INTRODUCTION
Bacteraemia caused by Escherichia coli are particularly frequent and severe, with a case-fatality rate of 5–30% [Reference Russo and Johnson1, Reference Lefort2]. Furthermore, the spread in recent years of isolates producing extended-spectrum β-lactamases (ESBLs) and carbapenemases that are often resistant to most of the available antibiotic classes have worsened the clinical impact of E. coli bacteraemia [Reference Rodriguez-Bano3–Reference Doi5]. In contrast to this, E. coli remains the most prevalent aerobic bacteria of the human digestive tract [Reference Tenaillon6], reaching a density of 108 colony-forming units per gram of faeces [Reference Smati7].
It is commonly accepted that the digestive tract is the reservoir of extra-intestinal pathogenic E. coli (ExPEC) strains [Reference Moreno8–Reference Krawczyk10], as well as of E. coli resistant strains [Reference Carlet11]. In contrast to the uniqueness of this reservoir, E. coli bacteraemia is characterized by the variability of its pathophysiological processes, which determine the severity of disease [Reference Lefort2]. Of these, two portals of entry are prevalent: the urinary tract during the course of pyelonephritis, prostatitis or renal abscess, and the digestive tract by translocation and various intra-abdominal infections [Reference Lefort2].
Several epidemiological studies comparing pathogenic and commensal E. coli strains have documented that urinary tract infection (UTI) strains, especially those isolated from pyelonephritis, belong to particular clones characterized by specific phylogenetic backgrounds, a high content of virulence factors (VFs) and a few O-types, arguing for a specialization of these clones in the high diversity of commensal strains that exhibit a low level of VFs [Reference Caugant12–Reference Kudinha20]. Less data are available for non-urinary portal-of-entry bacteraemic strains, especially from those with a digestive portal of entry [Reference Johnson21–Reference Sannes23]. Furthermore, these latter studies are based on a relatively small series of strains, and with one exception [Reference Sannes23], the patients and control subjects used for the commensal strains were not carefully matched.
To gain insight into the pathogenesis of E. coli bacteraemia, we performed a large epidemiological study comparing 243 commensal strains to 243 bacteraemic strains isolated from cases with a variety of portal of entry, including both urinary tract and digestive tract, from adult hosts matched in terms of gender and age, and from similar location and epoch. We demonstrate that the strains isolated from bacteraemia are more resistant to antibiotics than the commensal intestinal strains. Furthermore, the bacteraemic strains of urinary tract origin are a subset of the diversity of the commensal intestinal strains that has been selected for, whereas the strains isolated from bacteraemia of digestive origin encompass the same diversity as the commensal strains.
MATERIALS AND METHODS
Strain collections
Two previously published collections of E. coli sensu stricto were used. One is the COLIVILLE collection of 243 E. coli strains isolated from the stools of 243 community adult subjects living in the Paris area in 2010 [Reference Massot24], which was thereafter named ‘commensal collection’ (approved by the ethics evaluation committee of Institut National de la Santé et de la Recherche Médicale; CCTIRS no. 09·243). These can be considered as ‘true’ commensal strains, as strict inclusion criteria were applied to selection of the subjects, excluding individuals who had undergone any antibiotic therapy in the preceding month and any hospitalization in the 3 months preceding inclusion. In order to avoid any epidemiological bias, the recruited subjects belonged to independent families. The other is the COLIBAFI collection of 373 E. coli clinical strains isolated from the blood of 373 patients during the course of E. coli bacteraemia in the Paris area in 2005 [Reference Lefort2], which was thereafter named ‘clinical collection’ (approved by the ethics evaluation committee of the Hôpital Saint Louis, Paris, France; no. 2004–06). The patients included were from seven distinct hospitals and a large diversity of wards. The portal of entry of these bacteraemia was established according to the compatible clinical and/or radiographic features and the isolation of E. coli from the presumed source of infection. When E. coli could not be isolated from the presumed portal of entry, this latter was assigned on the basis of firm clinical suspicion, provided that all other possible sources of infection had been excluded. If the clinical data were ambiguous, the portal of entry was categorized as ‘undetermined’ [Reference Lefort2].
