In recent years, it has become clear that promoter recognition by bacterial RNA polymerase involves interactions not only between core promoter elements and the σ subunit, but also between a DNA element upstream of the core promoter and the α subunit. DNA binding by α can increase transcription dramatically. Here we review the current state of our understanding of the α interaction with DNA during basal transcription initiation (i.e. in the absence of proteins other than RNA polymerase) and activated transcription initiation (i.e. when stimulated by transcription factors).
Promoter recognition by RNA polymerase (RNAP; subunit composition α2ββ′σ) is a crucial step in gene expression and its regulation. Promoter recognition involves the interaction of σ70 with two elements in bacterial promoters located about 10 and 35 bp upstream of the transcription start site (reviewed by Record et al., 1996). However, almost 25 years passed after the identification of σ (Burgess et al., 1969) before it was discovered that α also plays a direct role in promoter recognition (Ross et al., 1993).
At least two factors contributed to the long delay in the recognition of α's role in promoter function: (i) early promoter sequence compilations identified the most conserved elements, the −10 and −35 hexamers, and therefore most emphasis was placed on understanding the σ-promoter interaction. (ii) Early studies of Escherichia coli promoter recognition most often concentrated on promoters (e.g. λPR and lac) for which we now know that α–DNA interactions contribute little to transcription (Ross et al., 1998).
The discovery of α–DNA interactions was also spurred by the recognition in the early 1990s that α was a target for the action of transcription factors (‘the Cinderella subunit of RNA polymerase’; Russo and Silhavy, 1992). This emergence of α as a critical determinant in transcription initiation was facilitated greatly by the identification and purification of RNAP holoenzymes containing mutant α subunits (Igarashi and Ishihama, 1991), and by the distribution of mutant RNAPs lacking the C-terminal one-third of α to laboratories around the world by the Ishihama laboratory.
These mutant RNAPs were shown to be defective in rRNA promoter function: they did not display ‘factor-independent upstream activation’ (Ross et al., 1993). Furthermore, purified α bound to the rrnB P1 upstream sequence, and the rrnB P1 α-binding site retained its effect on transcription when fused to other promoters (Ross et al., 1993). It was concluded that bacterial promoters can consist of not just σ binding sites, but of α binding sites as well, and the α recognition element was named the Upstream (or UP) element (Ross et al., 1993). Later, it was shown that the α binding sequence actually consists of two subsites, each of which binds a copy of α (Estrem et al., 1999).
The α–DNA interaction plays an important role in activated transcription as well as in basal transcription. The role of the α–DNA interaction in activated transcription is touched on briefly below, and readers are referred to two other recent reviews for more thorough treatments of activator interactions with α (Hochschild and Dove, 1998; Busby and Ebright, 1999).
The DNA binding patch in α
The 329-amino-acid α subunit from E. coli contains two independently folded domains (Blatter et al., 1994) whose structures have been determined at high resolution: an N-terminal domain (αNTD) extending to residue ∼235 (Zhang and Darst, 1998) and a C-terminal domain (αCTD) extending from residue ∼249 to the end of the protein (Jeon et al., 1995). The linker connecting the two domains is unstructured and flexible (Blatter et al., 1994; Jeon et al., 1997). The αCTD binds to the rrnB P1 UP element as a purified peptide (although with lower affinity than intact α and with much lower affinity than RNAP holoenzyme), confirming its identity as an independent domain responsible for DNA binding (Blatter et al., 1994). The specific αCTD residues required for DNA binding were determined using genetic screens for α mutants that interfered with UP element utilization in vivo, followed by alanine scans of the regions in α defined by the random screen. Effects of residues identified in vivo were analysed in vitro by transcription and footprinting with reconstituted mutant RNAPs (Gaal et al., 1996; Murakami et al., 1996). Seven residues in the αCTD are most critical for DNA interaction and for UP element function: L262, R265, N268, C269, G296, K298, S299. α Subunits containing alanine substitutions in any of these seven residues failed to complement temperative-sensitive α mutants at the restrictive temperature (Gaal et al., 1996). The one-to-one correspondence between the residues required for DNA binding and those required for viability in haploid suggests that UP element utilization is essential in E. coli. The DNA-binding residues form a patch on the αCTD surface (Fig. 1). The identities of the residues in this patch are almost 100% conserved in bacteria, suggesting that the UP element consensus sequence should apply throughout the bacterial kingdom.
