Genome sequencing of strains of the most prevalent clonal group of O1:K1:H7 Escherichia coli that causes neonatal meningitis in France

To describe the temporal dynamics, molecular characterization, clinical and ex vivo virulence of emerging O1:K1 neonatal meningitis Escherichia coli (NMEC) strains of Sequence Type complex (STc) 95 in France. The national reference center collected NMEC strains and performed whole genome sequencing (WGS) of O1:K1 STc95 NMEC strains for phylogenetic and virulence genes content analysis. Data on the clinical and biological features of patients were also collected. Ex vivo virulence was assessed using the Dictyostelium discoideum amoeba model. Among 250 NMEC strains collected between 1998 and 2015, 38 belonged to O1:K1 STc95. This clonal complex was the most frequently collected after 2004, representing up to 25% of NMEC strains in France. Phylogenetic analysis demonstrated that most (74%) belonged to a cluster designated D-1, characterized by the adhesin FimH30. There is no clinical data to suggest that this cluster is more pathogenic than its counterparts, although it is highly predominant and harbors a large repertoire of extraintestinal virulence factors, including a pS88-like plasmid. Ex vivo virulence model showed that this cluster was generally less virulent than STc95 reference strains of O45S88:H7 and O18:H7 serotypes. However, the model showed differences between several subclones, although they harbor the same known virulence determinants. The emerging clonal group O1:K1 STc95 of NMEC strains is mainly composed of a cluster with many virulence factors but of only moderate virulence. Whether its emergence is due to its ability to colonize the gut thanks to FimH30 or pS88-like plasmid remains to be determined.


Background
Escherichia coli is a commensal of the gastrointestinal tract of vertebrates, including humans [1]. It is also involved in intestinal and extraintestinal infections [2]. In particular, E. coli is the most frequent bacteria involved in preterm meningitis and the second most frequent in newborn meningitis, causing a high rate of mortality or sequelae [3,4].
Between 2001 and 2013, a French prospective national survey collected data for 325 children hospitalized with E. coli meningitis [5] and 141 E. coli isolates were sent to the national reference center (NRC) and studied. The highly pathogenic phylogenetic subgroup STc95 represented 56% of the 141 strains analyzed and among them, serogroup O1 (27.7%) was the most frequently identified during this period relative to serogroups O18 (19.1%) and O45 S88 (11.3%).
Whole genome sequencing (WGS) of a large collection of strains belonging to STc95 was recently performed [9], but no O1:K1 neonatal meningitis strain belonging to this complex has been comprehensively characterized. We performed WGS of all the 38 O1:K1 STc95 strains isolated from 1998 to 2015 to gain insight into the pathophysiology of neonatal meningitis and their disease potential. Moreover, we analyzed the virulence of representative strains using the Dictyostelium discoideum amoeba model as previously used to evaluate E. coli extraintestinal pathogenic strains [10].

Epidemiology
We characterized the temporal dynamics of the three major STc95-serogroups that cause neonatal meningitis, O1, O18, and O45 S88 , by determining their respective rates between 1998 and 2015 (Fig. 1). These three serogroups represented 30 to 50% of the E. coli meningitis strains among the 250 NMEC collected by the NRC during the study period. Serotype O45 S88 :K1 largely predominated between 1998 and 2003, whereas O1:K1 isolates became the main serotype in STc95 NMEC in 2004 and thereafter. The 38 O1:K1 STc95 strains that were subjected to WGS represented up to 25% of E. coli meningitis strains received each year in the NRC. O18:K1 serotype is still present in the epidemiological landscape from 1998 to 2015 but never in the majority, with a percentage oscillating between 6 and 21%.

Genotyping
All our O1:K1 NMEC belonged to ST95, except 12 strains belonging to ST421, ST1568, ST2619, and ST5484 (and three strains harbored an unnamed ST). All these STs differ from ST95 by one allele and are therefore considered to be part of STc95 (Fig. 2).
A WGS phylogenetic tree, based on the distribution of core-genome SNPs, clustered NMEC O1:K1 strains into two major subgroups, A and D. These are two of the five subgroups among STc95 E. coli described by Gordon et al. [9]. Within subgroup D, we identified three clusters called D-1, D-2 and D-3, with most NMEC isolates (74%) belonging to cluster D-1.
Strain S88, representative of the O45 S88 :K1 clonal group, appears to be more closely related to clusters D-2 and D-3 than the major cluster, D-1, whereas the representative strain of APEC O1:K1 seems to be paradoxically more distantly related (subgroup C) (Fig. 2).

