Regulation of Iron Storage by CsrA Supports Exponential Growth of Escherichia coli

Iron is an essential micronutrient for nearly all living organisms but is toxic in excess. Consequently, the maintenance of iron homeostasis is a critical biological process, and the genes involved in this function are tightly regulated. Here, we explored a new role for the bacterial RNA binding protein CsrA in the regulation of iron homeostasis. CsrA was shown to be a key regulator of iron storage genes in Escherichia coli, with consequential effects on cellular iron levels and growth. Our findings establish a model in which robust CsrA activity during the exponential phase of growth leads to repression of genes whose products sequester iron or divert it to unnecessary stress response processes. In so doing, CsrA supports E. coli growth under iron-limiting laboratory conditions and may promote fitness in the competitive iron-limited environment of the host large intestine.

The key component of the Csr system is CsrA, a homodimeric, sequence-specific RNA-binding protein, encoded by csrA (13)(14)(15). In general, CsrA represses the expression of genes associated with stress responses and stationary-phase growth (4,16), while activating the expression of genes that support exponential growth (16,17). CsrA binds to mRNA targets at sites containing a GGA motif surrounded by semiconserved sequences, with the GGA often located in the single-stranded loop of a hairpin (18,19). The binding of CsrA to sites located in the 5= untranslated region (5= UTR) or the early coding region can result in altered translation (20)(21)(22)(23), altered RNA stability (8,24), and/or changes in RNA secondary structure that affect Rho-dependent transcription termination (25).
CsrA itself is subject to multifaceted regulation (16). While csrA gene expression is regulated transcriptionally and posttranscriptionally (26), CsrA activity is controlled by the small RNAs (sRNAs) CsrB and CsrC. These sRNAs contain multiple high-affinity binding sites that mimic those of target mRNAs, allowing CsrB/C to bind to and sequester many CsrA dimers (27). CsrB and CsrC transcription is activated by amino acid limitation or other stresses via ppGpp/DksA (4), metabolic end products, e.g., acetate and formate, via BarA-UvrY (12,28), and extracytoplasmic stress via E (6). Their RNase E-mediated turnover is triggered by glucose, via interaction of EIIA glc with CsrD (29)(30)(31). Thus, CsrB/C sRNAs accumulate and sequester CsrA when preferred carbon resources have been expended, amino acids are limiting, metabolic end products have accumulated, and/or cells experience extracellular stress. Antagonism of CsrA by CsrB/C promotes the transition from glycolytic metabolism and active growth to gluconeogenesis, glycogen biosynthesis, and the formation of a stress-resistant phenotype (30).
A recent transcriptomics analysis in E. coli revealed that CsrA affects the expression of numerous genes involved in iron uptake and metabolism and confirmed its effects on levels of four such mRNAs (19). Several of the genes were of particular interest due to their importance in iron homeostasis and/or the large magnitude of CsrA effects (19). These effects of CsrA were suspected of occurring indirectly, as they were not confirmed by in vivo binding of CsrA to the mRNAs using cross-linking immunoprecipitation sequencing (CLIP-seq). However, CLIP-seq analyses have constraints, sources of bias (32), and are limited in sensitivity (33), potentially generating false-negative results.
Iron is an essential micronutrient for nearly all living organisms and required for fundamental processes such as respiration, central metabolism, genetic regulation, and DNA repair (34). It is the most common transition metal found in proteins, typically within heme or Fe-S prosthetic groups (35,36). Although iron is abundant in many environments, ferric iron (Fe 3ϩ ) predominates under aerobic conditions at neutral pH. Fe 3ϩ has poor aqueous solubility and exists as insoluble salts and oxides or bound to host iron-binding proteins (37). Soluble ferrous iron (Fe 2ϩ ) is biologically available but is rapidly oxidized to ferric iron at neutral or higher pH under aerobic conditions (35). Consequently, microbes have evolved high-affinity acquisition systems to scavenge iron (38). An added complication is that Fe 2ϩ can be cytotoxic due to its role in producing damaging hydroxyl radicals ( · OH) from H 2 O 2 via the Fenton reaction. Thus, free intracellular iron levels are tightly controlled (34,39). Additionally, iron availability can serve as an important cue regarding the environment of the bacterium, affecting virulence processes in many pathogens (38,40).
