Bacterial-Chromatin Structural Proteins Regulate the Bimodal Expression of the Locus of Enterocyte Effacement (LEE) Pathogenicity Island in Enteropathogenic Escherichia coli

ABSTRACT In enteropathogenic Escherichia coli (EPEC), the locus of enterocyte effacement (LEE) encodes a type 3 secretion system (T3SS) essential for pathogenesis. This pathogenicity island comprises five major operons (LEE1 to LEE5), with the LEE5 operon encoding T3SS effectors involved in the intimate adherence of bacteria to enterocytes. The first operon, LEE1, encodes Ler (LEE-encoded regulator), an H-NS (nucleoid structuring protein) paralog that alleviates the LEE H-NS silencing. We observed that the LEE5 and LEE1 promoters present a bimodal expression pattern, depending on environmental stimuli. One key regulator of bimodal LEE1 and LEE5 expression is ler expression, which fluctuates in response to different growth conditions. Under conditions in vitro considered to be equivalent to nonoptimal conditions for virulence, the opposing regulatory effects of H-NS and Ler can lead to the emergence of two bacterial subpopulations. H-NS and Ler share nucleation binding sites in the LEE5 promoter region, but H-NS binding results in local DNA structural modifications distinct from those generated through Ler binding, at least in vitro. Thus, we show how two nucleoid-binding proteins can contribute to the epigenetic regulation of bacterial virulence and lead to opposing bacterial fates. This finding implicates for the first time bacterial-chromatin structural proteins in the bimodal regulation of gene expression.

In pathogenic bacteria, stochastic gene expression can lead to distinct virulent states (8) or persistence (9,10) or heterogeneity in host immune responses (11). Under virulence-inducing conditions, bimodal expression patterns have been reported for several pathogenicity factors. These factors include expression of type 1 pili by Streptococcus pneumoniae (12) and type III secretion system (T3SS) by the phytopathogenic bacterium Dickeya dadantii (13) or Pseudomonas aeruginosa (14).
During Salmonella enterica serotype Typhimurium infection, division of labor occurs (15), with only some cells producing the T3SS. However, the fraction of bacteria producing SPI-1 T3SS acquires a growth penalty, resulting in loss of fitness (8). Most SPI-1-expressing bacteria die inside host cells, generating inflammation (16). In turn, in the gut lumen, inflammation confers a selective advantage to the mainly non-SPI-1expressing Salmonella over the microbiota and thereby promotes the stability of virulence in the evolutionary context (15,17). Similarly, phenotypically T3SS-expressingand non-T3SS-expressing bacteria coexist within the P. aeruginosa population in a murine model of acute pneumonia, suggesting that non-T3SS-expressing bacteria behave as cheaters, taking advantage of T3SS-expressing bacteria (14). Taken together, these studies highlight the importance of gene expression stochasticity to ensure the necessary phenotypes required for successful infection and survival.
In attaching/effacing (A/E) pathogens, such as enteropathogenic Escherichia coli (EPEC) and enterohemorrhagic Escherichia coli (EHEC), the expression of T3SS is central to pathogenesis and is associated with the locus of enterocyte effacement (LEE) pathogenicity island. LEE is a horizontally acquired AT-rich DNA locus and comprises 41 genes arranged in five polycistronic operons (designated LEE1 to LEE5) (18)(19)(20). The expression of all LEE genes is silenced by H-NS, an abundant nucleoid-associated protein. H-NS is a xenogeneic silencer that acts as a repressor of gene expression in elements recently acquired horizontally (21,22). Indeed, H-NS preferentially blocks transcription at these AT-rich acquired loci, facilitating foreign DNA incorporation into the chromosome. In addition to promoters of their own genes, AT-rich regions contain sequences that mimic polymerase-binding sites. Thus, transcription start sites have been mapped to unexpected locations in bacterial genomes, including the noncoding strand. H-NS also acts to silence these elements. Hence, a key function of H-NS is to ensure transcriptional specificity (23). H-NS organizes bacterial chromatin by binding to regions in vivo as long as 1,500 bp (24), forming nucleoprotein filaments organized in either stiffened or bridged DNA conformations depending on the presence of Mg 2ϩ (25)(26)(27)(28)(29). H-NS-bound regions are associated with low or no transcriptional activity (22,(30)(31)(32). At promoters, silencing by H-NS is often alleviated by H-NS antagonists that interfere with the H-NS-DNA complex structure, with or without concomitant displacement of H-NS (33,34). Among these antagonists, Ler, the first protein produced from LEE under the control of the products of the perABC operon, is an H-NS paralog. Ler relieves H-NS silencing specifically at LEE promoters and a few other targets (20,35). Recently, a growth rate bimodality, mediated by a hysteretic memory switch, was reported for EPEC (36). This bimodality results in the coexistence of nonvirulent and hypervirulent subpopulations. The hypervirulent subpopulation continues to express virulence after several generations of growth under nonactivating conditions. The main regulators of this hysteretic switch are the products of the perABC operon. Ler itself is not involved (36). This heterogeneity has been proposed to reflect a bet-hedging strategy (36). In this case, a subset of the cell population presents a phenotype considered nonoptimal or nonadapted that may be advantageous if environmental conditions change (e.g., sudden stress, rapid return to a previous situation). For example, in E. coli, such strategy has been reported for SOS genes and colicin expression (37,38).
