Promoter Substitution and Deletion Analysis of Upstream Region Required for rpoS Translational Regulation

Abstract

The RpoS sigma factor of enteric bacteria is required for the increased expression of a number of genes that are induced during nutrient limitation and growth into stationary phase and in response to high osmolarity. RpoS is also a virulence factor for several pathogenic species, including Salmonella typhimurium. The activity of RpoS is regulated at both the level of synthesis and protein turnover. Here we investigate the posttranscriptional control of RpoS synthesis by using rpoS-lac protein and operon fusions. Substitution of the native rpoS promoters with the tac or lac UV5 promoters allowed essentially normal regulation after growth into stationary phase in rich medium or after osmotic challenge. Regulation of these fusions required the function of hfq, encoding the RNA-binding protein host factor I (HF-I). Short deletions from the 5′ end of the rpoS transcript did not affect regulation very much; however, a larger deletion mutation that still retains 220 bp upstream of the rpoS ATG codon, including a proposed antisense element inhibitory for rpoS translation, was no longer regulated by HF-I. Several models for regulation of rpoS expression by HF-I are discussed.

In the enteric bacteria, including Escherichia coli and Salmonella typhimurium, the rpoS gene encodes an accessory sigma (specificity) factor for RNA polymerase (also called ς^S or ς³⁸ [36, 49]). RpoS is required for the transcription of many genes expressed during the onset of stationary phase. RpoS-dependent adaptations to nutrient limitation and starvation identified so far in E. coli include not only shifts in metabolic pathways but also resistance mechanisms protective against life-threatening stresses, such as high osmolarity, low pH, heat shock, elevated H₂O₂, and UV light (reviewed in references 19 to 21 and 29). RpoS is also a virulence factor for S. typhimurium (15) and other enteric bacteria.

RpoS abundance can be increased in exponential-phase cells by a variety of induction treatments, including osmotic challenge (22, 27, 34 [reviewed in reference 21]). High levels of RpoS are also seen in cells grown to stationary phase in rich medium (reviewed in reference 19). It has been demonstrated that RpoS abundance is increased by provoking an increase in the level of the alarmone ppGpp (18), and this may explain why so many treatments which induce RpoS also decrease the growth rate, at least transiently. For osmotic challenge and stationary phase, control of RpoS occurs both at the level of synthesis and by regulated proteolysis. Genetic analysis of RpoS regulation revealed a requirement for the energy-dependent ClpXP protease (41), which promotes RpoS turnover with the help of other factors (4, 32, 38). Both clpXP mutants and hns mutants lacking the abundant DNA-binding protein H-NS have an increased level of RpoS during the exponential phase (2, 53).

Comparative studies with rpoS-lac protein and operon fusions have shown that control of RpoS synthesis occurs mainly at a posttranscriptional level (27). Host factor I (HF-I) is an RNA-binding protein that was discovered through its role in the replication of Qβ, an RNA bacteriophage that infects E. coli (9, 16, 17, 23, 24, 42). The function of HF-I in uninfected cells has been unknown, but hfq mutants are quite pleiotropic (35, 51, 52). In recent work, it has been shown that S. typhimurium and E. coli hfq mutants lacking HF-I have substantially reduced expression of rpoS (7, 33). The defect in rpoS expression is posttranscriptional.

Some additional insight has been gained by analysis of mutations that restore expression of rpoS in hfq mutants (8; unpublished work cited in reference 33). Most of these mutations disrupt a predicted secondary structure that would sequester the rpoS ribosome binding site (RBS) (8). One attractive model is that control of rpoS synthesis at the translational level involves regulation of ribosome access by this inhibitory RNA secondary structure. The RNA-binding protein HF-I may be directly involved in relieving this inhibition, for example, as an RNA chaperone which promotes equilibration between different RNA secondary structures. However, other models of HF-I action are possible.

In this work, we tested two simple predictions of the translational control model for rpoS. Two different promoters were substituted for the native rpoS promoters: translational regulation was retained in these constructs and was still dependent on HF-I function. However, deletion analysis showed a requirement for sequences well upstream of the proposed secondary structure, suggesting that interaction of HF-I with this structure is probably not sufficient for regulation.

MATERIALS AND METHODS

Bacterial strains and construction.

