Duplication and diversification of a unique chromosomal virulence island hosting the subtilase cytotoxin in Escherichia coli ST58

Abstract

The AB₅ cytotoxins are important virulence factors in Escherichia coli . The most notable members of the AB₅ toxin families include Shiga toxin families 1 (Stx₁) and 2 (Stx₂), which are associated with enterohaemorrhagic E. coli infections causing haemolytic uraemic syndrome and haemorrhagic colitis. The subAB toxins are the newest and least well understood members of the AB₅ toxin gene family. The subtilase toxin genes are divided into a plasmid-based variant, subAB1, originally described in enterohaemorrhagic E. coli O113:H21, and distinct chromosomal variants, subAB2, that reside in pathogenicity islands encoding additional virulence effectors. Previously we identified a chromosomal subAB2 operon within an E. coli ST58 strain IBS28 (ONT:H25) taken from a wild ibis nest at an inland wetland in New South Wales, Australia. Here we show the subAB2 toxin operon comprised part of a 140 kb tRNA–Phe chromosomal island that co-hosted tia, encoding an outer-membrane protein that confers an adherence and invasion phenotype and additional virulence and accessory genetic content that potentially originated from known virulence island SE-PAI. This island shared a common evolutionary history with a secondary 90 kb tRNA–Phe pathogenicity island that was presumably generated via a duplication event. IBS28 is closely related [200 single-nucleotide polymorphisms (SNPs)] to four North American ST58 strains. The close relationship between North American isolates of ST58 and IBS28 was further supported by the identification of the only copy of a unique variant of IS26 within the O-antigen gene cluster. Strain ISB28 may be a historically important E. coli ST58 genome sequence hosting a progenitor pathogenicity island encoding subAB.

Data Summary

The completed assembly of the chromosome and raw sequencing reads for strain Escherichia coli IBS28 have been uploaded to GenBank under BioProject PRJNA591373, with the assembly under accession CP049979 and raw reads in the Sequence Read Archive under accession PRJNA591373.

Impact Statement.

Tracing the source, distribution and evolution of specific genetic components from the bacterial pan-genome is proving critical to understanding the development of infectious disease. Here we undertook a comparative analysis of a bacterial strain sourced from an ‘environmental’ niche hosting the Escherichia coli AB₅ protein, subtilase cytotoxin. This Australian strain was closely related to a set of sequences from North America. While only the Australian strain hosted the subtilase toxin, each related ST58 strain hosted the same IS26 allele within the O-antigen lipopolysaccharide region. This O-antigen region (including the insertion) was also observed in an unrelated E. coli ST58 strain, and a separate E. coli ST101 strain, both also from North America. The subtilase cytotoxin was hosted in a large genomic island alongside other virulence content often reported in E. coli . This genomic island appeared to be formed from a homologous duplication event leading to two large genomic islands with partial shared content. Finally, context of the subtilase toxin suggests that it is a genetic progenitor of many other instances of the toxin that have been reported based on the co-presence of invasin tia and other genes, as through this duplication we see the contextual formation of this tia-subAB gene configuration.

Introduction

Escherichia coli is a globally distributed commensal bacterium, colonizing most warm-blooded mammals and avian species, as well as diverse environmental niches [1]. The capacity to successfully colonize such a broad range of environments stems largely from its ability to acquire and integrate genetic cargo from lateral sources. The acquisition of plasmids [2–4], phagemids [5], phages [4, 6–8] and genomic islands [2, 9] has played an important role in shaping the evolution of E. coli and other clinically important proteobacteria. Lateral gene transfer (LGT) events underpin the formation of novel hybrid E. coli pathogens [10, 11]. The ability to capture genetic material horizontally gives the prospect of closely related bacterial lineages with very different phenotypes; outer-membrane polysaccharide structure (O-antigen) [12], metabolism [13] and the production of toxins and adhesins [14], to name a few. In E. coli , LGT has shaped pathotype lineages, including the differentiation of intestinal and extra-intestinal pathogens [15, 16]. The AB₅ toxins are particularly significant in this regard and include the Shiga toxins, Stx₁ and Stx₂, that distinguish enterohaemorrhagic E. coli [8] and the heat-labile (LT) enterotoxin family [17] that are responsible for significant enteric disease in both humans and porcine agriculture.