Virulence genotypes
The distribution of extraintestinal virulence factors identified by WGS in subgroup A and clusters D-1 and D-2/ D-3 is depicted in Table 1. Cluster D-1 possessed the most ExPEC genetic determinants, as almost all isolates harbored all iron transport systems, bacteriocins, and other ColV plasmid-related genes found in pS88-like plasmids. Conversely, toxin genes were more prevalent among subgroup A and clusters D-2/D-3. Several factors involved in the pathophysiology of meningitis, such as Cnf1, Sfa/ foc, and IbeA, were absent from our O1:K1 strains.
FimH is an adhesin of type I that is involved in the adhesion of endothelial cells preceding the invasion of cerebral spinal fluid (CSF) [11]. The distribution of the various alleles of FimH or the absence of this gene is in accordance with the genetic subgroups and clusters (Fig.  2). The major allele, FimH30, differs from those of the O45 S88 :K1 (S88) and O18:K1 (UTI89) reference strains, FimH54 and FimH18, respectively.

Antimicrobial resistance
No extended spectrum beta-lactamase or carbapenemase was found among these meningitis isolates. Seventeen strains (45%) produced a penicillinase, all harboring bla-TEM-1 except one strain harboring blaTEM-84. Alleles of blaTEM-1 were blaTEM-1-A (n = 1), blaTEM-1-B (n = 8) and blaTEM-1-C (n = 7). Allele blaTEM-1-B was associated with genes encoding resistance to streptomycin (strAB) and sulphonamide (sul2 and dfrA5) in all cases but one. Antimicrobial resistance was not associated with  [9]. Strains with names beginning with the letter S correspond to neonatal meningitis strains. APECO1, S88, and UTI89 were used as reference strains for comparison. The tree was rooted on UPEC strain CFT073. All NMEC strains are ST95 or ST421 except where indicated between brackets. Three strains carried unnamed STs (STUN). FimH alleles are also indicated a particular WGS subgroup or with pS88-associated genes. Thirteen isolates were devoid of any antimicrobial resistance gene.
Clinical and biological features of patients infected by O1:K1 STc95 E. coli strains We compared patients infected by strains belonging to the main epidemiological cluster D-1 to those infected by strains belonging to other clusters. Patients infected by cluster D-1 were younger, less mature, and of lower birth weight, although the differences were not statistically significant ( Table 2). Although cluster D-1 was largely predominant, no clinical data support higher pathogenicity of this cluster relative to its counterparts. Median CSF glucose was lower in patients infected by cluster D-1 and median CSF protein higher than for the other subgoups, as well as the number of cells in CSF, which was significantly higher (p = 0.02) ( Table 2).

Ex vivo model with amoeba Dictyostelium discoideum
Grazing scores of the amoeba virulence assay are not shown [see Additional file 1]. The ex vivo model with amoeba D. discoideum did not reveal clear associations between phylogenetic subgroups and grazing resistance. Generally, the O1:K1 strains were less virulent in this model than representative strains of the O45 S88 :K1 and O18:K1 serotypes and the virulent control strains. However, the different subclones were not equally virulent. Indeed, among the D-1 strains, the subclone containing strains S158, S208, and S358 killed D. discoideum more efficiently (mean score 0.06) than the subclone containing strains S136, S368, S384, and S386 (mean score 0.65).
We tested a representative strain of cluster D-1, S172, cured of its plasmid, and a transconjugant of J53 in our model to determine whether plasmid pS88-like may play a role in the virulence of O1:K1 D-1 strains in our model. The grazing score was clearly affected, with a loss of virulence for S172△pS172, while J53pS172 gained virulence [see Additional file 1].