The LEE5 promoter (P LEE5 ) controls the operon encoding the adhesin intimin (eae), its receptor (tir), and a chaperone (cesT). The intimin and Tir proteins are major virulence factors (39). The aim of the present study is to explore whether the opposing regulatory effects of Ler and H-NS on T3SS expression in EPEC at the individual cell level can be involved in a bimodal population pattern.
Here, we describe the bimodal expression pattern of P LEE5 under growth conditions generally considered mimicking conditions nonoptimal for virulence. This expression pattern is controlled by the interplay of H-NS and Ler. We show that H-NS and Ler, binding at the same nucleation DNA motif, induce different nucleoprotein structures in the isolated P LEE5 . Finally, we observe that under different environmental conditions, the level of Ler expression is a key element controlling the bimodality of LEE5 expression under different environmental conditions. Thus, the balance between H-NS silencing and Ler antisilencing activities generates nongenetic variability.

RESULTS
The expression from the LEE5 promoter is bimodal in exponential phase. Classically, infections of epithelial cells with EPEC are assayed in Dulbecco's modified Eagle's medium (DMEM). Indeed, the expression of EPEC virulence is generally considered to be active when grown in DMEM at 37°C. In such "activating" conditions, most virulence genes are expressed but not in Luria-Bertani liquid medium (LB) (see "Media" in Materials and Methods), "nonactivating" conditions (40)(41)(42)(43).
In order to explore a potential population phenotypic heterogeneity, we assessed P LEE5 (i.e., normally expressing intimin and Tir) activity in EPEC in these activating and nonactivating conditions. We wished to explore the heterogeneity of LEE5 expression at the individual cell level under these two conditions, since it might reflect either bet-hedging or division of labor strategies. In the case of bet-hedging, we could expect, for example, the presence of a subpopulation of LEE5-expressing bacteria in nonactivating conditions (LB). In contrast, a division of labor strategy could be indicated by bimodal expression of LEE5 in activating condition (DMEM).
To do so, we introduced a gfp reporter under the control of P LEE5 as a single copy on the EPEC chromosome at the attB Phi80 phage site and performed flow cytometry analysis (Fig. 1). Mean fluorescent measurement of the whole population ( Fig. 1A) confirmed that the upregulation of P LEE5 -gfp by Ler is eightfold higher in DMEM than in LB.
At the individual cell level in exponential phase, bimodal P LEE5 expression was observed in both LB and DMEM (Fig. 1B). Two subpopulations of bacteria were observed, bacteria expressing green fluorescent protein (GFP ϩ ) and bacteria expressing very low levels of GFP or not expressing GFP (GFP Ϫ ). The peak corresponding to GFP ϩ bacteria is visible only as a shoulder, presumably because the low cell fluorescence is closed to the sensitivity threshold. To amplify the signal and to confirm the presence of a bimodal phenotype in the cell population, we used a low-copy-number plasmid reporter (Ϸ10 copies per cell) carrying the P LEE5 -gfp cassette (44,45). This allowed us to clearly observe two subpopulations of cells expressing GFP either at a low level ("low state," with a distribution that differs slightly from the negative control without promoter) or expressing GFP at a high level ("high state"). The latter subpopulation displays a mean fluorescence intensity, as anticipated from the gene dosage effect, increased by 1 log unit compared to the subpopulation of GFP ϩ bacteria containing one chromosomal insertion (Fig. 1B). Using this reporter system thus yields a better discrimination of the different populations and confirms the bimodal population pattern in exponential phase in both LB and DMEM (Fig. 1B).