As it is known that E. coli epidemiology is related to the age and the sex of the subjects [Reference Gordon, Stern and Collignon25], each strain of the commensal collection was paired with a clinical collection strain (n = 243). Matching with a 1:1 ratio has been done on gender and minimizing age, with a median age difference of 1 year and an interquartile range (IQR) of 0–9 years. Moreover, two subsets according to the urinary (n = 138) and digestive (n = 60) portal of entry for the E. coli strains of the clinical collection were assessed with their pairs of the commensal collection.
Strain characterization
A successive refinement strain-typing strategy was applied on the two collections from the seven main phylogenetic groupings (A, B1, B2, C, D, E, F) [Reference Clermont26] to the characterization of particular clones of medical and epidemiological importance by a combination of PCR-based methods [Reference Skurnik27–Reference Johnson29] and O-typing [Reference Guinée, Jansen, Leeww and Guinée30, Reference Clermont31]. Thus, the B1 clonal complex 87 (CC87) (Institut Pasteur MLST schema nomenclature corresponding to ST58 and ST155 in the Achtman schema [Reference Skurnik27]), the 10 main B2 subgroups and the clonal group A (CGA) from the D phylogroup were determined in the strains. The exhaustive correspondence between this typing approach and STc membership according to the currently used Achtman MLST schema [Reference Wirth32] is available in Clermont et al. [Reference Clermont, Gordon and Denamur33].
The presence of 19 VFs representing the four main classes for E. coli extraintestinal virulence, including adhesins/invasin, iron capture systems, toxins and protectins was determined as in previous studies [Reference Clermont31, Reference Johnson34]. The ibeA gene is implicated in invasion but also indirectly in adhesion [Reference Cortes35].
The resistance to seven antibiotics was determined according to the recommendations of the Antimicrobial Committee of the French Society for Microbiology (http://www.sfm-microbiologie.org) [amoxicillin, amoxicillin-clavulanic acid, cefoxitin, cefotaxime, amikacin, ofloxacin, cotrimoxazole (sulfamethoxazole-trimethoprim)].
The bacterial characteristics listed above were already available for the strains of the commensal collection [Reference Massot24], except for CC87 membership. For the strains of the clinical collection, membership in the seven main E. coli phylogroups, membership in the B1, B2 and D main clonal complexes, presence of the clbQ VF-coding gene and O-antigen determination were performed in this study, with the remaining characters being already available [Reference Lefort2].
Statistical analyses
The bacterial characteristics described above were compared between the commensal and clinical collections. The same characteristics were compared in the specific clonal complexes/subgroups of the B2 and D phylogroups. A comparison was also performed according to the portal of entry (urinary- and digestive-source bacteraemia) for the E. coli strains of the clinical collection with their paired E. coli strains of the commensal collection.
Comparisons between the two collections were performed using non-parametric tests (Wilxocon rank-sum test or Fisher's exact test). As multiple tests were performed, the P values were adjusted using the Benjamini & Hochberg method [Reference Benjamini and Hochberg36] for each type of characteristic (phylogroup, VF, antibioresistance). For the phylogroup characteristics, if significant comparisons were found within all and each subgroup strains, the multiple test correction was performed at two levels: phylogenetic groups and subgroups. Analyses were performed with SAS v. 9.4 (SAS Institute Inc., USA). Data are shown as medians with IQR for continuous variables except when otherwise stated, and number/percentage for categorical variables. All tests were two-sided with a type-I error fixed to 0·05.
RESULTS
Host characteristics
Two hundred and forty-three subjects and 243 patients were included in this study from the commensal and clinical cohorts, respectively. The sex ratio was 0·83 in each cohort, corresponding to 110 men and 133 women for both commensal and clinical cohorts. The mean ages were 56·4 years [±standard deviation (s.d.) 11·3, range 21–86] and 59·0 years (±s.d. 15·1, range 20–86) for the commensal and clinical cohorts, respectively. The portal of entry of the bacteraemia episodes in the clinical collection were urinary tract (138, 56·8%), digestive (60, 24·7%) and miscellaneous and undetermined (45, 18·5%).