The strains used in this study are derived from the wild-type S. typhimurium strain LT-2 (originally obtained from J. Roth). S. typhimurium does not carry the lac operon. The LT-2 derivatives used here carry various lac fusions to the rpoS gene of E. coli and the hfq-1::Mud-Cam insertion (7), where indicated, but no other mutations with respect to the parental LT-2 strain. Note that some strains of LT-2 have been shown to carry a mutation in rpoS (a change of ATG→TTG in the rpoS initiation codon [3]) as well as a mutation inactivating mviA (the homolog of the E. coli gene rssB or sprE, which controls turnover of RpoS [4]). The LT-2 strain used here carries a mutation mapping in the mviA region that stabilizes RpoS (data not shown). The lac fusions used in this study do not encode the segment of the RpoS protein required for turnover (41), and the hybrid RpoS-LacZ protein produced is stable during exponential growth in minimal glucose medium (7).

The high-frequency generalized transducing bacteriophage P22 mutant HT105/1 int-201 (39) was used for transduction in S. typhimurium by standard methods (12). Fusions of DNA fragments derived from the E. coli rpoS gene to lac were constructed as described below; these lac fusions were transferred to the chromosome of an E. coli recD mutant by linear transformation as described previously (14). P22 phage lysates were grown in E. coli (13, 14) and used to transduce the fusions into the S. typhimurium chromosome. Each resulting strain carries a lac fusion in single copy as an insertion of a Kan^r promoter-lac fragment in the put operon.

Media and growth conditions.

Bacteria were grown at 37°C in Luria-Bertani (LB) medium (43) or in minimal MOPS (morpholinepropanesulfonic acid) medium (37), modified as described in reference 6, with 0.2% glucose as the carbon and energy source. For experiments employing continuous growth in high-osmolarity medium, a modified version of the basal medium of reference 25 was used. The modified medium contained 7 g of nutrient broth and 1 g of yeast extract (Difco) per liter of NCE minimal salts medium (per reference 5, but without MgSO₄). This medium also contained 0.2% glycerol, and where indicated, sucrose was added at 15%. Plates were prepared with nutrient agar (Difco) with 5 g of NaCl per liter. Antibiotics were added to the following final concentrations: sodium ampicillin, 100 μg/ml; chloramphenicol, 20 μg/ml; kanamycin sulfate, 50 μg/ml; and tetracycline hydrochloride, 20 μg/ml.

Construction of lac fusions.

The system we used has been described previously (14). Fusions were made in the pRS plasmid series of Simons et al. (44). Full-length rpoS-lac fusions have been described previously (7). Construct A was constructed by digesting pMMkatF2 (36) with ClaI and EagI, filling in the ends with the Klenow fragment of DNA polymerase I, and cloning the 1.6-kb fragment into pRS552 that had been digested with EcoRI and BamHI and filled in. To make construct B, an rpoS-lac operon fusion, the same 1.6-kb ClaI-EagI fragment of pMMkatF2 was filled in and inserted into the SmaI site of pRS415. The fragment was subsequently excised by digestion at the flanking EcoRI and BamHI sites and inserted into pTE583 (7). These parental lac protein and operon fusions to rpoS extend from the ClaI site upstream of nlpD to the EagI site at codon 73 of the rpoS gene. The operon fusion version includes an RNase III site directly upstream of the lacZ RBS (28).

Construct C was made from construct A by deleting the EcoRI-KpnI fragment including the promoters serving rpoS; construct D is similar, except that it is an operon fusion and contains extra nucleotides (specifying a SacI site and a KpnI site) at the deletion joint upstream of the rpoS sequences. Construct E is a fusion to codon 8 of rpoS (GTT CAG GAT CCG). (Sequences that are not derived from nlpD or rpoS are underlined here and below.) This fusion joint was constructed by PCR.

The tac promoter in construct F was generated by PCR to create the junction GAATT CTTGA CAATT AATCA TCGAC TAGTA TAATG TATTT GGGTG AACAG AGTGC TAACA AAATG TTG. The tac promoter in construct G was derived from pTM30 (31), yielding the tac promoter and lac operator followed by the junction sequence AGGAG TGTGA AATGC TGCAGGATCC GATAT CAAGC TTGGT ACCAA CAGCA AGCAC. This construct includes an RBS and ATG initiation codon. The BamHI site in this construct (in boldface) was filled in with DNA polymerase to produce the frameshifted version, construct H. The lac UV5 promoter in construct K was derived from pRS476 (44) to create the junction AGGAA ACAGG ATCCG ATATC AAGCT TGGTA CCAAC AGCAA GCAC. This version of the lac UV5 promoter is not equipped with an RBS. The deletion derivatives I and L were made from constructs G and K by substitution of a PCR-generated fragment which placed a KpnI site adjacent to bp 1 of the rpoS sequence numbered according to reference 8 and below (the deletion joint lies 221 bp upstream of the rpoS ATG codon). Constructs J and M are similar, but the deletion joint is 114 bp upstream of the rpoS ATG codon. This deletion removes material up to the boundary of the inhibitory secondary structure referred to above, which was described in detail previously (8). For deletions I and J, the correct reading frame was retained, so that ribosomes which initiate translation at the RBS just downstream of P_tac will terminate at the native nlpD stop codon.

