The regulation of the Escherichia coli mazEF promoter involves an unusual alternating palindrome

The Escherichia coli mazEF system is a chromosomal "addiction module" that, under starvation conditions in which guanosine-3',5'-bispyrophosphate (ppGpp) is produced, is responsible for programmed cell death. This module specifies for the toxic stable protein MazF and the labile antitoxic protein MazE. Upstream from the mazEF module are two promoters, P(2) and P(3) that are strongly negatively autoregulated by MazE and MazF. We show that the expression of this module is positively regulated by the factor for inversion stimulation. What seems to be responsible for the negative autoregulation of mazEF is an unusual DNA structure, which we have called an "alternating palindrome." The middle part, "a," of this structure may complement either the downstream fragment, "b," or the upstream fragment, "c". When the MazE.MazF complex binds either of these arms of the alternating palindrome, strong negative autoregulation results. We suggest that the combined presence of the two promoters, the alternating palindrome structure and the factor for inversion stimulation-binding site, all permit the expression of the mazEF module to be sensitively regulated under various growth conditions.


Introduction
In Escherichia coli (E. coli) programmed cell death is mediated through unique genetic elements called "addiction modules." These consist of two genes, where the second gene specifies for a stable toxin, and the first gene specifies for a labile antitoxin.
"Addiction modules" were first discovered in a number of extrachromosomal elements where they were found to be responsible for the postsegregational killing effect, that is, the death of cells from which these extrachromosomal elements have been removed. In other words, these cells are All known extra-chromosomal addiction systems have been shown to be negatively auto-regulated at the level of transcription. For example, such modules as ccdAB of the F factor [5; 6; 7], parD of the plasmid R1 [8], pemIK of the plasmid R100 [9], or phd-doc of the plasmid P1 [10; 11]. Magnuson and Yarmolinsky [11] suggested that the auto-regulation of addiction modules might prevent fluctuations in the levels of the antidote and the toxin that would result in the activation of the toxin.
During auto-regulation, both the toxin and the antidote bind to a palindrome sequence in their own promoter region thereby decreasing their own transcription. In a few cases, the binding of antitoxin by itself resulted in a low level of autoregulation; the concomitant binding of the toxic element increased the level of binding [8; 10]. In more complicated cases, as has been found for the pemIK [9] and phd-doc [11] modules, the promoter region of addiction module contains two separate palindrome sequences. Based on the stoichiometry and dynamics of binding, Magnuson and Yarmolinsky [11] suggested a model in which the palindrome sequence binds the antidote dimer independently but cannot bind the toxin. When the toxin interacts with the antidote it increases the binding affinity of the antidote to the palindrome sequence, and thus increases half-life of the complex. Pairs of genes homologous to some of the extra-chromosomal "addiction modules" have also been found on the E. coli chromosome [12; 13; 14; 15; 16]. As we have reported previously, the E. coli mazEF system is the first known regulatable prokaryotic chromosomal "addiction module" [12]. This system consists of the two genes, mazE and mazF, that are located in the rel operon downstream from the relA gene [16]. We found [12] that the mazEF gene pair has all the properties required for an addiction module. MazF is toxic and long-lived, while MazE is antitoxic and short-lived. MazE and MazF are co-expressed and they interact. In addition, the mazEF system has a unique property: its expression is regulated by guanosine-3',5'bispyrophosphate (ppGpp), which is synthesized by the RelA protein under conditions of amino acid starvation [17]. Furthermore, overproduction of ppGpp induces mazEF mediated cell death [12; 18]. These properties suggest that the mazEF module may be responsible for programmed cell death under conditions of nutrient starvation [12].
Here we studied the regulation of the expression of the mazEF system. The promoter region of the chromosomally borne mazEF addiction module is partially homologous to that of the promoter of the pemIK plasmid borne addiction module [14]. The promoters of both modules contain similar palindrome sequences, though, like the promoter of the phd-doc module [11], the pemIK promoter includes two separated palindromes [9] and the mazEF promoter was found to include only one [14]. Since pemIK is auto-regulated, Masuda and colleagues [14] hypothesized that mazEF might also be auto-regulated.
Results of our previous in vitro work [12] revealed that the chromosomal "addiction module" mazEF can be expressed from two promoters, P 2 and P 3 , which are located 13 bp apart. In in vivo studies, we found that, P 2 , the upstream promoter, was active in exponentially growing cells. Here we showed that the in vivo activity of the P 3 promoter is only one tenth of that of the P 2 promoter.
