σ^S-Dependent Gene Expression at the Onset of Stationary Phase in Escherichia coli: Function of σ^S-Dependent Genes and Identification of Their Promoter Sequences

Abstract

The σ^S subunit of RNA polymerase, the product of the rpoS gene, controls the expression of genes responding to starvation and cellular stresses. Using gene array technology, we investigated rpoS-dependent expression at the onset of stationary phase in Escherichia coli grown in rich medium. Forty-one genes were expressed at significantly lower levels in an rpoS mutant derived from the MG1655 strain; for 10 of these, we also confirmed rpoS and stationary-phase dependence by reverse transcription-PCR. Only seven genes (dps, osmE, osmY, sodC, rpsV, wrbA, and yahO) had previously been recognized as rpoS dependent. Several newly identified rpoS-dependent genes are involved in the uptake and metabolism of amino acids, sugars, and iron. Indeed, the rpoS mutant strain shows severely impaired growth on some sugars such as fructose and N-acetylglucosamine. The rpoS gene controls the production of indole, which acts as a signal molecule in stationary-phase cells, via regulation of the tnaA-encoded tryptophanase enzyme. Genes involved in protein biosynthesis, encoding the ribosome-associated protein RpsV (sra) and the initiation factor IF-1 (infA), were also induced in an rpoS-dependent fashion. Using primer extension, we determined the promoter sequences of a selection of rpoS-regulated genes representative of different functional classes. Significant fractions of these promoters carry sequence features specific for Eσ^S recognition of the −10 region, such as cytosines at positions −13 (70%) and −12 (30%) as well as a TG motif located upstream of the −10 region (50%), thus supporting the TGN_0-2C(C/T)ATA(C/A)T consensus sequence recently proposed for σ^S.

Bacterial cells undergo a variety of morphological and physiological changes as they enter stationary phase. Several global regulators, involved in a complex regulatory network, induce the expression of stationary-phase-responsive genes (reviewed in reference 34). Among these, the rpoS-encoded alternative sigma factor σ^S plays a major role as a regulator of genes involved in stress responses (32). The σ^S protein accumulates in stationary phase as well as in response to stress conditions and directs transcription from about 100 genes (41, 47, 85). The high conservation (61) and the tight control of RpoS accumulation in response to a wide range of environmental stimuli are consistent with its crucial role in bacterial physiology (36). rpoS-regulated genes encode a variety of proteins with unrelated physiological functions, and mutations in rpoS have pleiotropic effects. Strains carrying nonfunctional rpoS alleles fail to express acid phosphatase (appA) (91), as well as oxidative stress genes such as superoxide dismutase (sodC) and catalase (katE and katG). rpoS-dependent regulation of oxidative stress genes is often mediated by additional regulatory proteins such as FNR, OxyR, or Fur (69). The rpoS gene also regulates the expression of DNA repair enzymes such as the exonuclease xthA (83), the methyl transferase ada (90), and the nonspecific DNA binding protein dps (3). The σ^S protein is also required for expression of the acid resistance gadA and gadB genes (21), for the response to either osmotic or temperature shifts (32), and in the expression of genes determining cell morphology, such as bolA (51). Finally, it regulates the expression of several virulence factors in pathogenic Escherichia coli strains e.g., csgBA genes encoding curli (6).

Sigma factors associate with core RNA polymerase (RNAP) to determine specific promoter recognition. The σ⁷⁰ protein, encoded by the rpoD gene, is the main σ factor in Escherichia coli and is responsible for the transcription of most genes. Alternative σ factors compete with σ⁷⁰ for core RNA polymerase to form the corresponding holoenzymes (e.g., σ^S-associated RNAP, or Eσ^S) and direct transcription of genes belonging to specific functional classes (41, 58). Unlike other alternative σ factors, σ^S recognizes promoters with sequences very similar to σ⁷⁰-dependent promoters, raising the problem of specific promoter recognition in vivo. Indeed, promoter alignment of rpoS-dependent genes suggests a −10 consensus sequence for Eσ^S [CTATA(A/C)T] basically identical to the Eσ⁷⁰ consensus (TATAAT) (23, 55). This was confirmed by in vitro systematic evolution of ligands by exponential enrichment (SELEX), which also suggested recognition of an identical −35 sequence (TTGACA) (24). These data are consistent with the strong similarity in the DNA binding domains (2.4 and 4.2) between the two σ factors (57) and with the observation that Eσ^S can initiate transcription at many σ⁷⁰-dependent promoters in vitro (88). On the other hand, Eσ⁷⁰ can transcribe some σ^S-dependent genes in the presence of additional factors or in the absence of specific repressors such as H-NS (6, 22, 87; reviewed in reference 56). Thus, at least for a subclass of promoters, promoter selectivity between Eσ⁷⁰ and Eσ^S in vivo might be determined by factors other than DNA sequence, such as increased intracellular salt concentrations, degree of DNA supercoiling, modulation by additional regulators (37, 41; reviewed in reference 34), or signal molecules such as the ppGpp alarmone (43, 46). However, specific deviations from the common consensus sequence might favor either form of RNAP; for instance, the presence of a C immediately upstream of the −10 element (−13C [9]), flexibility in the location of the TG motif, and the presence of a C as the first nucleotide of the −10 sequence might favor promoter recognition by Eσ^S (47).

In this work, we have studied the expression of rpoS-dependent genes at the onset of stationary phase in cell grown in rich medium. We observed rpoS-dependent induction of amino acid and carbohydrate metabolism genes, consistent with the role of RpoS in response to starvation. We show that rpoS controls the production of indole, which plays a role as a signal molecule in stationary phase. Finally, our results show that rpoS also regulates the expression of genes involved in protein synthesis and that of a large number of genes with unknown function, underlining the complexity of the RpoS regulon. We characterized the promoter sequences in a selection of these genes by using primer extension. Our data strongly support the importance of the proposed σ^S-specific features in the −10 region (−12C, −13C, and TG) as well as the lack of conservation of the −35 promoter element.

