Genomic Comparison of Escherichia coli K1 Strains Isolated from the Cerebrospinal Fluid of Patients with Meningitis

Abstract

Escherichia coli is a major cause of enteric/diarrheal diseases, urinary tract infections, and sepsis. E. coli K1 is the leading gram-negative organism causing neonatal meningitis, but the microbial basis of E. coli K1 meningitis is incompletely understood. Here we employed comparative genomic hybridization to investigate 11 strains of E. coli K1 isolated from the cerebrospinal fluid (CSF) of patients with meningitis. These 11 strains cover the majority of common O serotypes in E. coli K1 isolates from CSF. Our data demonstrated that these 11 strains of E. coli K1 can be categorized into two groups based on their profile for putative virulence factors, lipoproteins, proteases, and outer membrane proteins. Of interest, we showed that some open reading frames (ORFs) encoding the type III secretion system apparatus were found in group 2 strains but not in group 1 strains, while ORFs encoding the general secretory pathway are predominant in group 1 strains. These findings suggest that E. coli K1 strains isolated from CSF can be divided into two groups and these two groups of E. coli K1 may utilize different mechanisms to induce meningitis.

The mortality and morbidity associated with neonatal gram-negative bacillary meningitis remain significant despite advances in antimicrobial chemotherapy and supportive care. The mortality rates have ranged between 5 and 50%, with the majority (30 to 50%) of surviving infants manifesting neurological sequelae (15, 25). A major contributing factor is the incomplete understanding of the pathogenesis of this disease.

Escherichia coli is the most common gram-negative organism that causes meningitis during the neonatal period, and E. coli strains possessing the K1 capsular polysaccharide are predominant (∼80%) among isolates from neonatal E. coli meningitis (26, 44, 58). At present, a few E. coli K1 strains isolated from cerebrospinal fluid (CSF) have been used to study the microbial basis of meningitis, but it is unclear whether the information derived from these E. coli K1 strains is relevant to other E. coli K1 strains isolated from CSF. For example, CNF1 (cytotoxic necrotizing factor 1) and IbeA have been shown to contribute to the pathogenesis of meningitis in strain RS218, i.e., invasion of human brain microvascular endothelial cells (HBMEC) in vitro and traversal of the blood-brain barrier (BBB) in vivo (35, 40), but it is unknown whether CNF1 and IbeA exist and play the same roles in other E. coli K1 strains isolated from CSF. Therefore, a better knowledge of the distribution pattern of known and putative virulence factors will improve our understanding of meningitis caused by E. coli K1.

The distribution of some known E. coli virulence factors has been reported in avian-pathogenic E. coli (APEC) (37), uropathogenic E. coli (21), enteropathogenic E. coli (EPEC), and enterohemorrhagic E. coli (EHEC) (50). For example, the distribution of these known and putative virulence factors has been used to determine phylogenetic relationships among meningitis-causing E. coli isolates (7, 39, 43). However, these results were based on analysis of relatively small numbers of virulence factors. To study the microbial basis of meningitis-causing E. coli K1 on the genome level, we constructed a comprehensive E. coli microarray which covers most of the known E. coli genes to examine the distribution of putative virulence factors and compare the genome contents of 11 representative strains of CSF-derived E. coli K1 belonging to serotypes common in meningitis (i.e., O18, O7, O1, O16, O12, and O45).

Comparative genomic hybridization (CGH) developed based on microarrays has shown differences in the contents of genes and genomic islands in closely related strains or species with different host ranges and virulence characteristics (18, 31, 59). CGH has also been used to study the distribution of specific genes among bacterial strains without the requirement of genome sequences (68). In this study, we employed CGH to investigate 11 representative strains of E. coli K1 isolated from the CSF of patients with meningitis. Based on genome profiles, our results revealed that the 11 strains can be divided into two groups and the two groups may utilize different mechanisms to cause meningitis.

MATERIALS AND METHODS

Bacterial strains.

The strains used in this study are listed in Table 1. Strains S88 and S95 were obtained from E. Bingen (8), and IHE3034 was obtained from T. Korhonen (49). The remaining E. coli strains were described previously (14, 42). Bacteria were grown overnight at 37°C in Luria broth.

TABLE 1.

E. coli K1 strains included in this study^a

Strain	Serotype	Origin	Reference(s)
RS218	O18:K1:H7	North America	32, 40, 64
C5	O18:K1:H7	North America	41
IHE3034	O18:K1:H7	Finland	49
EC10	O7:K1	North America	42
A90	O1:K1:H7	Finland	42
RS167	O16:K1:H6	North America	42
RS168	O1:K1:H−	North America	42
S88	O45:K1	France	8
S95	O45:K1	France	8
E253	O12:K1	North America	14
E334	O12:K1	North America	14

^aAll strains were judged virulent based on their ability to cause bacteremia and meningitis in 5-day-old rats.

Microarray descriptions.

A total of 8,144 50-mer oligonucleotides from E. coli and 95 negative-control oligonucleotides from human and Arabidopsis thaliana were spotted in replicate onto aminosilane slides. The oligonucleotides that are targeting backbone genes in E. coli genomes were derived from a commercially designed oligonucleotide set, which covers every open reading frame (ORF) present in E. coli strain MG1655, and EHEC O157:H7 strains EDL933 and Sakai (all together 6,268 50-mers). The extraintestinal pathogenic E. coli-specific (ExPEC) oligonucleotide probes were derived from uropathogenic E. coli strain CFT073 and meningitis-causing E. coli strain (RS218 and C5) sequences that were different from and/or absent in E. coli backbone (1,876 50-mers). In principle, oligonucleotides from the commercial source were constructed in accumulative fashion, i.e., oligonucleotides targeting every ORF present in MG1655 were constructed and O157:H7-specific probes were added subsequently. Similarly, ExPEC-specific 50-mer oligonucleotide probes were designed and constructed to adapt the existing oligonucleotide set to use in meningitis-causing or uropathogenic E. coli. The majority of conserved backbone ORFs or E. coli strain-specific ORFs were targeted with single probes. ORFs that were shared among different E. coli strains but appeared to be hypervariable were usually targeted with two or more oligonucleotide probes.