Intracellular iron storage proteins (ISPs) sequester iron, providing iron reserves and protection against toxicity (35). Three related classes of bacterial ISPs exist: archetypal ferritins, heme-containing bacterioferritins, and Dps proteins (DNA protection during starvation). Each of these proteins is composed of identical subunits that form a roughly spherical shell surrounding a central cavity that acts as an iron storage reservoir (34). Ferritins and bacterioferritins are composed of 24 subunits and can accommodate 2,000 to 3,000 iron ions; Dps proteins have 12 subunits and can store ϳ500 iron ions (34). A key feature of ISPs is the ferroxidase center, which binds two ferrous ions and oxidizes them using molecular oxygen or H 2 O 2 , forming a diferric intermediate that migrates to the central core for storage (35). Bacteria often possess multiple ferritin or bacterioferritin genes (34). E. coli expresses two ferritins (FtnA and FtnB), one bacterioferritin (Bfr), and one Dps.
Assembly of the iron-sulfur (Fe-S) clusters that serve as cofactors for enzymes is a crucial biological function, performed by Fe-S biogenesis systems (41,42). In E. coli, most Fe-S cluster formation under nonstress conditions is by the "housekeeping" Isc pathway, encoded by the iscRSUA-hscBA-fdx (isc) operon (42,43). Alternatively, the Suf pathway, encoded by the sufABCDSE (suf) operon, is more robust than Isc against oxidative damage and is favored for Fe-S assembly during oxidative and nitrosative stress or when iron is limiting (44)(45)(46)(47). The coordinated regulation of these pathways accommodates the changing requirements for Fe-S cluster biogenesis, which fluctuate with growth conditions (43,48).
In E. coli and most bacteria, the DNA-binding ferric uptake regulator protein, Fur, controls iron metabolism (49). Fur activity depends on cellular free iron, which acts as a corepressor of transcription (40). Fur also activates gene expression via RNA polymerase recruitment, blocking repressor access, or transcriptional repression of RyhB sRNA (36). RyhB base pairing stimulates the degradation of mRNAs encoding nonessential iron-dependent proteins, increasing iron availability for essential processes (50). The goal of the present study was to explore the effects of CsrA on E. coli iron homeostasis. Unlike Fur and RyhB, we show that CsrA is not responsive to iron availability but nevertheless exerts biologically important effects on iron homeostasis and growth by regulating the expression of ISPs.
we did not observe Fur-dependent regulation of the bfr=-=lacZ fusion. As the RyhB binding site in bfr mRNA has yet to be identified, it is conceivable that our fusion did not encompass that region. Finally, an ϳ3-fold increase in the csrA mutant was observed for dps=-=lacZ, which did not respond to Fur (Fig. 1C).
Moderate CsrA-dependent effects were observed for ftnA and sufA fusions ( Fig. 1D and E). The ftnA=-=lacZ showed positive effects of CsrA on its expression, which were retained in the Δfur mutant (Fig. 1D). Thus, CsrA has a minor role in activating ftnA gene expression that is mediated independently of Fur. The sufA=-=lacZ fusion exhibited modest repression via CsrA in the fur WT and mutant backgrounds (Fig. 1E). The entC, fhuE, fepA, fes, and fhuA translational fusions were all repressed by Fur but were not regulated by CsrA (Fig. S1A to E). Expression of fur=-=lacZ was unaffected by CsrA (Fig. S1F). Altogether, these findings demonstrate that CsrA effects on iron-related gene expression are mediated independently of Fur.
Growth phase-dependent regulation of ftnB, bfr, and dps expression by CsrA. Having identified three iron storage genes, ftnB, bfr and dps, that are repressed by CsrA, we assessed their expression during exponential phase, the transition from exponential to stationary phase, and stationary-phase growth at 24 h. Complementation tests were performed using plasmid-borne csrA (pCRA16) versus the empty vector (pBR322). The ftnB, dps, and bfr fusions all exhibited growth phase-dependent regulation by CsrA, which was abolished by complementation (Fig. 2). CsrA repressed the three genes maximally during exponential growth, with weaker effects in the transition to stationary phase and in stationary phase. The bfr and dps fusions were not regulated by CsrA during stationary phase ( Fig. 2B and C), while ftnB repression decreased from 14-fold in the exponential phase to ϳ4-fold at 24 h ( Fig. 2A). ftnB expression peaked during the transition to stationary phase and remained higher in the csrA mutant in all stages of growth ( Fig. 2A). bfr expression levels were lowest in exponential phase, higher at transition to stationary phase, and then decreased modestly in stationary phase Statistical significance was determined using unpaired t tests and denoted as follows: ***, P Ͻ 0.001; **, P Ͻ 0.002. (Fig. 2B). Finally, the growth-phase expression pattern of dps reflected what was already known; Dps levels are low during exponential growth and increase upon entry to stationary phase (Fig. 2C) (53). Thus, CsrA plays a particularly important role in limiting the expression of these three genes during exponential growth.
