Genome sequence of adherent-invasive Escherichia coli and comparative genomic analysis with other E. coli pathotypes

1Crohn's Disease (CD) is a chronic inflammatory bowel disease of the intestinal tract characterized by a strong activation of the intestinal immune system. A complex interaction of genetic, immunologic, and environmental factors contribute to the immunopathology of CD but despite intensive investigation over the last half-century, a unifying etiology of inflammatory bowel diseases (IBD) has not been uncovered. Abundant clinical and experimental data implicate luminal bacteria or bacterial products in both the initiation and perpetuation of chronic intestinal inflammation. Some pathological manifestations observed in CD, including ulcers of the mucosa, mural abscesses and macrophage recruitment and activation, also occur in well-recognized infectious diseases caused by Shigella, Salmonella and Yersinia, in which invasion into mucosal epithelial cells is an important virulence trait. However, a growing body of evidence indicates that the balance between host defence responses and the commensal microbiota plays a key role in the pathogenesis of IBD. Patients with CD display an increased number of coliforms in their feces, particularly during periods of active disease and E. coli antigens are found in most intestinal resection specimens from these patients. Furthermore, it has been shown that early and chronic ileal lesions of CD patients harbour high levels of E. coli that might participate in disease pathogenesis. E. coli strains isolated from the ileal lesions of CD patients can exhibit adherent and invasive capabilities in both gastrointestinal epithelial cells and macrophages, a phenotype that was the basis for a new pathogenic group called adherent and invasive E. coli (AIEC). AIEC are enriched in ileal lesions in human CD and are associated with expression of proinflammatory cytokines and inflammation in mice expressing human carcinoembryonic antigen-related cell adhesion molecule (CEACAM) receptors. The predominance of AIEC in human CD patients, in conjunction with a growing body of biological and animal model data has generated intense interest into the possible role of AIEC in the initiation or maintenance of chronic inflammation associated with CD.

2We previously reported on a clinical AIEC isolate with serotype O83:H1 (strain NRG857c) that was isolated from the terminal ileum of a patient with CD. NRG857c belongs to the same serogroup as the historical AIEC isolate called LF82 first described over a decade ago for which much of the experimental data on AIEC phenotypes have been documented. AIEC do not harbour common virulence factors found in various other pathogenic E. coli, and so the genetic basis for their invasive phenotype, proinflammatory nature and association with CD are not fully understood. Here, we report the complete genome sequence of AIEC NRG857c that includes a 150-kb plasmid. We found that AIEC are closely related to a group of extraintestinal pathogenic E. coli (ExPEC) associated with urinary tract infections and neonatal meningitis, a finding that confirms and extends previous work. The comparison of this genome with other ExPEC, enteropathogenic E. coli, AIEC LF82, and commensal E. coli facilitated the identification of 41 high-confidence genomic islands and 66 genes unique to E. coli displaying the adherent and invasive phenotype.

3AIEC strain NRG857c was shotgun sequenced to 40-fold coverage using pyrosequencing. Assembly of the raw sequence data generated 48 contiguous regions (contigs) greater than 2-kb with a total size of 4.84-Mb. Contigs were assembled by aligning the larger contigs to an optical restriction map using MapSolver and by BLASTX analysis of contigs ends. The majority of gaps between contigs were identified because contigs ends were syntenic with single-copy genes in previously sequenced E. coli genomes. PCR primers were designed to amplify across these gaps followed by sequencing to generate "super-contigs" (see Additional File 1, Figure S1). Final gap closure was achieved after incorporation of sequence data for the seven ribosomal RNA operons. Plasmid contigs were identified by BLASTX analysis. Gap closure for the plasmid was done using BLASTN analysis of the terminal sequences from which PCR primers were designed. Amplification and sequencing of these regions resulted in the assembly, but not closure, of a single plasmid contig.

4The chromosome of NRG857c is 4,747,819 bp (50.68% G + C content), encoding 4,431 genes (Figure 1, Table 1). The plasmid is 147,060 bp (50.92 G+C content) and encodes 155 genes (Table 1). The sequence of both the NRG857c chromosome and plasmid has been deposited in GenBank [GenBank: CP001855, GenBank: CP001856].

