The Bacterial Signal Molecule, ppGpp, Mediates the Environmental Regulation of Both the Invasion and Intracellular Virulence Gene Programs of Salmonella*

During infection of mammalian hosts, facultative intracellular pathogens have to adjust rapidly to different environmental conditions encountered during passage through the gastroin-testinal tract and following uptake into epithelial cells and macrophages. Successful establishment within the host therefore requires the coordinated expression of a large number of virulence genes necessary for the adaptation between the extracellular and intracellular phases of infection. In this study we show that the bacterial signal molecule, ppGpp, plays a major role in mediating the environmental signals involved in the regulation of both the extracellular and intracellular virulence gene programs. Under oxygen limiting conditions, we observed a strong ppGpp dependence for invasion gene expression, the result of severe reductions in expression of the Salmonella pathogenicity island (SPI) 1 transcriptional regulator genes hilA, C, and D and invF. Overexpression of the non-SPI1-encoded regulator RtsA restored hilA expression in the absence of ppGpp. SPI2-encoded genes, required for intracellular proliferation in macrophages, were activated in the wild type strain under aerobic, late log phase growth conditions. The expression of SPI2 genes was also shown to be ppGpp-dependent under these conditions. The results from this study suggest a mechanism for the alternate regulation of the opposing extracellular and intracellular virulence gene programs and indicate a remarkable specificity for ppGpp in the regulation of genes involved in virulence compared with the rest of the genome. This is the first demonstration that this highly conserved regulatory system is involved in bacterial virulence gene expression on a global scale.

Salmonella enterica serovar Typhimurium (Salmonella typhimurium) is a major pathogen of animals and man, causing gastroenteritis characterized by inflammatory diarrhea in humans. S. enterica serovars are responsible for up to 120 mil-lion cases of gastroenteritis each year in both industrial and developing nations (1). In the United States and Great Britain, Salmonella causes more human deaths than any other foodborne pathogen and may be responsible for an even higher mortality rate than previously realized (2).
As with a number of facultative intracellular pathogens, the course of infection and persistence of Salmonella in the host requires virulence gene products often found clustered within pathogenicity islands on the chromosome or on virulence plasmids (3,4). These virulence factors allow Salmonella to invade and colonize host cells, to surmount antimicrobial and physical barriers during infection, and to manipulate and modify host cell activities (5). Currently, five pathogenicity islands have been defined in S. typhimurium (6). Salmonella pathogenicity islands 1 and 2 (SPI1 and SPI2) 3 encode type III secretion systems required for invasion and replication within host cells, respectively (5)(6)(7)(8). SPI3 encodes a high affinity magnesium transport system required for macrophage survival and full virulence in mice (9). The role of SPI4 in Salmonella pathogenicity is not as well characterized, but it encodes products involved in the host specificity of S. typhimurium infections (10). One SPI5-encoded effector protein, SopB, activates the host cell Akt serine-threonine kinase, involved in the regulation of cell proliferation and survival of epithelial cells (11). S. typhimurium also harbors a 90-kb virulence plasmid containing the spv operon, which is required for full virulence in the murine typhoid model (12).
Guanosine tetraphosphate (ppGpp) is the mediator of a highly conserved regulatory system involved in control of global changes in gene expression patterns for adaptation to altered growth conditions. The accumulation of ppGpp in Gram-negative bacteria depends on the products of two genes, relA and spoT. The ribosome-associated RelA protein is responsible for the high level ppGpp synthesis during amino acid starvation (stringent response; reviewed in Ref. 13). In contrast to RelA, the SpoT protein is a cytoplasmic enzyme with both ppGpp synthetic and degradative functions (14,15). The SpoT-derived, basal ppGpp levels accumulate in response to stress and nutrient limitations other than amino acid starvation. In Gram-positive bacteria, a single RelA/SpoT protein is responsible for both functions (16).
There is growing evidence that ppGpp plays an important role in the virulence of pathogenic bacteria, including Mycobacterium tuberculosis (17), Listeria monocytogenes, (18) Legionella pneumophila (19,20), Vibrio cholerae (21), and Pseudomonas aeruginosa (22). A recent study found that an S. typhimurium ⌬relA⌬spoT strain was severely attenuated in susceptible BALB/c mice, was effectively noninvasive for epithelial cells in vitro, and played a crucial role in the regulation of SPI1 and the spv virulence plasmid genes in S. typhimurium (23,24). These results suggested that ppGpp may play a larger role in virulence gene expression in Salmonella.