CsrA binds with high affinity and specificity to 5= UTRs of ftnB, dps, and bfr transcripts. Although a CLIP-seq assay did not identify in vivo CsrA binding, previous transcriptomics data demonstrated that CsrA affects the stability and abundance of ftnB and dps mRNA (19), suggestive of posttranscriptional regulation. To explore the possible binding of CsrA to ftnB, bfr, and dps transcripts, electrophoretic gel mobility shift assays (EMSA) were performed. CsrA exhibited high-affinity binding to all three RNAs ( Fig. 3B to D). A nonlinear least-squares analysis of the data yielded apparent K D  (26,54). The bfr and dps mRNAs displayed a single shift as CsrA concentration was raised to 100 nM, while two distinct shifted species and a faint minor form were observed for ftnB at 50 nM and higher, suggesting that more than one CsrA dimer may bind to this RNA. Examination of the ftnB sequence identified four potential CsrA binding sites, while the bfr and dps sequences contained three and two potential CsrA binding sites, respectively (Fig. 3A). Competition assays performed with specific (self) and nonspecific (phoB) unlabeled competitor transcripts confirmed that binding is specific in all cases ( Fig. 3B to D).
Taken together with reporter fusion data ( Fig. 1 and 2) and transcriptomics studies (19), the high-affinity binding of CsrA to ftnB mRNA suggests that CsrA may directly regulate expression of this gene by binding to its 5= UTR. CsrA also bound tightly to the 5= UTR of dps mRNA, although the interpretation of this interaction is less clear. While CsrA negatively affects dps RNA abundance and translation (as measured by changes in ribosome occupancy), it increases dps mRNA stability (19). It is likely that CsrA binding to the dps 5= UTR is responsible for one or more of these effects. Finally, the high-affinity binding of CsrA to the 5= UTR of the bfr transcript, along with increased mRNA abundance in the csrA mutant (19), suggests that CsrA may directly repress bfr expression by binding its 5= UTR.
CsrA directly represses translation of ftnB, dps, and bfr via their 5= UTRs. To assess whether the regulatory effects of CsrA on ftnB, bfr, and dps are mediated posttranscriptionally, without the requirement for an intermediate regulatory gene, we used the PURExpress system to measure lacZ expression from reporter fusions carried on plasmid templates. Reporter sequences contained a T7 promoter fused to the 5= UTR of each gene and a few codons of the coding region (Fig. 3A). In all cases, with the exception of the control (pnp=-=lacZ), expression was inhibited by CsrA. Thus, CsrA represses translation of ftnB, bfr, and dps via their 5= UTRs (Fig. 4A). The in vitro translational repression exhibited by the dps reporter was similar to that by the ftnB reporter. This pattern of regulation differs from what was seen in the in vivo translational fusion assays, where dps demonstrated substantially less CsrA-dependent repression than ftnB (Fig. 2). This difference may result from increased dps mRNA stability, which was observed previously in the csrA mutant strain (19), or it might involve indirect effects of CsrA on dps transcription in vivo.
To identify the precise CsrA interaction site(s) for the most strongly regulated gene, ftnB, RNA footprinting experiments were performed. Three regions were protected from RNase T1᎑mediated cleavage with increasing concentrations of CsrA (Fig. 4B). The protected nucleotides correspond to three of the four GGA motifs, indicating that CsrA has three authentic binding sites in the ftnB 5= UTR (Fig. 4C) (BS1, BS2, and BS3). BS3, which overlaps the ftnB Shine-Dalgarno (SD) sequence, exhibited the strongest protection by CsrA. This finding provides a molecular mechanism for ftnB translational repression by CsrA, observed in vitro (Fig. 4A). The multiple protected regions are consistent with EMSA results, which show that more than one complex is formed between CsrA and the ftnB 5= UTR (Fig. 3B).
Overexpression of iron storage genes in the csrA mutant increases cellular iron levels. To determine the effect of CsrA regulation of iron storage genes on cellular iron levels, exponential-phase cultures were analyzed for iron content by inductively coupled plasma optical emission spectrometry (ICP-OES). Cellular iron content was ϳ25% higher in the csrA mutant (5.2 Ϯ 0.15 mol Fe/g protein) than in the WT (4.2 Ϯ 0.07 mol Fe/g protein) and was restored to the WT level upon complementation (Fig. 5A). This result is consistent with the increased expression of ftnB, bfr, and dps in the csrA mutant, which should increase its capacity for iron storage. To test this hypothesis, deletions of the chromosomal ftnB, bfr, and dps genes were introduced, and cellular iron levels were measured. Iron levels in the csrA mutant approached WT levels when FtnB and/or Bfr production was abolished (Fig. 5B), suggesting that increased iron accumulation in the csrA mutant is caused by the overproduction of these ISPs. The dps mutation had no substantial effect on iron levels, alone or in combination with ftnB or bfr. Notably, the increased iron in the csrA mutant may not be biologically available. In fact, excessive iron sequestration by FtnB and Bfr in the csrA mutant might inhibit iron-dependent cellular processes, to the detriment of cell growth.
