Rapid Determination of Escherichia coli O157:H7 Lineage Types and Molecular Subtypes by Using Comparative Genomic Fingerprinting

Abstract

In this study, variably absent or present (VAP) regions discovered through comparative genomics experiments were targeted for the development of a rapid, PCR-based method to subtype and fingerprint Escherichia coli O157:H7. Forty-four VAP loci were analyzed for discriminatory power among 79 E. coli O157:H7 strains of 13 phage types (PT). Twenty-three loci were found to maximize resolution among strains, generating 54 separate fingerprints, each of which contained strains of unique PT. Strains from the three previously identified major E. coli O157:H7 lineages, LSPA6-LI, LSPA6-LI/II, and LSPA6-LII, formed distinct branches on a dendrogram obtained by hierarchical clustering of comparative genomic fingerprinting (CGF) data. By contrast, pulsed-field gel electrophoresis (PFGE) typing generated 52 XbaI digestion profiles that were not unique to PT and did not cluster according to O157:H7 lineage. Our analysis identified a subpopulation comprised of 25 strains from a closed herd of cattle, all of which were of PT87 and formed a cluster distinct from all other E. coli O157:H7 strains examined. CGF found five related but unique fingerprints among the highly clonal herd strains, with two dominant subtypes characterized by a shift from the presence of locus fprn33 to its absence. CGF had equal resolution to PFGE typing but with greater specificity, generating fingerprints that were unique among phenotypically related E. coli O157:H7 lineages and PT. As a comparative genomics typing method that is amenable for use in high-throughput platforms, CGF may be a valuable tool in outbreak investigations and strain characterization.

Escherichia coli O157:H7 is one of the most significant human pathogens, responsible for outbreaks worldwide of food- and waterborne illnesses that range from diarrhea to the hemolytic uremic syndrome (15, 45, 46, 50). Two major lineages of this pathogen that differ in both genotype and phenotype have been identified using octamer-based genome scanning, with strains of lineage I more frequently linked to human illness (26). The lineage-specific polymorphism assay (LSPA6) is based on polymorphisms in six genetic loci (53) and was developed as a means to quickly identify strain lineage in a single PCR. As food production and distribution becomes increasingly centralized, outbreaks of infection associated with this pathogen from fresh produce and meats have an increased risk of being widely disseminated and infecting multiple populations from a single source, making the ability to quickly and effectively determine epidemiologically related strains of paramount importance to public health (10, 12).

Bacteriophage typing was among the first subtyping schemes developed to characterize E. coli O157:H7 isolates (1). While the method is still commonly used to screen E. coli O157:H7 isolates during outbreak investigations, the common occurrence of certain phage types (PTs), along with the low number of PTs overall, limits its resolution and frequently results in the classification of outbreak- and non-outbreak-related E. coli O157:H7 isolates within the same PT.

DNA-based subtyping methods, which rely on the unique genetic fingerprint of each bacterial strain, allow a higher level of discrimination among E. coli O157:H7 isolates than is possible with bacteriophage typing (51, 54). Among the DNA-based methods commonly used to genotype bacteria are multilocus sequence typing (MLST), amplified fragment length polymorphism (AFLP) analysis, ribotyping, PCR-restriction fragment length polymorphism (PCR-RFLP) analysis, multilocus variable-number tandem repeat analysis (MLVA), and the current standard, pulsed-field gel electrophoresis (PFGE).

MLST has been shown to be of inferior resolution to other genotyping methods such as PFGE because the small number of housekeeping genes analyzed by MLST do not contain sufficient variation to be useful as the sole epidemiological typing method (5, 33). Some groups have suggested AFLP as an alternative to PFGE (21, 56), and although AFLP could discriminate O157:H7 strains from non-O157:H7 strains, it was found to lack within-serogroup resolution compared to PFGE (18, 19). Likewise, ribotyping alone does not offer the discriminatory power needed to effectively fingerprint E. coli O157:H7 strains, although it has been successfully used in conjunction with PFGE to offer additional discrimination (3, 38). PCR-RFLP based on virulence regions (such as the regulatory region of Stx phage) has been used to unambiguously determine clonality among E. coli O157:H7 strains but lacks the resolution to differentiate among strains that are temporally or geographically distant (42, 43). The analysis of tandem duplication in the genome with MLVA has been useful in characterizing clonal organisms such as E. coli O157:H7 (20) with greater discriminatory power than PFGE (32, 36), although its usefulness in higher-level analyses, such as lineage typing, has not been determined. Unfortunately, MLVA requires the PCR products be sequenced or analyzed using capillary electrophoresis to accurately determine the number of tandem repeats (20), although estimation of size based on traditional agarose gels has also been used (24).

