Intestinal Persistence of Colonizing Escherichia coli Strains, Especially ST131-H30, in Relation to Bacterial and Host Factors

Abstract

Background

Superior gut colonization may underlie the pandemic emergence of the resistance-associated H30 subclone of Escherichia coli sequence type 131 (ST131-H30). Little is known about the associated host and bacterial characteristics, or the comparative persistence of non-ST131 intestinal E. coli.

Methods

Generic and fluoroquinolone-resistant E. coli isolates from volunteers’ serial fecal samples underwent clonal analysis and extensive polymerase chain reaction (PCR)-based characterization (phylogroup, selected sequence types, virulence genes). Kaplan-Meier survival analysis and Cox proportional hazards survival analysis using penalized regression (a machine-learning method) were used to identify correlates of strain persistence.

Results

Screening of 2005 subjects at the Minneapolis VA Medical Center identified 222 subjects (117 veterans, 105 human and animal household members) for longitudinal fecal surveillance. Analysis of their 585 unique-by-subject fecal E. coli strains identified multiple epidemiological, ecological, and bacterial correlates of strain persistence. ST131-H30, a strong univariable correlate of persistence, was superseded in multivariable analysis by outpatient status, fluoroquinolone resistance, and diverse (predominantly iron uptake-related) virulence genes.

Conclusions

ST131-H30 exhibits exceptional intestinal persistence, possibly due to a combination of fluoroquinolone resistance and virulence factors, which may be primarily colonization factors. This identifies both likely contributors to the ST131-H30 pandemic and potential targets for interventions against it.

Escherichia coli is a major extraintestinal pathogen that has become increasingly difficult to manage due to progressively extensive and prevalent antimicrobial resistance [1, 2]. The resistance problem is attributable largely to the emergence of a few highly successful, resistance-associated clonal groups [3]. The most notorious of these is the (fluoroquinolone resistance-associated) H30 subclone of sequence type 131 (ST131-H30), which currently causes most multidrug-resistant E. coli infections in the United States and globally [4, 5].

The basis for the current ST131-H30 pandemic remains undefined. Although early reports described this lineage as being hypervirulent compared with other E. coli, subsequent epidemiological and experimental studies failed to consistently support this concept [6, 7]. Antimicrobial resistance per se, although possibly contributory, also seems an insufficient explanation, given the abundance of E. coli lineages that are similarly extensively resistant as ST131-H30, but much less epidemiologically successful [8].

Because most extraintestinal E. coli infections arise from the gut reservoir, enhanced gut colonization—that is commensalism—by ST131-H30 might underlie its clinical prominence. Indeed, epidemiological studies have documented prevalent gut colonization with ST131 among hospital inpatients, long-term care facility residents, and community dwellers [6, 9–14].

To clarify the basis for the comparatively high intestinal prevalence of ST131-H30, longitudinal studies have assessed intestinal persistence for ST131-H30 versus other E. coli. In studies of fecal extended-spectrum β-lactamase–producing E. coli among hospital inpatients and long-term care facility residents, ST131 strains persisted longer in the host than did non-ST131 strains [15–17]. Similarly, a study of fluoroquinolone-resistant E. coli colonization among ambulatory women identified longer gut persistence for ST131-H30 than for other E. coli [18]. Notably, these studies did not assess the total gut E. coli population or considered only a limited range of strain characteristics, whereas a longitudinal study of clonal stability and plasmid variation in a persistent gut-colonizing ST131 strain from a woman with recurrent infections due to that strain involved only the 1 subject [19].

Accordingly, we undertook the present study to assess broadly the persistence of intestinal E. coli strains among US veteran inpatients and outpatients and their human and animal household members. We specifically compared intestinal persistence with selected features of the host and environment, plus a broad range of bacterial characteristics, including fluoroquinolone resistance, phylogenetic and clonal background, and virulence genes.

