Prokaryotic toxin–antitoxin (TA) loci consist of two genes in an operon that encodes a metabolically stable toxin and an unstable antitoxin. The antitoxin neutralizes its cognate toxin by forming a tight complex with it. In all cases known, the antitoxin autoregulates TA operon transcription by binding to one or more operators in the promoter region while the toxin functions as a co-repressor of transcription. Interestingly, the toxin can also stimulate TA operon transcription. Here we analyse mechanistic aspects of how RelE of Escherichia coli can function both as a co-repressor and as a derepressor of relBE transcription. When RelB was in excess to RelE, two trimeric RelB2•RelE complexes bound cooperatively to two adjacent operator sites in the relBE promoter region and repressed transcription. In contrast, RelE in excess stimulated relBE transcription and released the RelB2•RelE complex from operator DNA. A mutational analysis of the operator sites showed that RelE in excess counteracted cooperative binding of the RelB2•RelE complexes to the operator sites. Thus, RelE controls relBE transcription by conditional cooperativity.
Toxin–antitoxin (TA) loci are ubiquitous in free-living bacteria and Archaea (Pandey and Gerdes, 2005). A comprehensive search in more than 200 prokaryotic genomes led to the identification of ∼1400 TA loci belonging to seven different gene families (RelE, MazF, VapC, Doc, HicA, CcdB and ParE) that are distributed in a striking pattern: slowly growing free-living organisms have particularly many while obligatory intracellular organisms have few or none. For example, Mycobacterium tuberculosis has more than 60 and Sulfolobus spp. has ∼30 TA loci while Mycobacterium leprae and Rikettsiae have none (Pandey and Gerdes, 2005; Makarova et al., 2006). Most TA loci encode inhibitors of translation. Toxins of two families inhibit translation by mRNA cleavage (i.e. RelE and MazF) and were coined mRNA interferases by the Inouye group (Pedersen et al., 2003; Zhang et al., 2003; Munoz-Gomez et al., 2004; Kamada and Hanaoka, 2005; Christensen-Dalsgaard and Gerdes, 2006). Two other TA families (ccd and parDE) encode inhibitors of DNA gyrase (Bernard and Couturier, 1992; Miki et al., 1992; Jiang et al., 2002). The biological function of chromosome-encoded TA loci has been the subject of considerable debate. In all cases investigated, chromosome-encoded TA loci are induced by nutritional stress and accumulating evidence suggests that TA loci function to adjust rates of translation and replication to a given supply of nutrients from the environment (Buts et al., 2005; Gerdes et al., 2005). In further support of this stress–response hypothesis, transcription of TA loci was induced by heat shock in Sulfolobus solfataricus and exposure to chloroform in Nitrosomonas europaea (Tachdjian and Kelly, 2006; Gvakharia et al., 2007).
The relBE locus of Escherichia coli encodes mRNA interferase RelE that cleaves mRNA positioned at the ribosomal A-site (Christensen and Gerdes, 2003; Pedersen et al., 2003). RelB antitoxin forms a tight complex with RelE and thereby neutralizes its mRNA cleavage activity (Pedersen et al., 2002). RelB autoregulates relBE transcription via binding to the relBE promoter region (Gotfredsen and Gerdes, 1998). RelB alone can repress relBE transcription but repression is greatly enhanced by RelE. Thus, RelE functions as a co-repressor of relBE transcription.
Transcriptional autoregulation has been reported for members of all known TA gene families and seems to be a general characteristic of TA loci (Gerdes et al., 2005). The following common pattern has emerged: the antitoxins bind, via their N-termini, to operators located in the TA locus promoter region and thereby repress transcription. The toxins and antitoxins form tight complexes that bind avidly to the promoter regions. The toxins are co-repressors that do not themselves interact with operator DNA but increase the specific DNA binding affinity of the antitoxins. Thus, TA complexes very efficiently repress TA operon transcription during steady-state cell growth.
In E. coli, amino acid starvation strongly stimulates relBE transcription (Christensen et al., 2001; 2003). As Lon protease degrades RelB in vivo, the metabolic instability of RelB was suggested to explain the strongly increased transcription rate of relBE during amino acid starvation (Christensen et al., 2001). However, the post-starvation level of RelB was reduced by a factor of ∼2 only and these observations did not satisfactorily explain the high relBE transcription rate during amino acid starvation.
Here we show that an mRNA interferase can stimulate transcription of its operon in vivo and, using purified components, present insight into the mechanism of transcriptional derepression. We observe that overproduction of RelE greatly stimulates relBE transcription in living cells and that it is the RelB : RelE ratio rather than the absolute concentrations of the proteins that determine the ON/OFF state of the relBE promoter. The relBE operator (relO) was defined by in vitro footprinting analysis and consists of two operator sites of 6 bp inverted repeats located within each leg of a larger 26 bp inverted repeat. RelB2•RelE heterotrimeric complexes bind cooperatively to the two operator sites in relO. RelE in excess destabilizes binding of the RelB2•RelE complex to operator DNA by a titration mechanism that depends on the RelB binding site of RelE. Transcription analyses confirmed that two operator sites are essential for derepression of transcription in vivo. Thus, we provide strong evidence that RelE in excess counteracts the cooperative binding of RelB2•RelE to relO. This is the first evidence that transcription of a chromosome-encoded TA operon is regulated by conditional cooperativity and supports the notion that this remarkable type of transcriptional regulation is a common feature of TA loci.
High levels of RelE derepress the relBE promoter
To investigate the role of RelE in transcriptional control of relBE, wild-type RelE was overexpressed and the level of endogenously synthesized relBE mRNA measured (Fig. 1A, top left). As described previously, the level of relBE mRNA was very low during steady-state cell growth (−2 min sample). Interestingly, induction of relE conferred a rapid and strong increase in the relBE mRNA level already at 10 min (12-fold). The mRNA level stayed high for about 30 min (15-fold) and then levelled off. Concomitantly, the cellular amounts of RelB and RelE also increased (Fig. 1A, middle left and bottom left). This increase was expected, as the increased relBE mRNA level would lead to an increased synthesis rate of RelB, and RelE was produced ectopically by a co-resident plasmid.
