Excess of a toxin has been shown to destabilize binding of the toxin-antitoxin complex to operator DNA and to activate transcription of its own operon 35 . To test whether toxins can activate transcription of other TA operons, we measured the transcription of relBE in response to ectopic expression of toxins MazF, MqsR, YafQ, HicA, and HipA by northern hybridization (Figure 1). Since the relBE genes are co-transcribed with the downstream relF 45 , which encodes a hok-like toxin targeted against the inner membrane 46 , we analyzed the transcription of the full relBEF operon. In a reverse experiment, we over-expressed RelE and monitored the transcription of several chromosomal TA operons (Figure 2). Amino acid starvation is known to upregulate relBEF transcription 14 and was induced by addition of mupirocin (MUP) 47 as a positive control. Ectopic expression of RelE served as an additional positive control for activation of relBEF transcription whereas synonymous substitutions were introduced into the plasmidal relE sequence (Additional file 1: Table S2; primer relE-XbaUP) to enable unambiguous detection of the chromosomal relE transcript. Active RelE toxin could be expressed from the altered gene (Additional file 1: Figure S1) and the plasmidal transcript was not detectable in the ΔrelBEF strain, showing that our hybridization probes are specific and do not cross-hybridize (Additional file 1: Figure S3A,B,C lanes 1,2). Toxins were induced in log phase cultures and concomitant measurements of optical density confirmed growth inhibition in all cultures tested (Additional file 1: Figure S1). Samples for RNA isolation were collected before induction (−1 min) and during a two hour time-course post-induction (15, 60 and 120 min); mRNA of the chromosomal TA operon was analyzed by northern hybridization using DNA oligoprobes complementary to relB, relE, and relF (Figure 1; Additional file 1: Table S2).

As shown in Figure 1, we indeed saw a clear cross-activation of relBEF in response to all toxins tested except YafQ. Induction of RelE, MazF, MqsR, HicA and HipA conferred a clear increase in the relBEF mRNA level in an hour. Use of three separate probes revealed, however, that different mRNA species pile up in response to different toxins. Before induction and 15 min after, all three probes – relB, relE and relF – detected a transcript of the same size corresponding to the full-length mRNA of the operon 45 , as confirmed later by primer extension mapping of the 5^′ end (Additional file 1: Figure S4). Only after MazF expression a shorter transcript, a putative cleavage product, could be detected at the 15 min time point using relE probe (Figure 1B). At later time points, hybridization with relB (Figure 1A) and relE (Figure 1B) probes gave different signals: in response to induction of MazF, MqsR, and HicA we saw cleavage of the full-length mRNA and massive accumulation of the toxin-encoding part, while the antitoxin-coding portion could not be detected and was apparently degraded (Figure 1A,B). Such cleavage and accumulation of the toxin portion also occurred in response to RelE. Hybridization with relF probe revealed additional cleavage, both within relE and downstream, in response to expression of all these toxins, and the relF part accumulated as the most abundant portion of the relBEF transcript (Figure 1C). Also, some transcripts larger than the full relBEF mRNA appeared, particularly after induction of RelE and MqsR. Production of HipA, which is not a ribonuclease, conferred strong induction of full-length relBEF mRNA but cleavage and uneven accumulation of different mRNA fragments could not be seen. MUP treatment produced overproduction of the full relBEF mRNA as well as accumulation of some cleavage products. Production of YafQ did not lead to a clear cross-activation of relBEF transcription. However, relE probe showed accumulation of a short RNA fragment in response to this toxin. It is possible, that transcription of the operon is activated by YafQ but the transcript is degraded to small fragments. Clearly, these fragments cannot serve as templates for synthesis of RelE and, therefore, functional cross-activation does not occur. Modest induction of relBEF with no cleavage was evident in the 1h and 2h samples of control cultures, lacking artificial production of any free toxin. We have to consider that, at this stage, the control cultures were approaching stationary phase, and induction of toxin-antitoxin modules has been described in similar conditions 48 .