METHODS

Subjects and Samples

As described elsewhere [9], study subjects included US military veterans under care at the Minneapolis Veterans Affairs Medical Center (MVAMC) and all available household members. Veterans were recruited prospectively from May 2014 through May 2018 by sending invitations for study participation to all newly discharged MVAMC inpatients and randomly selected outpatients. Veterans who agreed to participate were encouraged to refer all available adult and child household members, and pets. Fecal swabs were collected by or for consenting or assenting subjects and pets according to an institutional review board-approved protocol. Swabs were mailed at room temperature in commercial transport medium to the research laboratory, along with basic demographic information. In pilot experiments, prolonged (72-hour) exposure of transport swabs to ambient temperature had a negligible effect on generic E. coli viability (not shown).

To assess strain persistence, subjects whose initial fecal swab, when processed as described below, yielded fluoroquinolone-resistant E. coli and/or ST131—and, during the initial phase of the study (May 2014 through June 2016), all other subjects (if any) from the same household, irrespective of their fecal culture results—were offered serial fecal sampling. Serial sampling was done monthly for 6 months, then every 3 months, until the first of the following occurred: the end of the study period, the subject/household declined further follow-up, or on 2 consecutive sampling occasions no household member yielded the household’s initial strain of interest. During the later phase of the study (July 2016 through July 2018) serial fecal sampling was limited to subjects whose initial specimen yielded fluoroquinolone-resistant ST131 E. coli.

Culture Methods

In the research laboratory, fecal swabs were streaked to gram-negative selective agar with and without ciprofloxacin (4 mg/L), for overnight incubation at 37°C. Indole-positive, citrate-negative colonies with a characteristic E. coli morphology were regarded presumptively as E. coli.

Molecular Methods

Up to 10 presumptive E. coli colonies per plate (as available) were screened for clonality by using random amplified polymorphic DNA (RAPD) analysis [20]. One representative per sample per unique RAPD profile, as determined visually, underwent standardized XbaI pulsed-field gel electrophoresis (PFGE) analysis [21]. PFGE profiles were analyzed within BioNumerics to classify isolates into pulsotypes based on 94% profile similarity, according to Dice similarity coefficients, in comparison with index profiles in a large private PFGE library [22]. Any novel profile (ie, < 94% similar to an established index profile) became the index profile for a new pulsotype.

Additionally, using duplicate boiled lysates for template DNA and relevant positive and negative controls, 1 representative colony per unique RAPD profile per sample underwent polymerase chain reaction (PCR)-based screening for ST131 [23]. Established PCR assays also were used to screen ST131 isolates for membership in the O16, H22, H30, and H30Rx ST131 subclones, and to screen non-ST131 isolates for E. coli phylogenetic group and key STs associated with virulence and/or resistance [24–27]. One representative per pulsotype per subject also underwent standardized PCR-based screening for 52 putative or proven extraintestinal virulence genes and variants thereof [28]. Established molecular definitions based on combinations of hallmark virulence genes were used to presumptively classify isolates as representing extraintestinal pathogenic E. coli (ExPEC) or uropathogenic E. coli (UPEC) [28].

Statistical Methods

The unit of analysis was the discrete strain, according to PFGE analysis, as detected in an individual subject, regardless of how many other strains the subject had, how many other subjects had the strain, or how long the subject or household was followed. Each strain identified in a subject was scored for its presence/absence at each sampling point.

Loss of persistence, per the strict definition, was assigned to the first of 2 consecutive sampling points at which a previously detected strain was no longer detected. For sensitivity analyses, alternate definitions were used, according to which loss of persistence required only a single negative sample, irrespective of possible subsequent positive samples (loose definition), or 2 consecutive negative samples, plus no subsequent positive samples (very strict definition).

With each definition, strains for which loss of persistence was not observed were censored as of the last sampling point at which they were detected. Because all 3 operational definitions of strain loss yielded qualitatively similar results that differed only for quantitative estimates of persistence duration (not shown), subsequent analyses relied on the base case (strict) definition.

