Bacterial “Virulence” Traits and Host Demographics Predict Escherichia coli Colonization Behaviors Within Households

Abstract

Background

Although intestinal colonization precedes most extraintestinal Escherichia coli infections, colonization-promoting factors are incompletely understood. We compared within-household E. coli colonization patterns with host and bacterial traits.

Methods

Twenty-two veterans with a clinical E. coli isolate and their 46 human and animal household members underwent longitudinal fecal sampling. Distinct E. coli strains were characterized for phylogenetic background, virulence genes, antibiotic resistance, and colonization behaviors. Host and bacterial traits were assessed statistically as predictors of colonization behaviors.

Results

Among the 139 unique-by-household fecal E. coli strains, univariable predictors of colonization behavior included (i) host demographics, (ii) matching the index clinical isolate, and (iii) bacterial characteristics (2 phylogroups, 5 clonal lineages, 18 virulence genes, and molecular extraintestinal pathogenic E. coli status). Multivariable predictors of colonization behavior included veteran host, spouse host, matching the index clinical isolate, phylogroup F, ST73, hlyD (alpha hemolysin), hlyF (variant hemolysin), H7 fliC (flagellar variant), vat (vacuolating toxin), and iha (adhesin-siderophore).

Conclusions

Host demographics, multiple bacterial “virulence” traits, and matching the index clinical isolate predicted E. coli fecal colonization behaviors. Thus, certain bacterial characteristics may promote both colonization and pathogenicity. Future interventions directed toward such traits might prevent E. coli infections both directly and by disrupting antecedent colonization.

Extraintestinal Escherichia coli infections are an ever-growing threat that is compounded by emerging antimicrobial resistance. How antimicrobial-resistant E. coli strains emerge and disseminate is unclear. Because the fecal microbiota is a reservoir for such organisms, sharing of fecal E. coli strains among individuals likely contributes to dissemination [1–3].

Specific molecular traits—including certain phylogenetic groups, clonal lineages, and virulence genes—are over-represented among the extraintestinal pathogenic E. coli (ExPEC) strains that cause most extraintestinal E. coli infections. For example, sequence type (ST) 131 (from phylogenetic group B2), and specifically its ST131-H30 subclone, is the leading cause of fluoroquinolone-resistant and extended-spectrum beta-lactamase-producing E. coli infections worldwide [4, 5]. ExPEC strains commonly exhibit multiple virulence-associated accessory traits such as adhesins, siderophore systems, toxins, protectins, and motility mechanisms, some of which—although regarded as contributing to pathogenesis—may also promote colonization of human and animal hosts [6–21].

Despite the gut being the main reservoir for ExPEC, factors that affect colonization by and sharing of E. coli strains are incompletely defined. Accordingly, we evaluated E. coli colonization patterns within veterans’ households and assessed host demographic subsets and bacterial traits as predictors of bacterial colonization behaviors.

METHODS

Study Population

This prospective observational cohort study involved veterans at the Minneapolis Veteran Affairs Medical Center (MVAMC) and their household members. Source veterans for sequential clinical E. coli isolates were sent a letter soliciting study participation, including any household members; veteran-only households were excluded. Eleven households, each with a fluoroquinolone-resistant or fluoroquinolone-susceptible index clinical isolate (hereafter, “fluoroquinolone-resistant households” and “fluoroquinolone-susceptible households”), were recruited, giving 22 total households. Of the 22 index isolates, 20 were from urine and 2 were from swabs; 15 represented infection, 4 represented colonization (urine, skin), and 3 were indeterminate (urine, wound). The 22 index subjects included 3 females and 19 males. Eighteen had notable comorbid health conditions, 13 reported antibiotic use in the past 6 months, and 6 reported antibiotic use in the past month (Supplementary Table 1).

Patient Consent Statement

The MVAMC Institutional Review Board approved the study before enrollment of participants. Written consent was obtained from all study participants.

Data Collection

The 68 total study participants (22 veterans, 27 human household members, 19 pets) underwent serial fecal sampling 2–6 times each between October 2012 and May 2014. Fecal samples were cultured on gram-negative selective agar, with and without ciprofloxacin (4 mg/L), to recover total and ciprofloxacin-resistant gram-negative bacilli. Morphologically consistent, indole-positive, citrate-negative colonies were regarded as E. coli. Up to 10 E. coli colonies per plate (as available) underwent random amplified polymorphic DNA (RAPD) analysis using primer 1254 according to an established procedure adapted from Berg et al. and optimized by the authors [22, 23]. In validation studies, when applied to replicate colonies from a given fecal sample, this approach yielded 85%–90% concordance with Xbal pulsed-field gel electrophoresis (PFGE) (unpublished, J.R.J.). One colony per unique RAPD profile per plate (as judged by visual inspection) and all index clinical isolates underwent standardized Xbal PFGE analysis [17, 24]. Pulsotypes were assigned based on 94% similarity to reference profiles within a large private PFGE library (2382 pulsotypes, 7092 E. coli isolates) [24]. This approach to resolving unique within-household shared and persisting colonizing strains has been validated by whole-genome sequencing [18].