PFGE is currently considered the “gold standard” of DNA-based subtyping methods and has been shown to discriminate between O157:H7 and non-O157:H7 isolates, among isolates from different geographic regions, between outbreak- and non-outbreak-related isolates, and to identify outbreak sources (4, 6, 7). Although PFGE has been an extremely powerful tool in E. coli O157:H7 genotyping, it requires complex computer-dependent graphical analysis using dedicated software and does not always produce consistent results, particularly when the analyses are conducted in different laboratories (35). In response to these challenges, the U.S. Centers for Disease Control and Prevention has attempted to improve the reproducibility of PFGE by developing PulseNet as a means of interlaboratory standardization and sharing of PFGE data.

While all of these subtyping methods are useful, none produces a stand-alone method for subtyping E. coli O157:H7 isolates and many indicate seemingly contradictory levels of diversity. Genomic variation in E. coli has been found in large part to be due to genomic islands and phage-related areas that often contain virulence-associated genes, which arise through horizontal gene transfer (17). Examples include the phage-encoded Shiga toxins (31, 37), the locus of enterocyte effacement pathogenicity island (29), and other genomic islands containing virulence-related factors, such as fimbriae (41) and iron uptake systems (9, 34). Recently a “seropathotype” designation based on the association between the occurrence and severity of human disease and certain serogroups has been proposed. It was found that those strains more virulent to humans also contain specific genes within the virulence-associated genomic island OI#122 (23). Genotyping methods, such as octamer-based genome scanning, have also shown that certain lineages of E. coli O157:H7 appear to be more commonly associated with human disease than others, suggesting differences in phenotype and possibly virulence within this serotype (26, 57). As all molecular subtyping methods are based on heterogeneity among DNA sequences, the ideal scheme would be based on comparative genomic analysis of the entire genome sequences of all isolates. However, the practical considerations of whole-genome sequencing do not as of yet make the analysis of multiple E. coli O157:H7 isolates feasible for rapid typing of outbreak strains.

Our previous work with microarray-based comparative genomics and subtractive hybridization led to the identification of genetic regions that are variably absent or present (VAPs) among E. coli O157:H7 strains (48, 55). In this study, VAPs were used to create a PCR-based comparative genomic fingerprinting (CGF) method capable of identifying taxonomically unique subtypes and epidemiologically related strains.

MATERIALS AND METHODS

Selection of loci.

Thirty-four E. coli O157:H7 strains which had previously been characterized by microarray-based comparative genome hybridization (CGH) were analyzed for genes that were variably absent or present (55). The microarray data contained 1,751 VAP loci, and those with binary log ratio distributions were selected for further examination, as they were expected to increase the probability that a locus was present or absent rather than heterogeneous in sequence. From the set of loci consisting of binary log ratio distributions, those that gave the greatest strain discrimination were selected preferentially over those that had very little discriminatory power. Thus, if the presence and absence of two loci were identical among all strains, only one was kept. No knowledge of lineage or other phylogenetic relationship was used in the selection of loci. A set of 34 loci were chosen that were distributed across the entire E. coli genome and offered the greatest combined resolution among the strains, without reference to any other typing method.

The VAP loci identified by our CGH work consisted of only O157:H7 lineage I open reading frames (55). In order to remove potential bias in subsequent analyses, 10 lineage II-specific VAP loci identified by subtractive hybridization experiments (48) were added to the 34 identified through microarray-based CGH, and this final set of 44 were used in testing. Although 10 loci were lineage II specific, no effort was made to preferentially keep them; all loci were treated equally, and those that offered the greatest discrimination were kept.

Isolation of DNA.

