Introduction

Lung is the first site of infection in intensive care unit (ICU) patients [1], and ventilator-associated pneumonia (VAP) is the most common life-threatening hospital-acquired infection despite considerable efforts to implement guidelines for its prevention [2, 3]. Consequences of VAP include increased duration of mechanical ventilation and length of ICU and hospital stay [4, 5], high morbidity and mortality [6], and increased antibiotic consumption and hospital costs [7]. Gram-negative bacteria (GNB) predominate in hospital-acquired pneumonia, and among them, Escherichia coli is the main Enterobacteriaceae involved in VAP [6]. E. coli is the most abundant aero-anaerobic commensal intestinal bacterium [8], with extraintestinal pathogenicity (extraintestinal pathogenic E. coli: ExPEC) [9]. Although a huge amount of data has accumulated on E. coli pathogenicity in intestinal, urologic, central nervous system, and bloodstream infections, there is, surprisingly, little data on lung infection [10], despite E. coli now being as often responsible for ICU infections as Pseudomonas aeruginosa [1]. Given the worrisome increase in multidrug resistance among GNB including E. coli [11], future therapeutic approaches will have to rely on other strategies including vaccine and immunotherapy [12, 13]. Greater knowledge of genotypic and phenotypic characteristics of E. coli responsible for VAP is the necessary first step in the search for potential vaccine candidates, as for P. aeruginosa [14, 15].

Phylogenetic analysis shows that E. coli population can be divided into four major groups (A, B1, B2, and D). Strains belonging to groups A and B1 are known to carry few genes encoding virulence factors (VF), with weak extraintestinal pathogenicity [16], and are commonly multidrug resistant [17]. On the other hand, group B2 and D strains typically express numerous VF mediating bacterial adhesion and protection, cell host invasion, toxicity, and iron uptake [10], usually exert high extraintestinal pathogenicity [16], and are generally more susceptible to antibiotics [17].

Phenotypic and genetic characteristics of ExPEC isolates have been extensively studied except for oropharyngeal and respiratory E. coli isolates in the ICU, for which there are no data. We, therefore, aimed to characterize E. coli respiratory isolates of ICU mechanically ventilated patients and compare them with their oropharyngeal and digestive counterparts. To better understand E. coli lung infection pathophysiology, an in vivo animal model of lung infection was used to assess pulmonary virulence of various representatives of these isolates.

Patients and methods

Patients

This prospective study was conducted in Louis-Mourier University Hospital medico-surgical ICU (Colombes, France) and approved by the ethics committee of the French National Society of Intensive Care (SRLF), which did not require informed consent since there was no change in practices and all procedures were already routinely performed. Patients or family/relatives were informed of the nature of the study, its purpose and objectives. During 22 months, all patients mechanically ventilated for 3 days or more were systematically sampled for simultaneous presence of E. coli in the digestive, oropharyngeal, and respiratory tract (and then once a week if still under mechanical ventilation). This screening was only possible during weekdays. Patients for whom day 3 occurred during the weekend were sampled on the Monday (unless they were extubated in the meantime). For each patient whose samples showed simultaneous colonization of the three sites, clinical, epidemiological, and laboratory data were recorded. Concomitant E. coli pneumonia was also reported.

E. coli isolates [see electronic supplementary material (ESM)]

Fecal isolates of E. coli were obtained by rectal swabbing performed during routine surveillance of multidrug-resistant bacteria carriage. Oropharyngeal isolates were obtained by swabbing the oropharyngeal cavity immediately before routine oropharyngeal care performed every 6 h on average in our unit.

Respiratory isolates were obtained from respiratory secretions retrieved either during routine tracheal suctioning or during fiberoptic bronchoscopy performed for VAP suspicion.

When all three sites were positive for E. coli, five colonies of each site were randomly picked [18], isolated, and stored at −80 °C in brain–heart infusion broth containing glycerol 20 %.

Isolates’ genetic background analysis (see ESM)

Triplex polymerase chain reaction (PCR) method [19] was used to determine the E. coli phylogenetic group membership (A, B1, B2, D). Multiplex PCR was used to detect genes encoding for seven frequently encountered extraintestinal VF, which belong to the main classes of VF (adhesin, toxin, and iron capture system) (see ESM Table 1) [20]. The VF score was defined as the number of VF present divided by the number of VF tested for each isolate.