Molecular Traits and Antibiotic Resistance

Each unique fecal strain (ie, pulsotype) per subject, plus the index isolates, underwent determination of major phylogenetic group (A, B1, B2, C, D, F), selected clonal lineages and subsets thereof (n = 15), and putative virulence genes (n = 47) with established multiplex polymerase chain reaction (PCR) assays [25, 26]. ExPEC status was defined operationally as presence of ≥2 of papAH and/or papC (P fimbriae), sfa/focDE (S and F1C fimbriae), afa/draBC (Dr-binding adhesins), iutA (aerobactin receptor), and kpsMII (group 2 capsules) [27]. The index isolates underwent molecular O typing using an established multiplex PCR assay that detects the 10 most common O types among invasive E. coli isolates [28]. Susceptibility to 24 common antibiotics was assessed by disk diffusion using Clinical and Laboratory Standards Institute–specified procedures and interpretive criteria. For strains with discrepant results among multiple representatives from a given household, the consensus result was used.

Colonization Outcome Definitions

For each unique E. coli strain per household, 4 colonization outcomes were determined. (i) The fecal predominance index (hereafter: fecal predominance) was the mean by-sample proportion of assessed E. coli colonies for which the strain accounted among those fecal samples that yielded the strain. (ii) The persistence index (hereafter: persistence) was the by-household mean proportion of a subject’s fecal samples that yielded the strain, among samples collected after the strain’s first detection in that subject. (iii) The sharing index (hereafter: sharing) was the proportion of potential sharing pairs within a household in which the strain was detected (synchronously or asynchronously) in both subjects. (iv) Household prevalence was the proportion of all of the household’s fecal samples that yielded the strain.

Statistical Analysis

Variables associated with ≥5 fecal strains were analyzed statistically. For univariable analysis, comparisons involving dichotomous variables were assessed using the (n-1) chi-square test [29] or binomial logistic regression; those involving continuous variables were assessed using the Mann-Whitney test or simple regression. Due to their strongly bimodal distribution (not shown), persistence and sharing were analyzed as dichotomous variables.

Multivariable analysis was used to assess jointly host demographic subset, clinical isolate–matching status, and molecular traits as predictors of the studied colonization outcomes. For each colonization outcome, statistically significant univariable predictor variables were assessed for correlations, and clusters were identified of highly correlated variables (correlation coefficient ≥ 0.7). Within each such cluster, the variable most highly correlated with the particular outcome was selected for multivariable analysis. Backward stepwise variable entry was used primarily; forced and forward stepwise variable entry was used for sensitivity analysis.

The final multivariable analyses did not include antibiotic resistance phenotypes, given the study’s main focus on the relationship between intrinsic bacterial characteristics (virulence genes, phylogenetic and clonal background) and colonization behaviors. In an exploratory multivariable analysis that included antibiotic resistance, streptomycin was the only resistance marker that significantly predicted any colonization outcome (household sharing: not shown).

Statistical analyses were performed using SPSS, version 19 (IBM Analytics). Throughout, the significance criterion was P < .05.

RESULTS

Index Clinical Isolates: Molecular Characteristics and Antibiotic Resistance

Of the 22 index clinical isolates, 18 were phylogroup B2; the rest were 1 each of phylogroups A, C, D, and F (Supplementary Table 2). ST and O-type could be determined for 18 of 21 and 16 of 21 index isolates, respectively. By ST (and O-type), in order of decreasing frequency, 6 isolates were ST131-H30R1 (O25b), 3 were ST73 (2 O25a, 1 O-undetermined), 2 were ST12 (O4), 2 were ST 95 (1 O2, 1 O1), and 1 each were ST1193 (O75), ST127 (O6), ST131-H30Rx (O25b), ST141 (O-undetermined), ST648 (O1), ST69 (O17). All but 1 index strain exhibited at least 1 virulence gene, and all but 2 showed phenotypic resistance to at least 1 of the 24 antibiotics (Supplementary Table 2).