The increased relBE transcription rate could be due to RelE-mediated inhibition of global translation that leads to a decreased level of the unstable RelB antitoxin (Christensen et al., 2001). To test this possibility, we overproduced a RelE variant that carries six C-terminal amino acid changes that inactivate RelE's mRNA cleavage activity (Pedersen et al., 2002). The His6-RelEcs6 non-toxic variant does not inhibit translation but forms a stable complex with RelB (data not shown). Overproduction of His6-RelEcs6 conferred an even stronger increase in the relBE mRNA level (35-fold; Fig. 1A, top right). Consistently, RelB and His6-RelEcs6 accumulated to higher levels (Fig. 1A, middle right and bottom right). Primer extension analysis of the relBE mRNA shown in Fig. 1A revealed 5′ ends corresponding to the relBE promoter. Moreover, the half-life of the relBE mRNA was unaffected by the overproduction of RelE or His6-RelEcs6 (data not shown). Taken together, these results show that overproduction of RelE derepresses the relBE promoter in living cells.
Derepression of the relBE promoter depends on the RelB : RelE ratio
We speculated that derepression of relBE transcription might depend on the RelB : RelE ratio. Therefore we varied the cellular levels of RelB and RelE using a dual plasmid induction system (relB induction with IPTG and his6-relEcs6 with arabinose). Moreover, we used amino acid starvation as a means for RelE-independent induction of relBE (Christensen et al., 2001). As seen before, the relBE mRNA level increased dramatically (> 30-fold) after induction of amino acid starvation (Fig. 1B, top left). The levels of RelB and RelE were in this case too low to be detected by Western analysis (Fig. 1B, middle left and bottom left). Next, we repeated this experiment using cells with an artificially increased level of RelB (Fig. 1B, middle). As seen, overproduction of RelB completely suppressed the induction of relBE transcription by amino acid starvation. RelB was, as expected, highly expressed and decayed after onset of amino acid starvation. Finally, we overproduced RelEcs6 in cells with an elevated RelB level (Fig. 1B, right). Strikingly, induction of RelE partially counteracted the RelB-mediated repression of relBE transcription. As expected, both RelB and RelE were in this case highly expressed. We conclude that the activity of the relBE promoter depends on the RelB : RelE ratio in living cells.
The RelB : RelE ratio controls relBE promoter activity in vitro
To investigate whether the observed derepression of the relBE promoter depended solely on RelB and RelE or if other factors were involved, we monitored the activity of the relBE promoter in vitro in the presence of different concentrations of purified RelB and RelE proteins using E. coli RNA polymerase and σ70 (Fig. 1C). Without addition of any of the proteins (lane 2) two transcripts (∼0.9 kb) were obtained: one corresponding to a relBEF run-off transcript (open arrow head) and a shorter one, probably terminating at the rho-independent terminator sequence downstream of the relBEF operon (closed arrow head) (Bech et al., 1985). This result confirms that relBE is transcribed from a strong σ70-activated promoter. Addition of RelB alone or of the purified RelB2•RelE complex (see later) repressed the formation of both transcripts (lanes 3 and 4). Addition of increasing (but low) concentrations of RelE resulted in weak but reproducible derepression of transcription (lanes 5–7). At higher concentrations, RelE inhibited the transcription reaction non-specifically (see, for example, lane 8). We were thus unable to use RelB : RelE ratios low enough to obtain full derepression of the relBE promoter in vitro. Nevertheless, we can conclude that the relBE promoter responds similarly in vivo and in vitro to changes in the RelB : RelE ratio.
RelE enhances RelB binding to operator DNA in vitro
To investigate how RelE derepresses relBE transcription, we turned to purified components. A relBE promoter–operator fragment of 166 bp (relO166) was incubated with increasing amounts of RelB and subjected to a gel-shift assay. A well-defined, upshifted band could be detected at the highest RelB concentration only (16.6 μM; Fig. 2A, left). RelB did not shift the migration of control DNA (Fig. 2A, right). Due to the relative weak binding, this method did not allow us to determine if the RelB•relO166 interaction involves cooperativity.
The RelB•His6-RelE (henceforth called RelB•RelE) complex was purified using nickel affinity chromatography. When increasing concentrations of the complex was incubated with relO166, a well-defined, shifted band was obtained, even at very low complex concentrations (12.5 nM; Fig. 2B). A plot of the shifted probe fraction versus RelB•RelE complex concentration yielded a sigmoid curve and could be fitted by a Hill-type equation (Fig. S1). The derived Hill coefficient for cooperative binding was 2.2 (Fig. S1, insert). Thus binding of the RelB•RelE complex to relO166 is cooperative and therefore most likely involves more than one binding site.
The RelB : RelE ratio controls RelB2•RelE•relO complex formation in vitro
Next, we varied the RelB : RelE ratio in the gel-shift assay. To this end we incubated fixed concentrations of RelB and relO166 with increasing concentrations of RelE. As seen in Fig. 2C, neither RelB alone nor RelE alone formed shifted complexes at 0.2 μM concentrations. When the proteins were mixed at a RelB : RelE molar ratio of 16:1 (RelB at 0.2 μM), two protein–DNA complexes were observed (bands 1 and 2). As the ratio of RelB : RelE approached 2:1, the faster migrating band (2) disappeared while the slower migrating band (1) increased in intensity. At a ratio of 2:1, only one intense band (1) was present. With a further increase in RelE, the slower migrating complex (1) disappeared and the free probe DNA reappeared. Results presented later indicate that the intense band (1) formed at the 2:1 ratio most likely reflects two RelB2•RelE complexes bound to operator DNA. The faint bands (2) and (3) may reflect single RelB2•RelE complexes bound to relO166 (see below).