Isolates’ intrasample genotypic diversity was investigated based on triplex PCR profile and/or VF gene content as in [21]. Different comparisons of intrasample diversity were made: between each site, between VAP and lung colonization, and between VAP and all other samples.

Pulsed-field gel electrophoresis (PFGE) was performed as in [22] on some isolates. VAP B2 phylogroup isolates were further characterized by presence of 11 additional VF [20], PCR O-typing [23], and B2-subgrouping [24].

Antimicrobial susceptibility (see ESM)

Antimicrobial susceptibility of each isolate was determined by disk-diffusion method according to the French Society of Microbiology. Susceptibility score was defined as the sum of active in vitro antimicrobial agents for each isolate. A score of 1 was attributed for a sensitive, 0.5 for an intermediary, and 0 for a resistant isolate, a higher score thus indicating a more sensitive isolate.

Infection/colonization definitions (see ESM)

Infection was defined as a microbiologically proven VAP. Colonization was defined as E. coli present in the respiratory sample but in the absence of clinically suspected or microbiologically confirmed VAP.

Animal study

Experiments were performed in pathogen-free 6–8-week-old male BALB/c mice (Janvier SA, France) in compliance with the recommendations of the French Ministry of Agriculture and approved by the French Veterinary Services. E. coli isolates were grown overnight (LBMedium®; BIO101 Inc., Carlsbad, CA, USA), and serial dilution was performed to obtain different inocula. To assess pulmonary virulence of various isolates, these microorganisms were inoculated via nasal route [25] under light anesthesia with pentobarbital (Sigma, France) and sevoflurane (Abbott, France). Preliminary experiments using a well-known ExPEC E. coli isolate (strain 536) [26] served to determine an inoculum that would yield 60 % mortality. This cutoff was chosen in order to be able to detect more or less virulent strains, being obtained with a 3.6 × 108 colony forming unit (CFU) inoculum per mouse. This value was therefore used for the experiments for each isolate.

Preliminary experiments with each strain tested were used to confirm histologically and microbiologically the presence of pneumonia. Animals were closely monitored for 96 h. Death never occurred after 48 h. This duration was subsequently used to assess mortality in our experiments. Each isolate was tested in 25 mice challenged intranasally with 3.6 × 108 CFU.

Statistical methods

Results are expressed as mean ± standard deviation (SD) if not indicated otherwise. The χ 2 test or Fisher’s exact test was used to compare categorical variables, and analysis of variance and t test were used for continuous variables. Probability of survival was assessed by Kaplan–Meier analysis. Correlation between content of VF genes and mortality was tested using nonlinear regression. GraphPad Prism 4 (GraphPad Software, San Diego, USA) was used. All tests were two-tailed, with P < 0.05 considered significant.

Results

Patients

During the study period, 891 patients were mechanically ventilated, a total of 381 for more than 3 days, of whom 132 patients were sampled. Others were not sampled mainly because they were extubated before the sampling could be performed. All rectal samples were positive for E. coli. Among those patients, 12 were also positive in the oropharynx (but negative for the lung) and 25 had positive cultures for E. coli at the three sites (rectum, oropharynx, and lung). Positive samples in the lung were always associated with positive oropharyngeal samples.

Five E. coli colonies were randomly picked from each sample. Figure 1 depicts the flowchart of the patients and the isolates. Based on the genotype and antibiotype characterization, a total of 115 distinct isolates were studied.

Fig. 1
figure 1

Flowchart of patients and isolates. One hundred fifteen distinct E. coli isolates were obtained from 343 studied isolates from 25 patients. Isolates were considered as identical when they exhibited identical PCR phylotyping pattern, VF gene content, and antibiotype

Full size image

Patients’ characteristics are displayed in Table 1. Twelve of these patients had documented E. coli VAP. No significant difference was found between patients with and without VAP in terms of characteristics (Table 1).

Table 1 Patient demographic and clinical characteristics
Full size table

Phylogenetic groups and VF genes

Considering all three sites, B2 was the main phylogenetic group (55 %), followed by group A (28 %), D (13 %), and B1 (3 %). The distribution pattern (Fig. 2) was similar among sites, although there was a trend towards a higher proportion of B2 in the respiratory tract than in the oropharyngeal and the rectal site (66, 51, and 49 %, respectively).

