Although more than 30 Escherichia coli promoters utilize the RNA polymerase holoenzyme containing σS (EσS), and it is known that there is some overlap between the promoters recognized by EσS and by the major E. coli holoenzyme (Eσ70), the sequence elements responsible for promoter recognition by EσS are not well understood. To define the DNA sequences recognized best by EσSin vitro, we started with random DNA and enriched for EσS promoter sequences by multiple cycles of binding and selection. Surprisingly, the sequences selected by EσS contained the known consensus elements (−10 and −35 hexamers) for recognition by Eσ70. Using genetic and biochemical approaches, we show that EσS and Eσ70 do not achieve specificity through ‘best fit’ to different consensus promoter hexamers, the way that other forms of holoenzyme limit transcription to discrete sets of promoters. Rather, we suggest that EσS-specific promoters have sequences that differ significantly from the consensus in at least one of the recognition hexamers, and that promoter discrimination against Eσ70 is achieved, at least in part, by the two enzymes tolerating different deviations from consensus. DNA recognition by EσS versus Eσ70 thus presents an alternative solution to the problem of promoter selectivity.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Escherichia coli RNA polymerase (RNAP) consists of β, β', σ, ω and two α subunits. The core enzyme (α2ββ'ω) is capable of transcription elongation, whereas σ is required for specific promoter recognition and transcription initiation. Escherichia coli has seven different σ subunits: σ70 (encoded by rpoD), σ54 (rpoN), σ32 (rpoH), σS (rpoS), σE (rpoE), σ28 (rpoF) and σFecI (fecI), each responsible for directing RNAP to a different set of promoters, and, in so doing, changing the pattern of gene expression (Gross et al., 1998; Ishihama, 2000).
Eσ70, the holoenzyme responsible for transcription of the majority of genes in E. coli, recognizes three DNA sequence elements (all sequences provided below are for the non-template strand, 5′ to 3′). The −10 element, consensus TATAAT, is centred approximately 10 bp preceding the transcription start site and is recognized by regions 2.3 and 2.4 of σ70. The −35 element, consensus TTGACA, is 16–19 bp upstream from the −10 element and interacts with region 4.2 of σ70 (Gross et al., 1998). The UP element, an AT-rich region upstream of the −35 hexamer, is recognized by the C-terminal domains of the α subunits (Ross et al., 1993; Blatter et al., 1994). In addition, some Eσ70 promoters contain an extension of the −10 hexamer, TGTGn, immediately upstream of the −10 motif, that is recognized by region 2.5 of σ70 (Barne et al., 1997; Burr et al., 2000). Generally, the more of these elements present in a particular promoter and the more similar their sequences are to the consensus sequences, the better the binding by RNAP (Gross et al., 1998; Ross et al., 1998).
We chose an in vitro approach to determine the sequences recognized best by EσS, reasoning that understanding the biochemical behaviour of the enzyme in the absence of other factors, and then integrating these data with the extensive information about EσS obtained in vivo, might facilitate understanding of the behaviour of the enzyme in the complex environment of the cell. We generated a DNA library containing every possible sequence variant in the σ recognition regions and used EσS to select those sequences that bound preferentially in vitro. The −10 and −35 hexamers obtained were identical to those recognized by Eσ70, raising an interesting problem about promoter selectivity of the two holoenzymes. Based on genetic and biochemical approaches, we show that the two forms of holoenzyme prefer the same consensus hexamers for transcription, and that specificity can be achieved by differential tolerance of the enzyme for deviations from the consensus. This model is consistent with the observation that many stationary phase-specific genes are transcribed in vitro by both Eσ70 and EσS (Tanaka et al., 1993; Tanaka et al., 1995), and that σS shows considerable sequence homology to σ70 in the regions responsible for specific promoter recognition, 2.3, 2.4, 2.5 and 4.2 (Lonetto et al., 1992), but it differs conceptually from the standard model for promoter recognition in which each holoenzyme has its own discrete DNA recognition elements responsible for transcribing discrete sets of promoters.
Rationale and experimental design
Footprinting experiments with an EσS-specific promoter, bolA1, indicated previously that EσS protects about 70 bp of promoter sequence (from about −57 to +17), similar in length and position to that protected by Eσ70 (Nguyen and Burgess, 1997). Based on this, and on the similarity in the amino acid sequences of the two σ factors in the regions responsible for DNA binding (regions 2.3–2.5, and 4.2), we decided to sample a population of DNA sequences containing all possible sequences from −38 to +2.
One copy of each possible 40 bp sequence variant would generate about 1 kg of DNA, far exceeding what can be manipulated experimentally even in vitro. Therefore, we performed the in vitro selection in two steps, randomizing a 20 bp segment to generate libraries containing at least one copy of 420 (∼1012) different sequences in each step. In the first step, a selection was performed with promoter fragments randomized from −18 to +2. This was followed by a second selection with promoter fragments randomized from −38 to −19 in the context of the selected −18 to +2 sequence.