Global comparison of the two collections of strains, i.e. bacteraemic vs. commensal strains
We first genotyped the strains of both collections at the phylogroup and clonal complex levels (Table 1). When the clinical collection strain phylogenetic distribution was compared to that of the commensal collection, clear differences were observed for all except B1 phylogroups. Phylogroups B2, C and D were over-represented in the clinical strains relative to the commensal strains, whereas the A, E and F phylogroup strains were under-represented. Of note, within the B2 phylogroup strains, only subgroup IV (STc141) strains were under-represented in the clinical strains. Within the B2 phylogroup, we had more thoroughly characterized the strains at the clonal level by a combination of subgroup and O-type, as proposed previously [Reference Bidet37, Reference Burdet38] (i.e. I-O25b) (Table 2, descriptive data). Some clones (I-O25b, VI-O4, VII-O23, IX-O1, IX-O45a) appeared to be over-represented in the clinical strains, whereas the IV-O2b clone was under-represented, and not a single VIII-O81 strain was isolated from the clinical samples. The proportion of D phylogroup strains that were CGA, an antimicrobial-resistant extra-intestinal pathogenic clonal complex, was not significantly different between the two collections, as well as the proportion of B1 phylogroup strains that were CC87, a newly reported antimicrobial-resistant clonal complex of animal origin spreading in humans [Reference Skurnik27] (Table 1).
Phylogenic group and subgroup distribution in E. coli strains of the commensal and clinical collections
* The correspondence with the Achtman MLST schema [Reference Wirth32] is as follow: CC87 = STc155; B2 subgroup I = STc131; B2 subgroup II = STc73; B2 subgroup IV = STc141; B2 subgroup IX = STc95; CGA = STc69 [Reference Clermont, Gordon and Denamur33].
† Global P values for the comparison of all phylogroups between the two collections: all strains P < 0·001; matched urinary source P < 0·001; matched digestive source P = 0·73.
‡ Global P values for the comparison of B2 subgroups between the two collections: all strains P = 0·002; matched urinary source P < 0·001; matched digestive source P = 0·83.
§ Non-adjusted P value for the comparison of CC87 subgroup between the two collections: all strains P = 0·32; matched urinary source P = 0·64; not done for matched digestive source.
|| Non-adjusted P value for the comparison of CGA subgroup between the two collections: all strains P = 0·89; matched urinary source P = 0·27; not done for matched digestive source.
Characterization of the main B2 clones using a combination of subgroup and O-type in E. coli strains of the commensal and clinical collections
* The roman number corresponds to the B2 subgroup. The correspondence with the Achtman MLST schema [Reference Wirth32] is as follow: subgroup I = STc131; subgroup II = STc73; subgroup III = STc127; subgroup IV = STc141; subgroup VI = STc12; subgroup VII = STc14; subgroup VIII = STc452; subgroup IX = STc95 [Reference Clermont, Gordon and Denamur33].
† The numbers in parentheses correspond to the percentages of each clone within the corresponding subgroup.
We next analysed the presence of 19 VFs in the two collections (Table 3). We observed a highly significant over-representation of the VFs involved in iron acquisition (all except ireA) in the clinical strains. This over-representation of VFs was less substantial for adhesins, protectins and toxins. Markedly, the ibeA invasin/adhesin coding gene was the only VF under-represented in the clinical strains (Table 3). Globally, the VF score was significantly higher in the clinical strains (8, IQR 5–11) than in the commensal strains (6, IQR 2–9). As it is well known that B2 strains are enriched in VFs [Reference Picard39], we then compared the VF proportions within the B2 strains (Table 3). It appeared that the over-representation of VFs in the clinical strains was less pronounced for the iron acquisition VFs and adhesins, and disappeared for the protectins and toxins. The virulence score was similar [11 (IQR 8–13) and 11 (IQR 9–14)] for the commensal and clinical collections of the B2 subgroups, respectively (Table 3).
Virulence factors detected in E. coli strains of the commensal and clinical collections for all strains (n = 243 in each collection) and in the B2 phylogroup strains (n = 84 and n = 129 in each collection, respectively)
* Except where otherwise stated.
† Adjusted P value for the comparison of the 20 virulence factors within each subgroup of strains.
‡ Distinct alleles of the papG virulence gene.
§ Wilcoxon rank-sum test.
Finally, we compared the antibiotic resistance of the two collections (Fig. 1). For all the studied antibiotics, we observed a higher frequency of resistance in the clinical strains than in the commensal strains.