Fig. 2
figure 2

Distribution of phylogenetic groups A, B1, D, and B2 at the three sites where E. coli was sampled: rectum, oropharynx, and respiratory tract. The distribution pattern was similar among sites, although there was a trend towards a higher proportion of B2 in the respiratory tract than in the oropharyngeal and the rectal site (66, 51, and 49 %, respectively)

Full size image
Fig. 3
figure 3

a Survival curves of mice instilled with 3.6 × 108 CFU of different E. coli clinical isolates (25 mice per isolate). A significant difference was found (P < 0.001) among survival rates. B1, A, and B2 isolates with zero, one, and three virulence factor genes, respectively, exerted significantly less mortality that one group A or two group B2 isolates with, respectively, four, five, and seven virulence factor genes. b Relationship between mortality and content of virulence factor genes: a very strong correlation (R 2 = 0.99) was found between the number of virulence factor genes carried by E. coli isolates in the pneumonia mouse model and mortality. The best fit was observed with a sigmoidal curve, indicating that a minimum of virulence factor genes are required to induce mortality, and that maximum mortality was observed with seven virulence factor genes

Full size image

Table 2 details the distribution of VF genes among phylogenetic groups. All screened VF genes were different in B2 compared with non-B2 isolates: all but traT and aer genes were more prevalent in B2 isolates. Furthermore, B2 isolates had a significantly higher virulence score than non-B2 isolates (0.70 ± 0.18 versus 0.37 ± 0.25; P < 0.0001). There was a trend towards more isolates with a greater virulence score at the respiratory and oropharyngeal sites than the rectal site (data not shown). When considering respiratory samples only (Table 3), there was a trend toward more B2 strains and fewer A strains in isolates from patients with VAP than from those without. In addition, VAP isolates had a significantly higher virulence score than those coming from patients with colonization only. This also held true when only considering non-B2 isolates (data not shown). Among VF genes, iroN and sfa were significantly more present in VAP isolates than colonization isolates (Table 3). Further characterization of the 16 VAP B2 isolates showed that only two belonged to the frequently recovered subgroup II [multilocus sequence type complex (STc)73)] [27]. Of note, five isolates from four patients belonged to subgroup III (STc127), which encompasses the highly virulent archetypal strain 536 [26] (Table 4).

Table 2 Distribution of virulence factor genes according to the phylogenetic group of the 115 E. coli isolates sampled from the three sites
Full size table
Table 3 Phylogenetic group, content in virulence factor genes, and antibiotic susceptibility score of E. coli isolates from VAP and lung colonization
Full size table
Table 4 Extended characteristics of the 16 VAP B2 E. coli isolates
Full size table

Individual characteristics of the respiratory isolates of the 12 VAP patients are given in ESM Table 3. Two patients were considered to have VAP despite bacterial count below threshold. Antimicrobial agents were given in these patients because VAP was strongly suspected in view of a low PaO2/FiO2 ratio with new radiological infiltrate, no other site of infection, and favorable outcome with appropriate anti-E. coli agents. Of note, lung microbiological sampling in these patients was performed while patients had already started receiving antibiotics for VAP suspicion, a situation in which thresholds of bacterial count should be reduced to maintain validity [28].

Isolates’ intrasample genotypic diversity was assessed, and no difference was found; For example, the mean and range of phylogenetic group/subgroups among the five isolates originating from rectum, oropharynx, and lung were 1.14 (1–3), 1, and 1, respectively.

Antimicrobial susceptibility

Antimicrobial susceptibility to seven clinically relevant antibiotics is summarized in ESM Table 4. Of all isolates, 11 % were piperacillin–tazobactam resistant or intermediate and third-generation cephalosporin susceptible.

Two patients had extended-spectrum β-lactamase (ESBL)-producing E. coli isolates at the three sites. B2 isolates were significantly more sensitive than non-B2 isolates with a greater susceptibility score (15.0 ± 2.7 versus 8.6 ± 3.7; P < 0.0001).

In 12 patients, all rectal, oropharyngeal, and respiratory isolates held the same antimicrobial susceptibility profile. In the remaining 13, intrasite diversity was observed (ESM Fig. 1). Four patients with various antibiotic susceptibility profiles in respiratory samples (ESM Table 5) had VAP. To differentiate further, PFGE were performed, and for three of them, similar genomic identities were evidenced (less than 11 % difference), suggesting a monoclonal population with microheterogeneity. For the last patient, the PFGE pattern evidenced polyclonal infection. No other case of antimicrobial heteroresistance was seen in the respiratory samples.

