In vitro, the σs subunit of RNA polymerase (RNAP), RpoS, recognizes nearly identical −35 and −10 promoter consensus sequences as the vegetative σ70. In vivo, promoter selectivity of RNAP holoenzyme containing either σs (Eσs) or σ70 (Eσ70) seems to be achieved by the differential ability of the two holoenzymes to tolerate deviations from the promoter consensus sequence. In this study, we suggest that many natural σs-dependent promoters possess a −35 element, a feature that has been considered as not conserved among σs-dependent promoters. These −35 hexamers are mostly non-optimally spaced from the −10 region, but nevertheless functional. A ± 2 bp deviation from the optimal spacer length of 17 bp or the complete absence of a −35 consensus sequence decreases overall promoter activity, but at the same time favours Eσs in its competition with Eσ70 for promoter recognition. On the other hand, the reduction of promoter activity due to shifting of the −35 element can be counterbalanced by an activity-stimulating feature such as A/T-richness of the spacer region without compromising Eσs selectivity. Based on mutational analysis of σs, we suggest a role of regions 2.5 and 4 of σs in sensing sub-optimally located −35 elements.
Bacterial RNA polymerase (RNAP) consists of a catalytic core (α2ββ’ω, abbreviated as E) and an additional subunit, the sigma factor, which is responsible for promoter recognition and DNA melting. Depending on their adaptive potential, different bacteria contain different numbers of sigma factors with each sigma factor regulating a distinct set of genes. The simplest way of accomplishing this is by programming the various sigma factors to recognize different promoter consensus sequences.
However, the case seems to be different for the house-keeping sigma factor in Escherichia coli, σ70 (RpoD), and σs (RpoS), a crucial sigma factor in stationary phase and in stressful conditions (Hengge-Aronis, 2002). These two sigma factors show a high degree of sequence similarity and were found to bind optimally to almost identical −35 and −10 elements in vitro (Gaal et al., 2001). In vivo, however, σs efficiently activates the expression of a distinct regulon, which, depending on the conditions, comprises up to 10% of the E. coli genes (Weber et al., 2005). What is the basis of this Eσs selectivity within the cellular context?
Based on sequence and experimental analysis, we propose here that many natural Eσs-dependent promoters exhibit spacers between the −35 and −10 regions that deviate from the optimal 17 bp spacer by one or two base pairs. We demonstrate experimentally that non-optimally located −35 elements are not only functional, but can be better used by Eσs than by Eσ70 and therefore contribute to Eσs selectivity of a promoter. The mechanism that allows Eσs to utilize non-optimally placed −35 regions more efficiently than Eσ70 is further investigated using specific mutations in relevant amino acids in σs. Finally, the frequent occurrence of transcription initiation-enhancing A/T stretches in the spacer region (centred at −20 and at −27, relative to the transcriptional start) may allow Eσs-dependent promoters to increase Eσs selectivity by using non-optimal spacers without a strong loss in overall activity.
Many natural Eσs-dependent promoters feature −35 regions, but not at a distance of 17 bp from the −10 region
In a recent study in our laboratory, we observed that destroying the −35 hexamer alone in a typical σ70-dependent promoter conferred little if any σs selectivity (G. Becker and R. Hengge, in preparation). This was somewhat surprising in view of the general concept that Eσs does not need a −35 region and that therefore, this element seems poorly conserved in σs-dependent promoters (Espinosa-Urgel et al., 1996; Lee and Gralla, 2001). This finding prompted us to make a new alignment of the ∼80 experimentally mapped σs-dependent promoter regions (see Table S1). The promoters were grouped in accordance to the presence and location of possible −35 elements with three or more matches to the consensus hexamer TTGACA (recognized in vitro both by Eσ70 and by Eσs; Gaal et al., 2001). A spacer length of 15–19 bp was allowed between the −10 and −35 boxes.
This alignment suggested that (i) −35 elements are likely to be more frequently present in Eσs–controlled promoters than previously thought, and (ii) Eσs does not appear to show the rigid preference of Eσ70 for a 17 bp distance between the −10 and −35 regions. Thus we proceeded with a statistical comparison of the data obtained from our list of σs-dependent promoter regions with the data recently published for 554 mapped, mostly Eσ70-dependent E. coli promoters (Mitchell et al., 2003). The distribution of −35 elements in relation to the range of spacer lengths between the −10 and −35 boxes was analysed (Fig. 1A). Mitchell et al. (2003) divided the promoters in extended −10 promoters, which contained a 5′-TG-3′ motif at positions −15 and −14, and promoters that did not feature this motif. Extended −10 promoters were 19% of the total and are designated here as ‘TG promoters’ (‘all promoters’ stands for TG and non-TG promoters together). It is worth mentioning that about 15% of the Eσs-dependent promoters also had a TG motif at positions −15 and −14 (most of these promoters did not exhibit a recognizable −35 element, or featured a misplaced one). A TG motif positioned one base pair further upstream (−16, −15) was recently proposed as a selective feature for σs-dependent promoters and part of the motif recognized by Eσs (Lacour et al., 2003; Lacour and Landini, 2004); however, only 10.1% of the confirmed σs-dependent promoters analysed here have such a TG element in comparison with 8% of the vegetative promoters in the study by Mitchell et al. (2003).
The comparison of Eσ70/Eσs-dependent promoters in relation to the spacer lengths between the −10 and −35 elements (Fig. 1A) outlined some of their opposing characteristics. First, Eσs-dependent promoters exhibit all spacer lengths between 15 and 19 bp with comparable frequencies. On the contrary, Eσ70-dependent promoters show a strong preference for a 17 bp spacer length, as expected. In addition to this, ‘TG-promoters’ tend to have 18 or 19 bp spacers at similar frequencies as Eσs-dependent promoters.
