Genomic Characterization of Enteroaggregative Escherichia coli From Children in Mali

Abstract

Background. Enteroaggregative Escherichia coli (EAEC) is a cause of epidemic and sporadic diarrhea, yet its role as an enteric pathogen is not fully understood.

Methods. We characterized 121 EAEC strains isolated in 2008 as part of a case-control study of moderate to severe acute diarrhea among children 0–59 months of age in Bamako, Mali. We applied multiplex polymerase chain reaction and comparative genome hybridization to identify potential virulence factors among the EAEC strains, coupled with classification and regression tree modeling to reveal combinations of factors most strongly associated with illness.

Results. The gene encoding the autotransporter protease SepA, originally described in Shigella species, was most strongly associated with diarrhea among the EAEC strains tested (odds ratio, 5.6 [95% confidence interval, 1.92–16.17]; P = .0006). In addition, we identified 3 gene combinations correlated with diarrhea: (1) a clonal group positive for sepA and a putative hemolysin; (2) a group harboring the EAST-1 enterotoxin and the flagellar type H33 but no other previously identified EAEC virulence factor; and (3) a group carrying several of the typical EAEC virulence genes.

Conclusion. Our data suggest that only a subset of EAEC strains are pathogenic in Mali and suggest that sepA may serve as a valuable marker for the most virulent isolates.

It is estimated that diarrhea causes at least 1.5 million deaths annually, mostly in children <5 years of age [1]. Although in aggregate the diarrheagenic Escherichia coli (DEC) pathotypes comprise the most common bacterial pathogens worldwide [2], each DEC pathotype is clinically, epidemiologically, and pathogenetically distinct. For some pathotypes, the key virulence factors are known, at least in part, whereas for other pathotypes, the key virulence genes and how they coordinately function in the setting of enteric disease remain elusive.

The enteroaggregative E. coli (EAEC) pathotype has been implicated in travelers’ diarrhea [3], in endemic diarrhea among children in both industrialized [4] and resource-poor countries [5], and in persistent diarrhea among individuals infected with human immunodeficiency virus. A recent outbreak of Shiga toxin–producing EAEC highlights its pathogenic potential [6]. Despite this, the molecular epidemiology of EAEC infection remains unclear, largely due to imperfect recognition of the true pathogenic factors within the broadly defined pathotype.

Most EAEC strains colonize the intestinal mucosa via the aggregative adherence fimbriae (AAFs), which include at least 4 major antigenic variants [7–10]. AAFs are transcriptionally regulated by an AraC/XylS family activator called AggR [7, 11]. AggR is also required for expression of genes encoding dispersin (the aap gene), the Aat dispersin translocator [12], and the chromosomal cluster termed Aai, encoding a type VI secretion system [13]. Factors not under AggR control include the Air adhesin, a regulator termed EilA, the EAEC heat-stable toxin EAST-1 (encoded by the astA gene), and a set of toxins termed the serine protease autotransporters of Enterobacteriaceae (SPATEs).

SPATEs have been organized phylogenetically into 2 classes. Members of class 1 are cytotoxic to epithelial cells [14]; class 1 SPATEs found in EAEC strains include the plasmid-encoded toxin (Pet) and its 2 homologs, Sat [15] and SigA [16]. The class 2, or noncytotoxic, SPATEs include Pic, a mucinase that promotes intestinal colonization [17, 18]. As with cytotoxic SPATEs and Pic, we have recently reported that the class 2 SPATE SepA is found commonly among EAEC strains [19]. SepA is a cryptic protease originally described in Shigella species, and is reported to contribute to intestinal inflammation [20]. Importantly, none of these factors are found in all EAEC isolates, and no single factor has ever been consistently implicated in EAEC virulence.

Here, we characterize 121 EAEC strains isolated as part of a case-control study of acute moderate to severe diarrhea among children aged 0–59 months in Mali. We report that the sepA gene and flagellar type H33 are strongly associated with illness, and we define additional sets of virulence genes and factors that are important in this population.

MATERIALS AND METHODS

Study Design

The strains utilized were isolated in the course of a prospective multicenter case-control study (Global Enteric Multi-Center Study, GEMS) of moderate to severe diarrhea among children <5 years of age. Full details of the GEMS design will be published elsewhere. In brief, children ≤59 months presenting to health centers for care with a complaint of diarrhea within the previous 7 days were considered eligible. Cases were enrolled upon parental consent if they met criteria for moderate to severe diarrhea comprising signs of moderate to severe dehydration (sunken eyes, decreased skin turgor), dysentery (blood in stool), or if they were deemed to require hospitalization or intravenous rehydration. Diarrhea was defined as the passage of ≥3 or more unformed stools within a 24-hour period. A stool sample was obtained at enrollment and analyzed comprehensively for bacterial, viral, and protozoal agents. An age-matched asymptomatic control from the same neighborhood was enrolled for each case; a stool sample was obtained from the control child and analyzed similarly.