To determine if disruption of the slower migrating complex (1) occurred suddenly by a threshold mechanism or gradually by titration, a series of more narrow RelB : RelE ratio changes were analysed (Fig. 2D). In this experiment, the slower migrating complex (1) disappeared gradually, consistent with a titration mechanism (lanes 2–5). The bottom panel in Fig. 2D shows a quantification of the band-shift experiment.
The reversibility of the RelE-mediated disruption of the RelB2•RelE•relO complex was tested by adding increasing amounts of RelB to the reaction containing RelB and RelE at a ratio of 1.17. Indeed, formation of the nucleoprotein complex was regained when the RelB : RelE ratio became higher than 1.33 (Fig. 2D, lanes 10–12). This result shows that the RelE-mediated disruption of the RelB2•RelE•relO complex does not change the DNA-binding properties of the components involved.
Identification of the relBE operator
The interaction of RelB and RelB2•RelE with relO166 was characterized by footprinting analysis. The DNase I protection pattern obtained with RelB (Fig. 3A, lanes 6–9) was indistinguishable from that obtained with purified RelB2•RelE complex (lanes 11–14), indicating that the interactions between the RelB2•RelE complex and DNA are mediated solely by RelB. RelE alone did not protect any nucleotides against DNase I cleavage (lane 15). Three regions of protection were visible in the region spanning from −10 to +17 (transcriptional start site is +1). Interestingly, this 27 bp region contains an inverted repeat consisting of two 12 bp ‘legs’: [CTTGTAATGACATTTGTAATTACAAG] (matching bp are underlined). Each 12 bp leg contains an imperfect inverted repeat with the 6 bp consensus sequence [T/A]TGT[A/C]A (Fig. 3B). Results presented later show that each leg of relO can bind one molecule of RelB2 or RelB2•RelE. The dimeric nature of the Ribbon–Helix–Helix (RHH) DNA-binding motif present in the N-terminus of RelB2 is consistent with the proposal that two RelB dimers recognize the two 6 bp inverted repeats in relO.
Mixing of RelB and RelE in a 2:1 ratio yielded an identical protection pattern as obtained with purified RelB2•RelE complex (compare lanes 14 and 16). Moreover, protection was lost when the RelB : RelE ratio was decreased to 1:1 or 0.25:1 (lanes 17 and 18), in agreement with the gel-shift analysis presented above.
Excess of RelE disrupts pre-formed (RelB2)2•relO and (RelB2•RelE)2•relO166 complexes
We used surface plasmon resonance (SPR) to further study the interaction of RelB2 and RelB2•RelE with relO. RelB2 or RelB2•RelE complexes were injected at increasing concentrations over relO166 immobilized on a sensor chip and each binding event monitored as a change in resonance units (RU). The maximal binding response for RelB2 was obtained at 400 nM (∼60 RU), while that for RelB2•RelE was obtained at 140 nM (∼108 RU) (Fig. 4A and B). RelE alone did not bind to relO166 and RelB2 did not bind to immobilized control DNA (data not shown). We were unable to derive the association rate constants ka from the binding data, most likely due to a complex cooperative binding mode. However, from the dissociation phase of each experiment, we obtained the dissociation rate constants kd of 3.7 × 10−3 s−1 and 3.9 × 10−4 s−1, corresponding to half-lives of 3 min and 30 min for RelB2•relO and RelB2•RelE•relO complexes respectively. Based on the maximal binding responses in each experiment (i.e. at equilibrium), we calculated the stoichiometry (n) of the complexes (see Experimental procedures). For RelB2•relO, we obtained n = 3.7 (∼4 molecules) of RelB monomers per relO166. For RelB•RelE, we obtained n = 2.0 or n = 1.4, assuming either RelB2•RelE heterotrimer or RelB2•RelE2 heterotetramer stoichiometry respectively. Thus, our SPR data favour a RelB2•RelE stoichiometry in the operator complex. This ratio in the RelB•RelE complex is supported by amino acid ratio analysis of the purified complex (M. Overgaard and K. Gerdes, unpubl. data). Consistently, the gel-shift analyses in Fig. 2 showed full binding of relO166 at a RelB : RelE ratio of ∼2:1. Taken together, our results show that two RelB2•RelE heterotrimers bind cooperatively and with high affinity to relO.
Next, we accomplished three series of SPR experiments with increasing RelE concentrations. In the first series, a constant amount of RelB (50 nM) was mixed with increasing concentrations of RelE. At the lower end of the concentration spectrum, an increase in RelE (0, 10 and 40 nM) led to an increase in binding response and a reduced off-rate in the dissociation phase (compare black, blue and yellow curves in Fig. 4C). However at the highest RelE concentration (160 nM) the binding response was much lower and the complex formed was highly unstable (Fig. 4C, red curve). These results show that the RelE level controls the binding of RelB to relO, consistent with the gel-shift analysis shown in Fig. 2C and D.
In the second series, we investigated if RelE could destabilize RelB2 pre-bound to relO. RelB (200 nM) was injected over the SPR chip to pre-form (RelB2)2•relO166 complexes. Injection of a low concentration of RelE (12.5 nM) in the dissociation phase resulted in a transient mass increase followed by reduced binding (Fig. 4D, blue curve). With higher RelE concentrations, the protein–DNA complexes were rapidly disrupted (Fig. 4D, green, yellow and red curves). These results show that RelE can destabilize a pre-formed (RelB2)2•relO166 complex in vitro.
Finally, 200 nM RelB2•RelE complex was injected over the chip surface to form (RelB2•RelE)2•relO166 complexes. Subsequent injection of RelE in the dissociation phase resulted in a rapid drop ∼2.5-fold in RU (Fig. 4E). These results show that RelE can disrupt the operator complex. The formation of a new, stable complex at ∼40 RU suggests that protein is still bound to relO166, even in excess of RelE. Consistent with the gel-shift analysis (Fig. 2C and D) and results presented below, this could reflect binding of a single RelB2•RelE complex to relO. Nevertheless, our SPR analyses support strongly that when RelE exceeds a certain level relative to RelB, then the binding of RelB2•RelE to relO is dramatically destabilized.
