Seven candidate genes encoding proteins with predicted esterase activity were identified, based on their respective PM and pI values, using the MaGe system 14 (aes 15 , yddV, glpQ, ndk, yzzH and cpdA). Of these, Aes exhibited several characteristics particularly reminiscent of esterase B: i) a major esterase domain, ii) a theoretical pI of 4.72 for the K-12 strain protein (esterase B₁, pI ranging from 4.5 to 4.8) and 5.18 for CFT073 protein (esterase B₂, pI ranging from 4.85 to 5.0), and iii) the presence of a serine in the active site 9 . The inactivation of aes by gene disruption in K-12 MG1655 and CFT073 strains and complementation of the mutant strains with the aes gene confirmed that Aes was esterase B (Additional file 1: Fig. S1 and data not shown).

We then studied the correlation between Aes sequences and esterase B electrophoretic polymorphism. The comparison of the Aes phylogenetic tree with the theoretical and observed pI values and the esterase B electrophoretic mobilities (Mf values) for the 72 ECOR strains 10 is shown in Fig. 1. Overall analysis of the tree confirmed separation of esterase B into two variants: esterase B₁ and esterase B₂. Indeed, the Aes tree showed a clear distinction between Aes from the phylogenetic group B2 strains and Aes proteins from other strains, separated by a long branch, well supported by bootstrap (83%). Moreover, the characterisation of the phylogenetic group B2, based on Aes polymorphism, was consistent with the pI and Mf values of esterase B₂ (pI: 4.85 to 5.0 and Mf 57 to Mf 62), which were previously demonstrated to be specific to the phylogenetic group B2. Likewise, the characterisation of the phylogenetic groups A, B1 and D, based on Aes polymorphism, correlated with the pI and Mf values of esterase B₁ (pI: 4.60 to 4.80 and Mf 68 to Mf 72) 10 . Amino-acid substitutions detected from the branches of the Aes tree were analysed taking into account variation in esterase B mobility and pI values 16 (Fig. 1). In most cases, for the Aes phylogenetic group B2 strains, substitutions of acidic to neutral, neutral to basic or acidic to basic amino acids corresponded to increases in pI (from 4.85 to 5.00) and decreases in Mf values (from 62 to 57) among esterase B variants. However, there were some discrepancies. For example, the substitution of a basic amino acid in the ECOR 53 and 60 strains by a neutral amino acid in the ECOR 61 and 62 strains (R?C) corresponded to a faster migration in the ECOR 61 and 62 strains (Mf values 62 versus 60), with no effect on pI (4.85) (Fig. 1).

A more complex pattern of polymorphism was found among the A, B1 and D phylogenetic group strains. Taking the most frequent esterase B electrophoretic variant (pI: 4.60 and Mf 70) detected in the phylogenetic group A and D strains, an acidic to neutral amino-acid change (E?G) led to an increase in pI (from 4.60 to 4.75) and a decrease of Mf (from 70 to 68) of the esterase B variant, as expected. This amino-acid change was detected in 11 strains in the phylogenetic group A (Fig. 1). In contrast, several discrepancies were found among strains belonging to the phylogenetic B1 group: Aes polymorphism included several substitutions of neutral to neutral amino acids but with increased pI values (from 4.60 to 4.75) and in some cases paradoxical increases of Mf values (from 70 to 72) was observed (Fig. 1). These apparent discrepancies may be due to the effects of conformational or post-translational modifications of the protein.

To determine the evolutionary history of aes, we tested for selection using the aes sequence from 78 studied strains. First, we used a one-ratio model (M0) to estimate the average ratio ω (dN/dS) for all sites and all lineages at 0.18. The likelihood ratio test suggested that aes was under strong global purifying selection (compared to the neutral hypothesis which is ω = 0). The M1a, M2a, M7 and M8 models, estimating the selection on specific codons, confirmed that the vast majority (91%) of the sites are under negative selection. Finally, the branch-site model A did not detect positive selection along the branch separating group B2 from group non-B2 strains. Thus, the use of several tests for selection has shown that the evolution of aes has been driven essentially by purifying selection, as observed for the housekeeping genes used in the MLST (data not shown); no positively selected site was identified for the branch separating esterase B₂ from esterase B₁ strains. Previous studies based on whole genome sequencing data using PAML have not identified aes to be under positive selection 17 18 .