For a control, a wild-type (WT) EPEC strain expressing gfp from the constitutive T5 phage P2 promoter was analyzed. The GFP expression pattern was unimodal throughout the bacterial population in all growth conditions (   NegaƟve controls: NegaƟve controls: LEE5 promoter expression progressively involves all the cells in activating conditions. In stationary-phase cultures expressing P LEE5 -gfp, two population patterns could be observed: a unimodal distribution, corresponding to the high state (growth in DMEM), and a bimodal distribution (growth in LB) (Fig. 1C). Under these conditions, the level of fluorescence in the cells results from GFP accumulation throughout the whole growth phase, since GFP is stable over the time of the experiment. We monitored the dilution of the GFP fluorescence to an undetectable level through cell division by shifting the culture to a nonpermissive temperature for LEE expression (Fig. S2) (20). We concluded that for the bimodal distribution in LB, the low-state subpopulation corresponds to bacteria that either never activated P LEE5 or activated it transiently during exponential phase. In the case of the unimodal population in DMEM, GFP accumulation thus indicates that all cells had expressed LEE5 at a high level in the experiment. Therefore, in activating conditions (DMEM), the switch on of P LEE5 is progressively spreading to the whole population.
To explore the hysteresis of the high state (i.e., its maintenance when the conditions that initially upregulate the promoter are not occurring anymore), we tested different culture inoculation conditions (Fig. S3). Notably, we observed that a unimodal cell population (inoculated from a culture in activating conditions, i.e., DMEM) reinoculated in fresh DMEM displayed a bimodal population pattern in exponential phase. This indicates that a resettable phenotypic switch controls the activation of LEE5 expression and that this phenotypic bimodal expression of LEE5 is not hysteretic.
Further analyses described below were all carried out in stationary phase, where the difference between LB (bimodal distribution) and DMEM (unimodal distribution) with respect to the pattern of LEE5 expression is observable (Fig. 1C).
The level of LEE5 promoter expression varies with the composition of the growth medium. To assess the impact of environmental conditions on the pattern of LEE5 expression and to mimic gastrointestinal repression or induction signals (39), we monitored LEE5 expression at stationary phase in various media. Notably, we tested the effect of ammonium chloride or sodium bicarbonate (the former acts as an inhibitor and the latter acts as an activator of LEE5 expression). Figure 2 shows the flow cytometry analysis of the WT EPEC with a P LEE5 -gfp reporter grown in eight different media.
The mean fluorescence of the whole cell population ( Fig. 2A) displayed a continuum of values for LEE5 expression levels according to the medium type. This apparent continuous variation reflects the average fluorescence of the entire population resulting from the distribution between the two subpopulations ( Fig. 2B). Indeed, depending on the composition of the medium, we again observed two patterns of LEE5 expression: a unimodal distribution corresponding to the high state (growth in DMEM or SF9 medium supplemented with sodium bicarbonate), and a bimodal distribution (growth in LB, SF9 medium, and CAA-Glc-M9 medium [M9 base with 0.5% Casamino Acids and 0.4% glucose] [see "Media" in Materials and Methods]) (Fig. 2B).
Altogether, our results indicate that describing media as activating and nonactivating does not adequately reflect the complexity of LEE5 expression. Henceforth, we shall therefore use the term "nonoptimal conditions" for P LEE5 -repressing conditions at the whole-population level (e.g., LB, SF9, and CAA-Glc-M9), conditions where LEE5 expression is low in most bacteria and high only in a small fraction of them. We will use the term "optimal conditions" when LEE5 expression was upregulated in all bacteria (e.g., DMEM, SF9 containing bicarbonate).
In conclusion, LEE5 expression was activated in all bacteria grown under optimal conditions but only in a subpopulation under nonoptimal conditions. This finding shows that under conditions classically considered to be repressive for LEE expression (40)(41)(42)(43), a small subpopulation is expressing LEE5 at high levels, suggesting a potential bet-hedging strategy.