MATERIALS AND METHODS

Bacterial strains, growth conditions, and determination of indole production.

The rpoS derivative (EB1.3 [78]) of E. coli strain MG1655 that was used in the present study was obtained by phage P1 transduction from strain MV2792 (93). For RNA extraction, bacterial cultures were grown overnight in Luria broth (LB) medium at 37°C with vigorous aeration, diluted 1:100 in LB medium, and then grown to an optical density at 600 nm (OD₆₀₀) of 0.2 and rediluted 1:100 in 15 ml of prewarmed LB in 100-ml flasks to avoid carryover of proteins accumulated in late-stationary phase. Samples for RNA isolation, either for gene array and primer extension or for real-time PCR experiments, were collected either at an OD₆₀₀ of 2.5 to 3, a cell density that in our growth curve corresponds to the onset of stationary phase, or at an OD₆₀₀ of 0.6 to 0.8 (mid-exponential phase of growth). For growth on limiting sugar concentrations, 1 ml of a bacterial suspension from an overnight culture in LB medium was centrifuged, washed, and resuspended in sterile phosphate-buffered saline. Twenty microliters of bacterial suspensions was used to inoculate 2 ml of M9 medium supplemented with different sugars as the sole carbon source, either at 10 mM (control cultures) or at 0.1 mM (carbon-limiting conditions), and bacteria were grown at 37°C with full aeration for 24 h. Aliquots of the overnight cultures prior to and after centrifugation were plated on L agar plates; plate counts showed very similar bacterial concentrations for strains MG1655 and EB1.3 (ca. 5 × 10⁹ CFU/ml).

For determination of indole production, we followed the method of indole conversion into indigo described in reference 71, with minor modifications. Both strains MG1655 and EB1.3 were transformed with the pStyABB plasmid, which constitutively expresses the styrene monooxygenase styAB genes from Pseudomonas strain S12 (71). In addition to styrene oxidation, the StyAB protein also catalyzes the transformation of indole into indigo, whose blue color can be determined spectrophotometrically at OD₆₂₀ after lysis of the cells with dimethyl sulfoxide (DMSO) (71). Aliquots (1 ml) of either MG1655/pStyABB or EB1.3/pStyABB cultures grown in LB medium were taken at different times; the bacterial cells were centrifuged and lysed in DMSO, and the presence of indigo was determined spectrophotometrically. In an alternative method, 1-ml supernatants from overnight cultures of either MG1655 or EB1.3 were used to resuspend an equal volume of MG1655/pStyABB culture from the early-exponential phase of growth (OD₆₀₀, 0.25 to 0.3). The cells were incubated with the spent medium for 30 min at 37°C before DMSO lysis and spectrophotometric determination of indigo production.

RNA isolation, cDNA labeling, and hybridization.

Cells were harvested by centrifugation, and RNA was isolated with the RNeasy mini kit (QIAGEN). Total-RNA purification was performed by on-column DNase I digestion according to the manufacturer's instructions. RNA samples were quantified by using a spectrophotometer (260 nm), checked by gel electrophoresis, and stored at −80°C until further use. For microarray hybridization, 25 μg of RNA samples was used as a template for reverse transcriptase to produce cDNA labeled with either Cy3- or Cy5-dCTP (Cyscribe first-strand cDNA kit; Amersham Biosciences). Four different array hybridizations were performed using RNA samples extracted from two independent cultures under analogous conditions. To correct for possible differences in Cy3 and Cy5 dye incorporation, the cDNAs were labeled with Cy3 dye in two hybridizations and with Cy5 dye in the other two (dye swapping). For each of the four experiments, the Cy3- and Cy5-labeled cDNAs, products of the reverse transcriptase reactions, were pooled, purified by using QIAGEN Qiaquick spin columns, and concentrated with a Microcon-30 concentrator (Millipore) prior to addition of the hybridization buffer. For our investigation we used the E. coli K-12 V2 array (MWG), which contains 4,286 genes (http://www.mwg-biotech.com). Hybridizations were run overnight at 42°C in the buffer supplied, and subsequent washing steps with SSC buffer (1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate) were performed at room temperature (22°C).

Procedure for microarray data analysis.

Sixteen-bit tagged-image format file (TIFF) images produced by the scan of the microarray slides were analyzed with Affymetrix 428 Scanalyse and Jaguar software (version 2; Affymetrix), and data were managed by using Microsoft Excel. No ribosomal DNA spots are available on the MWG array slides for normalization. Thus, normalization was computed by using Jaguar software (average method): the mean values of an array are corrected by a factor which equalizes the global intensity of the signals resulting from each channel. An average background was determined for each individual hybridization experiment from the background values given by Jaguar for each spot with a specific dye. The normalized intensities of the spots were sorted according to the signals obtained for the wild-type MG1655 control samples (Cy3- or Cy5-labeled cDNA), and values lower than 1.3 times the average background were eliminated. This subtraction was performed in order to eliminate spots with nonsignificant expression levels in the wild-type strain, which were therefore unsuitable for addressing down-regulation due to the rpoS mutation. Several known rpoS-dependent genes (e.g., bolA, katE) fell into this category and were not further considered in our analysis, despite a high wild-type/rpoS ratio. Data from the four array experiments were flagged, pooled, and sorted to identify reproducible spots. Genes corresponding to spots with a wild type/rpoS signal ratio (R) of >2.5 in at least three out of four hybridizations, which never showed an R of <1 and whose final average ratio was ≥2.5, were considered down-regulated in the rpoS strain and are listed in Table 2.

