B).

Standby RBSs Can Be Required for Translation

For some type I TA systems, the structural rearrangement induced upon the nucleolytic activation step does not change the accessibility of the SD sequence in the toxin mRNA. Those mRNAs require an additional element to be optimally translated: a ribosome standby site. While studying the translation of the highly structured MS2 coat protein mRNA, de Smit and van Duin (85, 86) encountered a paradox. They showed that ribosomes compete with mRNA folding and need to capture the SD in an unfolded state to initiate translation. However, the calculated energetic ΔG value of the SD-sequestering stem-loop was a problem: the fraction of time the stem-loop would spend in an unfolded state is in the microsecond range. Considering the time in an unfolded state and the ribosomal 30S subunit association rate, the calculations predicted a coat protein translation rate >10,000-fold lower than the observed one. This paradox was solved by the “standby” model (86). In this model, 30S subunits preloaded to a yet undefined single-stranded region can access the SD as soon as the structure is transiently unfolded (“breathing”), thus following a kinetically driven (first-order reaction) instead of a thermodynamically driven model (second-order reaction) (87).

This mechanism has been identified for the TisB toxin mRNA (58). The TisAB primary transcript (+1) is processed into a 5′-end shorter transcript (+42). However, the SD in both mRNA species is sequestered within a stable stem-loop (Fig. 3). The structural rearrangement upon processing of the TisAB mRNA exposes a region in the 5′ UTR, the standby site, thus allowing 30S accessibility to the sequestered RBS as soon as the stem “breathes” (Fig. 4). A similar translation mechanism has been recently suggested for the ZorO toxin mRNA (59); however, further studies are needed to confirm this hypothesis.

As mentioned above, “translational enhancer” sequences (often AC-, AU-, or U-rich) have been proposed to facilitate translation initiation mediated by protein factors such as the ribosomal protein S1. Strikingly, AC-rich motifs can be found in several toxin-encoding mRNAs, including the TisB standby site and the ZorO putative standby site, suggesting a potential relationship between the presence of such sequences and the translation efficiency mediated by standby sites (59, 87).

Antisense RNAs Bind to and Repress Activated mRNAs

Until now, we have highlighted the importance of RNA structural elements embedded in toxin mRNAs to prevent toxin expression. However, by definition, the inhibition of toxin synthesis in type I TA systems is achieved by direct base-pairing of an antisense RNA to the toxin-encoding mRNA. Depending on whether the antisense is encoded in cis or trans, the region and length of base-pairing can vary. Several organizations have been described in Gram-negative bacteria (Fig. 4). For instance, in E. coli, four TA loci have the antisense RNA encoded in a divergent orientation to the toxin-encoding gene: shoB/OhsC (16), zorO/OrzO (38), tisB/IstR (18), and dinQ/AgrAB (20). Remarkably, the antisense RNA of these TA systems targets a similar region in the 5′ UTR of the toxin-encoding mRNA displaying a short region of perfect complementarity (18 to 23 nucleotides). Indeed, a minimum of 15 continuous base pairs has been shown to be necessary for efficient repression of ZorO toxin expression (38). In the case of the DinQ toxin, the region of complementarity can be extended to 31 nucleotides containing 1 mismatch (60). However, antisense RNA antitoxins are usually encoded in cis, overlapping either the 5′ (in Gram-negative bacteria) or the 3′ end (in Gram-positive bacteria) of their toxin mRNA counterparts (Fig. 4 and 5). Although this organization can lead to the extensive base-pairing between the antisense RNA and its target mRNA, the initial pairing is often mediated by single-stranded regions such as kissing-loop interactions (31, 63). These single-stranded regions are often formed after the processing of the primary toxin-encoding mRNA (31, 76) and facilitate the pairing between both toxin and antitoxin RNAs. In particular, for some TA loci, the presence of a U-turn (YUNR) motif in one of the loops greatly enhances the pairing rate between the toxin and antitoxin RNAs (5, 7, 64, 82, 88). This initial short pairing is usually propagated into an extensive base-pairing forming a long RNA intermolecular duplex with increased stability. In most of the cases, this duplex is sufficient to mask the RBS, leading to the inhibition of toxin translation. Remarkably, in the case of the BsrG toxin-encoding mRNA, the SR4 antitoxin binding does not directly interfere with ribosome binding but instead generates an extended intramolecular SD sequestration (64). More surprisingly, RNAI, the Fst toxin mRNA in E. faecalis, displays two complementary regions, named DR (direct repeat) a and b, with the antisense RNAII (5, 7, 82). These two regions are in close proximity to the SD sequence of the RNAI. After the initial pairing of RNAII with RNAI in the mRNA 3′ end, this interaction is followed by an extended pairing within this region of complementarity in the 5′ UTR. As a consequence, the binding of RNAII leads to the formation of a stable duplex that blocks toxin translation. Surprisingly, the homolog of the par locus in S. aureus (Table 1), sprA1/asSprA1, seems to retain only the interaction with the 5′ UTR but not with the terminator stem-loop, to inhibit translation initiation (62). Another important point is that antitoxins seem to specifically target the activated toxin mRNAs (processed at the 5′ or 3′ end), the primary transcript being inert for interaction, as shown for the processed Hok (89), AapA1 (31), and TisB (58) mRNAs. Thus, in both Gram-negative and Gram-positive bacteria, RNA antitoxins act primarily at the translational level by efficiently impeding ribosome binding either directly or indirectly.