Highly purified EσS, containing no detectable σ70 or other σ factors (Nguyen et al., 1993), was added to the library of DNA fragments under limiting conditions so that only 0.5–5% of the DNA fragments formed complexes with RNAP. After separation of the complexes from free DNA on a non-denaturing polyacrylamide gel, DNA from the bound fraction was eluted, PCR-amplified and the procedure was repeated for multiple cycles to enrich for the best binding sequences (Fig. 1A). As EσS-DNA complexes migrated to a location different from Eσ70-DNA complexes (Fig. 1B), contamination with DNA fragments bound by undetectable levels of Eσ70 (if present) would be minimized at each cycle in the enrichment. We note that our binding assay selects for sequences resulting in the maximal rate of open complex formation, but these sequences need not be optimal for each step leading to individual intermediates in the transcription initiation pathway.
Selection of the −10 promoter region by EσS
To identify EσS promoter determinants in the region containing the −10 element, EσS was incubated with a promoter fragment library containing randomized sequences from −18 to +2 and bolA1 promoter sequences from −54 to −19 and from +3 to +16 (−10 rnd;Fig. 2A). After 17 cycles of selection by EσS, 16 DNA fragments were cloned and sequenced (see Experimental procedures). A close match to the 11 bp sequence TGTGCTATA(C/A)T was present on each of 12 promoter fragments (Fig. 2B and C; the other four fragments contained large deletions and probably were selected for artifactual reasons described in Experimental procedures). The 11-mer sequence (e.g. in −10 sel #1) showed limited similarity to the bolA1 promoter but contained the Eσ70-like −10 hexamers, TATAAT or TATACT (Fig. 2). The 11-mer also contained the consensus ‘extended −10′ motif TGTG, that is found upstream of the −10 hexamer in a subset of E. coli promoters (Burr et al., 2000), and a C between the extended −10 motif and the −10 hexamer. The other 9 bp of the randomized region differed among the selected promoter fragments, indicating the promoters were of independent origin.
The position of the selected 11-mer was relatively constant within the randomized region (from −17 to −7 in eight cases, from −16 to −6 in three cases, and from −18 to −8 in one case). These data suggest that sequences outside the randomized region contribute to positioning EσS on the DNA and, therefore, have an important role in recognition by EσS.
Selection of the −35 promoter region by EσS
To identify promoter determinants for EσS binding upstream of −18, a second selection was performed with a fragment library (−35 rnd;Fig. 3A) containing randomized basepairs from −38 to −19. Flanking sequences in the starting population derived from the −10 sel #1 promoter (−18 to +16; Fig. 2) and from the ‘SUB’ promoter (−54 to −39; Rao et al., 1994). The −54 to −39 SUB sequence does not contain RNAP recognition motifs (Rao et al., 1994) and was introduced to eliminate interference from −35-like sequences upstream of position −38 in bolA1 (Nguyen et al., 1993).
Twenty-two promoter fragments were cloned and sequenced after 17 cycles of selection. Nineteen promoter fragments of the correct length contained the sequence CTTGACA from positions −36 to −30, exactly 17 bp upstream from the −10 hexamer (Fig. 3B and C); the other three fragments contained large insertions and probably were selected for artifactual reasons, as described in Experimental procedures). In addition, there was a strong preference for A residues at positions −29 and −28 and a lesser preference for T at −27. Positions −38, −37 and −26 to −19 were different from clone to clone, confirming that the selected promoters were of independent origin. The selected −35 promoter region showed no similarity to the same region of the bolA1 promoter, but it contained the consensus −35 hexamer for recognition by Eσ70.
Characterization of the selected promoters in vitro
To analyse the effects of the selected sequences on RNAP binding and transcription, promoters were constructed containing the selected −38 to −19 and −18 to +2 regions (either individually or combined together). To facilitate direct comparison with the bolA1 promoter, each construct contained sequences from bolA1 at all positions outside the randomized regions (Fig. 4A). For simplicity, these constructs are referred to as the ‘in vitro selected’ promoters, including −10 con (containing the selected −18 to + 2 region from −10 sel #1; Fig. 2), −35 con (containing the selected −38 to −19 region from −35 sel #1; Fig. 3) and full con (containing both selected regions).
To verify that the selection resulted in DNA sequences that bound EσS better than bolA1, we compared the affinities of the DNA fragments for EσS under conditions identical to those used for the selection (see Experimental procedures). The three selected promoters bound EσS substantially better than bolA1 (data not shown), confirming that the selection worked as expected. The relative order of binding was full con > −35 con > −10 con >> bolA1.
To assess whether the increase in EσS binding by the selected promoters relative to bolA1 resulted in increased transcription, the activities of the three selected promoters were compared with the activity of the bolA1 promoter in vitro. When transcribed by EσS, all three selected promoters were much more active than bolA1, and the −35 con promoter was the strongest, in buffers containing 100–800 mM K-glutamate or 10–400 mM KCl (Fig. 4B and data not shown). The difference in activity between bolA1 and the selected promoters was less dramatic at high RNAP concentrations (data not shown). As the selected promoters also contained the known consensus sequences for Eσ70 recognition, we tested their transcription by Eσ70in vitro under the same conditions used for EσS. The selected promoters were transcribed very efficiently by Eσ70, whereas transcription from the bolA1 promoter was hardly detectable (Fig. 4C).
The selected promoters were transcribed similarly by both holoenzymes in buffers containing 10–400 mM KCl (Fig. 4D and data not shown) or 100–800 mM K-glutamate (Fig. 4E and data not shown). In no case were the promoters transcribed more than about twofold better by EσS holoenzyme than by Eσ70. Therefore, in contrast to the situation for some osmotically regulated promoters that are transcribed only by EσS at high concentrations of K-glutamate (Ding et al., 1995; Kusano and Ishihama, 1997), the selected promoters did not exhibit a large preference for EσS at high osmolarity.