Specimen Processing and Microbiological Analysis

A single, fresh, whole stool specimen was collected from cases and controls at enrollment for the recovery of potential enteropathogens. Various specific growth media were used for detecting the bacterial pathogens. Up to 3 colonies with the appearance of E. coli on MacConkey agar were selected from each sample and tested using multiplex polymerase chain reaction (PCR) for enterotoxigenic E coli (ETEC) (heat-labile [LT] and heat-stable [ST] enterotoxins), enteropathogenic E. coli (EPEC) (eae and bfpA), and EAEC (aaiC and aatA). Any colonies that were positive for either aaiC (chromosomally encoded) or aatA (encoded on the pAA plasmid) were considered EAEC for the purposes of this analysis.

Serotyping

Somatic (O) and flagella (H) antigens were identified as described elsewhere [21]; the following designations were included: “O rough,” the boiled culture auto-agglutinated, suggesting absence of O antigen; “O?,” it could not be determined whether the strain produces an O antigen (precipitation with Cetavlon indicates an acidic polysaccharide that could represent capsular K antigen); and “O+,” the O antigen is present but could not be typed. Serotyping was performed at the International Escherichia and Klebsiella Centre (World Health Organization), Department of Microbiological Surveillance and Research, Statens Serum Institut, Copenhagen, Denmark.

Polymerase Chain Reaction

Primers and conditions for detecting sequences encoding 21 putative virulence genes, which are described in Table 1, were used in 4 multiplex reactions. Multiplex PCR 1 was performed as previously described [19], with the addition of primers targeting astA. Multiplexes 2–4 were performed using PCR mastermix (2X) according to the manufacturer’s instructions (Fermentas International), with the addition of 1 μL 50 mM magnesium chloride per 50 μL reaction. A DNA template was prepared by boiling a suspension of 10 isolated colonies in 200 μL distilled water. PCR reaction cycles were as follows: (1) 2 minutes denaturation at 95°C, (2) 50 seconds denaturation at 94°C, (3) annealing for 1.5 minutes, and (4) extension for 1.5 minutes at 72°C with 35 cycles returning to step 2. The final extension was 10 minutes at 72°C. Products were amplified using an Eppendorf Mastercycler Gradient thermal cycler (Eppendorf North America) and separated in 2% agarose gels.

Table 1.

Primers Used for the 4 Multiplex Polymerase Chain Reactions (PCRs) and 3 Monoplex PCRs, Description of Target Gene, Product Size in Base Pairs, Annealing Temperature, and Concentration of the Primers