5The phylogeny of AIEC NRG857c was resolved in two ways. First, a phylogenetic tree based on the optical map data was constructed using the unweighted pair group method with arithmetic mean (UPGMA) along with the in silico derived NcoI fragments for other sequenced E. coli strains (Figure 2A). The second method involved multi-locus sequence typing (MLST) with seven housekeeping genes as described previously (Figure 2B; Additional File 2, Table S1), followed by comparison to sequences from other strains. In both analyses NRG857c clustered with avian pathogenic E. coli (APEC-O1), and the uropathogenic E. coli isolates 536 and CFT073. Also in this group was LF82, another AIEC strain of the same serotype as NRG857c (O83:H1) whose genome sequence was retrieved from Genoscope (http://www.genoscope.cns.fr see note added in revision). LF82 shows high sequence similarity to our strain as analyzed by MapSolver (Additional File 3, Figure S2), by BLASTN analysis (Figure 1), and by phylogenetic analysis (Figure 2).

6A general comparison of the total genome content of NRG857c with several other E. coli pathotypes is shown in Table 1. The majority of human ExPEC belong to phylogenetic group B2 and are categorized based on their clinical spectrum of disease, including urinary tract infections (UPEC) and neonatal meningitis (NMEC). AIEC strains cluster genetically with ExPEC and share some of their phenotypic traits including the ability to colonize mucosal epithelial cells, invade eukaryotic host cells, and to induce inflammatory responses in host animals. Although the prototype EPEC strain E2348/69 (serotype O127:H6) and other EPEC strains belong to the same phylogenetic group as the ExPEC strains, they are not generally considered to be invasive organisms. However, recent data suggests that at least two type III secreted proteins (EspT and EspF) can facilitate EPEC invasion into non-phagocytic cells and may define a new category of invasive EPEC.

7Genomic islands (GI) comprise a horizontally acquired flexible gene pool that is a major driver in evolution and niche specialization of pathogenic bacteria. Recent computational methods that take advantage of genetic signatures indicative of horizontal gene transfer enable the high-confidence prediction of GIs in annotated bacterial genomes. To identify putative genomic islands in NRG857c, we used IslandViewer, which uses three independent methods for island prediction, IslandPick, IslandPath-DIMOB and SIGI-HMM. Using the methods and established thresholds described previously, we identified 35 genomic islands (GI-1 to GI-35) on the NRG857c chromosome ranging from 4 to 25-kb, with G+C content differing significantly from genome mean and with poor conservation among the other non-AIEC pathotypes shown in Figure 1 (see Additional File 4, Table S2 for full list of genomic islands and gene content analysis). We limited our comparative analysis here to the strains most related to NRG857c and to two well-described E. coli strains of commensal and pathogenic nature. The conservation of these 35 islands between NRG8578c and LF82 was high, suggesting that they may encode traits unique to the adherent and invasive phenotype. Five of the genomic islands (GI-6, -7, -8, -10 and -16) code for defective prophages, three (GI-14, -22, -29) are fimbrial islands, and three (GI-20, -26 and -30) appear to be involved in lipopolysaccharide or capsular polysaccharide biosynthesis. GI-23 is noteworthy because it encodes an EmrKY-TolC multidrug resistance efflux pump and the sensor kinase, EvgA, involved in acid resistance and multidrug resistance in E. coli. GI-15 and GI-19 appear to be metabolic islands involved in the transport and metabolism of various sugars. An additional six genomic islands were identified on the large plasmid (PI-1 to PI-6 in Figure 3) (see Additional File 4, Table S2 for full list of plasmid islands and gene content analysis).

8To date, restriction profiles or other biased analyses such as pulse field gel electrophoresis (PFGE), MLST or typing for known virulence genes common to intestinal pathogenic E. coli have failed to uncover unique genetic determinants implicated in the AIEC phenotype. To begin to identify single genetic determinants unique to AIEC, we carried out whole-genome comparisons between NRG857c, LF82, and 29 other non-AIEC genomes of E. coli. NRG857c and LF82 show considerable sequence similarity and synteny (Additional File 3, Figure S2) with 46 chromosomal genes unique to NRG857c and 10 chromosomal genes unique to LF82 (see Additional File 5, Table S3 for full list of genes unique to AIEC). The large plasmids from NRG857c and LF82 show almost no conservation between them (see below), suggesting that they have different ancestry.