Because EσS and Eσ70 were prepared by adding saturating amounts of σ factor to aliquots of the same preparation of core enzyme, it is valid to compare the relative activities of EσS and Eσ70 on each of the promoters (see Experimental procedures). For each selected promoter, the level of transcription by EσS and Eσ70 was very similar, although the three selected promoters differed in activity from each other. In contrast, the wild-type bolA1 promoter was transcribed substantially better by EσS than by Eσ70, whereas the plasmid-derived RNA I promoter was transcribed much better by Eσ70 than by EσS. We conclude that the in vitro selected promoters are transcribed much better than the bolA1 promoter in vitro, but they do not exhibit specificity for EσS over Eσ70.
Transcription from the full consensus promoter (full con) was always weaker than from the promoter containing only the consensus −35 region, −35 con(Fig. 4B and C). The full con promoter's relative inactivity results from a defect in promoter escape (T. Gaal, R.L. Gourse and N. Shimamoto, unpublished).
Characterization of the in vitro selected promoters in vivo
We measured the activities of the bolA1, −10 con, −35 con and full con promoters as promoter–lacZ fusions to determine whether the relative activities observed in vitro were also observed in vivo, and also to determine whether the promoters displayed the increase in promoter activity in stationary phase characteristic of EσS-dependent promoters (Fig. 5). The relative activities of the promoters were consistent with the activities observed in vitro: the −35 con promoter was three to fivefold stronger than the −10 con or full con promoters, and all three were much stronger than the bolA1 promoter, both in exponential and in stationary phase (Fig. 5A and C).
The bolA1 promoter fragment was also introduced into a lacZ fusion system that produces higher β-galactosidase activity and lower background (‘System II’; Simons et al., 1987; Rao et al., 1994), allowing a more accurate estimate of transcription from weak promoters (see Fig. 5A and C inset). bolA1 promoter activity increased about 18-fold in stationary phase relative to exponential phase (Fig. 5B). In contrast, there was less than a twofold increase in bolA1 promoter activity in stationary phase in a strain lacking rpoS(Fig. 5D), confirming previous reports that the bolA1 promoter is EσS-dependent (e.g. Lange and Hengge-Aronis, 1991).
Although the −10 con, −35 con and full con promoters were selected for binding to EσS, they were transcribed efficiently even when EσS was not present (Fig. 5C), and their activities increased only slightly in stationary phase (Fig. 5D). As the −10 con, −35 con and full con promoters are transcribed very efficiently by both EσS and Eσ70in vitro, and as the promoters contain the consensus hexamers recognized by Eσ70, we conclude that these promoters are transcribed by Eσ70 in exponential phase and are probably transcribed by either EσS or Eσ70 (or both) in stationary phase. [σ70 is more abundant than σS, even in stationary phase (Jishage et al., 1996)]. The residual rpoS-independent small increase observed for all four promoters in stationary phase could derive from decreased competition for limiting Eσ70 when rRNA transcription drops in slow or non-growing cells (see Barker et al., 2001).
Interactions of EσS and Eσ70 RNAPs with specific promoter positions
The results from the in vitro selections described above suggested that both EσS and Eσ70 prefer the same −10, −35 and extended −10 consensus sequences. However, the identification of sequences in in vitro selection experiments does not imply that each selected position has the same relative importance to RNAP binding, as even small increments to binding will result in selection, nor do these experiments address whether EσS and Eσ70 interact with a particular promoter in the same manner. To address whether the two enzymes recognize the same promoter sequence differently, we used interference footprinting, a technique in which DNA fragments with modifications at different positions in the population can all be assayed for protein binding at the same time. Specifically, if a modification at a certain position interferes with RNAP binding, the promoter fragment containing that modification will be underrepresented in the bound population. Interference footprinting thus allows estimation of the relative contribution of individual positions in a promoter to RNAP binding (although of course the effects of every functional group or even every basepair are not tested by any one interference probe).
We tested the effects of two different modifications that alter the major groove surfaces of thymine or adenine in DNA on binding by EσS and Eσ70. Uracil is identical to thymine except it lacks the C5 methyl group in the major groove (Devchand et al., 1993), whereas 7-deaza-7-nitro-adenine (A*) introduces a bulky nitro group into the major groove (Min et al., 1996). dUTP, or dA*, was incorporated into promoter fragments at low frequency by polymerase chain reaction (PCR) (Ross et al., 2001). The effects of these substitutions on Eσ70 binding to the rrnB P1 promoter have been characterized previously (Ross et al., 2001). Preliminary experiments suggested that single base modifications in the full con promoter did not interfere enough with binding by either holoenzyme to alter the fraction bound in gel-shift assays (data not shown). Therefore, the −35 con and rrnB P1 promoters were used instead; these promoters bound both holoenzymes, but with lower affinity than the full con promoter, so that uracil or A* substitutions resulted in changes in the fraction of the DNA population bound by RNAP.