We found that the mazEF system is weakly auto-regulated by the antitoxic component MazE, and efficiently auto-regulated by the combined action of the Construction of plasmids bearing mazE or mazEF under the control of p tac .
Using appropriate DNA primers (EE-1 and EF-2 for mazE gene, or EE-1 and FG-1 for mazEF genes, see Table 2), we synthesized PCR fragments bearing the ORF of the mazE or the mazEF genes. These PCR products were used for cloning the corresponding genes under the tac promoter present on the expression vector pKK223-3. We called the resulting plasmids pKK-mazE and pKK-mazEF (Table 1).
We also cloned these PCR fragments into the modified pSK10∆6 compatible plasmid pLex1 that bears the IPTG inducible promoter p tac [20]. We could not use the plasmids of the pKK set for testing the influence of MazE and MazF proteins on their promoter because both pKK223-3 and pSK10∆6 are derivatives of the plasmid pBR322 and they are not compatible. Using pLex1 as our parent plasmid, we constructed the plasmids pLex-mazE and pLex-mazEF by introducing the promoter p tac , the chloramphenicol resistance gene, the p15A replication origin, and either mazE or mazEF such that they were under the control of the p tac promoter. We used these plasmids to study the influence of the proteins MazE and MazE-MazF on the mazEF promoter when it was present on the pSK10∆6-p ef plasmid. Proteins expressed either from the p tac promoter bearing plasmid pKK223 or from the chromosome of E.coli strain MC4100 were analyzed by electrophoresis on denaturing and native gels and by Western blot analysis using antibodies raised against MazE ( Fig. 1A and 1C) and MazF (Fig. 1B). As a control, we tested the proteins expressed from the chromosome of MC4100∆mazEF. As we found previously [12], we also found here that MazE and MazF interact directly. We observed no bands of the proteins from the cell extracts of the addiction module mazEF expressed from the cell chromosome, presumably because under such condition these proteins were expressed at very low physiological concentrations.
When both MazE and MazF were present on native gels, we observed a complex between the toxin and its antidote (indicated by an arrow on the Fig. 1C). We used these crude protein extracts to study the influence of MazE and MazF on their own promoter.
Preparation of DNA fragments. DNA fragments for the gel mobility shift assays and DNase I footprint analysis were obtained by PCR with appropriate primers ( Germany) and 2µl of labelled DNA fragments were added. The binding reactions were conducted at room temperature for 10min, after which they were loaded onto 6% native polyacrylamide gels (PAAG) and run in TAE buffer [22] at 200V.
DNase I footprinting analyses were done according to Giladi et al. [23].
β-galactosidase assays. β-Galactosidase assays were done according to Miller [24]. RNA levels, the gels were analysed and the bands were quantified using the Fujix BAS100 phosphoimager.
The mutagenesis of the promoter fragment. Point mutations were introduced into mazEF promoter region by PCR-based site-directed overlap extension mutagenesis [26] using appropriate primers ( Table 2). All introduced mutational changes were verified by DNA sequence determination.

Results
The mazEF promoter is negatively auto-regulated. Like most addiction modules, sequence analysis of the promoter region of mazEF suggests that it is autoregulated at the transcriptional level [15]. To test whether mazEF was indeed auto- To verify that the regulation of the mazEF promoter took place at the transcriptional level, we performed a series of primer extension experiments using plasmid pSK10∆6-p ef as the template. Using RNA extracted from cells carrying this plasmid, we estimated the relative efficiency of transcription from the two promoters P 2 and P 3 (Fig. 2B). Transcription initiated from promoter P 3 was about 10-fold weaker then that from P 2, located 13bp upstream (Fig. 2B). This explains why we were unable to observe initiation from P 3 by primer extension on RNA transcribed from a chromosome borne mazEF module, which is present in only one copy per cell [12].
Primer extension experiments under the same experimental conditions revealed that induction by IPTG led to repression of transcription from both P 2 and P 3 by MazE or by the MazE-MazF complex. Twenty minutes after induction, MazE repressed P 2 expression by 53% compared to the activity of the unrepressed promoter; MazE-MazF complex repressed P 2 by 92% ( Fig. 2B and 2C). We believe that these two promoters are inhibited similarly; however, after repression, the levels of the P 3 transcript may have been so low that we could not measure them (Fig. 2B).
These results from our primer extension experiments (Figs. 1B and 1C) confirmed the data that we obtained in our assays for β-galactosidase activity ( Fig. 2A). Thus, we concluded that the mazEF addiction module is auto-regulated at the transcriptional level.