Discussion
The clonal complex STc95 is one of the major E. coli lineages that causes human extraintestinal infections. The extraintestinal virulence of STc95 E. coli is exemplified by their ability to cause neonatal meningitis. In France, this group represents 56% of the neonatal meningitis strains collected by the national reference center [5]. WGS performed on several hundred STc95 strains from various parts of the world has provided a comprehensive view of the genetic organization of this clonal complex, as well as its geographical distribution and temporal dynamics [9]. However, among the 500 strains investigated in this study, only two were isolated from the CSF Among the three major subgroups (A, C, D) that contain O1:H7 strains, described by Gordon et al. [9], strains responsible for meningitis, collected throughout France, belonged exclusively to subgroups A and D. The absence of O1:H7 subgroup C strains in our collection may be related to their lower potential to cause disease relative to other subgroups or their specific geographical distribution, as they were only found in Australia in the world-wide collection [9]. O1:K1:H7 meningitis strains of our study were more frequently found in subgroup D than in subgroup A (n = 34 versus n = 4). These two subgroups were present in all continents studied (USA, Europe, and Australia) with a similar repartition (n = 8 and n = 12 for subgroup A and D, respectively [9]. Subgroup D which predominates among our collection may have a greater potential to cause neonatal meningitis than subgroup A. Phylogenetic analysis allowed us to distinguish three clusters within subgroup D, called D-1 (n = 28), D-2 (n = 4), and D-3 (n = 2). The closely related clusters D-2 and D-3 are minor relative to cluster D-1 and carry a different ST (ST421) and a different FimH (92) or no FimH gene (D-3). Cluster D-1, which carries FimH30, may represent a group of strains with a high capacity to induce neonatal meningitis. It also carries genetic determinants characteristic of the extraintestinal virulence plasmid pS88, which are absent from clusters D-2 and D-3, and rarely present in subgroup A. This plasmid may be key to the virulence of this group, as shown for the recently described clone O45 S88 :K1:H7 [8]. The representative strain of clone O45 S88 :K1:H7, S88, which carries the pS88 plasmid, appears to be closely related to D-2 and D-3 strains. Thus, it is possible that clone O45 S88 :K1:H7 was derived from D-2/D-3 strains, after switching its O antigen gene cluster and acquiring the pS88 plasmid.
Analysis of the clinical features of infected neonates did not provide evidence that cluster D-1 is more virulent than D-2/D-3, despite the large number of strains of cluster D1 and the presence of plasmid pS88. However, this cluster appeared to elicit a larger inflammatory response.
Several factors known to be involved in the pathophysiology of neonatal meningitis were completely absent from our collection, i.e. ibeA, cnf1, and sfa. This highlights the variation in the virulence factor repertoire that leads to acute bacteremia and crossing of the blood brain barrier, the two major steps of this infection. Several studies have attempted to define a potential NMEC pathotype [12,13]. For example, Wijetunge et al. compared 26 genes encoding virulence factors between 53 NMEC strains and 48 fecal strains of healthy individuals and found that the combination of K1 capsule, aerobactin siderophore, va cuolating cytotoxin (Vat), and the iron-binding protein (Sit) are typical traits of NMEC [13]. Among toxins, only vacuolating cytotoxin was present in almost all O1:K1 NMEC strains, irrespective of genetic subgroup, reinforcing the potential role of this toxin in the physiopathology of neonatal meningitis.
We assessed the experimental virulence of O1:K1:H7 and representative and control strains in the amoeba D. discoideum model, previously used to assess E. coli resistance to phagocytosis [10]. Our aim was to analyze possible fine differences between meningitis-causing clones and not to simulate the global pathophysiology of meningitis. This model avoids the use of animals and is of interest because it is performed at a low temperature (22°C). At this temperature, the K1 capsule, the major virulence factor of NMEC, which may mask other bacterial traits involved in virulence, is inactivated [14]. Its inactivation may facilitate analysis of the potential role of other factors. Indeed, we assessed the production of the capsule of our O1:K1 strains by the agglutination test after culture at 22°C and 35°C and found that the K1 capsule was undetectable at 22°C, but present at 35°C (data not shown). Generally, the O1:K1 strains appeared to be less virulent than representative strains of O45 S88 :H7 and O18:H7 serotypes and the control virulent strains. However, we found that they were not equally virulent upon analysis of each subclone. Among the D-1 strains, closely related subclones, with an identical repertoire of virulence genes behaved differently in the D. discoideum model. This highlights the complexity of the regulation of virulence and the involvement of various factors that are currently not known. Moreover, since most virulence factors implicated in human pathogeny are more efficient at 37°C, it is also likely that the amoeba model underestimates their role. Nevertheless, we were able to show that the pS88-like plasmid plays a role in the resistance against phagocytosis using isogenic strains, thus complementing its previously described role in survival to bactericidal activity of serum [8].
Another notable difference between the O1:K1 clusters is the adhesin FimH. It binds specifically to D-mannose residues attached to the surface of glycoproteins that line vaginal, perineal, and bladder cells, as well as enterocytes [15]. FimH is expressed by more than 95% of E. coli and genetic variation can change its tropism [16]. It also plays an important role in the adhesion and invasion of endothelial cells of brain capillaries by NMEC in humans [17]. However it may be dispensable, since we found two meningitis-causing strains, S229 and S245, with no fimH gene (this was confirmed by fimH-specific PCR, data not shown).
The strains of NMEC O1:K1:H7 described here, carry mostly the fimH30 allele, whereas the other main serotypes responsible for meningitis carry the FimH54 (O45:K1:H7) and FimH18 (O18:K1:H7) alleles. Of note, most strains of the multi-resistant epidemic clonal group of E. coli ST131 O25b:H4, found throughout the world, carry FimH30 [18]. A recent study has shown that this clone display a greater adherence to CaCo2 enterocytes compared to other ESBL-producing E. coli isolates, although the specific role of FimH30 was not assessed [19]. It is possible that FimH30 allele confer an advantage to strains of the D-1 cluster for gut colonization, thus aiding their expansion.
Antimicrobial resistance was limited to the production of penicillinase (encoded by blaTEM-1 except for one strain), resistance to streptomycin, tetracyclin and sulphonamid and about one third of strains were devoid of any acquired resistance mechanism. We found no association between resistance and a particular WGS-subgroup or with the presence of pS88 virulence factors, suggesting that virulence and resistance genes are harbored by different plasmids or genomic islands such as Integrative and Conjugative Elements (ICE).