The RelB operator consists of two RelB2 binding sites
We used fluorescence anisotropy to investigate the interaction of RelB2•RelE with relO variants. Double-stranded DNA oligonucleotides of 30 bp with relO (called relO-wt), relO with the upstream RelB2 binding site scrambled (relO-up) and relO with the downstream binding site scrambled (relO-down) were end-labelled with fluorescein (the oligonucleotides are shown in Fig. 3B). Equilibrium measurements using the wild-type relO fragment showed strong affinity binding of the RelB2•RelE complex (Kd = 3.4 nM), whereas the complex displayed lower affinity for relO-down (Kd = 28.9 nM) (Fig. 5A). With relO-up, saturation of binding could not be reached, even at μM ranges of RelB2•RelE. Nevertheless, we can conclude that the relBE operator contains two RelB2 binding sites of which the upstream binds RelB2 with higher affinity than the downstream site.
Both binding sites in relO are required for cooperative binding of RelB2•RelE
To calculate Hill coefficients, we used SPR to obtain equilibrium binding data of RelB2•RelE and RelB to wild-type relO (relO-wt) and relO with a downstream binding site scrambled (relO-down). As seen from Fig. 5B, RelB2•RelE bound with high affinity to wild-type relO and we obtained a Hill coefficient nH of ∼2.2, in good agreement with the Hill coefficient determined by the gel-shift analysis (nH ∼ 2.3). In contrast, RelB2•RelE bound to relO-down with reduced affinity (curve right-shifted) and reduced cooperativity (nH ∼ 1.5). The residual cooperativity of binding to relO-down may be explained by the recruitment of a second RelB2•RelE complex to the scrambled sequence present in relO-down. Compared with RelB2•RelE, RelB2 bound with reduced affinity and cooperativity to both wild-type and mutated relO targets. We obtained Hill coefficients of ∼1.3 and ∼0.9, respectively, indicating that RelB2 binds with weak cooperativity to relO and that this cooperativity depends on both binding sites in relO.
RelE-mediated disruption of RelB2•RelE•relO depends on cooperativity
The results presented above raised the possibility that RelE in excess would interfere with cooperative binding of RelB2•RelE to relO. Therefore, we compared the binding properties of relO-wt DNA containing both RelB2•RelE binding sites, and a truncated derivative, relO-Δdown, which contained the upstream binding site only (Fig. 3B). The latter target DNA was used to avoid any contribution to cooperativity that possibly could originate from non-specific binding of RelB2•RelE to downstream DNA. As before, RelB2•RelE bound relO-Δdown with high affinity, even at very low concentrations (Fig. 6A). RelB2•RelE bound with lower affinity to relO-Δdown, and with a very high dissociation rate (Fig. 6C). To investigate how these fragments responded to excess RelE, 100 nM or 400 nM RelB2•RelE complex was injected continuously for 120 s, reaching an equilibrium binding response of approximately 130 RU for relO-wt (Fig. 6B) and 35 RU for relO-Δdown (Fig. 6D). When the inlet flow was shifted from RelB2•RelE to RelB2•RelE plus increasing concentrations of RelE, a sudden drop to ∼40% of the initial response was obtained for relO-wt (Fig. 6B). Strikingly, however, increasing RelE concentrations had the opposite effect on the binding of RelB2•RelE to relO-Δdown. Thus, RelE counteracts RelB2•RelE binding only when both binding sites of relO are present. Furthermore, when the flow was immediately changed to running buffer (red lines in Fig. 6B and D), dissociation occurred with strikingly different kinetics: the RelB2•RelE complex formed with relO-Δdown was much more unstable than that formed with relO-wt. This result shows that both relO binding sites are required for RelE-mediated destabilization of the stable (RelB2•RelE)2•relO166 complex. Thus, RelE disrupts the operator complex by abolishing cooperative binding of RelB2•RelE to relO.
C-terminus of RelB counteracts RelE-mediated destabilization of RelB2•RelE•relO
Recently we showed that a C-terminal peptide of RelB that lacks the RHH DNA-binding and dimerization domain, RelB52−79, was sufficient to counteract RelE toxicity in vivo (Cherny et al., 2007). We inferred from that result that the RelB52−79 fragment interacts with RelE and thereby abolish RelE activity. Here we expressed and purified the RelB52−79•RelE complex and compared its ability to disrupt a pre-formed RelB2•RelE•relO complex with that of free RelE (Fig. 2E). In case of RelE, complex formation was lost when the RelB : RelE ratio became lower than 1.33 (Fig. 2E, left), in agreement with experiments described above. In contrast, RelB52−79•RelE did not destabilize RelB2•RelE•relO (Fig. 2E), even when its concentration was raised a further fivefold (data not shown). Thus, the C-terminus of RelB that binds to RelE is sufficient to block RelE's ability to destabilize the operator complex. This result suggests that RelE destabilizes the RelB2•RelE•relO complex by interacting with a C-terminal end of a RelB monomer present in the operator complex.
Operator mutations render PrelBE insensitive to RelE in vivo
To determine the effect of operator mutations on transcriptional regulation in vivo, we constructed lacZ fusions to wild-type or mutant operators carrying in cis wild-type relB and an allele of relE, relER81A (Fig. 7). Like RelEcs6, RelER81A forms a stable complex with RelB but its overexpression does not block translation (Pedersen et al., 2002). The activity of the wild-type promoter was very low (10 ± 2.5 U) consistent with transcriptional repression by RelB2•RelER81A. In this strain, RelB overexpression caused a further drop in promoter activity (6 ± 1 U). In contrast, overexpression of the non-toxic RelEcs6 variant led to a 14-fold increase in transcriptional activity (146 ± 23 U). This result is in agreement with the Northern analysis shown in Fig. 1A.