In the present study we show that ppGpp is required for the expression of nearly all known Salmonella virulence genes in response to growth conditions relevant to host infection. The mechanism for the alternate regulation of invasion genes and genes required for intracellular survival and proliferation of Salmonella within the host has eluded discovery for many years. The results of this study show that ppGpp is required for regulation of both the extracellular (invasion) and intracellular virulence gene programs. One of the environmental signals involved in activation of SPI1 (invasion) gene expression is low oxygen levels (25)(26)(27). In contrast, the environmental signals involved in activation of SPI2 gene expression include acidic pH, phosphate, and divalent cation limitations, conditions thought to resemble the phagosomal milieu (28 -30). Here, we show that the low oxygen activation of SPI1 requires ppGpp. Surprisingly, we found that SPI2 genes also show a growth phasedependent activation in the wild type strain under aerated growth conditions and are repressed under low oxygen conditions. However, SPI2 gene expression was severely reduced under all of the conditions in the absence of ppGpp. The results suggest that elevated oxygen tension may be an additional signal involved in a ppGpp-dependent activation of SPI2 gene expression. This novel regulatory role of ppGpp in virulence gene expression is shown to be largely independent of changes in the transcription of known global regulators. Because ppGpp is part of a highly conserved regulatory system for mediating the growth response to environmental conditions, this mechanism may represent a common strategy whereby facultative intracellular pathogens regulate the virulence gene programs required for invasion and later survival and persistence within host cells to match the capacity for growth.

EXPERIMENTAL PROCEDURES
Bacterial Strains and Growth Conditions-Strains and plasmids used in this study are listed in supplemental Table S4. The wild type, virulent S. typhimurium strain SL1344 (31) was provided by F. Norel (Institut Pasteur) and reisolated from the spleens of infected Balb/c mice. The SL1344 ⌬relA71::kan spoT ϩ and ⌬relA71::kan ⌬spoT281::cat strains have previously been described (32). SL1344 harboring a complete deletion of SPI1 was constructed by P22 transduction of the ⌬SPI1::kan allele from strain SD11 (33). The ⌬SPI2 deletion strain, KT3736, was constructed in the same manner, but using the ⌬SPI2::aph(kan) strain MvP371 (34) as the donor for the P22 lysates, followed by elimination of the kanamycin resistance cassette induced by plasmid pCP20 (35), yielding strain KT3824. The ⌬SPI1::kan ⌬SPI2 double mutant, KT3930, was constructed by transduction of the ⌬SPI1::kan allele into KT3824. The SL1344 phoP60::Tn10d mutant (strain KT3980) was constructed by P22 transduction of strain SL1344 using strain TT13215 (J. R. Roth) as the donor for the lysates.
Bacterial cultures were cultivated in Lennox broth (36) at 37°C. The precultures were inoculated from Ϫ70°C frozen stock cultures into L-broth (containing 0.085 M NaCl) and grown to mid-log phase (A 600 of ϳ0.5) for inoculation of experimental cultures. Oxygen-limited cultures were grown in filled, stoppered 15-ml Bijou bottles. Aerobic cultures were grown in 50 ml of L-broth in 500-ml flasks with shaking. The media were supplemented with carbenicillin (100 g/ml), kanamycin (50 g/ml), chloramphenicol (20 g/ml), or tetracycline (20 g/ml) for selection of strains during construction. The antibiotics were omitted from experimental cultures.
Nucleic Acid Extraction and Purification-Bacterial culture samples (5 or 10 ml) were added to one-fifth of the sample volume of a solution of 5% (v/v) phenol in ethanol maintained on ice to stabilize total RNA. Total RNA was extracted and purified as described (37). Chromosomal DNA was removed by digestion with 50 -100 units of RNase-free DNase (Roche Applied Science) and re-extracted with phenol and chloroform to obtain purified total RNA. The absence of contaminating DNA was verified by PCRs involving primers targeting bacterial housekeeping genes.
Microarray Construction-The S. typhimurium microarrays consisted of 4686 protein coding regions or open reading frames (CDS) derived from the complete S. typhimurium LT2 genome sequence (38). The construction of the S. typhimurium microarrays has been described previously (39 -41).