DNA isolation, labeling, and hybridization.

E. coli strains were grown overnight in brain heart infusion broth (64) at 37°C with shaking. A 1.5-ml amount of overnight culture was centrifuged for 5 min at 4°C at high speed in a tabletop centrifuge (Eppendorf, Westbury, NY). Supernatants were discarded, and cell pellets were resuspended in an equal volume of TE buffer (10 mM Tris, 1 mM EDTA; pH 8.0). Briefly, genomic DNA was isolated and labeled as described previously with a modification (2, 68). Genomic DNA was digested completely with EcoRV (New England Biolabs, Beverly, MA) and purified by using the QIAquick PCR purification kit (QIAGEN, Valencia, CA). Digested DNA (2 μg) was mixed with 4 μg random hexamer (Invitrogen, Carlsbad, CA) and 5 μl NEB buffer 2 (New England Biolabs, Beverly, MA) in a total volume of 44 μl and then denatured at 95°C for 10 min, followed by snap cooling on ice. Five microliters of deoxynucleoside triphosphate mixture (5 mM dATP, 5 mM dCTP, 5 mM dGTP, 1 mM dTTP, 4 mM aminoallyl-dUTP) and 1 μl Klenow enzyme (New England Biolabs) were added, and the labeling was carried out for 12 h at 37°C. The products were purified by using the QIAquick PCR purification kit (QIAGEN), eluted in 10 μl water, and dissolved with either Cy3 or Cy5 monofunctional N-hydroxysuccinimide ester (Amersham Biosciences, Piscataway, NJ) in 2 μl dimethyl sulfoxide. Two microliters of dye solution and 1 μl of labeling buffer (1 M sodium bicarbonate, pH 9.3) were added to the purified amine-modified DNA. The mixture was incubated in the dark at room temperature for 1 h. The labels were purified by using the QIAquick PCR purification kit (QIAGEN). The purified Cy3- and Cy5-coupled labels were combined and concentrated by centrifugation in a spin filter (Nanosep; molecular weight cutoff, 30,000; Pall, Ann Arbor, MI). The concentrated DNA was resuspended in 50 μl hybridization buffer (Pronto! universal hybridization kit; Corning, Corning, NY). The hybridization mixture was then denatured at 95°C for 2 min. The mixture was applied onto the array under a LifterSlip coverslip (Erie Scientific, Portsmouth, NH). The assembled slide was placed in a hybridization chamber (Corning, Park Acton, MA) and incubated at 42°C for 16 to 18 h. Following hybridization, slides were extensively washed for 1 min in 2× SSC (1× SSC is 0.15 M NaCl plus 0.015 M sodium citrate)-0.1% sodium dodecyl sulfate at 42°C, 5 min in 2× SSC-0.1% sodium dodecyl sulfate at 42°C, 10 min in 1× SSC at room temperature, and 2 min in 0.1× SSC at room temperature twice. Each experiment was run as a competitive hybridization by using Cy3-labeled DNA from one of the 10 strains and Cy5-labeled DNA from strain RS218. Then, experiments were repeated by reversing dyes.

Array scanning and analysis.

Arrays were scanned by using a GenePix 4000B microarray scanner (Axon Instruments, Fremont, CA) with 100% scan power and photomultiplier tube voltage set for each array to minimize saturated pixels and approximately equalize signal intensities in the two channels. Image processing and data extraction were performed by using GenePix Pro 6.0 (Axon Instruments). Spots with high background fluorescence and slide abnormalities were excluded from analysis. Firstly, the medians of spot intensities (MIs) of all slides were calculated. Secondly, the medians of spot intensities of two channels on each slide were normalized to the average of MIs of all slides. ORFs with intensity less than the negative-control value (the median intensities of 95 negative-control oligonucleotides) are considered “absent,” and ORFs with intensities greater than the average of MIs (validated by PCR and in silico analysis) are considered “present.” If both channels could not be determined by the above criteria, these ORFs were considered “divergent.” For remaining ORFs, if channel A showed “absent” and the ratio of intensities (B/A) was less than 3.0, these ORFs in channel B were considered “absent,” and if the ratio was greater than 3.0, then these ORFs were considered “divergent.” If channel A showed “present” and the ratio of intensities (B/A) was less than 0.33, these ORFs in channel B were considered “absent,” and if the ratio was greater than 0.33, then these ORFs were considered “present.”

Hierarchical clustering.

Data imported from GenePix were manipulated and clustered, using established algorithms implemented in the software program Cluster (19). Average linkage clustering with centered correlation was used. TreeView software generated visual representations of clusters (54).

PCR confirmation of microarray data.

Several selected genes were examined by PCR to confirm our microarray data. Primers were designed by Primer 3 (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi). PCR primers are listed in Table 2.

TABLE 2.

Sequences of PCR primers used in this study

Gene	Sequence
	Forward	Reverse
cnf1	GGACTCGAGGTGGTGGTCTA	CGTTGATGGCTCAGGAAAAT
aslA	GCGTGATGTTCATGTCAACC	ATCCGCCAGATCTACAATGC
fyuA	GGCTATACCACCGCTGAAAC	ACCCGGTTACCGTGATACAA
hlyA	CTTAGGAAAGGCAGGCAGTG	ACTTATCGGCAATGGACAGG
hlyE	CCCGCAGCAATAGAATAGGA	CCGCAGATGGAGCATTAGAT
ibeA	AGAAACGGCAAAATCAATGG	ACTCGGTGACCGTACTCCAG
iha	TGCAGGCTGACAGAATCATC	CCGGCTATGAGAAAAAGCTG
iucA	ACCACTCTCTCCCGGCTTAT	GTAGAGGGAACGGGAAGAGG

Triplex PCR for phylogenetic grouping.

The above 11 representative strains of E. coli K1 were also examined for E. coli phylogenetic group by using a combination of two genes (chuA and yjaA) and an anonymous DNA fragment (TspE4C2). PCR was carried out as described previously (11).