Household Category Comparisons: Fecal Sampling

The 22 study households comprised 68 total participants, including 49 humans (22 veterans, 27 others) and 19 pets. The 11 fluoroquinolone-resistant and 11 fluoroquinolone-susceptible households, so labeled based on the index veteran’s clinical isolate, comprised, respectively, 31 and 37 total subjects (Table 1). Households contributed a mean (range) of 3.1 (2–10) subjects each. Overall, subjects provided a mean (range) of 3.4 (2–6) fecal samples each, and households provided a mean (range) of 9 (4–30) total samples each, for 208 total samples (100 for fluoroquinolone-resistant households; 108 for fluoroquinolone-susceptible households). Fecal samples were submitted over intervals ranging, by subject, from 5 to 62 weeks (median, 31 weeks).

Table 1.

Household Composition in Relation to the Fluoroquinolone Phenotype of the Veteran’s Index Clinical Escherichia coli Isolate

	Fluoroquinolone Phenotype of Index Clinical Isolate (No. per Category)
Category	Resistanta	Susceptibleb
Households	11	11
Total subjects	31	37
Veterans	11	11
Spouses	9	10
Nonspouse adults	3	1
Children	0	4
Pets	8	11

^aHouseholds in which the veteran’s index clinical E. coli isolate was fluoroquinolone-resistant.

^bHouseholds in which the veteran’s index clinical E. coli isolate was fluoroquinolone-resistant.

Fecal samples yielded 139 unique-by-household E. coli strains. Of these, 16 (12%) matched the household’s index clinical isolate (hereafter, “clinical isolate–matching strains”), whereas 123 (88%) did not (hereafter, “fecal-only strains”).

Household Category Comparisons

Microbiological results differed between the fluoroquinolone-resistant and fluoroquinolone-susceptible households (Table 2). ST131 accounted for the index clinical isolate for 7/11 (64%) fluoroquinolone-resistant households but no fluoroquinolone-susceptible household (P < .001). Collectively, fluoroquinolone-resistant households yielded numerically fewer unique-by-household fecal E. coli strains (n = 53) than did fluoroquinolone-susceptible households (n = 86; P = .17).

Table 2.

Household Escherichia coli Fecal Strain Isolation in Relation to the Fluoroquinolone Phenotype of the Index Veteran’s Clinical E. coli Isolate

	Fluoroquinolone Phenotype of Index Clinical Isolate (No. or Proportion)
Category	Resistant Householdsa	Susceptible Householdsb	P c
Total No. fecal strainsd	53	86
Total No. fecal-only strainse	43	80
Fluoroquinolone-resistant fecal-only strains	13/43	1/80	<.001
ST131 fecal-only strains	7/43	1/80	<.001
Households with clinical isolate–matching strain in feces	10/11	6/11
Subjects with clinical isolate–matching strain in feces	17/31	9/37	.01
Veterans with clinical isolate–matching strain in feces	10/11	5/11	.03
Spouses with clinical isolate–matching strain in feces	5/9	2/10
Other adultsf with clinical isolate–matching strain in feces	0	0
Children with clinical isolate–matching strain in feces	0	0
Pets with clinical isolate–matching strain in feces	2/8	2/11

^aHouseholds in which the veteran’s index clinical E. coli isolate was fluoroquinolone-resistant.

^bHouseholds in which the veteran’s index clinical E. coli isolate was fluoroquinolone-susceptible.

^c P values shown if <.05. Comparisons involving dichotomous variables were assessed using the (n-1) chi-square test, whereas those involving continuous variables were assessed using the Mann-Whitney test.

^dUnique-by-household strains, including fecal strains that matched the index clinical E. coli isolate (ie, the “matching clinical isolate” strains).

^eUnique-by-household strains, excluding fecal strains that matched the index clinical E. coli isolate (ie, the “matching clinical isolate” strains).

^fNonveteran and nonspouse adults.

Colonization with a clinical isolate–matching strain was detected in proportionally more subjects from fluoroquinolone-resistant than fluoroquinolone-susceptible households, both among veterans (10/11 [91%] vs 5/11 [45%]; P = .03) and overall (17/31 [55%] from 10 households vs 9/37 [24%] from 6 households; P = .01) (Table 2). In veteran fecal samples, ST131 accounted for 6/10 (60%) of the clinical isolate–matching strains from fluoroquinolone-resistant households, but none from fluoroquinolone-susceptible households (P = .04).

As compared with fluoroquinolone-susceptible households, fluoroquinolone-resistant households yielded numerically fewer fecal-only strains (43 vs 80; P = .14) but, among those strains, a larger fluoroquinolone-resistant fraction (13/43 [30%] from 5 households vs 1/80 [1%] from a single household; P < .001) (Table 2). The fluoroquinolone-resistant households also yielded a larger fraction of ST131 fecal-only strains (7/43 [16%] from 5 households vs 1/80 [1%]; P < .001).