Besides translation inhibition, antisense RNA binding also leads to mRNA degradation (Fig. 4 and 5). Indeed, these duplexes are recognized and cleaved by the double-stranded specific RNase III, as reported, for instance, for the AapA1 (31), Hok (89), TisB (58), and BsrG (64) toxin-encoding mRNAs. However, this degradation is, in most cases, not essential to suppress toxin expression. This feature can be easily confirmed, for instance, by the viability of strains deleted for RNase III, as shown for H. pylori and E. coli (18, 31). Nevertheless, there are two exceptions to this paradigm. In Bacillus subtilis, the antitoxin binding is not always sufficient to avoid toxin expression. The binding of RatA and SR6 antitoxins to TxpA and YonT mRNAs, respectively, does not block toxin translation (63, 68). In contrast, the antisense pairing to the toxin mRNA 3′ end leads to the formation of a long, extended duplex cleaved by RNase III, whose activity is essential to avoid toxin expression (Fig. 5).

Type I TA Systems May Also Be Modulated by RNA Editing

RNA editing is a widespread posttranscriptional regulation layer known to recode RNAs and proteins in plants, animals, fungi, protists, bacteria, and viruses (90). The most common modification is adenosine (A)-to-inosine (I) conversion, mediated by enzymes of the ADAR (adenosine deaminase, RNA-specific) family. In bacteria, RNA editing was so far thought to occur exclusively in a single nucleotide site within a tRNA for arginine (tRNA^Arg). It was known to be mediated by a tRNA-specific adenosine deaminase (tadA) that recognizes a 4-base-long motif, TACG, with the edited A in second position (91). Interestingly, TadA activity requires a specific mRNA secondary structure on which the edited base is embedded within a loop (92).

Remarkably, a recent study in E. coli revealed 15 novel A-to-I RNA-editing events (93). Twelve of these occur within ORFs, contrasting with the mostly noncoding RNA editing reported in eukaryotes. Four such events occur within ORFs belonging to the Hok family of host-killing toxins, which may not be a surprise, as the original tadA mutant strain was obtained during the search for suppressors upon overexpression of HokC toxin (gef protein) (91, 94). The HokB mRNA showed the highest levels of RNA editing. Sequence alignment reflected that HokB mRNA editing could occur even outside E. coli, in different bacterial species such as Klebsiella pneumoniae and Yersinia enterocolitica, thanks to the conservation of the TadA enzyme. Interestingly, HokB orthologs either have a tyrosine (Tyr) encoded by the editable TAC codon at position 29 and a cysteine (Cys) at position 46, or on the opposite, a Cys at position 29 and an editable Tyr codon at position 46. Notably, both positions reside in one of the two β-strands predicted in the HokB peptide, suggesting that they may play a functional role. Interestingly, all the editable positions lie within the TACG TadA-recognition motif in a loop that is the only mRNA region identical to the tRNA^Arg.

Overexpression experiments to study the effect of the different variants at position 29 showed that all three studied variants (Tyr encoded by the editable TAC codon; Tyr encoded by the noneditable TAT codon; and Cys TGC, mimicking the postediting situation) are toxic. However, the strongest effect was observed for the peptide mimicking the postediting situation (Cys at position 29), indicating that RNA editing in HokB mRNA enhances peptide toxicity.