The nucleoid-associated proteins Ler and H-NS are essential regulators of LEE5 bimodality. To assess the roles of Ler and H-NS in the bimodal expression of LEE5, gfp expression under the control of P LEE5 was monitored in WT EPEC and Δler, Δhns single or double mutant EPEC strains grown in nonoptimal (LB) and optimal (DMEM) media ( Fig. 3). In the absence of Ler and the presence of H-NS, only one population of bacteria was observed, and the peak was in the low state, confirming that Ler is required in some way for LEE5 activation in both LB and DMEM. Because Ler is required for virulence (46,47), we suggest that the low state likely corresponds to nonvirulent bacteria. This hypothesis is in accordance with the identification of a hypervirulent bacterial subpopulation that expressed Ler and T3SS at high levels (36). These experiments suggested a direct link between the level of Ler expression and virulence in an EPEC subpopulation.
We observed that the mean fluorescence of the Δler bacterial population was 1 log unit higher in DMEM than in LB, indicating that the medium composition affects the basal activity of P LEE5 , independent of Ler (Fig. 3). In the absence of H-NS, Ler was no longer required for P LEE5 activation, confirming that the main role of Ler on this promoter is to relieve H-NS silencing. In DMEM, deletion of hns has no apparent effect on LEE5 expression, indicating that in this growth medium, there is no repression by H-NS (due to Ler antisilencing activity). Interestingly, in LB medium, genetic inactivation of H-NS led to an upregulated unimodal distribution of fluorescence in the bacterial population, with all cells being in the high state (Fig. 3). Taken together, these observations indicate that both the Ler and H-NS proteins are required for the bimodal expression of LEE5.
Ler and H-NS bind common sites on the LEE5 promoter but affect differently the local DNA structure in vitro. To assess the molecular mechanisms underlying the effects of H-NS and Ler, we compared the binding of these two proteins to the P LEE5 region. We performed DNase I footprinting in the P LEE5 core region extending from In the present study, H-NS was found to cover a larger region of P LEE5 than Ler (Fig. S5). A comparable observation was previously reported, with H-NS covering larger DNA regions in vitro than Ler at the lpf1 promoter (50). This finding suggests that Ler binding may not spread along DNA as much as H-NS, which typically covers up to 1,500 bp (24).
H-NS binding to DNA is initiated at the level of consensus sequences (51,52). The P LEE5 region displays nine consensus sequences at positions Ϫ195 to Ϫ190, Ϫ160 to Ϫ150 (2 sites), Ϫ123 to Ϫ105 (2 sites), ϩ10 to ϩ20, ϩ75 to ϩ85, ϩ250 to ϩ270 (2 sites) (Fig. S4). When their consensus scores were compared using Virtual footprint software (53), the best predicted site was centered at position ϩ5, the second site was located at position Ϫ110, and the third site was located at position ϩ254 (Fig. S4). These three sites were simultaneously disrupted by substituting the central AT-rich motif with a CG-rich motif. In the resulting promoter, referred to as "P LEE5-3M ," these mutations altered the binding of both H-NS and Ler in vitro (Fig. S6). In vivo, examination of the expression of P LEE5-3M showed that the altered binding of both H-NS and Ler resulted     in weaker silencing by H-NS and weaker antisilencing by Ler compared with the native P LEE5 (Fig. 4C). Consequently, when measured at the individual cell level, the two bacterial subpopulations (low and high states) are closer to each other (Fig. 4D). Taken together, these results indicate that H-NS and Ler recognize the same or a very similar nucleation DNA motif, but H-NS induces different DNA structural changes at the RNA polymerase-binding site and covers a longer DNA region than Ler. This suggests that the roles of Ler and H-NS in the bimodal expression of LEE5 involve competitive binding and distinctive modifications of local DNA structure organization.
LEE5 expression is finely tuned by Ler. We next explored variations in expression by measuring the fluorescence of cells harboring the gfp gene under the control of the LEE1 promoter ("P LEE1 "). For P LEE1 , similar to the results with P LEE5 , we observed a bimodal pattern of expression under nonoptimal conditions (LB) and a unimodal distribution of highly expressing cells under optimal conditions (DMEM) (Fig. 5A). These results are consistent with previous measurements of ler promoter expression using a chromosomal ler-gfp transcriptional fusion (36). At the whole-population level, P LEE1 activity, like P LEE5 activity, varied depending on the presence or absence of either Ler or H-NS. LEE1 expression was reduced in the absence of Ler (regardless of the medium) and increased in the absence of H-NS under nonoptimal conditions (LB). Under optimal conditions (DMEM), the H-NS silencing of P LEE1 (as observed for P LEE5 above) was completely relieved (Fig. 5A). These observations were confirmed by reverse transcription-quantitative PCR (RT-qPCR) (Fig. S7).