Multiplex PCR	Gene/Target	Description of Target	Primer Sequence (5′- 3′)	PCR Product, bp	Annealing Temperature Primer Concentration (°C), pmol/μL	GenBank Accession No.
1	astA	EAST-1 heat-stable toxin	ATGCCATCAACACAGTATAT [22]	110	58/20	L11241
			GCGAGTGACGGCTTTGTAGT [22]
	pet	Plasmid-encoded toxin	GGCACAGAATAAAGGGGTGTTT [23]	302	58/25	AF056581
			CCTCTTGTTTCCACGACATAC [23]
	sigA	IgA protease-like homolog	CCGACTTCTCACTTTCTCCCG [19]	430	58/30	NC_004337
			CCATCCAGCTGCATAGTGTTTG [19]
	pic	Serine protease precursor	ACTGGATCTTAAGGCTCAGGAT [23]	572	58/25	AF097644
			GACTTAATGTCACTGTTCAGCG [23]
	sepA	Shigella extracellular protease	GCAGTGGAAATATGATGCGGC [23]	794	58/25	Z48219
			TTGTTCAGATCGGAGAAGAACG [23]
	sat	Secreted autotransporter toxin [15]	TCAGAAGCTCAGCGAATCATTG [19]	932	58/25	AE014075
			CCATTATCACCAGTAAAACGCACC [19]
2	ORF3	Cryptic proteina	CAGCAACCATCGCATTTCTA	121	57/35	…
			CGCATCTTTCAATACCTCCA
	aap	Dispersin, antiaggregation protein [12]	GGACCCGTCCCAATGTATAAb	250	57/25	Z32523
			CCATTCGGTTAGAGCACGATb
	aaiC	AaiC, secreted protein	TGGTGACTACTTTGATGGACATTGTb	313	57/25	…
			GACACTCTCTTCTGGGGTAAACGAb
	aggR	Transcriptional activator	GCAATCAGATTAARCAGCGATACAb	426	57/25	Z18751
			CATTCTTGATTGCATAAGGATCTGGb
	aatA	Dispersin transporter protein	CAGACTCTGGCRAAAGACTGTATCATb	642	57/35	AY351860
			CAGCTAATAATGTATAGAAATCCGCTGTb
3	agg4A	AAF/IV fimbrial subunit	TGAGTTGTGGGGCTAYCTGGAb	169	57/25	EU637023
			CACCATAAGCCGCCAAATAAGCb
	aggA	AAF/I fimbrial subunit	TCTATCTRGGGGGGCTAACGCTb	220	57/20	Y18149 AY344586
			ACCTGTTCCCCATAACCAGACCb
	aafA	AAF/II fimbrial subunit	CTACTTTATTATCAAGTGGAGCCGCTAb	289	57/25	AF012835
			GGAGAGGCCAGAGTGAATCCTGb
	agg3A	AAF/III fimbrial subunit	CCAGTTATTACAGGGTAACAAGGGAAb	370	57/25	AF411067
			TTGGTCTGGAATAACAACTTGAACGb
	agg3/4C c	Usher, AAF/III-IV assembly unit	TTCTCAGTTAACTGGACACGCAATb	409	57/35	AF411067 AB255435 EU637023
			TTAATTGGTTACGCAATCGCAATb
			TCTGACCAAATGTTATACCTTCAYTATGb
	aafC	Usher, AAF/II assembly unit	ACAGCCTGCGGTCAAAAGCb	491	57/25	AF114828
			GCTTACGGGTACGAGTTTTACGGb
4	ORF61	Plasmid-encoded hemolysina	AGCTCTGGAAACTGGCCTCT	108	57/10	…
			AACCGTCCTGATTTCTGCTT
	eilA	Salmonella HilA homolog	AGGTCTGGAGCGCGAGTGTTb	248	57/30	…
			GTAAAACGGTATCCACGACCb
	capU	Hexosyltransferase homolog	CAGGCTGTTGCTCAAATGAAb	395	57/25	AF134403
			GTTCGACATCCTTCCTGCTCb
	air	Enteroaggregative immunoglobulin repeat protein [24]	TTATCCTGGTCTGTCTCAAT	600	57/25
			GGTTAAATCGCTGGTTTCTT
Singleplex PCR	espY2	Non-LEE-encoded type III secreted effector	CGCAAAAGATCCGGAAAATAb	216	59/25	ECSP_0073
			TCAGCATTGCTCAGGTCAACb
	rmoA	Putative hemolysin expression-modulating protein	TTACCTTACATATTTCCATATCb	210	60/25	ECUMN_0072
			CGAAAACAAAACAGGAATGGb
	shiAd	shiA-like inflammation suppressord	CAGAATGCCCCGCGTAAGGC [25]	292	57/25	ECB_03517
			CACTGAAGGCTCGCTCATGATCGCCG [25]

Abbreviations: bp, base pair; PCR, polymerase chain reaction.

^aUnpublished.

^bDesigned for this study.

^cTwo forward primers and 1 reverse primer were used for the amplification of agg3/4C. This primer set was designed to amplify the usher gene from both AAF/III and IV, hence the name.

^dPrimers used to amplify the shiA gene were forward primer from sisA gene and reverse primer from sisB gene, as described by Lloyd et al [25].

Individual amplification reactions to detect genes designated rmoA, espY2, and shiA were done in 25 μL reaction volumes using crude bacterial cell lysates; PCR reactions were performed as multiplex 2–4. The final extension was 10 minutes at 72°C.

The phylogenetic groups A, B1, B2, and D were determined using triplex PCR methods employing phylogenetic group-specific primers for 2 genes, chuA and yjaA, and a cryptic DNA fragment, TspE4C2. The grouping was coupled to a dichotomous decision tree according to Clermont et al [26].

The following strains where used as controls for detection of target genes: JM221 (aggA, sat) [27], 042 (aatA, aggR, aaiC, aap, ORF3, pic, pet, astA, aafA, aafC, air, capU, eilA) [28], 55989 (agg3A, agg3/4C) [8], H223-1 (sigA) [29], C1010-00 (agg4A, agg3/4C, sat, sepA) [30], MC1061 (negative control), J96 (chuA, yjaA) [26], CFT073 (chuA, yjaA, TspE4.C2) [31], C452-97 (TspE4.C2) [32], and EDL933 (chuA) [33].

Genomic Hybridization

Comparative genome hybridization (CGH) was performed on all the sepA-positive strains as well as sepA-negative strains C801-09 and C46-10 and reference EAEC isolates as previously described [34]. The pan-genome microarrays used in this study were designed by FDA-ECSG Array Probe Set Design and represent the genomes of 32 diverse E. coli and Shigella species, as well as 46 enteric plasmid sequences [35]. Initial data analysis was performed with the Gene Chip Operating System suite of tools provided by Affymetrix. Additional analysis was performed using the Affymetrix power tools software. The MAS5 algorithm was utilized with the perfect match and mismatch calculations and a Tau of 0.150 to detect which probes were present or absent. Features that were present or absent in all samples were removed from further analysis. The resulting features, known as the variable gene set, were analyzed using Multiple Experiment Viewer, version 4.5. The cladogram was constructed using the 12 673 variable features in this dataset, which contained hybridization data from 36 strains. The relationship was determined using hierarchical clustering with Pearson correlation, using the absolute distance and complete linkage run with 500 bootstrap calculations.