These results established RNA editing as an alternative mode to regulate toxin activity, one more posttranscriptional regulation layer in bacteria, which importantly may contribute to proteome diversity and phenotypic heterogeneity between genetically identical cells. Remarkably, it also represents one more example in which an RNA secondary structure rules a posttranscriptional regulatory event. In the case of HokB, it additionally opens the question of whether RNA editing could represent a key adaptation mechanism, which could be implied in situations where phenotypic diversity is advantageous, such as antibiotic persistence, thus reflecting a more complex bet-hedging mechanism than the previously suggested one (23). However, while RNA editing has been shown for HokB, it still remains to be seen how widespread such a modification may be in other type I toxins.

PERSPECTIVES

Identifying Type I TA Systems

As already mentioned, type I TA systems were initially discovered on plasmids based on their biological function (i.e., PSK). However, only a few families of type I TA systems have been identified on plasmids, while many have been found on chromosomes (hok/Sok, fst/RNAI, and, more recently, aapA/IsoA) (15, 31, 95) (Table 1). As highlighted by Coray and colleagues (96), there is a narrower distribution of type I and III TA systems compared to type II TA systems. Type I systems are more likely found within single clades of bacteria rather than dispersed across phyla (96).

Much of the known distribution of TA systems relies on in silico searches based on sequence conservation (12, 15). Historically, type I systems have been difficult to predict in silico, often due to the small size of the toxin (<60 amino acids) and to the lack of a prediction tool for their RNA antitoxin counterpart. Pioneering work done by Fozo and colleagues greatly expanded the number of known TA loci across 774 bacterial genomes and predicted new families of TA systems (22). They based their search on toxin hydrophobicity and highly structured sRNA based on RNA-folding energy profiles. They also looked for loci duplication (based on ibs/Sib and ldr/Rdl examples) and polar C-terminal residues of the potential toxins.

However, even if some strategies have succeeded in the de novo prediction of type I TA systems (22), it remains a big challenge. Our group has recently implemented a new database that includes all the known type I TA loci (Table 1) (N. J. Tourasse and F. Darfeuille, unpublished data). The search for new type I TA systems focused on the mRNA folding features of the toxin-encoding transcripts rather than on the hydrophobicity and sequence of the predicted toxin. The mRNA features common to each family of type I TA systems (AapA, BsrE/G/H, DinQ, Fst, Hok, Ibs, Ldr, ShoB, TisB, TxpA, YonT, and ZorO) included the mRNA length, SD sequence sequestration by an upstream complementary sequence (within a stem-loop or via 5′-to-3′ LDI), as well as the antitoxin sRNA secondary structure. This algorithm independently recovered the mRNAs and sRNAs of most TA loci that were previously identified based on peptide features and greatly increased the number of newly identified ones, demonstrating that RNA features are key determinants for the identification of type I TA systems.

Understanding the Functions of Chromosomally Encoded TA Systems

As we have seen in the first part of this review, toxin expression needs to be tightly regulated and this regulation strongly depends on antitoxin RNA levels. For PSK function, as in the hok/Sok and fst/RNAII systems, constitutive expression from both toxin and antitoxin promoters is required (10). The same is true for the two chromosomally encoded txpA/RatA and yonT/SR6 loci in B. subtilis that are involved in the stabilization of the Skin and SPβ prophages, respectively (63, 97). However, the link between chromosomal TA systems and PSK has not yet been shown for other TA systems. For instance, we observed constitutive expression of both toxin-encoding mRNA and antitoxin sRNA in the aapA1/IsoA1 TA system in H. pylori (31). However, we have not been able to demonstrate its role in PSK despite its location near integrative and conjugative elements.