The two plasmid constructs lacking either the upstream or downstream binding site of relO exhibited highly increased basal transcription rates (42- to 47-fold). This was expected as DNA fragments carrying these mutations bind the RelB2•RelE complex less strongly in vitro (Fig. 5A). Importantly, RelE donated in trans did not significantly derepress transcription of the plasmids carrying mutated relO binding sites. The mutations in pMO4005 and pMO4006 did not significantly reduce relBE promoter activity and reduced PrelBE activity therefore cannot explain the lack of transcriptional induction by RelE (data not shown). Thus, the two RelB2 binding sites of relO are required both for repression and for derepression of the relBE promoter in vivo.
Previous investigations showed that RelB is a repressor of relBE transcription and that RelE functions as a co-repressor such that the relBE promoter is efficiently repressed during steady-state cell growth (Fig. 1A) (Gotfredsen and Gerdes, 1998; Christensen et al., 2001). We show here that overexpression of RelE dramatically induced relBE transcription and that the RelB : RelE ratio rather than the absolute RelE level controls the relBE transcription rate both in vivo and in vitro (Fig. 1). RelB is a dimer that binds to the relBE promoter region via its RHH domain (M. Overgaard and K. Gerdes, in preparation). The RelB2 binding sites in the promoter region are located within a 27 bp region, the relO operator, which contains a large, 26 bp imperfect inverted repeat to which two RelB2 dimers bind (Figs 3 and 5). At RelB : RelE ratios higher than 2:1, RelE dramatically increased cooperative binding of RelB2 to relO and the stability of the RelB2•relO complex (Figs 2–4). Our data indicate that the repressor complex consists of two RelB2•RelE heterotrimers bound to relO. Verification of this inference requires structural analyses of the nucleoprotein complex. Binding of RelB2 without RelE was only weakly cooperative (Figs 2 and 5).
At RelB : RelE ratios lower than 2:1, RelE did not enhance binding of RelB2 to relO. Moreover, RelE in excess disrupted a pre-formed repressor complex (Figs 2C, D and 4E). Two distinct SPR experiments indicated that RelE disrupted a pre-formed RelB2•relO complex: dissociation of RelB2•RelE at high RelE concentration (Fig. 4C, red line) occurred faster than dissociation of RelB2 (Fig. 4A, red line), and RelE added to RelB2•relO led to rapid dissociation of RelB2 (Fig. 4D).
Each leg of relO can bind RelB2 (Fig. 5) and contains a 6 bp inverted repeat that is potential RelB2 recognition element (Fig. 3B). A relO target with one scrambled recognition element bound RelB2•RelE with reduced cooperativity, indicating that cooperative binding of the complex depends on both binding sites (Fig. 5B). Destabilization of the (RelB2•RelE)2•relO complex also depended on the presence of two RelB2•RelE binding sites (Fig. 6). Thus, destabilization of the repressor complex depends on cooperative binding of RelB2•RelE to relO. This proposal is consistent with the finding that both RelB2•RelE binding sites in relO were required for RelE-mediated derepression of relBE transcription in vivo (Fig. 7). In fact, RelE in excess did not stimulate transcriptional activity of the two relBE promoter derivatives containing one binding site. Thus, the RelE-mediated increase in binding of RelB2•RelE to a short DNA fragment containing the upstream relO binding site only (Fig. 6D) may not be relevant to gene regulation in vivo. In any case, the understanding of this phenomenon will require further analysis.
Transcriptional triggering by excess of toxin is general
The complexes of the plasmid-encoded TA loci ccd of F, parD of R1, parDE of RK2 and phd/doc are all destabilized by excess toxin in vitro (Johnson et al., 1996; Magnuson and Yarmolinsky, 1998; Afif et al., 2001; Madl et al., 2006; Monti et al., 2007). The results presented here show that this is also the case for the chromosome-encoded relBE locus and we suggest that toxin-mediated destabilization of TA•operator complexes is a general phenomenon. Our observations further raise the possibility that an excess of the toxins in general counteracts cooperative binding of TA complexes to their cognate operators.
As in the case of relBE, the toxins in excess may interact with the TA•operator complexes via their unoccupied antitoxin binding sites and thereby destabilize cooperative interactions between adjacent complex units (Fig. 8). Thus, even though the architectures of the operator–DNA complexes may vary, conditional cooperativity may provide a general mechanism to derepress TA operon transcription.
Mechanism of (RelB2•RelE)2•relO complex disruption
We showed previously that the C-terminal end of RelB interacts with RelE (Cherny et al., 2007). We tested the possibility that RelE recognizes a C-terminus of RelB2 in the repressor complex by adding RelB52−79•RelE to a pre-formed (RelB2•RelE)2•relO complex (Fig. 2E). In this case, RelE did not destabilize the repressor complex, indicating that extra RelE molecules destabilize the (RelB2•RelE)2•relO complex by interacting with its RelB C-termini. The schematic in Fig. 8 shows a model of how this might occur at the molecular level. Thus the model assumes that RelE has a low- and a high-affinity binding site for RelB. When the promoter is repressed, each RelE interacts with the C-termini of two different RelB2 molecules bound to relO and stabilizes RelB2 binding. With RelE in excess, the RelE will invade the repressor complex via interaction with its high-affinity RelB binding site. This interaction somehow obliterates cooperative binding of the RelB2•RelE complex to relO. One possible way this could occur is that a RelB2•RelE2 complex forms and that this complex binds weakly to relO. However, we have not yet been able to detect such a complex, neither by gel filtration nor by chemical cross-linking (data not shown). The gradual disruption of the repressor complex seen with increasing RelE concentrations (Fig. 2D) is, however, consistent with the proposal that RelE releases the complex by a titration mechanism (one molecule of RelB2•RelE released per RelE molecule in excess) rather than by a cascade mechanism (many molecules of RelB2•RelE released per RelE molecule). It should be noted that the crystal structure of an archaeal complex revealed a RelB2•RelE2 stoichiometry (Takagi et al., 2005).