Visual comparison of the phylogenetic history of aes with that of the six concatenated housekeeping genes, reflecting the species phylogeny, revealed a similar topology with four main phylogenetic groups (Fig. 2). Indeed, all strains belonging to the B2 phylogenetic group were clustered in a monophyletic group (bootstrap 99%) with ECOR 66 at its base, as observed in the MLST tree. Likewise, two sub-groups were observed for phylogenetic group D, one of which was associated with the phylogenetic group B2 (ECOR 35, 36, 38, 39, 40, 41) (bootstrap 85%), also observed in the MLST tree. Phylogenetic group A also constituted two sub-groups, although these were not sister groups. By contrast, the B1 phylogenetic group was monophyletic overall, with only two strains (ECOR 4 and ECOR 47) clearly misclassified (Fig. 2).

We used a recently developed technique ("TreeOfTree") allowing the level of congruence between phylogenetic trees to be tested 19 . We tested each individual housekeeping gene tree, the MLST tree, and the aes tree. All the bootstraps are low enough (less than 67%) to suggest that all the gene trees can be view as not incongruent, the aes gene tree itself clustering with pabB and trpA gene trees with very low bootstrap (44%) (Fig. 3). Thus, aes tree topology showed that aes is a powerful marker of the species phylogeny, as observed for each housekeeping gene used in the MLST scheme.

Aes B₁ and B₂ protein variants were then compared by protein modelling. We found that residues S 157, D 254 and H 284 had a geometry similar to that of the esterase catalytic site. Thirty-eight polymorphic sites were identified by comparing the 319 Aes amino-acid sequences obtained for the 72 ECOR strains and six reference strains. For four of these sites, variation has become fixed in both B₁ and B₂ types, with the identified residues differing between the two types at each site. These polymorphisms could thus be used to distinguish between the types: the B₁ conserved amino acids A 53, M 64, E 73 and C 78 correspond to the B₂ conserved amino acids V, R, K and Y, respectively. These four polymorphic sites were found on the long B2/non B2 branch in the proteic tree, explaining the observed high bootstrap (83%) (Fig. 1). Fig. 4 shows the location of 24 additional sites at the protein surface with observed amino-acid variants for either type B₁ (green) or type B₂ (red). No one site was polymorphic for both B₁ and B₂ types. But for all the polymorphic sites within types B₁ and B₂, some of the amino-acid variants are shared by the two types. Consequently, these sites cannot be considered to be specific to either one type or the other and cannot be used to distinguish between the two types of protein. Polymorphic sites were clustered, localised at the surface and were not found in the active site, consistent with previous observations of similarity in the catalytic activity of B₁ and B₂ esterases with synthetic substrates 7 9 . These differences in location of the polymorphic sites between the two variants support the divergence of the B2 phylogenetic group strains from the A, B1 and D phylogenetic groups strains within this species.

The previously observed correlation between electrophoretic esterase B polymorphism and the distinction between B2 and non-B2 phylogenetic group strains 10 - and thus with the extraintestinal virulence of the strains - suggested a putative role for the enzyme, or certain variants, as a virulence factor. The esterase B hydrolase function may have a direct role in the colonization or invasion of the eukaryotic cells as it was observed for esterases in other bacteria 20 21 . Indeed, esterase B₂ variants belonging to phylogenetic group B2 may confer higher levels of virulence to the strain during extraintestinal infection. There are several examples of proteins with variants playing different roles in extraintestinal infections: the adhesins FimH 22 , PapG 23 and the somatic antigen O 24 25 .