Probes complementary to yiaF and rpsS were used for control because the levels of transcription of these genes did not differ between log phase cells and the ampicillin-refractory non-growing subpopulation, where TA operons were highly expressed 38 . rpsS is a part of the large S10 ribosomal protein operon with an estimated transcribed length of 5181 bp 49 ; yiaF (711 bp ORF) encodes for a putative membrane protein of unknown function; it is located between genes pointing in the opposite direction and must form a single-gene operon. The control mRNAs were not induced by toxins (Additional file 1: Figure S2B,C). After induction of toxins, the yiaF transcript was degraded without accumulation of any stable fragments. (Additional file 1: Figure S2B). Surprisingly, mupirocin initially induced transcription of yiaF whereas the level of the transcript dropped after longer incubation (Additional file 1: Figure S2B). The S10 transcript was degraded as well. Some accumulating stable fragments of the S10 transcript were detectable after MazF, RelE and MqsR production (Additional file 1: Figure S2C).

To be sure that the accumulating RNA fragments, which correspond to the 3^′ portion of the relBEF mRNA, are not initiated from toxin-inducible cryptic promoters within the operon, we deleted the promoter of the relBEF operon. In the promoterless BW25113 ΔP _relBEF strain, we did not see induction of the relBEF mRNA nor the characteristic accumulation of its 3^′ portion (Additional file 1: Figure S3). We still saw a transcript that could be detected by the relE and relF probes (Additional file 1: Figure S3B,C) but the level of this transcript did not depend on the RelE production. It might be initiated from a constitutive promoter that was newly created by deletion of P _relBEF . Transiently induced smear of RNA that was detected in BW25113 ΔP _relBEF with the relB probe (Additional file 1: Figure S3A, lanes 6 and 7) is transcribed from the RelB-expression plasmid pKP3033. That is the reason why we omitted this plasmid when we studied induction of relBEF in response to RelE (Figure 1, Additional file 1: Figure S3, lanes 8–11). Thus, we can be sure that the shorter transcripts that massively pile up in response to toxins are indeed cleavage products and are initiated at the genuine P _relBEF promoter.

Next, we tested whether over-production of the toxin RelE activates other toxin-antitoxin genes in the chromosome. The northern hybridization results show strong induction of the mqsRA, mazEF, dinJ-yafQ, hicAB, yefM-yoeB, and prlF-yhaV TA systems (Figure 2). Similarly to relBEF, the induced transcripts were cleaved and the toxin-encoding parts seem to accumulate preferentially while the antitoxin-coding parts are more effectively degraded. That appears to be true irrespective of whether the toxin is encoded by the first (mqsRA, hicAB) or the second (mazEF, yefM-yoeB, prlF-yhaV) gene of the operon (Figure 2). Reliable testing of this phenomenon requires characterization of the cleavage products and additional experiments in the future.

Additional experiments indicated that transcriptional cross-activation of TA operons does not occur between all possible TA combinations. Northern hybridization using mqsR probe showed that overproduction of MazF and HicA does not induce the mqsRA promoter while YafQ and HipA induce it (data not shown), as well as RelE (Figure 2).

We wanted to test whether TA cross-activation happens also during natural physiological stresses. Amino acid starvation has been shown to induce transcription of the relBE 14 and mazEF 17 genes. We induced amino-acid starvation by addition of mupirocin to the cultures of BW25113 (wild type) and BW25113ΔrelBEF. Northern analysis indicated that transcription of mazEF is upregulated only in wild type bacteria and not in the relBE deficient strain (Figure 3B). Transcription of mqsRA, the other TA operon that we tested, was induced in both strains, independently of the RelBE system (Figure 3A). Thus, RelBE system activates another TA system, MazEF, in response to amino acid shortage. This evidences that TA cross-activation is not a mere artifact of toxin overexpression but occurs as a part of a real physiological response.