Over-expressing the MazE-MazF complex leads to a gel mobility shift of the mazEF promoter fragment. To further investigate the mechanism of the action of MazE-MazF on the promoters, we studied how MazE and MazF bind to the promoter region of the mazEF module (see map in Fig. 3A). As the source of proteins for these assays we used crude cell extracts enriched for either MazE or MazE and MazF ( Fig. 1). For our electrophoretic mobility shift assay we used the 74bp fragment of the mazEF promoter that extends from the multi-linker to the residue +2 of the P 2 promoter (Fig. 3A). This DNA fragment was labeled and exposed to every one of the cell extracts that we had prepared (defined in the legend to Fig. 1). We found that the mazEF promoter was bound by the MazE-MazF complex (Fig. 4A, lane 4), confirming our hypothesis that mazEF is negatively auto-regulated (Fig. 2).
The crude extract containing only MazE but lacking MazF also bound the promoter fragment (Fig. 4A, lane 3). Though MazE was present in approximately equal amounts when by itself or in the presence of MazF, here the shift was much weaker (Fig. 1A, compare lanes 3 and 4). Thus, MazE could bind to its own promoter, but, like the antidotes from most other addiction modules of plasmid origin, the binding affinity of MazE to its promoter was very low. The cooperative binding of the toxic protein, here MazF, greatly enhanced the binding of MazE (Fig. 4A, lane 4).
We also found that exposing the promoter fragment to the extract of MC4100∆mazEF, that contained neither MazE nor MazF, resulted in an additional retarded band with a very low shift (Fig. 4A, lane 2). This same band was present when the promoter fragment was exposed to the extracts that either contain no MazF or contained it at very low level (Fig. 4A, lanes 1 and 3). This unexpected result suggested that in addition to the binding by its own proteins MazE and MazF, the promoter region of mazEF also bore the binding site of another as yet Further characterising the mazEF promoter region by dividing it into two overlapping fragments. To define the binding sites of the mazEF promoter more precisely, we divided it into two overlapping fragments: the upstream fragment (-72 to -19) and the downstream fragment (-38 to +2) (Fig. 3B). These two overlapping fragments shared a common element (-38 to -19). When we exposed this common element to each of our cell extracts, we observed no shift in its electrophoretic mobility (data not shown), suggesting that the sequence that it bears is not long enough to permit binding. In contrast, using this same set of cell extracts to expose each of the overlapping fragments, the upstream fragment and the downstream fragment, caused each of them to be shifted very differently ( MazF, we observed both a supershift and an additional band above it. However, this highest band was missing from the gel when this same cellular extract was used to retard the downstream fragment (compare Fig. 4A and 4C, lanes 4). We propose that this overshifted band may have been formed by the mazEF promoter fragment binding to all three proteins: MazE, MazF, and the unidentified protein factor. The mobility pattern of the downstream DNA fragment exposed to the cell extract of E.
coli strain from which the mazEF genes had been deleted and (Fig. 4C, lane 2) looked like that of the probe alone (Fig. 4C, lane C), that is, there was no binding.
It did not seem to matter if the downstream fragment were exposed to MazE and MazF expressed from the E. coli chromosome (Fig. 4C, lane 1) or to MazE alone expressed from a plasmid (Fig. 4C, lane 3). We propose that because the promoter region binding affinity of the MazE-MazF complex is much higher than that Integration Host Factor (IHF), rpoS (stationary phase sigma factor), gyrase, and histone-like proteins H-NS and HU. Gel mobility shift assays on the upstream DNA fragment of the mazEF promoter exposed to each of these extracts (Fig. 6A) revealed retardation in every case except when FIS was absent (Fig. 6A, lane 2).
The addition to that same DNA fragment of increasing amounts of the pure protein promoter as it acts on most other promoters that it regulates [28]. A mutation in the putative FIS-binding site of the mazEF promoter, in which the T residue at position (-40) is replaced by a G residue, caused the mazEF promoter to be insensitive to activation by FIS ( Fig. 6D and 6F).
Analysis of the auto-regulation region of the mazEF promoter. Since we were able to show that the downstream fragment of the mazEF promoter was responsible for mazEF auto-regulation (Fig. 4C), the next step was to analyze its sequence (Fig.   7A). The auto-regulation regions of all known addiction modules contain palindrome structures, and mazEF was no exception: the palindrome sequence "a-b," analogous to the palindrome of pemIK, had already been predicted by Masuda and colleagues [14].
Sequence analysis of the area protected by the MazE-MazF complex revealed an unusually complicated structure (Fig. 7): (i) fragment "a" of the palindrome could be complemented not only by fragment "b", but also by fragment "c". We called this "c-a-b" component an "alternating palindrome", (ii) the centre of an additional palindrome, "d-e," is located 4bp upstream from the centre of the "c-a" palindrome ( Fig. 7).