While this statistical analysis clearly indicates that in σs-controlled promoters 17 bp spacers have not been selected for, the finding that spacers with apparent lengths between 15 and 19 bp are found at similar frequencies has to be interpreted with caution. The probability of matching 3/6 bases in a random hexamer is 17% for an unbiased genome. At the same time, the occurrence of a −35 element at any physiological distance (i.e. between 15 and 19 bp) from the −10 box in σs-dependent promoters is approximately of this size. This may mean that (i) Eσs does not use a −35 region at all; or (ii) Eσs uses a −35 region, but functions equally well with spacers between 15 and 19 bp; or (iii) Eσs works better with non-optimal spacers than Eσ70, but still has some preference for 17 bp spacers, which, however, is not reflected in natural Eσs-controlled promoters, as these promoters have been selected by evolution not for maximal activity but for being Eσs-selective in the presence of Eσ70. These possibilities can only be distinguished experimentally.
In order to test whether putative −35 elements with non-optimal sequence and spacing are used at all by Eσsin vivo, three sequence-predicted −35 elements in natural σs-dependent promoters were selected for further testing (with their deviations from optimal sequence and spacing, these promoters can be regarded as representative cases). The cfa P2-promoter is entirely σs-dependent (Wang and Cronan, 1994), and possesses two possible overlapping −35 elements which can be mutated separately. Thereby we could show that the two elements enhance transcriptional activity in an additive manner, with the −35 hexamer situated 18 bp upstream of the −10 element (‘−35 element 2’) playing a somewhat more profound role than the −35 element with a 16 bp spacer (‘−35 element 1’). For the csiD promoter, the presence of the −35 element (located 17 bp upstream of the −10 element) turned out to be crucial not only for high transcriptional output, but also for enhancing σs selectivity of the promoter (Fig. 1B). Furthermore, direct or indirect evidence for the existence of functional −35 elements in σs-dependent promoters can also be found in literature for aidB (Lacour et al., 2002), osmB (P2) (Jung et al., 1990), osmE (Checroun et al., 2004) and osmY (Wise et al., 1996).
In summary, we conclude that −35 elements are likely to be also important for Eσs-controlled promoters, but for them no strong preference for an optimal spacing between the −10 and −35 regions is apparent. Questions that arose from these findings were whether Eσs can use spacers of different length equally well, and/or whether ‘misplaced’−35 regions are conserved because they can contribute to Eσs selectivity of a promoter by being non-optimal for Eσ70. In order to answer them, effects on Eσs and Eσ70 after systematically varying promoter spacer lengths had to be compared.
Spacer length affects Eσs selectivity of synthetic promoters
As naturally evolved σs-dependent promoters can exhibit complex and not yet well-recognized combinations of different Eσs selectivity-enhancing elements (cis or trans-acting), we decided to systematically study the contribution of ‘non-optimal’ spacer length in the context of less complex and better characterized synthetic promoters.
Therefore, a synthetic promoter with a 17 bp spacer between −10 and −35 (synp213; see Experimental procedures), fused to lacZ and integrated in single-copy at the chromosomal att(λ) site, was used as a basis for analysing the effect of varying the spacer length between 15 and 19 base pairs. The −35 element in this promoter carries one mismatch to the consensus sequence, as a ‘perfect promoter’ (consensus −10 and −35 elements with 17 bp spacing) was not tolerated by the cells, which immediately picked up suppressor mutants (the majority had the single mismatch nucleotide in the −35 region used here; G. Becker and R. Hengge, in preparation). This initially 17 bp spaced promoter possesses a −10 region optimal for Eσs[with a C(−13) and a TAA sequence downstream of the −10 region; Weber et al., 2005], that allows its predominantly σs-dependent expression in stationary phase. Thus, the promoter activity was approximately 2.5-fold lower in the absence of σs (Fig. 2). When the −35 region was moved to sub-optimal positions relative to the −10 region, i.e. spacer lengths of 15, 16, 18 or 19 bp were generated, overall promoter activity was reduced as expected, but the residual expression (deriving from the house-keeping Eσ70 alone) in the rpoS mutant was now 4–4.5-fold lower than when Eσs was also present (Fig. 2). In other words, the introduction of longer or shorter spacers into the promoter reduced its overall activity but improved its Eσs selectivity. Eσs selectivity was even more pronounced after the complete deletion of the −35 hexamer (eightfold lower activity in the rpoS mutant; Fig. 2). It is worth mentioning that in terms of absolute activities, Eσ70 slightly benefited from the existence of −35 region even with a 19 bp spacer. For Eσs, however, this arrangement was rather counterproductive, as Eσs could function better in the total absence of −35 element than with a 19 bp spacer (Figs. 2 and 5A).
It is important to clarify at this point that in this study when we refer to in vivo Eσs-dependency or selectivity we always compare transcriptional activity in two different physiological conditions, i.e. in the presence of σs (Eσs + Eσ70-derived transcription) and in its absence (Eσ70-derived transcription). Measuring pure Eσs-driven transcriptional activation at a promoter also recognized by the vegetative RNAP is impossible in vivo, and calculating it by simply subtracting the Eσ70-derived activity (in the rpoS mutant) from the Eσs + Eσ70-derived activity (in the wild-type strain) would be an oversimplification and an underestimation of the actual value (although formally Eσs/Eσ70-derived transcription ratios would be higher than (Eσs + Eσ70)/Eσ70-derived ratios, that we are using). Due to sigma factor competition for limiting amounts of core RNAP, Eσ70 levels and therefore Eσ70-mediated promoter activities are not identical in the presence and absence of σs. This becomes most visible with certain weak Eσ70-controlled promoters that show ‘negative’ regulation by EσS., i.e. are upregulated in an rpoS mutant (Farewell et al., 1998).