Uracil incorporation at three non-template strand positions in the −35 con promoter (one in the −10 region at −12, and two in the −35 region at −34 and −35) reduced binding by both EσS and Eσ70(Fig. 6A). Uracil at −12 virtually eliminated binding by both enzymes. In contrast, uracil at either −34 or −35 greatly reduced Eσ70 binding, but only moderately reduced EσS binding. Similar differential effects of uracil substitution at −34 and −35 on Eσ70 versus EσS binding were also observed on the rrnB P1 promoter (data not shown). These results suggested that even though both holoenzymes prefer the same promoter sequence, they interact with that promoter differently.
The effects of A* substitutions for adenine in the non-template strand on Eσ70 and EσS binding to the rrnB P1 promoter are illustrated in Fig. 6B; qualitatively similar results were obtained for the −10 con and −35 con promoters (data not shown). Modification of either of two positions in the −10 hexamer reduced RNAP binding, whereas no effects were observed in the −35 region. Binding of the two holoenzymes was affected differentially at one of the −10 region positions, the highly conserved A−11 in the −10 hexamer. At this position, A* decreased EσS binding by about 60% but decreased Eσ70 binding by only about 30%. As with the uracil interference experiments, the A* footprints indicate that EσS and Eσ70 bind differently to the same promoter sequence. The less than total inhibition observed from introduction of A* for the highly conserved A−11 could be attributable to the fact that the N1 position on the base is most crucial for A−11 function (Matlock and Heyduk, 2000), not the 6 and 7 positions on the base altered in A*.
A* substitution at A−9 had a smaller effect than at A−11, and A* substitution at A−8 had a larger effect than at A−11, but in both cases the effects were similar for the two holoenzymes. The strong inhibition of RNAP binding by A* at −8 could be attributable to steric clash by introduction of the bulky nitro group, as typically mutations at −8 have less severe effects on RNAP interactions than at −11.
Basis for promoter selectivity by EσS versus Eσ70
The similarity in the intrinsic promoter recognition properties of EσS and Eσ70 raises the issue of how some promoters could be almost entirely dependent on EσS. That is, what accounts for some promoters being transcribed only by EσS when both EσS and Eσ70 are present in stationary phase, and for these same promoters not being transcribed by Eσ70 in exponential phase? In this section, we propose a model to address these questions and then we provide experimental support for this proposal.
Our data suggest that the preferred hexamer sequences might be identical for EσS and Eσ70. However, naturally occurring promoter sequences rarely match the consensus perfectly for a particular RNAP. Differences from the Eσ70 consensus sequence, in conjunction with the effects of specific activators and repressors, are responsible for differences in levels of transcription initiation from individual Eσ70 promoters. The interference footprints shown above indicate that certain modifications of the −10 and −35 hexamers affect binding by EσS and Eσ70 differently. Therefore, to explain differences in transcription by EσS versus Eσ70, we suggest that non-preferred basepairs in a promoter sequence might have different effects on recognition by the two enzymes.
To test this general concept, we chose two positions in the −35 element for detailed examination. The choice of these positions was dictated by the results of our interference footprinting studies (Fig. 6A) indicating that the loss of the methyl groups at positions −34 and −35 affect discrimination by Eσ70 versus EσS and by previous genetic evidence suggesting that C substitutions at −34 and −35 result in specificity for EσS (Wise et al., 1996; Bordes et al., 2000). To confirm that the identities of −34 and −35 contribute to discrimination by Eσ70 versus EσS, we constructed a CC-35 con promoter (i.e. a −35 con promoter containing T-34C and T-35C substitutions) fused to lacZ. The promoter DNA fragment used in this fusion extended only to −40 to eliminate any contribution to activity in vivo from a weak Eσ70-dependent promoter upstream of the −35 hexamer that derives from bolA1 sequences (see Fig. 4C).
CC−35 con was considerably weaker than −35 con and displayed the characteristics of an EσS-specific promoter in vivo(Fig. 7A). Transcription from CC−35 con greatly increased in stationary relative to exponential phase, and this increase was almost completely dependent on the presence of rpoS. (The promoter endpoints and the lacZ reporter system used in Fig. 7 differ from those used in Fig. 5. Thus, the absolute β-galactosidase activities in the two figures should not be compared directly (see also Experimental procedures). We also tested transcription from CC−35 con with EσS and Eσ70in vitro(Fig. 7B). Consistent with the results obtained in vivo, CC−35 con was preferentially utilized by EσSin vitro. EσS tolerated the CC substitution much better than did Eσ70 under every condition tested.
To test whether EσS can tolerate only the CC substitution at −34 and −35 or whether other substitutions at these positions are tolerated as well in −35 con, we generated all possible single and some two basepair substitutions for the TT at −34 and −35, and we compared transcription from each mutant by EσSin vitro to that from −35 con(Fig. 8A). None of the mutant promoters was more active than −35 con, confirming that TT at these two positions is optimal for recognition by EσS. None of the substitutions decreased transcription by EσS more than about twofold. CC−35 con was transcribed by EσS worse than most of these promoters, indicating that CCGACA (Wise et al., 1996) is not the consensus −35 hexamer for recognition by EσS. As the promoter with the single G substitution at −35 was transcribed as well as −35 con in vitro, but promoters with this substitution were not obtained in the in vitro selection, it is possible that the in vitro transcription assay is not sufficiently sensitive to detect the slight preference for the consensus T at this position, or there is an effect of promoter context in this case.