Full UP element and subsite consensus sequences
The ‘consensus’ UP element sequence was determined (Estrem et al., 1998) using a strategy different from that used for defining the consensus σ binding sites for Eσ70 holoenzyme (the −10 and −35 hexamers). The upstream sequences that most increased binding of RNAP to the rrnB P1 core promoter were selected from a library of about 1012 random DNA sequences using an in vitro selection procedure. After multiple cycles of RNAP binding in vitro and separation of RNAP bound from unbound DNA fragments, surviving sequences were screened for their ability to stimulate transcription in vivo using promoter–lacZ fusions. The best UP elements obtained from this in vitro selection increased transcription about 330-fold in vivo, five- to 10-fold more than the natural rrnB P1 UP element where the role of α binding was first detected. Like the rrnB P1 UP element, the consensus UP element derived from the in vitro selection (Fig. 2) was extremely A+T-rich (−59 to −38; 5′-NNAAAWWTWTTTTNNNAAANNN-3′; W = A or T, N = any base), consisting of a region of about 20 bp that was essentially alternating A and T-tracts.
Several results suggested that UP elements might function in two parts. Hydroxyl radical footprints indicated that α protected two sections of the UP element backbone, centred at about −42 and −52 on the same face of the helix in rrnB P1 (where the −35 hexamer extends from residues −36 to −31; Newlands et al., 1991; Ross et al., 1993; Estrem et al., 1998; 1999). Deletion of the upstream half of the rrnB P1 UP element, corresponding to one of the two regions protected by α in the hydroxyl radical footprints, only partially reduced UP element-dependent transcription (Leirmo and Gourse, 1991; Rao et al., 1994), and the remaining half of the UP element was fully protected in footprints with RNAP (Newlands et al., 1991). Finally, separation of the two protected regions of the rrnB P1 UP element by an insertion of 11 bp did not interfere with its function, and each separated region remained protected in footprints. (Newlands et al., 1992).
Therefore, in vitro selections and in vivo screens for the optimal sequences in each region, referred to as the proximal and distal subsites, were carried out in the absence of optimal sequences in the other region (Estrem et al., 1999; Fig. 2). The consensus proximal subsite (positions −46 to −38; 5′-AAAAAARNR-3′; R = purine) was similar, but not identical, to the corresponding part of the consensus full UP element, and by itself increased transcription of the rrnB P1 core promoter as much as 170-fold in vivo. In contrast, the consensus distal subsite (positions −59 to −46; 5′-NNAWWWWWTTTTTN-3′) was essentially identical to the corresponding portion of the consensus full UP element and, by itself, increased transcription only about 16-fold. When combined, the two consensus subsites together increased transcription about 340-fold, similar to the effect of the in vitro selected full UP element containing both subsites, and not by the product of the effects of the individual subsites. This indicated there is an upper limit to stimulation of the rrnB P1 core promoter, probably dictated by full promoter occupancy. We speculate that the consensus proximal subsite increases transcription more than the consensus distal subsite because of its proximity to the −35 element and the σ subunit (Estrem et al., 1999). Recent genetic evidence suggests that α bound to the proximal subsite and σ bound to the −35 element may interact (see below).