The E. coli strains used in this study were obtained from a variety of human, bovine, and environmental sources (Table 1). Growth of the bacteria was carried out in 10 ml of brain heart infusion broth, at 37°C for ∼16 h in a shaking incubator (200 rpm). The cultures were centrifuged at 8,000 rpm for 10 min, and the bacterial pellet was dissolved in 15 ml of 10 mM NaCl, 20 mM Tris-HCl (pH 8.0), 1 mM EDTA, 100 μg/ml proteinase K, and 0.5% sodium dodecyl sulfate. This suspension was incubated at 50°C for 2 h and extracted with an equal volume of phenol-chloroform-isoamyl alcohol (25:24:1). Following centrifugation for 10 min at 8,000 rpm, the upper phase was removed and precipitated by adding 0.1 volume of 3 M NaO-acetate (pH 5.2) and 2 volumes of 99% ethanol. The DNA precipitate was then spooled out of solution using a sterile glass rod, washed with 70% ethanol, and dissolved in 3 ml of Tris-EDTA (10 mM Tris-HCl, 1 mM EDTA, pH 8.0) buffer.

TABLE 1.

The E. coli O157:H7 strains studied and their source, date of isolation, phage type, LSPA6 lineage, XbaI PFGE digestion pattern, and CGF profile

Name	Source	Date isolated	Phage type	LSPA6 lineage	PFGE XbaI	CGF profile
R1388	Human	1996	1	I/II	X01.0091	CGF55
59243	Human	10-Jun-95	2	I/II	X01.0102	CGF53
71074	Human	19-Jul-95	2	I/II	X01.0110	CGF52
09601Fe046.1	Bovine	NAa	2	I/II	X01.0274	CGF52
Zap0046	Bovine	NA	2	I/II	X01.0209	CGF54
S2302-1	Human	1996	14	I	X01.0002	CGF18
S3722	Human	1997	14	I	X01.0156	CGF40
S2628	Human	1998	14	I	X01.0002	CGF26
97701	Human	1995	14	I	X01.0143	CGF41
F5	Bovine	8-Aug-95	14	I	X01.0109	CGF42
F1299	Bovine	26-Aug-96	14	I	X01.0528	CGF34
TS-97	Bovine	11-Oct-01	14	I	X01.0109	CGF19
LRH73	Human	7-Dec-99	14	I	X01.0119	CGF31
23339	Human	1996	14	I	X01.0003	CGF38
F30	Bovine	8-Aug-95	14	I	X01.0056	CGF23
LN637-4	Environmental	2001	14	I	X01.0074	CGF24
LS110	Bovine	22-Aug-01	14	I	X01.0132	CGF39
AA1002-1	Environmental	10-Jun-02	14	I	X01.0176	CGF23
93111	Human	1993	14	I	X01.0001	CGF50
EDL933	Human	1982	21	I	X01.0138	CGF21
LRH13	Human	1-Nov-96	23	II	X01.0095	CGF10
F12	Bovine	8-Aug-95	23	II	X01.0095	CGF8
F1081	Bovine	25-Jun-96	23	II	X01.0097	CGF5
F1305	Bovine	26-Aug-96	23	II	X01.0093	CGF11
R1797	Human	1996	23	II	X01.0527	CGF12
LS68	Bovine	2001	23	II	X01.0049	CGF9
FRIK920	Bovine	NA	23	II	X01.0160	CGF2
FRIK1999	Bovine	NA	23	II	X01.0162	CGF1
EC19920026	Bovine	NA	23	II	X01.0046	CGF7
63154	Human	25-Jun-95	31	I	X01.0002	CGF20
8-1360-1	Human	NA	31	I	X01.0191	CGF47
58212	Human	1996	31	I	X01.0144	CGF20
F2	Bovine	8-Aug-95	31	I	X01.0192	CGF49
F744	Bovine	18-Mar-96	31	I	X01.0047	CGF48
F1095	Bovine	25-Jun-96	31	I	X01.0002	CGF22
R1195	Human	1-Jul-95	31	I	X01.0002	CGF47
F1103	Bovine	25-Jun-96	32	I	X01.0150	CGF44
Sakai	Human	ATCC	32	I	X01.0139	CGF27
AA1000-2	Environmental	10-Jun-02	32	I	X01.0174	CGF43
M01MD3265	NA1	NA	32	I	X01.0154	CGF46
APF59-3	Environmental	3-Jun-04	32	I	X01.0247	CGF25
70490	Human	9-Jul-95	34	I	X01.0002	CGF36
F732	Bovine	18-Mar-96	34	I	X01.0088	CGF45
AA619-2	Environmental	4-Jul-01	34	II	X01.0092	CGF13
LS236	Bovine	10-Oct-01	34	I	X01.0008	CGF37
AA995-2	Environmental	10-Jun-02	34	II	X01.0173	CGF15
FRIK1985	Bovine	NA	45	II	X01.0161	CGF4
FRIK1990	Bovine	NA	54	II	X01.0111	CGF16
FRIK2001	Bovine	NA	54	II	X01.0163	CGF6
EC970520	NA1	12-Dec-02	67	II	X01.0525	CGF3
12491	Human	03-Feb-95	74	II	X01.0526	CGF14
E2328	Human	12-Jan-95	87	I	X01.0062	CGF29
F1082	Bovine	25-Jun-96	87	I	X01.0062	CGF32
H432	Bovine	21-May-96	87	I	X01.0062	CGF29
H435	Bovine	21-May-96	87	I	X01.0062	CGF29
H451	Bovine	21-May-96	87	I	X01.0080	CGF33
H453	Bovine	21-May-96	87	I	X01.0062	CGF29
H454	Bovine	21-May-96	87	I	X01.0062	CGF29
H568	Bovine	17-Jun-96	87	I	X01.0062	CGF29
H571	Bovine	17-Jun-96	87	I	X01.0062	CGF29
H572	Bovine	17-Jun-96	87	I	X01.0062	CGF29
H573	Bovine	17-Jun-96	87	I	X01.0062	CGF29
H574	Bovine	17-Jun-96	87	I	X01.0063	CGF35
H2160	Bovine	24-Mar-97	87	I	X01.0065	CGF28
H2161	Bovine	24-Mar-97	87	I	X01.0065	CGF28
H2163	Bovine	24-Mar-97	87	I	X01.0065	CGF29
H2164	Bovine	24-Mar-97	87	I	X01.0343	CGF29
H2176	Bovine	24-Mar-97	87	I	X01.0086	CGF29
H2704	Bovine	9-May-97	87	I	X01.0522	CGF30
H2718	Bovine	9-May-97	87	I	X01.0086	CGF29
H2723	Bovine	9-May-97	87	I	X01.0065	CGF29
H2727	Bovine	9-May-97	87	I	X01.0065	CGF29
H2731	Bovine	9-May-97	87	I	X01.0065	CGF29
H4420	Bovine	23-Dec-97	87	I	X01.0086	CGF30
EC980120	Bovine	23-Dec-97	87	I	X01.0086	CGF30
EC980121	Bovine	23-Dec-97	87	I	X01.0086	CGF30
EC980122	Bovine	23-Dec-97	87	I	X01.0086	CGF30
EC980125	Bovine	23-Dec-97	87	I	X01.0086	CGF30
OK1	Human	1996	Atypical	I	X01.0529	CGF51
K12 MG1655	ATCC	NA	NA	NA	NA	CGF17