Since it is widely accepted that TA loci are activated by proteolytic degradation of antitoxins, we tested whether transcriptional cross-activation is affected by Lon, ClpP or HslV proteases. Besides, we tested the requirement of polyphospate, which has been shown to activate Lon 50 . We expressed RelE, MazF, and MqsR toxins in BW25113 strain lacking lon or ppk, which encode for Lon and polyphosphate kinase, respectively, and observed chromosomal relBEF transcript by northern hybridization using probes relE and relF (Figure 4). Deletion of lon or ppk did not abolish cross-induction of relBEF by MqsR, and as seen on relF probed blot (Figure 4B), by MazF. We further tested relBEF activation in a double-knockout strain lacking Lon and ClpP, and a triple-knockout lacking Lon, ClpP and HslV proteases. Again, expression of MazF and MqsR obviously induced relBEF in the strains deficient for multiple proteases (Figure 4). Accumulating RelE-, MazF- and MqsR- specific cleavage intermediates produced similar patterns in all tested strains (Figure 1B,C, Figure 4). Production of YafQ did not cause a clear activation of relBEF transcription in the protease-deficient strains, similarly to the wt strain. Accumulation of a small fragment hybridizing to the relE probe can be detected in the ΔclpPXΔlonΔhslVU strain (Figure 1B, Figure 4A). Ectopic production of RelE induced transcription of chromosomal relBEF in all strain backgrounds, as expected. Essentially, we can conclude that cross-activation of TA transcription occurs also in lon ^- , ppk ^- , clpPX ^- lon ^- , and clpPX ^- lon ^- hslVU ^- backgrounds.

To characterize the in vivo cleavage of relBEF mRNA in more detail, we mapped the 5^′ ends of the cleavage products using primer extension analysis (Figure 5, Additional file 1: Figure S4, Table S3). As seen in Figure 5, the cleavage sites in the mRNA, which was purified from the cells with over-expression of the nucleases MqsR and HicA, are distributed all over the operon. Several specific cutting sites of the MazF nuclease are found in the RelB-encoding part. No cleavage is detected in response to production of the protein kinase HipA, as expected. Most of the cutting sites were unique for each toxin indicating that the cleavage in vivo was a result of primary activity of the over-produced toxin. RNA from MazF and MqsR over-expression samples was mostly cleaved at the specific cutting sites of these toxins, i.e. ACA 51 and GCU 16 . However, several unique cleavage sites in the MazF and MqsR over-expression samples do not contain these sequences and might be generated by unidentified ribonuclease(s), possibly cross-activated toxins (Additional file 1: Table S3). We also observed that not all ACA and GCU sequences were cleaved in the relBEF mRNA by MazF and MqsR, respectively. As before 19 , the cleavage preferences of HicA could not be identified.

To confirm our notion of TA cross-activation, we hoped to see some cleavage hotspots. At those sites, strong cleavage by an overproduced toxin occurs at its specific cutting sequence (e.g. ACA in the case of MazF). Cleavage at the same site in response to expression of another toxin would indicate activation of the primary cutter by the over-produced toxin. We tested possible cross-activation at three of these sites. At position 174 (ˇACA), the relBEF transcript is cut by MazF and in response to the over-produced HicA. The MqsR-specific cleavage sites at positions 399 (GCˇU) and 431 (GˇCU) are also cleaved in the samples from HicA over-production (Additional file 1: Figure S4). We found that these cuts were not due to the activation of MazF and MqsR, since they occurred in RNA extracted from the BW25113ΔmazEF and BW25113ΔmqsRA cells (data not shown). ChpBK, a homolog of MazF with similar but relaxed sequence specificity 52 may be accountable for the cleavage at 174 (ˇACA).

The toxin-encoding parts of the TA transcripts seem to be generally more stable than the antitoxin-encoding parts and accumulate after cleavage (Figures 1, 2). If the toxin open reading frame (ORF) on these cleavage products is intact and translated into a functional protein, the T:A balance must be shifted towards toxin followed by more cleavage, cross-activation of other TA systems, and inhibition of protein synthesis. That creates the possibility of a positive feedback circuit and even a network of them. A positive autoregulatory loop, in turn, could explain the bistability of bacterial growth observed in response to toxin expression 53 54 .