In 2004, a new AB₅ toxin known as the subtilase toxin SubAB was described [18]. The genes were found in a Shiga-toxigenic E. coli (STEC) strain 98NK2 with serotype O113:H21 that caused haemolytic uraemic syndrome (HUS). The SubAB AB₅ toxin is a serine protease and the endoplasmic reticulum chaperone protein BiP is one of its known targets [19]. Immunoprecipitation experiments using the SubAB toxin as bait with a Vero cell lysate identified several interaction partners, including α2 and β1 integrin, leading the authors to suggest that α2β1 integrin is a receptor for SubAB [20]. The ability of SubAB to bind to α2β1 integrin is likely important to how SubAB enters host cells, engages with the endosomal pathway and traffics intracellularly. Furthermore, the SubB component of the toxin is atypical for AB₅ toxins in that it has specificity for glycans terminating with α2–3-linked N-glycolylneuraminic acid (Neu5Gc) [21].

SubAB is increasingly seen as a significant addition to the virulence genes of E. coli . Although the clinical significance of SubAB toxin in humans is poorly understood, the pathological effects of SubAB toxins on mice [22, 23] and rats [24] are substantial and mimic the pathological features of enterohaemorrhagic uraemic syndrome . The subAB operon is often described in association with strains that are either LEE-negative STEC or Shiga toxin-negative E. coli [25]. The isolation of an intimin-negative (eae ⁻), subAB-positive (subAB ⁺) O128:H2 STEC strain from a patient with prolonged bloody diarrhoea [26] may be a significant observation. O128:H2 is a serotype of E. coli typically sourced from sheep and deer but not cattle [27, 28]. Ovine STEC strains are considered to be less pathogenic because they lack intimin and carry variants of stx, such as stx _1c and stx _2b [29–32], which are infrequently associated with STEC that cause HUS. It was suggested that carriage of subAB linked to a chromosomal pathogenicity island (PAI) in the human O128:H2 isolate may have ramifications for its ability to cause disease in humans [26], but more comprehensive genomic epidemiological studies are needed.

Currently, four variants of subAB (subAB1, subAB2-1, subAB2-2 and subAB2-3) have been described [25, 33]. subAB1 is a plasmid-encoded variant found on plasmid pO113 that was originally described in E. coli O113:H21 strain 98NK2. STEC with serotype O113:H21 have a reservoir in cattle [28] that spills over into humans, fresh produce and in the environment but they are rarely isolated from sheep [27, 32, 34, 35]. The remaining three variants (subAB2-1, subAB2-2 and subAB2-3) reside chromosomally. In Shiga toxin-negative E. coli strains ED32 and ED 591 from unrelated cases of childhood diarrhoea, pathogenicity island SE-PAI, which carries subAB2-1, spans about 8 kb and is located between yjhS and tRNA–Phe [25, 36]. The SE-PAI also carries several putative virulence-associated genes (VAGs) including tia that encodes an epithelial cell adhesin and invasin [37], ShiA, whose function in Shigella flexneri is to attenuate inflammation by suppressing the innate T-cell immune response [38, 39], a putative sulfatase and an integrase that presumably is involved in the mobilization/capture of the PAI [36].

E. coli ST58 belongs to a phylogroup B1 commensal lineage that has been isolated from diverse animal, environmental and human hosts [40–42], demonstrating links with human urinary tract infections and urosepsis, and it can carry a colV-like virulence resistance plasmid [40]. Recently we characterized an environmental ST58 isolate recovered from the faeces of a straw-necked ibis nest in inland wetlands, New South Wales, Australia [43].