RESULTS

Overview of 11 strains.

We selected 11 strains of E. coli K1 which were isolated from the CSF of patients with meningitis. These strains included serotypes that are shown to be common in E. coli meningitis, i.e., O1, O7, O12, O16, and O18. In addition, E. coli K1 isolates belonging to a new O45:K1 group have been included, which have been shown to be predominant in neonates with E. coli meningitis from France (8). After data analysis, the ORFs of these 11 representative strains have been estimated. The total ORF numbers in the genomes of the 11 strains are close to each other, from 4,852 (strain RS167) to 5,147 (strain E253) ORFs (Fig. 1). Some strains may harbor specific ORFs which may not be detectable by our microarray, and thus, the actual ORF numbers may be higher than those shown in Fig. 1. All the microarray data are provided in Table S1 in the supplemental material. To validate our microarray data, 32 ORFs which have been sequenced in either strain RS218 or strain C5 were chosen for in silico analysis (see Table S2 in the supplemental material). The results showed consistency between microarray and in silico analysis, indicating that our microarray data are reliable.

FIG. 1.

Genome of 11 representative strains of E. coli K1. The total ORFs were calculated based on microarray hybridization. Spots were excluded from analysis because of high local background fluorescence and slide abnormalities. The determination of cutoff values has been described in Materials and Methods.

Genetic relationship between strains.

To evaluate the relationship among these 11 representative strains, we performed hierarchical clustering, using the data from comparative genomic hybridization (Fig. 2). Clustering revealed that some E. coli K1 strains with the same O serotypes are closely related to each other. For example, O18:K1:H7 strains RS218, IHE3034, and C5 are closely related to each other and the two O45 strains, S88 and S95, are most closely related to each other. Clustering analysis also linked E. coli strains with different O serotypes. For example, strains EC10 (O7), RS168 (O1), and E253 (O12) were clustered together. Strain RS168, which belongs to O1, was found to be close to strain EC10 (O7), not to strain A90 (O1). Similarly, strains E252 and E334, belonging to O12, were clustered into different groups. Overall, strains EC10, RS168, and E253 represent a distinct cluster and appear to be most distantly related to the RS218 group. Here, based on genome similarity, we named the cluster which contains strains RS218, IHE3034, C5, RS167, A90, S88, S95, and E334 as group 1 and the other strains as group 2.

FIG. 2.

Clustering of 11 representative meningitis-causing E. coli K1 strains based on microarray data. Data imported from GenePix were manipulated and clustered, using established algorithms implemented in the software program Cluster (19). Average linkage clustering with centered correlation was used. TreeView software generated visual representations of clusters (54).

Phylogenetic grouping.

All 11 representative strains of E. coli K1 were classified as to E. coli phylogenetic group by using a combination of two genes (chuA and yjaA) and an anonymous DNA fragment (TspE4C2) (11). The triplex PCR results showed that strains RS218, IHE3034, C5, A90, RS167, E334, S88, and S95 belong to group B2 while strains EC10 and RS168 are in group D and strain E253 is in group A.

Specific virulence factors.

We next examined the distribution of selected E. coli virulence factors printed on a microarray among these 11 representative E. coli K1 strains based on the available information that may be relevant to meningitis (Table 3). The gene ibeA, previously identified from RS218 to contribute to the invasion of HBMEC in vitro and traversal of the blood-brain barrier in vivo (35), was also found to be present in strains C5, RS167, IHE3034, and E334 by microarray analysis. This was confirmed by PCR analysis (Fig. 3A). CNF1 is a 115-kDa protein toxin produced by extraintestinal E. coli strains which include uropathogenic and meningitis-causing E. coli (9). A cnf1 mutant had decreased virulence in a mouse model of ascending urinary tract infection compared to the isogenic cnf1⁺ strain (57). We have previously shown that CNF1 is a virulence factor contributing to E. coli K1 invasion of HBMEC in vitro and traversal of the BBB in vivo (40). cnf1 was found to be present in only 2 of the 11 strains (C5 and RS218) by both microarray and PCR analysis (Fig. 3B). The iron-regulated gene homologue adhesin (iha), an E. coli O157:H7 outer membrane protein, was shown to confer the adherence phenotype upon nonadherent laboratory E. coli and shown to be a virulence factor in uropathogenic E. coli (38). iha encodes a 67-kDa protein in E. coli O157:H7 similar to iron-regulated gene A (IrgA) of Vibrio cholerae (22). Microarray and PCR analysis showed that iha is present in all group 2 strains (EC10, RS168, and E253) and in only one group 1 strain, E334 (Fig. 3C). The yersiniabactin receptor (fyuA) gene was the most highly prevalent iron utilization system in enteroaggregative E. coli and APEC (20, 53). fyuA was found to be uniformly present in all 11 representative strains by microarray analysis, which was confirmed by PCR (Fig. 3D). The iuc (iron uptake chelate) gene locus encodes enzymes necessary for synthesis of the siderophore aerobactin, whereas the iutA (iron uptake transport) gene product represents the TonB-dependent outer membrane receptor for ferric aerobactin. The first step in aerobactin synthesis is hydroxylation of l-lysine by IucD, whereas IucB is responsible for acetylation of N-hydroxylysine. Two N-acetyl-N-hydroxylysine molecules are then attached to the carboxylic groups of citric acid by IucC and probably IucA, resulting in aerobactin (16, 17). iuc was shown to be present in all strains, except for strains C5, RS167, IHE3034, and RS218, and PCR verified the microarray data (Fig. 3F). However, iutA is present in all group 2 strains (EC10, RS168, and E253) and one group 1 strain, E334. Hemolysin production was described in approximately 23% of the E. coli strains associated with neonatal meningitis (43), and our microarray showed the presence of hlyABCD in strains C5, E334, and RS218 but not in others (Table 3). PCR analysis of hlyA also confirmed the microarray result but revealed a different amplification product (about 1 kb compared with the expected 550-bp product) in strain S88 (Fig. 3E2). This 1-kb band from strain S88 was sequenced, and BLAST showed that it is similar to ORF Z1789 (O157 EDL933), which encodes a putative AraC-type regulatory protein. The distribution of hlyE is completely different from that of hlyABCD. Only group 2 strains EC10 and RS168 showed the potential presence of hlyE by microarray analysis (Table 3). However, PCR and sequencing analysis showed that all group 2 strains (EC10, RS168, and E253) harbor hlyE (Fig. 3E1). These differences might be related to a design problem with an oligonucleotide from MWG (High Point, NC).