In vitro, however, Eσs- and Eσ70-driven expression can be separated. With respect to using promoters with different spacer lengths, a pattern similar to the in vivo results was observed in an in vitro transcription assay (Fig. 3). In terms of absolute activities, Eσ70 had a clear preference for a 17 bp spacer not shown to this extent by Eσs. As a consequence, Eσs used the promoter configurations with extreme spacing (15 or 19 bp) or lack of a −35 element more efficiently than Eσ70. Sixteen or 18 bp spaced promoters were equally utilized by the two holoenzymes (Eσs and Eσ70), whereas the 17 bp spacer promoter yielded higher activity with Eσ70 than with Eσs (Fig. 3). This indicates that Eσs can work nearly equally well with promoters exhibiting spacers of any length between 15 and 19 bp, whereas Eσ70 is highly optimized to operate with 17 bp spacer promoters. It is worth mentioning that the results were identical no matter if a 1:1 or a 5:1 molar ratio of sigma factor to core RNAP was used for the reconstitution of the holoenzymes (to compensate for different affinities of sigma factors to core RNAP), as reconstituted RNAP was added in considerable excess over the template DNA in the assay. Under these conditions of excess RNAP, Eσs and Eσ70 can equally transcribe RNAI (used for normalization; for more details see Experimental procedures). In order to draw the former conclusions, however, two more issues have to be clarified: (i) the shift/mutation of the −35 element should not accidentally introduce any new transcriptional start points; and (ii) it has to be shown that the −35 elements in sub-optimal places are truly used by the RNAPs. With respect to the first question, both the in vitro transcription assay (Fig. 3) and primer extension experiments (data not shown) did not reveal the generation of any transcript with an alternative start site (as expected because the insertions/deletions were designed so that they would not introduce alternative possible −10 or −35 elements and the existing −10 element is optimal for Eσs and can also be used by Eσ70).
As for the second conundrum, the fact that at least 15, 16 and 18 bp spaced promoters produced higher activities than the promoter without a −35 region (Figs 2 and 5A), indicated utilization of the −35 region in these non-optimally spaced promoters. In addition, we degenerated the −35 region in the non-optimal 16 bp spaced promoter in a step-wise manner (Fig. 4A). Our initial construct carries already one mismatch to the consensus sequence (TTCACA compared with TTGACA), and the insertion of further mismatches led to a gradual drop of the overall promoter activity (both Eσs and Eσ70-dependent), but at the same time increased Eσs selectivity of the promoter (Fig. 4B). Degenerating a −35 element is known to increase σs selectivity (Gaal et al., 2001; Lacour et al., 2003). Moreover, for the promoter carrying only a 3/6 match to the −35 element consensus sequence (‘16/3’) the absolute overall transcriptional levels were still higher than that for the ‘no-35’ construct (compare with Fig. 2), but Eσs-dependence was similar (eightfold more in the presence or Eσs). This result suggests that promoter efficiency and σs selectivity can be combined by retaining sub-optimally positioned and partially degenerate −35 elements, instead of completely eliminating the −35 element (which has more devastating effects in terms of overall promoter activity). The construct with 2/6 matches to the −35 hexamer consensus (referred as 16/4), expected to function similar to the ‘no-35’ construct (Fig. 2), showed slightly higher overall transcriptional activity and also σs-dependency (10-fold more in the presence or Eσs, versus eightfold of the ‘no-35’ construct). This may be due to the accidental introduction of a cCGG sequence in the place of the canonical −35 element; the GCGG motif, 16 bp spaced from the −10 box, has been suggested to be utilized as an alternative −35 element consensus sequence in osmotically induced σs-dependent promoters (Lee and Gralla, 2004).
All these in vivo data indicated that non-optimally placed and even degenerate −35 elements are indeed used by RNAP. This was also confirmed by in vitro evidence obtained in DNaseI footprinting experiments. Both the template and the non-template strands (shown in Fig. 4C and D respectively) revealed decreasing protection of the −35 region upon addition of any of the holoenzymes (Eσ70 and Eσs) with increasing sequence degeneration of the −35 region. Protection patterns for both holoenzymes extend from −60 to +20 (for both strands; in order to increase the resolution of the relevant promoter region, only a part of the footprint pattern is shown for the template strand in Fig. 4C). Differences between the two holoenzymes are subtle and restricted to the template strand, e.g. Eσs provides better protection downstream of the −10 box than Eσ70 (Fig. 4C). Interestingly, the constructs with more degenerate −35 elements (especially the 16/4) exhibited a more hypersensitive site around position −25 in the non-template strand upon RNAP binding (Fig. 4D; position −25 is marked by an arrow). That would imply a different DNA configuration in the complex with RNAP when a canonical −35 hexamer is not present (see Discussion).