Relationship of DNA curvature to UP elements
Long before it was discovered that upstream sequences in bacterial promoters interacted with α, it was found that some naturally occurring promoters contained phased A-tracts upstream of the −35 hexamer, and that A-tract sequences increased promoter activity when fused to core promoters in synthetic promoter hybrids (McAllister and Achberger, 1988; Bracco et al., 1989; Gartenberg and Crothers, 1990; Travers, 1990). Because A-tracts phased with the helical repeat resulted in DNA bending by about 18° per A-tract (Koo et al., 1986), it was proposed that DNA curvature per se facilitated transcription (reviewed by Perez-Martin et al., 1994). However, the discovery that UP elements often contain A-tracts suggested an alternative explanation: that direct effects of A-tracts on transcription result from DNA–α interactions. It was found that phased A-tracts failed to stimulate expression when promoters were transcribed with RNAPs lacking the DNA-binding domain of α, and protection of the A-tract sequences in footprints with RNAP was αCTD dependent (Aiyar et al., 1998). In addition, the extent of stimulation of transcription by wild-type RNAP did not correlate directly with the extent of curvature, and a single A-tract at the position of the proximal subsite increased transcription much more than phased A-tracts located further upstream. Some UP elements (e.g. rrnB P1; Gaal et al., 1994) do not display much intrinsic curvature, indicating that macroscopic DNA bending is not a prerequisite for UP element function. Together, these results demonstrated that, in most cases, A-tracts increase transcription through DNA–protein interactions (i.e. by binding to the αCTD of RNAP), not simply by bending DNA, and that usually phased A-tracts should be considered a subset of UP element sequences.
A-tract DNA structure is unusual and some aspect of this structure (for example, narrow minor groove width; see below) might facilitate α binding. In addition, the presence of DNAse I hypersensitive sites in footprints of the UP element region suggests that, upon RNAP binding, there is DNA distortion within the α binding site and/or between the α and σ-binding sites (Gourse, 1988; Newlands et al., 1991; Estrem et al., 1999). Some portion of the wrapping of DNA around RNAP in the transcription initiation complex (Record et al., 1996) is probably attributable to α–DNA interactions.
The αCTDs are not resolved and σ is not present in the current X-ray structure of bacterial RNAP (Zhang et al., 1999). However, the positions of the α N-terminal domains in the structure predict the positions of the αCTDs, and models have been proposed for core RNAP contacts with σ and for the path of DNA in the initiation complex (Mooney and Landick, 1999). The αCTDs are located on the opposite side of the complex from where DNA enters and exits the catalytic pocket. In order for RNAP to interact simultaneously with the core promoter and with the UP element, it is clear that a sharp bend in the DNA must be induced by interactions with the enzyme somewhere upstream of the transcription bubble. The presence of a DNAse I hypersensitive site just upstream of the −35 hexamer (position −38 in rrnB P1) is consistent with the existence of such a bend between the binding sites for αCTD and σ. Therefore, intrinsic curvature may not be directly responsible for the effects of DNA sequences upstream of the −35 element, but UP element DNA structure and RNAP-induced DNA bending are nevertheless likely to play a role in transcription initiation.
UP element distribution in E. coli and other bacteria
The rules that govern the magnitude of the effects of particular UP element sequences on specific promoters are likely to be extremely complex, and thus the predictive value of the UP element consensus sequences is somewhat limited. Factors contributing to the effect of a particular UP element sequence on transcription would include not only the number of matches to consensus within each subsite, but also the positions of those matches within a subsite, whether they are in the distal compared with the proximal subsite, where the substituted base pair fits in the hierarchy of favoured base pairs at that position, context effects of neighbouring sequences and the identity of the core promoter. Nevertheless, when a set of upstream sequences from different promoters was fused to the same core promoter, a rough correlation was observed between the number of matches to the consensus sequence and the effect of an UP element on transcription (Ross et al., 1998). The extent of stimulation varied widely with different upstream sequences, from less than twofold to 90-fold in naturally occurring UP elements tested to date (Ross et al., 1993; 1998).
To determine the frequency of potential UP elements in naturally occurring promoters, the E. coli genome sequence was screened for matches to the consensus full UP element and to the consensus proximal or distal subsites (Estrem et al., 1999). Although the number of promoters in a genome that appear to contain UP elements depends on how many matches are set as a threshold and/or on how large an effect on transcription is defined as significant, several important conclusions could still be drawn. First, numerous E. coli promoters, driving expression of a wide array of gene products, contain near-consensus full UP elements or subsites. Second, promoters with a close match to consensus in only one subsite are far more common than promoters with close matches to consensus in both subsites. Third, rRNA and tRNA promoters are significantly enriched for UP elements.