TABLE 2.

Genes induced at the onset of stationary phase with σ^s -dependent expression as identified by microarray analyses

Gene	Gene product	Functional classa	Mean WT valueb	Ratio (WT/rpoS) SDc	Regulationd
Starvation and stress response genes
dpsS	DNA protection protein, bacterial ferritin	SSR	1,597	2.5 ± 0.4	IHFS, OxyRS, rssBTC, MMC, PQ, NaSal, Lrp, BG, DHCP, OS
osmES	Stress-inducible outer membrane lipoprotein	SSR	1,728	9.6 ± 2.1	FisS, NaSal, Lrp, OS
osmYS	Hyperosmotically inducible periplasmic protein	SSR	424	4.4 ± 1.3	CRPS, LrpS, IHFS, H-NSS, arcBTC, MMC, Lrp, Rcs, BG, DHCP
psiF	Induced by phosphate starvation		656	4.8 ± 0.4	ompR/envZTC, CD
sodCS	Superoxide dismulase precursor (Cu-Zn)	SSR	459	2.7 ± 0.5	uvrYTC, arcBTC
Protein synthesis
rpsVS	30S ribosomal subunit protein S22	PS	3,466	5.2 ± 2.8	FISS, IHFS, ppGppS
infA	Protein chain initiation factor IF-1	PS	1,011	2.7 ± 0.8
Amino acid transport and metabolism
artP	Arginine transport, ATP-binding component	AATM	714	2.6 ± 0.7	PQ, NaSal, Lrp
artI	Arginine transport, periplasmic binding protein	AATM	346	3.8 ± 1.9	PQ, Lrp
ansP	L-Asparagine permease	AATM	572	2.5 ± 0.4	DHCP
ilvD	Dihydroxyacid dehydratase	AATM	364	4.8 ± 2.1	ompR/envZTC
tnaA	Tryptophanase	AATM	7,808	13.7 ± 4.3	phoQTC, ompR/envZTC, CRP, uvrYTC, arcATC, arcBTC, rcsBTC, citABTC, MMC, PQ (ilvG: Lrp)
wrbAS	Flavodoxin-like protein; binds the trp repressor	SSR	364	5.0 ± 2.0	NaSal, Lrp, BG, DHCP, OS
Iron uptake and storage
bfr	Bacterioferrin, iron storage homoprotein	IUS	905	5.1 ± 2.1	DHCP, ryhB RNA
entD	Enterochelin synthetase, component D	IUS	508	4.2 ± 0.5	acrBTC
Carbohydrate metabolism
crr	Phosphotransferase system, glucose-specific IIA component	CHM	443	2.7 ± 0.6	CD
fbaB	Fructose 1,6-bisphosphate aldolase (b2097)	CHM	1,264	4.6 ± 1.4	MMC, Lrp
gip	Glyoxylate-induced protein		384	3.6 ± 1.0	kdpABCDETC
glcG	Putative oxidase, glycolate utilization (b2977)	CHM	462	3.0 ± 1.7	uvrYTC, arcATC, arcBTC
nagB	Glucosamine-6-phosphate deaminase	CHM	860	2.6 ± 0.6	ompR/envZTC, arcBTC
pfkB	Phosphofructokinase-2	CHM	512	2.5 ± 2	MMC, CD, DHCP
talA	Transaldolase A	CHM	1,097	2.9 ± 1.8	rssBTC, MMC, DHCP, CreBC
Miscellaneous function or unknown function
deoD	Purine nucleoside phosphorylase	NM	349	4.8 ± 3.0
moeB	Molybdopterin biosynthesis	CB	1,100	3.4 ± 1.2
msyB	Membrane protein, protein export (b1051)	MF	516	6.1 ± 4.0	CD, Lrp, DHCP
yafN	Hypothetical (adherence) protein (b0232)	(MF)	354	2.6 ± 0.1	MMC, GadX
yahOS	Hypothetical protein (b0329)		933	4.5 ± 2.2	Lrp
ybgA	Hypothetical (inner membrane) protein (b0707)		2,038	4.5 ± 1.7
ybjP	Putative enzyme (b0865)		1,317	4.2 ± 2.5	rssBTC, baeSRTC, atoSCTC, Lrp
yddX	Hypothetical protein, biofilm dependent (b1481)	(EF)	376	4.2 ± 1.4
ydiZ	Hypothetical protein (b1724)		707	3.1 ± 1.7	rssBTC, OS
yfeT	Hypothetical protein (b2427)		341	3.5 ± 0.9	ompR/envZTC
ygaF	Putative dehydrogenase protein (b2660)		453	2.8 ± 0.5	rssBTC, NarXLTC
yggE	Putative immunogenic protein (b2922)	(EF)	813	4.8 ± 1.7	rssBTC, MMC; Rcs, phoQP
ygiS	Putative transport periplasmic protein (b3020)	(MF)	436	2.7 ± 0.8
ygiW	Hypothetical (outer membrane) protein (b3024)	(MF)	892	4.0 ± 1.9	rssBTC, citABTC, arcATC, NaSal
yiaG	Hypothetical (regulatory) protein (b3555)		1,227	2.7 ± 0.5	rssBTC, ntrBCTC, NaSal
yjbJ	Hypothetical protein (b4045)		601	4.5 ± 1.2
ynfA	Hypothetical (membrane) protein (b1582)	(MF)	3,334	3.5 ± 1.1
yqjC	Hypothetical (periplasmic) protein (b3097)		1,121	5.9 ± 1.3
ytfK	Hypothetical protein (b4217)		556	3.9 ± 3.0	PQ, Fur, BG

^aSSR, starvation and stress response; PS, protein synthesis; AATM, amino acid transport and metabolism; IUS, iron uptake and storage; CHM, carbohydrate metabolism; NM, nucleoside metabolism; CB, cofactor biosynthesis; MF, membrane function; EF, extracellular function. Parentheses indicate a putative function.

