CFA/I-positive ETEC H10407 was obtained from the global collection deposited at Gothenburg University, Sweden. This isolate was used for the construction of a transposon library and the mutants developed in this study. The CFs and enterotoxin profiles of H10407 were confirmed using previously described multiplex PCR [19]. The expression of CFA/I was also evaluated using a previously described colony blot assay [20] and inhibitory ELISA [21].

A derivative of the transposon EZ-Tn5 (Epicentre) was phosphorylated with polynucleotide kinase (New England BioLabs) and incubated with EZ-Tn5 transposases (Epicentre) at 37 °C for 1 h to prepare the transposome that was stored at −20 °C until use. An overnight culture of H10407 was diluted 1:100 in LB broth and the bacteria were allowed to grow to an OD₆₀₀ of 0.3–0.4. The bacterial cells were then harvested, washed three times in 0.5 volume of 10 % chilled glycerol and re-suspended in 1:1000 of the volume of the initial culture. Then, 40 µl of freshly prepared H10407 electrocompetent cells was mixed with 0.3–0.5 µl of transposomes and electrotransformed in a 0.2 cm cuvette (Bio-Rad Laboratories) using bacteria pre-setting parameters (2.5 kV, 25 µF and 200 Ω) in the Gene Pulser Xcell Electroporation System (Bio-Rad Laboratories). Cells were immediately re-suspended in 1 ml of SOC medium (Invitrogen) and incubated at 37 °C for 2 h before being spread on LB agar supplemented with kanamycin (20 µg ml⁻¹). Individual colonies from 20 transformation batches were counted, scraped from plates using LB, and pooled to construct an initial transposon library consisting of ~1.1 million mutants of H10407. The library was aliquoted and kept in 25–30 % glycerol stock at −80 °C.

Genomic DNA (gDNA) of the original (input) and screened (output) mutant pools was extracted using a Qiagen Genomic-tip 100/G (Qiagen) and used to prepare TraDIS sequencing libraries as described previously [22]. Briefly, ~2 µg of gDNA was fragmented to an insert size of ~300 bp using a Covaris Focused-ultrasonicator (Covaris). The fragmented DNA was end repaired, A-tailed and adapter-ligated using an Illumina DNA fragment library preparation kit (New England BioLabs) according to the manufacturer’s instructions. The libraries were then enriched by 10–20 cycles of PCR using a transposon-specific forward primer and a reverse primer that included Illumina flow cell binding sites [22]. The enriched libraries were purified using Agencourt AMPure XP beads (Beckman Coulter) and quantified on an Agilent DNA1000 chip (Agilent Technologies) following the manufacturer’s instructions. The peak size of the libraries was checked on a Bioanalyzer (Agilent Technologies) and the concentration of the libraries was quantified by quantitative PCR (qPCR). The libraries were sequenced on a HiSeq2000 Illumina platform as single-end reads using a specialized recipe. The sequencing was conducted at the KAUST Bioscience Core Laboratory.

The reads were processed using the Bio::Tradis toolkit [23]. Briefly, reads were filtered for 10 bases matching the expected transposon tag sequence. Filtered reads were then mapped against H10407 reference chromosomal (accession number FN649414) and accessory genomes (accession numbers: FN649415, FN649416, FN649417 and FN649418) using SMALT version 0.7.2, and transposon insertion sites and insertion indices (obtained by dividing the number of unique insertion sites by the gene length) were determined for every gene. The essentiality of every gene in the chromosome and accessory genomes of H10407 was determined as described previously [24]. Gamma distributions were fitted to the two modes observed in the insertion index distribution, and Log2 likelihood ratios (LLRs) were calculated between the fitted distributions to classify genes as essential or non-essential, with cutoffs set at LLRs of less than −2 or greater than 2, respectively. For the calculation of fold change, Log2 fold changes and significance of changes in mutant abundance before (input) and after selection (output; biological replicates of mutant pool screened under the prescribed condition) were calculated using edgeR [25] after normalization using the TMM method [26].