Construct N was made by substitution of an XhoI-EagI fragment from construct K, containing the lac UV5 promoter and rpoS sequences, into the operon fusion vector pTE583. Construct O was made from a plasmid that is identical to pRS475 (44), except that it contains a Kan^r marker from pUC4K upstream of the lac UV5 promoter. Construct P was made by substitution of an XhoI-BamHI fragment from construct G, containing the P_tac promoter, into the operon fusion vector pTE583. All PCR-derived fragments were sequenced completely; all constructs were sequenced to verify the predicted junctions.

β-Galactosidase assays.

Cells were centrifuged and resuspended in Z-buffer (100 mM NaPO₄ [pH 7.0], 10 mM KCl, 1 mM MgSO₄) and then permeabilized by treatment with sodium dodecyl sulfate and chloroform (7). Assays were performed in Z-buffer containing 50 mM β-mercaptoethanol by a kinetic method using a plate reader (Molecular Dynamics). Activities (change in optical density at 420 nm [ΔOD₄₂₀] per min) are normalized to actual cell density (OD₆₅₀) and were always compared to those of appropriate controls assayed at the same time. The results shown are from a single experiment; each experiment was repeated several times with similar results.

RESULTS

Deletion analysis of sequences required for translational control of RpoS.

If HF-I or stimuli such as osmotic challenge regulate RpoS synthesis at the translational level, then we expect that the native rpoS promoters can be replaced by another promoter without loss of regulation. A more restrictive translational control model (discussed above) predicts that HF-I interacts only with the segment of mRNA containing a putative secondary structure that inhibits translation. The ultimate result of this interaction is increased availability of the rpoS RBS for binding to ribosomes. According to this model, transcript sequences upstream of the secondary structure should not be required for HF-I regulation of RpoS.

We constructed rpoS-lac fusions containing deletions of the native promoters and substituted new promoters to test these ideas. In the first set of experiments, expression of β-galactosidase was measured in cultures grown overnight to saturation in LB medium. The rpoS gene is highly expressed under these conditions. Construct A (Fig. 1) is the parental lac fusion; it includes both of the identified promoters that are used to initiate transcripts of rpoS. One of the native promoters serves both the upstream nlpD gene and rpoS and is referred to here as P_nlpD; the second promoter lies within the nlpD gene and is referred to as P_rpoS (26, 48). The lac fusion in construct A is a protein or translational fusion, and it is identical to the fusion used in our previous studies (7, 8). Construct B is the same, except that it is an operon or transcriptional fusion. As shown previously, expression of the rpoS-lac (protein) construct A was stimulated in wild-type cells (containing a functional hfq gene) compared to that in an otherwise isogenic strain that carries a Mud-Cam insertion mutation at codon 68 of hfq (7). In contrast, the rpoS-lac operon fusion (construct B) containing the same sequences was expressed at a similar level in the presence or absence of HF-I.

FIG. 1.

rpoS-lac fusions. The top line shows a restriction map of the nlpD and rpoS genes of E. coli, including the restriction sites used to make lac fusions in this study. The positions of two promoters that serve rpoS are also shown by bent arrows. Each line below the top one represents a different fusion; these are referred to in the text as rpoS-lac fusions and are labeled (A to P). The extent of the nlpD-rpoS sequence present in each fusion is shown, together with the identity of the promoter substitution where appropriate: P_tac or P_{lac UV5}. The nature of the lac fusion (protein [pr] or operon [op]) is also indicated. ATG start codons (solid squares) and their cognate stop codons (arrowheads) are shown. One-letter abbreviations are used for restriction sites (the full name of each site is on the top line). Additional sequence and construction details are given in Materials and Methods. The lac region is not drawn to scale. Expression of β-galactosidase was measured in overnight LB cultures of S. typhimurium strains bearing each fusion in single copy in the bacterial chromosome and was compared to the activity seen in an otherwise isogenic hfq mutant. Values are in arbitrary units, normalized to that of construct A in a wild-type background (set as 100). The ratio of the enzyme activity in the wild type divided by that in the hfq mutant is also given. N.D., not determined.