^bMean relative expression level in the wild-type (WT) strain; average values of the signal intensities obtained in the wild-type strain in four experiments without correction (i.e., without dye swapping).

^cAverages and standard deviations from four experiments are presented. The genes listed in this table presented a ratio of expression in the wild-type versus the rpoS strain (R) greater than 1.9 in at least three out of four experiments and never presented an R of <1. Genes with an average R lower than 2.5 were excluded from this table, except for ytfK and pfkB, which were confirmed by primer extension, and artI, located in a cluster with ybjP and artP, both of which are down-regulated in the rpoS strain.

^dGenes known to be regulated by RpoS are marked with a superscript capital S. Relevant references for these genes are given in parentheses dps (3), osmE (11, 17), osmY (8, 50, 101), sodC (27), rpsV (sra [42]), wrbA (100), and yahO (40). Growth conditions or regulatory pathways known (by either genomic or genetic analysis) to affect expression of the genes listed include the following: (i) disruption of known two- component (TC) regulatory systems (74); (ii) knockout of the specific regulatory gene(s) fur (10), rcs or phoQP (30), lrp (39, 89), gadX (64), creBC (7), or ryhB (63); (iii) specific growth conditions, including exposure to mitomycin C (MMC) (44), oxidative stress induced by either paraquat (PQ) or sodium salicylate (NaSal) (76), biofilm growth (BG) (86), 4,5-dihydroxy-2-cycloenten-1-one (DHCP) treatment (antimicrobial agent [75]), high-cell-density (CD) culture (102), osmotic stress (OS) response (supercoiling- dependent genes [16]), and minimal versus LB medium (MD) (97).

RT PCR.

For real-time PCR (RT-PCR), reverse transcription from RNA samples (extracted as for the microarray experiments) was carried out using 62.5 U of MultiScribe (Applied Biosystems) reverse transcriptase per μg of total RNA, in the presence of 1.25 μM random hexamers. Amplification reactions were carried out in an Applied Biosystems ABI Prism 7000 sequence detection system using the SYBR Green PCR master mix according to the manufacturer's standard protocol. cDNA produced from 20 ng of total RNA was used for each reaction. The internal-control 16S rRNA gene, used to normalize the values obtained for the genes analyzed, was amplified by primers rrsBfw (5′-GAATGCCACGGTGAATACGTT) and rrsBrev (5′-ACCCACTCCCATGGTGTGA). The PCR products of the genes of interest (shown in Fig. 1) were generated by using the primers listed in Table 1. All reactions were performed in triplicate, in addition to samples using DNase I-digested RNA as templates, to verify the lack of residual DNA. Real-time PCR data were analyzed by using ABI PRISM 7000 SDS software (version 1.0) and according to Relative Quantitation of Gene Expression (P/N 4303859), issued by the manufacturer.

FIG. 1.

Real-time PCR experiments. Values are shown as induction factors for expression in the wild-type (WT) strain MG1655 versus the rpoS mutant strain EB1.3 in the exponential (white bars) and stationary (black bars) phases of growth. The genes tested are indicated in the figure. Values were normalized to those for the 16S rRNA gene, and results shown are averages of three experiments. The horizontal line shows the 2.5-fold threshold used to select rpoS-dependent genes in the microarray experiments.

TABLE 1.

Primers used in this study

Name	Sequence
For RT-PCR
osmYfw	5′-GTCGATAGCTCTATG-3′
osmYrev	5′-GAAGGCGGCCCTGGG-3′
artPfw	5′-GAGTATTCAATTAAACGGC-3′
artPrev	5′-CCCTGTGGGCAATCCAGCGTG-3′
infAfw	5′-CGGTCACGTGGTTACTGCAC-3′
infArev	5′-GTCGCCCGTCAGGATGCG-3′
fbaBfw	5′-CGGGAGCATAGTAATGAC-3′
fbaBrev	5′-ACGGTGCTGTAAAAGGTTGTCG-3′
nagBfw	5′-CTCAACGGCAACGCCCCGG-3′
nagBrev	5′-GGCTCGTTACCTACACCGCCC-3′
bfrfw	5′-CTGGGGTCTCAAACGTCTC-3′
bfrrev	5′-CGGCGTGTTTCATCTCATC-3′
ybgAfw	5′-GATTTGATGACAGCACCCTAATC-3′
ybgArev	5′-AGCGGCTGTGATAATCGAGAAGC-3′
yggEfw	5′-GTGAAGTTCAAAGTTATCGCC-3′
yggErev	5′-TGACAATATGCGGTCCATCCG-3′
yiaGfw	5′-GATCCAATGCATGAGCTGTTGAG-3′
yiaGrev	5′-AGGACGTTGTTCTGTGCGTCAGG-3′
ytfkfw	5′-GGTACAACCCACTTCAGGTG-3′
ytfKrev	5′-TCCTTGATGTATAACCGTCC-3′
For determination of transcription start sites
ansP	5′-AATGGCGATCATCTGCACCTGG-3′
artP	5′-CCCTGTGGGCAATCCAGCGTG-3′
dps	5′-AACTGGATAACCTGGCGATTCAG-3′
fbaB	5′-ACGGTGCTGTAAAAGGTTGTCG-3′
mysB	5′-TTTGGGCATTGAACTGTTGCAC-3′
pfkB	5′-AACACCGGTGCGGTACAGCG-3′
rpsV	5′-TGCCGGTTGGGTTATTTACTACG-3′
talA	5′-TGAGTAACAGCGAAGGATTGGTG-3′
ybjP	5′-TTCAACGCAAGGACCACTGCG-3′
ybgA	5′-AGCGGCTGTGATAATCGAGAAGC-3′
yggE	5′-TGACAATATGCGGTCCATCCG-3′
yiaG	5′-AGGACGTTGTTCTGTGCGTCAGG-3′
ytfK	5′-TCCTTGATGTATAACCGTCC-3′

Primer extension.