Statistics

We utilized classification and regression tree (CART) Pro Version 6.0 (Salford Systems) software inputting 21 or 24 factors of interest as binary (present/absent) independent predictive variables along with a continuous “factor total” that was a sum of all factors including flagellum type H33. Case/control status was the binary dependent outcome variable.

RESULTS

Initial Characterization of EAEC Strains

After 1 year of surveillance, EAEC strains were isolated as the sole DEC pathogen from 60 children with diarrhea and 61 asymptomatic controls. The lack of association of EAEC with diarrhea among the cases persisted even when the presence of other potential pathogens, stratifying for age, was considered or when either aatA or aaiC alone or in combination were considered.

One EAEC isolate was selected from each stool sample that yielded EAEC by multiplex PCR. The 121 EAEC strains belonged to diverse serotypes (Table 2; cases listed in Table 2A and controls listed in Table 2B). Examination of the correlation between serotype and case/control status revealed that only flagellum type H33 was significantly more common among cases than controls (12 cases, 2 controls; odds ratio [OR], 5.9; P = .0138) (Table 3). EAEC cases and control strains were localized to similar positions within a previously published general E. coli phylogenetic tree (Tables 2 and 3).

Table 2a.

Characteristics of 60 Enteroaggregative Escherichia coli Strains Isolated From Children With Diarrhea

Table 2b.

Characteristics of 61 EAEC Strains From Asymptomatic Control Children.

Table 3.

Distribution of Enteroaggregative Escherichia coli Virulence Factors in Cases and Controls

EAEC	Cases (n = 60)		Controls (n = 61)		Total (N = 121)			Risk Estimate
Virulence Gene	No. (%)		No. (%)		No. (%)		Odds Ratio	[95% CI]	χ2	P Value
aatA	37	(61.7)	46	(75.4)	83	(68.6)	0.5	[.24–1.15]	2.7	.10
aggR	38	(63.3)	46	(75.4)	84	(69.4)	0.6	[.26–1.23]	2.1	.15
aaiC	32	(53.3)	26	(42.6)	58	(47.9)	1.5	[.75–3.15]	1.4	.24
aap	39	(65.0)	48	(78.7)	87	(71.9)	0.5	[.22–1.19]	2.8	.09
ORF3	49	(81.7)	55	(90.2)	104	(86.0)	0.5	[.17–1.39]	1.8	.18
sat	24	(40.0)	33	(54.1)	57	(47.1)	0.6	[.28–1.16]	2.4	.12
sepA	20	(33.3)	5	(8.2)	25	(20.7)	5.6	[1.92–16.17]	11.7	.0006
pic	29	(48.3)	27	(44.3)	56	(46.3)	1.2	[.58–2.41]	0.2	.66
sigA	8	(13.3)	7	(11.5)	15	(12.4)	1.2	[.40–3.51]	0.1	.76
pet	4	(6.7)	6	(9.8)	10	(8.3)	0.7	…	…	.74
astA	32	(53.3)	30	(49.2)	62	(51.2)	1.2	[.58–2.41]	0.2	.65
aafC	5	(8.3)	5	(8.2)	10	(8.3)	1.0	…	…	>.999
agg3/4C	42	(70.0)	40	(65.6)	82	(67.8)	1.2	[.57–2.63]	0.3	.60
agg3A	1	(1.7)	5	(8.2)	6	(5.0)	0.2	…	…	.21
aafA	3	(5.0)	3	(4.9)	6	(5.0)	1.0	…	…	>.999
aggA	11	(18.3)	21	(34.4)	32	(26.4)	0.4	[.18–.99]	4.0	.04
agg4A	5	(8.3)	1	(1.6)	6	(5.0)	5.5	…	…	.11
air	20	(33.3)	29	(47.5)	49	(40.5)	0.6	[.26–1.15]	2.5	.11
capU	48	(80.0)	51	(83.6)	99	(81.8)	0.8	[.31–1.99]	0.3	.61
eilA	50	(83.3)	53	(86.9)	103	(85.1)	0.8	[.28–2.09]	0.3	.58
ORF61	28	(46.7)	44	(72.1)	72	(59.5)	0.3	[.16–0.72]	8.1	.004
espY2	13	(21.6)	20	(32.8)	33	(27.3)	0.6	[.25–1.28]	1.9	.17
rmoA	30	(50.0)	23	(37.7)	53	(43.8)	1.7	[.80–3.39]	1.9	.17
shiA	21	(35)	22	(36.1)	43	(35.5)	0.5	[.45–2.01]	0.01	.92
EAEC Serogroup
O99	5	(8.3)	2	(3.3)	7	(5.8)	2.7	…	…	.27
O153	6	(10.0)	1	(1.6)	7	(5.8)	6.7	…	…	.61
H-	7	(11.7)	17	(27.9)	24	(19.9)	0.3	[.12–.98]	4.9	.04
H5	6	(10.0)	2	(3.3)	8	(6.6)	3.3	…	…	.16
H9	5	(8.3)	1	(1.6)	6	(5.0)	5.5	…	…	.21
H30	9	(15.0)	4	(6.6)	13	(10.7)	2.5	[.73–8.66]	2.2	.13
H33	10	(16.7)	2	(3.3)	12	(9.9)	5.9	[1.23–28.19]	6.7	.01
Phylogenetic Group
A	23	(38.3)	14	(22.9)	37	(32.2)	2.1	[.95–4.61]	3.4	.07
B1	15	(25)	13	(21.3)	28	(23.1)	1.2	[.53–2.78]	0.2	.63
B2	8	(13.3)	9	(14.7)	17	(14)	0.9	[.32–2.48]	0.05	.82
D	14	(23.3)	25	(40.9)	39	(32.2)	0.4	[.19–.96]	4.3	.04