TABLE 3.

Distribution of selected E. coli virulence factors among 11 E. coli K1 strains^a

ORF	Function	Presence of virulence factor in strainb:
		RS218	C5	IHE3034	S88	S95	A90	RS167	E334	EC10	RS168	E253
cnf1	Cytotoxic necrotizing factor 1	+	+	−	−	−	−	−	−	−	−	−
aslA	Arylsulfatase-like gene	+	+/−	+/−	+/−	+/−	+/−	+	+	+/−	+/−	+
fhuA	Ferrichrome-iron receptor precursor; fhuA	+	+	+	+	+	+	+	+	+/−	+	+
fyuA	Pesticin/yersiniabactin receptor protein	+	+/−	+	+/−	+	+	+	+	+	+	+
hlyABCD	Hemolysin	+	+	−	−	−	−	−	+	−	−	−
hlyE	Hemolysin E	−	−	−	−	−	−	−	−	+/−	+/−	−
ibeA	Invasin IbeA	+	+	+	−	−	−	+	+	−	−	−
ibeB	Invasin IbeB	+	+	+	+	+	+	+	+	+	+	+
iha	Adhesin	−	−	−	−	−	−	−	+	+	+	+
iroN	Ferric uptake protein	+	+	+	+	+	+	+	+	−	−	−
iucABCD	IucABCD protein	−	−	−	+	+	+	−	+	+	+	+
iutA	IutA protein	−	−	−	−	−	−	−	+	+	+	+
malX	PTSc family enzyme IIC	+	+	+	+	+	+	+	+	+	+	+
ompA	Outer membrane protein	+	+	+	+	+	+	+	+	+	+	+
sfa/foc	S fimbria	+	+	+	−	−	−	−	−	−	−	−
sfaG	F1C minor fimbria	+	+	+	+/−	+/−	+/−	+/−	+/−	+	+	+
papA	P fimbria	+	−	−	+	+	+	+/−	+	+	+	+
hecA	Hemolysin/hemagglutinin-like protein	+	+	−	−	−	−	−	−	−	−	−
flu	Antigen 43	+	+	+	+	+	+	+/−	+	−	+	+
hek	Adhesin/virulence factor	+	+	−	+/−	+/−	+/−	−	+/−	+/−	+/−	+/−
traJ	TraJ protein	+	+	−	−	−	−	+	+	−	−	+
neu	K1 capsule	+	+	+	+	+	+	+	+	+	+	+
sitABCD	Sit proteins	+	+	+	+	+	+	+	+	+	+	+

^aOnly microarray data are given, not PCR confirmation results.

^b+, present; −, absent; +/−, divergent.

^cPTS, phosphotransferase system.

FIG. 3.

PCR verification. Selected genes were examined by PCR to confirm microarray data. Primers listed in Table 2 were designed by Primer 3 (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi). PCR programs for each selected gene were optimized by using the same amount of genomic DNA as template. Microarray data for selected genes (P, present; A, absent; D, divergent) are shown at the bottom of each panel. (A) ibeA; (B) cnf1; (C) iha; (D) fyuA; (E1) hlyE; (E2) hlyA; (F) iucA.

Lipoprotein.

Lipoprotein is a major component of the outer membrane of members of the family Enterobacteriaceae. Lipoprotein induces proinflammatory cytokine production in macrophages and lethal shock in lipopolysaccharide-responsive and nonresponsive mice (67). Bacterial lipoproteins comprise a unique set of proteins modified at their amino-terminal cysteines by the addition of N-acyl and S-diacyl glyceryl groups (67). In E. coli, this lipid serves to anchor these proteins to the inner or outer membrane so that they can function at the lipid-aqueous interface. These proteins can be identified by the presence of a leader with a common consensus sequence (10). The leader is typically between 15 and 40 amino acid residues in length and has at least one arginine or lysine in the first seven residues. The leader is cleaved by signal peptidase II on the amino-terminal side of the cysteine residue, which is then enzymatically modified (67). Our data revealed that about 100 lipoproteins are present in E. coli K-12 (46). Fifty-four ORFs which encode lipoproteins or putative lipoproteins were printed on our microarray, and about half of them (20 of 54) are shown to be present in all 11 representative strains of E. coli K1 (Table 4). For example, cutF (nlpE), encoding an outer membrane lipoprotein, is involved in copper transport in E. coli, and a mutation of cutF results in an increased copper sensitivity in E. coli (28). cutF is shown to be present in all 11 E. coli K1 strains. rlpA and rlpB, encoding two lipoproteins, are located in the leuS-dacA region (15 min) on the E. coli chromosome (63). A truncated rlpA gene in E. coli was able to rescue a conditionally lethal mutation in the prc gene (involved in C-terminal processing of penicillin-binding protein 3) (6). The prc mutants are sensitive to heat and osmotic stress (29). rlpA is shown to be present in all 11 E. coli K1 strains (Table 4). LolCDE, an ATP-binding cassette transporter, releases outer membrane-specific lipoproteins from the inner membrane to form a complex between the released lipoproteins and the periplasmic molecular chaperone LolA (51). LolA was shown to release other outer membrane lipoproteins such as Pal, NlpB, Slp, and RlpA, whereas inner membrane lipoproteins AcrA and NlpA were not released even in the presence of LolA (69), indicating that LolA plays a critical role in the sorting of lipoproteins. lolA is shown to be present in all 11 E. coli K1 strains (Table 4). An outer membrane lipoprotein encoded by nlpI may be important for the process of cell division in E. coli K-12 and has been shown to be involved in adherence to and invasion of intestinal epithelial cells by E. coli strain LF82 (5, 52). This nlpI gene is also found to be present in all 11 E. coli K1 strains (Table 4). Of interest, aec24, encoding a hypothetical lipoprotein, is shown to be present in group 1 strains but absent in group 2 strains.