Growth under iron-limiting conditions is compromised by csrA mutation and restored by deletion of ftnB and bfr. To test the above hypothesis, we monitored the growth kinetics of WT, csrA, ftnB, bfr, and/or dps mutants under iron-replete and iron-limiting conditions in both rich and minimal media. Growth in rich medium was assessed by tracking total cell protein of cultures grown in LB (for iron-replete conditions) and LB containing 400 M 2,2=-dipyridyl (DIP) (for iron-limiting conditions). Growth in minimal medium was evaluated by measuring the optical density at 600 nm (OD 600 ) in MOPS (morpholinepropanesulfonic acid) medium supplemented with 100 M FeSO 4 (iron replete) or 0.05 M FeSO 4 (iron limiting) and 0.2% glucose (55). However, biofilm formation compromised growth measurements (A 600 ) in MOPS medium. Under this condition, biofilm was apparently dependent upon the formation of curli fimbriae and was abolished by deletion of csgA (data not shown). Thus, all strains examined in MOPS medium were constructed in a csgA background. Growth of the WT and csrA strains was inhibited under iron-limiting conditions in both media, and these growth defects were particularly severe in the csrA mutant ( Fig. 6; Fig. S2). To assess the contribution of each iron storage gene to the growth inhibition exhibited by the csrA mutant, deletions in the chromosomal ftnB, bfr, and dps genes were introduced into the WT and csrA mutant. Each strain was grown under iron-replete and iron-limiting conditions in both the rich ( Fig. 6B and C; Fig. S2B and C) and minimal ( Fig. 6D and E; Fig. S2D and E) media. The inactivation of ftnB, bfr, or dps  (Fig. S2B to D). However, introduction of these knockouts into the csrA mutant led to partial or complete restoration of the WT growth rate in MOPS medium with limiting iron (Fig. 6E). In rich medium, the csrA mutant exhibited reduced growth only when iron was limiting, and this growth defect was completely recovered upon inactivation of ftnB alone; dps or bfr alone did not restore growth ( Fig. 6B and C). Interestingly, the csrA mutant displayed a growth defect in the minimal medium under both iron conditions, although the effects were more pronounced in the iron-limited MOPS ( Fig. 6D and E). Inactivation of ftnB in the csrA mutant recovered the growth rate completely in both iron-limited and iron-replete minimal media. The introduction of a bfr mutation to the csrA mutant partially restored growth in iron-replete and iron-limited MOPS ( Fig. 6D and E). The dps mutation only improved growth of the csrA mutant in MOPS medium with limiting iron, and this effect was weak ( Fig. 6D and E).
Iron availability has negligible effects on CsrA/B/C levels. A fundamental feature of Csr systems is their ability to integrate a variety of nutritional cues and signals and to interact with other regulators in diverse circuitry (10,16,29). Therefore, we sought to determine whether iron availability affects levels of Csr system components. Identically growing exponential-phase cultures of WT E. coli were treated with or without DIP (250 M), and CsrA/B/C levels were measured. RyhB levels served as a positive control for the iron starvation response (Fig. 7A) (52). CsrA was determined by Western blotting using polyclonal anti-CsrA antibodies and normalized to RpoB (Fig. 7B). CsrA levels were unaffected by DIP addition, indicating that CsrA expression is not regulated in response to a change in iron availability. Northern blotting (Fig. 7C and D) showed that CsrB levels were unaffected by iron availability, although CsrC levels dropped slightly ϳ30 min after the addition of DIP. CsrB/C sRNAs function via CsrA, and CsrB is a more effective antagonist of CsrA than CsrC (56). Thus, the minor change in CsrC levels at 30 min might not have a significant impact on cellular physiology. In summary, it appears that iron availability has negligible effects on the key components of the Csr system and is not one of the cues to which the Csr system responds.

CsrA regulates resistance of E. coli against killing by H 2 O 2 .