^aNA, not available.

PCR.

PCRs were carried out in a reaction volume of 50 μl containing 1× PCR buffer (Qiagen), 1 mM each deoxynucleoside triphosphate (Invitrogen), 0.2 μM each primer (Alpha DNA), 1 U Taq DNA polymerase (Qiagen), and distilled H₂O to fill the remaining volume. Amplification was performed using either a GeneAmp PCR System 9700 (Applied Biosystems) or a Mastercycler epGradient (Eppendorf), with an initial denaturing step of 95°C for 5 min followed by 30 cycles of 95°C for 30 s, an annealing step of 20 s at the appropriate temperature and an extension step of 1 min/kb expected product size at 72°C, and completed with a final extension of 72°C for 5 min. Visualization of PCRs was carried out following agarose gel electrophoresis. Briefly, 8 μl of PCR mixture combined with 2 μl loading dye (0.25% [wt/vol] bromophenol blue, 40% [wt/vol] sucrose) were run on a 1% (wt/vol) agarose gel stained with ethidium bromide for 40 min at 110 V and visualized under UV light.

Phage, pulsed-field gel electrophoresis, and lineage typing.

Bacteriophage typing was carried out as described by Ahmed and colleagues and extended by Khakhria and colleagues (1, 25). PFGE was carried out at the Laboratory for Foodborne Zoonoses in Guelph, Ontario, according to the Centers for Disease Control and Prevention manual standard 1-day protocol, as previously described (40). Lineage typing was carried out using the E. coli O157:H7 lineage-specific polymorphism assay (53).