In contrast to their small effects on transcription by EσS, the TG, AA, CC, or GG substitutions decreased transcription by Eσ70 as much as 10-fold, resulting in a six to 10-fold preference for EσS over Eσ70(Fig. 8B and C). These results are consistent with the model that discrimination between EσS and Eσ70 is accomplished by differential tolerance of the two enzymes for specific deviations from the consensus sequence.
Eσ70 and EσS recognize the same consensus hexamer sequences
We have studied the intrinsic promoter recognition properties of EσS using an in vitro selection to identify the DNA sequence determinants for EσS binding. In contrast to expectations based on previous proposals (see Introduction), the binding sites selected by EσS contained the consensus hexamer sequences for binding Eσ70, and they initiated transcription at high levels by both holoenzymes.
Many EσS-dependent promoters rely predominantly on recognition of the −10 region (e.g. Tanaka et al., 1995; Espinosa-Urgel et al., 1996; Colland et al., 1999; Lee and Gralla, 2001). However, the results of the in vitro selection and the extraordinary strength of the −35 con promoter, relative to the bolA1 promoter, when transcribed by EσSin vitro, suggest that the −35 element can also play a prominent role in recognition by EσS in some promoter contexts. In fact, some EσS-dependent promoters (e.g. osmE;Bordes et al., 2000) have almost consensus −35 regions and appear to be strongly dependent on −35 hexamer interactions.
Differential tolerance of RNAP holoenzymes for deviations from the same consensus hexamers
As both Eσ70 and EσS prefer the same consensus hexamer sequences, there must be a mechanism for preventing transcription from stationary phase-specific promoters by Eσ70 in exponential phase. (Preventing transcription from these promoters by EσS in exponential phase is accomplished by limiting expression of σS.) We suggest that holoenzyme specificity might rely, at least in part (see below), on tolerance of the two enzymes for different deviations from the same consensus sequences. This hypothesis is supported by interference footprinting studies showing that EσS and Eσ70 recognize the same promoter sequence differently (Fig. 6) and by the effects of certain promoter mutations constructed as a test of this concept (Figs. 7 and 8). We found that, in the context of the −35 con promoter, substitutions of TT to TG, CC, AA, or GG at the first two positions of the −35 hexamer resulted in transcription in vitro that was strongly dependent on EσS, in contrast to −35 con. These results are consistent with genetic studies on the proU promoter, in which TT to CC substitutions at the same positions strongly reduced transcription and made it dependent on rpoS in vivo, and on the osmY promoter, in which CC to TT substitutions decreased discrimination between EσS and Eσ70in vivo (Wise et al., 1996).
We did not evaluate the relative preferences of EσS and Eσ70 for each possible basepair at every position in the consensus hexamers. However, several results strongly suggest that, in addition to positions in the −35 hexamer, there are positions in the −10 hexamer at which interactions with the two RNAPs are differentially affected by the same base modification or base substitution. These include the results of our interference footprints (Fig. 6), the fact that the promoter differing from bolA1 only in the −10 region (−10 con) is recognized by EσS but has lost its EσS specificity (Fig. 4), and recent studies showing there is differential binding of EσS and Eσ70 to fork-junction templates with substitutions in the −10 region (Lee and Gralla, 2001).
At position −8, A and C were about equally represented in the sequences selected by EσS, whereas A is considered the consensus for Eσ70-dependent promoters. Although there are some differences between σ70 and σS in the region predicted to interact with position −8 (region 2.3; Lonetto et al., 1992; Malhotra et al., 1996; Fenton et al., 2000), we suggest that the identity of −8 does not play a major role in discrimination between the two holoenzymes. We constructed promoters differing only by A or C at this position and found they were transcribed similarly by the two enzymes in vitro (data not shown), and previous reports suggest that both Eσ70 and EσS tolerate A or C at −8 (Oliphant and Struhl, 1988; Kolb et al., 1995; Tanaka et al., 1995; Espinosa-Urgel et al., 1996). Recognition of either A or C at −8 by both enzymes could potentially be attributable to an interaction with the same functional group on both bases, for example the amino group on position 6 of A, or 4 of C.
Other potential contributions to differential promoter recognition by EσS and Eσ70
Recent studies have proposed that a C at position −13 (i.e. the position just upstream from the −10 hexamer) is preferred by EσS, and G−13 is preferred by Eσ70 in certain contexts, for example in the csiD and osmY promoters (Becker and Hengge-Aronis, 2001) and in fork-junction templates (Lee and Gralla, 2001). The results of our in vitro selection strongly support the preference of EσS for C at −13. However, C−13 is clearly not sufficient to discriminate against recognition by Eσ70 in the context of the in vitro selected promoters described here (Figs. 4 and 5). In addition, the rrn P1 promoters that are transcribed much more efficiently by Eσ70 than by EσS (data not shown) contain a C at this position.