Template Labeling and Hybridization-RNA for microarray analysis was reverse transcribed into cDNA and labeled according to protocols described on the IFR microarray website (www.ifr.bbsrc.ac.uk/safety/microarrays/). The hybridizations were performed as indirect comparison (type II) experiments using genomic DNA as the common reference and internal hybridization efficiency control (42).
Data Acquisition and Analysis-Fluorescence intensities of scanned microarrays were quantified using GenePix Pro software, version 3.0 (Axon Instruments, Inc.). The data were filtered and spots showing a reference signal lower than background plus 2 standard deviations were discarded. Unequal dye incorporation was compensated by median centering. The significance of the data at p ϭ 0.05 was determined using a parametric-based statistical test adjusting the individual p value with the Benjamini and Hochberg false discovery rate multiple test correction (43). All of the expression data for genes discussed in the text have passed this filter and are therefore statistically significant. These tests are features of the Gene-Spring TM 6.2 (Silicon Genetics) microarray analysis software package, which was used for both data visualization and analysis. All of the microarray data used in this study has been made available at NCBI (GEO accession number GSE4631).
Cell Culture and Invasion Assays-LoVo (ATCC CCL-229) intestinal epithelial and J774A.1 (ATCC TIB-67) murine macrophage-like cell lines were grown and maintained in Dulbecco's modified Eagle's medium/Ham's F-12 salts medium or Iscove's modified Dulbecco's modified Eagle's medium (Biochrom), respectively, supplemented with 10% fetal calf serum at 37°C and 5% CO 2 . The bacteria were grown with aeration to an A 600 of between 2 and 3 (late log phase), collected by centrifugation, resuspended in cell culture medium, and diluted to the appropriate concentrations for infection. Duplicate wells of monolayers of cells grown in 12-well culture plates (2 ϫ 10 5 cells/well) were infected to yield an multiplicity of infection of Յ 1/host cell except for the ⌬relA⌬spoT and ⌬SPI1⌬SPI2 strains (multiplicity of infection of 10). After a centrifugation step (5 min, 250 ϫ g), infected cells were incubated for 30 min (macrophage) or 60 min (epithelial cells) prior to a change of medium containing 50 g/ml of gentamycin. The medium was again replaced after 30 or 60 min, respectively, with medium containing 10 g/ml of gentamycin for the remainder of the experiment. At the times indicated in the figures, the wells were washed twice with phosphate-buffered saline and lysed by addition of 0.1% Triton X-100 in distilled water. Serial dilutions of lysates were plated for determination of intracellular bacteria. The data shown are representative of at least three independent assays for all strains.

A ppGpp-dependent Switch between Expression of Virulence Genes Involved in Invasion and Intracellular
Growth-In a previous study, a ⌬relA⌬spoT strain of S. typhimurium showed severe reductions in the expression of genes involved in invasion of host cells and an avirulent phenotype in susceptible mice (23). However, because noninvasive strains of Salmonella are also capable of causing infection (44,45), it was likely that the expression of additional virulence factors were affected in this mutant. We therefore used microarray gene expression profiling to assess the full effects of ppGpp on the expression of S. typhimurium virulence genes as well as global gene expression patterns. Because it was not known which of the environmental signaling pathways involved in virulence gene expression were affected by the absence of ppGpp, we compared the expression profiles obtained from wild type and ⌬relA⌬spoT strains grown under four different growth conditions: low oxygen cultures grown in L-broth containing 0.3 M NaCl, previously reported to activate SPI1 invasion genes (25,26), and the separated growth conditions of either L-broth with low oxygen, or aerated L-broth, with or without 0.3 M NaCl. The role of ppGpp on pathogenicity island gene expression was first placed into context by comparing transcript levels of the wild type strain for pathogenicity island genes encoded within SPI1 through SPI5 (Fig. 1). Genes encoded within SPI1 and SPI5 were expressed under all of the growth conditions examined. Additionally, SPI4 was found to be activated under aerated L-broth growth conditions (Fig. 1A). SPI3 signal intensities were low under all growth conditions. Surprisingly, SPI2encoded genes showed significant levels of transcription under aerated growth conditions, irrespective of the salt concentration in the medium (Fig. 1, A and C). These observations were unexpected, because SPI2 genes are not thought to be activated under these growth conditions (28,29). Under low oxygen con-  Table S2.