TABLE 4.

Distribution of lipoproteins among meningitis-causing E. coli K1 strains^a

ORF	Function	Presence of ORF in strainb:
		RS218	C5	IHE3034	S88	S95	A90	RS167	E334	EC10	RS168	E253
aec24	Hypothetical lipoprotein	+	+	+	+	+	+	+	+	−	−	+/−
apbE	Putative thiamine biosynthesis lipoprotein	+	+	+	+	+	+	+	+	+	+	+
blc	Outer membrane lipoprotein (lipocalin)	+	+	+	+	+	+	+	+	+	+	+
cutF	Outer membrane lipoprotein	+	+	+	+	+	+	+	+	+	+	+
hmsF	Putative lipoprotein	+	+	+	+	+	+	+	+	+	−	+
lgt	Phosphatidylglycerol-prolipoprotein diacylglyceryl transferase	+	+	+	+	+	+	+	+	+	+	+
lnt	Apolipoprotein N-acyltransferase	+	+	+	+	+	+	+	+	+	+	+
lolA	Outer membrane lipoproteins	+	+	+	+	+	+	+	+	+	+	+
lpp	Murein lipoprotein	+	+	+	+	+	+	+	+	+	+	+/−
nlpB	Lipoprotein 34	+	+	+	+	+	+	+	+	+	+	+
nlpC	Lipoprotein	+	+	+	+	+	+	+	+	+	+	+
nlpD	Lipoprotein	+	+	+	+	+	+	+	+	+	+	+
nlpI	Lipoprotein	+	+	+	+	+	+	+	+	+	+	+
osmB	Lipoprotein, osmotically inducible	+	+	+	+	+	+	+	+	+	+	+
rlpAB	Minor lipoprotein	+	+	+	+	+	+	+	+	+	+	+
slyB	Putative outer membrane lipoprotein	+	+	+	+	+	+	+	+	+	+	+
spr	Suppresses thermosensitivity of prc mutants at low osmolarity	+	+	+	+	+	+	+	+	+	+	+
vacJ	Lipoprotein precursor	+	+	+	+	+	+	+	+	+	+	+
yafL	Putative lipoprotein	+	+	+	+	+	+	+	+	+	+	+
yajG	Putative lipoprotein	+	+	+/−	+/−	+/−	+	+	+/−	+/−	+/−	+/−
yehR	Putative lipoprotein	+	+	+	+	+	+	+	+	+	+	+
yfiO	Putative lipoprotein with tetratricopeptide repeat domain	+	+	+	+	+	+	+	+	+	+	+
yqhH	Putative outer membrane lipoprotein	+	+	+	+	+	+	+	+	+	+	+
yhiU	Multidrug resistance protein	+/−	+	+	+	+	+	+	+/−	+/−	+/−	+/−

^aOnly microarray data are given, not PCR confirmation results.

^b+, present; −, absent; +/−, divergent.

Protease.

Proteases were shown to be present in many pathogenic bacteria, where they play critical functions related to colonization and evasion of host immune defenses or tissue damage during infection (47, 65). Protease functions as an endopeptidase, signal peptidase, or aminopeptidase. Proteolysis in E. coli serves to rid the cell of abnormal and misfolded proteins and to limit the time and amounts of availability of critical regulatory proteins. Most intracellular proteolysis is initiated by energy-dependent proteases, including Lon, ClpXP, and HflB (24). Oligonucleotides which represent 137 proteases were printed on our microarray. Approximately 70% of them (102 of 137) are shown to be present in all 11 representative strains. For the remaining proteases, some ORFs exhibit nonrandom distribution among representative strains. Of interest, vat.2, encoding hemoglobin protease (55), is present in all strains in group 1 but absent from group 2 strains (Table 5). Similarly, yeaZ, encoding a putative glycoprotein endopeptidase, is absent only in group 2 strains EC10 and RS168 (Table 5). The gene sohA, encoding a putative protease, allows temperature-sensitive htrA mutant E. coli to grow at 42°C (4). The microarray showed that sohA is present in all group 2 strains and one group 1 strain, RS167 (Table 5).

TABLE 5.