A KpnI site 73 bp downstream of the P_rpoS transcription initiation site (Fig. 1) was used to delete the native P_nlpD and P_rpoS promoters. Loss of DNA upstream of this site in constructs C and D eliminated most lac expression, confirming previous work by other groups showing that most transcription of rpoS initiates upstream of this point. Two different promoters were substituted. The promoters used were the tac and lac UV5 promoters; the sequences used are detailed in Materials and Methods. Construct F contains the tac promoter; this version of the tac promoter does not include the lac operator. The promoter is placed so that the resulting transcript should be identical to the transcript initiated from the P_rpoS promoter that normally serves rpoS. HF-I regulation of rpoS expression from this transcript was the same as that for native rpoS. Although the absolute levels of expression were elevated in both wild-type and hfq mutant hosts, the stimulation ratio for HF-I was similar to that seen for construct A. Thus, as predicted by all translational models, the native rpoS promoters are not required for regulation by HF-I. Control experiments with operon fusions to both the P_tac and P_{lac UV5} promoters (constructs N, O, and P) showed no HF-I regulation.

The transcript produced from construct F does not contain an identified RBS to allow translation of the nlpD gene fragment upstream of rpoS. Constructs G and H differ from construct F in two respects: there is a further deletion of 73 bp extending to the KpnI site, and there is an added 5′ leader sequence which includes a strong RBS. In construct G, translation initiated at the upstream RBS will terminate at the natural nlpD stop codon, whereas construct H contains a frameshift in the leader sequence leading to termination of translation 346 nucleotides (nt) upstream of the natural stop (83 nt downstream of the KpnI site). Sites of translation initiation and termination are indicated in the figure by solid squares and arrowheads, respectively. Both constructs G and H show substantial HF-I stimulation, suggesting that the small 73-bp deletion does not remove important sequences and also that translation to the nlpD stop codon does not affect regulation significantly. This result is surprising, given that the nlpD stop codon lies between the two stems of the proposed regulatory secondary structure, and ribosomes translating nlpD would be expected to disrupt the structure at least transiently. A similar construct (K) with the lac UV5 promoter placed at the KpnI site also shows normal regulation by HF-I.

The effect of deletion of additional DNA from the region upstream of rpoS was also investigated. Constructs I and L have deletions of DNA extending to a position 271 bp downstream of the KpnI site and 221 bp upstream of the rpoS ATG codon. Part of the sequence of the rpoS region which is retained in these deletions is shown in Fig. 2, including the postulated secondary structure. The boundary of the deletion in constructs I and L lies about 115 bp upstream of the postulated secondary structure (Fig. 2). Therefore, if HF-I interaction with the structure is all that is needed, we would predict that this deletion should be regulated normally. Instead, the fusion containing the tac promoter (construct I) is partially defective in the HF-I response (Fig. 1), and the fusion to the lac UV5 promoter (construct L) has nearly identical expression in the presence and absence of HF-I. Further deletions (to bp 111 in Fig. 2; constructs J and M in Fig. 1) extending to a point immediately upstream of the secondary structure element do not change this result. Finally, the requirement for sequences downstream of the rpoS ATG codon was also tested. Much of the HF-I control was retained by a lac fusion to codon 8 of rpoS (construct E), suggesting that the relevant target for regulation lies upstream of codon 8.

FIG. 2.

Model of the antisense-RBS structure near the rpoS start codon. The structure presented here is that suggested on the basis of genetic analysis (8), and the previous numbering scheme is retained in this figure. Arrows point to the nucleotides altered in the compensatory mutations which support the structure (C126G and G206C, as well as a second pair not shown). The U at nt 111 (also marked with an arrow and denoted Δ2) corresponds to the first rpoS-specific nucleotide in constructs J and M; material upstream of this point is deleted in these fusions. Not shown are the additional 111 nt retained by Δ1 (constructs I and L of Fig. 1) or the additional ≈400 nt extending to the 5′ end of the P_rpoS transcript. Two stems which pair the nucleotides connecting the antisense and RBS elements have also been added to this model. S.D., Shine-Dalgarno sequence.