ppGpp and the Regulation of Gene Expression
ditions, the wild type strain showed continued expression of SPI1-encoded invasion genes as expected (Fig. 1, B and D) but a near complete repression of SPI2-encoded genes.
The transcriptional profiles of the ppGpp-deficient ⌬relA⌬spoT strain was compared with the wild type strain to determine the role of ppGpp in virulence gene expression under the same four environmental conditions. The ⌬relA⌬spoT mutant showed reduced transcript levels of SPI1encoded (invasion) genes under aerated growth conditions ( Fig.  1, B and F) but essentially a complete loss of invasion gene expression under oxygen-limited growth conditions (Fig. 1, D and H). In the high salt, low oxygen (SPI1-activating) growth conditions, genes encoding the major transcriptional activators of SPI1 genes, hilACD and invF, showed 2-6-fold reduced expression in the ⌬relA⌬spoT strain, explaining the loss of invasion gene expression (23). These results indicated that the severe reductions in SPI1 gene expression in the ⌬relA⌬spoT strain were independent of the salt concentration in the medium; rather, the response to low oxygen levels appeared to have been lost. This was of particular interest because no regulatory factor has been identified to explain the activation of hilA expression under low oxygen conditions (26,46). Large reductions in the expression of the SPI5 effector protein genes sopE2, pipC, and sopB were also observed (Fig. 2D). The secretion of these proteins is dependent on the SPI1 type III secretion apparatus (46).
Although the low oxygen growth conditions mainly affected the expression of SPI1-encoded invasion genes in the ⌬relA⌬spoT strain (Fig 2, B and D), under oxygenated growth conditions loss of ppGpp had the greatest effect on virulence genes involved in intracellular growth and/or survival (Fig 2, A  and C). We observed up to 10-fold reduced expression of SPI2encoded genes and reduced expression of other virulence-related genes (pipB, pipB2, sopE2, sifA, sopD, and the PhoPQ-regulated pagCK) involved in phagosome biogenesis or intracellular survival (Refs. 8 and 47; supplemental Table S2). Whereas the wild type strain showed activation of SPI2encoded genes in the presence of oxygen, the ⌬relA⌬spoT strain showed no expression of SPI2 genes under any of the growth conditions tested. These observations indicated that ppGpp was required for the response to environmental signals involved in the regulation of both SPI1 and SPI2 pathogenicity islands. Under low oxygen growth conditions, SPI1 gene expression was strongly ppGpp-dependent, whereas under aerated conditions SPI1 genes showed only reduced expression, but there was no activation of SPI2 gene expression as seen in the wild type. Therefore, ppGpp wasrequiredforboththelowoxygendependent activation of SPI1 genes and the oxygen-dependent activation of SPI2 genes. However, because the growth rates for the wild type strain were different under these two conditions, the involvement of other previously reported factors or components of the media in the ppGpp-dependent effect on SPI2 activation cannot be excluded. Nevertheless, the results indicated that the expression of both SPI1-and SPI2-encoded genes were dependent on ppGpp. One curious exception to the general reduction in virulence gene expression was the SPI3-encoded slsA gene, which has been annotated only as a conserved, hypothetical inner membrane protein. Under aerated growth conditions, slsA showed up to 2-5-fold higher levels of expression in the ⌬relA⌬spoT strain (Figs. 1, B and F, and 2, A and C, and supplemental Table S2).
In contrast to virulence-related genes, no major reductions in the expression of genes encoding components of the transcriptional and translational apparatus were observed in the ⌬relA⌬spoT strain (supplemental Fig S5 and supplemental Table S5). There were also no significant differences in the growth rates under all four conditions examined (data not shown). These results indicated that the severe ppGpp-dependent reductions in SPI1 and SPI5 gene expression were not the result of a general repression of gene expression.