For global transcriptomic analysis, the prototype CFA/I-positive strain H10407 was grown under different in vitro conditions to reach the early exponential growth phase (OD₆₀₀ of 0.4). Briefly, H10407 was cultured aerobically in regular LB media (pH 7) (or adjusted to pH of 5 or 9 using appropriate buffers) at 37 °C with shaking. For anaerobic culturing, H10407 was grown overnight in 150 ml of regular LB media at 37 °C in a 200 ml conical flask fitted with a permeable filter in the flask lid (to allow gaseous exchange) and shaken in an anaerobic chamber (Whitley A35 workstation). One millilitre of this culture was added to 100 ml of LB media in a 200 ml conical flask and shaken at 37 °C in the anaerobic cabinet.

For water survival experiments, H10407 that was grown aerobically in regular LB media was washed once before being suspended in water to a final concentration of 1×10⁶ c.f.u. ml⁻¹ and incubated at ambient temperature (22–25 °C) without shaking for 48 h. For each culturing and survival condition, duplicate independent biological replicates were used to extract bacterial RNA using the RNeasy Mini Kit (Qiagen) according to the manufacturer’s instructions. rRNA was depleted using Invitrogen’s RiboMinus Prokaryotic Kit (Invitrogen) and contaminating DNA was removed using RNase-free DNase (Invitrogen). The cDNA libraries were prepared using an Illumina TruSeq RNA sample preparation kit, pooled in equimolar concentrations and sequenced on an Illumina HiSeq2000 platform at the KAUST Bioscience Core Laboratory.

For each sample, the reads were mapped against H10407 reference chromosomal (accession number FN649414) and accessory genomes (accession numbers: FN649415, FN649416, FN649417 and FN649418) using BWA v0.7.12 [27] to produce a BAM file using default parameters but with the quality threshold for read trimming set to 15 (q=15) and maximum insert size of 300 bp. Gene expression values were computed from the read alignments to the coding sequencing to generate the number of reads mapping and RPKM (reads per kilobase per million) [28]. Only reads with a mapping quality score of 10 were included in the count.

For the purpose of clustering, read counts were converted into transcripts per million (TPM) [29], and averaged across replicates. Genes that were highly expressed in at least one condition (TPM>100) were considered for further analysis, along with a small selection of biologically relevant genes that did not meet this stringent threshold. Hierarchical clustering was performed on log-transformed TPM values using the base R function hclust with default parameters and visualized using heatmap 2.

The Bio-RNASeq pipeline, which makes use of the DESeq Bioconductor R package [30], was used to perform the differential gene expression analysis between pairs of conditions (https://github.com/sanger-pathogens/Bio-RNASeq). The differential gene expression was visualized using heatmapper [31].

H10407 mutant derivatives harbouring targeted independent deletions in the chromosomal (atpB, atpC, atpD, galU, gmhD, waaC, waaF and waaO) and plasmid genes (aatC, cexE and cfaE) were constructed by replacing the wild type gene with a DNA cassette encoding a kanamycin gene using the λ–Red recombination method [32].

To understand the basis of ETEC persistence in the gut of mice, the transposon input library was used to orally inoculate a group of six streptomycin-treated C57BL/6 mice (5–6 weeks old). Briefly, the mice were orally inoculated with 20 mg of streptomycin 24 h before being infected with the mutant library at a dose of 10¹⁰ c.f.u. per mouse. The caeca of the infected mice were harvested 48 h post-infection, washed with PBS, homogenized and plated on LB agar supplemented with appropriate antibiotics.

Competitive infection assays were used to examine the impaired capability of null mutant derivatives in aatC, cexE and cfaE to colonize the caecum of streptomycin-treated mice. For each mutant, a group of five C57BL/6 mice (5–6 weeks old) were infected by oral gavage with an approximately 1:1 mixture of mutants and isogenic parent strains at a dose of approximately 1×10¹⁰ c.f.u. per mouse. The caeca of the infected mice were harvested 48 h post-infection, washed with sterile PBS and homogenized. Faecal pellets were also collected and homogenized in sterile PBS. Serial dilutions of homogenized caecum, caecal contents (wash) and faecal pellets were plated on LB agar supplemented with appropriate antibiotics and enumerated.