The short-read assembly analysis of this ST58:ONT:H25 isolate (IBS28) indicated that it was closely related to 4 North American strains from a collection of 69 diverse ST58 genomes [43]. Strain ISB28 was distinct from known Australian human ST58 pathogen E. coli 2009–52 [40] and avian pathogen E. coli A193 [44], which have been described recently. It hosted various VAGs that were uncommon among the other strains recovered from the ibis nests. The profile included immune survival factor iss, microcins mch, major subunits of the F17 adhesin and the AB₅ subtilase cytotoxin genes subAB. Strain IBS28 hosted an ICE with no known VAGs; a relative of an ICE identified in an Australian E. coli ST38 strain taken from hospital-sourced human infection. Strain IBS28 was host to a single IS6 family element, an insertion sequence (IS) heavily involved in the evolution and dissemination of antimicrobial resistance genes [45]. This insertion was chromosomal within the acquired O-antigen island (non-typeable in silico). Here we report the long-read sequencing, assembly and annotation of E. coli ST58 strain IBS28 for the purpose of resolving the genetic context of subAB, IS26 and numerous other virulence genes.

Methods

The isolation and initial analysis of strain E. coli IBS28 has been reported previously [43]. The strain was derived from a desiccated faecal sample initially taken from an ibis nest at an inland wetland during 2012, in New South Wales, Australia.

To isolate genomic DNA for long-read sequencing, the strain was first plated onto Lysogeny broth (LB) agar for single colonies and incubated at 37 °C for 16 h. A single colony was then sub-cultured in 2 ml LB broth at 37 °C, with shaking at 220 r.p.m. for 16 h. Genomic DNA was then extracted from lysed cells using XS buffer [46] and purified gently with a phenol/chloroform procedure designed to preserve the integrity of DNA strands for sequencing.

Long-read sequencing was performed on a PacBio Sequel at the Ramaciotti Centre for Genomics at the University of New South Wales, as part of a multiplex protocol. Assembly was performed by the service provider using the HGAP4 pipeline. Genome sequence data were deposited under accession PRJNA591373.

Genomic analysis was performed using several software packages and online servers. Primarily, sequence comparisons and gene identification were performed using the National Center for Biotechnology Information (NCBI) blastn/blastx (https://blast.ncbi.nlm.nih.gov/Blast.cgi) servers and progressiveMauve [47]. Bacteriophages were identified using PHASTER [48], insertion sequences were identified using ISFinder [49] and genomic islands were identified using IslandViewer 4 [50]. Genomic islands were only annotated if they appeared in outputs from at least two detection algorithms. Single-nucleotide polymorphism (SNP) analysis of subAB was performed using AliView [51] utilizing the muscle aligner [52]. GC content analysis was performed using GC Content Calculator (https://www.biologicscorp.com/tools/GCContent) Reference sequences for comparative alignments and phylogenetic analyses were sourced from GenBank (https://www.ncbi.nlm.nih.gov/nucleotide/) and Enterobase (https://enterobase.warwick.ac.uk/).

Results

Sequence summary

The long-read assembly of E. coli strain IBS28 resolved two contigs, a 4 934 027 bp chromosome, plus the ICE pIBS28_1, as described previously [43]. Automated annotation identified 4877 open reading frames (ORFs) and 112 RNA coding sequences in the chromosome. Automated genomic island analyses identified 25 acquired regions, and comparisons to the closely related North American ST58 genomes, Australian ST58 genomes and ST101 strain KSC9 (Fig. 1) suggests the presence of others highlighted by a lack of blastn coverage amongst the references. The genomic islands identified in the IBS28 chromosome encode previously identified operons including secretion systems, VgrG-Rhs protein operons, putative adhesins and DNA repair enzymes. Three complete phages were similarly identified, two reported as lambda enterophages with one encoding the serum resistance gene iss/bor, and one mEp460-like enterophage, each approximately 50 kb in size.

Fig. 1.

Map and phylogeny of the E. coli strain IBS28 chromosome highlighting genomic islands and other insertions. Coloured rings on the map visualize blastn alignment data of other E. coli sequences, with the black centre ring indicting GC content for IBS28. A snapshot of a larger ST58 SNP phylogeny including IBS28 is presented in the top right. blastn alignments were performed against the most closely related sequences from North America, plus two additional Australian ST58 pathogens.