The Effects on Regulatory Gene Expression in the ⌬relA⌬spoT Strain Is Limited-One goal of this study was to determine whether the ppGpp-dependent loss of virulence gene expression reflected altered levels of known or previously unidentified regulators of virulence gene expression. As noted above, regulatory genes that showed reduced expression in the ⌬relA⌬spoT strain under low oxygen conditions included the major SPI1 and SPI2 pathogenicity island regulators hilACD, invF, and ssrA/B (Fig. 3, B and D). In addition to pathogenicity island-encoded regulatory genes, many other global regulators  Table S2. OCTOBER 6, 2006 • VOLUME 281 • NUMBER 40 play significant roles in maintaining the virulence of Salmonella (reviewed in Refs. 26, 46, 49, and 50). A comparison of the expression patterns of known regulatory genes revealed only four global regulatory genes with significant ppGpp-dependent reductions in expression under the low oxygen conditions (Fig.  3, B and D, and supplemental Table S3). Of these four global regulators (Dps, Lrp, and RtsA/B), only RtsA/B have been shown to be involved in hilA gene expression (51,52), and both genes showed 3-5-fold reductions in expression in the absence of ppGpp under low oxygen conditions (Fig. 3, B and D).

ppGpp and the Regulation of Gene Expression
To determine whether the loss of expression of dps, lrp, or rtsAB contributed to reduced invasion gene expression in the ⌬relA⌬spoT strain, we examined the effects of both mutations and overexpression of Dps, Lrp, and RtsA/B on the expression of a hilA-lacZ fusion in the wild type and ⌬relA⌬spoT strains. As shown in Fig. 4, transcriptional fusions with rtsA and rtsB verified a strong ppGpp dependence for expression, consistent with the microarray data. Overexpression of RtsA or RtsA/B only partially compensated for the loss of hilA gene expression in the ⌬relA⌬spoT strain (Fig. 5), presumably because of the reduced expression of hilA gene coactivators, HilC and HilD (52). These results identified RtsA and/or RtsA/B as the first, non-SPI1-encoded regulators whose expression could be directly linked to loss or severe reductions in hilA and SPI1 gene expression in the ⌬relA⌬spoT strain.
In contrast, neither the overexpression of Dps or Lrp, nor the introduction of mutations in dps or lrp showed any effect on the loss of hilA expression in the ⌬relA⌬spoT strain, indicating that the reduced expression of Dps and Lrp was not involved in the loss of SPI1 gene expression. Indeed, overexpression of these regulatory genes resulted in reduced hilA expression in the wild type strain, suggesting that both can act as repressors of hilA expression (supplemental Figs. S1 and S2). The SirA/BarA two-component system has also been found to play a role in the activation of hilA expression (53). Although the microarray data indicated no significant effects on the expression of the sirA(uvrY)/barA genes in the ⌬relA⌬spoT strain under any of the growth conditions, we also examined the effects of SirA overexpression. Despite an up to 2-fold activation of hilA expression in the wild type strain, the overexpression of SirA did not compensate for reduced hilA expression in the ⌬relA⌬spoT strain, suggesting that SirA levels were not limiting (supplemental Fig. S3).
It remained possible that the overexpression of a number of global regulatory genes in the ⌬relA⌬spoT strain might also have been responsible for the loss of SPI1 gene expression (Fig.  3). The two-component PhoP/Q system functions as a negative regulator of hilA and invasion gene expression (reviewed in Ref. 46), and phoP gene expression showed significant elevations under the SPI1-activating growth conditions in the ⌬relA⌬spoT strain (Fig. 3D). However, a previous study had shown that the elevated phoP expression in the ⌬relA⌬spoT strain did not contribute to loss of hilA expression (23), observations that we verified (data not shown). Genes encoding the global regulators fis and fur also showed elevated expression levels in the ⌬relA⌬spoT strain under low oxygen growth conditions (Fig. 3, B and D). Whereas Fis has been found to activate hilA expression (26), both Fis and Fur are also known to have repressor activity. The introduction of mutations in either of these genes reduced hilA expression in both the wild type and ⌬relA⌬spoT strains, indicating that the loss of hilA expression in the ⌬relA⌬spoT strain was not due to repression by these regulators (supplemental Fig. S4). The loss of hilA and invasion gene expression in the ⌬relA⌬spoT strain therefore appeared to

ppGpp and the Regulation of Gene Expression
result from the concomitant loss of expression of both SPI1encoded (hilC, hilD, and invF; Refs. 23 and 24; and this study) and the non-SPI1-encoded regulators, RtsA/B (Figs. 3 and 4).