Colonization Outcomes: Host Demographic Subset

Of the 139 total (unique-by-household) fecal strains, 63 occurred in veterans, 50 in spouses, 14 in children, and 43 in pets (some strains had multiple hosts). Host subset corresponded significantly with persistence, sharing, and/or household prevalence, but not fecal predominance (Table 3). Specifically, spouse source predicted persistence (P = .01), sharing (P = .02), and household prevalence (P < .001); veteran source predicted sharing (P = .02) and household prevalence (P < .001); and child source negatively predicted household prevalence (P = .001).

Table 3.

Relationship Between Host Demographic Subset and Colonization Outcomes^a Among 139 Fecal Escherichia coli Strains

		Persistence				Sharing				Prevalence
		Univar.		Multivar.a		Univar.		Multivar.b		Univar.		Multivar.c
Host Subsetd,e	Strains, No.	ORf	P	OR	P g	OR	P	OR	P	Co	P	Co	P
Veteran	63					1.88	.02	2.93	.02	.31	<.001	.19	.01
Spouse	50	2.63	.01	2.63	.02	4.56	<.001	6.21	<.001	.32	<.001	.40	<.001
Child	14									.30	.001

Abbreviations: Co., correlation; Multivar., multivariable; OR, odds ratio; Prevalence, household prevalence; Univar., univariable.

^aMultivariable candidate predictor variables were spouse, matches CI, ST131-H30, sat (secreted auto-transporter toxin), usp (uropathogenic-specific protein), kpsMII (group 2 capsules), ompT (outer membrane protease), and ExPEC (extra-intestinal pathogenic E. coli, functionally defined as ≥2 of the following: papAH and/or papC [counted as 1: P fimbriae], sfa/focDE [S and F1C fimbriae], afa/draBC [Dr-binding adhesins], iutA [aerobactin receptor], and kpsMII).

^bMultivariable candidate predictor variables were veteran, spouse, matches CI, ST95, vat (vacuolating toxin), and kpsMII.

^cMultivariable candidate predictor variables were veteran, spouse, child, matches CI, phylogroup B1, ST73, ST131-H30, papAH, iha (adhesin-siderophore receptor), usp, and kpsMII.

^dHost subset and outcome variables shown are those that yielded P < .05 for at least 1 (host subset vs outcome variable) comparison. No statistically significant associations were observed for the outcome variable “fecal predominance” or for host subsets: adult (nonspouse; n = 19) and pet (n = 19).

^eHost subset refers to the subject from whom the fecal sample yielded the Escherichia coli strain. Some strains were present in multiple host subsets.

^fOdds ratios and correlations are shown only for significant associations.

^g P values are shown where P < .05. Comparisons involving dichotomous variables (persistence and sharing) were assessed using the (n-1) chi-square test or binomial logistic regression, whereas those involving continuous variables (predominance, household prevalence) were assessed using the Mann-Whitney test.

Colonization Outcomes: Clinical Isolate Matching

Of the 16 total (unique-by-household) clinical isolate–matching fecal strains, 15 were identified in veterans, 7 in spouses, and 4 in pets. As compared with the 123 fecal-only strains, the 16 clinical isolate–matching fecal strains exhibited significantly greater persistence (P < .001), sharing (P = .05), and household prevalence (P < .001), but similar fecal predominance (Table 4).

Table 4.

Relationship Between Molecular Traits and Colonization Outcomes Among 139 Fecal Escherichia coli Strains

			Predominance				Persistence				Sharing				Prevalence
			Univar.		Multivar. a		Univar.		Multivar.b		Univar.		Multivar.c		Univar.		Multivar.d
Traitse		Strains, No.	Cor.	P f	Co	P	ORg	P	OR	P	OR	P	OR	P	Co	P	Co	P
Matches CI		16					5.66	<.001	4.20	.02	3.24	.005			.37	<.001	.33	<.001
Phylogroup	B1	22													–.35	<.001
	F	7	–.20	.02	–.17	.04
Lineages	ST131	14					1.36	.002							.18	.04
	ST131-H30	13					5.12	.002							.20	.02
	ST73	8													–.22	.01	–.14	.04
	ST95	16									3.24	.02
	ST648	5	–.19	.02
Genesh
Adhesin	papAH	37	.28	.001											.17	.048
	papC	37	.28	.001											.17	.048
	papG II	25	.22	.01
	papG	39	.17	.04
	papEF	36	.26	.002
	sfaS	20	.18	.04
	iha	23					2.22	.02							.17	.04	.15	.04
Toxin	hlyD	22	.20	.02	.21	.01
	hlyF	9	.22	.01	.19	.02
	sat	27					2.45	.01
	vat	47									2.27	.04	2.42	.04
Siderophore	fyuA	78					1.67	.04
	ireA	21	.21	.02
Protectin	iss	6	.19	.03
	kpsMll	70					2.59	.004			2.08	.04			.25	.004
	usp	59					1.69	.02							.21	.02
Mobility	H7 fliC	18	.22	.01	–.17	.04
Misc.	ompT	87									1.82	.02
	ExPEC	61					2.19	.04