Construction of dendrograms.

Results of the PCR amplifications were converted to binary values (0 for absence, 1 for presence) and clustered using Bionumerics version 5.1 (Applied Maths) with the simple matching distance metric and the average linkage method of clustering. The dendrogram was rooted with K-12 strain MG1655.

PFGE banding patterns were analyzed with Bionumerics version 5.1 (Applied Maths) and clustered by UPGMA. The dendrogram was created using the Dice similarity coefficient with an optimization of 1.5% and a tolerance of 1.5%.

RESULTS

Forty-four loci that were previously found to be variously absent or present in genomic islands of E. coli O157:H7 and which were distributed throughout the entire genome (Tables 2 and 3) were selected for initial testing and assessed by PCR assay across 79 E. coli O157:H7 isolates and the laboratory strain K-12 MG1655 (Table 1). Of the initial 44 loci targeted by PCR, 13 generated nonbinary data (e.g., multiple bands or bands of various sizes). The remaining 31 loci produced binary results and generated 54 unique fingerprints. Each of the 31 loci was assessed for its ability to discriminate among strains. It was found that 23 loci differentiated strains that only differed at one, two, or three loci. The remaining eight loci only offered additional discrimination between strains where four or more loci already differed and did not contribute additional fingerprints (data not shown); exclusion of these loci from the final set maintained the resolution of the 54 unique fingerprints (Table 4).

TABLE 2.

The E. coli O157:H7 lineage II-specific VAP loci initially tested^a

Region	Function
A01	EC93 contact-dependent inhibition (cdiB) gene
A05	Bacteriophage P2 genes J and I, encoding baseplate assembly proteins
A06b	E. coli Sakai intergenic region between ECs1159 and ECs1160
A07	E. coli Sakai ORF ECs1928, encoding a hypothetical protein, adjacent to ECs1997, encoding a putative filament protein
A12	DNA sequence without significant homology to annotated DNA sequences
B01	Shigella flexneri bacteriophage V ORFs orf39, orf40, and dam, encoding a replication protein, a hypothetical protein, and a DNA adenine methylase
B02	Bacteriophage P2 genes J and I, encoding baseplate assembly proteins
B05	ColE6-CT14 colicin plasmid, colicin E6 gene
B13	DNA sequence without significant homology to annotated DNA sequences
B18	pColD-157 plasmid, colicin D-157 activity protein

^aIdentified by Steele et al. (48).

TABLE 3.

The E. coli O157:H7 lineage I VAP loci initially tested, the primers targeting them, and their locations in the sequenced genomes of K-12 MG1655 and E. coli O157:H7 strains EDL933 and Sakai