UP elements are not limited to only Eσ70 promoters or to only E. coli promoters. UP elements have been shown to activate certain promoters transcribed by Eσ32 (rrnB P1; Newlands et al., 1993) or EσF (fliC: Fredrick et al., 1995). As described above, the extreme evolutionary conservation of the amino acids in α involved in UP element interactions suggests that UP elements are prevalent in promoters throughout the bacterial world. Promoters from some Gram-positive bacteria (e.g. Clostridium and Bacillus) may utilize UP elements even more frequently than E. coli (Graves and Rabinowitz, 1986; Helmann, 1995). In only a few cases, however, have the effects of upstream sequences been proven to result from interactions with RNAP rather than from a transcription factor, and in even fewer cases has the role of α been tested directly (e.g. Fredrick et al., 1995).
Mechanism of transcription stimulation/inhibition by UP elements
RNAP binds to promoters and initiates transcription in a series of steps characterized by conformational changes (‘isomerizations’), but complex formation is often simplified to a two-step mechanism in which initial binding is defined by the equilibrium constant KB, and later steps are combined into the forward rate constant kf (Record et al., 1996). Kinetic measurements of the effects of UP elements have been performed for only a limited number of promoters. The rrnB P1 UP element clearly recruits RNAP to the promoter, i.e. it increases KB (Leirmo and Gourse, 1991; Rao et al., 1994). In addition, the rrnB P1 UP element appears to increase the rate of a later step in the transcription mechanism (Rao et al., 1994; Strainic et al., 1998).
In some cases, promoter–α interactions occur but have either no effect or a negative effect on promoter function (Ross et al., 1998). Therefore, α binding should not be taken as proof of UP element function in the absence of other supporting data. Although transcription from the lac core promoter was stimulated by various UP elements in synthetic hybrids, sequences with limited similarity to consensus resulted in no increase in transcription despite their protection in footprints (Ross et al., 1998). Furthermore, inhibition of malT transcription was correlated with an inappropriately positioned UP element (Tagami and Aiba, 1999), and an A-tract at the position of a proximal UP element subsite (now interpreted as a probable α binding site) inhibited RNAP escape from a clearance-limited promoter (Ellinger et al., 1994). Therefore, α–DNA interactions can mediate the formation of both productive and non-productive complexes, depending on the identity of the core promoter sequence and on the location of the UP element.
αCTD is a minor groove binding protein
Because no X-ray or NMR structure exists of a complex containing the αCTD bound to DNA, detailed information about the α-UP element interaction is based primarily on footprinting analyses with holoenzyme and with purified α. α alone and holoenzyme produce similar footprints on UP element DNA (Ross et al., 1993; Estrem et al., 1998; W. Ross, unpublished results), suggesting that αCTD interacts with UP elements similarly when part of RNAP and when present as a purified dimer. In hydroxyl radical footprints, the protected regions on the two DNA strands of the UP element were offset by 3 bp, suggesting that α binds across the minor groove (Newlands et al., 1991; Ross et al., 1993; Estrem et al., 1998; 1999).
α Protected the N3 position of adenine residues in the UP element minor groove from modification by dimethyl sulphate (DMS), and distamycin, an antibiotic that binds snugly in the minor groove of A+T-rich DNA, prevented α from binding to the UP element (W.R., and R.L.G., in preparation). In addition, distamycin inhibited UP element-dependent transcription under conditions where UP element-independent transcription was unaffected (W.R., and R.L.G., in preparation). We conclude that α makes intimate contacts to bases in the minor groove. This conclusion is also consistent with results from site-specific protein–DNA cross-linking experiments (N. Naryshkin, A. Revyakin, Y. Kim, and R. H. Ebright, unpublished results). It is possible that narrow minor groove width, or some other aspect of the unusual structure of A-tract sequences, could also play a role in α binding.