Conclusions
This work describes the recent dominance of the meningitis-causing E. coli O1:K1:H7, belonging to the highly virulent STc95, in France. We used WGS to describe its genetic diversity and found the predominance of a major cluster (D-1), characterized by the presence of the pS88-like plasmid. Clinical data and an ex vivo model failed to show higher pathogenicity of strains belonging to this major cluster D-1. Whether its high prevalence results from a high capacity to colonize the gut, due to adhesion through the FimH30 allele and/or the presence of the pS88-like plasmid, are yet to be determined.

E. coli isolates and clinical data
Between January 1, 1998 and December 31, 2015, 250 E. coli strains responsible for neonatal meningitis in France were collected by the E. coli national reference center (NRC). All strains were subjected to serogroup and svg PCR [20] [see Additional file 2]. By this method, 38 O1:K1 E. coli of STc95 were identified among these isolates and included in the study for WGS.
Clinical and biological data (age and sex of the patient, birth term, birth weight, date of onset of the infection, blood culture, and biological characteristics of the lumbar puncture) were collected.

Whole-genome sequencing
The whole genomes of the 38 E. coli isolates were sequenced. The Nextera XT kit (Illumina) was used to prepare libraries. Sequencing was performed on a MiSeq instrument using reagent kit V3 600 cycles (Illumina technology). Sequences were analyzed using the Center of Genomic Epidemiology (CGE) website (www.genomicepidemiology.org). Genes were annotated using Rast® software (version 2.0), and the sequences aligned and compared using BioEdit® Sequence Alignment Editor (V7.2.5).

Sequencing data quality
The SPAdes assembler was used to construct assemblies. Contigs < 500 bp were removed. The quality of the de novo assemblies was estimated using standard metrics [see Additional file 3]. Mean N50 were of 127,597 and mean N75 were of 70,794. Finally, reads used to construct the assemblies were remapped against the assembly contigs to visualize coverage. Mean coverage was of 37.

STc95 and subgroups
The CGE website was used to confirm the ST of the strains using the MLST tool [21]. In silico PCR was performed to determine STc95 subgroups (A,B,C,D and E), as described by Gordon et al. [9].
Phylogeny SNP (single nucleotide polymorphism) calling was performed with CSI Phylogeny (version 1.4) available on CGE web site (https://cge.cbs.dtu.dk/services/CSIPhylogeny/). CFT073 strain (O6:K5:H1) was used as outgroup in order to restrict our analysis to core genome SNPs. A phylogenetic tree, based on concatenated alignment of the 19,547 SNPs identified, was constructed by the Neighbor-Joining method available on MEGA software (version 3.1) with 100 bootstrap replicates [22].

Detection of putative virulence genes and genes encoding antimicrobial resistance
Genes encoding antimicrobial resistance and putative virulence factors were identified using the Resfinder and VirulenceFinder tools available on CGE website [23]. Moreover, 166 genes [Additional file 4] were also assessed by local blast analysis using the NCBI blast tool (Blast+ version 2.2.31). Genes with an alignment coverage higher than 90% and a homology > 90% were considered to be present. Strains harboring > 90% of plasmid pS88 sequence were considered as harboring a pS88-like plasmid. The FimH allele nomenclature of Sokurenko et al. was used [16].

Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.