The expression of SPI2-encoded genes, involved in establishing intracellular growth, is also dependent upon a number of regulators. Encoded within SPI2, SsrA/B are the key regulators for the majority of SPI2-encoded genes (7,54,55), and ssrA/B expression is in turn dependent upon the OmpR/EnvZ twocomponent regulators (30,56) and SlyA, a global transcriptional regulator affecting a number of functions including flagellar biosynthesis (57). To determine whether overexpression of SsrA/B would compensate for the reduced SPI2 expression observed in the microarray study for the ⌬relA⌬spoT strain (Figs. 1, B and F, and 2, A and C), we examined the effects of SsrA/B overexpression on a chromosomally encoded SPI2 ssaG promoter-GFP transcriptional fusion in the wild type and ⌬relA⌬spoT strains (Fig. 6). The expression from the ssaG promoter was low in both strains compared with a control, ribosomal protein gene promoter (Fig. 6A). However, there was an apparent growth phase-dependent activation late in the growth phase in the wild type as previously reported (58) that was at least 2-fold reduced in the ⌬relA⌬spoT strain (Fig. 6B). The reduced expression of ssaG in the ⌬relA⌬spoT strain was fully compensated by the overexpression of plasmid-encoded SsrA/B (Fig. 6C). The high background level of expression of ssaG in both the wild type and ⌬relA⌬spoT strains was likely due to incomplete repression of the arabinose-inducible promoter. Nevertheless, these results supported the suggestion that the reduced levels of SPI2-encoded gene expression observed in the microarray analysis for the ⌬relA⌬spoT strain was likely due to the parallel loss or reduction in ssrAB expression.
Because ssrA/B expression is dependent on OmpR, we also determined the expression levels of ompR as well as a nonviru-   OCTOBER 6, 2006 • VOLUME 281 • NUMBER 40 lence related OmpR-dependent gene, ompC, in both the wild type and ⌬relA⌬spoT strains. Under normal, aerated growth conditions where the levels of expression of SPI2-encoded genes were highest in the wild type strain (Fig. 1A), the levels of ompR expression in the ⌬relA⌬spoT strain were similar to that of the wild type (supplemental Fig. S5). Likewise, the level of expression of ompC was essentially identical in both strains, indicating that OmpR was both expressed and active under these growth conditions. These results indicated that the reductions observed for ssrA/B and SPI2 gene expression in the ⌬relA⌬spoT strain were not the result of loss of either the expression or activities of OmpR. Interestingly, under the high salt, low oxygen growth conditions, where SPI2 gene expression was repressed in the wild type strain (Fig. 1G), both ompR expression and ompC expression were activated in the wild type strain (supplemental Fig. S5, B and D). In contrast, the expression levels in the ⌬relA⌬spoT strain remained the same regardless of the growth conditions.

ppGpp and the Regulation of Gene Expression
ppGpp Is Required for Intracellular Growth-A number of Salmonella virulence genes in addition to those encoded by SPI2 are required for intracellular proliferation within macrophage and other cell types. To determine whether the expression of genes involved in this aspect of pathogenesis were also affected in the ⌬relA⌬spoT strain, we compared the intracellular growth phenotypes of the wild type and ⌬relA⌬spoT strains in both intestinal epithelial and murine macrophage cell lines. Consistent with the reduced levels of SPI1-encoded invasion genes, at least 10-fold higher levels of infecting ⌬relA⌬spoT bacteria were required to achieve initial uptake levels similar to those of the wild type strain in both cell types. As shown in Fig. 7 (A and B), the intracellular growth phenotype of the ⌬relA⌬spoT strain varied with cell type. In epithelial cells, the mutant showed only a growth defect, with no increase in intracellular colony forming units for up to 24 h (Fig. 7A), whereas the wild type strain showed a 20-fold increase in the same time period.
We also compared the intracellular growth kinetics with a number of known mutants affecting proliferation and survival within macrophage. As shown in Fig. 7 (C and D), strains harboring mutations or complete deletions of phoP or SPI2, respectively, showed phenotypes distinct from that of the ⌬relA⌬spoT strain. Whereas the former strains showed an intracellular growth defect, i.e. no increase in intracellular bacteria with time, the ⌬relA⌬spoT strain showed a rapid loss in viable, colony-forming units, indicating that both intracellular growth as well as survival mechanisms were affected (Fig. 7B). These results indicated that the loss of ppGpp affected mechanisms involved in proliferation and intracellular survival in addition to those dependent upon either the PhoP regulon or the SPI2 type III secretion system.