Characterization is IS6 family insertion sequence

A key observation regarding the IBS28 chromosome was the presence of an IS6 family element, an IS26 variant with a single C573T SNP (820 bp sequence). The long-read assembly confirms that this element lies within the variable O-antigen region. Alignment analysis revealed that this same O-antigen and IS6 element were present across the entire local clade of North American ST58 strains most closely related to IBS28, which range from isolations in Canada 1988 to the USA in 2018. Coincidentally, this insertion was identified in another ST58 strain, short-read assembly ESC_LA6312AA_AS (USA, mammal, 1997) from the Enterobase collection utilized in our prior phylogenetic analysis. This additional ST58 strain is phylogenetically distinct from the local clade IBS28 belongs to. This unusual insertion signature also led to the identification of E. coli strain KSC9, an ST101 from a 2014 porcine sample from the USA, which also hosted this specific LPS system carrying the IS26 allele. Our comparative analysis in Fig. 1 demonstrates its presence there. These examples place this chromosomal IS6 element in human, porcine, bovine and environmental contexts isolated over a 30 year span.

Summary of pathogenicity islands PAI-1 and PAI-2

Critically, amongst the genomic islands we identified in IBS28 were two large related PAIs of approximately 140 and 88 kb. These genomic islands, defined from here as PAI-1 and PAI-2, share much of their genetic content and we hypothesize that they were once a single genomic island that underwent duplication and then diversification through the capture of additional genetic content.

From what could be observed in the sequence data, the progenitor island sat at one of two separate tRNA–Phe–GAA sites in the E. coli ST58 chromosome, generated by a DUF4102 domain-containing integrase. At the time of duplication, this island was approximately 21 kb in size. One scenario suggests a recombination event involving a second tRNA–Phe–GAA sequence elsewhere in the chromosome led to the duplication of this region, leaving a second copy in the reverse orientation at the second tRNA–Phe–GAA site. This event also apparently led to the inversion of a 1.3 Mb region of chromosome that sits between the two PAIs (See Fig. 1). Over the course of time, PAI-1 and PAI-2 appear to have subsequently captured diverse sets of unrelated genes. Shared genetic content encoded by both islands includes 12 coding sequences (CDS) and 16 hypothetical proteins, including invasin tia, autotransporter adhesin Ag43 (Antigen 43) involved in biofilm formation, a YeeVU toxin/antitoxin system and a haemolysin expression-modulating protein. Both islands also have an ISSfl8-like element inserted into a Z1226 protein (putative RNA-directed DNA polymerase activity), likely a once unique feature now shared by both PAI-1 and PAI-2, supporting the contention these islands were once a single element. Given these observations, we suggest that PAI-1 and PAI-2 were once a single genomic island, the inferred structure of which will be referred to as PAI-0.

A third partial copy of PAI-0, named PAI-3, was found inserted near the acquired O-antigen lipopolysaccharide (see Fig. 1), but this copy did not include the Z1226 protein or the insertion, so the evolutionary relevance of its presence remains unresolved. A comparison of GC content of this potentially homologous region (Fig. S1, available in the online version of this article) demonstrates some similarity, with each sitting at 52 % GC over the 14–17 kb included.

The integrase and much of the other shared content initially forming PAI-0 sat at 94 % nucleotide identity between PAI-1 and PAI-2, suggesting the duplication was not a recent event, but the tRNA sequences remained identical.

PAI-1 (139 839 kb) encoded 91 ORFs plus 93 hypothetical proteins, with some ORFs broadly separated into virulence-associated (n=21), membrane-associated (n=11) and metabolism-associated (n=10) genes (Fig. 2). The virulence genes of primary concern that PAI-1 has acquired include the subtilase cytotoxin, subtyped as a subAB2-1, invasin tia, Ag43 and a serine protease autotransporter (espL), an RTX toxin, haemagglutinins and immunoglobulin-binding protein eibA, an F17 fimbrial adhesin operon (fimACDG) and microcin H47 (mch/MccH47). In addition to these, it has acquired outer membrane-associated proteins of the TonB, TolC, OmpF and ABC transporter families, amongst others. Metabolism genes included iroBD, PTS system components and an [NiFe]-hydrogenase operon. The remaining ORFs comprising the island included polymerases, a toxin/antitoxin system, gene co-activators and neurotensin receptor r8. Insertion sequences of the following families were also present: IS110, IS66, IS3 and ISSfl8, as mentioned.