The finding that CsrA represses sufA (Fig. 1E), which encodes a protein involved in Fe-S assembly under oxidative stress (43), suggested that CsrA may play a role in regulating resistance against oxidative damage. To test this hypothesis, we exposed several isogenic strains to H 2 O 2 during the exponential phase of growth and monitored its effect on viability. As suspected, the csrA mutant was considerably more resistant to oxidative stress than the WT strain (Fig. 8). Complementation restored sensitivity of the csrA mutant to H 2 O 2 , confirming a role for CsrA in repressing oxidative stress resistance. While deleting ftnB and/or bfr did not restore H 2 O 2 sensitivity to the csrA mutant, the loss of dps caused a marked increase in H 2 O 2 susceptibility, resulting in greater sensitivity even than in the csrA WT strain (Fig. 8B). Perhaps this is not surprising, given the importance of Dps in protection against oxidative stress (57). Finally, the WT and csrA mutant strains were tested for survival against H 2 O 2 during early stationary phase. Unexpectedly, the csrA mutant was considerably more sensitive than WT to H 2 O 2 at this stage of growth (see Fig. S3), opposite of its response in the exponential phase (Fig. 8). These observations are consistent with recent findings from Salmonella enterica that CsrA is a flexible regulator whose regulon varies under different growth or physiological conditions (58).

DISCUSSION
Here, we show that CsrA regulates at least 5 genes that participate in iron storage and utilization (Fig. 1). Repression of ftnB and bfr by CsrA affects cellular iron content (Fig. 5) and is required for optimal growth under iron limitation and in minimal medium (Fig. 6). In other words, repression of ftnB and bfr by CsrA positions the Csr system to guide intracellular iron flux toward growth-promoting processes. CsrA is known to broadly repress gene expression involved in stress responses and in the stationary phase of growth. Furthermore, the Csr system itself does not appear to respond to iron availability (Fig. 7). In contrast, CsrB and CsrC sRNAs are known to be expressed under conditions of stress or nutrient limitation, and by sequestering CsrA, these sRNAs diminish CsrA-mediated regulation of gene expression when growth is no longer a priority (4, 5, 30). This regulatory arrangement no doubt allows the cellular capacity for

Regulation of Iron Storage by CsrA
® iron storage to be governed in part by nutritional and growth phase conditions that influence the Csr system (29).
To our knowledge, the present studies represent the first evidence that the repression of gene expression by CsrA can be necessary for robust exponential growth, although modest growth support via CsrA-mediated repression has been observed (59). While the csrA::gm allele used in this study results in a much less severe growth defect than seen in a ΔcsrA strain, even when the csrA::gm strain was grown under limiting iron conditions, we posit that the findings obtained with the csrA::gm mutant are nevertheless important for the following reasons. (i) Most likely, there is no environmental condition under which CsrA activity is absent, but there may be conditions under which reduced CsrA activity impacts iron requirements. (ii) In considering the development of CsrA inhibitors (60), complete inhibition of CsrA activity may not be required to achieve useful therapeutic effects, because human and animal body compartments tend to contain little available iron (see related discussion below).
The high-affinity binding of CsrA to ftnB, dps, and bfr transcripts (Fig. 3), taken together with in vitro translation results (Fig. 4A), indicates that CsrA represses translation of ftnB, bfr, and dps mRNAs by binding to the 5= UTR of each transcript. Furthermore, RNase T1 footprinting revealed that CsrA binds to the ftnB SD sequence (Fig. 4B), which should directly impede ribosome loading. Translational repression of hfq, glgC, cstA, and pgaA expression is similarly mediated by CsrA binding to sites overlapping the SD sequences (21,54,61). Both bfr and dps transcripts contain potential CsrA binding sites overlapping their SD sequences (Fig. 3A), which might mediate translational repression by a similar mechanism (Fig. 4C).
The overexpression of ftnB and bfr in the csrA mutant caused excessive accumulation of cellular iron (Fig. 5). The csrA mutation also inhibited growth under iron-limiting conditions, apparently by altering iron bioavailability (Fig. 6). The relative magnitude of CsrA-dependent repression of genes for ISPs (ftnB Ͼ bfr Ͼ dps) mirrored the effects of these genes on cellular iron content and growth (ftnB Ͼ bfr Ͼ dps) (Fig. 1A to C, 5B, and 6C to E). We propose that during exponential growth, repression of the ISPs, particularly FtnB and Bfr, prevents the sequestration of iron away from iron-dependent growth processes. Importantly, the regulatory effects of CsrA on these genes became weaker or were eliminated as cultures entered the stationary phase of growth (Fig. 2), during which time E. coli becomes resistant to a variety of stressors (62).