Under nonoptimal conditions, the distribution of the Δler bacterial population was unimodal, and its fluorescence level fell between the low and high states of the WT strain (Fig. 5A). These findings are consistent with previous reports and explain why either a negative effect (54,55) or a positive effect (35) of Ler on its own promoter was previously observed depending on growth conditions. Since GrlA, PerA, and PerC are major activators of LEE1 expression, depending on growth conditions (36, 56), we assessed the precise roles of these activators on both P LEE1 and P LEE5 activity in the different media used here (Fig. 5B). The deletion of ler reduced P LEE1 activity but had a lower impact than the double inactivation of both grlA and perC, regardless of the medium composition (Fig. 5B). This indicates that expression from P LEE1 is highly dependent on these activators, while Ler plays a secondary role. As previously described (56), PerA, PerC, and GrlA independently activate ler expression in DMEM. Additionally, these factors were required for the optimal expression of both LEE1 and LEE5 in all tested media (Fig. 5B). The variation of P LEE5 expression therefore correlates with Ler production according to both the medium composition and the control by PerA, PerC, and GrlA. In nonoptimal conditions, LEE1 and LEE5 expression remained bimodal in the ΔgrlA mutant but was unimodal and at a lower level in the ΔperC and ΔgrlA ΔperC mutant strains (data not shown). To formally show that the Ler protein directly controls the variation of P LEE5 expression, we constructed a synthetic, tunable promoter (Tet-ON) controlling ler in the commensal E. coli K-12 strain. Increasing the inducer concentration resulted in a shift between the two populations expressing LEE5 at low and high states (Fig. 6). Importantly, a bimodal pattern, with two subpopulations of cells, was also observed at an intermediary dose of the inducer (Fig. 6). Thus, in the absence of a complete LEE island and additional virulence factors, modulation of Ler protein levels is sufficient to induce and modulate a bimodal pattern of LEE5 expression.

DISCUSSION
In the present study, we showed that H-NS and Ler, which regulate the LEE1 and LEE5 promoters, are essential for generating a bimodal pattern of expression. The key parameter, depending on growth conditions, is the modulation of Ler expression (Fig. 7). Under appropriate environmental conditions (e.g., DMEM or Glc-CAA-M9* [Glc-CAA-M9 without NH 4 Cl] plus NaHCO 3 ), GrlA, PerA, and PerC activate LEE1 transcription. Moreover, Ler exerts a dual regulatory effect on its own promoter, P LEE1negative autoregulation (54, 55) and/or positive indirect activation, via the stimulation of GrlA expression (57) (Fig. 5 and 7). Accordingly, at the single-cell level, the two subpopulations (in high and low states) in the WT strain merge into a single unimodal population presenting an intermediate level of LEE1 expression in the Δler mutant (Fig. 5). Therefore, the bimodal distribution observed within a population expressing LEE1 under nonoptimal conditions may reflect the balance between these two opposing feedback loops ( Fig. 5 and 7), a type of network that has been shown to lead to bimodality (1,58). Moreover, we showed that stochastic expression of Ler propagates to its downstream target LEE5. Fluctuations in Ler levels, possibly due to the bimodal expression of the perABC operon (36) lead to stochastic LEE5 expression resulting from an imbalance between Ler and H-NS levels. This imbalance manifests when bacteria are grown under nonoptimal conditions, where the quantity of Ler determines the fate of LEE5 expression. We propose that if the concentration of Ler is sufficiently high, Ler overrides the silencing of P LEE5 through H-NS. Otherwise, H-NS silencing predominates (Fig. 7). High and low states of expression depend upon amplification phenomena (i.e., H-NS or Ler cooperative binding, positive-feedback loops). We also propose that in the subpopulation in a high state, a positive-feedback loop maintains Ler expression at a high level, while the second population is in a low state. In this case, H-NS repression of P LEE1 and P LEE5 predominates (Fig. 7).