Fig. 2.

Maps of large pathogenicity islands PAI-1 and PAI-2. Annotations of the virulence islands with ORFs coloured by function. blastn similarity between islands is indicated in black, with reversed sequence indicated in red. Key features are annotated. The region of sequence identified in PAI-3 is indicated by the orange bar under the PAI-2 diagram.

PAI-2 (87 839 bp) likewise encoded 64 ORFs and 48 hypothetical proteins, with relatively reduced virulence-associated content (n=7) and membrane-associated proteins (n=4), but an increased number of metabolism-associated genes (n=20). Primary virulence genes include putative haemolysin activator hecB, an AidA-1 like adhesin, a large filamentous haemagglutinin coupled with a haemolysin activator, and another copy of Ag43 as part of the original PAI-0 sequence. Comparisons between PAI-1 and PAI-2 are shown in Fig. 2.

The full list of genes identified in PAI-1 and PAI-2 can be found in Data File S1, tables S1 and S2 respectively. Additionally, the sporadic presence of these genes can be observed in the ST58 reference sequence blastn data presented in Fig. 1.

Characterization of the subtilase cytotoxin operon

To characterize the subAB2-1 allele and context identified here, available subAB2 sequences (both CDS-only and those present with context) were downloaded from GenBank. Fig. S2 displays a full SNP alignment of available subA2 gene sequences. The majority of these sequences were initially sourced from two separate studies identifying subAB2 from E.coli in deer samples from Germany (CDS-only, with some reportedly hosted on plasmids) and human samples from Norway. As reported in other studies, this analysis demonstrated both type 1 and type 2 subA2 alleles, with the IBS28 allele resolving as the type 1 (See Figs 3 and S3). It should be noted that subB2 shows no such type differentiation (Fig. S4).

Fig. 3.

Phylogeny of subA2 sequences. Visualization of muscle alignments of subA2 gene sequences and associated metadata of E. coli hosts. Alignment highlights the split between type 1 and type 2 alleles and various subgroups therein. Co-presence of subA with tia and shiA is indicated with a red dot. Grey dots indicate that only the subAB CDS were available.

Michelacci et al. [36] identified the type 1 subA2 allele co-localized with the genes tia and shiA and a sulfatase ORF (orf) inserted at a phage integrase in the order: tRNA–integrase–shiA–orf–tia–subAB2. This region was described as SE-PAI (JQ994271). This same gene co-localization and organization are observed in every available example of contextual type 1 subA2 genes, including in PAI-1 of IBS28. Initial blastn analyses indicate the presence of different integrases at different tRNA sites hosting this virulence island (data not shown). Here, within this PAI duplication example, we appear to have identified a potential source/progenitor of this shiA–orf–tia–subAB2 virulence region, and thus the now widely distributed type 1 subA allele in general. Shared sequence between PAI-1 and PAI-2 (described as PAI-0) encoded the orf–tia sequence that was observed centrally in SE-PAI. Our analysis suggests that within PAI-1 the shiA and subAB2 were inserted before and after these genes post-duplication. PAI-2 still encodes the sulfatase ORF and tia from PAI-0, although the invasin is highly degraded now at 75 % nucleotide identity, a comparatively high rate of nucleotide variation to the rest of shared PAI-0 sequence. Added to this is the fact that regions of at least 10 kb of PAI-1 can be found distributed around the chromosomes of strains carrying SE-PAI (Data File S1, table 3)) and also type 1 subAB alleles, which implies that this ST58 chromosome may be the source of this and other large gene organisations, which then disseminated as a coherent genomic island through the chromosomal integrase system. The highest coverage of PAI-1 was seen in CP027672.1, which carried 68 % of the island at an average 96.75 % nucleotide identity, with the highest fragment size being 16501 bp. As a final point, subA2 genes reported as type 3 (subA2-3; LK985413.1, LK999956.1) show the same SE-PAI context as other type 1 subA genes, but these alleles share certain SNPs with the type 2 alleles, giving them an observable subclade amongst the type 1 alleles in Fig. 3.