The complex and dissimilar regulation of the E. coli ISPs seemingly complicates the interpretation of their individual physiological roles (34). ftnA expression is primarily regulated by Fur-mediated transcriptional activation, coupling FtnA levels to iron availability (63). At high iron levels, FtnA serves as the primary reservoir in E. coli, storing up to 50% of cellular iron (64). Under low to moderate iron levels, Bfr can become the principal ISP (65). Notably, bfr expression is induced by stresses, including osmotic and heat stress, while ftnA expression is not (66). Likewise, ftnB transcription can be driven by a E -dependent promoter (67), linking FtnB production to extracytoplasmic stress, or from a D -dependent promoter, as part of the Cpx regulon, connecting it to cell envelope damage (68). dps transcription is induced under oxidative stress by the transcriptional activator OxyR (69) and by S during the stationary phase of growth (70). This regulatory diversity implies that individual ferritins are beneficial under different physiological conditions. The strongest regulation by CsrA observed here was that of ftnB (Fig. 1A). FtnB is the major Fe 2ϩ donor for the repair of oxidatively damaged Fe-sulfur clusters in Salmonella, although this is yet to be confirmed in E. coli (71). These findings raise the possibility that a primary role of CsrA in iron homeostasis is related to oxidative stress and repair response. CsrA also repressed bfr and dps ( Fig. 1B and C), whose protein products appear to play modest roles in iron storage in E. coli and are linked to stress responses (64). CsrA repressed expression of sufA (Fig. 1E) and negatively affected the levels of sufABCDE operon transcripts (19), which encode proteins responsible for Fe-S cluster assembly under stress conditions. Recent research suggests that FtnB, Bfr, and Dps may each function to donate iron to the Suf Fe-S cluster biogenesis pathway (72). Thus, the present findings are consistent with a regulatory role of CsrA in the prevention and repair of oxidative damage.
CsrA modestly activated ftnA expression, in contrast to its stronger negative effects on the genes for the other ISPs. Unlike FtnB, Bfr, and Dps, FtnA does not appear to donate iron to the Suf Fe-S cluster biogenesis pathway (72). In fact, FtnA facilitates the assembly of iron-sulfur clusters via the Isc pathway under normal physiological conditions (73). Although Suf Fe-S biogenesis is preferred to Isc assembly during stress, the Suf pathway fails to fully mature some Fe-S proteins and may only meet the minimal Fe-S requirements for growth (43,48,74). As such, the activation of FtnA and repression of FtnB, Bfr, Dps, and SufA by CsrA may assist in directing iron stores toward the Fe-S assembly pathway that is most beneficial for exponential growth (Fig. 9). These findings suggest an explanation for the csrA growth defect in minimal medium, even when iron is replete. The loss of CsrA activation of FtnA and repression of FtnB, Bfr, Dps, and SufA apparently result in an inefficient utilization of available iron, which is particularly detrimental in minimal medium, where the demand for Fe-S proteins involved in biosynthetic processes may be high (75).
Interestingly, the ferritins are not the first example of CsrA mediating opposing regulatory effects on proteins having identical or highly related activities. For example, expression of the major phosphofructokinase, PfkA, is activated by CsrA, while the minor and s -inducible enzyme, PfkB, is repressed by CsrA (1, 19, 76). The different allosteric responses of these enzymes to key metabolites are most likely the basis of their distinct regulatory patterns. To our knowledge, in such cases, CsrA consistently serves to repress the expression of genes associated with stationary-phase growth and/or stress responses, while tending to activate genes that are needed to support exponential growth.
The dramatically increased resistance of the csrA mutant to H 2 O 2 during exponential growth demonstrates a regulatory role of CsrA in the resistance of E. coli to oxidative damage (Fig. 8). This is further supported by transcriptomics studies, which found increased steady-state mRNA levels for genes involved in antioxidant defense in the csrA mutant during exponential-phase growth, including the transcriptional activator of the superoxide response regulon, soxS, superoxide dismutases sodA and sodC, and catalase katE (19). CsrA also copurifies with mRNAs encoding the superoxide dismutase SodB and the catalase-peroxidase KatG (4). A recent study also provided evidence for a role for CsrA in repressing the oxidative stress response in Salmonella (58). The repression of many genes involved in oxidative stress response pathways by CsrA suggests that ungoverned expression of these processes may be detrimental during exponential growth and in the absence of stress, when conditions favor high CsrA activity (30). Finally, the unexpected finding that the csrA mutant became less resistant than the WT to H 2 O 2 in the stationary phase of growth (see Fig. S3 in the supplemental material) suggests that CsrA also performs important yet unexplained functions required for oxidative stress resistance in the stationary phase.