Previous studies of Aes have not demonstrated a role of the protein in virulence. Firstly, experimental studies characterising Aes as an enzyme with esterase activity have demonstrated the inhibitory interaction of Aes with MalT, a transcriptional regulator of the maltose regulon. These findings suggested a role for Aes in the regulation of maltose metabolism 15 26 . Aes may also play a role in the regulation of raffinose metabolism by inhibiting α-galactosidase 27 . However, these data were obtained from overexpression of aes from plasmids, thus raising the question of their relevance in vivo. An illustration of aes overexpression from the plasmid pACS2 28 is shown in Additional file 1: Fig. S1. Secondly, a previous study of aes expression in the K-12 strain in vitro did not find significant effects on expression under the various metabolic, stress or environmental conditions tested http://genexpdb.ou.edu/, with the exception of aes overexpression observed in strains cultured in the presence of acetate 29 . Interestingly, esterase B exhibits Michaelis-Menten kinetics for the hydrolysis of 1-naphtyl acetate 9 . Finally, aes expression was found to be homogeneous across 10 representative strains of E. coli/Shigella cultured in 869 medium 30 .

Our previous findings from the study of the genetic sequence surrounding aes did not suggest a role for the encoded protein in virulence. Indeed, comparisons, using the MaGe system, of 75 kbp of sequence upstream and downstream from aes in the 20 strains of E. coli 31 showed that aes is not located in/or adjacent to any regions linked to extraintestinal pathogenicity specific to B2 strains (Additional file 2: Table S1).

To gain insight into Aes function we tested the mutants under different conditions. Firstly, we studied the in vitro growth of parent-type strains and their respective mutants on several carbon sources. We did not observe any difference between parent-type strains K-12 or CFT073 and their respective mutants K-12 Δaes and CFT073 Δaes in competition studies with LB and gluconate minimum media (data not shown). Additionally, growth of the strains CFT073, K-12, CFT073 Δaes and K-12 Δaes, in the presence of different carbon sources, was the same for parent and mutant strains. These results suggested that Aes does not play a role in regulation of the growth of the strains in these conditions. Secondly, we studied whether Aes is involved in the virulence of E. coli in vivo using a septicaemia mouse model. Kaplan-Meyer curves obtained for CFT073 and its mutants CFT073 Δaes and CFT073 Δaes:Cm were similar, suggesting that Aes is not involved in the virulence process (p = 0.87) (Additional file 1: Fig. S2).

We used E. coli K-12 MG1655 (phylogenetic group A) and CFT073 (phylogenetic group B2) reference strains, their mutants, K-12 Δaes (obtained from the KEIO collection 34 ) and CFT073 Δaes (obtained during the course of this study) and the aes complemented mutant strains K-12 Δaes pACS2 28 andCFT073 Δaes pACS2 for the identification of the esterase B-encoding gene. The strains K-12 MG1655, CFT073 and their aes mutants were also used for the investigation of the putative role of esterase B. We used the 72 strains from the E. coli reference (ECOR) collection, encompassing commensal and pathogenic strains representative of the genetic diversity of the species 35 , and four additional pathogenic reference strains (536, UTI89, Sakaï and EDL 933) for the sequencing of aes. The E. fergusonii strain ATCC 35469^T, the most closely related species to E. coli 36 , was used as an outgroup.

The MaGe (Magnifying Genome) software program 14 was used for candidate gene selection and comparative analysis of genetic sequences surrounding aes. The MaGe software program allows gene annotation and comparative analysis of available E. coli and closely related genomes, with visualisation of E. coli genome annotations enhanced by a synchronized display of synteny groups in the other genomes chosen for comparison 14 . Protein motifs and domains can be identified using the InterPro databank 37 . Candidate genes were obtained after the selection of proteins showing esterase motifs and compatible molecular weights (from 15,000 to 60,000 Da) and pI values (from 4.0 to 5.5) 9 .