9Panseq, a Web-based tool designed to analyse the "pan-genome" of closely-related genome sequences, was used to identify genes common to AIEC strains NRG857c and LF82, but absent in other members of this phylogenetic cluster (i.e. APEC-O1, 536, and CFT073). We programmed Panseq to find unique sequences of at least 2-kb present in NRG857c and LF82 but absent in APEC-O1, 536 and CFT073. In this analysis, we found 21 sequences with a combined length of 155-kb that are unique to AIEC strains. Several of these sequences code for prophage elements including a 19.7-kb region encoding the morphogenesis and packaging modules of a P22-like prophage (NRG857_04720 - NRG857_04815). A second interesting region of 47.2-kb extends, with one interruption, from NRG857_09990 to NRG857_10240 and codes for several proteins involved in intermediary metabolism including transport of propanol/propanediol and galactitol. BLASTN analysis of this region revealed two sub-regions, one 20.3-kb and the other 4.4-kb, which are not found in the complete genome sequence of any other E. coli strain. The latter region shows 71% sequence coverage to a region from the complete genome of Citrobacter rodentium ICC168, while approximately half of the longer sequence is also found in an uncharacterized E. coli strain ATCC 8739. This 10.7-kb region has no nucleotide similarity with any other fully sequenced bacterium. BLASTX revealed similarity in this region to two hypothetical Vibrio coralliilyticus ATCC BAA-450 proteins [GenBank: ZP_05883689, GenBank: ZP_05883688] adjacent to orthologs in Burkholderia cenocepacia HI2424 [GenBank: YP_833853, GenBank: YP_833854], which are described as hypothetical proteins.

10The 150-kb plasmid in NRG857c is different from the plasmid found in LF82. Whereas plasmid pNRG857c shows significant regions of identity to plasmids in other seropathotypes of E. coli, the 110-kb plasmid of strain LF82 (pLF82) has very little similarity to pNRG857c or pAPEC-O1 (APEC-O1), pColBM (APEC-O103), pUTI189 (UPEC UTI189) and pO157 Sakai (EHEC O157:H7) (Figure 4). The extrachromosomal plasmid in NRG857c is a antimicrobial resistance plasmid with a suite of genes encoding resistance to aminoglycosides, β-lactams, chloramphenicol, mercury, quaternary ammonium salts, sulfonamides, tetracycline, and trimethoprim, several of which appear to be enclosed as transposon blocks. The plasmid may be capable of conjugal transfer as it encodes several tra genes, although we have not experimentally tested this. In addition, there are genes for colicins M and V production and immunity. The antibiotic resistance genes are clustered in three regions of the plasmid in PI-2, PI-3 and PI-4 (Figure 3B). The mercury resistance cassette is identical to IS5075 found in IncA/C2 plasmids pRYC103T24 [GenBank: GQ293500.1], pLEW517 [GenBank: DQ390455.1], NR1 [GenBank: DQ364638.1] and R100 [GenBank: AP000342.1]. The β-lactam-macrolide region is identical to sequences present in plasmid pTZ3721 [GenBank: AB020531.1] and pTZ3723 [GenBank: AB038654.1]. Also of interest to us were several genes involved in siderophore production and iron metabolism. Plasmid pNRG857c has the sitABCD operon that encodes proteins involved in the periplasmic and inner membrane transport of iron and manganese. Two outer membrane proteins (IutA and FepA) are also encoded by the plasmid and are involved in translocation of iron across the membrane. IutA (NRG857_30235) is the ferric-aerobactin receptor, while FepA (NRG857_30015) is an iron-enterobactin outer membrane transporter, both of which are involved in the tonB-dependent transport pathway for iron and also the OM receptor for the colicins. IutA and FepA are encoded on plasmids pAPEC-O103-ColBM, pAPEC-O1-ColBM, pCVM29188_146 (from Salmonella enterica serovar Kentucky,), pVM01 (from the APEC strain E3,), and pLVPK (from Klebsiella pneumoniae CG43,). Interestingly, the chromosome contains a FepA paralog (NRG857_02640). The presence of several iron-acquisition genes suggests that Fur regulation of these plasmid-encoded genes occur. As predicted, the consensus DNA sequence for Fur binding (WAATDRNWNYNAWTW) is found in the upstream regulatory region,) of the iroBCDE, sitABCD, iucABCD-iutA operons, and the shiF and fepA genes.