Primera	K-12 ORF	EDL933 ORF	Sakai ORF	Genomic island	Function
fprn01	B0546	NAb	NA	KI#37	Putative ARAC-type regulatory protein; ybcM [B]
fprn02	B1201	NA	NA	KI#74	Putative sensor-type regulator
fprn03	B2082-B2083	NA	NA	KI#103	Hypothetical protein
fprn04	B2083	NA	NA	KI#103	Prophage P2 ogr protein and hypothetical protein
fprn05	B2361	NA	NA	KI#121	Hypothetical protein
fprn06	NA	NA	ECs4943-ECs4944	S-loop#274	Putative regulatory protein; putative DNA-binding protein
fprn07	NA	NA	ECs4955-ECs4956	S-loop#274	Hypothetical proteins
fprn08	NA	Z0020	ECs0020	OI#1/S-loop#1	Hypothetical protein
fprn09	NA	Z0309	ECs0275	OI#8/S-loop#16	Similar to regulatory protein Cro (phage lambda)
fprn10	NA	NA	ECs1184	S-loop#69	Hypothetical protein
fprn11	NA	Z1203	ECs1386	OI#43/S-loop#72	Hypothetical protein
fprn12	NA	Z1770	ECs1506	OI#50/S-loop#77	Hypothetical protein
fprn13	NA	Z1963	ECs1695	OI#55/S-loop#83	Hypothetical protein
fprn14	NA	Z2400	ECs1941	S-loop#91	Putative transcriptional regulator
fprn15	NA	Z2386-Z2387	ECs1954	S-loop#93	Hypothetical protein
fprn16	NA	Z3358	ECs2990	OI#93/S-loop#153	Putative prophage repressor CI
fprn17	NA	Z3939	ECs3507	OI#108/S-loop#186	Hypothetical protein
fprn18	NA	Z4188	ECs3724	OI#115/S-loop#197	Type III secretion protein EpaQ
fprn20	NA	NA	ECs4954	S-loop#274	Hypothetical protein
fprn21	NA	Z5690	ECs5072	OI#167/S-loop#276	Putative carbohydrate ABC transport system permease
fprn22	B1289	NA	NA	KI#87	Hypothetical protein
fprn23	NA	Z1204	ECs1387	OI#43/S-loop#72	Hypothetical protein
fprn24	NA	Z1447	NA	OI#45	Putative repressor protein CI of bacteriophage BP-933W
fprn25	NA	Z1481-Z1483	ECs1225-ECs1228	OI#45/S-loop#69	Hypothetical protein and putative tail fiber protein
fprn26	NA	Z1835	ECs1574	OI#51/S-loop#78	Putative integrase
fprn27	NA	Z1841	ECs1579-ECs1580	OI#51/S-loop#78	Hypothetical protein
fprn28	NA	Z1963-Z1965	ECs1695-ECs1698	OI#55/S-loop#83	Hypothetical protein; putative ferric enterobactin transport ATP-binding protein; putative ABC transporter permease protein; putative ATP-binding protein of ABC transporter
fprn29	NA	Z2090	ECs2209	OI#57/S-loop#108	Putative repressor protein encoded within prophage CP-933O
fprn30	B1519	Z2186	ECs2126	NA	Putative enzyme
fprn31	NA	Z2311-Z2312	ECs2016-ECs2017	OI#66/S-loop#98	Hypothetical protein
fprn32	NA	Z2322	ECs2007	OI#67/S-loop#95	Putative outer membrane protein
fprn33	NA	Z2969	ECs2620	OI#76/S-loop#119	Putative transcriptional regulator
fprn34	NA	Z4092	ECs3637	OI#112/S-loop#194	Hypothetical membrane protein
fprn35	NA	Z5885	ECs5249	OI#172/S-loop#286	Putative resolvase

^aForward and reverse primer sequences are given in Table 4.

^bNA, not applicable.

TABLE 4.

The 23 CGF primers and the conditions for their amplification

Primer	Expected size (bp)	Annealing temp (°C)	Forward primer (5′→3′)	Reverse primer (5′→3′)
fprn02	413	55.0	ACGGGCAATCACTGAACTC	TGACGGCGATATAAGGACC
fprn30	614	55.0	TCGCCATTACCAACTGAC	GATAACGGGCATCGACTC
fprn03	347	57.0	CAATGTATCGCTGTACCGACTC	GCAGGACTGGACAATCACTAAG
fprn05	422	55.0	GCCGGTCATCTTTCTCAC	GCGCATTTGCGTAACTCC
fprn10	408	55.0	TTGAGGCCGAACGTAAAC	CGCCTCTTAACGGTAGTC
fprn11	442	55.0	GCGGATGAATCCCTCAATCG	CCAGCAGGAACATCGCATAG
fprn13	208	55.0	ACCGCCCTGGATATCTTC	ACGTCAGCTGGAGTCTTG
fprn28	954	57.0	CCGCAGCCCATATCGAATAAGG	TCATGCGTCTCTGGGTTACTCC
fprn14	150	55.0	CTCGTCCAGAAGGTCAATGATG	CCTCGCCGAGAGAAATTAAG
fprn15	648	55.0	GGTCGCATGCCTTAGATAACTC	TTGGTGGTCCGATGCTTGAAG
fprn32	2210	55.0	CGTTGCGGTTTGTGTATC	CTGCGGTAGTGAATAACG
fprn29	1002	55.0	CCATCGTCAACCTCATC	GCTCCGTAAGTCAATGG
fprn33	359	55.0	TGTCACCTCCCTACTATCAC	TGCCATGAGCGACATTG
fprn16	512	55.0	GAAAGGTTAGCGCGGATG	TCAGCCAAACGTCTCTTC
fprn34	1007	55.0	AACCGGGACATTAACGAC	ACGAAGAGCGAGAATGAG
fprn18	101	55.0	TTCACGCCAAATGGCAAC	TCTGCAGGACCTATTGCAGTAG
fprn06	807	57.0	ACCCGCTTTAGATTCTACCG	AAAGAGCGCAGTGATAAGCC
fprn07	926	57.0	GCAGGACTGCCAGAAAGTTAAG	CGCCGATTTCCCTAAAGGTAAG
fprn35	489	55.0	ACCAGCGCATGGAACTTG	GCGTCGCTCTTGAAACAC
fprn24	691	55.0	CACTGGTTACCAGCTATG	CGCAAGCCTGTAAAGAAG
A01a	432	60.8	ACCAAGGCATCCCCCGTGTGAA	ATAATCCGCTGGGGCTGGCTGAC
B01a	316	60.7	AAGCCCCCACGTAATTCCCTGACA	CATTTCCCGCGCCTGACTGAGA
A12a	446	55.9	GGGCGGACTTTGTTTGGTTGAA	GCCTGGCGGAAATGGACTGTAT