A very similar regulatory pattern was seen with the parD locus of plasmid R1 that encodes the Kis/Kid TA couple (Monti et al., 2007). With excess Kis antitoxin, a (Kis2•Kid2)2 complex bound cooperatively to operator DNA. In contrast, with excess of Kid toxin, a multimeric (Kid2•Kis2•Kid2)n complex that bound weakly to operator DNA, was observed. Thus even though different stoichiometries and operator complexes are formed, excess of the toxins seem to induce TA operon transcription in most, if not all TA loci. We infer that transcriptional triggering has evolved independently in unrelated TA families, suggesting that this property is fundamental to the biological function of TA loci.
Biological functions of toxin-mediated transcriptional triggering of TA loci
The chromosome of E. coli K-12 codes for four relBE-like and two mazEF-like loci (Gerdes et al., 2005; Schmidt et al., 2007) and several putative TA loci (Makarova et al., 2006; Sevin and Barloy-Hubler, 2007). A seminal observation was that nutritional stress, such as amino acid starvation, induces strong transcription of TA loci. Consistently, amino acid starvation also induces RelE and MazF mRNA cleavage activity and thereby reduces the global level of translation (Christensen et al., 2001; 2003). Lon protease degraded RelB in vivo, suggesting that the metabolic instability of RelB could explain the strongly increased transcription rate of relBE during amino acid starvation. However, the post-starvation level of RelB was reduced by a factor of ∼2 only (Fig. S2A). The persistent level of RelB during amino acid starvation was difficult to reconcile with the concomitant high rate of relBE transcription.
The transcriptional trigger mechanism described here yields a possible explanation of these apparent inconsistent observations. As RelB is metabolically unstable, amino acid starvation obviously leads to a transiently reduced RelB level that, in turn, increases PrelBE activity (Christensen et al., 2001). The resulting increase in relBE mRNA level increases the synthesis rates of both RelB and RelE and without further regulatory components the increased level of RelB would rapidly shut down relBE transcription. Obviously, this was not what we observed (Fig. 1B). One possible explanation for the sustained high relBE transcription level is that the post-starvation ratio of RelB : RelE is reduced. This proposal is supported by our overexpression data shown in Fig. 1B. Here, the RelB level decreased dramatically during amino acid starvation while the RelE level stayed constant. By inference, the transcriptional trigger mechanism ensures a sustained high relBE transcription rate during starvation even with RelB being present at a significant level. A final proof of this proposal requires measurements of the RelB : RelE ratios before and after amino acid starvation without overproduction of the proteins. So far we have not been able to do this because of the very low levels of the proteins. However, in cells harbouring the relBE operon on a medium-copy-number plasmid, RelE concentration remained constant during starvation (data not shown). As RelB decays during amino acid starvation, we infer that the RelB : RelE ratio indeed changes during starvation.
Here we have presented evidence that RelE of E. coli can trigger transcription of its operon by counteracting cooperative binding of RelB2•RelE to operator DNA and that this property is relevant to the physiological function of relBE during nutritional stress. The finding that transcription of both plasmid- and chromosome-encoded TA loci can be derepressed by excess toxin suggest that the mechanism elucidated here is general. Moreover, the transcriptional trigger mechanism may enable individual bacterial cells to respond fast and efficiently on environmental changes such as amino acid starvation, carbon starvation and heat shock, a hypothesis that we now aim to test.
Media, antibiotics, strains and plasmids
Luria–Bertani (LB) and 2× YT medium were prepared as described (Sambrook et al., 1989). When required, the medium was supplemented with 30 or 100 μg ml−1 ampicillin, 50 μg ml−1 chloramphenicol, 50 μg ml−1 kanamycin and/or 10 μg ml−1 tetracycline. Indicator plates were added Xgal (5-bromo-4-chloro-3-indoyl-β-d-galactoside) to a final concentration of 40 μg ml−1. Expression of proteins from the PA1/O4/O3 or PBAD promoters was induced by 1 mM isopropyl-β-d-thiogalactopyranoside (IPTG) or 0.2% arabinose respectively. Strains and plasmids are listed in Table 1 and oligonucleotides in Table S1.
Table 1. Bacterial strains and plasmids used in this study.
aphA: kanamycin resistance gene; tet: tetracycline resistance gene; cat: chloramphenicoltransacetylase gene; bla: β-lactamase gene. PBAD denotes the arabinose-inducible promoter of the pBAD series of plasmids (Guzman et al., 1995).
pMO2203. The bla gene, contained on a 1.85 kb AvaI fragment, was excised from pOU254, treated with Klenow polymerase and ligated to similar blunt-ended pMG2203 opened with BamHI. The resulting plasmid confers ampicillin resistance.
pMGJ25. The cat gene, contained on a PvuI fragment, was ligated to pMG25 opened with PvuI in the bla gene. The resulting plasmid is thus a cat derivative of pMG25.
pMO3103. A PCR fragment of relEcs6 was amplified by PCR using primers pBAD33-BamHI-KpnI-f and BAD-CCW using pKP3103 as template The resulting fragment was digested with BamHI and HindIII and inserted into BamHI- and HindIII-digested pMGJ25, yielding pMO3103.
pMO331. The relB gene was amplified by PCR using primers relB1-XbaI and relB2-HindIII-SalI using pBD2430 as template. The resulting fragment was digested with XbaI and HindIII and inserted into XbaI- and HindIII-digested pBAD33, yielding pMO331.
pMGJ4004. A DNA fragment of relBER81A was generated by three rounds of PCR using primers: pOU254-CW and relEmut81f with pKG4003 as template (PCR 1); pOU254-CCW and relEmut81r with pKG4003 as template (PCR 2); pOU254-CW and pOU254-CCW with PCR1 and PCR2 as template (PCR 3). The resulting 0.8 kb PCR fragment was digested with BamHI and EcoRI and inserted into BamHI- and EcoRI-digested pOU254, yielding pMGJ4004.