To test whether proteins are translated from the cleaved relBEF mRNA, we used the T7 promoter for expression of two transcripts, which begin at the sites of MazF-inflicted cleavage, at positions +28 and +148 from the 5^′ end of the full-length transcript, and extend downstream of the relE ORF. The +28 RNA starts immediately upstream of the relB ORF (Additional file 1: Figure S4). Thus, the relB ORF is leaderless and lacks the upstream untranslated region with the ribosome binding site (RBS). The +148 RNA starts in the middle of the relB ORF. To allow RelE to be detected, we added the His6 tag to the C-terminus of the toxin and introduced substitutions R81A and R83A, which reduce its toxicity 55 . Expression of these RNAs in BL21(DE3) resulted in production of the toxin RelE(R81A/R83A)-C-His, although in smaller quantities than from the control transcript with the intact 5^′ end (Figure 6). Thus, the accumulating cleavage products of TA mRNA can be translated into proteins, although less effectively than full transcripts with intact RBS in front of relB. Reduced translation of the downstream relE(R81A/R83A)-C-His open reading frame in shorter transcripts suggests that relE lacks its own RBS and it is produced due to translational coupling of relBE genes. Translational coupling of polycistronic TA mRNA has been demonstrated previously for parD (kis-kid) of plasmid R1 56 .

Production of toxins causes an extensive rearrangement of bacterial physiology. It can inflict dormancy and antibiotic tolerance 57 if the toxin level exceeds a threshold 54 . Fluctuations in toxin levels above and below the threshold have been used to explain the coexistence of dormant and growing cells in a population 54 . The possibility of positive feedback by the generation and selective buildup of the toxin-encoding mRNA fragments may explain this heterogeneity in growth. Therefore, we wanted to evaluate the recovery of single bacteria and test possible growth heterogeneity after over-production of a toxin and the resulting activation of the chromosomal TA loci. We monitored growth resumption by individual cells using dilution of previously synthesized green fluorescent protein (GFP) 58 . The plasmid pTM11 was inserted into the chromosome of BW25311 to allow IPTG-inducible GFP to be expressed, and this strain was transformed with plasmids for L-arabinose-inducible production of toxins RelE, MazF, MqsR and HipA. Expression of GFP was induced for 2.5 h; thereafter, the cells were transferred into medium containing L-arabinose to induce the toxins. After 90 min, the growth medium was changed again to shut down toxin synthesis and allow recovery (Additional file 1: Figure S5). Analysis of the bacterial GFP content by flow cytometry (Additional file 1: Figure S6) showed that after temporary expression of RelE and HipA the bacteria resumed growth rather uniformly, while after expression of MazF and MqsR a subpopulation started to grow with a delay. Thus, expression of these toxins created bistability in a population. Most importantly, all bacteria resumed growth after the transient expression of toxins. Although inhibition by MazF and MqsR was apparently stronger and induced growth heterogeneity, it did not generate a subpopulation of persistently non-dividing bacteria (Additional file 1: Figure S6).

Sequential or simultaneous activation of different TA systems has been reported elsewhere. Transcription of several TA operons was induced in the persister-enriched subpopulation 38 39 . Amino acid starvation in E. coli activated both RelE and MazF (ChpAK) 14 17 . We observed induction of the mqsRA system in response to HipA activation 59 , whereas overproduction of MqsR induced transcription of relBE and relF(hokD) 60 . Also, ectopic expression of VapC toxins originating from Salmonella and Shigella activated YoeB 61 and production of the Doc toxin activated RelE in E. coli 62 . Here, we show that overexpression of several toxins can activate transcription of the other TA operons. Since toxins and TA operons in this study present a random sample, such cross-interactions might be common and be the rule rather than the exception. Consequently, TA systems have a potential to form a cross-activation network, which operates at the transcriptional level (Figure 7). The presence of such network versus lone and uncoordinated TA systems must have an impact on TA activity during the stress response and setup of dormancy.

