Bacterial strains and plasmids are summarized in Supplementary Table S1. Mutant strains were generated by the Red recombinase system (Datsenko and Wanner, 2000; Murphy and Campellone, 2003), and verified by PCR and sequencing. Complemented strains were constructed by cloning the appropriate genes into low-copy plasmid pACYC184, and then by electroporating the resulting constructs into the corresponding mutants. Strains for protein purification were constructed by cloning genes of interest into the pET28a expression vector, and then by electroporating the resulting constructs into E. coli BL21. All genetic manipulation on virulent bacterial strains was performed according to standard biosecurity and institutional safety procedures. Primers for all manipulations are listed in Supplementary Table S2. Unless otherwise specified, all strains were grown in Luria-Bertani (LB) broth supplemented as needed with 100 μg mL^-1 ampicillin, 15 μg mL^-1 chloramphenicol, and 50 μg mL^-1 kanamycin.

Adherence was assayed as previously described (Dibb-Fuller et al., 2001). Briefly, HeLa and Caco-2 cells, obtained from Shanghai Institute of Biochemistry and Cell Biology, Chinese Academy of Sciences (Shanghai, China), were grown at 37°C in 5% CO₂ until confluent, washed three times with pre-warmed PBS, and the medium was replaced with fresh DMEM without antibiotics and fetal bovine serum. Cells were then infected with bacterial cultures in exponential phase (10⁸ CFU/well). After 3 h, unattached bacteria were removed by washing the wells six times with PBS. Cells were then lysed with 0.1% SDS, and resulting lysates were serially diluted and plated on LB agar. Attachment efficiency was calculated as the numbers of adherent bacteria per cell.

Total RNA was prepared using TRIzol^® LS Reagent (Invitrogen: 15596018) following the manufacturer’s instructions, and digested with RNase-Free DNase I (Qiagen: 79254) to eliminate contaminating genomic DNA. First-strand cDNA was synthesized using PrimeScript 1^st Strand cDNA Synthesis Kit (Takara: D6110A), and analyzed by qRT-PCR on an ABI 7500 system (Applied Biosystems) using SYBR Green PCR master mix (Applied Biosystems: 4367659). The 16S rRNA gene rrsH was used as reference, and relative differences in gene expression were calculated by the cycle threshold method (2^-ΔΔct) (Livak and Schmittgen, 2001; Tasara and Stephan, 2007). Data were collected from at least three biological replicates.

Overnight cultures were adjusted to optical density 1.0 at 600 nm, of which 1 μL was then stab-inoculated using a sterile pipette tip into 0.25% LB-agar plates supplemented with ampicillin as needed. Agar plates were then incubated at 30 or 37°C for 10 h, at which point the diameter of the swimming zone around the inoculation site was measured. All strains were tested in triplicate, and each experiment was carried out on three separate occasions.

Strains were cultured in LB broth at 37°C until optical density 0.8 at 600 nm. Samples (1 mL) were then harvested at 2,000 rpm for 5 min, resuspended in an equal volume of distilled water, of which 10 μL was dropped and adsorbed for 3 min to carbon-stabilized Formvar supports on 200-mesh copper grids. Cells were then stained by submerging the grids for 3 min in 2% wt/vol sodium phosphotungstate, and imaged on a HITACHI HT7700 transmission electron microscope operating at 100 kV and fitted with a high-sensitivity real-time CCD camera.

Overnight bacterial cultures were diluted 1:100 to optical density 1.0 at 600 nm. Whole-cell lysates were resolved on 12% sodium dodecyl sulfate-polyacrylamide gels, and transferred to polyvinylidene difluoride membranes. Subsequently, membranes were probed with a 1:10,000 dilution of antibodies to flagellin (Abcam: 93713) or DnaK (Abcam: ab69617), followed by a 1:2,000 dilution of goat anti-rabbit (Abcam: ab6721) or anti-mouse immunoglobulin G (Abcam: ab205719) conjugated to horseradish peroxidase. Blots were visualized on a chemiluminescence detection system following reaction with ECL enhanced chemiluminescence reagent. Proteins were quantified using Amersham Imager 600 software (GE Healthcare).