DISCUSSION
Virulence gene regulation in Salmonella is complex, and overlapping regulatory networks have evolved to enable Salmonella to adapt and survive in both the intracellular and intestinal environments (8,28,30,46,49,50). Previous studies have shown that subsets of virulence genes are expressed in response to various extra-and intracellular signals. The current study shows that ppGpp is required for the expression of these genes and most of the other known virulence-related genes in Salmonella. In addition, the reduced expression of virulence genes occurs in the absence of major effects on the expression of other, known regulatory genes. Indeed, for the two growth conditions of aerated L-broth and high salt, low oxygen, 56 and 37% of the total virulence genes, respectively, showed ppGpp-dependent reductions of greater than 2-fold. This was in marked contrast to the remainder of the genome, which showed 2-fold reduced expression for only 8 and 10% of the total genes under the same growth conditions (supplemental Table S1). The results of this study therefore define ppGpp as the first regulatory effector found to play a comprehensive role in the expression of virulence genes in Salmonella.
The results show that ppGpp is required to mediate the environmental signals involved in the activation of either the invasive or intracellular virulence gene programs in Salmonella (Figs. 1 and 2) and identify the low oxygen response for SPI1 activation as a key ppGpp-dependent pathway. The data also identify a possible mediator for the ppGpp-dependent regulatory pathway involved in SPI1 expression under low oxygen conditions. To date, no regulatory factor has been identified to explain the activation of hilA expression under low oxygen conditions (25,26,46). Under low oxygen conditions, the large reductions in SPI1 gene expression correlated with up to a 10-fold loss in expression of the SPI1-encoded regulators hilA and hilC,D (Fig. 3, B and D). Prior studies showed that RtsA/B was capable of increasing the expression of hilA, hilC, hilD, and invF as well as other virulence genes dependent upon these regulators (51,52), results also verified in this study (Figs. 4 and 5). RtsA/B are the first non-SPI1-encoded regulators shown to be directly involved in hilA and SPI1 gene expression in a ppGppdependent manner. The observation of a strong ppGpp dependence for the expression of the rtsA/B genes under low oxygen growth conditions may therefore provide the missing, additional components explaining the loss of SPI1 (invasion) gene expression in the ⌬relA⌬spoT strain.
It was unexpected that significant levels of transcription of SPI2 genes were observed in the wild type strain under aerated growth conditions (Fig. 1A), because the environmental signals involved in regulation of this pathogenicity island are not assumed to be present in these growth media. Acidic pH, limiting divalent metal ions, and phosphate limitation have all been reported as signals involved in SPI2 gene activation (28,29,59). The signals involved in the observed activation of SPI2 gene expression in the wild type strain in aerated, L-broth cultures is not clear; however, the activation appears to require oxygen and is clearly ppGpp-dependent (Figs. 1 and 2). SPI2-encoded genes are activated intracellularly in macrophages, an apparently oxygen-rich environment (40,54). One of the ppGpp-dependent components in the signaling pathway leading to SPI2 activation was found to be SsrB, the response regulator for SPI2 genes. The loss of activation of SPI2 expression in the ⌬relA⌬spoT strain under aerated growth conditions correlated with a 2-fold reduction in ssrB gene expression (Fig. 3), and SsrA/B overexpression was able to compensate for reduced levels of ssaG (Fig. 6C). Because the expression levels and activities of the ompR gene product did not appear to be affected under these conditions ( Fig. 3 and supplemental Fig. S5), it remains unclear whether the reductions in ssrB expression represent a direct or indirect result of loss of ppGpp. The ⌬relA⌬spoT strain also showed extensive defects in macrophage survival and proliferation in host cells, as indicated by the distinct intracellular phenotype of the ⌬relA⌬spoT strain compared with other known regulatory and virulence mutants ( Fig. 7 and supplemental Fig. S6). These observations support a major role for ppGpp in regulation of the intracellular gene expression program in addition to its role in invasion gene expression.