A role of the interaction between C(−13) and K173 in σs for enhancing Eσs selectivity of promoters with spacers longer than 17 bp
As a next step, we addressed the question whether altering the spacer length per se can increase σs selectivity, or whether there is a prerequisite for some σs-selective features in the initially optimally spaced promoter. Therefore, a promoter very similar to synp213 was constructed, this time though without a hallmark of σs selectivity, i.e. the C(−13), which is involved in a direct interaction to amino acid K173 in σs and is counter-selective for Eσ70, which has a glutamate at the corresponding position (Becker and Hengge-Aronis, 2001). In addition, C(−13) was suggested to contact also amino acids Q152 and E155 of σs, although evidence is not conclusive (Checroun et al., 2004). Synp215, which carries TT(−14, −13), yields about half of the activity as synp213, which carries CC (compare Figs 2 and 5A) and in stationary phase the activity for synp215 with the 17 bp spacer was only 1.5-fold stimulated by the presence of Eσs (Fig. 5A). Shifting the −35 hexamer 1 or 2 bp further upstream did not change the relative ability of Eσs to utilize the promoter (18 or 19 bp spacers), whereas reducing the spacer length to 15 or 16 bp enhanced the contribution of Eσs to overall transcriptional activation, i.e. the promoter activity was more than twofold reduced in the rpoS mutant background (Fig. 5A). Removing the −35 box entirely improved Eσs selectivity as expected, by increasing the ratio of expression in rpoS+/rpoS backgrounds to 2.5-fold. In other words, in the absence the C(−13) and therefore its direct interaction to σs, the ability of Eσs to recognize better than Eσ70 a promoter with spacers longer than 17 bp was lost, whereas its advantageous use of shorter spaced promoters or promoters without a −35 region was maintained. This is also apparent in Fig. 5B, where Eσs-dependence (i.e. expression ratios in the presence versus absence of σs) is plotted against the different spacer lengths.
In order to further establish that the interaction between C(−13) and K173 in σs is involved in increased Eσs dependence of promoters with longer spacers, we used the σs(K173E) variant for transcription from the differently spaced derivatives of synp213 promoter [which carries CC(−14, −13)]. This amino acid exchange in σs destroys the C(−13)–σs interaction (Becker and Hengge-Aronis, 2001) and therefore reduces overall expression as expected (data not shown). In addition, we observed a differential effect depending on promoter spacer length: for the promoters with 17, 18 or 19 bp spacers, the K173E variant of σs did not exhibit the increased Eσs-dependence observed for wild-type σs. On the other side, increased Eσs-dependence of promoters with 15 and 16 bp spacer lengths, or in the complete absence of a −35 element was still observed (Fig. 5B and data not shown). Thus, we conclude that the interaction of C(−13) and K173 in σs (i.e. a contact right upstream of the −10 region) plays a role in enhancing Eσs selectivity of a promoter by increasing its spacer length, whereas it is of less importance for the relative advantage Eσs gains when the spacer is shortened by 1 or 2 bp from the standard length of 17 bp.
Genetic analysis of the divergent abilities of Eσs and Eσ70 to utilize promoters with spacers shorter or longer than 17 bp
In order to further explore the molecular mechanism that allows Eσs to better tolerate deviations from the standard 17 bp spacer length than the house-keeping Eσ70, we constructed a series of single amino acid substitutions, mainly in region 4 of σs. The effects of different σs mutants were monitored with the aid of a set of plasmids encoding σs variants in a rpoS– background. The plasmids are derivatives of pRL40.1, which express the σs variants from the tac promoter. In the absence of inducer the levels of expression of σs in stationary phase are similar to those obtained in wild-type strains that express σs from the rpoS gene in the chromosome under its own promoter control (for details see Experimental procedures).
The amino acids of σs that were replaced by the corresponding residues of σ70 are indicated in bold face in Fig. 6A. Our goal was to identify amino acid residues that contribute to the ability of σs to better utilize non-optimal spacers in comparison with σ70. Therefore, the logic behind selecting the residues of σs to be mutated was the following: (i) they had to be different in the two sigmas; (ii) they could be directly associated with the different interactions of domain 4 of σs (σ4) with either the promoter region (−35 element) or the core RNAP (flap domain of the β subunit of RNAP), or would be located in the close proximity of those amino acids.
All mutant variants of σs were tested for their ability to utilize the synp213 promoter series with different spacer lengths. Depending on the effects that the mutant forms of σs produced, they fell into three different categories. The first category consisted of mutants that did not exhibit changes in either overall transcriptional activity or in the relative superiority of σs in using non-optimally spaced promoters (i.e. ratios of expression in wild-type and rpoS mutant remained unchanged; data not shown). The second category consists of σs variants that produced general transcriptional defects with Eσs and therefore reduced Eσs selectivity (i.e. ratios in wild-type and rpoS mutant strains) independent of spacer lengths (data not shown). Most of these amino acid exchanges are situated in region 4.1 or the linker between regions 4.1 and 4.2, and the reverse exchanges of some of them (i.e. I219P, I259T, L280I, L281D) are known to affect the σ4–core interaction when introduced into σ70 (Sharp et al., 1999; Campbell et al., 2002; Gregory et al., 2004; Nickels et al., 2005). The third group consists of two mutants (I316P; and a deletion downstream of G321, i.e. of the last nine C-terminal amino acids of σs) that result in both a defect of Eσs transcriptional ability and reduction of Eσs selectivity, predominantly with non-optimal spacers (Fig. 6B). Thus, these two mutant variants of σs exhibit a reduced ability to utilize promoters with non-optimal spacers. These variants are either lacking or most likely have altered (by introduction of a proline) the fifth helix of the C-terminal domain of σs and therefore are partially impaired in the interaction with the RNAP (see also Discussion). Finally, a few mutants (Q270E, Y283N, AA285DY and Q306E) did not fit to any of these groups as they affected specifically only one or two of the differentially spaced promoters (data not shown). From these data we conclude that the C-terminal region of σs is involved in its ability to take advantage of non-optimal spacers.