Animal study

Representative photos of histological features of pneumonia obtained during preliminary experiments are displayed in ESM Fig. 2. Characteristics of the six representative isolates that were used are detailed in Table 5. Twenty-five mice were used per isolate, thus a total of 150 mice were tested. No death occurred before 6 h after inoculation. We found significant differences in survival rates in mice challenged with these isolates (Fig. 3a). The B2 isolate with the greatest content in VF genes (n = 7) induced the highest mortality. A B2 isolate carrying five VF genes had significantly higher mortality power than A, B1, or B2 isolates with fewer VF. We found a very strong positive correlation (R² = 0.99) between the numbers of VF genes carried by E. coli isolates and mortality induced in the pneumonia model (Fig. 3b).

Table 5 Characteristics of the six isolates assessed for mortality in the mouse pneumonia model
Full size table

Discussion

To the best of our knowledge, this is the first study showing genotypic and phenotypic characteristics of E. coli isolates responsible for VAP in ICU patients and comparing them with their rectal and oropharyngeal counterparts. We also assessed pathogenicity and virulence of representative isolates in a pneumonia mouse model. The main results can be summarized as follows: (1) multisite colonization with E. coli is frequent, observed in 19 % of our patients; (2) the majority of isolates belong to phylogenetic group B2, characterized by higher susceptibility to antimicrobial agents and greater content of VF genes than the other phylogenetic groups; (3) VAP isolates belong more often to group B2 and have higher VF content than colonization isolates; (4) VAP B2 isolates exhibit infrequent ExPEC sequence types; (5) antimicrobial resistance diversity is observed in respiratory samples of patients with E. coli VAP, an unprecedented finding; (6) high extraintestinal virulence of these clinical isolates responsible for VAP is confirmed in a mouse model of E. coli pneumonia, and content of VF genes and mice mortality are strongly correlated. These results may have significant impact in VAP prevention and treatment, given the growing prevalence of hospital-acquired infections due to GNB [29], by providing for the first time a detailed analysis of E. coli strains colonizing and infecting ICU patients. They may provide potential leads to investigate future therapeutic strategies such as those based on vaccine candidates [30–33].

A large-scale epidemiological study on infections in the ICU showed for the first time that E. coli was more frequent than P. aeruginosa in North America and as frequent in Western Europe [1]. More recent data even suggest that Enterobacteriaceae (and among them E. coli) are now more often responsible for VAP than P. aeruginosa and Staphylococcus aureus [34, 35]. Our local epidemiology of VAP is in agreement with these recent findings. Paradoxically, little is known about E. coli lung pathogenicity, apart from the inflammatory role of E. coli capsule and O-antigen [30] and hemolysin-induced surfactant dysfunction [36].

In order to investigate potential differences in E. coli isolates between intestinal commensalism, oropharyngeal colonization, respiratory tract colonization, and subsequent lung infection, we included in our study only patients whose samples were simultaneously positive for E. coli at these three sites. With this restriction in mind, we found that 25/132 (19 %) of our patients had multisite positive samples. Among them, 12 were diagnosed with E. coli VAP. Because of the specific aims of our study and therefore its design, comparison with existing data is difficult. However, our results expand those previously published dealing with GNB colonization, indicating increasing levels of colonization with increasing patient severity and subsequent risk for VAP [37–39].

The B2 group was the most frequent phylogenetic group at all sites (Fig. 2) with a trend towards a higher proportion in respiratory samples than oropharyngeal and rectal samples. Furthermore, these B2 isolates harbored more VF genes than isolates from other phylogenetic groups, in agreement with former findings [16]. This distribution is similar to what has been described in urinary tract infection, septicemia or neonatal meningitis [16, 40, 41], and reflects the greater virulence of B2 strains.