Abbreviations: CI, clinical isolate; Cor., correlation; ExPEC, extraintestinal pathogenic E. coli; Misc. miscellaneous; Multivar., multivariable; OR, odds ratio; Prevalence, household prevalence; Univar., univariable.

^aMultivariable candidate predictor variables included phylogroup F, papAH (P fimbriae), hlyF (variant hemolysin), sfaS (S fimbriae adhesin), hlyD (alpha hemolysin), ireA (siderophore receptor), and H7 fliC (flagellar variant).

^bMultivariable candidate predictor variables included spouse, matches CI, ST131-H30, sat (secreted auto-transporter toxin), usp (uropathogenic-specific protein), kpsMII (group 2 capsules), ompT (outer membrane protease), and ExPEC (defined as ≥ 2 of the following: papAH and/or papC [counted as 1: P fimbriae], sfa/focDE [S and F1C fimbriae], afa/draBC [Dr-binding adhesins], iutA [aerobactin receptor], and kpsMII).

^cMultivariable candidate predictor variables included veteran, spouse, matches CI, ST95, vat (vacuolating toxin), and kpsMII.

^dMultivariable candidate predictor variables were veteran, spouse, child, matches CI, phylogroup B1, ST73, ST131-H30, papAH, iha (adhesin-siderophore receptor), usp, and kpsMII.

^eMolecular traits shown are those that yielded P < .05 for at least 1 outcome variable. Three categories of molecular traits were excluded from the table: (i) Traits detected in ≥5 strains but yielding P > .05: ST127, papGIII (P fimbriae), sfa/focDE, focG (F1C fimbriae adhesin), hra (heat-resistant agglutinin), fimH (type 1 fimbriae adhesin), cnf1 (cytotoxic necrotizing factor), cdtB (cytolethal distending toxin), pic (protein associated with intestinal colonization), iroN (salmochelin receptor), iutA, neuB (K1 capsule variant), kfliC (K5 capsule variant), K2/K100 (capsule variants), rfc (O4 lipopolysaccharide synthesis), cvaC (colicin/microcin V), traT (serum resistance–associated), ibeA (invasion of brain endothelium), malX (pathogenicity island marker), clbB and clbN (colibactin polyketide synthesis). (ii) Traits detected in <5 strains (too few for statistical analysis): phylogroup C, ST12, ST141, ST372, ST14, ST405, 016, CGA, ST131-H30Rx, papGI (P fimbriae), kpsMTIII (group 3 capsules), afa/draBC, tsh (temperature-sensitive hemagglutinin), and astA (enteroaggregative Escherichia coli toxin). (iii) Traits sought but not detected: ST144, O15 clonal group (ST31 complex), afaE8 (variant afimbrial adhesin), bmaE (M fimbriae), gafD (G fimbriae), F17 (adhesin variant), clpG (adhesin variant), K15 kpsM (capsule variant).

^f P values are shown where P < .05. For univariable analysis, comparisons involving dichotomous variables (persistence and sharing) were assessed using the (n-1) chi-square test or binomial logistic regression, whereas those involving continuous variables (predominance, household prevalence) were assessed using the Mann-Whitney test or regression. Multivariable analysis used backward stepwise variable entry.

^gOdds ratios and correlations are shown only for significant associations (P < .05).

^hVirulence genes: papAH, papC, papGII, papG, and papEF (P fimbriae) sfaS, iha, hlyD, hlyF, sat, vat, fyuA (yersiniabactin receptor), ireA, ompT, iss (increased serum survival), kpsMll, usp, H7 fliC, and molecular ExPEC status.