GmrA N-terminally tagged with 6× His was expressed in E. coli BL21, using the expression vector pET28a and purified from soluble extracts using nickel columns (GE Healthcare: 17057501). Protein concentration was determined by Bradford assay, and stored in aliquots at -70°C. PCR fragments encompassing regulatory regions of fliA (495 bp, -412 to +83) and flhD (541 bp, -336 to +205), with respect to the corresponding transcriptional start sites, were amplified using genomic DNA of E. coli O157:H7 EDL933 as template (for the regulatory regions of fliA and flhD, see Salgado et al., 2013). A rpoS fragment (384 bp, +1052 to +1435 relative to the transcriptional start site) was also amplified, and used as negative control. The DNA fragments were then gel-purified and labeled with DIG using terminal transferase. Eectrophoretic mobility shift assays were performed using DIG Gel Shift Kit, 2^nd Generation (Roche: 03353591910) according to the manufacturer’s instructions. Briefly, labeled DNA fragments (1 nM) were incubated at 37°C for 20 min with various concentrations of purified GmrA -His6 (0–120 nM), in 20 μL reactions containing band-shift buffer (20 mM Tris-HCl pH 7.5, 80 mM NaCl, 0.1 mM EDTA, and 1 mM DTT). For competition assays, various concentrations of unlabeled DNA fragments (10–150 nM) were added. Samples were separated by 10% native polyacrylamide gel electrophoresis, and transferred to nylon membranes. Labeled fragments were visualized on a chemiluminescence detection system following an enzyme immunoassay using anti-digoxigenin-AP, Fab-fragments, and the chemiluminescent substrate CSPD.

Chromatin immunoprecipitation was performed as previously described (Lucchini et al., 2006; Davies et al., 2011) with some modification. Briefly, an inducible expression vector (pTRC99a) carrying 3× FLAG-tagged gmrA was constructed and transformed into ΔgmrA mutant. Bacterial cultures were grown to mid-logarithmic phase until optical density 0.4 at 600 nm, and protein expression was induced with 1 mM IPTG for 30 min at 37°C. To crosslink protein to DNA, formaldehyde was added to cultures to a final concentration of 1%, and the mixture incubated at room temperature for 25 min. Cross-linking was quenched by adding glycine at a final concentrations of 0.5 M. Cross-linked cells were then washed three times with ice-cold TBS, and sonicated extensively to generate DNA fragments of average size ∼500 bp. Cell debris was removed, and the resulting supernatant was used as cell extract for immunoprecipitation. Protein-DNA complexes were enriched with 3× FLAG antibody (Sigma: F1804) and protein A magnetic beads (Invitrogen: 10002D), following the manufacturer’s instructions. As negative control, chromatin immunoprecipitation was performed using different aliquot without addition of antibodies. RNA were removed by incubation with RNaseA for 2 h at 37°C, and proteins were removed by incubation with proteinase K for 2 h at 55°C. The DNA sample was then purified using a PCR purification kit (Qiagen: 28104). To measure enrichment of fliA and flhDC promoters in immunoprecipitated DNA samples, relative-abundance quantitative PCR (qPCR) was performed with SYBR green mix. Relative enrichment was calculated by the ΔΔCt method (Livak and Schmittgen, 2001). Results shown represent average enrichment for three biological replicates.

Orthologous groups were identified using OrthoFinder (Emms and Kelly, 2015), by which all nucleotide sequences were compared using a BLASTN all-against-all search with an E-value cutoff of <10^-4. Nucleotide sequences used to construct the phylogenetic tree were aligned in MAFFT (Katoh and Standley, 2013), and a maximum likelihood tree was constructed in PhyML (Guindon and Gascuel, 2003) based on the GTR model of nucleotide substitution with c-distributed rates among sites.

Previously, we showed by comparative transcriptomics that genes in OI-29 are significantly downregulated 3 h after incubation of E. coli O157:H7 with HeLa cells (Yang et al., 2015). We have now confirmed this result by qRT-PCR (Supplementary Table S3). To further investigate whether OI-29 is associated with virulence, we constructed a ΔOI-29 mutant and assessed its ability to adhere to host epithelial cells. This mutant was found to similarly adhere to HeLa cells (Figure 1A) and Caco-2 intestinal epithelial cells (Figure 1B) as the parental strain, while ΔescC, a mutant of LEE genes and used as positive control, adhered at much lower levels (Figures 1A,B). Accordingly, transcripts of seven representative LEE genes (ler, escT, escC, escN, eae, tir, and espB) were similarly abundant between parental and ΔOI-29 strains (Figure 1C). These results suggest that OI-29 is not required for E. coli O157:H7 adherence and LEE gene expression.