Samples were collected, and RNA was extracted, in the same way as for microarray hybridization (see above). Twelve micrograms of RNA was used by the primer extension assay with ImpromII reverse transcriptase (Promega) according to the manufacturer's instructions. RNA and labeled primers were annealed at 70°C (for 5 min), and the reaction proceeded at 42°C for 1.5 h. The labeled extended fragments were separated from mRNA by heating at 95°C and were placed on ice prior to addition of formamide dye (MWG); the samples were then loaded onto a 7% polyacrylamide-urea denaturing sequencing gel, and transcription start sites were determined by comparison with a known DNA sequence used as a molecular weight marker. Reaction products were quantified with ImageQuant (Molecular Dynamics) after background subtraction from the TIFF images generated by the laser scanner (Li-Cor 4200 long reader; MWG). Primers used in this assay are shown in Table 1.

Other genetic methods.

The tnaA gene was inactivated by the λ red gam method as described in reference 20, producing strain PL614. Insertion was verified by direct amplification of the tnaA gene. To measure tnaA gene expression, its promoter region, including the short leader peptide for the TnaA protein encoded by the tnaL gene, was amplified by PCR using primers tnaA1 (CTGAAGCTTGATTGTGATTCGATTC, annealing at positions −392 to −376 relative to the tnaA start codon) and tnaA2 (CGGAATTCCTCTTCACGATAAGC, annealing at positions +76 to +90 relative to the tnaA start codon). The primers introduced a HindIII and an EcoRI site (underlined in the primer sequences) into the amplified fragment, which was then subcloned into the multicopy plasmid pRS1274 (49) by using the corresponding restriction sites. β-Galactosidase activity was determined as described in reference 66.

Sequence analysis and gene function.

The pattern search tools of Colibri (http://genolist.pasteur.fr/Colibri/genome.cgi) (65) were used to search for putative −10 promoter elements upstream of the primer extension products. The Colibri, PromEC (38) (http://bioinfo.md.huji.ac.il/marg/promec/prom.seq.final.html), and RegulonDB (84) (http://www.cifn.unam.mx/Computational_Genomics/regulondb) databases, and J. E. Mitchell's collection of σ⁷⁰-dependent promoters described in reference 67 (http://www.biosciences.bham.ac.uk/labs/minchin/mitchell2003), were used for analysis of promoter and operon structures. Finally, the Swiss-Prot (http://www.expasy.org/) and ProDom (http://prodes.toulouse.inra.fr/prodom/current/html/home.php) databases were used to gather more information about genes encoding hypothetical proteins.

RESULTS AND DISCUSSION

rpoS-dependent genes expressed at the onset of stationary phase.

Whole-genome expression in either E. coli MG1655 or its rpoS mutant derivative was determined at the onset of stationary phase (i.e., cells were harvested as soon as an increase in cell turbidity was no longer detectable) for cells grown in LB (rich) medium. At this stage of bacterial growth, we expected that the subset of rpoS-regulated genes mostly involved in the establishment of stationary-phase physiology would be maximally expressed. Through microarray experiments we detected 41 genes whose expression was significantly reduced in the rpoS strain (Table 2); only 7 of these had already been described as rpoS dependent (dps, osmE, osmY, rpsV, sodC, wrbA, and yahO). Known rpoS-dependent genes not detected in our experiment might be expressed only at later stages of stationary phase or in different growth media and conditions; however, lack of detection of some rpoS-dependent genes might be due to shortcomings of the microarray technique and to the criteria chosen for microarray analysis (see Materials and Methods). The known rpoS-dependent genes found in our experiments mainly encode stress response proteins with either regulatory or detoxification functions (35, 41; reviewed in reference 33). Among the newly identified σ^S-regulated genes, almost 50% have unknown functions. The others belong to four main functional classes: genes involved in amino acid transport and metabolism, iron uptake and storage, protein synthesis, and carbohydrate and nucleoside metabolism. It is noteworthy that the newly identified rpoS-dependent genes do not appear to be essential genes as defined by Gerdes et al. (26), except for infA (18), a finding consistent with the lack of major effects of the rpoS mutation on cell viability.

To confirm the results of the whole-genome expression analysis, we performed RT-PCR experiments on a selection of differentially expressed genes belonging to each functional category (Fig. 1). RT-PCR experiments confirmed the increased expression of all the genes tested in strain MG1655 compared to the rpoS mutant strain EB1.3. Increased gene expression in MG1655 was detectable only at the onset of stationary phase; samples taken in mid-exponential phase (OD₆₀₀, 0.6 to 0.8) showed no significant differences in gene expression between the two strains (Fig. 1).

Genes involved in cell metabolism.