Colonization Outcomes: Molecular Traits

Three categories of molecular traits (phylogroups, clonal lineages, and virulence genes) were assessed as predictors of fecal colonization outcomes. By phylogroup, the fecal E. coli strains were distributed as follows (% of 139): B2, 48%; A, 21%; B1, 17%; D, 9%; F, 5%; and C, 1%. Phylogroups B2 and D ranked numerically highest for fecal predominance, persistence, and household prevalence, and for sharing were surpassed only by phylogroup F (Table 5). However, statistically significant associations were limited to phylogroups F (fecal predominance [negative]: P = .02) and B1 (household prevalence [negative]: P < .001) (Table 4).

Table 5.

Rank Order of Phylogroups for Colonization Outcomes Among 139 Fecal Escherichia coli Strains

		Rank Order of Phylogroups for Each Outcome Variableb
Phylogroupa	Strains, No.	Predominance	Persistence	Sharing	Prevalence
A	29	3	5	4	3
B1	23	4	2	5	5
B2	67	2	1	2	2
D	12	1	2	3	1
F	7	5	5	1	4

Abbreviations: Prevalence, household prevalence.

^aPhylogroups shown are those that accounted for at least 5 strains. Phylogroup C (not shown) accounted for only 1 strain. Phylogroup E did not account for any strains.

^bNumerical rank order for the corresponding colonization outcome (1 = highest, 5 = lowest).

Of the 15 studied clonal lineages and sublineages, 6 qualified numerically for statistical analysis, including (number of strains) ST95 (16), ST131 (14), ST131-H30 (13), ST73 (8), ST127 (8), and ST648 (5). Each such lineage except ST127 was associated significantly with ≥1 colonization outcome (Table 4). Specifically, ST131 and ST131-H30 predicted persistence (for each, P = .002) and household prevalence (P = .04 and P = .02, respectively), ST95 predicted sharing (P = .02), ST73 (negatively) predicted household prevalence (P = .01), and ST648 (negatively) predicted fecal predominance (P = .02).

Of the 49 studied virulence genes, 43 were detected; 38 qualified for statistical analysis. Of these, 18 (47%) were associated significantly with ≥1 colonization outcome, including fecal predominance (11 genes), persistence (5 genes), household prevalence (5 genes), and sharing (3 genes) (Table 4). Of the 18 colonization-associated genes, 1 was associated with 3 outcomes, 4 with 2 outcomes each, and 13 with 1 outcome each. Similarly, molecular ExPEC status (n = 61) predicted persistence (P = .04). Notably, the 11 fecal predominance–associated genes overlapped minimally (only papAH and papC) with those that predicted within-household colonization patterns (Table 4).

Colonization Outcomes: Antibiotic Resistance

Of the 24 resistance markers, 14 were detected in ≥5 fecal strains each, so they could be analyzed statistically. Of these, 7 (50%) corresponded significantly with ≥1 outcome variable in univariable analyses (Table 6). For persistence, the 4 significant predictors were piperacillin (P ≤ .001), amoxicillin-clavulanic acid (P = .02), ciprofloxacin/levofloxacin (P = .006), and nalidixic acid (P = .02). For prevalence, the 5 significant predictors were piperacillin (P ≤ .001), amoxicillin-clavulanic acid (P = .001), nalidixic acid (P = .04), ampicillin (P = .02), and ampicillin-sulbactam (P = .046). For sharing, the only significant predictor (negative) was streptomycin (P = .03). No antibiotic predicted fecal predominance.

Table 6.

Relationship Between Antibiotic Resistance and Colonization Outcome

		Persistence		Sharing		Prevalence
Antibiotica	No.	Odds Ratiob	P	Odds Ratio	P	Correlation	P
Ampicillin	50	-	-	-	-	.21	.02
Piperacillin	28	3.63	<.001			.34	<.001
Amoxicillin-clavulanic acid	35	1.87	.02	-	-	.30	.001
Ampicillin-sulbactam	23					.17	.05
Ciprofloxacin/levofloxacin	24	2.76	.006	-	-	-	-
Naladixic acid	29	1.65	.02			.17	.04
Streptomycin	36	-	-	.09	.03	-	-

^aAntibiotics and outcome variables shown are those that yielded P < .05 for at least 1 (antibiotic-vs-outcome variable) comparison. No statistically significant associations were observed for the outcome variable “fecal predominance” or for the remaining 17 antibiotics (amikacin, aztreonam, cefazolin, cefotaxime, cefoxitin, ceftazidime, ceftriaxone, cephalothin chloramphenicol, ertapenem, fosfomycin, gentamicin, imipenem, piperacillin-tazobactam, teicoplanin, trimethoprim, trimethoprim-sulfamethoxazole).

^bAll associations were outcomes of univariable analysis. Notably, in an exploratory multivariable analysis including antibiotics along with all other strain characteristic variables (ie, host demographic and molecular traits), streptomycin was the only resistance variable that significantly predicted any colonization outcome (negatively predictive of household sharing: not shown).