Secondary structure in rpoS transcripts from deletion constructs.

The lack of response to HF-I by the deletion constructs, as described above, could have a simple explanation by analogy to the behavior of RNA II, the primer for ColE1 plasmid replication. RNA II folds into substantially different structures in its downstream half, depending on either the presence of upstream sequences or their sequestration in a complex with the antisense RNA, RNA I (29a). Thus, if the inhibitory structure did not form in an rpoS transcript which lacks important upstream sequences, then a failure to respond to HF-I would be understandable but not enlightening. As a test of this possibility, we prepared a set of rpoS-lac fusions from construct L to use the method of compensatory mutations as previously described (8). These fusions carry either of two mutations, C126G (SD2) or G206C (SD3), as well as the double mutant together with a wild-type control. Expression of rpoS-lac in wild-type and hfq mutant derivatives of these fusion strains is reported in Table 1. Since the elevated β-galactosidase activity seen in the single mutants is restored nearly to wild-type levels in the double mutant (measured in an hfq mutant background), we conclude that at least the identified RNA secondary structure forms in transcripts from this construct. None of the constructs is significantly stimulated by the presence of HF-I, and the lack of response to HF-I cannot be ascribed to loss of this structural element.

TABLE 1.

Expression of rpoS-lac in derivatives of fusion construct L that carry compensatory mutations in the rpoS antisense-RBS region

Strain constructa	β-Galactosidase activity (arbitrary units)b
	Wild type	hfq mutant
Plac UV5-rpoS-lac [pr] (Δ1)	46	45
Plac UV5-rpoS-lac [pr] (Δ1, C126G)	204	261
Plac UV5-rpoS-lac [pr] (Δ1, G206C)	184	244
Plac UV5-rpoS-lac [pr] (Δ1, C126G/G206C)	93	70

^a[pr], protein fusion to lac. 

^bDetermined as described in Materials and Methods. Activities are normalized as described in the legend to Fig. 1. 

Osmotic challenge.

We extended the analysis to include osmotic challenge because it has been demonstrated that HF-I is required for osmotic control of rpoS translation in E. coli (33). Several deletion constructs were tested for their response to challenge with high salt. The first experiment employed an osmotic challenge in which cultures growing in minimal MOPS glucose medium were treated with 0.3 M NaCl. The wild-type rpoS-lac protein fusion (construct A) showed a 3.5- to 4-fold increase in lac expression over 40 min and then reached a plateau by 60 min (Fig. 3A). The operon fusion (construct B) was not induced by this treatment (Fig. 3F). When an hfq mutation was present, the response of construct A to the osmotic challenge was partially defective (Fig. 3D). The kinetics of the response were changed, with an ≈2-fold increase at 40 min, but lac expression continued to increase up to at least 90 min.

FIG. 3.

Induction of rpoS-lac after osmotic challenge. Cultures of S. typhimurium strains carrying a lac fusion (as identified in Fig. 1 and the text) and an hfq mutation where indicated were grown in minimal MOPS glucose medium to an OD₆₀₀ of 0.4 and then challenged with 0.3 M NaCl. The activity of β-galactosidase was determined at 20, 40, 60, and 90 min after challenge. Solid squares, cultures receiving NaCl; open squares, control cultures. Data are reported as a percentage of the initial value for each fusion. [pr], lac protein fusion; [op], operon fusion.

Construct F, in which the tac promoter expresses the normal P_rpoS transcript, showed regulation similar to that of wild-type rpoS-lac (Fig. 3B), and expression peaked at a level about threefold higher than that of the untreated control. The hfq mutant derivative of this fusion was also defective. Similar to the result for native rpoS-lac, very little increase in β-galactosidase was seen at 20 min following the addition of 0.3 M NaCl. Construct I contains a deletion of 343 nt of the P_rpoS transcript leader, but retains 115 nt upstream of the proposed secondary structure. This fusion was defective in osmotic stimulation of lac expression (Fig. 3C). The results are similar to those for overnight growth in LB medium. The tac promoter gives normal regulation of rpoS-lac (protein fusion to lac) but this requires sequences well upstream of the proposed secondary structure. A larger set of constructs was also examined in a fixed-time assay (data not shown). The results confirm the picture from the kinetic assays. In particular, since several constructs with different upstream sequences but retaining the secondary structure are all defective for response to osmotic challenge, this property of the response (as illustrated in Fig. 3C) is not due to inhibition by one particular leader sequence.