We found several genes involved in amino acid transport (artI, artP, and ansP) or metabolism (tnaA and ilvD) to be induced in a σ^S-dependent fashion at the onset of stationary phase. In particular, rpoS appears to promote tryptophan degradation by inducing the tryptophanase gene tnaA, which results in the production of indole, and by stimulating expression of the WrbA protein, which might strengthen negative regulation of the trp biosynthetic operon via binding to the TrpR repressor protein (100). Expression of the wrbA gene has already been shown to be rpoS dependent (41, 100). Interestingly, indole has been proposed to act as an extracellular signal in stationary cells of E. coli (95), and tnaA-mediated indole biosynthesis appears to stimulate biofilm formation (62). We investigated the possibility that indole production is negatively affected by rpoS mutation. Indole is converted into indigo (which is not further degraded in E. coli) by several monooxygenases, thus providing an easy method for its determination (71). We transformed both strains MG1655 and EB1.3 with pStyABB, a plasmid expressing styrene monooxygenase, and monitored indole production through its conversion into indigo (Fig. 2A). The production of indole closely follows the growth curve in MG1655, as revealed by indigo accumulation; in contrast, indole production is totally abolished in a tnaA-null mutant and severely affected by inactivation of the rpoS gene (Fig. 2A). To rule out the possibility that indigo production might be affected by different expression of styrene monooxygenase in the rpoS mutant strain, we exposed MG1655/pStyABB cells in early-exponential phase to filtered supernatants of either MG1655 or EB1.3 cultures. MG1655/pStyABB produced indigo only when treated with MG1655 conditioned medium, confirming that the rpoS mutant strain EB1.3 is deficient in indole production (data not shown). The results of the microarray experiments strongly suggest that lack of indole production in the rpoS mutant strain depends on reduced tnaA expression, which was further confirmed by tnaA gene expression measurement using the lacZ reporter gene (Fig. 2B).

FIG. 2.

(A) Indole production in strains MG1655 (wild type) (black diamonds), EB1.3 (rpoS) (black squares), and PL614 (tnaA) (black triangles), all carrying the pStyABB plasmid. The indole concentration was determined through its conversion into indigo by the StyAB monooxygenase as described in Materials and Methods. The growth curve in LB for MG1655 is shown by the dashed line and open circles. The growth rate of either the EB1.3 or the PL614 strain was not affected by the mutation. Samples for RT-PCR (Fig. 1) from exponential- and stationary-phase cells were collected at time points t₁ (OD₆₀₀, 0.6 to 0.8) and t₂ (OD₆₀₀, 2.5 to 3), respectively. (B) tnaA promoter activity from a multicopy plasmid in strains MG1655, EB1.3, and PL614. Symbols are as explained for panel A.

While tryptophan is degraded in stationary phase, accumulation of the amino acids arginine and asparagine appears to be stimulated by rpoS through activation of the artP and artI genes, whose products are components of the arginine transport system and of the asparagine transporter ansP. Thus, the rpoS gene seems to play an important role in determining intracellular amino acid concentrations in stationary phase. This function might be related to the synthesis of amino acid- or peptide-derived signal molecules, as for indole, or to the need to redirect general amino acid metabolism in stationary phase. It is noteworthy that both the artP and the wrbA promoter are controlled by the global regulator Lrp (leucine-responsive protein) (Table 2), the main regulator of amino acid biosynthesis and metabolism. The unknown ybjP gene, located immediately upstream of the art operon but controlled by its own promoter (see Fig. 4), is also regulated by lrp (39) as well as by rpoS. Concerted regulation by RpoS and Lrp has already been reported for a number of genes (osmC, osmY, aidB, and csiD [13, 48, 50, 59]), suggesting tight integration of Lrp- and RpoS-dependent regulation.

FIG. 4.

Promoter sequences of σ^S-regulated genes induced at the onset of stationary phase, characterized by primer extension. The −10 sequence of each promoter is boldfaced, as is the conserved −35 element when present. Nucleotides that match the proposed consensus for the −10 element [(TGN_0-2)CYANNMT, where Y stands for C or T, M stands for A or C, and N stands for any nucleotide (47)] or the −35 element (TTGACA) at a functional location (15 to 17 nucleotides away from the −10 sequence) are underlined (for the −10 sequence) or double underlined (for the −35 sequence). The transcription start sites are lowercased, boldfaced, and italicized. Recurrent CG dinucleotides in the −35 region and the CCG motifs proposed by Wise et al. (98) are also underlined. Promoters marked with an asterisk are known promoters (3, 19, 42). P₊₁ indicates the position of the transcription start site relative to the start codon. WT and rpoS designate the products of primer extension obtained with total mRNA from the wild-type strain MG1655 and its rpoS derivative, respectively; the sizes of the primer extension products were determined by use of a DNA sequencing ladder of a known sequence run on the same gel. R_PE indicates the WT/rpoS expression ratio as determined from the primer extension experiments. R_MA indicates the WT/rpoS expression ratio as determined from the microarray experiments (Table 2).

We also detected rpoS-dependent activation of genes encoding carbohydrate metabolism proteins such as the carbohydrate phosphotransferase systems (crr) or glycolysis enzymatic proteins (fbaB and pfkB). Interestingly, both the fbaB and talA genes encode isoenzymes of fbaA and talB, genes that do not appear to be regulated by rpoS, suggesting that σ^S could up-regulate the expression of isoenzymes that are possibly more functional during the stationary phase of growth. The rpoS gene also appears to be involved in alternative metabolic pathways for carbohydrate utilization such as the pentose phosphate or the glyoxylate shunt to the carboxylic acid pathway (glcG or gip). In addition, rpoS activates the catabolism of N-acetylglucosamine through induction of nagB. We directly tested the possible role of rpoS in the expression of carbohydrate metabolism genes by growing strains MG1655 and EB1.3 on limiting amounts of different sugars as sole carbon sources. Both the parental and rpoS strains were able to grow to an OD₆₀₀ of >1.0 on 10 mM concentrations of glucose, fructose, N-acetylglucosamine, or sucrose, with similar growth rates (data not shown). However, in the presence of a growth-limiting sugar concentration (0.1 mM), the rpoS mutant showed significantly impaired ability to grow, ranging from a twofold reduction for growth on glucose to an eightfold reduction for growth on sucrose (Fig. 3), despite the fact that the inocula contained similar numbers of viable cells (data not shown) (see Materials and Methods). Although rpoS dependence of sugar uptake or catabolism genes would be consistent with their induction in response to starvation, it is in contrast with previous observations suggesting that rpoS might negatively regulate the nagB and crr genes in continuous cultures grown under sugar-limited conditions (92). It is possible that rpoS might be involved in feedback regulation of carbohydrate metabolism genes. rpoS could activate the expression of sugar uptake and metabolism genes at the onset of stationary phase, to scavenge for the presence of low concentrations of sugars. However, in the absence of the specific inducers, expression would be shut down by rpoS itself through indirect regulation at a later stage of the stationary phase. An example of feedback loop regulation by rpoS has already been described for the curli-encoding csg operon (6, 78).

