ABSTRACT
Escherichia coli bacteremia is caused mainly by sequence type complex 131 (STc131) and two clades within its fluoroquinolone-resistance-associated H30 subclone, H30R1 and H30Rx. We examined clinical and molecular correlates of E. coli bacteremia in two geographically distinct centers. We retrospectively studied 251 unique E. coli bloodstream isolates from 246 patients (48 from the Mayo Clinic, Rochester, MN [MN], and 198 from Tan Tock Seng Hospital, Singapore [SG]), from October 2013 through March 2014. Isolates underwent PCR for phylogroup, STc, blaCTX-M type, and virulence gene profiles, and medical records were reviewed. Although STc131 accounted for 25 to 27% of all E. coli bacteremia isolates at each site, its extended-spectrum-β-lactamase (ESBL)-associated H30Rx clade was more prominent in SG than in MN (15% versus 4%; P = 0.04). In SG only, patients with STc131 (versus other E. coli STc isolates) were more likely to receive inactive initial antibiotics (odds ratio, 2.8; P = 0.005); this was true specifically for patients with H30Rx (odds ratio, 7.0; P = 0.005). H30Rx comprised 16% of community-onset bacteremia episodes in SG but none in MN. In SG, virulence scores were higher for H30Rx than for H30R1, non-H30 STc131, and non-STc131 isolates (P < 0.02 for all comparisons). At neither site did mortality differ by clonal status. The ESBL-associated H30Rx clade was more prevalent and more often of community onset in SG, where it predicted inactive empirical treatment. The clonal distribution varies geographically and has potentially important clinical implications. Rapid susceptibility testing and clonal diagnostics for H30/H30Rx might facilitate earlier prescribing of active therapy.
INTRODUCTION
The past 15 years have seen an unprecedented and rapid global expansion of a single Escherichia coli lineage, sequence type complex 131 (STc131), and specifically its fluoroquinolone-resistance-associated H30R subclone. H30R comprises sister clades H30R1 (or C1) and H30Rx (or C2), which are associated with the CTX-M-15 extended-spectrum β-lactamase (ESBL) (1, 2). The emergence of STc131 is the main basis for the rise in fluoroquinolone and ESBL resistance in E. coli. STc131 isolates typically harbor multiple virulence factors (1, 3) and are associated with diverse extraintestinal infections, including bacteremia (1, 2).
Although infections due to ESBL-producing Enterobacteriaceae, including STc131, have traditionally been associated with health care exposure (i.e., health care associated) (4), the boundaries between health care- and community-associated infections with respect to the associated clones are becoming increasingly blurred. In areas of low ESBL prevalence, community-onset infections due to ESBL-producing E. coli (including STc131) may be related to antibiotic use and travel to regions of higher ESBL endemicity, in addition to health care exposure (5, 6). Indeed, community transmission of STc131 is well documented (7), including between family members (8), and appears to be more common between household than nosocomial contacts (9). In areas of high ESBL endemicity, such as countries in Asia, STc131 is well established in the community and accounts for up to one-third of community-onset ESBL E. coli bacteremias (10, 11). The emergence of extensively resistant lineages of STc131 E. coli, such as H30Rx, as a cause of community-onset infections is reminiscent of the emergence of another concerning multidrug-resistant (MDR) pathogen, community-associated methicillin-resistant Staphylococcus aureus.
Few data are available regarding the molecular epidemiology of antimicrobial-resistant E. coli infections in Singapore (SG). ESBL-producing Enterobacteriaceae emerged in Singapore in the mid-1980s, associated initially with TEM/SHV-type enzymes, which by the late 1990s were supplanted by CTX-M-type enzymes. Among 46 recent cefotaxime-nonsusceptible E. coli bloodstream isolates from Singapore, STc131 accounted for more of the community-onset isolates (23/34; 68%) than the health care-associated isolates (6/12; 50%) (12). Additionally, few studies from any locale have attempted to correlate the E. coli clonal distribution with detailed, patient-level clinical data.
Here we sought to define the molecular epidemiology (clonal background, resistance, and virulence genotype) of contemporaneous E. coli bloodstream isolates from two distinct geographical locations with different antimicrobial resistance rates: Rochester, MN (here MN), which has a low prevalence of ESBL-producing Enterobacteriaceae, including E. coli (13, 14), and Singapore, which has comparatively high rates (15). Specifically, we sought to determine if the locale-specific differences in resistance prevalence reflect differences in clonal distribution and whether clonal background corresponds with health care- or community-associated origin or clinical outcomes.