Testing of antibiotic susceptibility showed significantly higher antimicrobial susceptibility in B2, compared with non-B2, for both respiratory and nonrespiratory isolates. This is also in agreement with former findings in other settings [17]. Furthermore, this trade-off between virulence and resistance was also observed within the B2 isolates (ESM Fig. 3). Importantly, we identified 11 % of isolates that were resistant to the combination piperacillin–tazobactam but were still susceptible to third-generation cephalosporin (in relation to secretion of a high level of penicillinase). This was observed in both respiratory (2/35; 6 %) and nonrespiratory isolates (11/80; 14 %) and confirms previous reports on this phenomenon [42]. This finding bears potential therapeutic impact if piperacillin–tazobactam is used as first-line empirical therapy for Gram-negative VAP. Interestingly, we found that E. coli isolates retrieved from a single respiratory sample could yield different antimicrobial susceptibility patterns, highlighting antimicrobial resistance polymorphism in E. coli. Of note and bearing potential impact on VAP treatment, this diversity was only evidenced in respiratory samples from patients with pneumonia (4/12 patients; 33 %). PFGE was performed in order to test the clonality of the isolates and confirmed that three of these patients had monoclonal infection. In all other respiratory samples, colonies had the same antimicrobial resistance phenotype. Antimicrobial diversity resided in sensitivity to β-lactams in three patients, and in antimicrobial agents of lesser medical interest (trimethoprim or chloramphenicol) for two of these patients and for the remaining patient (ESM Table 5), and resulted probably from plasmid loss. This finding may explain clinical failure of antimicrobial treatment when the more susceptible strain has been picked for the antibiogram. We recently described the same phenomenon in other E. coli extraintestinal infections [43], and it has been reported for other microorganisms [44–47]. We, therefore, believe that several colonies of a single respiratory sample should be tested for antimicrobial susceptibility in case of GNB VAP.

We found among the respiratory isolates that those responsible for VAP had greater content of VF genes than those implicated in simple colonization. This held true for both B2 and non-B2 isolates (data not shown). VF genes of interest included iroN and sfa, indicating that, alongside host factors, pathogen factors also play a major role in lung infection development. iroN codes for a siderophore receptor of extraintestinal pathogenic E. coli, and has been implicated in the virulence of these strains. Studies have shown that iroN may be a potential vaccine candidate to prevent urinary tract infections [48]. We found that E. coli isolates responsible for VAP harbored iroN more often than those responsible for respiratory tract colonization, showing the unprecedented potential role of iroN in lung pathogenicity. Further studies are required to confirm and clarify the respective roles played by these VF, but our findings are consistent with the choice of iron uptake proteins as potential candidates for vaccines against E. coli as recently published [32].

Given the scarcity of experimental data with clinical respiratory isolates of E. coli (since the bulk of in vitro and in vivo extraintestinal experimental data focus on bloodstream, meningitis, and uropathogenic isolates), we decided to test representative strains in an animal model of pneumonia. We found that isolates with the greatest content of VF genes were associated with the highest mortality. Interestingly, even a rectal isolate was capable of very high mortality, in agreement with the notion that virulence is a by-product of commensalism [24] (Fig. 3a). Not surprisingly [16], but never reported before in the setting of pneumonia, we found a very strong correlation between the number of VF genes carried by isolates and mortality (Fig. 3b). The additive effects of VF have been recently reported experimentally using mutant strains [49].

Limitations and strength of the study

One limitation of the study is the number of patients screened, and therefore studied. This screening (which includes extended-spectrum β-lactamase GNB and methicillin-resistant S. aureus) is part of our routine ICU admission procedure. Any incoming patient (whether under mechanical ventilation or not) is screened for the pathogens listed above on ICU admission and then once a week. As mean duration in our ICU is 6 days, a limited number of patients were screened repeatedly during their stay (n = 20, 15 %), thus it is difficult to assess the dynamics of colonization, and the sequential acquisition of various isolates at the different sites. We can only speculate on the qualitative order of E. coli acquisition sequence: rectum, then oropharynx, then lung. The quantitative order (exact time to acquisition) requires another study, which we are about to perform. Because studies on E. coli VAP are very limited (and because P. aeruginosa has been already extensively studied), we deliberately focused solely on E. coli. Despite a limited number of patients, a major strength of the study is that it provides for the first time an in-depth phenotypic and genotypic analysis of E. coli VAP in comparison with their oropharyngeal and rectal counterparts.

Taken together, our results show that respiratory isolates and especially those responsible for VAP are more often B2 E. coli, with greater VF gene content than other isolates, including iron uptake genes such as iroN. Targeting these proteins by way of vaccines or monoclonal antibodies may help establish novel therapeutic strategies.