Multivariable Analysis

Given the numerous univariable host and molecular predictors of colonization outcomes, multivariable analysis was used to identify primary associations using a single representative for each cluster of correlated variables (correlation coefficient ≥ 0.7). In backward stepwise multivariable models, veteran source remained predictive of sharing and household prevalence (P = .02 and P = .01, respectively) and spouse source remained predictive of persistence, sharing, and household prevalence (P = .02, P < .001, and P < .001, respectively) (Table 3). Clinical isolate–matching (vs fecal only) status remained predictive of persistence and household prevalence (P = .02 and P < .001, respectively). One or more molecular traits predicted fecal predominance (phylogroup F: P = .04; hlyF: P = .02; hlyD: P = .01; H7 fliC: P = .04), sharing (vat: P = .04), and household prevalence (ST73: P = .04 [negative], iha [adhesin-siderophore receptor]: P = .04). By contrast, none predicted persistence (Table 4).

Additional multivariable analyses involving molecular traits only (ie, excluding demographic subset and clinical isolate–matching status) were used to further assess the relationship between molecular traits and colonization outcomes. These models identified slightly different predictive molecular traits than did the initial models: For persistence, ST131-H30 was newly predictive; for sharing, vat lost significance, whereas ST95 was newly predictive; and for household prevalence, iha lost significance, ST73 remained predictive, and phylogroup B1 was newly predictive (Supplementary Tables 3–6). Models using forced or forward stepwise variable entry yielded results largely similar to those of the backward stepwise entry models, although forced entry usually identified fewer predictors than did the stepwise approaches (Supplementary Tables 3–6).

DISCUSSION

Our longitudinal surveillance of fecal E. coli strains from 22 veterans with a clinical E. coli isolate and their 46 human and animal household members yielded 3 main findings, with potentially important implications regarding the ecology of this important extraintestinal pathogen. Specifically, within-household fecal colonization behaviors were associated with (i) multiple virulence-associated clonal lineages and genes, (ii) host demographic subset, and (iii) matching the household’s index clinical isolate. This suggests that traits associated traditionally with virulence likely also contribute to fecal colonization behavior.

Our first main finding was that extraintestinal virulence-associated lineages and genes predicted fecal colonization behavior. The observed colonization behaviors of ST131 and its H30 subclone are consistent with prior reports of ST131’s remarkable ability to disseminate and persist [7, 11, 13, 15, 17, 18, 30]. Likewise, here ST95, a prominent endemic ExPEC lineage, also exhibited increased within-household strain sharing. Conceivably, superior colonization/sharing ability contributes to these lineages’ epidemiological success. By contrast, ST73, another prominent cause of extraintestinal E. coli infections [31] that in previous studies exhibited increased sharing and persistence [12, 32], here was associated negatively with household prevalence. These between-study and between-lineage differences suggest that although virulence and commensalism may align generally, exceptions occur, perhaps related to specific lineages and/or host populations or stochastic factors.

Additionally, 18 virulence genes and molecularly defined ExPEC status predicted 1 or more colonization outcomes. These genes represented diverse functional categories, precluding simple mechanistic interpretations and suggesting instead a multifactorial effect.

Of the 18 colonization-associated genes, 8 (iha, hlyF, sat, vat, ireA, ompT, iss, and H7 fliC) were newly identified as predicting colonization outcomes, whereas 10 had been so identified previously. Of these 10, 5 represent the pap (P fimbriae) operon, which previously was associated with gut colonization and/or strain sharing among men with febrile urinary tract infection and their sex partners [12], infants and schoolgirls [9, 33], members of a 3-subject household [34], and women with cystitis and their sex partners [8]. The other 5 previously identified genes (fyuA, kpsMII, sfaS, hlyD, and usp [uropathogenic-specific protein]) likewise were associated previously with colonization and/or strain sharing in diverse contexts and hosts [12, 34–36].

By contrast, the 8 newly identified colonization-associated genes, which similarly represented diverse functional categories, were assessed previously in only 1 or 2 colonization studies each [12, 19]. Our study’s larger number of households and more extended follow-up may underlie our novel identification of colonization-related associations for these genes. Collectively, the present findings confirm and extend previous evidence that so-called extraintestinal “virulence factors,” or traits linked to them, may contribute to gut colonization and/or strain sharing.