Residues in addition to the consensus hexamers and C-13 were selected by EσS in our in vitro binding experiments, including a cytosine at the upstream flank of the −35 hexamer (−36), two adenines at the positions just downstream of the −35 hexamer (−29 and −28) and the sequence TGTG at positions −14 to −17 (the extended −10 region); −36, −29, and −28 are not usually considered as being determinants of Eσ70 binding and, therefore, it is formally possible that they might contribute to specific EσS recognition. However, we consider it improbable that these basepairs, by themselves, are responsible for discrimination between EσS and Eσ70, as (i) C is also preferred by Eσ70 at −36 in the lac and rrnB P1 promoters (Reznikoff, 1976; Josaitis et al., 1990), and (ii) A to G substitutions for the A residues at −29 and −28, at least in the context of the −35 con promoter, did not affect activity substantially in vitro with either holoenzyme (data not shown). [We note, however, that an A-tract downstream of the −35 hexamer was identified as a requirement for efficient rRNA promoter activity by Chlamydia trachomatis RNAP (Tan et al., 1998)]. The interaction between the extended −10 motif (TGTG) and region 2.5 of σ has been well documented as an important determinant of both EσS- and Eσ70-dependent transcription (Voskuil et al., 1995; Barne et al., 1997; Becker et al., 1999; Colland et al., 1999; Burr et al., 2000).
A spacer length of 17 bp between the −10 and −35 hexamers was strongly favoured for recognition by EσS in our in vitro selection. We found that Eσ70 tolerated a 16 bp spacer in the context of the rrnB P1 promoter much better than EσS, but a 16 bp spacer reduced the activity of the full con promoter approximately the same for both holoenzymes (data not shown). Therefore, spacer-length could contribute to EσS versus Eσ70 discrimination at some promoters, but our preliminary studies indicate that the effect of spacer length on the two holoenzymes is context-dependent and complex.
Finally, our hypothesis does not exclude a role for additional DNA-binding proteins in promoter discrimination by EσS and Eσ70. That is, expression of some promoters (although not those investigated here) is influenced by differential effects of transcription factors or nucleoid-associated proteins on the two holoenzymes (Arnqvist et al., 1994; Bouvier et al., 1998; Colland et al., 2000).
σ determinants for differential recognition of the same consensus hexamers
The similarity between the EσS and Eσ70 consensus sequences is consistent with the similarity of the motifs in the two σ factors responsible for promoter recognition. Sixteen out of 28 amino acids in region 4.2 (residues 572–599 of σ70) and 25 out of 40 amino acids in regions 2.3 and 2.4 (residues 417–456 of σ70) are identical in the two σ factors (Lonetto et al., 1992), including several residues implicated in direct interactions with specific basepairs in the −10 and −35 hexamers (e.g. Q437 in region 2.4, and R584 and R588 in region 4.2; σ70 numbering, Gross et al., 1998).
Superimposed on the overall conservation between the two σ factors in the regions responsible for DNA binding are discrete regions of divergence that could contribute to promoter discrimination. Two clusters of four amino acids in region 4.2 (residues 578–581 and 591–594 of σ70) are virtually identical among σ70 members from different species and among different σS members from different species, but these sequences differ between σ70 and σS (Lonetto et al., 1992) and could potentially play a role in discrimination between deviations from the consensus −35 hexamer. K173 of σS has been implicated in recognition of C−13 (Becker and Hengge-Aronis, 2001). The corresponding residue in σ70 is different (E458), consistent with their different DNA recognition preferences. As interactions with the −10 element are complex, involving binding to both single-stranded and double-stranded DNA in multiple steps during the process of transcription initiation, it is more difficult with present information to predict specific amino acid residues in σ70 and σS that might play a role in differential −10 region recognition.
Promoters recognized by both EσS and Eσ70
The observed overlap in the sequences recognized by EσS and Eσ70 potentially allows some promoters to be transcribed by both holoenzymes under some conditions, and yet to be transcribed by only one or the other holoenzyme under other conditions. Even though a promoter might be transcribed by both enzymes under some conditions, changes in the environment that alter template geometry (e.g. superhelicity) or solute concentrations (e.g. anions or cations) might be sufficient to alter the ratio of transcription by the two holoenzymes (Ishihama, 2000).
It has been reported that the two holoenzymes possess different tolerances for high concentrations of K-glutamate or trehalose, and that this favours specific promoter recognition by EσS (Ding et al., 1995; Kusano and Ishihama, 1997; Nguyen and Burgess, 1997). We compared the activities of EσS and Eσ70 on the bolA1, −10 con, −35 con, and full con promoters in buffers containing 100–800 mM K-glutamate or 10–400 mM KCl (data not shown). As predicted from previous studies (Leirmo et al., 1987), the anion glutamate was less disruptive to RNAP–promoter interactions than chloride, so transcription by both holoenzymes was generally higher at the same cation concentration in glutamate buffers. Tolerance for high salt concentrations may contribute to the ability of certain promoters to be transcribed only by EσS (Ding et al., 1995; Kusano and Ishihama, 1997). However, the in vitro selected promoters described here are transcribed by both holoenzymes at high osmolarity; high salt does not result in holoenzyme specificity.
The degree of negative supercoiling of the bacterial chromosome, on average, is somewhat lower in stationary phase than in exponential growth, and it has been suggested that reduced supercoiling may enhance transcription by EσS rather than by Eσ70 (Kusano et al., 1996). Our promoter selections were performed on DNA fragments (linking number = 0), but the promoters were also transcribed efficiently by both holoenzymes on supercoiled plasmid templates (linking number = ≈−0.06). Therefore, it is probable that the same consensus hexamers are recognized best by both EσS and Eσ70 on templates with a wide range of superhelicities. Nevertheless, as with changes in osmolarity, conditions that alter template geometry could potentially favour transcription by one holoenzyme or the other in a specific promoter context.