Motility was repressed in the ΔOI-29 mutant compared with that in wild-type E. coli O157:H7, with growth radius after 10 h at 30°C on motility agar 1.86-fold smaller in the former than in the latter (Figures 2A,B). The motility defect was rescued by complementation with a low-copy plasmid (pACYC184) encoding OI-29 (Figures 2A,B). The motility of ΔOI-29 was not affected by introducing an empty pACYC184 into the mutant (Supplementary Figures S1A,B). The decreased motility of ΔOI-29 was also detected when the bacterium was grown at 37°C (Supplementary Figure S2). Of note, wild-type, ΔOI-29, and the complemented strain grew at similar rates in LB medium (Figure 2C), indicating that the decreased motility in ΔOI-29 was not due to slower growth. Strikingly, electron microscopy revealed that approximately 85% of wild-type and complemented cells (n = 500 cells per strain) possessed 1–3 flagella, while approximately 80% of ΔOI-29 cells (n = 500) were aflagellar. Representative transmission electron micrographs were shown in Figure 2D. These results suggested that loss of OI-29 represses flagellar biosynthesis.

Hence, expression of the flagellar genes fliC (encoding the major flagellin) and motA (encoding the flagella motor) was analyzed by qRT-PCR in both the parental strain and the ΔOI-29 mutant. Deletion of OI-29 was found to decrease the abundance of fliC and motA transcripts by fourfold and threefold, respectively (Figure 2E). The transcription of flagellar genes in ΔOI-29 was not affected by introducing an empty pACYC184 into the mutant (Supplementary Figure S1C). Western blotting also revealed that FliC expression was significantly decreased in the mutant (Figure 2F). Transcription and translation were restored to wild-type levels upon complementation with a functional copy of ΔOI-29 (Figures 2E,F), confirming that OI-29 enhances E. coli O157:H7 motility by upregulating flagellar synthesis.

OI-29 in E. coli O157:H7 is a 2,643-bp island that contains z0638, z0639 (named gmrA), and z0640, which encode hypothetical proteins of unknown function (Supplementary Figure S3A). Notably, the gmrA gene product contains a putative DNA-binding domain (Supplementary Figure S3B), and therefore may act as a transcriptional regulator. Accordingly, the growth radius of ΔOI-29 on motility plates was restored to wild-type levels when complemented with gmrA but not with z0638 and z0640 (Supplementary Figure S4), suggesting that the motility defect is due only to gmrA. A ΔgmrA mutant was also constructed, and was found to have significantly reduced motility on semi-solid LB agar and flagellar biosynthesis, as well as suppressed transcriptional and translational expression of flagellar genes in comparison to wild-type (Supplementary Figures S1D, S5). These defects were comparable to those in ΔOI-29, and were restored to wild-type levels when complemented with a low-copy plasmid carrying gmrA (Supplementary Figure S5). On the other hand, the growth radius and transcriptional expression of fliC and motA were not affected by deletion of z0638 and z0640 (Supplementary Figure S6), implying that these genes do not regulate bacterial motility. Collectively, these results confirm that GmrA is an activator of motility and flagellar synthesis in E. coli O157:H7.

As expression of flagellar genes is directly controlled by FlhDC and FliA, we tested whether GmrA interacts with either or both. We found that fliA transcripts were significantly less abundant in the ΔgmrA mutant, a defect rescued by a low-copy plasmid carrying gmrA (Figure 3A). In contrast, expression of flhD and flhC was comparable among E. coli O157:H7 wild-type, ΔgmrA, and complemented strain (Figure 3A). Electrophoretic mobility shift and competition assays suggested that GmrA binds specifically to the fliA promoter in vitro, but not to the flhDC promoter and rpoS (negative control) (Figure 3B). ChIP-qPCR also showed that the fliA promoter was enriched 7.75-fold in GmrA-ChIP samples than in mock-ChIP control samples, confirming that GmrA binds to the fliA promoter in vivo (Figure 3C). However, flhDC promoter and rpoS were not enriched in GmrA-ChIP samples (Figure 3C). These results suggest that GmrA binds directly and specifically to the fliA promoter to upregulate transcription.