Here, phylogroups B2 and D ranked highest for fecal predominance, persistence, and prevalence and second highest (after phylogroup F) for sharing. This aligns generally with evidence that phylogroup B2—the most extraintestinal pathogenic E. coli phylogroup—corresponds with within-household strain sharing [12, 14, 19, 32, 37, 38]. Likewise, here phylogroup B1, generally regarded as less virulent, was associated negatively with fecal colonization. Paradoxically, phylogroup F, a recently delineated phylogroup associated with antimicrobial resistance and virulence [39, 40], behaved similarly to group B1. Collectively, these phylogroup-related associations provide added support for a linkage between colonization and extraintestinal virulence.

Our second main finding was that even in multivariable models 1 or more host demographic subsets predicted persistence (spouse), strain sharing (spouse and veteran), and overall household prevalence (spouse and veteran) but not fecal predominance. Prior studies have associated E. coli strain sharing with sexual behaviors [3, 8, 12], which we did not assess; sexual or other types of host–host contact may underlie our observed associations of strain sharing with spouses and veterans. Our findings regarding demographic subsets suggest that diverse factors other than inherent strain characteristics, including host biology and behaviors, also influence E. coli colonization phenotypes.

Our third main finding was that clinical isolate–matching fecal strains exhibited increased household colonization behavior, specifically sharing, persistence, and overall household prevalence. This additionally supports a link between commensalism and virulence. Notably, clinical isolate–matching fecal strains were mostly (10/16) fluoroquinolone-resistant, and 6/10 such strains were ST131-H30, underscoring the clinical importance of colonizing ExPEC strains.

Some propose that E. coli virulence genes are byproducts of commensalism, with certain traits promoting both colonization and infection-causing ability [6], as supported by human epidemiological observations and animal model studies [7–9, 12–14, 19, 21, 32, 41]. Our work supports this paradigm of co-evolution of virulence and commensalism, expands the number of clonal lineages and virulence factors known to be associated with gut colonization, and suggests that host factors also may play an important role in E. coli household colonization dynamics.

In addition to host and molecular factors, we found that resistance to several common antibiotics predicted increased persistence of strains in serial fecal samples (piperacillin, amoxicillin-clavulanic acid, ciprofloxacin/levofloxacin, nalidixic acid) and overall prevalence in household fecal samples (piperacillin, ampicillin, amoxicillin-clavulanic acid, ampicillin-sulbactam, nalidixic acid). By contrast, antibiotic resistance predicted neither increased predominance of a strain within individual fecal samples nor increased sharing among household members.

Our study has limitations. First, the large number of candidate predictor variables, although appropriate for an exploratory study, risked finding associations by chance alone, whereas small numbers in certain subgroups risked missing true associations. Second, the study’s observational nature precluded causality assessments. Third, use of ciprofloxacin-supplemented plates may have biased toward detection of ciprofloxacin-resistant strains, although only 6 (4% of 139) strains came only from ciprofloxacin-supplemented plates. Fourth, the focus on veterans with a clinical E. coli isolate may have biased toward detecting associations involving clinical isolate–matching strains. Fifth, we studied only a fraction of all possible epidemiological and bacterial variables that might influence colonization behavior, for example, host comorbidities, exposures (including antibiotic use), and behaviors, and other genes, gene expression/regulation, and the gut microbiota and phageome.

Our study also has strengths. First, it uniquely assessed virulence genes and phylogenetic subsets as colonization predictors within >20 multimember households. Second, the extensive household fecal sampling permitted robust analyses of sharing and persistence of fecal E. coli strains. Third, use of generic index clinical E. coli isolates provided clinical relevance, unbiased by focusing on dramatic clinical scenarios [18]. Fourth, in contrast to prior studies involving children [21, 30, 33, 36, 42], our study involved mostly adults (45/49 [92%] of human subjects)—making its results more generalizable to adults and to veterans, who usually are elderly men [43], an understudied population.

In summary, multiple host and bacterial characteristics predicted E. coli fecal colonization behaviors among veterans and their household members. These findings support and extend the concept that recognized pathogenic lineages (eg, ST131) and putative extraintestinal virulence factors (eg, P fimbriae, other adhesins, toxins, siderophores, and protectins) may feature importantly in both individual- and household-level gut colonization. Thus, many “virulence factors” may also be “colonization factors,” and many “virulent lineages” also “elite colonizing lineages.” Additionally, host demographic subset strongly predicted household colonization; thus, host biology and/or behaviors likely contribute importantly to E. coli colonization patterns. Lastly, clinical isolate–matching fecal strains colonized most extensively, underscoring the relationship between superior gut colonization and infection-causing proficiency. Given the important clinical implications of E. coli colonization, our findings suggest that future interventions that modulate colonization behavior might represent new ways of preventing human disease.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.