The CFA/I-positive ETEC H10407 was transformed with DNA encoding Tn5 transposomes and transformants were selected on LB plates supplemented with kanamycin. Individual colonies from different transformation batches were pooled to construct an initial transposon library of approximately 1.1 million mutants. DNA from two independent samples was prepared and these were sequenced from the transposon ends using the TraDIS method (Fig. S1, available in the online version of this article). High-quality sequencing reads that contained the 10 base transposon tag sequence were then mapped onto the ETEC H10407 genome (FN649414; 5 153 435 bp) including the associated plasmids p52 (FN649415; 5175 bp), p58 (FN649416; 5800 bp), p666 (FN649417; 66 681 bp) and p948 (FN649418; 94 797 bp) in order to simultaneously identify all of the transposon insertion sites.

The transposon unique insertion (UI) sites were densely and randomly distributed throughout the chromosome of H10407 and associated plasmids (Fig. 1). Approximately 520 000, 5000 and 22 000 UIs were identified in the chromosome of H10407 and plasmids p948 and p666, respectively. This resulted in an average of one insertion every 9.9, 18 and 3 bp of H10407 chromosome, plasmids p948 and p666, respectively, which confirms the generation of a highly saturated transposon library (Table S1). To identify the essential genes of CFA/I-positive ETEC H10407, an insertion index was calculated for each gene, normalizing the number of insertions to gene length. Statistical analysis of the bimodal distribution of insertion indices (fitted to a gamma distribution) identified 425 chromosomal genes as candidates essential for H10407 (Table S2). Of these, 247 genes have previously been predicted by TraDIS to be essential in both E. coli K12 BW25113 [33] and the uropathogenic E. coli EC958 [34]. Kyoto Encyclopedia of Genes and Genomes (KEGG) analysis demonstrated that all of the genes that were deemed essential for H10407 have previously been found to be involved in key pathways and vital cellular processes (Table S2).

The RNA expression profiles of H10407 cultured under different conditions were next determined, including in water, a common source of ETEC infection and persistence in the environment. For each gene, TPM values across duplicates were averaged; log TPM values were calculated and these were used to perform an unsupervised hierarchical clustering (Fig. 2). A relatively conserved gene expression profile was detected when E. coli H10407 was cultured under different in vitro conditions including aerobic [neutral (pH 7), alkaline (pH 9) and acidic media (pH 5)] and anaerobic growth. In contrast, many of these genes were significantly downregulated when the bacteria were placed in water for 48 h (see Methods for conditions) (Fig. 2). These included genes associated with cell division (n=16), translation (protein biosynthesis), production of cellular energy/ATP (n=16) and ribosomal subunits (n=56). This analysis indicates that H10407 uses different genes for water survival compared to growth in richer medium.

In order to identify H10407 genes that facilitate the survival of H10407 in water, the transposon library consisting of 10⁶ mutants was suspended in sterile water at a density of approximately 10¹⁰ c.f.u. ml⁻¹ (input) and incubated at ambient temperature (22–25 °C) without shaking for 5 weeks. At this time point, duplicates of water samples were collected and centrifuged to recover bacterial cells (output). gDNA was isolated from the input and output samples and subjected to TraDIS analysis. In this regard, a comparison was made between input and 5 week screened libraries to identify the genes required for long-term survival in water.

A total of 56 genes with at least a two-fold decrease in the number of transposon insertions per gene in output versus input pools (log₂ equal to or lower than −2, P<0.05 and false discovery rate of 5 %) were identified as candidates for the long-term survival of H10407 in water (Table S3). The majority of genes identified as essential candidates for the long-term survival of H10407 in water had 1.3–7 log₂ decrease in the unique transposon insertions identified per gene in 5 week screened libraries as compared to 48 h screened libraries. This highlighted the essentiality of these candidate genes for the survival and persistence of H10407 in water.