11The chromosome of AIEC strain NRG857c encodes a variety of potential virulence factors (Table 2). As mentioned above, the plasmid carries several potential virulence factors including genes for iron acquisition. This would suggest that the plasmid contributes to the overall virulence of this bacterium, however we have demonstrated previously that a plasmid-cured variant was still able to attach to and invade epithelial cells in vitro.

12We identified genes for a complete type VI secretion system (T6SS) that are associated with virulence in other invasive organisms (Table 3). T6SS are phage-related secretion systems found in many Gram-negative pathogens and are thought to be involved in supporting an intracellular lifestyle, although their distribution is not restricted to pathogenic bacteria. The T6SS in NRG857c is found in GI-2, a low GC region of the chromosome directly downstream from a tRNA which is a common integration site for mobile genetic elements. This T6SS island encodes the conserved core elements of the secretion apparatus, including the valine-glycine repeat protein G (VgrG/NRG857_01165), the ClpV ATPase (NRG_01105) and the hemolysin coregulated protein (Hcp/NRG857_01155) that is 100% identical to Hcp in APEC-O1 and the UPEC strains UT189 and 536. We also identified a second Hcp upstream of this conserved locus (NRG_01080) that is 100% identical to Hcp in E. coli S88 (O45:K1:H7) that causes neonatal meningitis, suggesting that this T6SS island is a mosaic with different ancestries. Other organisms, including Vibrio cholerae, have two hcp genes in different parts of the genome, which may impart different functionalities on the secretion apparatus. Whether the T6SS in AIEC facilitates intracellular survival and/or growth will require additional experimentation that we are currently pursuing.

13NRG857c contains genes that are important for adhesion and invasion of AIEC LF82, including nlp1, htrA, yfgL, and dsbA. The SPAAN program as well as BLASTP with relaxed stringency was used to identify and extensive list of additional predicted adhesins (Table 4). The majority of the fimbrial operons in NRG857c are found in other E. coli strains, with the exception of the long polar fimbriae (Lpf; NRG857-17915-17923), which might be important for tissue tropism. A second Auf fimbrial system with a potential role as a colonization factor is encoded by genes NRG857_16960 through _17005. Other potential mediators of invasion include a hemagglutinin/invasin (NRG857_17920 to _17923) and an Ibe invasin (NRG857_21885 to _21890). In previous work, the invasion of brain endothelial cells was found to be mediated by the Ibe invasin, and was located on a genomic island called GimA. The presence of GimA was almost exclusive to ExPEC strains of phylogroup B2, and we now show that ibe is also present in AIEC, suggesting it may be involved in invasive properties of certain strains.

14In mouse models of AIEC-induced colitis, inflammation requires type I pili expression by the bacterial cells, as no colitis is induced by ΔfimH mutant bacteria. Colitis in this model requires the expression of human CEACAM receptors by transgenic mice, suggesting that the type I pili of AIEC can induce a proinflammatory response via CEACAM receptors in the gut mucosa. In support of this, FimH, the adhesin tip protein, is necessary but not sufficient for adhesion of AIEC strain LF82 to Intestine-407 cells. Polymorphisms in the FimH sequence have been identified in E. coli isolated from IBD patients and healthy individuals. In particular, 7 amino acid variants are associated with E. coli from IBD tissue and 2 variants are associated with E. coli from healthy individuals. Interestingly, FimH in NRG857c contains two disease-associated amino acid variants (N91S, S99N, and none of the SNPs associated with healthy tissue (A48V, A140V). Whether or not these variants are associated with different inflammatory responses or subtle differences in adherence in vivo will be important areas for future work.