Bacterial resistance to seven antibiotics in E. coli strains of the commensal (COLIVILLE) and clinical (COLIBAFI) collections for all strains (n = 243 in each collection). Histograms represent proportion of resistance to amoxicillin (AMX), amoxicillin-clavulanic acid (AMC), cefoxitin (FOX), cefoxatime (CTX), amikacin (AMK), cotrimoxazole (SXT) and ofloxacin (OFX). The asterisk (*) indicates a significant difference between the two collections.
Comparisons of commensal vs. urinary portal of entry and vs. digestive portal of entry bacteraemic strains
The two main portals of entry of bacteraemia are the urinary tract and the digestive tract. We therefore undertook a comparison according to the portal of entry for the E. coli strains of the clinical collection (urinary tract = 138 strains, digestive tract = 60 strains) with their paired E. coli strains from the commensal collection.
When comparing the phylogenetic distribution, the differences observed between the strains of the two collections were even more marked for the urinary tract origin bacteraemia, with B2 phylogroup strains reaching 64·5% and A phylogroup strains decreasing to 5·1% (Table 1). Similarly, within the B2 phylogroup, clones I-O25b, II-O6a, VI-O4, IX-O1 and IX-O45a were over-represented in the bacteraemic strains of urinary origin (Table 2, descriptive data). In contrast, no significant difference was observed in the representation of the different phylogroups and subgroups between the commensal strains and bacteraemic strains of digestive origin (Table 1).
The comparison of the content of 19 VFs yielded similar trends to those observed for the phylogroup distribution. All five VFs involved in iron acquisition two adhesins, two protectins and two toxins were over-represented in urinary tract origin bacteraemia strains, whereas the ibeA gene was under-represented (Fig. 2a ). Globally, the VF score of the urinary clinical strains was 11 (IQR 7–12), vs. 5 (IQR 2–9) (P = 0·006) for commensal strains. When comparing only B2 phylogenetic group strains, the over-representation of VFs was overall less pronounced, with a virulence score of 11 (IQR 8–12) and 11 (IQR 11–14) (P < 0·01) in the commensal and urinary clinical strains, respectively. The protectin, toxin and iroN gene content was not found to be differentially represented (Supplementary Table S1). Contrasting sharply with the strains of urinary tract origin, the clinical strains of digestive origin were not found to be different from the commensal strains. Globally, the VF scores were 6 (IQR 2–10) and 5 (IQR 2–8) (P = 0·68) for the digestive clinical and commensal strains, respectively. This held true whether individual factors of all the strains were analysed (Fig. 2b ), or when only B2 group strains were taken into consideration (Supplementary Table S1).
Virulence factor genes detected in E. coli strains of the commensal (COLIVILLE) and clinical (COLIBAFI) collections according to the portal of entry of the bacteraemia. (a) Urinary portal of entry (n = 138 in each collection). (b) Digestive portal of entry (n = 60 in each collection). Histograms represent proportion of virulence factors grouped by function. The asterisk (*) indicates a significant difference between the two collections.
The differences observed for antibiotic resistance prevalence between the commensal and clinical collections were found to be present in the same proportions between the commensal and the bacteraemic strains of urinary and digestive origins (Table S2), contrasting with the analogous finding regarding virulence gene content.
DISCUSSION
In this work, we compared commensal and bacteraemic E. coli strains to gain insight into the pathogenesis of E. coli bacteraemia. The strengths of our epidemiological study were (i) the size of the sampling (243 commensal vs. 243 bacteraemic strains), allowing statistical analysis of sub-populations, (ii) the unrelated character of the strains within each collection, (iii), the ‘true’ commensal character of the commensal collection [Reference Massot24], (iv) the matching of the two populations of hosts in terms of sex and age, and (v) the homogeneity of epoch and geographical location of the study.
The first characteristic that sharply differentiated the populations of commensal and clinical strains was their acquired resistance to all the antibiotics tested, whatever the plasmidic or chromosomal origin of this resistance (Fig. 1) and despite the fact that the acquired resistance of E. coli increases with time [Reference Massot24] and that the strains of the clinical collection were recruited 5 years before those of the commensal collection. Moreover, this difference was observed whatever the portal of entry of the bacteria (Supplementary Table S2). This difference can be due to different ecological sources (commensal vs. clinical) or more probably to the antibiotic selective pressure applied on the microbiota of the patients or to the nosocomial contamination of the patients by resistant strains before the bacteraemic episode [Reference de Lastours40]. Thus, we evidenced 9·7% of patients receiving antibiotics within 2 weeks preceding the bacteraemia and 23·9% of nosocomial infections in the 243 COLIBAFI patients (Supplementary Table S3).