Role of the A/T-richness of the spacer region in the competition between Eσ70 and Eσs for the same promoters
The data shown above indicate that spacer length matters for Eσs promoter selectivity. Another question is whether the specific sequence of the spacer also plays a role in sigma selectivity of a promoter. A DNA sequence selected in vitro as best recognized by Eσs (Gaal et al., 2001), contained stretches of A/Ts at specific positions of the spacer (see also Fig. 7). The presence of short poly(A) or poly(T) tracts in the spacer region of E. coli promoters has long been known to increase the DNA curvature and at the same time aid transcriptional initiation (Collis et al., 1989; Lozinski et al., 1989). In addition, an A/T rich region directly upstream of the −10 was recently shown both to increase RNAP affinity and facilitate open complex formation, allowing CRP-independent activity of the lac promoter (Liu et al., 2004).
Sequence analysis revealed that there is a significant higher tendency for AAA or TTT tracts at distinct positions in the spacer region of the experimentally confirmed σs-dependent promoters (Fig. 7A) compared with σ70-dependent promoters (Fig. 7B, which was produced using the data in the Mitchell et al. study). In σs-dependent promoters, AAA and TTT trinucleotides are often found centred around positions −27 and −20 (Fig. 7A), whereas for σ70-dependent promoters their occurrence is less frequent and evenly distributed (though significantly higher than randomness: 1.56%; Fig. 7B).
In order to enquire the contribution of these A/T stretches to promoter activity and/or Eσs selectivity, we introduced or deleted A/T tracts in the spacer region of synp213 promoter (with either a 17 or an 18 bp spacer; Fig. 7C). When the already existing A/T region downstream of the −35 hexamer was replaced by a G/C stretch, expression was somewhat reduced, whereas adding another A/T region upstream of the extended −10 region resulted in a significant stimulation of the promoter activity (Fig. 7D and E). EσS. selectivity, however, was not altered, as Eσ70-dependent activation (in the rpoS mutant background) retained a similar fraction of the overall, independently of spacer length. In other words, A/T richness in the spacer region (especially directly upstream of the extended −10 region) resulted in a higher general promoter activity without compromising σs selectivity gained by increasing the spacer length to 18 bp.
In summary, Eσs promoter selectivity can be established by the ability of Eσs to better tolerate sequence or spacing deviations from the optimal, i.e. maximally active promoter (which is very similar to that of σ70). The concurring reduction in promoter strength, however, can be counteracted by the presence of A/T tracts in the spacer region that stimulate transcriptional activity without compromising Eσs selectivity. Consistently, such A/T tracts are significantly conserved in σs-dependent promoters.
Presence, positioning and role of −35 elements in σs-dependent promoters
The initial goal of this study was to settle the question whether −35 elements are present and play any role in σs-dependent promoters. The data existing up to that point were conflicting. On one hand, the apparent lack of a conserved −35 element was perhaps the first noticed feature that seemed to distinguish promoters recognized by Eσs from those recognized by Eσ70 (Hiratsu et al., 1995; Espinosa-Urgel et al., 1996; Lee and Gralla, 2001). This correlated with the weak interaction of region 4.2 of σs with a −35 box observed in FeBABE cleavage studies (Colland et al., 1999). On the other hand, in vitro selection of a sequence tightly bound by Eσs produced a −35 hexamer like that for Eσ70 (Gaal et al., 2001), and a direct interaction between region 4.2 of σs and the −35 box was observed, when the latter was present (Checroun et al., 2004). In addition, most of the amino acids that mediate the interaction of region 4.2 of σ70 to the −35 hexamer are conserved in region 4.2 of σs.
The results of our study reported here allow us to draw some basic conclusions with respect to the presence and the role of −35 regions in σs-dependent promoters. First, many σs-dependent promoters carry putative −35 elements that seem to be functional in the cases studied. Second, σs-dependent promoters exhibit a clearly higher flexibility regarding the position of the −35 box in relation to the −10 region, which is in contrast to the strong preference of Eσ70 for a 17 bp distance. This high incidence of 15, 16, 18 or 19 bp spacers exhibited in natural σs-dependent promoters contributes to σs selectivity, despite the reduction in general promoter activity that the non-optimal spacing produces in vivo. A complete lack of a −35 hexamer can offer even stronger Eσs promoter selectivity but the reduction in general promoter activity is pronounced (Figs 2 and 5). Alternatively, a combination of a non-optimally located −35 hexamer with small deviations from the consensus sequence can arrange a compromise of relatively high promoter activity and σs-dependency (Fig. 4B). This is perhaps why relatively degenerate −35 elements are often found in natural σs-dependent promoters.
But why did some previous studies fail to acknowledge the presence of −35 elements in a significant number of σs-dependent promoters? One reason was the expectance that −35 elements, if present, should be found in a restricted, well-defined position, i.e. 17 bp upstream of the −10 region. Another point is that most of the σs-dependent promoters that carry a −35 box show a relative low match to the consensus (3–4/6 match) and therefore can be overlooked. Only 10% have a 5/6 or higher match to the consensus, opposed to 30% of the house-keeping genes (Mitchell et al., 2003). Thus, the tendency of σs-dependent promoters to possess a ‘misplaced’−35 element goes hand in hand with a certain deviation from the consensus sequence. This fits well to the observations that mismatches in the −35 region are far more deleterious for Eσ70 than for Eσs (Fig. 4B; Bordes et al., 2000), and that Eσs can recognize equally well some −35 elements that are slightly deviated from the consensus sequence, e.g. with other nucleotide combinations instead of two thymidines in the first two positions of the −35 element (Wise et al., 1996; Gaal et al., 2001). It also explains why sequence pattern search algorithms did not identify a conserved sequence around the −35 position while readily finding a conserved extended −10 element (KCTATACTTAA) with high statistical significance upstream of a set of 140 strongly σs-dependent genes recently identified by genome-wide microarray analysis (Weber et al., 2005).