A recent study showed that SPI2 genes are activated extracellularly prior to invasion of the intestinal epithelium, indicating that at least some of the signals involved in activation of intracellular virulence genes are also present in the intestinal lumen (60). Based on the observed loss of SPI1 activation under oxygen-limiting conditions, and the lack of activation of SPI2 genes under aerated growth conditions in the ⌬relA⌬spoT strain, we speculate that oxygen tension may serve as an additional signal mediating the switch between the SPI1 and SPI2 gene expression programs. Activation of SPI2-encoded genes outside of the intracellular environment may result from proximity to the intestinal wall, where diffusion of oxygen from underlying tissues may be sensed by Salmonella. Invasive Salmonella in close contact with the intestinal epithelia would therefore activate genes required for intracellular growth and survival prior to or at the same time as invasion. This might also explain why in vitro growth conditions have generally not been able to show simultaneous activation of both of these pathogenicity islands. As noted previously (26,46), it is likely that the role of oxygen in the regulation of SPI1, and possibly SPI2, is indirect, a result of changes in bacterial metabolism and global regulatory responses. The observation of a growth phase (cell density) dependence for both SPI1 and SPI2 gene expression would be consistent with this idea.
Apart from reduced ssrA/B and rtsA/B gene expression, we observed no consistent ppGpp-dependent differences in the expression of other, known regulatory genes that would explain the observed loss of global virulence gene expression. Indeed, for a number of regulatory systems, we noted large reductions in virulence-related gene expression but little or no effects on housekeeping genes controlled by the same regulators. Supportive of this observation is a recent study that examined the immunogenicity of the ⌬relA⌬spoT strain in mice (61). Despite the avirulent phenotype of the mutant in mice, immunization with the strain was able to elicit both systemic and mucosal antibody responses against a subsequent challenge with the wild type strain. This observation also indicates that the majority of all other epitopes and bacterial components were expressed and available for antigen presentation as part of an immune response. The observation that deletion of either or both of the major pathogenicity islands, a gene involved in establishment and maintenance of the phagosome (sifA) or a regulator of genes involved in bacterial defense mechanisms (phoP), were unable to reproduce the reduced survival and/or intracellular growth of the ⌬relA⌬spoT strain ( Fig. 7 and supplemental Fig. S6), supports a global role for ppGpp in the regulation of both extra-and intracellular virulence gene expression.
The results of this study are consistent with a model wherein the horizontal acquisition of virulence determinants by facultative, intracellular pathogens is brought under control of the same regulatory system involved in growth rate control and regulation of global gene expression in response to changing growth conditions. As noted in a previous study (23), the simplest explanation for the strong ppGpp dependence for invasion gene expression might be related to the higher AT content of these acquired genes and islands relative to the remainder of the genome. AT-rich promoters have been found to show a ppGpp dependence for expression (62). Consistent with this idea, a whole genome comparison of gene expression levels versus GC content indicates a skew toward reduced expression of genes with high AT content in the ⌬relA⌬spoT strain. Although the AT content of coding regions may not necessarily reflect the base composition of promoter regions, for the few virulence gene promoters that have been mapped, the correlation appears to hold true for the promoters as well. 4 This model would also explain the apparently inconsistent expression patterns of genes dependent upon the various regulators. Because the major function of ppGpp is the reallocation or partitioning of limiting RNA polymerase among promoters in the cell (63), the specificity for the expression of a given gene will also depend on the presence of accessory proteins and regulators involved in RNA polymerase recruitment for activation of promoters in the various regulons. However, in a ⌬relA⌬spoT strain, promoters that require either ppGpp-bound RNA polymerase for transcription or unusually high local concentrations of RNA polymerase would still be poorly expressed despite the presence of accessory proteins, because of insufficient free RNA polymerase available for mRNA synthesis.
Attention has recently been focused on an additional RNA polymerase-associated protein involved in the regulation of gene expression by ppGpp, DksA (64). DksA has been described to enhance both the inhibitory (rRNA promoters) and activating (e.g. amino acid gene promoters) regulatory effects of ppGpp (48). The absence of significant changes in dksA expression under any of the growth conditions examined here (Table  S1) is consistent with observations in Escherichia coli and further suggests that the effects observed here for virulence gene expression are due to the loss of ppGpp, although DksA likely plays a co-regulatory role in the wild type strain.
The lack of significant effects on the expression of housekeeping genes coupled with large reductions in virulence-related genes dependent upon the same regulators may therefore simply reflect the high AT content of the promoters of the latter, horizontally acquired gene sets. We suggest that promoters for housekeeping genes may have been optimized during evolution for a certain degree of ppGpp independence to assure basal levels of expression under all growth conditions, whereas recently acquired virulence gene promoters remain strongly dependent on ppGpp.