The bimodal expression of LEE5 and LEE1 in nonoptimal conditions in this study are reminiscent of a previously described bimodal growth rate phenotype, illustrated by small and large colonies of EPEC on DMEM plates (36). Small colonies correspond to hypervirulent bacteria expressing Ler and T3SS at a high level (36). This growth rate phenotype that results in bimodality of host cell infectivity required the per operon but not Ler and T3SS (36). In contrast, here we show the existence of a distinctive bimodality controlling T3SS expression that requires Ler and H-NS. Therefore, these two bimodal phenotypes (growth rate [36] and T3SS expression [this study]) appear to be under the control of different regulatory mechanisms. Accordingly, in coculture experiments using WT EPEC and Δler strains, T3SS expression in the WT strain did not apparently influence fitness under the experimental conditions of the present study (see Fig. S8 in the supplemental material). Notably, the variability in various phenotypic states observed here under conditions (LB) that were nonoptimal for virulence may be advantageous for rapid adaptation to changes in environmental conditions. Consequently, the results of these ex vivo experiments suggest a bet-hedging strategy (1, 4, , where a small subpopulation is primed to take advantage of environmental changes. In the case of virulence, a bet-hedging strategy might bring a selective advantage by increasing the chance of successful infection or host-to-host spreading. A potential bet-hedging strategy for growth rate control is also supported as previously described by Ronin et al. (36) by the observation of large and small colonies, even after many generations of growth in nonoptimal conditions. However, we cannot exclude the possibility that bimodal expression of LEE1 and LEE5 observed here might belong to a division of labor scenario, i.e., that during the course of an infection, both populations (in low and high states) may cooperate. This hypothesis may be relevant, since we observed transiently these two subpopulations in optimal conditions (DMEM and cells in exponential phase), which merged into a single upregulated population during growth of the cultures. In this case, the different phenotypes in the population may participate in specific tasks that ensure the survival of the shared genotype. Thus, the potential importance of coexisting bimodal patterns for bet-hedging or division of labor strategies for A/E pathogens remains to be further explored. From our studies, the interplay between two proteins from the H-NS family appears to be at the heart of the stochastic gene expression regulating virulence expression. H-NS, described as a chromatin organizer protein, is highly abundant and is constitutively bound to the nucleoid (61)(62)(63). Conversely, Ler is only transiently expressed at variable levels, depending on environmental stimuli (Fig. 5), and thus, this protein could play a role as a "chromatin remodeler" of the promoter that it regulates. Notably, despite frequently being described as a fairly nonspecific protein, H-NS controls sophisticated regulatory networks in coordination with Ler, one of its paralogs. Finally, this study shows for the first time that the H-NS protein family is involved in the stochastic regulation of gene expression. Other bacterial pathogenicity islands are similarly regulated through the interplay between H-NS and antagonist proteins, such as SlyA in Salmonella, RovA in Yersinia, or ToxT in Vibrio cholerae (64). Future studies regarding potential bimodal expression in these organisms under specific growth conditions could provide further evidence that bacterial-chromatin structure plays an important role in the epigenetics and virulence of bacteria.

MATERIALS AND METHODS
Strains, plasmids, promoter fragments, and primers. E. coli K-12 and the EPEC E2348/69 strains, plasmids, and primers are listed in Tables S1 and S2 in the supplemental material. For recombinant DNA manipulation, standard techniques were used.
Promoter fragments were amplified through PCR. By convention, the promoter sequences are numbered with respect to the transcription start point (ϩ1), with upstream and downstream locations denoted by the "-" and "ϩ" prefixes, respectively. Fragments of the EPEC P LEE5 (Ϫ249 to ϩ273) (Fig. S4) and EPEC P LEE1 (Ϫ257 to ϩ285), where ϩ1 refers to the transcription start site of the LEE1 P1A promoter (65), were amplified from genomic DNA using the primers proLEE5S/proLEE5R and proLEE1S/proLEE1R, respectively, and subcloned into the pGEM-T Easy vector (Promega). For footprinting experiments, the extended P LEE5 fragment (Ϫ228 to ϩ273) was amplified using the primers proLEE5S4 and proLEE5R. The central P LEE5 (Ϫ80 to ϩ104) fragment was amplified using the primers proLEE5SVI and proLEE5RII. In all cases, the pGEM-T-Easy-P LEE5 construct was used as the template for PCR. To generate the mutated promoter fragment (Fig. S4), the plasmid pMRQ-P LEE5-3M (Genart; Life Technology) was used as the template. To assay promoter activities, P LEE5 (Ϫ249 to ϩ273) and P LEE1 (Ϫ257 to ϩ285) fragments were cloned into pKK-gfp to obtain pKK-P LEE5 -gfp and pKK-P LEE1 -gfp (51). Notably, the pKK-gfp plasmid has a medium-to-low copy number (Ϸ10 per bacterium) (44,45). The pKK-gfp promoter-less gfp plasmid was used as a negative control to determine the basal fluorescence level of the bacteria. The phage T5 constitutive promoter (a kind gift from Pascale Boulanger), referred to as "P2" (Table S1), was cloned into the pKK-gfp plasmid using XhoI and XbaI and used as a control.