The second characteristic that distinguished the strains according to their commensal or clinical origin was their phylogenetic distribution and VF content, two traits previously demonstrated to be linked [Reference Picard39]. The newly described phylogroups E and F were under-represented in the clinical strains, whereas phylogroup C was over-represented (Table 1). This latter phylogroup encompasses strains virulent in a mouse model of sepsis, which was isolated during the course of a cluster of neonatal meningitis [Reference Moissenet41]. Moreover, this clonal complex is also prone to be antibiotic resistant [Reference Massot24]. An over-representation of strains belonging to the B2 and D phylogroups was noted in the clinical strains, as reported previously [Reference Sannes23, Reference Picard39].
Within phylogroup B2, the subgroup IV, corresponding to STc141 according to the Achtman schema [Reference Wirth32], was under-represented in the clinical isolates, and no subgroup VIII (STc452) strain was identified in the clinical strains (Table 2). These particular subgroups could correspond to commensal clones with low level of human invasiveness. This hypothesis is corroborated by the fact that the strains of subgroup VIII are avirulent in the mouse model of sepsis [Reference Clermont42]. The proportions of subgroups I (STc131), II (STc73) and IX (STc95) within the B2 phylogroup and of CGA (STc69) within the D phylogroup were similar in both commensal and clinical strains, despite that group B2 (and to a lesser extent group D) as a whole were more prevalent in the clinical isolates than the commensal isolates (Table 1). This is corroborated to the fact that these four STcs are also those most frequently isolated in extra-intestinal infections worldwide [Reference Bengtsson43–Reference Alhashash45].
As expected, the proportion of VFs, especially those involved in iron acquisition, was higher in the clinical than commensal strains. However, this difference was less marked within the B2 phylogroup strains (Table 3). Of note, the strains of B2 subgroup IV often possessed the ibeA gene (data not shown), explaining its under-representation within the clinical strains.
The most marked feature of our study was that, except for antibiotic resistance, the observed differences between the characteristics of the commensal and bacteraemic strains depend on the portal of entry of bacteraemia, i.e. urinary vs. digestive tract. Previous studies have identified differences in phylogenetic background and VF repertoire between urine-source versus other-source blood isolates [Reference Lefort2, Reference Johnson21, Reference Wang22]. A clear specialization of the strains isolated during a bacteraemia of urinary tract origin was revealed by a marked reduction in prevalence of A phylogroup strains and an increase of specific phylogenetic group B2 clones (Table 2), exhibiting numerous VFs (Supplementary Table S1). These clones exhibiting O-specific antigens were previously found to be associated with UTI (O-UTI group) [Reference Johnson46]. Conversely, no significant difference in terms of phylogroup distribution or VF content was found between commensal and bacteraemic strains of digestive tract origin (Table 1 and Fig. 2b ). Additionally, a relationship between the portal of entry and host characteristics was evidenced. The patients having a bacteraemia of digestive origin were more often males and with predisposing conditions such as solid cancer, tobacco addiction and immunosuppression than those having a bacteraemia of urinary origin (Supplementary Table S3).
Our data provide evidence for two distinct models [Reference Moreno8] of pathophysiology in bacteraemia, according to the portal of entry and the host characteristics: (i) the special pathogenicity model in the case of urinary tract origin, implicating a subset of highly specialized strains selected from the intestinal tract which are able to cross several anatomical and physiological barriers and occurring mostly in females, and (ii) the prevalence model in the case of digestive origin, implying that more prevalent faecal E. coli clones are able to translocate into the bloodstream in hosts with predisposing conditions. This duality in pathogenesis is also linked to death at 28 days; with a urinary tract portal of entry and the presence of the ireA gene involved in iron capture being negatively correlated with death [Reference Lefort2] (Supplementary Table S3).
This study had several limitations. First, despite the substantial size of the cohorts, their divisions into subgroups according to several variables resulted in the comparison of series of small size, which is not always statistically relevant. Second, we did not performed an exhaustive characterization of the clones, which could be refined by complete multilocus sequence typing and ultimately whole-genome sequencing. This fine-scale characterization of the strains will be the topic of a further study. Last, we did not focus our study on the relationship of host characteristics with bacterial characteristics and clinical outcomes.