Inactivation was carried out as previously described 38 , using a PCR product obtained with primers aesW1 (5'-TTTCATGGCAGTGGTTCCTTACAATGACGTAATTTG AAAGGAGTTTTTGCGTTAGGCTGGAGCTGCTTC-3') and aesW2 (5'-GCCACGCCG GAACATATCGAAATGATGGCTAATCTTGTTGCCGCGTATCGCATATGAAATATCCTCCTTAG-3'). The PCR product contained (i) the FRT-flanked chloramphenicol acetyltransferase (cat) gene responsible for chloramphenicol resistance and (ii) the adjacent sequences homologous to the 5' and 3' flanking regions of aes. Inactivation of the aes gene was confirmed by PCR using the following primers: aes1 (5'-ACTGAGCGGCGAATGTTAACA-3') and aes2 (5'-ATTGTCTGGAGACGCTGGAA-3'), targeting sequences upstream and downstream from the aes gene, respectively; and c1 (5'-TTATACGCAAGGCGACAAGG-3') and c2 (5'-GATCTTCCGTCACAGGTAGG-3'), targeting sequences within cat gene. The antibiotic resistance gene was removed using the pCP20 plasmid 38 .

Complementation analysis of the mutant strains was carried out by electroporation of the multicopy plasmid pACS2 28 containing the aes gene under its native promoter.

The esterase B phenotype was investigated by vertical slab polyacrylamide gel electrophoresis of crude extracts of parent type, mutant and complemented mutant strains, using 12% (w/v) acrylamide and discontinous Tris/glycine buffer, pH 8.7. Esterase activity was detected by testing for the hydrolysis of 1-naphtyl acetate, as previously described 39 .

The aes gene was amplified by PCR, using the primers aes1 and aes2 (see above). The resulting 1250 bp PCR product was then sequenced by the Sanger method 40 . We compared aes sequences of 894 bp by sequence alignment using the ClustalW program 41 . The 72 aes sequences of the ECOR strains have GenBank accession numbers GQ167069 to GQ167140.

Amino-acid sequences deduced from the nucleotide sequences of aes were also analysed. After the generation of the maximum likelihood tree (see below), amino-acid substitutions for each branch of the Aes tree were identified by comparison of consensus sequences between different branches using the SEAVIEW program 42 .

We tested for selection with code ML, implemented in PAML 43 44 . Using a maximum likelihood algorithm, PAML assigns likelihood scores to the data according to the various models of selection. Assignment of a higher likelihood score to a model incorporating selection than to a null model without selection and a significative likelihood ratio test are indicative of selection. The overall Ka/Ks ratio (or ω, dN/dS), reflecting selective pressure on a protein-encoding gene, was estimated using the M0 model (one-ratio) 45 for all isolate sequences, with the E. fergusonii sequence as an outgroup. We also used the M1a (null) and M2a (positive selection) models 46 47 and the more powerful M7 and M8 models 46 48 to detect positive selection on specific codons (sites). We used the branch-site model A 47 49 for the B2/non-B2 partition. This model is based on the hypothesis that positive selection occurs only in certain branches/lineages.

Maximum-likelihood phylogenetic trees were all reconstructed using the PHYML program 50 and the GTR+G+I model. This general model is not necessarily the most parsimonious one. However, we also wanted to obtain the bootstrap support values for each partition. Given that (i) the most parsimonious model may differ from one bootstrap resampling to another, and (ii) a very long computer processing time would be required to choose the best model among the 88 possible models for each of the 500 resamplings, we chose a less time-consuming strategy, simply selecting the most general model (GTR+G+I) for all resamplings. We checked that the trees remain the same for the different models, whether the most parsimonious or the most general model is used. Additional file 2: Table S2 gives the different parsimonious models, and their estimated parameters, selected by the Akaike criterion (jMODELTEST version 0.1.1, written by Posada 51 , available at http://darwin.uvigo.es/software/jmodeltest.html).