15NRG857c contains global transcriptional regulators including phoP-phoQ, envZ-ompR, slyA and the negative regulators hns, hha, and fis involved in genome architecture and transcriptional regulation. Although these transcriptional factors are common to many bacterial species, in most Gram-negative pathogens they coordinate transcription of virulence genes including secretion system, toxins, adhesins and flagellar biosynthesis machinery. With this completed genome sequence, functional genomics approaches are now possible to understand the regulons of these transcription factors and their roles in intracellular survival and growth of AIEC. Indeed, Fis levels in the cell have already been associated with regulating the adhesive properties of AIEC strain LF82.

16Iron acquisition is an essential virulence trait in other ExPEC and these systems are expressed during urinary tract infections in vivo. Since NRG875c had an abundance of iron uptake systems, we designed experiments to test the role of iron acquisition during infection. We made an aerobactin transport mutant by deletion of iutA and tested whether this iron transport system was important for intracellular survival and the ability to colonize animals. We found that the iutA mutant was able to synthesize but not transport aerobactin (Additional file 6, Table S4). To investigate the invasive properties of ΔiutA, we conducted standard gentamicin protection assays in J774.1 macrophage cells, which did not reveal a significant difference in the uptake at 2 h of the wild type and the iutA mutant (Figure 5A). However, by 4 h after infection and thereafter, the iutA mutant had a significant defect in intracellular survival and/or replication compared to wild type cells. To determine whether the transport of aerobactin was important for bacterial infection in vivo, streptomycin pre-treated mice were infected with wild type NRG857c and the isogenic iutA mutant as described previously for a Salmonella infection model. Wild type NRG857c was recovered in ~50-fold more abundance in the intestinal tissue compared to ΔiutA (Figure 5B).

17The two broad hypotheses accounting for the immunopathology of IBD, including deregulation of the intestinal immune system, and dysbiosis of the commensal microbiota, are likely not mutually exclusive. Both pathways could be operationalized at the same time and in response to known genetic and environmental triggers. Regarding the genetic correlates of the AIEC phenotype, our genome sequence and comparative analyses provide many testable hypotheses to uncover the adhesive, invasive, and proinflammatory nature of AIEC. The fact that the 35 genomic islands in NRG857c are, in many cases, highly orthologous in LF82 but weakly conserved or absent in other E. coli pathotypes and commensal organisms is suggestive that these genomic islands may have an influential role in the expression of the AIEC phenotype. It is also likely that evolved differences in gene expression, or regulatory evolution, has played a pivotal role in generating phenotypic diversity involved in pathogen-like behaviour of AIEC, as we have shown previously for another intracellular pathogen. Functional genomics studies enabled by this work will be forthcoming.

18Escherichia coli AIEC strain NRC857c was isolated from a biopsy of a Crohn's disease patient at the Charite Hospital, Germany. A mutant in aerobactin transport (designated RAA002) was created by disruption of the iutA gene using allelic exchange from a suicide plasmid as described previously. For preparation of genomic DNA, wild type NRC857c cells were grown on solid Luria-Bertani (LB) agar at 37°C. Genomic DNA was extracted from 10 mg of bacteria scraped from a plate using the BioRobot EZ1 with the EZ1 DNA kit (Qiagen, Hilden, Germany). For plasmid purification, bacteria were grown in 4 L of LB broth and plasmid was isolated using a Maxi-prep kit (Qiagen, Hilden, Germany) according to the manufacturer's instructions. Total genomic DNA was sequenced using a Genome Sequencer FLX System (454 Life Sciences, Branford, CT, USA) at the McGill University and Genome Quebec Innovation Centre (Montreal, QC, Canada).

19Phylogenetic determinations were performed by in silico MLST using seven housekeeping genes (aspC, clpX, fadD, icdA, lysP, mdh and uidA). Analysis was performed using the software package MEGA4 and the Neighbour-Joining method under the Tajima-Nei model. An optical map of NRG857c was generated using the restriction enzyme NcoI (OpGen Inc., Madison, WI) and used for contig ordering. Unweighted Pair-Group Method using Arithmetic averages (UPGMA) similarity clustering of the restriction fragments generated in the whole genome optical map of NRG857c with in silico maps of publicly available E. coli isolates was performed using MapSolver version 2.1.1 (OpGen Inc., Madison, WI).