Continuous growth at high osmolarity.

In addition to changing the osmotic strength of the medium, the osmotic challenge method also changes the bacterial growth rate (at least transiently). Therefore, we tested the effect of continuous growth in high-osmolarity medium. These experiments (modeled on those described in reference 38) were carried out in a phosphate-buffered rich medium including glycerol; sucrose was the solute used to vary the osmolarity. Assays for β-galactosidase were performed with cells sampled at three densities: OD₆₀₀ of 0.15, OD₆₀₀ of 0.6, and after growth overnight to stationary phase. Similar to the observations with E. coli (38), cells grown at high osmolarity showed an increase in rpoS-lac expression in exponential phase (OD₆₀₀ of 0.15 or 0.6), but not in stationary phase (data for OD₆₀₀ of 0.6 are shown in Table 2).

TABLE 2.

Effect of continuous growth at high osmolarity on expression of rpoS-lac

Con- structa	Fusion	β-Galactosidase activity (arbitrary units)b
		−Sucrose	+Sucrose	Ratio
A	rpoS-lac [pr]c	12	39	3.3
A	rpoS-lac [pr] hfq mutantd	1.5	4.4	2.9
B	rpoS-lac [op]e	190	210	1.1
C	rpoS-lac [pr] (KpnI)	0.41	0.44	1.1
E	rpoS-lac [pr] (codon 8)	9.8	30	3.1
F	Ptac-rpoS-lac [pr]	29	78	2.7
G	Ptac-rpoS-lac [pr] (KpnI)	8.1	18	2.2
J	Ptac-rpoS-lac [pr] (Δ2)	8	12	1.5
K	Plac UV5-rpoS-lac [pr] (KpnI)	12	28	2.3
K	Plac UV5-rpoS-lac [pr] (KpnI), hfq mutantd	1.9	4.6	2.4
L	Plac UV5-rpoS-lac [pr] (Δ1)	14	32	2.3
N	Plac UV5-rpoS-lac [op] (KpnI)	260	250	1.0
O	Plac UV5-lac [op]	170	146	0.9
P	Ptac-lac [op]	760	740	1.0

^aConstructs are described in the legend to Fig. 1. 

^bCells grown in a phosphate-buffered rich medium with glycerol (see Materials and Methods). 

^c[pr], protein fusion. 

^dUnless indicated, all other strains are hfq⁺. 

^e[op], operon fusion. 

Expression from construct A (native rpoS-lac) was increased ≈3-fold in high-osmolarity medium during exponential phase. Most tac or lac UV5 promoter constructs with rpoS-lac protein fusions (E, F, G, K, and L in Fig. 1) showed an induction ratio of 2- to 2.5-fold. The exception was construct J, deleted to just upstream of the proposed secondary structure, with an induction ratio of only about 1.5-fold. In contrast to the protein fusions, operon fusions (B, N, O, and P) were not detectably induced by high osmolarity (induction ratios of 0.9-1.1). Surprisingly, the hfq mutant derivatives of the wild-type fusion (construct A) and tac promoter construct (F) did not show a significant defect in the response to high osmolarity when tested by this method. These results suggest that continuous growth in high-osmolarity medium increases rpoS expression at a posttranscriptional level, but the mechanism is independent of HF-I. This contrasts with the result in osmotic challenge experiments, where the hfq mutant was partially defective, most severely in the case in which the native rpoS promoters were substituted with the tac promoter.

DISCUSSION

Previous work with S. typhimurium and E. coli has shown that the rate of synthesis of RpoS protein is reduced four- to sixfold in hfq mutants which lack HF-I (7, 33). Comparison of rpoS-lac protein and operon fusions suggests that the defect in hfq mutants lies at a posttranscriptional step, but it is not established whether HF-I specifically increases the rate of translation initiation or affects mRNA stability instead. The lack of an HF-I requirement for expression of rpoS-lac operon fusions is not incompatible with a role in mRNA stabilization, since the lacZ RBS of operon fusions could be insulated from such effects (28). There is as yet no in vitro system showing HF-I dependence of RpoS expression, so it is not certain that HF-I acts directly.