Relative degeneration combined to relative variability of the position of the −35 element in σs-dependent promoters offers the possibility that more than one sequence around this position can be used as a contact site for region 4.2 of σs (10 promoters in Table S1). An example is the cfa promoter, which features two potential overlapping −35 regions with 16 and 18 bp spacers, both of which exhibit a 4/6 match to the −35 consensus sequence. Point mutational studies indicate that both are functional (Fig. 1B) and it is conceivable that the two −35 elements may even have different relative importance under various stress conditions.
Besides the general existence of −35 regions in Eσs-dependent promoters, the second important finding of this paper is that non-optimal spacer lengths are common for Eσs-dependent promoters (Fig. 1A), a feature that in vivo increases Eσs selectivity by simply impairing Eσs-dependent activity less drastically than Eσ70-dependent activity (Figs 2 and 5). In vitro, transcriptional activities obtained with Eσs are even very similar with spacers of any length between 15 and 19 bp, which is in pronounced contrast to Eσ70, which has a strict 17 bp requirement for maximal activity (Fig. 3). The difference between the in vivo and in vitro results (in vivo, Eσs can clearly benefit from an optimally spaced promoter in terms of absolute promoter activity) may be due to the ‘isolated’ conditions in which the single-round in vitro transcription is performed. In the in vitro assay, excess RNAP is added together with template DNA (plasmid), and then the long pre-incubation time ensures that RNAP reaches an equilibrium state for recruitment and open complex formation at each promoter. As a next step, a mixture of nucleotides and heparin is added and enough time is allowed for a single round of transcription. Thereby differences in the kinetics of binding or open complex formation of the holoenzymes at the different promoters are masked. On the contrary, such kinetic differences might strongly affect the end result (high transcriptional activity) in vivo, where weaker promoters have to compete with stronger ones over limiting amounts of RNAP, sigma factors have to compete for core RNAP, and Eσ70 and Eσs for the same promoters.
In summary, σs, which in evolutionary terms is still highly related to σ70 and therefore does not have a full consensus sequence ‘of its own’ as ‘true’ alternative sigma factors do, has achieved specific promoter selectivity by resorting to a broader tolerance for deviations from the optimal promoter, be it in sequence or exact geometric alignment of the −35 and −10 regions, i.e. spacer length.
The molecular mechanism behind the ability of Eσs to tolerate longer or shorter spacer lengths
What is the structural and mechanistic basis of this enhanced tolerance for deviations from the optimal promoter? The presence of a C(−13) in the promoter (the only highly EσS.-specific promoter element not present in Eσ70-controlled promoters) as well as K173 in σs, i.e. the interaction between these two elements (Becker and Hengge-Aronis, 2001), is crucial for Eσs to utilize promoters with spacers of 18 bp and 19 bp better than Eσ70 (Fig. 5). But why is this interaction important?
It is known that the intervening region between the −10 and −35 boxes is relatively long to be accommodated by the sigma factor simultaneously at both positions. Therefore two bends are induced in the DNA: (i) at position −16, the DNA makes a sharp kink (37°) towards RNAP (partially due to a direct interaction between the extended −10 region and σ702.2/2.3/3.0); and (ii) the N-terminal part of the Zinc-binding domain of the β′ subunit contacts the DNA backbone around position −25, inducing a 8° kink (Murakami et al., 2002a; Borukhov and Nudler, 2003). At the same time, C(−13), which is crucial for σs selectivity (Becker and Hengge-Aronis, 2001), is directly interacting with residue K173 in σs (E458 is the corresponding amino acid in σ70; Becker and Hengge-Aronis, 2001). In the case of Eσs, this strong interaction could facilitate an optimal DNA bending, similar to that induced when an interaction between region 2.2–3 of sigma and the extended −10 region is available (Murakami et al., 2002a). Thereby, σ4.2 could reach even more distantly situated −35 elements. This scenario would also explain the strong resemblance between Eσs-dependent promoters and vegetative promoters possessing extended −10 elements, with respect to their preference for longer spacers (see Fig. 1A).
It is noteworthy that in our DNaseI footprinting experiments the hypersensitive site at −25 is more pronounced at the promoters carrying more degenerate −35 elements (Fig. 4D). Careful examination of previous DNaseI footprinting data revealed a strong correlation between the spacer length and the hypersensitivity at −25 (Murakami et al., 2002a). Promoters with a 16 bp spacer rarely showed any hypersensitivity at −25, whereas 18 bp spaced promoters nearly always do. This fits well to our footprinting results for the construct with a good −35 element consensus sequence (5/6 or 4/6 matches) and a 16 bp spacer, for which low hypersensitivity at −25 is apparent (Fig. 4D). On the contrary, the promoter derivatives with less well-conserved −35 elements exhibit increased hypersensitivity at −25 (probably due to DNA-bending), indicating a need for a rearrangement of the binding of RNAP at the promoter. Without a strong, specific recruitment of σ4.2 at the −35 element, the contact of the N-terminal part of the Zinc-binding domain of the β′ subunit at the DNA backbone might be of stronger importance for an efficient RNAP binding at the promoter.