20Outward facing primers annealing to adjacent contigs were designed using Primer3Plus, synthesized by SigmaGenosys (Oakville, ON, Canada) and used to amplify DNA of NRG857c using the Expand Long Template PCR system (Roche, Mannheim, Germany). PCR products were analysed on agarose gels, purified with a Montage PCR purification kit (Millipore, Billerica, MA, USA) and sequenced using Sanger sequencing (University of Guelph, ON, Canada). Finished sequence was assembled using SeqManPro (DNASTAR Inc., Madison, WI). For ribosomal RNA (rRNA) operons, primers were designed using the syntenic flanking sequences of each rRNA operon in the E. coli strain CFT073. These seven rDNA amplicons were sequenced using the flanking primers and specifically designed 16S (rrs) and 23S (rrl) primers based on sequence alignment with CFT073 rDNAs.

21The genome sequence was subjected to automated annotation using the NCBI Prokaryotic Genomes Automatic Annotation Pipeline with the resulting GenBank data incorporated into Kodon (Applied Maths Inc., Austin, TX) for manual curation. A protein database was constructed from 22 Escherichia coli genomes available in GenBank. All of the open reading frames of NRG857c predicted by Glimmer 3 were searched against the protein database using BLASTX running locally. The same comparison was performed using the LF82 nucleotide sequences. A script written with the BioPerl toolkit was used to parse the BLAST output files for sequences that did not have any matches, or sequences with only weak matches using the criteria: (E-value ≥ 0.01), or (Percent Identity < 50%), or (<50% of the query length was used in the BLAST alignment). The predicted ORFs of NRG857c were compared against those of strain LF82 to identify those unique to each strain. Additional comparative genomics analyses were carried out using Panseq and 29 publicly-available E. coli genome sequences (see Additional File 7, Table S5 for list of E. coli genomes and accession numbers used for comparative analyses). The functions of identified sequences were predicted using the annotation engine AutoFACT. Circular genome atlases were generated using CGView or Circos.

22J774A.1 macrophage cells were seeded at 5 × 105 cells/well in DMEM with L-glutamine and 10% FBS for 16 h prior to infection. Cells were infected at a multiplicity of infection of 10 with wild type NRG857c or the iutA mutant. Infected cells were incubated at 37°C for 2 h, then washed and treated for 2 h with 100 μg/ml gentamicin. At various times post-infection, cells were washed and lysed with 0.1% Triton X-100 in PBS, followed by serial plating on LB agar. Gentamicin protection experiments were performed in triplicate and reported as the percent survival with standard error with statistical significance determined by Student's t test.

23All animal experiments were performed in accordance with protocols approved by the local animal ethics committee at the University of Texas Medical Branch, Galveston, Texas. Female ICR mice of 20-25-g (Charles River Laboratories) were used after 72 h of quarantine as described previously. Briefly, food-restricted animals received streptomycin (5 g/L in drinking water supplemented with 7% fructose) for 48 h prior to oral inoculation with NRG857c or the iutA mutant. Groups of mice (n = 6) were orally inoculated with a suspension of NRG857c bacteria in a final volume of 0.4 mL delivered by gavage (20-gauge needle). The animals were maintained for 72 h, after which the animals were killed and the small intestines removed for homogenization and enumeration of the bacterial load. Groups were compared using the Mann Whitney non-parametric test.

24The synthesis of siderophores by AIEC O83:H1 was analyzed by the colorimetric Arnow assay to detect catechol siderophores and the ferric perchlorate assay for hydroxamates. To restrict the iron availability in liquid or solid medium, the iron chelator 2,2'-dipyridil was used. To examine the ability to use various siderophores or iron compounds as iron sources, overnight cultures of AIEC O83:H1 were diluted to 1 × 105 bacteria per ml and seeded into L agar containing 2,2'-dipyridil. Plates were spotted with 5 μl of 8 μM hemin or 5 μl of an overnight culture of a siderophore-producing strain. A sterile disk containing 20 μl of 10 mM FeSO4 was placed on each plate. Growth was monitored around the spots or disk after 18 to 24 hours at 37°C.

25All authors contributed to the writing of this manuscript as well as overall project design; AK developed the gap closure strategy and manually screened the NCBI pipeline data; MM, PK and KZ carried out the laboratory experiments for closing the chromosome and plasmid sequences; AV, JN and PK performed the bioinformatics analyses.