While the function of TA systems encoded on plasmids or near mobile genetic elements can be easily associated with PSK, elucidating their function when they are encoded on the chromosome is still challenging. However, from a biological point of view, TA systems could potentially be beneficial at the population level (e.g., increase phenotypic heterogeneity, which enhances fitness and adaptation to adverse environmental conditions). Therefore, bacteria may have domesticated TA systems, allowing them to fully take over the control of their expression and make use of it for their own advantage. One way to address this question is to search for possible regulation of the antitoxin RNA expression. There are a number of examples where TA expression may be related to stress responses. In E. coli tisB, symE, and dinQ, toxin promoters are preceded by LexA boxes, making them inducible during SOS response via the LexA transcriptional repressor, whereas their sRNA antitoxins are constitutively expressed (18, 20, 36) (Table 1). While both toxin-encoding mRNA and antitoxin RNA of the par locus (pAD1 plasmid) have been found to be constitutively expressed, the antitoxin RNA of its chromosomal copy in E. faecalis (98), as well as the S. mutans and S. aureus homologs, are regulated in response to various stress conditions (13, 62, 83). SprA1 antitoxin expression decreases under acidic pH stress and in response to oxidative stress (99). In B. subtilis, the expression of several TA loci is modulated in response to multiple stresses including ethanol, heat shock, and anaerobic stresses (65, 80, 100) (Table 1). Finally, it has recently been shown that upstream of the promoter of IsoA3 antitoxin in H. pylori there is a binding site for the essential orphan response regulator HP1043, which has been related to oxidative stress and growth phase-dependent responses (101). This is an interesting observation, and further studies need to be performed to confirm a potential transcriptional regulation of IsoA3 expression.

The levels of RNA antitoxins are also adjusted at the posttranscriptional level by RNA degradation. Depending on the bacterial species, several RNases are responsible for antitoxin turnover. RNase E controls Sok RNA degradation in E. coli (102), RNase J controls IsoA3 RNA stability in H. pylori (103), and RNase Y controls many antitoxin sRNAs (e.g., SR4 and RatA) in B. subtilis. Therefore, it could be very exciting to evaluate to what extent bacteria can modulate the level of antitoxin RNA by adjusting the accessibility of these degrading enzymes to their substrate. For instance, several antitoxin sRNAs (IstR, Sib, and Rdl) were found to be associated with the RNA-binding protein ProQ in Salmonella (77). Whether this interaction affects their stability or their ability to regulate their toxin-encoding mRNA counterpart is not known yet.

CONCLUSIONS

Overall, in this review we wish to highlight that for type I TA systems, antisense RNA binding is not sufficient to control toxin expression. Due to the transcription/translation coupling that occurs in bacteria, tight control of toxin expression requires the presence of cis-encoded elements, which co- and/or posttranscriptionally occlude the RBS, as already discussed in earlier reviews (10, 13, 28). This sequestration can occur within the stem-loop or by an LDI between both mRNA ends. In the first case, the aSD sequence is encoded just upstream from the SD sequence in order to form a stable stem-loop, leading to the sequestration of the RBS at early stages of transcription. In the second case, the SD sequestration requires sequences that are encoded at the 3′ end of the mRNA. Thus, when SD sequestration occurs by mRNA end-pairing (LDI strategy), additional intermediate structures must be formed to impede premature toxin translation (72). These unstable and successive interactions lead to the formation of so-called metastable structures, avoiding cotranscriptional translation of the toxin mRNA. Why and how these two mechanisms of SD sequestration evolved and were kept is still unknown. From an energetic point of view, the stem-loop strategy may be advantageous, as the first SD-sequestering structure formed will be the most stable one. On the opposite, the LDI strategy may be more difficult to make, as one or several metastable structures need to be successively formed before reaching the final and stable folding state of the mRNA. Additionally, the stem-loop strategy may impede premature RNA antitoxin binding in a more efficient way than the metastable structures formed in the LDI strategy. From an evolutionary point of view, the stem-loop strategy seems to offer greater sequence flexibility, as only one aSD sequence needs to be conserved, contrary to the LDI strategy. Nevertheless, a potential advantage of the LDI strategy may be that it allows the formation of a highly stable primary toxin-encoding mRNA (1). This consequence might be really advantageous for the PSK mechanism, as the primary transcript needs to be transmitted to the daughter cell. Irrespective of the employed mechanism, both SD-sequestering strategies require an activation step. While it is clear that the activation needs 3′ exonucleolytic activity in the case of the LDI strategy, no general mechanism has been shown for the stem-loop strategy. Yet an activation step via 5′ endonucleolytic cleavage has been shown or suggested for only a few cases (25, 58–60). In many cases, no processing event has been described (13, 63–65, 68). The combination of new sequencing technologies to precisely map the mRNA 3′ end (Term-seq) (104) with classical techniques, such as rifampin assays to analyze RNA processing (31), should help in deciphering these mechanisms. Alternatively, no processing event may be required to activate these primary transcripts. The involvement of RNA helicases or RNA chaperones (such as ProQ), to facilitate the removal of the antisense RNA paired to the toxin-encoding mRNA, could be a good alternative to RNA processing. Further studies will be required to address this question in more detail.