P < .05 is significant. Fisher exact test was applied when the comparisons between cases and controls were <5 observations.

Abbreviation: EAEC, enteroaggregative Escherichia coli.

Frequencies of Virulence-Related Genes

In order to assess the roles of putative virulence factors in EAEC epidemiology, we developed 4 multiplex PCR assays for the characterization of 21 genes previously found in EAEC strains. The results of the PCR assays for all strains are listed in Table 2. Of the 21 genes scored, hypothetical ORF3 was the most frequently detected (86%) followed by eilA (85.1%), capU (81.8%), aap (71.9%), aggR (69.4%), and aatA (68.6%) (Table 3). There was a high degree of concordance of these genes, which has been demonstrated previously for the plasmid-encoded aap, aggR, and aatA genes [7, 12, 36–38]. Sixty-eight percent of the strains were positive for the usher-encoding gene agg3/4C (the ushers for AAF/III and AAF/IV variants are closely related). The most frequent AAF pilin gene was that of AAF/I, encoded by aggA (26.4%), followed by those of AAF/II (aafA), AAF/III (agg3A), and AAF/IV (agg4A) at 5% each (Table 3). The agg4A gene was found more frequently among cases than controls (5 cases and 1 control), although this difference did not reach statistical significance. A total of 71 strains (58.7%) were negative for a known AAF variant.

Of the 5 genes encoding SPATEs, the most frequent were sat (47.1%) and pic (46.3%). The least common SPATEs were pet (8.3%) and sigA (12.4%). The sepA gene was found in 25 strains (20.7%): 20 from cases and 5 from controls, yielding an OR of 5.6 (P = .0006) (Table 3). Among all the putative virulence factors scored, sepA was the only one significantly associated with moderate to severe diarrheal illness.

Significance of Combinations of EAEC Genes

In addition to considering each factor individually, we pursued a number of approaches to consider the importance of combinations of potential EAEC virulence factors. When crudely considering the collective number of virulence loci present (generating a virulence factor score, VFS), the average number of virulence genes from cases was 8.75 versus 9.5 from control isolates.

To consider combinations of factors, we employed CART analysis, which builds a model in stepwise fashion to yield the combination of factors most strongly associated with the queried outcome. Each branch of a CART output tree ends in a terminal “node”; each observation falls into exactly 1 terminal node; and each terminal node is uniquely defined by a set of rules, such as having or not having a certain factor.

We considered all genotypic and phenotypic assays performed and interrogated the association with case status. Figure 1 illustrates the best CART fit for the dataset. The analysis demonstrates that the presence of sepA, regardless of the presence or absence of any other scored genotype or phenotype among the sepA-positive strains, provides a strong association with diarrhea.

Figure 1.

Classification and regression tree (CART) classification tree topology reveals combinations of factors most strongly associated with moderate to severe diarrhea. We considered all genotypic and phenotypic assays performed: aatA, aggR, aaiC, aap, ORF3, sat, sepA, pic, sigA, pet, astA, aafC, agg3/4C, aafA, agg3A, aggA, agg4A, air, capU, eilA, ORF61, virulence factor score (VSF), and flagellum type H33. Each branch of the CART tree ends in a terminal “node” (red boxes), and each terminal node is uniquely defined by the presence or absence of a predictive factor such as a gene or VFS. The tree is hierarchical in nature. C701-09, C718-09, C801-09, and C46-10 are also shown on the dendrogram.

Among the sepA-negative strains, CART analysis suggested 2 additional trait clusters that were associated with moderate to severe diarrheal illness: 1 cluster included those strains harboring the flagellum H33 and the toxin EAST-1, whereas a second cluster lacked H33 but featured a VFS of 9, suggesting a combination of typical EAEC factors in addition to the Sat toxin.