Strains were classified as initial (ie, present in the subject’s first sample) or acquired (ie, newly detected in the subject during follow-up). Because selective plating using ciprofloxacin-supplemented agar enhanced detection of fluoroquinolone-resistant strains, separate analyses were done for strains recovered from antibiotic-free plates (whether fluoroquinolone-susceptible or fluoroquinolone-resistant) versus for fluoroquinolone-resistant strains only, as recovered from ciprofloxacin-supplemented plates. Additionally, because acquired strains were anticipated to possibly differ from initial strains, they also were analyzed separately. This gave 3 total strain groups, for separate analysis.

Strain persistence in relation to relevant bacterial and host characteristics was assessed initially in a univariable manner by using Kaplan-Meier survival analysis. The log rank test was used to assess between-group differences. The mean hazard rate for strain loss was used to summarize (inversely) persistence. Additionally, crosstabs analysis (n − 1 χ² tests) was used to assess for associations of ST131 with other study variables.

Because the univariable analysis identified multiple statistically significant correlates of persistence, many of which have known associations [27, 29], Cox proportional hazards models using penalized regression were used to allow the variables to compete with one another while addressing the problem of multicollinearity. Penalized regression [30] is an alternative to stepwise procedures for reduction of predictors in high dimensional data. It minimizes residual sum of squares through a penalty imposed on the size of regression coefficients. This is a regularization method.

In the present Cox proportional hazards models, we employed LASSO (least absolute shrinkage and selection operator), using the shrinkage parameter L1-Norm, which is the sum of the absolute value of the coefficients. In a LASSO model, larger penalties (L1-Norm) result in coefficients closer to zero [30]. We used the R package GLMNET [31] to run models across a grid of 100 λ values (the parameter for tuning shrinkage). The product of this set of models was then subject to k-fold cross validation, which then produces a final model in which the λ value is selected that minimizes partial likelihood deviance. We ran the cross-validation 10 times for each dataset, with a different start value each time, and selected the most stable (ie, most replicated) model across the 10 validation runs. Results were reported as weighting coefficient and the corresponding hazard ratio (HR) for strain loss. Based on the variables identified in the penalized regression models we ran Cox proportional hazards models without clustering and models using the R package coxme to account for clustering at the subject and household levels. Results from these models were very similar and suggest a relatively small effect due to clustering at either level (data not shown).

Throughout, the criterion for statistical significance was P < .05. Because of the study’s exploratory nature, no adjustment was made for multiple comparisons.

RESULTS

Study Population

Initially, 2005 subjects (humans and animals), from 1125 households, were screened for carriage of fluoroquinolone-resistant E. coli or ST131. Of these, 133 (7%) had no detectable fecal E. coli, 1580 (79%) were from households without detected fecal fluoroquinolone-resistant E. coli or ST131, and 70 (3.5) provided only 1 sample. This left for the strain persistence analysis the 222 subjects who had submitted 2 or more serial fecal specimens and were from households in which at least 1 member was colonized with fluoroquinolone-resistant E. coli or ST131.

The 222 subjects included 117 veterans (84 outpatients, 33 inpatients) and 105 nonveteran household members (65 spouses, 2 other adults, 7 children, and 31 pets). The 184 adults included 114 men, 69 women, and 1 unknown-gender subject. Each household (n = 127) contributed from 1 to 7 subjects (median, 1), including, for 5 households, from 1 to 3 children (median, 1) and, for 16 households, from 1 to 6 pets (median, 2; all dogs or cats). Their 3073 total fecal samples yielded 585 unique-by-subject E. coli strains for persistence analysis.

Study Categories

The 585 strains were classified into 3 groups based on isolation method and timing of appearance. The first group (n = 224), that is initial strains from antibiotic-free agar, represented 140 subjects from 78 households. The second group (n = 273), that is acquired strains from antibiotic-free agar, represented 105 subjects from 57 households. The third group (n = 158), that is initial strains from ciprofloxacin-supplemented agar (70 of which also appeared in the first strain group), represented 131 subjects from 109 households.