Induction of the chromosomal relBEF in response to the ectopically produced RelE can be explained by conditional cooperativity (dependence of transcriptional regulation on the T:A ratio) 35 . However, according to our current knowledge, such mechanism is not applicable to cross-induction. Activation of YoeB by VapC depended on Lon protease 61 . Also, Lon was required for induction of TA operons in response to amino acid starvation and chloramphenicol 14 17 18 61 . Our experiments do not provide a solid support for the role of Lon and ClpP in cross-regulation between TA systems of E. coli (Figure 4). Since the cross-induction was present in the knock-out strains, an additional, Lon-, ClpP-, HslV-, and polyphosphate-independent mechanism of regulation must be involved. Unlocking this mechanism remains a task for future studies. The simplest explanation to activation of TA systems would be depletion of antitoxins. It must inevitably happen when protein synthesis decreases. That predicts nonselective induction of all TA operons in response to inhibition of translation, no matter if it is caused by starvation or artificial production of a toxin. Requirement of relBE for transcriptional activation of mazEF during amino acid starvation (Figure 3) contradicts this prediction as well as the lack of mqsRA induction in response to overproduction of MazF and HicA (data not shown). An option for a mechanism of cross-activation is positive feedback regulation due to selective accumulation of toxin-encoding fragments upon mRNA cleavage. As we saw, after cleavage by overproduced toxin, the antitoxin-encoding RNA fragments are rapidly degraded while the toxin-encoding fragments may serve as templates for translation of toxin. Different toxins produce different cleavage products. That can potentially explain why they cause unequal level of trans-activation when overproduced.

Another intriguing issue of TA cross-reaction is the possible cross-inhibition due to non-cognate interactions. Some authors report such cross-reactions 63 64 65 66 67 68 while others have tested but not found them 69 70 . As a part of this study, we examined non-cognate inhibition between E. coli toxins and antitoxins of the RelBE, MazEF, MqsRA, and HipBA systems in vivo. In this attempt, we run into a previously described phenomenon that may become a source of erroneous results. If toxins are expressed from the arabinose-inducible P _BAD promoter and antitoxins from an IPTG-inducible promoter, it is important to consider that IPTG inhibits P _BAD directly 71 . When we used an expression vector that encoded for the IPTG-insensitive C280* version of AraC transcriptional activator, we could not see any cross-inhibition. Based on that, a recent report on functional non-cognate TA interactions in Mycobacterium tuberculosis 67 may require retesting.

In the current study, we found that the cleavage products produced by TA toxins differ in stability. Selective targeting of mRNAs by endoribonucleolytic toxins and different stabilities of the resulting cleavage products may constitute another layer of gene regulation in the bacterial stress response. Differences in half-life and translational efficiency of mRNA cleavage products, along with generation of a pool of ribosomes lacking the anti-Shine-Dalgarno sequence (as shown for MazF 22 ), could profoundly affect the proteome composition. An example of such an effect is the occurrence of a MazF-resistant protein pool in E. coli 72 . The accumulation of toxin-encoding mRNA fragments may have potential use as a marker of toxin activation in studies of stressed and non-growing bacteria. Increase of the T/A ratio may possibly trigger a positive feedback loop consisting of transcriptional activation of the TA operon, successive cleavage of the TA transcript, buildup of the toxin-encoding mRNA fragments, and translation of them, shifting the T/A balance (Figure 7). Thus, it can be related to TA-linked growth heterogeneity in bacterial populations (Additional file 1: Figure S6) 38 39 54 .

All strains and plasmids are listed in Additional file 1: Table S1. Conditions of bacterial cultivation and construction of strains and plasmids are described in Additional file 1: Supporting information.