(These data were presented, in part, at the 27th European Congress of Clinical Microbiology and Infectious Diseases, Vienna, Austria, 22 to 25 April 2017, and the Singapore Health and Biomedical Congress, 12 to 14 October 2017.)
RESULTS
Study isolates.Forty-eight unique E. coli isolates were obtained from the 48 MN patients, while 203 unique E. coli isolates were obtained from the 198 SG patients, 194 of whom had a single morphotype and 4 of whom had multiple morphotypes (3 with 2 morphotypes each and 1 with 3 morphotypes). Hence, 251 isolates underwent molecular characterization for phylogenetic group, STc, virulence factors, and blaCTX-M (Table 1; see also Table S1 in the supplemental material).
Molecular characteristics of 251 Escherichia coli blood isolates
Phylogenetic groups, sequence types, subclones, blaCTX-M status, and extended-virulence genotypes.A significantly higher proportion of E. coli bacteremia isolates qualified molecularly as extraintestinal pathogenic E. coli (ExPEC) in SG (75%; 153/203) than in MN (54%; 26/48) (P = 0.003), with almost all of the H30Rx isolates being ExPEC at both sites, but 92% versus 56% of H30R1 isolates qualified as ExPEC in SG and MN, respectively (see the supplemental material). Phylogenetic and clonal groups were distributed similarly at the two sites (Table 1), with phylogroup B2 accounting for 60 to 62% and STc131 accounting for 25 to 27% of isolates. In both SG and MN, phylogroup B2, and especially STc131, was overrepresented compared to other phylogroups or clonal groups (P < 0.001 for all comparisons [one-sample chi-square test]). Likewise, among the STc131 isolates from SG, compared to other subclones and clades, the H30 subclone was overrepresented, as was H30Rx within the H30 subclone (P values of <0.001 and 0.004, respectively [one-sample chi-square test]). Although STc131 accounted for about one-quarter of the E. coli bacteremia episodes at both sites, H30Rx was significantly more prevalent overall in SG than in MN (15% versus 4%; P = 0.04). blaCTX-M-15 was associated closely with H30Rx (P < 0.001) among the SG but not the MN isolates.
Overall, SG isolates had higher virulence scores than the MN isolates (medians of 12 versus 10.5; P = 0.03 by a Mann-Whitney U test). Virulence gene profiles and scores varied significantly in relation to H30 subclone status (see the supplemental material). At both sites, H30Rx isolates had significantly higher virulence scores than the H30R1 isolates (medians of 14 [range, 13 to 15] versus 11 [range, 9 to 12] [P = 0.036] for MN and 14 [range, 10 to 16] versus 13 [range, 11 to 14] [P < 0.001 by a Mann-Whitney U test] for SG).
Clinical characteristics, bacterial variables, and outcomes at both sites.Analysis of associations between MN and SG clinical data included only patients with a single E. coli morphotype (n = 242; 48 from MN and 194 from SG) (Tables 2 to 4). Genders, ages, Charlson comorbidity indexes, and Pitt bacteremia scores were similar at both sites (Table 2). A higher proportion of MN patients received immunosuppressive medications (P = 0.006), were neutropenic (P = 0.004), had undergone solid-organ or stem cell transplantation (P < 0.001), or were admitted to the intensive care unit (P < 0.001). At both sites, the most common source of bacteremia was the urinary tract, followed by an intra-abdominal infection (Table 2). Non-E. coli organisms were recovered from blood along with E. coli more commonly in SG than in MN (9.3% versus 0%; P = 0.028). At both sites, community-onset episodes predominated (88% for SG and 85% for MN) over hospital-onset episodes (12% for SG and 15% for MN).
Clinical characteristics of 242 patients with Escherichia coli bacteremia in Singapore and Rochester, MN
Characteristics of Escherichia coli bloodstream infections by locale and STc131 and H30Rx status
Univariable and multivariable predictors of 30-day all-cause mortalitya
In SG, H30Rx appeared to have undergone community expansion in that it accounted for 27 (16%) of 170 community-onset bacteremia episodes (Fig. 1). Of these 27 community-onset episodes, 11 were fully community associated, while 16 were health care associated. In contrast, in MN, H30Rx caused none of the 41 community-onset bacteremia episodes (16 community and 25 health care associated) (Fig. 1).