Commensal E. coli is ubiquitous in the gastrointestinal tracts of mammals and seemingly has colonized every mammalian species on the planet (77). While much research has been devoted to the role of CsrA in regulating virulence gene expression in pathogenic Gammaproteobacteria (10), little is known about its role in commensal colonization. E. coli inhabits the large intestine, which is primarily anoxic and nutrient limited. In this environment, soluble free iron is a contested and often limited resource (78). In addition, reactive oxygen species are scarce in the healthy intestinal lumen, although this can change during inflammation. For commensal microbes, iron is critical for enzyme function and energy generation. By repressing genes with the potential to sequester iron or divert iron to unnecessary stress response processes, CsrA fosters E. coli growth under iron-limiting laboratory conditions. We propose that it may play a similar role during colonization and survival in the large intestine.

MATERIALS AND METHODS
Bacterial strains and culture conditions. The bacterial strains and plasmids used in this study are listed in Table S2 in the supplemental material. All media and medium components were prepared using Nanopure water, 18.2 M⍀/cm (Barnstead). Bacterial strains were grown in LB broth (5 g yeast extract, 10 g Bacto tryptone, and 10 g NaCl per liter double-distilled water [ddH 2  When necessary, the following antibiotics were added to growth media: ampicillin (100 g/ml), tetracycline (15 g/ml), gentamicin (10 g/ml), kanamycin (50 g/ ml), and chloramphenicol (25 g/ml). Overnight cultures were routinely used to inoculate LB broth or minimal medium unless otherwise indicated. For strains carrying a dps deletion, exponentially growing cultures were used to prepare frozen glycerol stocks and to inoculate LB broth to eliminate/minimize the time spent under stationary-phase conditions. Transduction with P1vir was used to introduce gene deletions and disruptions from E. coli donor strains constructed in previous studies (31,80) and from the Keio library (81) (Table S2). Plasmids pBR322 (82) and pCRA16 (54) (csrA cloned into pBR322) were used in complementation tests. The Flp recombinase encoded in pCP20 (83) was used to eliminate the kanamycin resistance cassette, as required.
Considerations for strain construction. Because complete loss of CsrA activity causes severe growth defects and genetic instability (84,85), our experiments were performed with strains carrying a mutation (csrA::gm) producing a protein that contains the first 50 of 61 amino acids of the native protein and has residual CsrA activity (31). This csrA allele can be transduced by selection for a distally encoded gentamicin resistance (gm) marker. Strains containing fur csrA double mutations produced copious biofilm, which interfered with growth and reporter assays (data not shown). CsrA represses pgaABCD expression, required for synthesis and secretion of the biofilm adhesin poly-␤-1,6-N-acetyl-D-glucosamine (PGA) (54), and studies have also documented a role for Fur in biofilm formation (86,87). Deletion of pgaC, encoding the PGA glycosyl transferase, abolished biofilm production by the strains in LB medium. Consequently, studies in LB were performed in the pgaC mutant background.
Construction of =lacZ reporter fusions. Chromosomal translational fusions to =lacZ were constructed using the CRIM system (88) and plasmid vector pLFT (4) and integrated at the att site. Single-copy integrants were confirmed by PCR, as described previously (88). Constructions were performed as follows: ϳ500 nucleotides (nt) of DNA upstream of and including the promoter region through one or more codons downstream of the translational start site was amplified by PCR using the associated primers (see Table S3). The PCR products were gel purified, digested with PstI and BamHI, ligated into PstI-and BamHI-digested and dephosphorylated plasmid pLFT, and electroporated into DH5␣ pir cells. The fusion sequences were verified, and plasmids were isolated and integrated into the att site of strain MG1655 ΔlacZ using the helper plasmid pFINT (4).
␤-Galactosidase assay. Strains containing =lacZ fusions were grown at 37°C in Luria-Bertani (LB) broth. Tetracycline (15 g/ml) was used to ensure plasmid maintenance in strains bearing pBR322 or pCRA16. Cells were harvested at various times and ␤-galactosidase activity was measured as described previously (4). Total cell protein was measured following precipitation with 10% trichloroacetic acid, using the bicinchoninic acid (BCA) assay (Pierce Biotechnology) with bovine serum albumin as the protein standard.