We compared the phylogenetic history of aes to the phylogenetic history of the strains, based on the concatenated nucleotide sequences of six housekeeping genes (trpA, trpB, pabB, putP, icd and polB) and individual gene sequences, as described elsewhere 19 . Briefly, each phylogenetic tree T _i is firstly transformed into a tree-distance matrix D _i , the distance between two strains being the number of branches with positive length connecting them along the tree. The resulting tree distance matrix D _i allows the initial tree structure T _i to be recovered, independently of branch length. Two tree distance matrices (D _i and D _j ) (corresponding to two gene trees i and j) can be compared by calculating the Euclidian distance between them (δ _ij ) 52 . A low δ _ij value means that the similarity between the two tree distance matrices D _i and D _j is high, and, consequently, that their tree structures T _i and T _j are close. As several gene tree structures are compared through this Euclidian distance metric, a new distance matrix Δ can be built with the δ _ij elements. This Δ matrix can then be transformed into a "tree of gene trees" using a neighbour-joining algorithm 53 .

To obtain a support value for each partition of this tree, we applied this same procedure to 500 bootstrapped sets of data, obtaining 500 Δ matrices and finally, a bootstrapped consensus "tree of gene trees". A high bootstrap support value separating two sets of gene trees allows incongruent sets of gene trees to be identified; however, a low bootstrap value suggests that the two sets of trees are not incongruent or that there is insufficient phylogenetic information to reject the hypothesis of incongruence. The "TreeOfTree" package is available from the website http://bioinformatics.lif.univ-mrs.fr.

Modelling of the Aes protein structure was based on comparison of the available models from MODBASE 54 with models previously obtained using the Tasser-Lite homology modelling server 55 56 . Although some differences were observed between the models obtained by these two independent approaches, in particular in the N terminus region, the best models proposed by Tasser-Lite and MODBASE were similar overall. Given that our aim was to determine only the approximate location of the Aes polymorphism within the protein structure, the MODBASE model was used for further analysis. The model was finally tested to ensure that it contains an active site consistent with esterase activity. This was carried out using the 3D MSS-Sites program http://bioserv.rpbs.jussieu.fr/ 57 and the Catalytic Site Atlas 58 . Polymorphic sites were identified by sequence alignment using ClustalW 41 for B₁ and B₂ variants separately.

Theoritical pIs of Aes were calculated using the program compute pI of the ExPASY home page http://www.expasy.ch/tools/pi_tool.html.

Competition studies of parent strains K-12 and CFT073, with their respective mutants K-12 Δaes:Kan and CFT073 Δaes:Cm (1/1 ratio), were performed in Luria Bertani (LB) and gluconate minimum liquid media. Gluconate minimal medium mimics the intestinal environment 59 . For each medium and for each competition experiment, bacteria were plated on media with or without the appropriate antibiotic and counted after 2 h (exponential phase) and 18 h (stationary phase). Each experiment was repeated twice. Biolog GN2 (Biolog, Inc., Hayward, CA) plates were used to detect carbon utilisation of 95 substrates. Utilisation of various C sources is coupled to the reduction of a tetrazolium dye and generation of a purple colour 60 . Each strain was grown in LB medium, washed and resuspended to an optical density of 0.01 at 600 nm in mineral medium 60 . Plates were incubated at 37°C and colour changes were measured by changes in optical density (measured on a Tecan microplate reader) at a wavelength of 600 nm. The cut-off for positive results was an optical density of 0.2.

A mouse model of systemic infection was used to assess the intrinsic virulence of the strains 11 . For each strain, 10 outbred female swiss OF1 mice (3-4 weeks old, 14-16 g) were challenged with a standardized subcutaneous bacterial inoculum (2 × 10⁸ CFU of E. coli). Mortality was assessed over seven days following the challenge. Assays were performed using the CFT073 strain as a positive control (killing 10/10 mice), the K-12 strain as a negative control (killing 0/10 mice) 61 and the CFT073 Δaes and CFT073 Δaes:Cm mutant strains. Data were analysed using the StatView software to obtain Kaplan-Meyer curves; statistical analysis was carried out using the logrank test, with p values < 0.05 considered as significant.