CONCLUSIONS
By comparing bacteraemic to rigorously selected community commensal E. coli strains, we were able to demonstrate two levels of specialization in the bacteraemic strains: (i) an acquired resistance to antibiotics, which concerns the bacteraemic strains regardless of their portal of entry and (ii) an intrinsic virulence characterized by membership of specific clonal groups which are adapted to the urinary portal of entry. These epidemiological data provided additional evidence for niche specialization within the E. coli species.
APPENDIX
The COLIVILLE group is composed of the following medical practitioners who were involved in the recruitment of subjects for the study: Monique Allouche, Jean-Pierre Aubert, Isabelle Aubin, Ghislaine Audran, Dan Baruch, Philippe Birembaux, Max Budowski, Emilie Chemla, Alain Eddi, Marc Frarier, Eric Galam, Julien Gelly, Serge Joly, Jean-François Millet, Michel Nougairede, Nadja Pillon, Guy Septavaux, Catherine Szwebel, Philippe Vellard, Raymond Wakim, Xavier Watelet and Philippe Zerr.
The COLIBAFI group includes the following individuals. Clinical investigators are Michel Wolff, Loubna Alavoine, Xavier Duval, David Skurnik, Paul-Louis Woerther, Antoine Andremont (CHU Bichat-Claude-Bernard, Paris); Etienne Carbonnelle, Olivier Lortholary, Xavier Nassif (CHU Necker-Enfants Malades, Paris); Sophie Abgrall, Françoise Jaureguy, Bertrand Picard (CHU Avicenne, Bobigny); Véronique Houdouin, Yannick Aujard, Stéphane Bonacorsi, Edouard Bingen, Chloé Lemaitre, Romain Basmaci (CHU Robert-Debré, Paris); Agnès Meybeck, Guilène Barnaud, Catherine Branger (CHU Louis-Mourier, Colombes); Agnès Lefort, Bruno Fantin, Claire Bellier, Frédéric Bert, Marie-Hélène Nicolas-Chanoine (CHU Beaujon, Clichy); Bernard Page, Julie Cremniter, Jean-Louis Gaillard (CHU Ambroise-Paré, Boulogne-Billancourt); Bernard Garo, Séverine Ansart, Geneviève Herry-Arnaud, Didier Tandé (CHU Brest, Brest); Jean-Claude Renet, René Ze Bekolo, Renaud Verdon, Roland Leclercq (CHU Caen, Caen); Claire de Gialluly, Jean-Marc Besnier, Laurent Mereghetti, Roland Quentin (CHU Tours, Tours); Achille Kouatchet, Alain Mercat, Marie Laure Joly-Guillou (CHU Angers, Angers); Catherine Dalebroux, Pascal Chavanet, Catherine Neuwirth (CHU Dijon, Dijon); Camille Colliard, Martin Dary, Gilles Potel, Jocelyne Caillon (CHU Nantes, Nantes); Françoise Leturdu, Jean-Pierre Sollet, Gaëtan Plantefève (CH Argenteuil, Argenteuil); Agnès de Patureaux, Pierre Tattevin, Pierre-Yves Donnio (CHU Rennes, Rennes), all in France. Those responsible for bacterial genotyping are Erick Denamur, Olivier Clermont, Christine Amorin, Jeremy Glodt (INSERM, UMR722, Université Paris-Diderot, Paris, France). Those responsible for methodology are Xavière Panhard, Ludovic Lassel, Quentin Dornic, France Mentré (AP-HP, Hôpital Bichat, UF de Biostatistiques, Paris, France), Estelle Marcault, Florence Tubach (CHU Bichat-Claude-Bernard, Paris, France).
SUPPLEMENTARY MATERIAL
For supplementary material accompanying this paper visit https://doi.org/10.1017/S0950268816003010.
ACKNOWLEDGMENTS
This work was partly funded by recurrent grants from INSERM and Université Paris Diderot. This work was partially supported by the Consellería de Cultura, Educación e Ordenación Universitaria, Xunta de Galicia (grant CN2012/303) and the European Regional Development Fund, ERDF.
We are grateful to Harry Kemble for careful editing of the English language.
DECLARATION OF INTEREST
None.