Distribution of Escherichia coli bloodstream infection episodes in Minnesota (MN) and Singapore (SG) by location of onset. (A) Hospital-onset infections; (B) community-onset infections. Statistically significant differences and associated P values are indicated with *.
Regarding outcomes, the SG and MN cohorts did not differ significantly for either the proportion of patients who received active antimicrobials 0, 24, or 48 h after the availability of the Gram stain result (data not shown) or 30-day mortality (Table 2). However, in SG, patients with an STc131 or an H30Rx isolate were less likely to receive an active antimicrobial agent at the time when the Gram stain was reported (Table 3).
Epidemiological, clinical, and bacterial variables were examined as predictors of mortality using both univariable and multivariable logistic regression analyses (Table 4). In the final models, mortality was predicted significantly only by the Pitt bacteremia score in SG and only by the age-adjusted Charlson comorbidity score in MN.
DISCUSSION
We assessed the clinical and associated bacterial characteristics of contemporaneous patients with E. coli bacteremia in two locales that differ greatly for the background prevalence of antimicrobial resistance. Our findings support four main conclusions. First, since its emergence in approximately 2000 (1), STc131 and its H30 subclone have disseminated and expanded impressively, now comprising approximately 25% of bloodstream isolates at each of our geographically separated study sites. Second, despite the relatively similar proportions of E. coli bacteremia episodes caused by STc131 at each site, the ESBL-associated H30Rx clade was significantly more prevalent in SG than in MN and concerningly accounted for 16% of community-onset infections, whereas in MN, no community-onset H30Rx bacteremia was detected. Third, while the H30Rx subclone had a larger number of virulence factors and a correspondingly higher virulence score, mortality did not correlate with the clonal background, specific virulence factors, or the virulence score, suggesting that other factors (e.g., host characteristics) may play a more important role in outcomes. Fourth, H30Rx bacteremia was associated with delayed receipt of active antimicrobial therapy in SG.
The finding that at both sites, phylogroup B2 accounted for ∼60% and STc131 accounted for ∼25% of isolates agrees with older and more contemporary estimates (16–19), including a recent U.S.-wide 2011–2012 survey (20). This concordant observation reaffirms that STc131 has disseminated successfully as a global clone, possibly facilitated by some combination of enhanced virulence (for which empirical support is largely lacking), high metabolic potential, and increased transmissibility and colonization ability (1, 21). These characteristics, along with a propensity to accumulate antimicrobial resistance genes without incurring a significant fitness cost, pose a challenge to clinicians managing infections caused by E. coli, a leading Gram-negative pathogen in both community and hospital settings.
H30Rx was significantly more prominent in SG than in MN (15% versus 4% overall; 70% versus 15% of ST131-H30 isolates), although the prevalences of STc131 and H30 clones were similar at both study sites. This suggests that within different geographic locales, certain unique factors may drive the differential expansion of distinct H30 clades. We hypothesize that the larger H30Rx fraction that we observed in SG may be due to comparatively greater antimicrobial pressure in SG, leading to preferential clonal expansion of the more-extensively resistant H30Rx lineage. Numerous studies document associations of prior antibiotic exposure with the subsequent isolation of multidrug-resistant E. coli (22–24). In the SG cohort, the proportions of patients with prior antibiotic use were 36% for the preceding year and 11% for the preceding month. Comparable data were not available for the MN cohort and although estimates of standardized antimicrobial use suggest higher rates for the United States overall than for SG (25), MN has comparatively lower antibiotic utilization rates than other U.S. states (26, 27) and, thus, may have a lower rate of utilization than SG.
Another possible contributor to the higher prevalence of H30Rx in SG is that SG is an international hub, visited frequently by travelers from within and beyond Asia, that relies heavily on imported food sources. Thus, ESBL-producing Gram-negative organisms (including ST131-H30Rx) from more highly endemic locales may be introduced through these routes (28).