We present a possible alignment of the palindrome fragments (Fig. 7B) and a possible base pairing among them (Fig. 7C). The "d" and "e" parts of the palindrome show a perfect complementation ( Fig. 7B and C). The common element (-38 --19) of the overlapping upstream and downstream fragments, tested before, contained almost the whole "d-e" palindrome, except for its last nucleotide T (-18). In the gel shift mobility assay the cell extracts contain either MazE alone or MazE-MazF complex did not show retardation with this DNA fragment (data not shown).
Within the alternative palindrome the homology was high, especially between parts "c" and "b." We hypothesized that at any given moment the "alternating palindrome" might exist in one of two possible configurations: "c-a" or "a-b" (Fig. 7C).
To test our model, we performed gel mobility shift experiments using the two by guest on November 6, 2017 http://www.jbc.org/ Downloaded from overlapping parts of the double palindrome, "a-b" (-20 to +6) and "a-c" (-34 to -7) and of their common fragment "a" (-23 to -7) alone. The MazE-MazF protein complex bound each of the two palindrome sequences ("a-b" or "a-c"), while none of the cell extracts caused a shift of the middle fragment "a" alone (Fig. 8A).
To verify our model further, we constructed a variation of the downstream promoter fragment "c-m-b" (-38 to +6), in which we replaced the middle part "a" with an unrelated multi-linker sequence that we called "m." After exposing this "c-m-b" fragment to our standard set of crude cellular extracts, we ran the mixtures in a gel mobility shift assay. As we observed for fragment "a" alone, none of the cell extracts led to retardation of the "c-m-b" fragment ( Fig. 8A). In these experiments, we observed no binding of MazE by itself to any other fragment than to the complete "ca-b" fragment ( Fig. 8A).
We asked: what was the factor that was important for binding? Was it simply the sequence of the double palindrome or was the secondary structure of the promoter region required? To distinguish between these two possibilities we introduced several mutations into the sequence of the mazEF promoter region (indicated by the black points in Fig. 7C and specified in Table 2). Every mutation we introduced destroyed a hydrogen bond in the proposed structure, and in the case of mutations (-25; -26; -27) three such bonds were destroyed simultaneously. Each mutated fragment was exposed to cell extracts enriched for the MazE-MazF complex and was run on a gel mobility shift assay. It appears that introducing these mutations caused no changes in the mobility of the promoter fragment (data not shown).

Discussion
In previous studies we have shown that the chromosomal genes mazE and mazF, located in the E. coli rel operon, have all the properties required to be an addiction module [12]. Along with properties shared with other known addiction systems, the mazEF addiction module has two additional properties: (a) it is directed from two promoter is about 10-fold stronger then is the P 3 promoter; (ii) expression from both P 2 and P 3 is repressed by MazE and is highly repressed by the MazE-MazF complex; (iii) MazE and MazF could bind to an "alternating palindrome" that we found in the promoter region (-34 to +6). This alternating palindrome, which in fact is the operator of mazEF, could exist in one of two alternative states: its middle part "a" complemented with either of the outer parts "b" or "c" (Fig. 7); (iv) expression from mazEF promoters is activated by FIS.
When β-galactosidase was expressed under the control of the mazEF promoters, we observed as much as 7000-8000 Miller units at mid-logarithmic growth ( Fig. 2A). This high level of synthesis indicated that the mazEF promoter was very strong, as are the promoters of most addiction modules [5; 8; 9; 10]. Moreover, the presence in trans of the gene products of the mazEF module led to negative auto-regulation, and about ten-fold repression of transcription (Fig. 2).
In further experiments, we obtained cell extracts containing the MazE and MazF proteins, either transcribed from the chromosome or over-expressed from a plasmid under the control of the tac promoter. When MazE and MazF are present together in a cell extract they form a complex. Here, and in other studies, we observed the formation of such a complex in native gel electrophoresis with ( 35 S)methionine labelled proteins [12], with antibodies against MazE (Fig. 1C), and also in an E. coli two hybrid system (Marianovsky and Glaser, in preparation).