FIG. 3.

Growth of MG1655 (wild type) (white bars) or EB1.3 (rpoS) (grey bars) on limiting concentrations (0.1 mM) of either glucose, sucrose, fructose, or N-acetylglucosamine (N-Ac-Glu). The growth experiments were performed as described in Materials and Methods. Values represent the final OD₆₀₀ reached at the end of the 24-h incubation and are averages from three experiments.

Iron-dependent genes.

Iron is an essential cofactor of many enzymes, and bacteria have evolved a range of strategies to acquire and store iron, which is often not bioavailable in the environment. E. coli possesses two main iron storage proteins: bacterioferritin and ferritin (4). Bacterioferritin, whose precise function is still unclear (1), is induced during slow growth or at the transition to stationary phase, consistent with its regulation by rpoS. The entD gene encodes enterochelin, a synthase of a catechol siderophore from chorismic acid; the EntD protein is located in the inner membrane, but its function is not fully understood (5, 29). The gene is located downstream of the fepA gene, apart from the rest of the ent cluster, suggesting possible differential regulation. Regulation of ferritins and iron-dependent genes by rpoS is not entirely surprising, since rpoS is already known as the main regulator of dps expression. Although the Dps protein was originally described as a nonspecific DNA binding protein involved in resistance to oxidative stress (60), it is actually a bacterial ferritin (28), whose main function could be to chelate intracellular iron and thus to prevent DNA damage through Fenton reaction-catalyzed oxyradical formation (31). Many iron-dependent genes are regulated by Fur (ferric uptake regulator), which, in the presence of iron, represses the transcription of iron-responsive genes (25, 63, 80). A recent report proposes that Fur accumulation would be modulated by rpoS in Vibrio vulnificus (52), pointing to a possible overlap of fur- and rpoS-dependent gene regulation in enterobacteria.

Protein synthesis.

We found two genes involved in protein synthesis to be activated in an rpoS-dependent fashion under the conditions tested. Interestingly, in addition to the already known rpsV gene, we could show that transcription of the infA gene is also stimulated by rpoS (Table 2; Fig. 1). The infA gene encodes the initiation factor IF-1, whose precise role has not yet been assigned (77). Similarly, no precise function has yet been assigned to the product of rpsV, a small, stationary-phase-induced, ribosome-associated protein (SRA) which appears to be specific to enterobacteria (42). The RpsV protein might play a role similar to that of the ribosome-modulating factor (RMF), i.e., to stabilize ribosomes in stationary phase. The RMF protein is also produced in stationary phase, although not in an rpoS-dependent fashion (94), and E. coli mutants with defects in rmf cannot survive long periods of starvation (99). Recent findings point to a close relationship between protein synthesis and entrance into stationary phase: aberrant proteins result from increased mistranslation in nonproliferating bacteria and are likely to accumulate in stationary phase (70). We propose that σ^S-dependent induction of IF-1 and SRA might either prevent translation errors or modulate protein synthesis by blocking ribosomes in an inactive form.

Genes of unknown function.

The rpoS mutation affects the expression of 16 genes encoding hypothetical proteins of unknown function, several of which might have functions associated with membrane trafficking (ybgA, ygiS, ygiW, ynfA, and yqjC) or might be located at the cell surface (yafN, yddX, and yggE). Some of these proteins are common to a wide number of enterobacteria or, in the case of ynfA and yiaG, even to a larger group including gram-positive bacteria. Interestingly, the yahO gene has previously been reported to be regulated by RpoS in Salmonella enterica (40); the YahO protein shares 66% identity and 79% similarity with its E. coli homologue. The yggE gene, conserved in other enterobacteria such as Shigella flexneri and Edwardsiella ictaluri at an identical locus (68, 96), encodes a putative extracellular protein, considered a potential antigenic protein for development of a vaccine against E. ictaluri in catfish (68). Two rpoS-dependent genes encode either a potential adhesion factor (yafN) or a factor involved in biofilm formation (yddX [79]). Another gene, ytfK, is up-regulated in E. coli cells growing as a biofilm (86). These observations would support a general positive role of rpoS in biofilm formation (2), possibly mediated by indole production (Fig. 2) (62), consistent with the dependence on rpoS of known adhesion and biofilm formation determinants such as the csg genes in Salmonella and in pathogenic E. coli strains (73).

Promoter sequences of σ^S-regulated genes.