Univariable Correlates of Persistence

For the first strain group, that is initial strains from antibiotic-free agar, the overall mean hazard rate for strain loss was 0.21/month (Table 1). Univariable survival analysis identified 29 persistence-associated variables: 4 host categories, shared strain status, fluoroquinolone resistance, and 23 molecular characteristics, including 3 phylogroups, 3 clonal markers, and 16 virulence genes or combinations thereof (2 adhesins, 2 toxins, 3 iron acquisition systems, 4 protectins, 4 miscellaneous traits, and both pathotypes). All but 3 such associations were positive, indicating longer persistence for strains with the characteristic. The 3 exceptions involved phylogroup A, phylogroup B1, and tsh (temperature-sensitive hemagglutinin).

Among the persistence-associated variables the lowest observed mean hazard rate for strain loss (ie, longest persistence) of 0.09/month attached to 4 molecular variables: ST131-H30, sat (secreted autotransporter toxin), kpsM K1 (K1 capsule), and kfiC (K5 capsule) (Table 1). Following these, with nearly as low a rate of strain loss (0.10/month), came fluoroquinolone resistance, STc131, O25b, and iha (adhesin-siderophore receptor). Strain loss rates were slightly higher for the pathotypes (lowest, ExPEC: 0.12/month) and host categories (lowest, outpatient: 0.13/month).

The second strain group, that is acquired strains from antibiotic-free agar, had a similar overall mean strain loss rate as did initial strains (0.24/month) (Table 1). By contrast with the initial strains, however, they exhibited no significant persistence associations with host category, and had an opposite-direction association of persistence with strain sharing. Nonetheless, as with the initial strains, numerous bacterial traits—including most of those identified among the initial strains—were associated with persistence, including phylogroup B2, 21 virulence genes, UPEC, and ExPEC. The lowest associated rate of strain loss for any variable (0.06/month) attached to papG II (P fimbriae adhesin variant) and ireA (siderophore receptor), followed closely by other pap (P fimbriae) elements. Notably, fluoroquinolone resistance, STc131, and O25b exhibited similarly low rates of strain loss as among the initial strains, but here did not reach statistical significance, whereas ST131-H30 was too scarce for analysis.

The third strain group, that is initial strains from ciprofloxacin-supplemented agar, had a lower overall mean rate of strain loss (0.05/month) than did the first 2 strain groups. Markedly fewer significant predictors of persistence emerged here than with the first 2 strain groups (9 total; 3 host categories, 6 virulence genes). With these 9 predictors, the lowest associated mean rate of strain loss was for outpatient (0.03/month), followed closely by the 6 virulence genes (0.04–0.05/month each). Strain loss rates were numerically but not statistically significantly lower in association with STc131, ST131-H30, and O25b. Because all strains were fluoroquinolone-resistant, fluoroquinolone resistance was not assessable in relation to persistence.

Univariable Associations with ST131

Crosstab comparisons of ST131 with other study variables showed that, for each strain group, multiple variables within most categories were associated with ST131 (Table 2). Most such associations (53/66, 80%) were positive, that is the variable was significantly more prevalent among ST131 strains.

Multivariable Correlates of Persistence

Because of the suspected collinearity of many variables, including the multiple associations involving ST131 (Table 2), penalized regression analysis was used to identify independent correlates of persistence within each strain group. This analysis identified 11 correlates of persistence for the first strain group and 4 each for the second and third strain groups, giving 19 in total, which collectively involved 15 unique variables (Table 3). All identified variables except phylogroup A had a negative coefficient for strain loss and a corresponding HR < 1.0, indicating longer persistence.