KEGG analysis demonstrated that many of the candidate water survival genes fell into distinct pathways, including lipopolysaccharide biosynthesis (n=12; waaC, waaF, waaG, waaL, waaO, waaP, waaR, waaV, gmhA, gmhC, gmhB and gmhD), energy metabolism, oxidative phosphorylation (n=8; atpC, atpD, atpG, atpA, atpH, atpF, atpE and atpB), biosynthesis of antibiotics (n=5; galU, rfbA, rfbB, pgi and fbp), microbial metabolism in diverse environments (n=2; pgi and fbp), biofilm formation (n=2; rpoS and dksA), and peptidoglycan biosynthesis and degradation (n=2; ldcA and lpp) (Table S4).

The screen also identified four genes implicated in transcription machinery (rpoS, rseA, rseB and dksA). rpoS codes for the major alternative sigma factor 38 (σ³⁸) and regulates stationary phase transcription under stress conditions [35–37]. Additionally, rseA and rseB that code for σ^E-negative regulator proteins and dksA encoding a suppressor protein (transcription factor), which have been shown to be key determinants of the E. coli stringent response (i.e. sensing and responding to nutritional status), were identified [38, 39].

The majority of these candidate water survival genes (n=41) were significantly downregulated under water stress compared with aerobic growth in LB (Tables S3 and S5). These include the genes associated with lipopolysaccharide biosynthesis, energy metabolism (oxidative phosphorylation), transcription machinery and biofilm formation. The expression analysis also revealed differential expression of genes associated with flagellar assembly, chemotaxis, motility and fimbrial formation. The majority of fimbrial (n=26), phage (n=94), chemotaxis (n=6; cheZ, cheB, cheR, tap, tar and motA), curli assembly (n=5; csgA, csgB, csgD, csgF and csgG) and two CRISPR genes (ETEC_2945 and ETEC_2951) were relatively upregulated under water stress compared with aerobic growth (Table S6). In contrast, genes associated with flagellar assembly including those coding for flagellar hook (flgD and flgE), distal (flgG) and proximal rods (flgB, flgC and flgF) were significantly downregulated (Table S5). The analyses also identified differential expression in the chemotaxis genes cheZ and cheY, which interact with the E. coli flagellar motor switch complex and thus control the direction of flagella rotation, (impacting smooth swimming – tumbling) [40–43]. Both cheY and cheZ were upregulated under anaerobic growth while only cheZ was upregulated under water stress, compared to aerobic growth (Tables S6 and S7). Previous studies showed that motile behaviour of E. coli is a random motion of straight swimming (anticlockwise rotation of flagella) interrupted by tumbling (clockwise rotation of flagella induced by phosphorylated CheY) [40, 41]. Recently, an elevated level of CheZ (which increases the dephosphorylation of CheY-P) has been associated with low tumbling bias of swarming E. coli [44, 45]. This suggests that under water survival H10407 can perform high-speed swimming through altering the run tumble bias.

In order to confirm if genes identified by the TraDIS screen influenced the survival of H10407 in water, we constructed null mutant derivatives in eight genes and tested their viability in water. The genes gmhD, waaF, waaC, waaO (lipopolysaccharide biosynthesis), atpC, atpD, atpB (F₀F₁ ATPase involved in oxidative phosphorylation) and galU were inactivated in H10407 with a DNA cassette encoding the kan gene, conferring kanamycin resistance. At day 2 post-water incubation, the numbers of c.f.u. recovered from the mutants were slightly lower than that recovered from the wild type parent strain. All the mutants examined exhibited an impaired ability to survive in water following incubation for 5 weeks (Fig. 3). In each case, the c.f.u. recovered from mutant screening were significantly lower (1–2.5 log₁₀ decrease) than those recovered from the wild type parent equivalent derivative, confirming the TraDIS data (Fig. 3).