Distribution of proteases among meningitis-causing E. coli K1 strains^a

ORF	Function(s)	Presence of ORF in strainb:
		RS218	C5	IHE3034	S88	S95	A90	RS167	E334	EC10	RS168	E253
ampE	Putative transmembrane protein, putative protease	+	+	+	+	+	+	+	+	+	+	+
clpA	Serine protease	+	+	+	+	+	+	+	+	+	+	+
degQ	Serine endoprotease	+	+	+	+	+	+	+	+	+	+	+
degS	Periplasmic serine endoprotease	+	+	+	+	+	+	+	+	+	+	+
ecfE	Membrane-associated protease	+	+	+	+	+	+	+	+	+	+	+
envC	Periplasmic protease	+	+	+	+	+	+	+	+	+	+	+
gcp	Putative O-sialoglycoprotein endopeptidase	+	+	+	+	+	+	+	+	+	+	+
hflB	ATP-dependent zinc metalloprotease	+	+	+	+	+	+	+	+	+	+	+
hslU	ATPase component of the HslUV protease	+	+	+	+	+	+	+	+	+	+	+
htrA	Periplasmic serine protease	+	+	+	+	+	+	+	+	+	+	+
hycI	Protease involved in processing C-terminal end of HycE	+	+	+	+	+	+	+	+	+	+	+
iap	Aminopeptidase in alkaline phosphatase isozyme conversion	+	+	+	+	+	+	+	+	+	+	+
lepB	Signal peptidase I	+/−	+/−	+/−	+/−	+/−	−	−	+/−	+/−	+/−	−
lexA	Transcriptional repressor for SOS response (signal peptidase of LexA family)	+	+	+	+	+	+	+	+	+	+	+
lon	DNA-binding ATP-dependent protease La	+	+	+	+	+	+	+	+	−	−	+
lspA	Prolipoprotein signal peptidase (SPase II)	+	+	+	+	+	+	+	+	+	+	+
map	Methionine aminopeptidase	+	+	+	+	+	+	+	+	+	+	+
mepA	Murein dd-endopeptidase	+	+	+	+	+	+	+	+	+	+	+
ompX	Outer membrane protease, receptor for phage OX2	+	+	+	+	+	+	+	+	+	+	+
pbpG	d-Alanyl-d-alanine endopeptidase	+	+	+	+	+	+	+	+	+	+	+
pepA	Aminopeptidase A	+	+	+	+	+	+	+	+	+	+	+
prc	Carboxy-terminal protease	+	+	+	+/−	+	+	+	+	+	+	+
ptrA	Protease III	+	+	+	+	+	+	+	+	+	+	+
ptrB	Protease II	+	+	+	+	+	+	+	+	+	+	+
radA	Putative ATP-dependent protease	+	+	+	+	+	+	+	+	+	+	+
recA	DNA strand-exchange and recombination protein with protease and nuclease activity	+	+	+	+	+	+	+	+	−	+	+
rzpR	Putative prophage lambda endopeptidase	+	+	+	+	+	+	+	+	+	+	+
sohA	Putative protease; htrA suppressor protein	−	−	−	−	−	−	+	−	+	+	+
sppA	Protease IV, a signal peptide peptidase	+	+/−	+/−	+/−	+/−	+/−	+/−	+/−	+	+	+/−
sspB	Stringent starvation protein B	+	+	+	+	+	+	+	+	+	+	+
tesA	Protease I	+	+	+	+	+	+	+	+	+	+	+
vat.2	Hemoglobin protease	+	+	+	+	+	+	+	+	−	−	−
yeaZ	Putative glycoprotein endopeptidase	+	+	+	+	+	+	+	+	−	−	+
yegQ	Putative protease	+	+	+	+	+	+	+	+	+	+	+
yghQ	Putative serine protease	+	+	+	+	+	+	+	+	+	+	+
yhbU	Putative protease	+	+	+	+	+	+	+	+	+	+	+
yhbV	Putative protease	+	+	+	+	+	+	+	+	+	+	+

^aOnly microarray data are given, not PCR confirmation results.

^b+, present; −, absent; +/−, divergent.

Outer membrane protein.

There is increasing evidence that outer membrane proteins contribute to adhesion and invasion of the host cells in several gram-negative organisms. For example, outer membrane protein A (OmpA), a highly conserved protein in E. coli, was shown to contribute to E. coli K1 association with HBMEC and penetration into the central nervous system (66). One hundred forty-three oligonucleotides which represent outer membrane proteins and outer membrane protein-related proteins were printed onto our microarray. Microarray data showed that many outer membrane proteins were absent in group 1 (strains RS218, IHE3034, C5, RS167, A90, S88, S95, and E334) but present in group 2 (strains EC10, E253, and RS168) (Table 6). For example, ycbS, encoding a putative outer membrane protein; yejO, encoding a putative outer membrane protein with a pectin lyase-like domain; and yiaT, encoding a putative outer membrane protein, are present in all group 2 strains but absent in group 1 strains. Similarly, sfmD, encoding a putative outer membrane protein, is present in all group 2 strains but absent in group 1 strains. In contrast, some other outer membrane proteins are present only in group 1 strains. For example, bglH, encoding a carbohydrate-specific outer membrane porin, is one gene of the bgl operon, which is responsible for uptake and fermentation of β-glucosides in E. coli (1). Our results showed that bglH is present in all group 1 strains but absent in group 2 strains except for strain E253 (Table 6).

TABLE 6.

Distribution of outer membrane proteins among meningitis-causing E. coli K1 strains^a

ORF	Function	Presence of ORF in strainb:
		RS218	C5	IHE3034	S88	S95	A90	RS167	E334	EC10	RS168	E253
bglH	Carbohydrate-specific outer membrane porin in cryptic operon	+	+	+	+	+	+	+	+	−	−	+
btuB	Cobalamin transport, receptor for E colicins	+	+	+	+	+	+	+	+	+	+	+
cirA	Receptor for colicin I	+	+	+	+	+	+	+	+	+	+	+
eaeH	Putative outer membrane protein important for attachment to cell	+	+/−	+/−	+/−	+/−	+/−	+/−	+/−	+	+	+
fadL	Transport of long-chain fatty acids	+	+	+	+	+	+	+	+	+	+	+
fecA	Receptor for ferric citrate	−	−	−	+	+	+	−	−	+	+	+
fepA	Receptor for ferric enterobactin and colicins B and D	+	+	+	+	+	+	+	+	+	+	+
hmsH	Putative outer membrane protein	+	+	+	+	+	+	+	+	+	−	+
nfrA	Bacteriophage N4 receptor	+/−	−	−	−	−	−	+	+	+	+	+
nmpC	NmpC precursor	+	+	+	+	+	+	+	+	+	+	+
ompA	Outer membrane protein 3a	+	+	+	+	+	+	+	+	+	+	+
ompC	Outer membrane pore protein 1b	+	+	+	+	+	+	+	+	+	+	+
ompF	Outer membrane pore protein 1a	+	+	+	+	+	+	+	+	+	+	+
ompG	Outer membrane pore protein	+	+	+	+	+	+	+	+	+	+	+
ompN	Outer membrane pore protein N	+	+	+	+	+	+	+	+	+	+	+
ompT	Protease VII, outer membrane protein 3b (a), putative porin	+	+	+	+	+	+	+	+	+	+	−
ompW	Colicin S4 receptor	+	+	+	+	+	+	+	+	+	+	+
pldA	Outer membrane phospholipase A	+	+	+	+	+	+	+	+	+	+	+
sfmD	Putative outer membrane protein	−	−	−	−	−	−	−	−	+	+	+
slp	Outer membrane protein, induced after carbon starvation	+	+/−	+	+	+	+	+	+/−	+	+	+
yaeT	Putative outer membrane antigen	+/−	+/−	+/−	+/−	+/−	+/−	+/−	+	+/−	+/−	+
yaiV	Putative outer membrane protein with cyclic AMP-binding domain	+	+	+	+	+	+	+	+	+	+	+
ybiL	Putative ferrisiderophore receptor	+	+	+	+	+	+	+	+	+	+	+
ycbS	Putative outer membrane protein	−	−	−	−	−	−	−	−	+	+	+
ydeT	Putative outer membrane protein	−	+	+	−	+	+	+	+/−	+	+	−
yehB	Putative outer membrane protein	+	+/−	+/−	+/−	+/−	+/−	+/−	+/−	+/−	+/−	+
yejO	Putative outer membrane protein with pectin lyase-like domain	−	−	−	−	−	−	−	+/−	+	+	+
yfiB	Putative outer membrane protein	+	+	+	+	+	+	+	+	+	+	+
yiaD	Putative outer membrane protein	+	+	+	+	+	+	+	+	+	+	+
yiaT	Putative outer membrane protein	−	−	−	−	−	−	−	+/−	+	+	+
yjcP	Putative outer membrane protein	+	+	+	+	+	+	+	+	+	+	+
yjhA	Conserved hypothetical protein, outer membrane domain	+	+	+	+	+	+	+	+	+	+	+
yjiK	Putative outer membrane protein	+	+	+	+/−	+/−	+	+	+	+	+	+
yncD	Putative outer membrane porin protein	+	+	+	+	+	+	+	+	+	+	+
yohG	Putative outer membrane protein	+	+	+	+	+	+	+	+	+	+	+
ypjA	Putative outer membrane protein with pectin lyase-like domain	+	+	+	+	+	+	+	+	+	+	+
ytfM	Putative outer membrane protein	+	+	+	+	+	+	+	+	+	+	+