The 15 persistence-associated variables were diverse: considering all 3 strain groups, 1 or more significant correlations with persistence were identified within each variable category excepting ecological, STs, and toxins. Each of the corresponding associations excepting 3 (those for outpatient, fyuA, and ireA) appeared in only 1 strain group. The associations with persistence varied greatly in strength, with HRs for strain loss ranging from < 0.80 (0.72 for K1 capsule and ireA [first strain group]; 0.75 for iha; 0.78 for adult) to > 0.95 (0.96 for clbB [colibactin]; 0.97 for ireA [second strain group]; 0.99 for group B2).

Overall, among the 12 persistence-associated bacterial characteristics the functional category that accounted for the most associations was iron uptake (5 associations, or 6 including iha), followed by protectins and miscellaneous traits (2 associations each) (Table 3). By contrast, neither ST131 nor ST131-H30 was significantly associated with persistence. Notably, however, 2 of the 3 persistence-associated epidemiological variables, and 7 (58%) of the 12 persistence-associated bacterial characteristics, as identified in 1 or more strain groups (Table 3), were associated with ST131 in the corresponding strain group (Tables 2).

DISCUSSION

In this study we sought epidemiological, ecological, and bacterial correlates of intestinal persistence among fecal E. coli strains from veterans and their household members. The findings provide novel insights into the clonal dynamics of the intestinal E. coli reservoir, a critical proximal source for extraintestinal E. coli infections.

Univariable analysis identified multiple persistence-associated variables, including representatives of all studied variable categories. Of the epidemiological variables, outpatient was the strongest correlate of persistence for the first strain group (nonselected initial strains), superseding human, adult, and veteran. Conceivably, outpatient veterans experience less intestinal strain turnover that do hosts from other categories, possibly because they have less exposure to new strains than do nonveterans or household pets, and less antibiotic exposure than inpatient veterans. By contrast, for the second strain group (acquired strains), no epidemiological correlates were identified, whereas for third strain group (fluoroquinolone-resistant initial strains), veteran was the strongest correlate of persistence, possibly because veterans are more likely than other hosts to use antibiotics, which select for resistant strains.

The study’s single ecological variable, shared strain, has been associated in previous studies with clinical infections and virulence-associated lineages and genes [32–35]. Here, the weak and conflicting univariable associations of strain sharing with persistence left these relationships unresolved, despite intuitive mechanisms whereby strain sharing and persistence might be mutually reinforcing.

Of the 3 studied categories of bacterial characteristics (fluoroquinolone resistance, phylogenetic subsets, and virulence traits), fluoroquinolone resistance exhibited, within the first strain group, the strongest univariable association and a medium-strength multivariable association with persistence. This phenomenon is consistent with—and may partially explain—the global emergence of fluoroquinolone-resistant E. coli, notably ST131-H30.

The basis for the greater persistence of fluoroquinolone-resistant strains is unclear. Resistance per se may be critical. Presumably, however, for fluoroquinolone resistance to enhance persistence directly the host would need to be exposed to fluoroquinolones; here, the unavailability of drug exposure data precluded testing of this hypothesis. Alternatively, fluoroquinolone resistance may be a marker for linked traits that confer enhanced persistence. Conceivably, genome-wide association studies could search for such primary factors, which, if identified, could be assessed in animal models.

Regarding phylogenetic subsets, ST131-H30 status was one of the strongest univariable correlates of persistence among the nonselected initial strains, and exhibited similar trends among the acquired and ciprofloxacin-selected initial strains. ST131-H30’s exceptional intestinal persistence capability has been proposed to underlie its pandemic emergence [18]. By contrast with the strong positive persistence associations of ST131, ST131-H30, and group B2 (which accounts for most ExPEC-associated lineages), negative associations with persistence were observed with phylogroups A and B1, which among clinical isolates are usually minor contributors that infect primarily compromised hosts. These contrasting phylogroup and ST-specific associations identify parallels between commensalism and virulence, as seen also with virulence genes (discussed below) [36].