The interaction between C(−13) and K173 in σs, on the other hand, is not essential for Eσs to tolerate shorter than optimal spacers (15, 16 bp; Fig. 5). Therefore, we tried to construct a more ‘σ70-like’σs, which would be incompetent of taking advantage of non-optimal spacers, using a genetic approach. The results indicate that the C-terminal domain of σs is important for establishing a preferential use of non-optimal spacers by Eσs. A truncated form of σs, missing the C-terminal nine amino acids (σs-321), and the I316P derivative are both shown to exhibit a reduced ability to operate with non-optimal spacer lengths (Fig. 6B), but are considerably more adequate than σ70 in utilizing a promoter completely devoid of a −35 hexamer (data not shown). I316P is a mutation introduced at the very end of the second helix–turn–helix motif in the 4.2 region of σs (Fig. 6A); the introduction of a proline at this position (i.e. the corresponding residue of σ70) is probably distorting the structure of the following last helix at the C-terminal end, which therefore would be different for the two sigmas. This short fifth helix in domain 4 of σ70 (Lambert et al., 2004), not observed in all structures of RNAP available (usually this region appears disordered; Campbell et al., 2002), is accounted to aid region 4–core RNAP interaction (Vassylyev et al., 2002). The region 4–core RNAP interaction initiates the conformational change that repositions σ4.2 in order to interact with the −35 element (Campbell et al., 2002; Kuznedelov et al., 2002; Murakami et al., 2002b). Therefore, we propose that σs exhibits a different kind of interaction with RNAP core that enables it to reach less ‘well-positioned’−35 hexamers. Evidence that support the first part of this hypothesis already exists, as σs supports a stronger interaction with the β-flap than σ70 does (Kuznedelov et al., 2002).
On the other hand, some of the residues in region 4.1 that are different for the two sigmas and were associated previously with the region 4–core RNAP interaction were found here to produce only a general loss of Eσs activity but no alteration in its ability to use non-optimal spacers (data not shown). An explanation for this can be that the docking of the β-flap in the hydrophobic cavity of domain σ4 is mediated through two distinct areas of σ4: the first centred in region 4.1 and the other being the C-terminal region of sigma. Changes in both regions of σs alter the interaction with core RNAP but only changes in the C-terminus of σs may alter the interaction in a way that affects the ability of Eσs to productively interact with non-optimally placed −35 elements.
It is worth mentioning that it has been previously reported that a truncation of the C-terminus of σs leads to a severe loss in its in vivo activity (Ohnuma et al., 2000), but has almost no effect in in vitro transcription from supercoiled templates (Gowrishankar et al., 2003). This implies an inability of the truncated σs to compete in vivo with the house-keeping sigma factor for core RNAP binding and probably also for promoter recognition. At the same time it adds credibility to our hypothesis that the C-terminus of σs assists its binding to the core RNAP and probably this interaction allows σ4.2 to find −35 elements situated further upstream or downstream.
Promoter features that compensate for the reduction in promoter activity associated with generating σs selectivity
As discussed above, increased Eσs selectivity of a promoter can be generated by deviations from the optimal promoter sequence and spacing. However, this strategy comes at the price of a reduced absolute promoter activity. As σs is active in stressed cells that grow slowly or not at all, this may not be a general problem. Yet, there might be Eσs-controlled promoters that nevertheless have to be relatively strong, and therefore strategies exist to overcome this constraint. One is the presence of a C(−13) in nearly all Eσs-dependent promoters (Weber et al., 2005), which is involved in an interaction (with K173 in σs), which not only is σs-specific but also produces a gain in absolute promoter activity (Becker and Hengge-Aronis, 2001).
Another point being established in this report is a high conservation of A/T tracts at certain positions of the spacer region of natural σs-dependent promoters (Fig. 7). In our synthetic promoters, expression is clearly enhanced by the introduction of A/T richness in the spacer region, especially when present directly upstream of the extended −10 region (Fig. 7). This gain in activity is the same for Eσs and Eσ70 (i.e. in the wild-type and in the rpoS mutant), which correlates well with a recent study, in which an A/T rich region, introduced upstream of the −10 region, resulted in transcriptional stimulation and CRP-independency of the lac promoter (Liu et al., 2004). This sigma factor independence of the effect of the A/T tracts also means that the relative ability of Eσs and Eσ70 to utilize these promoters, i.e. the sigma factor selectivity balance, remains unchanged when A/T tracts are introduced (Fig. 7). Thus, an 18 bp spaced promoter with an A/T tract upstream of the −10 region has increased Eσs selectivity, but nearly the same absolute activity as a 17 bp promoter that lacks such an A/T tract. In a study published during the revision of this paper, it was shown that after constructing a library for strong stationary phase-inducible and σs-dependent promoters by randomizing the sequence upstream of −10, A/T nucleotides were selected upstream of the extended −10 motif (TRTG) and downstream of the −35 region (Miksch et al., 2005). We therefore suggest that the frequent display of A/T tracts in the spacer regions of natural σs-dependent promoters (Fig. 7A) probably reflects the potential of A/T tracts to compensate for reduced promoter strength that comes along with the generation of Eσs selectivity by the ‘deviation from the optimum’ strategy.
Finally, Eσs can also exploit additional protein–DNA or protein–protein contacts that occur outside of the core promoter region for such compensatory increases in promoter activity. In the presence of an intact −35 element (and therefore relatively high overall activity), a promoter can increase its Eσs selectivity by possessing a distal UP-element half site (Typas and Hengge, 2005). Even when these cis-features are not present or not sufficient, further σs selectivity or promoter strength can be introduced by co-regulation by trans-acting factors. In this case Eσs can take advantage of sub-optimally placed activators to introduce selectivity (Germer et al., 2001), or exploits repressors that do not exert such a devastating influence on Eσs as they do on Eσ70 (Colland et al., 2000; Shin et al., 2005). In other words, Eσs again explores ‘sub-optimal territory’ in which the vegetative RNAP will have a serious disadvantage or even will be unable to function.