^aOnly microarray data are given, not PCR confirmation results.

^b+, present; −, absent; +/−, divergent.

Secretion system.

General secretory pathway (GSP) systems, which can export the majority of bacterial exoenzymes and toxins, have been identified in many gram-negative bacteria, such as Haemophilus influenzae, V. cholerae, uropathogenic E. coli, and Helicobacter pylori (56, 60). Secretion by the GSP has been shown to play an important role in bacterial pathogenesis. For example, many virulence factors including extracellular toxins, pili, curli, adhesins, invasins, and proteases are exported to the extracellular environment via the GSP (62). Our microarray data showed the presence of the GSP operon (13 ORFs) in all group 1 strains but not in group 2 strains except for strain E253 (Table 7).

TABLE 7.

Distribution of type III secretion system and GSP genes among meningitis-causing E. coli K1 strains

Gene	Presence of gene in straina:
	RS218	C5	IHE3034	S88	S95	A90	RS167	E334	EC10	RS168	E253
sepb	−	−	−	−	−	−	−	−	−	−	−
escb	−	−	−	−	−	−	−	−	−	−	−
eprb	−	−	−	−	−	−	−	−	+	+	+
epab	−	−	−	−	−	−	−	−	+	+	+
eivb	−	−	−	−	−	−	−	−	+	−	−
GSPc,d	+	+	+	+	+	+	+	+	−	−	+

^a+, present; −, absent.

^bPCR confirmation results.

^cIncluding all 13 ORFs of the GSP operon.

^dOnly microarray data are given, not PCR confirmation results.

Type III secretion systems (TTSSs) allow Yersinia spp., Salmonella spp., Shigella spp., Pseudomonas aeruginosa, and enteropathogenic E. coli to adhere to the surface of eukaryotic cells by injecting bacterial proteins across the two bacterial membranes and the eukaryotic cell membrane to destroy or subvert the target cell (12, 45). These systems consist of a secretion apparatus, made of approximately 25 proteins, and an array of proteins released by this apparatus. Some of these released proteins are “effectors,” which are delivered into the cytosol of the target cell, whereas the others are “translocators,” which help the effectors cross the membrane of the eukaryotic cell. Most of the effectors act on the cytoskeleton or on intracellular signaling cascades (13). A protein injected by the enteropathogenic E. coli serves as a membrane receptor for the docking of the bacterium itself at the surface of the cell (23). Interestingly, by using comparative genomic hybridization, we found that all group 2 strains harbor the type III secretion system locus (Table 7). This locus contains ORFs whose amino acid sequences show high degrees of similarity with those of the proteins that make up the type III secretion apparatus of the inv-spa-prg locus on a Salmonella SPI-1 pathogenicity island (61). This locus was designated ETT2 (E. coli type III secretion 2) and consisted of the epr, epa, and eiv genes. ETT2 was found in enteropathogenic E. coli and some non-O157 Shiga toxin-producing E. coli strains, but most of them contained a truncated portion of ETT2 (30). Strains RS168 and E253 harbor a part of ETT2 (epr and epa operon) but lack eiv genes. However, strain EC10 was found to harbor all the genes needed to encode type III secretion apparatus proteins. This is the first demonstration that the type III secretion system is found to be present in E. coli K1 strains isolated from CSF.

DISCUSSION

E. coli is the most common gram-negative organism that causes meningitis during the neonatal period and presents a major burden for the public health system. Further improvement in the treatment of neonatal meningitis will require new therapeutic approaches and a complete understanding of the underlying pathogenic mechanisms. Several microbial factors such as the K1 capsule, OmpA, type I fimbriae, Ibe proteins, AslA, and CNF1 were shown to be involved in E. coli meningitis and have been characterized (3, 32, 34, 40, 64, 66). However, most of the meningitis-causing E. coli K1 isolates do not harbor all of those known microbial determinants. For example, we have previously reported that CNF1 is a virulence factor contributing to E. coli K1 invasion of HBMEC in vitro and traversal of the BBB in vivo (40), but cnf1 was found to be present in only two (strains C5 and RS218) of 11 representative meningitis-causing E. coli K1 strains (Fig. 3B). Thus, it is likely that the remaining strains utilize other microbial determinants to invade HBMEC and traverse the BBB. At present, the distribution of potential microbial virulence factors in meningitis-causing E. coli K1 strains is poorly understood.