Electrophoretic gel mobility shift assays for RNA binding. The binding of CsrA to ftnB, dps, and bfr transcripts was determined by EMSA using in vitro-synthesized ftnB, dps, and bfr transcripts (MAXIscript SP6/MEGAshortscript T7 kits; Ambion) and recombinant CsrA-His 6 (14). The template DNAs for in vitro transcription of ftnB and bfr were generated by PCR from MG1655 genomic DNA, using oligonucleotide pairs ftnB SP6 fwd EMSA/ftnB SP6 rev EMSA and bfr T7 fwd EMSA/bfr T7 rev EMSA. The template DNA for in vitro transcription of dps was generated by annealing the oligonucleotide pair dps T7 EMSA/dps T7 EMSA comp. RNA was synthesized from the ftnB (127 nt, consisting of 127 nt of the 5= UTR), dps (50 nt, consisting of 39 nt of the 5= UTR and 11 nt of the coding region), and bfr (81 nt, consisting of the 23 nt 5= UTR and 58 nt of the coding region) templates in vitro using the MEGAshortscript kit (Ambion). The resulting transcripts were purified via denaturing polyacrylamide gel electrophoresis (PAGE) followed by phenol-chloroform extraction and ethanol precipitation. Transcripts were treated with Antarctic phosphatase (NEB) and radiolabeled at the 5= end using [␥-32 P]ATP and T4 polynucleotide kinase. Binding reaction mixtures contained 0.5 nM RNA, 10 mM MgCl 2 , 100 mM KCl, 32.5 ng total yeast RNA, 20 mM dithiothreitol (DTT), 7.5% glycerol, 4 U SUPERasin (Ambion), and various concentrations of recombinant CsrA and were incubated at 37°C for 30 min. Reaction mixtures were separated on 9% native polyacrylamide gels (for ftnB mRNA) and 12% native polyacrylamide gels (for dps and bfr mRNAs) with 1ϫ Tris-borate-EDTA (TBE) buffer. Competition assays were performed in the presence or absence of unlabeled specific (self) and nonspecific (phoB) RNA competitors using the minimum CsrA concentration required for a full shift. Labeled RNA was analyzed using a phosphorimager equipped with Quantity One software (Bio-Rad), as previously described (4). The apparent equilibrium binding constant (K D ) for CsrA-RNA complex formation was calculated according to a previously described cooperative binding equation (14).
Coupled transcription-translation assay. In vitro coupled transcription-translation assays using PURExpress (New England BioLabs) were performed according to a published procedure (89). Plasmid pYH333 contains a T7 promoter driving transcription of the bfr translational fusion (nt ϩ1 to ϩ50 relative to the bfr transcriptional start site). Plasmid pYH334 contains a T7 promoter driving transcription of the dps translational fusion (nt ϩ1 to ϩ81 relative to the dps transcriptional start site). Plasmid pYH336 contains a T7 promoter driving transcription of the ftnB translational fusion (nt ϩ1 to ϩ152 relative to the ftnB transcriptional start site). These plasmids were used as the templates for coupled transcriptiontranslation reactions using the PURExpress in vitro protein synthesis kit, according to the manufacturer's instructions. Each 6.7-l reaction mixture contained 250 ng of plasmid DNA template, various concentrations of purified CsrA-H 6 , 1 U of RNasin (Promega), 2.5 mM DTT, 2.7 l of solution A, and 2 l of solution B. The mixtures were incubated for 2.5 h at 37°C, and ␤-galactosidase activity was determined according to the manufacturer's instructions.
Cellular iron measurement. Total cellular Fe concentrations were measured by inductively coupled plasma optical emission spectrometry (ICP-OES) at the University of Florida Institute of Food and Agricultural Sciences Analytical Services Laboratories. LB medium (500 ml) was inoculated with overnight cultures and incubated at 37°C with shaking. Two hundred fifty milliliters of exponential-phase cultures (OD 600 of 0.5) were harvested by centrifugation and washed three times in cold phosphate-buffered saline (PBS). The resulting bacterial pellets were resuspended in 1 ml 35% HNO 3 (trace-metal grade) treated at 95°C and diluted 1:10 with Invitrogen UltraPure distilled water. Fe concentrations were normalized to total protein. For quantification of iron in the medium, 2 ml trace-metal-grade 35% HNO 3 was added to 18 ml medium for ICP-OES analysis. Glassware for these experiments was pretreated overnight to remove trace metal contamination with 20% trace-metal-grade HNO 3 and multiple rinses with Nanopure water.
Growth kinetics assay. Growth under Fe-limiting and -replete conditions was monitored by OD 600 measurements in MOPS minimal medium (see "Bacterial strains and culture conditions") or by BCA protein assay in Luria-Bertani (LB) medium. All medium components were filter sterilized. Prior to use, labware was treated overnight in a 20% trace-metal-grade HNO 3 and rinsed multiple times with Nanopure water. ICP-OES analysis indicated that final Fe concentration of the MOPS medium before Fe supplementation was below 0.01 M. For growth experiments performed in MOPS, overnight cultures were grown in MOPS with 0.05 M FeSO 4 and were diluted 1:100 (except for cultures carrying the dps mutation, which were inoculated directly from Ϫ80°C glycerol stocks) and grown to exponential phase in MOPS with 0.05 or 100 M FeSO 4 . Growth experiments in LB were started from overnight cultures in LB, which were diluted 1:100 and grown to exponential phase in LB. The growth curves in all media were  The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.