The high prevalence of H30Rx that we observed among community-onset E. coli bacteremia isolates in SG (16% overall; 10% of community-associated and 25% of health care-associated isolates) (Fig. 1) poses a therapeutic challenge for providers, especially for patients without health care-associated risk factors. Such patients may be at risk of receiving ineffective initial antimicrobial therapy. In SG, at the time when Gram-negative bacilli were reported in blood cultures, patients with STc131 were 2.8-fold more likely than those with non-STc131 E. coli to be receiving inactive antibiotic therapy. The main driver of this phenomenon was H30Rx, with its 7-fold-higher likelihood of inactive therapy. Assuming that blood cultures typically flag positive within 24 to 72 h of incubation, our data suggest that in SG, patients with bacteremia due to STc131, and specifically H30Rx, were less likely than those with bacteremia due to other strains of E. coli to receive active antimicrobial therapy for up to ∼48 to 96 h after blood cultures were drawn. Whereas empirical antibiotic therapy was ineffective in 23% of SG patients overall, this rose to 55% for those with H30Rx, which suggests that clinicians are unaware of risk factors for H30Rx. In our study, while we did not detect a difference in 30-day mortality, other studies have found poorer clinical outcomes (e.g., prolonged hospitalization and clinical or microbiological persistence) with H30 infection and inactive antibiotic therapy (22, 29, 30). Further efforts should be directed toward identifying physician knowledge gaps in local antimicrobial resistance patterns and improving empirical prescribing practices via educational and stewardship efforts.
Rapid diagnostics to identify resistant subclones may facilitate the more timely institution of effective therapy. Previous studies have shown that PCR-based rapid clonal typing for E. coli directly on urine from patients with urinary tract infections may be rapid enough to inform the prescription of empirical antibiotics (31, 32). By using a clonotype-specific antibiogram on urine specimens, antibiotic/pathogen mismatch was reduced by more than 60% (31). Besides decreasing the use of ineffective antibiotics, such a strategy may allow the use of narrower-spectrum agents. Extrapolating from this, rapid clonal typing may also be useful for patients with bacteremia (e.g., positive blood cultures), possibly as part of a multiplexed molecular pathogen detection panel with other resistance markers, although further studies are needed.
Neither clonal background nor virulence genotype was associated with 30-day mortality. In contrast, multivariable analysis identified host factors, including the Pitt bacteremia score and Charlson comorbidity index, as correlates of mortality, which is in keeping with previous studies that found that host factors are more important than bacterial factors in determining the severity of E. coli bacteremia (19). Notably, other recent studies found that ST131-H30 was associated with older patients, compromised hosts, antimicrobial resistance, and clinical/microbiological persistence but not with key adverse outcomes (e.g., recurrence or mortality) (22, 33).
Limitations of this study include its retrospective nature, its inclusion of only a subset of E. coli bacteremia isolates from the two study sites, and the comparatively small number of isolates from MN, which limited statistical power. Study strengths include the relatively large overall study population and SG cohort, the comparisons of standardized clinical data with extensive molecular typing results, and the attention to key clonal subsets within STc131.
In summary, STc131 accounted for approximately 25% of bloodstream isolates at major medical centers in both MN and SG in 2013 to 2014. However, unlike in MN, where the most extensively resistant subclone, H30Rx, was not embedded in the community, in SG, H30Rx was prevalent and accounted for 16% of community-onset bacteremia and may thus be difficult to contain. Given that in SG, STc131, H30, and H30Rx were also associated with delayed institution of effective antimicrobial therapy, rapid susceptibility and clonal diagnostics for H30 and H30Rx may facilitate the earlier administration of active therapy. The clinical impact of rapid diagnostics for the detection of these subclones is likely to be greatest in locales like SG with a high endemic prevalence of resistant strains. The global dissemination of successful and antimicrobial-resistant clones such as STc131-H30 and H30Rx mandates continued vigilance and attention to local clonal epidemiology.
MATERIALS AND METHODS
Patients and E. coli bloodstream isolates.We retrospectively studied a convenience sample of 246 patients with E. coli bacteremia from October 2013 through March 2014 at the Mayo Clinic, Rochester, MN (here MN) (n = 48) and the Tan Tock Seng Hospital (TTSH), Singapore (here SG) (n = 198). The MN isolates were collected as part of a previous study and represented all available isolates from the second half of that study, when isolate collection was done (34). The SG isolates were a subset selected randomly with a computer-generated list from the hospital's larger collection of all E. coli bloodstream isolates during that period.