Most of the present persistence-associated accessory traits are associated with virulence, to an extent that corresponds roughly with the strength of their associations with persistence [28]. This supports the concept that most so-called virulence traits also enhance colonization fitness, and that virulence may be a secondary, or spillover, effect conferred by traits that evolved mainly under selection for enhanced commensalism, that is success in the species’ primary ecological niche, the intestinal tract [36].

Additionally, most of the present persistence-associated bacterial characteristics, including fluoroquinolone resistance and specific virulence genes, were associated with ST131. This left in question whether persistence is linked primarily to clonal background or accessory traits.

To address this, we used penalized regression analysis, which considers all variables concurrently [30]. This approach identified as multivariable correlates of persistence 3 epidemiological variables, plus group B2 (weakly; acquired strains only) and diverse accessory traits (fluoroquinolone resistance, multiple virulence genes, and ExPEC status), but not ST131 or ST131-H30. This suggests that the greater persistence of ST131 and ST131-H30 [18] is attributable—or linked—to a suite of specific persistence-enhancing traits, including but not limited to fluoroquinolone resistance, rather than to some unique unmeasured characteristic of ST131’s core or accessory genome.

As such, interventions against the identified persistence-associated virulence traits—mainly siderophore receptors, which implies that iron is important for gut persistence—conceivably could protect against ST131 infections in 2 ways: by diminishing the gut ST131 reservoir and by interfering directly with pathogenesis [37, 38]. Such interventions conceivably could have a broader impact than would lineage-specific interventions because they would affect any strain that possessed the cognate trait, not just a specific lineage of concern, which is important because of shifting clonal dynamics. Likewise, reduced fluoroquinolone use conceivably could blunt the persistence of fluoroquinolone-resistant strains.

The study has notable limitations. First, all subjects were from households in which at least 1 member initially carried ST131 or fluoroquinolone-resistant E. coli. Second, few epidemiological data were available, especially regarding exposures and behaviors. Third, 10 or fewer E. coli colonies were picked per plate, doubtless missing minor strains. Fourth, the definitions used for loss of persistence were arbitrary, although in sensitivity analyses the definition used affected results minimally. Fifth, the gut microbiota was not addressed. Sixth, a biased set of bacterial characteristics was studied, and only for presence/absence. Seventh, ambient-temperature sample shipment may have affected different strains differently. Eighth, only association, not causation, was studied.

The study also has notable strengths. These include the large sample size, extended follow-up, minimally restrictive inclusion criteria, detailed strain characterization (including for clinically relevant traits), and use of a machine learning approach to create more parsimonious and reliable survival models with high dimensional data.

In summary, ST131-H30 status was one of the strongest univariable correlates of intestinal persistence among fecal E. coli strains from veterans and their household members, which suggests a possibly important contributor to ST131-H30’s pandemic emergence. Notably, however, specific ST131-associated bacterial traits—including multiple virulence genes and fluoroquinolone resistance—dominated ST131-H30 status as multivariable correlates of persistence. This suggests possible mechanisms for ST131-H30’s exceptional intestinal persistence and potential targets for preventive interventions.

Notes

Acknowledgments. Muhanad Mohamed and Billy Slater provided technical assistance.

Financial support. This work was supported in part by the Office of Research and Development, Department of Veterans Affairs (grant numbers 1 I01 CX000920-01 and 2I01CX000920-04 to J. R. J.); and the National Institute of Allergy and Infectious Diseases of the National Institutes of Health Antibacterial Resistance Leadership Group (grant number UM1AI104681 to J. R. J.).

Potential conflicts of interest. J. R. J. has received grants or consultancies from Achaogen/Cipla, Allergan, Janssen/Crucell, Melinta, Merck, Shionogi, Syntiron, and Tetraphase, and has patent application for tests to detect specific E. coli strains. All other authors report no potential conflicts.

All authors have submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest. Conflicts that the editors consider relevant to the content of the manuscript have been disclosed.

References