We have shown for the first time that two RNAP holoenzymes present in the same organism recognize the same consensus hexamer sequences. This poses a potential problem for promoter specificity, but we have proposed a model that might resolve this dilemma. The principle of tolerance for different deviations from the same consensus sequence was recognized long ago as the basis for differential binding of the Cro and cI repressor proteins from bacteriophages λ and 434 (e.g. Hochschild et al., 1986; Harrison and Aggarwal, 1990; Albright and Matthews, 1998). Cro and cI bind to the same consensus operator sequences, but the two proteins prefer different non-consensus bases at particular positions in the DNA sequence.
Differential tolerance of two holoenzymes for certain deviations from the same consensus is apparently not the method used for discrimination between the other E. coli RNAP holoenzymes, as the other E. coliσ factors have diverged from each other more than σS and σ70 and recognize qualitatively different DNA sequences. Even in the case of promoters recognized by EσS versus Eσ70, context is important to the effects of individual promoter positions on RNAP recognition, and thus it is difficult to predict which deviations from consensus account for discrimination simply by inspection of a particular promoter sequence. However, we suggest that the concept of tolerance for different deviations from the same consensus sequence as a mechanism for achieving specificity should be considered in transcription studies of organisms with multiple closely related RNAP holoenzymes, a common feature in bacteria. For example, recent genome analysis suggests that Streptomyces coelicolor might have as many as 40 ECF σ factors (http://www.sanger.ac.uk/Projects/S_coelicolor/). We speculate that the principle described here for discrimination between E. coli EσS and Eσ70 might be utilized for discrimination between members of σ factor families in other bacteria.
Promoter fragment library
In vitro selections with EσS were performed with libraries of 86 bp promoter fragments containing 20 bp of random DNA sequence. Non-template strand oligonucleotides (Integrated DNA Technologies) for the −10 region selection contained (from 5′ to 3′) an EcoR1 site, bolA1 promoter sequences from −54 to −19, random sequences from −18 to +2 generated from equimolar mixtures of the four nucleotides, bolA1 sequences from +3 to +16, and a HindIII site. Non-template strand oligonucleotides for the −35 region selection contained an EcoR1 site, the ‘SUB’ sequence (Rao et al., 1994) from −54 to −39, random sequences from −38 to −19, selected −10 region sequence #1 (Fig. 2B) from −18 to +2, bolA1 sequences from +3 to +16 and a Hind III site. The template strand for each of the two selections was generated by annealing 60 pmol of the non-template strand with 60 pmol of a 21-mer complementary to the sequence from +3 to the Hind III site. The reactions were incubated in 25 µl Sequenase buffer (US Biochemical) for 5 min at 95°C, slowly cooled to room temperature, dNTPs were added to 1 mM, and the annealed oligos were extended with 20 units of Sequenase for 20 min at 37°C. The DNAs were digested with EcoRI and HindIII, and aliquots were cloned into M13mp18 for sequence analysis before selection with EσS to ensure that the randomized regions contained approximately equal percentages of all four bases. The rest of the DNA was labelled at both ends with [α−32P]-dATP using Sequenase before use in the selection.
EσS was prepared from core RNAP and highly purified σS (Nguyen et al., 1993). The preparations contained no detectable other σ factors as judged on silver-stained gels (data not shown). EσS was reconstituted by mixing purified core with a fivefold excess of σS (1 µM) for 45 min at 30°C. Eσ70 was reconstituted using the same core RNAP preparation and a fivefold excess of purified σ70. Increasing the concentration of either σ factor further did not increase transcription (data not shown). Therefore, the two holoenzyme preparations contain the same number of active RNAPs (unless there are σ molecules that bind to core but are not active in transcription).
In vitro selection
The selection consisted of repeated cycles of RNAP binding, separation of the bound population from free DNA by electrophoresis on 4% polyacrylamide gels in 0.5X Tris-Borate-EDTA buffer for 2–3 h at ∼10 V cm−1 and PCR amplification (Fig. 1). For the initial 50 µl binding reaction, 2 µg (0.7 µM) of the double-stranded DNA fragment library was incubated with 20 nM EσS in 50 mM HEPES (pH 7.0), 100 mM KCl, 10 mM Mg-acetate, 0.1 mM dithiothreitol (DTT), 100 µg ml−1 BSA, 5% glycerol at room temperature for 20 min. Then, 25 µg ml−1 heparin was added to prevent further RNAP binding, and the samples were immediately loaded on the gel. The RNAP–promoter complexes (0.5–5% of the total DNA) were visualized by phosphorimaging, excised, eluted, extracted with phenol and precipitated with ethanol. The recovered DNA was used as a template for PCR amplification using Taq DNA polymerase for 15 cycles (95°C for 1 min; 55°C for 1 min; 72°C for 1 min). For the −10 selection, the PCR primers were complementary to 20 nt at each end of the promoter DNA fragment, whereas in the −35 selection, the primers were complementary to all except the randomized positions (−38 to −19). Aliquots were checked by gel electrophoresis on 4% agarose gels and staining with ethidium bromide for successful amplification. The amplified DNA was digested with HindIII, end-labelled and used for the next cycle of RNAP binding. The binding time was progressively shortened, and the RNAP concentration was gradually decreased, in subsequent selection cycles, until in the last round the reaction time was 2 min and the RNAP concentration was 4 nM. The progress of the selection was followed after 10, 14 and 17 cycles by cloning samples into M13 and sequencing a number of individual clones from the selected population. The selection was stopped after 17 cycles because inspection indicated that an obvious consensus sequence had been reached.