Discussion

Here we characterized an environmentally associated E. coli ST58 (ISB28) from the straw-necked ibis. Our initial analysis utilizing short-read whole-genome data indicated that IBS28 was closely related to a series of strains from North America dating back to 1988, and we detected several virulence genes plus an unusual chromosomal IS6 family element [43]. Members of the IS6 family are frequently involved in the dissemination of antimicrobial resistance genes [53], although none were detected within the strain. The only extra-chromosomal element IBS28 carried was an integrative conjugative element, a homologue to an element we identified from an Australian human clinical E. coli isolate, giving it a tentative link to the Australian human microbiome.

In resolving the completed chromosome, it became apparent that IBS28 hosts significant virulence gene content, as aside from hosting smaller, more commonly observed, genomic islands such as secretion systems [54, 55], it had undergone an interesting chromosomal duplication of an integrase-based genomic island, leading to the formation of two extremely large genomic islands. At 148 kb, the pathogenicity islands we describe here are, to the best of our knowledge, the largest integrase-based genomic islands observed in an E. coli . We hypothesize that this extreme level of genetic capture occurred due to the duplication, which led to two assumedly functional copies of the integrase. The islands PAI-1 and PAI-2 are composed of many previously reported chromosomal virulence factors, including SubAB2 [36, 56], H47 siderophore microcin [57] and F17 adhesin [58], and many individual genes such as an RTX toxin [59], haemagglutinins and a protease autotransporter [60]. Whilst these genes encoding these factors are not novel, the appearance of large aggregations of these virulence genes suggests high acquisition rates within this chromosome. The potential fitness costs associated with carrying these large islands may have resulted in the lack of plasmid content [61], although this is strictly conjecture.

The IBS28 genome has both a partial and a complete copy of the previously identified virulence island SE-PAI. Encoding shiA–orf–tia–subA–subB, the full set of genes are present in the order previously described [36] at the start of PAI-1, with only orf–tia found in PAI-2, with tia showing a high level of nucleotide variation compared to the rest of the island. We suggest that the sulfatase ORF and tia were the initial genes inserted in PAI-0, and we then see shiA and the subAB2 ORFs inserted before and after in PAI-1 Fig. 2), generating this SE-PAI structure observed in multiple strains across continents. We consider this constructive acquisition scenario to be more likely than if the toxin genes were lost within PAI-2, as there is no evidence of mobile genetic element activity, and nor is there any evidence to suggest that the chromosomal integrase gene can extract or rearrange the captured sequence. As a final indicator, sections of PAI-1 that include the SE-PAI region can be found within the reference strains analysed, suggesting that the SE-PAI element described initially is a much-reduced version of the PAI-1 described here.

Here we have identified and characterized an unusual O-antigen in strain IBS28 hosting an IS26 allele that is present within the local clade of phylogenetically related ST58 strains, and incidentally observed in unrelated ST58 strain LA6312AA_AS from Enterobase plus an ST101 strain KSC9 from the GenBank nuccore database, each from the USA. The role IS26 plays in the generation and dissemination of AMR is well described and, combined with its ability to capture and rearrange small genomic regions into new configurations, it has aided in the development of multiple-drug resistance [62]. Recent work has also demonstrated that IS26 readily self-associates [63], meaning that the likelihood of an IS26 element inserting at an already present IS6 family element is much higher than at a random locus. Given this, we view the O-antigen cassette identified here as a potential chromosomal hotspot for the capture of genetic cargo mobilized by IS26, including antimicrobial resistance genes, that can be disseminated alongside the lipopolysaccharide variant.

Neither PAI-0 nor its now duplicated islands are present in any of the closely related North American strains, but the IS26 allele is, indicating that several closely related lineages of this ST58 strain have been isolated since 1988, and as recently as 2018 in the USA. In identifying this strain from a nest in New South Wales wetlands circa 2012, we demonstrate the breadth of dissemination of bacterial genes that can play a role in human pathogenesis, and suggest that epidemiological work is performed to characterize the spread of this ST58 strain as a source of chromosomal virulence content in E. coli .