Genomic Analyses

We hypothesized that the strain sets belonging to the nodes most strongly associated with diarrhea would reveal the presence of additional virulence determinants, which themselves might explain the observed clinical correlations. We therefore performed CGH analysis using a previously described microarray containing the full genomes of 32 E. coli and Shigella strains and the genes of an additional 46 E. coli plasmids [35]. For this analysis, we chose all 25 sepA-positive strains, 2 additional strains (C46-10 and C801-09) representing CART (Figure 1) nodes 2 (SepA absent, H33 present, EAST-1 present) and 3 (SepA absent, H33 absent, >8.5VFS, Sat present), and a set of archetype EAEC reference strains. Standard cluster analysis was performed on the microarray data (Figure 2). All isolates belonging to a common serotype clustered together in this analysis. Although cluster analysis did not suggest genomic differences discriminating cases and controls, the analysis did suggest that sepA-positive strains segregated into 2 major clusters (indicated as A and B in Figure 2). We chose for further genomic examination archetypal strains representing sepA-positive clusters A and B, as well as the 2 additional nodes (2 and 3) indicated by CART analysis (Figure 1).

Figure 2.

Cladogram of comparative genomic hybridization data of sepA-positive isolates (C34-10, C35-10, C679-09C, C682-09, C693-09, C701-09, C703-09, C716-09, C718-09, C719-09, C729-09, C736-09, C745-09, C764-09, C765-09, C769-09, C771-09, C778-09, C783-09, C796-09, C697-09, C707-09, C710-09, C734-09, and C748-09), sepA-negative isolates (C46-10 and C801-09), and reference isolates (C1096, 042, 55989, JM221, 17-2, 34b, 101-1, and HS). Notably, C46-10 was most closely related to Mexican enteroaggregative Escherichia coli (EAEC) strain JM221 (isolated from an adult traveler to Guadalajara [27]), and strain C801-09 was most closely related to EAEC strain 042, isolated from a child with diarrhea in Lima, Peru [28]. The phylogenetic comparison was performed using the 12 673 variable features of the 36 hybridizations included. The tree is built using a hierarchical clustering with Pearson correlation using both the absolute distance and complete linkage and viewed in FigTree (http://tree.bio.ed.ac.uk/software/figtree/). Isolates represented in black are reference isolates, controls are indicated in blue, and cases in red. The serotypes of the strains are indicated to the right. The gray boxes identify clusters of serotypes within the context of the larger tree, indicating that those serogroups are genomically similar.

Genome Analysis of sepA-Positive Strains

To represent the sepA-positive strain cluster A (Figure 2), we chose strain C718-09 for further genome analysis; to represent cluster B, we chose strain C701-09. Results are presented in Table 4. Genome analysis of strain C-718-09 did not reveal the presence of additional genes that were not also carried by nonpathogenic E. coli strains. C701-09 hybridized to an open reading frame that was 99% identical to the rmoA gene encoded on plasmid R100 (GenBank accession number Y13856.1); rmoA encodes a predicted 69 amino acid putative hemolysin expression-modulation protein that is 100% identical to protein RmoA found on plasmid R100 from E. coli [39]. The protein sequence of R100 RmoA exhibits 52% identity and 75% amino acid similarity with Hha protein from E. coli K12 (GenBank accession number NP_414993).

Table 4.

Comparative Genomic Hybridization of Strains C701-09, C718-09, C801-09, and C46-10 Against a Microarray That Comprises the Full Genomes of 32 Escherichia coli and Shigella Strains and the Genes of Additional 46 E. coli Plasmids