Both sites used BD Bactec Plus Aerobic/F and Anaerobic/F blood culture vials (Becton Dickinson, Sparks, MD), incubated on the Bactec 9240 system (SG) or the Bactec FX system (MN), and identified E. coli based on typical colonial morphology on selective agar plates (MacConkey agar [SG] or eosin methylene blue [MN]), supplemented by biochemical testing and matrix-assisted laser desorption ionization–time of flight mass spectrometry (MALDI-TOF MS) with the Bruker MALDI Biotyper (Bruker Daltonics).
Susceptibility testing was performed at SG by disk diffusion directly from positive blood cultures, using an in-house, laboratory-validated method similar to that described by Andrews et al. (35), and at MN by agar dilution using pure subcultures (36). Both sites used CLSI interpretive criteria (37). Screening for ESBL production was done in SG using the double-disk approximation method (38) and in MN based on cefpodoxime MICs, with values of >4 μg/ml being considered positive (37).
Molecular characterization and virulence scores.Phylogroups (phylogroups A, B1, B2, C, D, E, and F) were determined with the updated PCR-based Clermont method (39). Clonal groups that were resolved by PCR included STc131 and its sister clades H30Rx and H30R1, STc12, STc14/81/550, STc31 (O15:K52:H1), STc69 (“clonal group A” [CGA]), STc73, STc95, STc127, STc141, STc144, STc405, STc420, and STc648 (16, 40–42). Isolates were tested for blaCTX-M (universal), blaCTX-M-15, and the blaCTX-M-9 group (3, 43) and underwent extended virulence genotyping for 42 suspected or proven extraintestinal virulence-associated genes (44).
Clinical and epidemiological data.Medical records were reviewed for patient demographics, effectiveness of antimicrobial therapy, and mortality. Active antimicrobial therapy was defined as the presence in the electronic medical record of an order for an antibiotic to which the E. coli isolate was susceptible. This was assessed at the point when the blood culture Gram stain result was reported (0 h) and 24 and 48 h later.
Based on criteria modified from those described previously by Friedman et al., bacteremia episodes were classified as health care associated if the patient had been hospitalized ≥2 days in the preceding 90 days, was a resident in a nursing home or extended-care facility, received chronic dialysis within the preceding 30 days, or received any antibiotic within the 90 days prior to the E. coli bacteremia episode (45); all other episodes were classified as community associated. Community-onset bacteremia was defined as a bloodstream infection diagnosed from a blood culture drawn up to 48 h after admission; all other bacteremia episodes (positive blood cultures drawn beyond 48 h postadmission) were considered hospital-onset episodes. The severity of the bacteremia was calculated by using the Pitt bacteremia score (46).
Statistical methods.Categorical variables were compared with Fisher's exact test or a chi-square test, as appropriate. A one-sample chi-square test was used to examine differences between observed and expected (assuming an equal distribution of clonal groups) frequencies of E. coli clonal distributions at each site. Continuous variables were compared with the Mann-Whitney U test. To identify risk factors associated with all-cause 30-day mortality at each site, backward stepwise multivariable logistic regression was performed considering multiple models, including initially all variables with a univariable P value of <0.20. The final multivariable models included, as candidate predictors, all factors with a univariable P value of <0.05, plus H30Rx status (despite its lack of univariable significance). The Hosmer-Lemeshow statistic was used to assess goodness of fit. Analyses were performed by using SPSS version 21. P values of <0.05 were considered statistically significant. This study was approved by the institutional review boards of the Mayo Clinic and the National Healthcare Group, Singapore.
ACKNOWLEDGMENTS
This work was supported in part by Office of Research and Development, Department of Veterans Affairs, grants 1I01CX000920-01 and 2I01CX000920-04 (to J.R.J.) and National Institutes of Health grant 2KL2RR024151-07 (to R.B.).
We thank Vincent Chua for extracting data on recent antimicrobial use for the SG patients.
FOOTNOTES
- Received 6 May 2018.
- Returned for modification 6 June 2018.
- Accepted 26 July 2018.
- Accepted manuscript posted online 6 August 2018.
Supplemental material for this article may be found at https://doi.org/10.1128/AAC.00937-18.
REFERENCES
- Copyright © 2018 American Society for Microbiology.