We also used the LEE5 reporter cassette as a single copy at the EPEC attB Phi80 site on the chromosome. The fragment of pKK-P LEE5 -gfp containing the XmaI site and EcoRV P LEE5 -gfp was subcloned into the pBBint⌽ integrative base vector using the AgeI and HincII sites. Chromosomal integration with Phi80 phage integrase was performed in WT EPEC and Δler strains, as previously described (66).
Purification of H-NS and Ler proteins. The H-NS protein was purified as previously described (67), and its concentration was measured according to a previous study (61). The Ler expression plasmid was constructed from the EPEC genome through ler gene PCR amplification using the primers "Ler F3" and "Ler R3" (Table S2) and subsequently subcloned into pET 28, generating pET-ler. E. coli BL21(DE3)/pLysS (Invitrogen) cells were transformed with pET-ler and used for Ler protein overexpression. The cells were grown in LB at 37°C until reaching an optical density at 600 nm (OD 600 ) of 0.8. Subsequently, protein production was induced by adding 1 mM isopropyl-␤-D-thiogalactopyranoside (IPTG), and the cells were harvested 1 h later by centrifugation at 5,000 ϫ g at 4°C. The cells from 1 liter of culture were resuspended in 10 ml of buffer A (20 mM phosphate, 0.5 M NaCl, 5 mM dithiothreitol [DTT], and 80 mM imidazole [pH 7]), and a cocktail of protease inhibitors (Roche) at the concentration recommended by the manufacturer. The cells were disrupted using a French press at 1,500 lb/in 2 , followed by centrifugation (18,000 ϫ g, 40 min, 4°C), and the supernatant was applied to a nickel-nitrilotriacetic acid (Ni-NTA) affinity column (GE Healthcare) equilibrated with buffer A containing 80 mM imidazole. The column was subsequently washed with buffer A containing 100 mM imidazole, and the protein was eluted with buffer A containing 500 mM imidazole. The fractions were dialyzed against buffer A containing 25% glycerol, and proteins were quantified using the Bradford assay with H-NS as a standard. Aliquots were frozen in liquid N 2 and stored at Ϫ20°C until further use. The quality of the purification was determined after SDS-PAGE analysis and staining with InstantBlue.
DNase I footprinting. Fragments were generated by PCR using one primer end labeled with [␥-32 P]ATP (3,000 Ci mmol Ϫ1 ) and the phage T4 polynucleotide kinase (NEB). DNase I footprinting was performed after incubating a 2 to 5 nM concentration of the end-labeled promoter fragment with the proteins at the indicated concentrations at 20°C in a buffer containing 10 mM HEPES (pH 7), 50 mM K H-NS-Like Proteins Regulate LEE Bimodal Expression ® glutamate, 8 mM Mg aspartate, 4 mM DTT, 10 g/ml of bovine serum albumin, and 0.01% NP-40. The digested products were then migrated in denaturing 7% acrylamide (19:1) gels. The analysis was performed as previously described (51).
Flow cytometry analysis. Bacteria were precultured overnight in 4 ml of LB supplemented with ampicillin (100 g/ml) at 37°C under agitation. The samples were then diluted 1:1,000 in 4 ml of the appropriate medium in 15-ml conical tubes and incubated at 37°C in a shaking incubator (160 rpm, INFORS AG CH-4103). After 3 h or 24 h, single-cell fluorescence was measured using either a BD FACSCalibur (BD Biosciences) or CyFlowCube8 (Partec) flow cytometer and analyzed using FlowJo software. Bacteria harboring pKK-gfp were employed to calibrate appropriately the FL-1 voltage. In parallel, we measured the turbidimetry at 600 nm of each sample. We used a magnetic gate (FlowJo) selecting Ϸ30% of the bacterial population corresponding to the most-frequent side scatter (SSC)-forward scatter (FSC) pattern (Ϸ10,000 events). This kind of filtering minimizes the analysis of cells differing in size and complexity that could affect the variability of fluorescence (68). The magnetic gate (FlowJo), centered on each population, allows an accurate gate on populations that may shift slightly between samples. The data were normalized to the mode and smoothed using FlowJo software.