The potential role of major groove contacts in α-UP element interactions was tested by interference footprinting, using RNAP and promoter fragments prepared by limited incorporation of dUTP or 7-deaza 7-nitro dATP into the DNA (W.R., A. Ernst, G. Verdine, and R.L.G., in preparation). These experiments indicated that major groove contacts contribute little, if at all, to α-UP element interactions. In contrast, incorporation of dUTP or 7-deaza 7-nitro dATP at positions in the core promoter had much larger effects on RNAP binding, interfering with σ contacts to the −10 and −35 elements. Together with the results cited above, the data indicate that interactions with bases in the minor groove and with the DNA backbone are primarily responsible for specific binding of α to UP elements.
Arrangement of α subunits on UP elements
RNAP contains two α subunits and therefore two αCTDs. Although αCTD purifies as a dimer in solution (Blatter et al., 1994), there is no evidence that the two αCTDs interact when bound to DNA. In fact, the DNA binding and dimerization interfaces appear to overlap, and therefore DNA binding and dimerization may be mutually exclusive events (R. H. Ebright, personal communication).
Whether each UP element subsite constitutes a site for interaction with one copy of αCTD was tested using oriented-α RNAPs lacking one αCTD (Murakami et al., 1997a; Estrem et al., 1999). One α binds primarily to β and one to β′ (Zhang et al., 1999), thereby permitting the two αCTDs in RNAP to be differentiated using mutations in the αNTD that orient the α subunits with respect to the β subunit. Our experiments resulted in a relatively straightforward model for the arrangement of the two αCTDs in transcription complexes containing UP elements (Fig. 3; Estrem et al., 1999), although more complex interpretations are possible. First, transcription from promoters containing only a consensus proximal UP element subsite requires only one αCTD, and this subsite is preferred by αCTD in UP elements containing good matches to consensus in both subsites. Second, each UP element subsite interacts with a copy of αCTD and as a result, UP elements containing two good subsites require both αCTDs for maximal stimulation of transcription. Third, the two αCTDs are functionally interchangeable with respect to recognition of an UP element consisting of only a consensus proximal subsite. Fourth, promoters containing only a consensus distal subsite require both αCTDs for efficient transcription, perhaps because of sequence-non-specific interactions between the proximal subsite region and the second αCTD or because the second αCTD in some other way affects the overall stability of the complex.
In a few cases, more than two αCTD-dependent protected regions appear in footprints upstream of a single promoter (e.g. Newlands et al., 1992; Kolb et al., 1993; Aiyar et al., 1998). We speculate that in these cases multiple potential binding sites for αCTD are present, but different combinations of sites are occupied in fractions of the DNA population. Although generally the sequences closest to the −35 hexamer appear to exert the greatest effects on transcription in the absence of transcription factors (Aiyar et al., 1998), multiple targets for αCTD could also aid the binding of adjacent transcription factors (e.g. see below and Murakami et al., 1997b).
Potential interactions between α and σ
Because the proximal UP element subsite is adjacent to the −35 element, the DNA-binding surface of one αCTD is likely to be in close proximity to σ. Recent genetic evidence (W.R., D. Schneider, B. Paul, A. Mertens, and R.L.G., unpublished results; H. Chen, A. Kapanidis, H. Tang, and R.H. Ebright, unpublished results) supports previous proposals that the αCTD and the C-terminal region of σ interact (Ross et al., 1993; Busby and Ebright, 1994; Kuldell and Hochschild, 1994; Estrem et al., 1999). Alanine substitutions were identified in the C-terminal domain of α and in the C-terminal region of σ that resulted in selective inhibition of promoters with consensus proximal subsites without affecting core promoter activity. The alanine substitutions in αCTD affecting proximal subsite-dependent transcription included E261A, which resides in a negatively charged, surface-exposed patch located adjacent to the DNA-binding surface. The alanine substitutions in σ affecting proximal subsite-dependent transcription were for positively charged residues just C-terminal to region 4.2, the region of σ that interacts with the −35 element (see also Lonetto et al., 1998). Thus, the α and σ regions defined by these mutants might interact. While these preliminary conclusions should be viewed with caution, we note that σ–αCTD interactions have also been proposed previously based on cross-linking evidence (McMahan and Burgess, 1994), although the residues responsible for the cross-links were not identified.