Primer extension assays were performed to determine the promoter sequences of the genes identified and to confirm the results of the gene array experiment. We selected 11 genes belonging to different functional classes and showing variable expression levels and wild-type/rpoS mutant ratios in the gene array experiment. In addition, we performed primer extension assays on the dps and rpsV genes, whose σ^S-dependent promoters have already been characterized, as a control for our experimental conditions. The promoter sequences found for both dps and rpsV correspond to those previously described (3, 42); this was also the case for the known pfkB promoter, which, however, had not yet been identified as a σ^S-dependent promoter (19). Primer extension was performed from total RNAs of both wild-type and rpoS strains, and transcription start sites were determined by comparison with a DNA sequencing reaction ladder (see Fig. S1 in the supplemental material). Quantification of primer extension products consistently showed lower levels of expression in the rpoS mutant, confirming the results of the gene array experiments (Fig. 4). For several rpoS-dependent genes, multiple signals were detected in the primer extension experiments (ansP, artP, talA, pfkB, ybgA, and yggE), which might depend either on secondary RNA structures resulting in pausing of reverse transcriptase, or on the presence of multiple promoters. Indeed, well-conserved promoter-like sequences could be detected upstream of most reverse transcriptase extension products, suggesting that the presence of multiple promoters might be a rather common feature for rpoS-dependent genes.

Our results do not allow us to ascertain that all rpoS-dependent genes identified are directly transcribed by Eσ^S. However, we expect that most promoters induced at the onset of stationary phase should belong to the class of promoters recognized with high affinity by the σ^S subunit of RNA polymerase and thus should display promoter features favorable for Eσ^S. Indeed, the sequences of the −10 regions of the promoters tested are in good agreement with previous compilations of σ^S-dependent promoters (23, 47, 54) and reflect the recently proposed motif TGN_0-2CYATAMT (47). The −10 sequence of around 70% (12 of 18) of the newly characterized σ^S-dependent promoters is immediately preceded by a C at the position conventionally indicated as −13; −13C is a widely conserved feature for promoter recognition by Eσ^S (9, 15). Six (30%) promoters (ansPp₁, ansPp₂, fbaB, talAp₂, pfkBp₁, and yggEp₂) display a C as the first nucleotide of the −10 promoter element (conventionally −12C), a feature that might prevent their transcription by Eσ⁷⁰ (47). A TG motif is present upstream of the −10 element in 10 (50%) of the newly determined promoters, at positions from −17 to −14. In a previous study (47), we proposed that σ^S can recognize the TG motif at different positions, in contrast to σ⁷⁰, which can interact only with a TG placed at −15/−14 (14, 67). In the novel σ^S-dependent promoters identified in this report, the distribution of the TG motif is not as wide as that which we had found at the 56 previously characterized σ^S-dependent promoters (47), as it is more restricted to the −15/−14 position. Interestingly, the artPp₃ promoter, one of the promoters with the lowest σ^S/σ⁷⁰ activity ratio, harbors a double TG, which is supposed to further enhance promoter recognition and transcription initiation by both forms of polymerase (24, 67).

The conservation of the amino acids of region 4.2 of both σ factors and the results obtained by the SELEX approach (24) suggest that Eσ^S might be able to recognize a −35 element and would have the same optimal sequence as σ⁷⁰ (TTGACA). At some rpoS-dependent promoters, such as osmE (12), disruption of the −35 sequence (TTGAAA) has a negative effect on σ^S-mediated transcription. However, data on the importance of a −35 sequence for Eσ^S are controversial, and very few known σ^S-dependent promoters possess a −35 sequence similar to that recognized by Eσ⁷⁰ (12, 72, 87, 98). Among the newly identified promoters, only three (pfkB, ytfK, and yggE) have a discernible −35 element with four of six matches to the σ⁷⁰ consensus sequence, possibly suggesting that Eσ^S-specific −35 sequences might strongly diverge from the Eσ⁷⁰ consensus (Fig. 4). It has been proposed that DNA bending upstream of the promoter region as well as motifs rich in C/G in the −35 region would favor σ^S selectivity (45, 98). Although our results do not allow us to confirm this hypothesis, we did find fairly high occurrence of GC or CG dinucleotides in the −35 area (Fig. 4). Indeed, the artPp₁, artPp₂, and yiaG promoters carry a CCG sequence, proposed to be a σ^S-specific −35 promoter sequence by Wise et al. (98), while dps and ytfK display the GCGG motif recently proposed as an osmotic shock-responding module in a subset of σ^S-dependent promoters (53).

Finally, five promoters (ansPp₁, fbaBp₁, pfkB, talA, and ytfK) possess UP-like elements, i.e., putative binding sites for the α subunit of the RNAP (82). The role of the UP element in promoter recognition by Eσ^S has not yet been addressed in detail; at the aidB promoter, an UP-like element functional for Eσ⁷⁰ also contributes to Eσ^S-dependent transcription in vitro, suggesting that UP elements might indeed play a role in Eσ^S-promoter interactions (S. Lacour and P. Landini, unpublished data).

Conclusions.

In this report, we compared the expression profiles of strain MG1655 and its rpoS derivative in LB (rich) medium. Since we tested rpoS-dependent expression at only one time point (the onset of stationary phase) and in one growth medium, our results give only an incomplete picture of the rpoS regulon; however, they provide information on genetic and physiological adaptation to stationary phase under the conditions tested, as well as on the structure of rpoS-dependent promoters. Indeed, in addition to known stress response genes, we found genes involved in a variety of metabolic functions to be induced in an rpoS-dependent fashion. The rpoS gene appears to be directly involved in the biosynthesis of signal molecules such as indole, which plays a role in biofilm formation in some E. coli strains (62) and might regulate gene expression in coordination with other cell-signaling systems directly related to quorum sensing and cell density (81). Our results also point to a direct role of σ^S in response to nutrient starvation, suggesting that the σ^S regulon overlaps with other nutrient-specific regulatory pathways such as cAMP/CRP, NtrB/NrtC/σ⁵⁴, and PhoB/PhoR. We could confirm with primer extension experiments that, for a large number of the newly identified σ^S-dependent promoters, the −10 promoter element matches the proposed consensus TGN_0-2CYATAMT. In contrast to strong conservation in the −10 region, little similarity to the −35 sequence for Eσ⁷⁰ and no significant presence of UP element sequences could be detected.