Suppressor mutations that decrease the in vivo dependence of rpoS-lac expression on HF-I function were found to map to a region encompassing ≈100 bp near the rpoS ATG codon, and genetic analysis of compensatory mutations suggests that an RNA secondary structure formed in this region limits rpoS expression (i.e., sequestration of the rpoS RBS by an intramolecular antisense RNA [8]). Perhaps the simplest model would be that HF-I binds to a site(s) in or near this region and disrupts the antisense pairing to allow ribosomes access to the rpoS mRNA. But although suppressor mutations may be suggestive of a potential mechanism, they do not necessarily recapitulate the role of HF-I for wild-type rpoS.

In this work, we tested whether such potential interactions of HF-I and the rpoS RBS region are sufficient for HF-I function by making promoter substitutions and by deletion of upstream segments from rpoS-lac transcripts expressed from the P_tac and P_{lac UV5} promoters. The results indicate that nonnative promoters still allow correct regulation by HF-I during growth into stationary phase and after osmotic challenge. However, in contrast to the prediction of the simple model, some sequences required for HF-I regulation of rpoS lie >100 nt upstream of the rpoS transcript antisense element. Reapplication of the method of compensatory mutations indicates that correct folding of the antisense-RBS structure is likely to be preserved in these deletion variants. Thus, HF-I must do more than simply melt this duplex.

We favor the idea that HF-I acts in the rpoS system very similarly to its function in the replication of E. coli RNA phage Qβ. There, HF-I is specifically required for copying of Qβ plus strands (1, 17, 47). Electron microscopy of HF-I protein bound to Qβ RNA reveals doubly looped structures that indicate specific and simultaneous interaction with two widely separated internal sites (independently bound by the replicase) which are then brought together with the RNA 3′ end (30). Such interactions are consistent with the multimeric nature of HF-I protein (17, 24) and build on Senear and Steitz’s demonstration of site-specific RNA binding activity by HF-I directed at RNA targets from phages R17 and Qβ (42). Mutations that disrupt the terminal RNA duplex of phage Qβ overcome the requirement for HF-I protein in phage replication (40), which suggests that HF-I might facilitate melting of the phage RNA 3′ end to allow initiation of replication. This terminal stem of 5 bp is not by any means the most stable element in the highly structured phage RNA, which suggests a specific role for HF-I in replication initiation.

Thus, we could imagine that HF-I binds to a specific upstream site in the rpoS mRNA and from that position interacts with downstream elements to melt the antisense-RBS duplex. This type of model could also accommodate recent genetic studies showing that the DsrA RNA, a small untranslated RNA of E. coli (45), acts directly to increase rpoS expression by pairing with and sequestering the upstream rpoS antisense element (18a, 46). DsrA-rpoS antisense RNA pairing requires HF-I, and this suggests that HF-I might act like the Rop (Rom) protein, a facilitator of the RNA-antisense RNA pairing that controls ColE1 plasmid copy number (50). Or perhaps, as suggested previously, HF-I is actually a chaperone for RNA and promotes structural rearrangements (35).

It seems less likely but still possible that HF-I has its primary effect on rpoS mRNA stability. RNase E cleavage sites are rich in A and U residues (reviewed in reference 10); these are also favored by HF-I. Another possibility is that bound HF-I might direct rpoS mRNA down an alternative folding pathway by preventing the formation of particular duplexes which are targets for RNase E. Effects on mRNA turnover have been suggested to explain the autoregulation of hfq gene expression in E. coli (52). Consistent with this, we have found that HF-I is much more promiscuous in its RNA binding activity than predicted from earlier studies (unpublished data). However, changes in mRNA turnover may also be a secondary consequence of changes in the rate of initiation of translation (11), so only a demonstration that rpoS mRNA turnover is not affected by HF-I would be definitive. Finally, any model must explain why the benefits of the postulated mRNA stabilization are not evident for an mRNA having an unfettered, strong RBS or for a variety of other rpoS single mutants that alter the antisense RNA element and thereby have become completely independent of HF-I (unpublished data).

In summary, current evidence in this system still allows a number of possible models including: (i) a looping interaction between HF-I complexed to rpoS mRNA at far upstream sites and at sites closer to the AUG codon; (ii) the existence of additional proteins, acting together with HF-I or indirectly under its control, which might bind to upstream sequences; (iii) a subtle influence of upstream sequences on the ultimate folding pattern of rpoS mRNA in the region near the AUG initiation codon; and (iv) a requirement for other components, such as small regulatory RNAs, including DsrA RNA, to observe tight complex formation between HF-I and the rpoS mRNA.