Interactions between αCTD bound to the proximal UP element subsite and σ bound to the −35 hexamer could potentially stabilize initiation complexes and contribute to the large effect of the consensus proximal subsite on transcription. αCTD–σ interactions could also explain the preference of holoenzymes containing only one αCTD for binding to the consensus proximal subsite rather than to the consensus distal subsite. Furthermore, it is likely that altered αCTD–σ interactions are responsible for previously reported effects of substitutions for E261 in α on MetR-independent transcription from the metE promoter (Jafri et al., 1995) and on CRP-dependent transcription from the lac promoter (Tang et al., 1994).
Involvement of αCTD–DNA interactions in activator function
αCTD is a frequent target on RNAP for transcription activators that bind to DNA upstream of the σ-binding site (Hochschild and Dove, 1998; Busby and Ebright, 1999). Because the linker separating the αNTD and the αCTD is flexible and might be predicted to stretch as far as 44 Å when fully extended (Blatter et al., 1994), interactions between αCTD and an activator can occur in quite different spatial arrangements. Furthermore, distortion of the DNA between the activator-binding site and the core promoter can further facilitate interactions between αCTD and activators bound at distant locations. As a result, activator complexes are found with several different architectures: (i) one or both αCTDs situated between a bound activator and the rest of RNAP; (ii) one or both αCTDs situated upstream of a bound activator; (iii) αCTDs flanking a bound activator; or (iv) permutations of the above arrangements in the presence of multiple activators (see, for example, Murakami et al., 1997b; Belyaeva et al., 1998; Bertoni et al., 1998). In each case, interactions between αCTD and DNA adjacent to the activator-binding site are likely to contribute to the overall stability of the activation complex. αCTD–DNA interactions in these complexes can involve either UP element-like DNA sequences, as in the rrnB P1 complex containing the transcription factor FIS (Bokal et al., 1995), or non-specific DNA sequences, as in the lac complex containing CRP (Kolb et al., 1993). As a result, screens or selections whose purpose was to identify mutations in rpoA defining activator-αCTD contacts have sometimes identified residues in αCTD required for DNA interactions.
Prospects for the future
Several issues relevant to UP element structure/function remain unresolved. We have alluded to the first already: the precise details of the RNAP–UP element interaction await a high resolution structure of the αCTD–DNA complex. Furthermore, structures of αCTD bound in a ‘specific’ high affinity complex (such as present when αCTD is bound to a consensus UP element subsite) and of αCTD bound in a ‘non-specific’ low affinity complex (such as present when αCTD interacts with non-UP element DNA) would be informative.
A second issue concerns the effect of one αCTD on the interaction of the other αCTD with DNA in UP element complexes containing two subsites. The details of the interaction between αCTD and the proximal UP element subsite seem to be altered by the presence of αCTD bound to the distal subsite (W.R. and R.L.G., unpublished results; Estrem et al., 1999). The basis for these effects remains unclear.
A third issue relevant to UP element-dependent transcription is whether cells or phage modify αCTD or UP element DNA in a way that regulates UP element utilization, or whether there are repressors that specifically target UP element function. At least one case has been documented: upon infection, bacteriophage T4 ADP-ribosylates residue R265 of α, the residue in αCTD most crucial for DNA binding. This large modification virtually eliminates UP element-dependent and CRP-dependent transcription (K. Severinov, W.R., H. Tang, L. Snyder, A. Goldfarb, R.L.G., and R.H. Ebright, in preparation). Thus, a fertile subject for future investigation is whether mechanisms in uninfected cells ever target the UP element αCTD interaction for regulatory purposes. Such regulatory systems could have important biological consequences.
Research in our laboratory is supported by the National Institutes of Health (GM37048). The authors thank members of our laboratory, past and present, for their contributions to the research described in this review. We thank R. Ebright for permission to cite unpublished results and for comments on the manuscript. We apologize to investigators whose work was not adequately cited because of space limitations.