pMGJ4004. A DNA fragment encoding relBER81A was generated by three rounds of PCR using primers: pOU254-CW and relEmut81f with pKG4003 as template (PCR 1); pOU254-CCW and relEmut81* with pKG4003 as template (PCR 2); pOU254-CW and pOU254-CCW with PCR1 and PCR2 as template (PCR 3). The resulting 0.8 kb PCR fragment was digested with BamHI and EcoRI and inserted into BamHI- and EcoRI-digested pOU254, yielding pMGJ4004.
pMO4004 relO-up. To mutate the upstream operator site in pMGJ4004, three rounds of PCR were performed using primers: rel-80f-EcoRI and relO-mut-up-r2 on pMGJ4004 (PCR 1); relO-up-f and relB2-XhoI-r on pMGJ4004 (PCR 2); rel-80f-EcoRI and relB-XhoI-r on PCR 1 + 2. The resulting 0.23 kb fragment (PCR3) was inserted into pMGJ4004 using the EcoRI and XhoI restriction sites.
pMO4004 relO-down. To mutate the downstream operator site in pMGJ4004, three rounds of PCR were performed using primers: rel-80f-EcoRI and prel-mut-down-r on pMGJ4004 (PCR 1); relO-mut-down-f and relB-XhoI-r on pMGJ4004 (PCR 2); rel-80f-EcoRI and relB-XhoI-r on PCR 1 + 2. The resulting 0.23 kb fragment (PCR 3) was inserted into pMGJ4004 using the EcoRI and XhoI restriction sites. All PCR-generated plasmids were verified by DNA sequencing.
Protein expression and purification
RelB and His6-RelE proteins were purified in complex and separated as described (Cherny et al., 2007). To determine the concentrations of RelB, His6-RelE and the RelB2•His6-RelE complex, the proteins were subjected to amino acid analysis according to Thaysen-Andersen et al. (2007). The extinction coefficients were calculated to be 4924 M−1 cm−1, 59 345 M−1 cm−1 and 20 666 M−1 cm−1 for RelB, His6-RelE and the RelB2•His6-RelE complex respectively.
Overnight cultures were diluted to OD450 ∼ 0.005 in LB with selection and grown at 37°C. When the cultures reached an OD450 of 0.05 appropriate inducer was added and the cultures continued to grow to OD450 ∼ 0.3. Amino acid starvation was induced by addition of 1 mg ml−1 serine hydroxamate and protein synthesis was inhibited by the addition of 100 μg ml−1 chloramphenicol. For protein half-life measurements, proteolysis was terminated by the addition of 10% ice-cold TCA followed by 2 h incubation on ice. Cells were pelleted by centrifugation, concentrated in sample loading buffer to OD450 = 5 and boiled at 95°C for 5 min with rotation. Proteins were separated on 10–20% gradient SDS polyacrylamide gels (Cambrex) and transferred to an Immobilon PVDF membrane (Millipore) using a semi-dry blotting apparatus (Pharmacia) and probed with primary antibodies raised against RelB or His6-RelE at 1:2000 or 1:5000 dilutions respectively. Following washing in TTBS, 2% dry milk the membrane was incubated with secondary antibodies for 1 h with horseradish-conjugated goat anti-rabbit immunoglobulin G (DAKO). Specific proteins were visualized by incubating the membrane with ECLplus (GE Healthcare) to generate a chemiluminescence signal. The membrane was immediately exposed to a BioMax MS film (Kodak) in a film cassette for a short amount of time and subsequently developed manually in the dark.
Northern blotting analysis
Cells were grown in LB at 37°C. At an OD450 of 0.05 the cultures were induced by addition of 0.2% arabinose and harvested at an OD450 of 0.5. For Northern analysis, total RNA was fractionated by PAGE (4.5% low-bis acrylamide), blotted to a Zeta-probe nylon membrane and hybridized with a single-stranded 32P-labelled riboprobe, complementary to relB. The radiolabelled probe was generated using linearized plasmid DNA of pMG421.
In vitro transcription with E. coli RNA polymerase
Multiple-round transcription of the relBEF operon was carried out using a 1 kb EcoRI–BamHI restriction fragment of pMG1904 (Table 1) as template. This DNA fragment contains the entire relBEF operon including promoter and terminator sequences. Each reaction was set up in a total volume of 20 μl in transcription buffer [40 mM Tris-HCl (pH 7.9 at 25°C), 6 mM MgCl2, 2 mM spermidine, 10 mM NaCl] supplemented by 1 mM DTT, 0.2 mM ATP, GTP and UTP, 10 μM CTP, 1 μCi of [α-32P]-CTP, 1 nM relBEF DNA fragment and RelB, RelE and RelB•RelE at the concentrations indicated in the legend to Fig. 1. All reactions were pre-incubated for 20 min at 37°C and transcription was initiated by addition of 28 nM RNA polymerase holoenzyme (Epicentre Biotechnologies). After 30 min, the reactions were terminated by addition of an equal volume of formamide loading buffer and subsequent boiling at 95°C for 5 min. Finally, the resulting transcripts were separated on a 6% denaturing acrylamide gel and subjected to phosphoimaging.
Electrophoretic mobility shift assay and DNase I footprinting
Purified proteins were incubated with 2–5 nM 32P- or Cy5-labelled DNA probe at 37°C for 20 min in a 20 μl standard reaction of 1× binding buffer (20 mM Tris-HCl pH 7.5; 100 mM KCl; 2 mM MgCl2, 1 mM DTT, 50 μg ml−1 BSA) and 0.1 mg ml−1 sonicated salmon sperm DNA (ssDNA). For EMSA analysis, samples were added 1× loading buffer (10% glycerol in binding buffer) and subjected to native PAGE on 5% acrylamide gels in 0.5× TBE, followed by phosphoimaging or direct fluorescent scanning on a Typhoon Trio scanner (GE Healthcare). Gel-shift data of Fig. 2B were quantified using ImageQuant 5.0 (Molecular Dynamics) and the fraction of relO166 shifted was plotted versus the total protein concentration. The binding isotherm was fitted to Eq. 1 using Sigmaplot 8.02:
where AB/Atotal represents the fraction bound, measured at a given total protein concentration B, and fmax is the fraction bound at saturating B. The Hill coefficient nH was determined using non-linear least squares analysis.