Heteroduplexes (which bind RNAP very efficiently) sometimes arose in the final cycles of the PCR from annealing of single-stranded DNAs that contained different sequences in the randomized region or from annealing of insertions and deletions that arose from mistakes during oligonucleotide synthesis or amplification. These appear on agarose gels as a blurred region of stained material, migrating slower than the band of the expected size. If contamination from heteroduplexes was detected, 5 µl of the reaction was used as template for an additional two cycles of PCR amplification in a 50 µl reaction.
Strains and plasmids
Plasmids and strains are listed in Table 1. Cloning with M13 and plasmids was carried out using standard techniques. Plasmid templates for in vitro transcription were constructed by insertion of promoter fragments into pRLG770 (Ross et al., 1990). Site-directed promoter mutations in −35 con were generated by PCR using plasmid pRLG3748 as template and primers containing the desired sequence alterations. DNA sequencing was performed using a Sequenase kit supplied by US Biochemicals.
λ lysogens were constructed containing promoter–lacZ fusions as described (Rao et al., 1994). Two promoter–lacZ fusion systems were used; ‘System II’ (Rao et al., 1994) was used for measuring transcription from the bolA1 and CC−35 con promoters as indicated. This fusion system has a very low background, but it cannot tolerate very strong promoters. ‘System I’ (Rao et al., 1994), which has higher background but can accommodate strong promoters, was used in all other cases. The same promoter makes at least 6.7-fold more β-galactosidase in System II than in System I. The CC−35 con and bolA1 promoters in the lacZ fusions used in Fig. 7(Table 1; RLG5852, RLG5857, RLG5861 and RLG5862) contain sequences only from −40 to +2 to eliminate interference from the weak Eσ70 binding site upstream of the −35 hexamer (see Fig. 4 and Nguyen et al., 1993). As the reporter systems and the downstream endpoints in the promoters used for these fusions (+2) are different from those used for the fusions in Fig. 5 (+16), probably resulting in different mRNA half-lives for the resulting transcripts, the β-galactosidase activities shown in Fig. 7 should not be compared directly with those shown in Fig. 5. Strains lacking σS were constructed by transduction of rpoS::Tn10 from RLG3237 (= UM122; Xu and Johnson, 1995) with P1vir.
Measurement of promoter activity in vivo
Cells were grown in Luria–Bertani (LB) medium at 30°C, and β-galactosidase activity was measured at regular intervals throughout a growth cycle. To estimate promoter activity in exponential phase, cells were grown to an A600 of ≈ 0.25–0.5 and then diluted 100-fold and grown again to ensure that β-galactosidase, accumulated previously during stationary phase, was minimized. The promoter activity at A600 0.25 and 0.5 was essentially identical. To estimate promoter activity in stationary phase, β-galactosidase activity was measured from cells grown to an A600 of 2.6 and 4.0 (in which β-galactosidase activity was essentially identical).
Interference footprinting was performed as described in detail elsewhere (Ross et al., 2001). The DNA contained, on average, one modified base (U instead of T; Devchand et al., 1993; or A* instead of A; Min et al., 1996) per fragment, incorporated by PCR. In brief, RNAP (2 nM EσS or Eσ70) was incubated with an excess of 32P end-labelled DNA fragment (10 nM) for 10 min at 22°C in 50 µl of 10 mM Tris-Cl (pH 7.9), 100 mM KCl, 10 mM MgCl2, and 1 mM DTT, so that about 5% of the DNA formed complexes. Heparin (final concentration 25 µg ml−1) was added, RNAP–promoter complexes were separated from unbound DNA and the DNA was eluted from the gel. DNA fragments were cleaved at the position of A* incorporation by treatment with 1 M piperidine at 90°C for 30 min. Fragments containing U were first treated with uracil DNA glycosylase (UDG; NE Biolabs) and then cleaved with piperidine. Fragments were analysed by phosphorimaging (Molecular Dynamics). After electrophoresis on 10% denaturing polyacrylamide gels, lanes were normalized to correct for loading differences and graphed using SigmaPlot (Jandel Scientific).
In vitro transcription
Reactions contained supercoiled plasmid DNA (20 ng), 10 mM Tris-Cl (pH 7.9), 10 mM MgCl2, 1 mM DTT, 100 µg ml−1 of BSA, 200 µM ATP, GTP, and CTP, 10 µM UTP, 4 µCi [α-32P]-UTP (NEN) and 10–800 mM KCl or K-glutamate. Transcription was initiated by addition of RNAP (1–4 nM EσS or Eσ70) and terminated by addition of stop solution (Ross et al., 1993) after 15 min at 22°C. Samples were electrophoresed on 5.5% polyacrylamide 7 M urea gels and quantified by phosphorimaging.
We thank A. Ernst and G. Verdine (Harvard University) for providing 7-deaza-7-nitro-adenine (A*), and M. Barker, J. Gralla, J. Helmann, A. Hochschild and R. Hengge-Aronis for helpful discussions. This work was supported by National Institutes of Health Grant GM37048 to R.L.G., by a Hatch grant from the US Department of Agriculture to R.L.G. and by NIH grant GM28575 to R.R.B.