By gel mobility shift assays we clearly showed that the retardation of the promoter fragment depended on exposure to either the MazE or to MazE-MazF complex. We were surprised to find that this DNA fragment could also be retarded by an initially unidentified protein present in the crude cellular extract of a strain from which mazEF had been deleted (Fig. 4A). On the basis of our results reported above, we have concluded that this regulation protein is FIS. FIS is one of the nucleoid-associated proteins that regulate various processes, including transcription, recombination, and replication [28; 29]. Here we found that FIS increased the activity of the mazEF promoter by 1.6-fold ( Fig. 6C and 6E). However, it is possible that under certain specific stressful conditions the effect of FIS on mazEF could be more profound. In this regard, we suggest that FIS may affect the role of mazEF in programmed cell death. Under various physiological conditions, the cellular levels of FIS vary over a large interval (up to 100-fold) and they depend on both the growth phase and on nutritional conditions [29; 30] In rich medium, the concentrations of FIS are very high in the early exponential phase, but sharply decrease towards stationary phase. FIS is known to act as a homodimer [27]; the molecular weight that we calculated for the initially unidentified protein from the results of our gel filtration experiments corresponded to the molecular weight of FIS as a homodimer.
It has been shown that by binding to the DNA region upstream from promoters, this homodimer causes the DNA to bend, thus increasing the binding efficiency of the RNA polymerase [31]. Thus, positive regulation of the mazEF promoter by FIS must be maximal under conditions of rapid growth on rich media.
We found a high level of conformity between the sequence of the upstream fragment of the mazEF promoter and the consensus sequence of the known FISbinding sites [28]. When we introduced a point mutation in the FIS binding site of the mazEF promoter, the influence of FIS on the promoter was abolished ( Fig. 6D and   6F), further confirming that FIS participates in mazEF gene regulation. We were able to ascertain the precise location of the FIS binding site in the mazEF promoter (underlined in the schematic diagram in Fig. 3C). FIS regulation of the promoter of this addiction module seems to be a unique feature of the mazEF module.
While FIS caused positive regulation of the mazEF promoter, auto-regulation of mazEF promoter was strongly negative as it is for most known addiction modules.
We found that the auto-regulation site of mazEF was longer and more complicated than would have been predicted by the results of previous studies [14]. In our DNA footprint experiments, we found that an area more extended then the "a-b" palindrome was protected against DNase I digestion by the MazE-MazF protein complex (Fig. 5). In addition to this "a-b" palindrome, which is similar to the palindrome in the pemIK promoter [14], we discovered an unusual structure that we have called an "alternating palindrome." Thus, our results suggest that the middle fragment "a" may complement not only the downstream fragment "b", but also fragment "c" located upstream from "a" (Fig. 7C). Comparing fragments "b" and "c" revealed that they were highly similar (Fig. 7B). The "alternating binding model" that we have proposed here is supported by the results of our gel mobility shift experiments: the MazE-MazF complex could bind both alternate structures, "c-a" and "a-b," but not the central "a" fragment by itself. Moreover, the MazE-MazF complex could not bind the "c-m-b" fragment in which the "a" fragment was replaced by an unrelated sequence (Fig. 8A). Based on mutational analyses, we propose that the mazEF promoter can exist in two possible alternate states (Fig. 8). The MazE-MazF complex can bind either of these structures, resulting in strong negative autoregulation (see below).
The role of the additional "d-e" palindrome is not yet clear. The regulation areas of the promoters of many known addiction modules contain a palindrome sequence [9; 10; 11; 32]. Moreover, some addiction modules, like phd-doc and pemIK, also contain two palindromes [9; 10; 11; 14]. It has been suggested that in the regulation of the phd-doc addiction module, the toxic and the anti-toxic proteins bind to the two palindromes cooperatively. This binding process is accompanied by an increased affinity of the protein for the DNA, and hence an increasing stability of the DNA-protein complex [11].
The numerous mutations that we introduced into the "alternating palindrome" did not at all affect the binding efficiency of the Maze-MazF complex, suggesting that the secondary structures of the regulating region is more important than its DNA sequence per se. The "alternating palindrome" that we have described seems to be a unique feature of the mazEF promoter. In this structure, it is as though the two palindromes often found in the promoters of other addiction modules [9; 10; 11; 14] have been collapsed, thus minimizing the space required for the regulating elements without losing efficiency.
We suggest that the combined presence of two promoters, a complicated palindrome structure, and the FIS binding site permits regulation of expression that is simultaneously safe and dynamic, enabling quick responses to changes in physiological conditions. The duplication of the structural elements (promoters or binding sites for auto-regulation) assures that mazEF regulation will be adequate even in the case that one of these elements may be destroyed. The action of two promoters, P 2 and P 3 , is additive; during exponential growth they are repressed by Thus, the mazEF promoter is elegantly engineered to respond to any possible changes in the nutritional environment of the bacterium.  promoter, was transformed with pLex-mazE or with pLex-mazEF (see Table 1). At