		Hybridization by Genome
Putative Virulence Genea	Accession No.	Nonpathogenicb	C701-09	C718-09	C801-09	C46 -10
Adhesins
csgA; cryptic curlin major subunita	SBO_2026	+	+	+	+	–
csgA; major curlin subunitb	LF82_0360	+	+	–	+	–
csgC; putative autoagglutination proteinb	ECUMN_1217	+	+	+	+	–
ecpD; putative chaperone protein EcpDb	SBO_0126	+	–	+	+	–
Fimbrial usher family proteinb	SbBS512_E2717	+	+	+	–	+
Flu; antigen 43 (Ag43)b	ECUMN_3400	+	+	–	–	+
Hemagglutinin familyc	SbBS512_E4026	+	–	+	+	–
Putative AidA-I adhesin-like protein	ECO26_3415	–	–	–	+	+
Putative AidA-I adhesin-like proteind	ECO26_1353	+	–	–	+	–
Putative chaperone protein EcpD	ECUMN_0137	–	+	–	+	–
Putative fimbrial biogenesis outer membrane usher protein	ECUMN_0019	–	+	–	+	–
Putative fimbrial proteinb	SbBS512_E2376	+	+	+	+	–
Putative fimbrial-like proteinb	SD.Y_0915	+	+	+	+	–
Putative invasinb	EcSMS35_1146	+	+	+	+	–
Putative type 1 fimbrial protein	ECSP_0022	–	–	–	+	–
sfmD; putative outer membrane export usher protein SfmDb	ECO26_0565	+	–	+	+	–
sfmF; putative fimbrial-like adhesin protein SfmFb	ECO26_0567	+	–	+	+	–
sfmH; putative fimbrial-like adhesin proteinb	ECUMN_0573	+	–	+	+	–
siiEA; adhesin for cattle intestine colonization	ECUMN_0527	–	–	–	+	–
yfaL; adhesin YfaLb	ECO26_3226	+	+	+	+	–
yfcP; putative fimbrial-like adhesin proteinb	BWG_2107	+	+	–	–	–
yfcQ; putative fimbrial-like adhesin proteinb	BWG_2108	+	+	–	–	–
yfcR; putative fimbrial-like adhesin proteinb	BWG_2109	+	+	–	–	+
yfcS; putative periplasmic pilus chaperoneb	BWG_2110	+	+	–	+	+
yfcS; putative periplasmic pilus exported chaperoneb	ECUMN_2676	+	+	–	+	+
yfcT; outer membrane export usher proteinb	ECDH10B_2499	+	+	–	+	+
yfcU; export usher proteinb	ECDH10B_2500	+	+	–	+	+
yfcU; outer membrane usher protein	E2348C_2477	–	+	–	+	+
yfcV; predicted fimbrialprotein–like protein	E2348C_2478	–	+	–	+	–
Toxins
Hcp-like proteinb	SSON_0233	+	–	+	+	–
hlyE; hemolysin Eb	ECO26_1695	+	+	+	+	–
Secretion Systems
espY2; Non-LEE-encoded Type III Secreted Effector	ECSP_0073	–	–	–	+	–
Hypothetical protein; type VI secretion system secreted protein VgrGb	ECSP_0240	+	+	–	+	+
Putative type II secretion protein (GspI-like)b	ECIAI1_3105	+	+	–	+	–
Putative type III secretion protein EpaRb	ECUMN_3195	+	–	–	+	–
T3SS effector-like protein EspL-homologb	ECO111_4829	+	–	+	+	–
tolC; outer membrane channel proteinb	SDY_3205	+	+	+	–	+
Type III secretion protein EpaQb	ECO26_3940	+	–	+	+	–
Type III secretion protein EpaRb	ECO103_3428	+	–	+	+	–
Type III secretion protein EprJb	ECO26_3933	+	–	+	+	–
Other
Hemolysin expression-modulating protein	EC55989_3351	–	+	+	+	–
Putative hemolysin expression-modulating protein RmoA	ECUMN_0072	–	+	+	–	+
Putative hemolysin co-regulated proteinb	SSON_0255	+	–	–	+	+
ShiA-like protein	ECB_03517	–	–	–	–	+

^aEAEC genes are listed in Table 2A and 2B.

^bOne hundred percent identities with HS and/or K12.

^cEighty-eight percent identities with HS.

^dFifty-five percent to 62% identities with HS and/or K12.

Genome Analysis of sepA-Negative Strains

Strains C46-10 and C801-09 were representative of the 2 sepA-negative nodes that were associated with diarrhea in Figure 1.

C46-10 Genome Analysis

Strain C46-10 best represented the confluence of factors identified in node 2 (Figure 1), characterized as sepA absent, H33 present, and EAST-1 present. By CGH, strain C46-10 hybridized to a large number of genes found among DEC pathotypes and Shigella speices (Table 4) and encodes a complete yfc gene cluster, which has been proposed to encode a novel usher-chaperone fimbrial adhesin [40]. C46-10 harbored elements of a type VI secretion system homologous to VgrG of Agrobacterium. However, this component of the newly described type VI systems is also found among nonvirulent isolates and has not yet been assigned any virulence function among DEC or Shigella strains.

C46-10 DNA was found to hybridize with the 347 amino acid ShiA-like protein from E. coli strain REL606 (GenBank accession number YP_003046696). The latter protein exhibited 97% identity with the ShiA protein initially described in Shigella flexneri 5a strain M90T (GenBank accession number AF141323) [41]. ShiA and related proteins identified in uropathogenic E coli and Shigella strains have been found to suppress the inflammatory response in animal models [42].

C801-09 Genome Analysis

C801-09 is closely related to the virulent archetype EAEC strain 042 and harbors many of the same virulence genes, including a near-complete plasmid-borne AggR regulon. It represents the most common serotype found in our study (O153:H30). Like C701-09, C801-09 harbored homologs of a large number of adhesins, including the siiEA locus that is associated with colonization of cattle by E. coli strain UMN026. C801-09 also harbored EspY2, a non-LEE-encoded type III secreted effector from E. coli O157:H7 strain TW14359 (GeneID: 8214639). Five proteins, EspY1-5 from the E. coli O157:H7 Sakai strain, possess an N-terminal WEX5F domain, which has been linked to type III secretion and is conserved in several well-characterized Salmonella effectors and in putative effectors from Edwarsiella and Sodalis [43].