Interestingly, distinct expression patterns were identified between the genes encoded by plasmids p948 (that encodes ETEC-associated virulence genes, including the cfaABCE operon and heat-stable enterotoxin st) and p666 (a self-transmissible plasmid that encodes the heat-labile enterotoxin elt) [46] under anaerobic growth and water stress compared with aerobic growth. The majority of p666 genes, including those associated with conjugal transfer, were significantly upregulated under water stress, whereas the majority of p948 genes were significantly upregulated under anaerobic growth (Fig. 4). Anaerobic growth relatively upregulated the expression of many plasmid-encoded, virulence-associated genes, including the cfaABCE operon and eatA, encoding a mucin-degrading serine protease. This is in addition to genes associated with plasmid stability and maintenance (replication, partitioning and segregation) (Fig. 4). Differential transcriptional profiles associated with the anaerobic growth of H10407 are provided in Tables S7 and S8.

We were interested to see if genes associated with survival in water were involved in colonization of and persistence in the mammalian host. Consequently a streptomycin pretreatment model was used to explore the ability of E. coli H10407 and associated mutants to colonize mice. To this end, 5–6-week-old C57BL/6 mice were orally inoculated with 20 mg of streptomycin 24 h before infection to disrupt the resident microbiota. These mice were then challenged, by oral gavage, with wild type H10407 at a dose of ~10¹⁰ c.f.u. per mouse. On average, 2×10⁶ and 10⁵ c.f.u. g⁻¹ were recovered from the faecal pellets collected from infected mice 2 and 4 days post-challenge, respectively. The colonizing capability of H10407 to colonize the mouse caecum, colon and small intestine was evaluated on days 1, 2 and 4 post-challenge. At all time points, high c.f.u. values were recovered from the caecum followed by colon and small intestine (Fig. S2). All mice survived the infection and, as expected, showed limited signs of pathology and other infection-associated phenotypes.

The H10407 mutant library was used to orally inoculate mice, and the output pools of bacteria that were capable of survival in the caeca of mice for 48 h post-infection were recovered from homogenized caeca. DNA from six independent output pools was prepared, pooled and subjected to massively parallel sequencing of transposon flanking regions. The bacterial load in mice challenged with the H10407 mutant pool was comparable to that in mice infected with the wild type H10407 challenge (data not shown). However, there was significant loss of the diversity of the mutant pool (complexity of H10407 mutant output pool) (~36 000, 189 and 2458 UIs were identified in the chromosome of H10407 and plasmids p948 and p666, respectively), compromising unbiased TraDIS analysis. An average of a unique insertion was identified every 153, 890 and 29 bp of the H10407 chromosome, and plasmids p948 and p666, respectively (Table S1). This indicated that the loss of pool genetic diversity was random in nature, which precluded TraDIS analysis of H10407 persitence in the mouse gut.

A selected gene targeting approach was used to determine if introducing specific mutations associated with ETEC pathogenicity would compromise mouse colonization. To this end, E. coli H10407 null mutant derivatives in plasmid genes aatC, cexE (coding for non-fimbrial proteins whose implication for pathogenesis are unknown) and cfaE (coding for CFA/I minor pillin) that were upregulated under anaerobic growth were constructed and used to orally inoculate a group of streptomycin-treated C57BL/6 mice. Competitive infection assays were used to examine the ability of these mutants to colonize the caeca of mice for 48 h post-infection. These assays demonstrated that all mutants tested were significantly attenuated, characterized by an impaired capability to colonize the caecum of streptomycin-treated mice (Fig. 5). For all mutants examined, the output ratio (mutant:isogenic parent) recovered from the caeca (Fig. 5a), caecum contents (Fig. 5b) and faecal pellets (Fig. 5c) was significantly lower than that present in the inocula used for infection, confirming a role for these genes for H10407 colonization. Interestingly, cfaE mutants of H10407 were attenuated in mice. This is consistent with published data that have suggested that the minor pilin CfaE is one of the major virulence factors that mediate colonization of the host intestine by CFA/I-positive ETEC [47, 48].