For DNase I footprinting, samples were incubated with 2 U DNase I and 1× DNase I buffer (Ambion) at 37°C for 2 min followed by addition of 100 μl of Stop buffer (2 M ammonium acetate, 20 mM EDTA, 10 mg ml−1 ssDNA). The resulting fragmented DNA was extracted once in phenol, once in chloroform, precipitated in ethanol and separated on an 8% denaturing acrylamide gel along with a dideoxy sequencing ladder. The gel was dried and finally subjected to phosphoimaging.
Surface plasmon resonance analysis was carried out on a Biacore 3000 instrument (Biacore AB) with streptavidin-coated SA sensor chips. In short, 70 RU of a 60 bp double-stranded oligonucleotide, containing a 5′biotin moiety in the forward strand, was captured on flow cell 2 by injecting running buffer (150 mM KCl, 4 mM MgCl2, 1 mM DTT, 0.005% Tween 20, 20 mM HEPES, pH 7.5) supplied with 0.5 M NaCl, whereas flow cell 1 was left blank for reference subtraction. A continuous flow of running buffer at 10 μl min−1 was used to create a stable base line over the two flow cells. RelB or RelB•RelE complex, diluted to the specified concentration in running buffer, was then injected over the two flow cells for 2 min. To release RelB or RelB•RelE complex from the immobilized DNA, the flow was increased to 40 μl min−1 and two 15 s pulses of 6 M guanidinium hydrochloride were injected over both flow cells leaving the immobilized DNA ready for another binding cycle. The binding data were fitted using BIAevaluation 3.1 software to obtain rate constants.
As the SPR response is proportional to mass bound to the sensor chip, the relative numbers of RelB monomers or RelB•His6-RelE complexes binding per immobilized rel DNA molecules can be calculated according to:
where Ranalyte is the SPR response caused by the analyte (RelB or RelB•His6-RelE complex) binding to the immobilized ligand (rel DNA), Rligand is the response contributed by the immobilized ligand, and Mrligand and Mranalyte are the molecular weights of the ligand and analyte respectively. By substituting the following values into Eq. 1: R(RelB) = 60 RU; R(rel DNA) = 70 RU; Mr(RelB) = 9071.5 g mol−1; Mr(rel DNA) = 39 600 g mol−1, we obtain n = 3.7, thus the number of RelB monomers per rel DNA is close to 4. When using Ranalyte = 108 RU we obtain for the RelB•His6-RelE complex n = 2.0, assuming a 2:1 RelB : RelE trimer stochiometry (30 154.0 g mol−1). Thus, at a RelB concentration of 400 nM ∼4 monomers of RelB bind to rel operator DNA and at 140 nM RelB2•RelE ∼2 trimers of RelB2•RelE complex binds to rel operator DNA. Hill coefficients were obtained by determination of the slopes of log[Y/(1 − Y)] plotted as a function of log[protein], where Y is fractional saturation of the binding response at each protein concentration at equilibrium and protein was either RelB as monomer or RelB2•RelE as heterotrimer.
Fluorescence polarization experiments were performed using a LS 55 Fluorescence Spectrometer (Perkin Elmer). One strand of each DNA duplex contained a 5′ fluorescein label, and the increase in polarization of the fluorophore upon protein binding was measured. Oligonucleotide pairs are listed in Table S1. Each pair of complementary fragments was diluted in 10 mM Tris pH 8, 100 mM NaCl to a 5 μM concentration, annealed following denaturation at 95°C for 5 min, and slowly cooled to ambient temperature over a period of 60 min. Each binding experiment contained 2 nM fluorescein-labelled duplex DNA in binding buffer (20 mM Tris-HCl pH 7.5; 100 mM KCl; 2 mM MgCl2, 1 mM DTT). For each set of experiments the RelB2•RelE complex was titrated into the DNA. After each addition of protein, the samples were incubated for 60 s to reach equilibrium before a measurement was taken. Each titration was repeated at least three times. To create a baseline, RelB2•RelE complex up to 2500 nM was titrated into DNA with both binding sites being scrambled. These values were subtracted from values obtained using relO-Δdown, relO-up and relO-down. Each binding isotherm was fit to Eq. 2 using Sigmaplot 8.02:
where A is the normalized anisotropy measured at a given total protein concentration and Amax is the maximum anisotropy of specifically bound DNA. Non-linear least squares analysis was used to determine Amax and Kd.
Overnight cultures were diluted to an OD450 ∼ 0.005 in LB supplemented with 30 μg ml−1 Amp and 50 μg ml−1 chloramphenicol and grown at 37°C to an OD450 of 0.3–0.5. relBE promoter activity was measured by monitoring β-galactosidase activity from pMGJ4004 and derivatives using the chloroform protocol variant as described (Miller, 1972). The activity of each test-strains was subtracted that of the control strain carrying the pOU254 vector plasmid assayed in parallel. All assays were performed in duplicate and each experiment was carried out three times to ensure reproducibility of the data. The stability of the R1 plasmids containing the relBE–lacZ fusions were tested by plating on LA plates with 40 μg ml−1 Xgal. The R1 plasmids were all found to be stable during the experiments.
Note added in proof
A recent publication showed that RelB dimers recognize the hexad repeats in the relBE operator through a ribbon–helix–helix motif (Guang-Yao Li, Yonglong Zahng, Masayori Inouye and Mitsuhiko Ikura. Structural mechanism of transcriptional autorepression of the E. coli RelB/RelE antititoxin/toxin module. J Mol Biol (2008); doi:10.1016/j.jmb.2008.04.039).
We thank Remy Loris, Dagmar Iber, Jakob Møller-Jensen and the members of the Gerdes group for stimulating discussions, Jan-Willem Veenig for critical comments to the manuscript and Pia Hovendal for technical assistance. This work was supported by The Danish National Research Foundation via the Centre for MRNP biogenesis and metabolism.