Motility and expression of fliC and motA were significantly reduced in the ΔfliA mutant compared with that in wild-type (Figures 3D–F), confirming that FliA is a positive regulator of bacterial motility and flagellar synthesis. We also found that the growth radius and abundance of fliC and motA transcripts were comparable between ΔfliA and the double mutant ΔfliAΔgmrA (Figures 3D–F), viz., deletion of gmrA has no impact on bacterial motility and flagellar gene expression in the ΔfliA background. In addition, both bacterial motility and flagellar gene expression were restored to wild-type levels when an inducible plasmid carrying fliA was introduced into the ΔgmrA mutant (Figures 3D–F). Collectively, the data indicate that GmrA regulates E. coli O157:H7 motility and flagella synthesis through fliA.

Expression of gmrA in E. coli O157:H7 grown to exponential phase or stationary phase in LB or DMEM (the tissue culture medium used for adherence assays) was examined by qRT-PCR. While the transcript level of gmrA was higher in LB-grown than the level in DMEM-grown E. coli O157:H7 under the same growth stage (Figure 4A), gmrA expressed higher in exponential phase than stationary phase E. coli O157:H7 grown in either medium (Figure 4A). To determine optimal conditions for gmrA expression, E. coli O157:H7 was grown in LB or DMEM at different temperatures, pH and osmolarity to exponential phase for qRT-PCR analysis. In either LB or DMEM, the optimal temperature for gmrA expression is 37°C, with transcript levels slightly reduced at 33 or 39°C, but significantly diminished at 30 or 42°C (Figure 4B). Buffering the medium at pH 7.0 elicited the highest levels of expression, while an increase or decrease in pH greatly reduced transcripts (Figure 4C). Expression was also maximal at physiological osmolarity, such that increased osmolarity from NaCl or KCl also significantly reduced transcription (Figures 4D,E). These results suggest that gmrA expression is maximal in conditions similar to those in the intestinal tract. However, gmrA expression was not affected by the presence of bile salts and sodium bicarbonate, which are predominantly found in small intestine and common signals for virulence gene regulation (Supplementary Figure S7). Remarkably, expression was strongly induced in the first 2 h of co-incubating E. coli O157:H7 with Caco-2 or HeLa epithelial cells, but was significantly repressed 3 h post-infection (Figure 4F), indicating that GmrA regulates motility and flagella synthesis mainly at the initial phase of infection. The expression of gmrA was at similar levels during the course of 6 h growth in DMEM in the absence of host epithelial cells (Figure 4F), indicating the induction of gmrA in the first 2 h and repression 3 h onwards is dependent on the presence of host cells.

Bioinformatics analysis of 231 available E. coli genome sequence showed that OI-29 and gmrA are highly conserved and widely distributed in various E. coli lineages. Phylogenetic analysis also revealed that E. coli with OI-29 fall predominantly into four distinct clades, consisting of all pathogenic strains with only one exception (64 in total) (Supplementary Figure S8 and Supplementary Table S4). Clade 1 contains enteropathogenic E. coli O55:H7 and EHEC strains O157:H7 and O145:H28. Other three clades contain neonatal meningitis-associated E. coli strains CE10, IHE3034, RS218, and S88; uropathogenic E. coli strains IAI39, MS6198, and PMV-1; extraintestinal pathogenic E. coli strains UTI89, UMN026, PCN033, and PPECC42; avian pathogenic E. coli strains O1 and IMT5155; enteroaggregative E. coli 042; adherent invasive E. coli UM146; and other clinical isolate E. coli strains (Supplementary Figure S8 and Supplementary Table S4). To investigate whether orthologous gmrA genes also regulate motility, we deleted such genes from eight representative strains, consisting of two strains each of O157:H7 and O55:H7 and one strain each of O127:H6, avian O2 and O2:H8, and neonatal meningitis-associated O18. The growth radius on motility plates and expression of fliC in these mutants were significantly decreased compared with the corresponding wild-type strains (Figure 5), confirming that GmrA is a widespread regulator of E. coli motility and flagellar synthesis.