It has been known that some virulence factors can be encoded by mobile genetic islands, which are capable of lateral gene transfer. Besides the core genetic contents to maintain primary metabolism, some strains may acquire different extra genetic elements which can help them adapt to specific niches. Meningitis-causing E. coli K1 strains have been shown to obtain some strain-specific genetic elements (33). To understand the roles of these genetic elements in the pathogenesis of meningitis, determining the distribution of known and putative virulence factors should be an important first step. CGH is a powerful tool for this purpose. For CGH, the more oligonucleotide probes that are printed on the slide, the more data that can be obtained. We constructed a comprehensive E. coli microarray harboring 8,144 ORFs. CGH data showed that the distribution of known and putative virulence factors among 11 representative strains is not random. Some of them are present in particular strains and absent in other strains or vice versa. These findings provided novel insights into the evolution and relationship of meningitis-associated E. coli K1 strains and suggested potential targets for prevention and therapy of E. coli meningitis.

Clustering data revealed that these 11 representative meningitis-causing E. coli K1 strains can be divided into two different groups (Fig. 2). Strains RS218, IHE3034, C5, A90, RS167, E334, S88, and S95 are grouped together and named group 1. The other group includes strains EC10, RS168, and E253 and is named group 2. This finding based on our comprehensive microarray analysis revealed that E. coli K1 strains with different O serotypes may belong to the same group (e.g., O18, O1, O16, O12, and O45 for group 1 versus O1, O2, and O12 for group 2), while E. coli K1 strains with the same O serotype may belong to different groups (e.g., O1 and O12). We also carried out triplex PCR to determine the phylogenetic relationship of these 11 representative meningitis-causing E. coli K1 strains (11). Results showed that strains RS218, IHE3034, C5, A90, RS167, E334, S88, and S95 belong to group B2 (73%) while strains EC10 and RS168 are in group D (18%) and strain E253 is in group A (9%). The proportions of each phylogenetic group in our E. coli strains matched those of the previous study (7), indicating that our E. coli K1 strains used in this study are indeed representative of the meningitis-causing E. coli K1 strains. Moreover, there is a similarity between our genome-based grouping and the previously defined phylogenetic grouping. For example, phylogenetic group B2, which is predominant in CSF isolates of E. coli K1, belongs to our group 1, while less common phylogenetic groups A and D belong to our group 2.

Most remarkably, we showed that 3 of the 11 representative strains harbor some genes from ETT2 (Table 7) and these three strains belong to group 2, while none of the group 1 strains harbor ETT2. The function and mechanism of the TTSS have been well documented in Yersinia spp., Salmonella spp., Shigella spp., and P. aeruginosa (12, 45). The TTSS is found exclusively among gram-negative bacteria and is responsible for the transport of proteins across the inner bacterial membrane, the peptidoglycan layer, and the outer bacterial membrane, as well as across host cell barriers such as the plasma membrane and in some instances such as the plant cell wall, into the host cell interior (12, 45). The TTSS apparatus used to deliver these effectors is conserved and shows functional complementarity for secretion and translocation. ETT2 was also found in EPEC and some non-O157 Shiga toxin-producing E. coli strains, but most of them contained a truncated portion of ETT2 (30).

The existence of a degenerate ETT2 gene cluster was found in septicemic E. coli strain 789 (36). Sequence analysis of the ETT2 genes showed premature stop codons in eprI and eprJ, encoding the needle structure, and deletion of the invG gene, which encodes a conserved component of the outer membrane ring. This ETT2 lacks the gene (eivC) for the cytoplasmic ATPase that energizes secretion and some other conserved components of the TTSS (e.g., epaS). However, a deletion mutant of genes coding for the putative inner membrane ring of the secretion complex showed significantly reduced virulence in a 1-day-old chick model, even though the mutation does not seem to affect the secreted proteome (36). Of interest, strain EC10 from group 2, which was isolated from the CSF of a neonate with meningitis, was found to harbor all the genes needed to encode type III secretion apparatus proteins compared to the above-mentioned septicemic E. coli strain 789. We speculate that strain EC10 may utilize the TTSS to invade and subvert the signal transduction pathway in HBMEC to induce meningitis. However, our microarray data failed to reveal the presence of effectors and translocators in group 2 strains such as sep, esc, and tir which were identified in the locus for enterocyte effacement (LEE) (48). Thus, group 2 strains may employ some other effectors and translocators to utilize the TTSS. For example, effectors which are non-LEE-encoded type III translocated virulence factors have been identified in EHEC (27). Studies are in progress to determine the potential contribution of ETT2 to the pathogenesis of E. coli meningitis caused by strain EC10.

The general secretory pathway is used by many gram-negative bacteria to transport exoproteins from the periplasm to the outside milieu (56). We showed that all group 1 strains harbor the general secretory pathway system, but group 2 strains do not have such a system, except for strain E253 (Table 7). It is tempting to speculate that group 1 strains utilize the general secretory pathway system during infection. We searched all the sequenced E. coli genomes and found that uropathogenic E. coli strain CFT073 also harbors the GSP system instead of the TTSS. However, EHEC strains DEL933 and Sakai (O157) do not have GSP systems and harbor the typical TTSS. We speculate that either GSP or TTSS may have been maintained during evolution because of their similar function. Additional studies are needed to clarify this speculation.

In conclusion, we carried out comparative genomic hybridization in representative strains of meningitis-causing E. coli K1 and provide evidence that these E. coli K1 strains can be divided into two groups. Based on the distribution of putative virulence factors, lipoproteins, proteases, outer membrane proteins, and secretion systems, we speculate that these two groups of E. coli K1 strains may cause meningitis by using different mechanisms. This speculation is also supported by our demonstration of the TTSS mainly in group 2 strains of E. coli K1. Additional studies with a larger collection of E. coli K1 strains isolated from CSF are needed to determine whether our proposed grouping can be applied to all the CSF isolates of E. coli K1.