Procedures for blotting and hybridization are described in 59 . E. coli BW25113 was transformed with two plasmids, one bearing an antitoxin gene and the other bearing a toxin gene. Cultures containing the empty vector plasmids pBAD33 and pOU82 were used for negative controls. When bacteria contained plasmids for toxin expression, the LB medium for overnight cultures was supplemented with 0.2% glucose and 50 μM IPTG (for HicA with 1mM L-arabinose). Overnight cultures were diluted 1000-fold into 200 ml of LB and grown to OD₆₀₀ ≈ 0.2 (for ~ 2.5 h). To induce toxins, 1 mM L-arabinose, 1 mM IPTG (for HicA) or 30 μg ml⁻¹ mupirocin was added. Overnight cultures of BW25113 ΔrelBEF and BW25113 ΔP _relBEF containing plasmids were diluted into LB supplemented with 0.2% glucose and 50 μM IPTG; at OD₆₀₀ ≈ 0.2, bacteria were collected by centrifugation (5 min, 5000g, at 20°C) and resuspended in prewarmed LB supplemented with 1 mM L-arabinose. Total RNA was extracted using two different protocols: in Figures 2, 6 and S3 we used Trizol reagent 59 and in all other experiments we used hot phenol (for details see Additional file 1: Supporting information). Samples of total RNA (10 μg) were subjected to electrophoresis on denaturing gels. The DNA oligoprobes used for hybridization are listed in Table S2 (Additional file 1). For re-hybridization, the membranes were stripped by boiling for 2×10 min in 0.1% SDS, 5mM EDTA. Chemiluminescent signals were captured using ImageQuant RT ECL imager (GE Healthcare) and X-ray film (Agfa).

RNA samples were collected as for northern blotting. Extension primers (Additional file 1: Table S2) were labeled with [γ³²P]ATP by T4 polynucleotide kinase (Thermo Scientific) and purified with a Nucleotide Removal Kit (Qiagen). Total RNA (15 μg) was mixed with labeled primer and incubated at 75°C for 2 min followed by slow cooling for 25 min. Extension reactions were carried out at 44°C for 30 min using 200U of RevertAid^TM H minus reverse transcriptase (Thermo Scientific) and stopped with 10 μl of formamide loading buffer 73 . Reaction products were concentrated by ethanol precipitation before gel electrophoresis. DNA was sequenced using a Sequenase Version 2.0 Kit (USB Products, Affymetrics). A PCR product amplified using primers relBEFup and relFdwn, and treated with Exonuclease I and shrimp alkaline phosphatase (ThermoScientific), was used as the template for the sequencing reactions. Samples were analyzed by 7M urea-6% polyacrylamide gel electrophoresis.

To prepare lysates, bacteria were grown to an OD₆₀₀ of ~0.7 and expression of T7 RNA polymerase was induced for 1 h by adding 1mM IPTG. Control cultures were grown without IPTG. Bacteria were spinned down and lysed in Laemmli sample buffer. Proteins were separated by tricin–SDS–13% polyacrylamide gel electrophoresis 74 . For detection of the His6-tagged toxins, the proteins were electroblotted onto Hybond-ECL nitrocellulose membrane filters (GE Healthcare) and probed with nickel-activated horseradish peroxidase (HisProbe^TM-HRP; Thermo Scientific).

Overnight cultures were grown from fresh single colonies for 17–18 h in LB supplemented with 0.2% glucose and diluted 500-fold, into 3 ml of broth. After 2 h of incubation, 1 mM IPTG was added to initiate synthesis of green fluorescent protein (GFP). Expression of GFP was induced for 2.5 h. Then, cells were collected by centrifugation and transferred into LB supplemented with 0.2% L-arabinose to induce toxin synthesis. During the change of the medium, the culture was diluted 10-fold. After 90 min, the growth medium was changed to LB containing 0.2% glucose to stop the production of toxins, this time with 2-fold dilution. Starting from the induction of toxin synthesis, samples were taken for flow cytometry analysis and OD₆₀₀ measurement. For flow cytometry analysis, the samples were mixed with an equal volume of 30% glycerol in PBS and stored at −80°C pending analysis. After dilution with sterile PBS, the samples were analyzed using an LSRII and a high-throughput sampler (BD) with a laser beam maximum wavelength of 488 nm. The results were analyzed by using FlowJo 7.2.1software.

All growth inhibition (Additional file 1: Figure S1) and growth resumption experiments (Additional file 1: Figure S5, S6) were repeated at least three times. All northern blot (Figures 1, 2, 3, 4, 6 Additional file 1: Figures S2, S3), primer extension mapping (Additional file 1: Figure S4) and in vivo translation experiments (Figure 6) were repeated at least twice with similar results. Typical results are presented for these experiments and for the FACS analysis of growth resumption (Additional file 1: Figure S6).