Screening of the EAEC Collection for Presence of espY2, rmoA, and shiA Genes

Based on the CGH analysis from strains C46-10, C701-09, C718-09, and C801-09, we inferred that espY2, rmoA, and shiA were the factors most plausibly associated with virulence. Using PCR, we found that 35.5% of the EAEC strains from cases and a similar percentage from controls harbored the shiA gene and 27.3% of each group harbored espY2. The rmoA gene was found in 43.8% of the EAEC strains (50% from cases and 37.7% from controls). None of the 3 genes were independently associated with diarrheal illness (Table 3). However, when we repeated the CART analysis including the espY2, rmoA, and shiA genes (Figure 3), sepA once again exhibited a strong association, yet strains that were both sepA- and rmoA-positive were most strongly associated with disease (13 out of 14 strains positive for this combination were present among cases).

Figure 3.

Classification and regression tree analysis described in Figure 1, adding the genes shiA, espy2, rmoA (hemolysin expression-modulation protein). See Figure 1 legend for details of analysis

DISCUSSION

EAEC is a common diarrheal isolate, yet apart from those outbreak-associated, identification of truly pathogenic strains remains difficult. A large number of virulence factors and combinations have been associated with clinical illness in epidemiologic studies, and it is possible that either the principal determinants of pathogenicity vary by site and population or that the true determinants have not yet been identified.

We report the most detailed genomic characterization of EAEC performed to date, targeting a collection of 121 EAEC strains isolated from children in Mali with or without moderate to severe diarrhea. In agreement with previous reports [44–46], our strains belonged to a diverse range and combination of O:H and phylogenetic types. Although no specific O:H combination was associated with diarrhea, strains expressing the H33 flagellar antigen were found significantly more often in cases than in controls. This association may signify the existence of a specific set of virulence genes in strains of this H type.

To profile the virulence genes of our strain set, we developed and applied 4 multiplex PCR assays targeting 21 putative virulence genes. We found our EAEC strains to be astonishingly diverse. The only factor associated individually with diarrhea in these analyses was the Shigella SPATE toxin SepA. Recognizing that pathogenicity represents the concerted action of multiple virulence factors, which can sort independently throughout the E. coli population, we assessed combinations of virulence factors using CART analysis. This analysis reinforced the association of sepA with diarrhea, independent of any of the other 20 genes scored. We then performed comprehensive genomic analyses on the sepA-positive strains using CGH against a reference set of E. coli genomes. These studies identified the hemolysin expression-modulating protein RmoA as commonly present in combination with SepA and served to strengthen the association of SepA with clinical illness (Figure 3). Our data demonstrate the importance of strains encoding a combination of virulence factors (here SepA and RmoA), although additional factors may colocalize with these genes.

Among the sepA-negative strains, CART analysis suggested 2 combinations of factors that indicate virulent strains (Figure 1). CGH analysis of strains representative of these combinations (strains C801-09 and C46-10) revealed 2 additional factors: T3SS effector EspY2 and ShiA, the latter being associated with modulation of the inflammatory response. However, screening the complete strain set for the presence of these 2 factors, followed by revised CART analysis, did not suggest that these 2 genes strengthened the association with moderate to severe diarrhea.

The association of the toxin EAST-1 with diarrhea only occurred among strains that lacked the majority of the AggR regulon, suggesting that they may require virulence factors not yet apparent; these may occur predominantly in strains harboring flagellar type H33 (Figure 1). EAST-1-positive strains have previously been implicated in pediatric diarrhea [47], so these strains may warrant continued investigation.

The terms typical EAEC and atypical EAEC have been suggested to refer to EAEC strains harboring or lacking AggR, respectively. Some studies have demonstrated an association of typical EAEC with diarrhea [5, 48]. We did not observe any correlation of AggR regulon genes with moderate to severe illness in this study. It is possible that our focus on moderate to severe diarrhea overlooks mild illness due to EAEC and that true determinants of pathogenicity are not recognized. Alternatively, illness may be obscured by epidemiologic factors, such as previous exposure. Also, we note that our EAEC definition included two AggR-related genes, potentially introducing strain selection bias.

This study is notable for the association of the Shigella virulence factor SepA with clinical illness, an association that persisted when the effects of other pathogens were considered (OR, 5.6; P = .0006; data not shown). SepA was first described by Benjelloun-Touimi et al [20] and is a prominent extracellular protein secreted by S. flexneri strains. SepA is produced during infection [20] and has been shown to confer increased epithelial cell exfoliation from human intestinal explants infected with S. flexneri [49]. We also note that SepA is produced by the Shiga toxin–producing outbreak strain from Germany in 2011. The previously unsuspected role of SepA in EAEC warrants further investigation.

Leveraging a large epidemiologic study and powerful genomic techniques, our study sheds additional light on the complex nature of diarrheagenic E. coli genomes and their association with human disease.