Bacterial strains and plasmids
The strains used in this study are derivatives of MC4100 (Casadaban, 1976). All the lacZ reporter fusions are located in single-copy at the chromosomal att(λ) site. Thus, phenotypically all the strains correspond to MC4100, except those in which rpoS mutations were introduced by P1 transduction (rpoS359::Tn10; Lange and Hengge-Aronis, 1991). The region of the gene fused to lacZ, mentioned in parenthesis, refers to the translational start as +1. For a detailed list of the strains used in this study see Table S2.
The synthetic promoter plasmids used in this study are all variants of psynp213, where a synthetic promoter derived from the core ptac is fused to lacZ in a translational fusion (for more details see Typas and Hengge, 2005). psynp215 is a plasmid carrying exactly the same promoter fused to lacZ, except for the extended −10 region, where it carries a TT(−14, −13) instead of CC(−14, 13). For constructing spacers with different lengths or mutating the −35 region, the same series of insertions/deletions as for synp213 were introduced also in synp215. The exact sequences of all promoters constructed with synp213 as a basis are shown in Figs 2, 4 and 7. psynpT213, a suitable psynp213 derivative for in vitro transcription, was constructed by cloning a rrnB (T1,T2) terminator in the place of lacZ (Typas and Hengge, 2005).
The σs variants used in this study were generated using pRL40.1 (rpoS cloned under the control of the ptac promoter in pRH800; Lange and Hengge-Aronis, 1994). When using this plasmid in a rpoS– background the levels of σs in stationary phase are similar to those obtained when σs is encoded in the chromosome.
Site-directed and insertional mutagenesis
For mutagenesis of the −35 element in the synthetic or natural promoter regions or changing the length and/or sequence of the spacer, a four-primer or two-step PCR protocol was used (described by Germer et al., 2001). The mutations generated can be viewed in the relevant figures (always the mutations introduced are indicated as bold) and the primers used for introducing them are listed in Table S2.
For constructing the different σs variants the four-primer or two-step PCR protocol was used again. The primers used for some of the mutants have been already described (Typas and Hengge, 2005); the rest can be found in Table S2.
Purified proteins for in vitro transcription
σ70 was purified as described previously (Gribskov and Burgess, 1983). RNAP core enzyme was purchased from Epicentre Technologies. Native σs without any tag was purified as described earlier (Typas and Hengge, 2005). In order to obtain reconstituted RNAP holoenzyme, equimolar amounts of core RNAP and sigma factor (or 1:5 ratio alternatively) were mixed in buffer A (10 mM Tris-HCl pH 8, 0.1 mM EDTA, 10 mM MgCl2, 200 mM KCl, 50% glycerol) and incubated for 20 min at 37°C. The reconstituted enzymes were then diluted to a final concentration of 350 nM in buffer B [40 mM Hepes (pH 8), 10 mM MgCl2, 100 mM potassium glutamate, 2.5 mM DTT and 0.5% BSA] and used further in the assay.
In vitro transcription assay
For the in vitro transcription assays with the synthetic promoters carrying different spacer lengths or no −35 box, a rrnB (T1,T2) terminator was cloned after the promoter in the place of lacZ in the derivatives of psynpT213 (Typas and Hengge, 2005). Transcription reactions contained 4 µl of 20 nM plasmid DNA in buffer B (without DTT), 4 µl reconstituted 350 nM RNAP in buffer B, and 4 µl prewarmed rNTP solution (1 mM ATP, CTP and GTP, 50 nM of non-labelled UTP, 0.8 µCi/250 nM of [α-32P] UTP, 0.725 mg ml−1 heparin). RNAP and plasmid DNA were pre-incubated for 8–12 min at 37°C in order to ensure open complex formation of both polymerases. Reactions were initiated by the addition of rNTPs and were terminated after 5 min at 37°C by the addition of loading buffer. Finally the samples were loaded onto a 7% denaturing sequencing gel. Radiolabelled transcripts ending at the rrnB (T2) terminator (no apparent transcripts were observed for the T1 terminator) were quantified on a Fuji phosphoimager using IMAGE GAUGE software and normalized against the vector encoded RNA I transcripts present in each lane.
DNaseI footprinting assay
In order to obtain reconstituted RNAP holoenzyme, tenfold excess of sigma factor and core RNAP were mixed in buffer A (see above) and incubated for 30 min at 37°C. 50 nM of reconstituted RNAP and 2 nM DNA (PCR fragment or plasmid DNA) were mixed in buffer B and then incubated for at least 20 min at 37°C. Complexes were then treated with DNaseI (0.6 µg ml−1, Sigma) for 20 s and the reaction was stopped by addition of non-specific DNA (salmon sperm; 1.75 ng µl−1) and rapid chilling on ice. Primer extension with 32P-radiolabelled primer followed. The primers used for the template and the non-template strands were 5′-CGCCAGCTGGCGAAAGGG-3′ and 5′-AGTGC CACCTGACGTCTAAG-3′ respectively. Finally the samples were loaded onto an 8 M urea/7% acrylamide sequencing gel.
β-Galactosidase was assayed using o-nitrophenyl-β-D-galactopyranoside (ONPG) as a substrate and is reported as µmol of o-nitrophenol min−1 mg−1 cellular protein (Miller, 1972). For all the fusion constructs, β-galactosidase activities were determined along the growth curve, but due to space restriction, only the data of middle stationary phase (4–6 h after entry) samples are presented (similar effects throughout the stationary phase).
The following supplementary material is available for this article online:
Table S1. σs-dependent promoters grouped according to the existence/positioning of the −35 element.