^aIdentified by Steele et al. (48).

Phage typing, PFGE analysis, and lineage typing of the 79 E. coli O157:H7 strains resolved 13 separate PTs, 52 unique PFGE profiles, and 58 LSPA6-LI, 5 LSPA6-LI/II, and 16 LSPA6-LII strains. The 23-locus binary fingerprint of every strain was hierarchically clustered using Bionumerics version 5.1 (Fig. 1). As can be seen, at a 60% similarity cutoff, four unique clusters of strains were observed. One cluster contained the non-O157:H7 strain K-12 MG1655, which was used as an outgroup, while the three remaining clusters corresponded to the three major O157:H7 lineages, LSPA6-LI, LSPA6-LI/II, and LSPA6-LII. The first cluster contained the 16 LSPA6-LII strains of PT23, PT34, PT45, PT54, PT67, and PT74; the second contained the 58 LSPA6-LI strains of PT14, PT21, PT31, PT32, PT34, and PT87; the third contained the 5 LSPA6-LI/II strains of PT1 and PT2. Every PT was exclusive to a cluster, except for PT34, which was observed in both LSPA6-LI and LSPA6-LII clusters. When strains were grouped according to PFGE banding pattern (Fig. 2), the LSPA6 lineage groupings were no longer absolute. LSPA6-LI strain AA1000-2 grouped with LSPA6-LII strains, LSPA6-LI/II strain R1388 grouped apart from the other four LSPA6-LI/II strains, and LSPA6-LII strains AA995-2 and 12491 grouped among the LSPA6-LI strains.

FIG. 1.

The dendrogram created when the 23-locus binary fingerprint of each of the 79 E. coli O157:H7 strains and non-O157:H7 strain K-12 MG1655 were clustered using Bionumerics version 5.1. A black square represents the presence of a locus and a white square represents the absence of a locus, as determined by PCR. Sequence data for EDL933 indicate that locus fprn16 of OI#93/S-loop#153 is homologous to that of Sakai; however, repeated efforts to amplify fprn16 in EDL933 were unsuccessful. Similarly, previous work with the Sakai strain used in this study has shown a deletion of fprn29 in OI#57/S-loop#108.

FIG. 2.

The dendrogram created when the XbaI PFGE banding pattern of each of the 79 E. coli strains studied were clustered by UPGMA using the Dice coefficient in Bionumerics version 5.1, with an optimization of 1.5% and a tolerance of 1.5%.

Strains with the same CGF profile were found to share the same PT, a trend that was observed for all seven CGF profiles represented by multiple strains in the data set (CGF20, CGF23, CGF28, CGF29, CGF30, CGF47, and CGF52). By contrast, PFGE profile X01.0002 contained strains of PT14, PT31, and PT34.

In order to determine whether CGF could distinguish epidemiologically relevant subpopulations, our data set included 25 PT87 isolates that had been obtained from a closed herd of cattle at the Animal Diseases Research Institute (ADRI) in Lethbridge, Alberta, Canada, over a period of 18 months. Five separate isolates were obtained from each of five sampling dates from 21 May 1996 through 23 Dec 1997. As Fig. 1 shows, a discrete branch in the dendrogram representing strains with fingerprints having greater than 91% identity (i.e., fewer than three differences in a CGF profile) contained all PT87 strains in the data set and included the 25 isolates from the ADRI cattle herd. The cattle strains could be further subtyped into two major groups based on a 100% similarity threshold, one of which (CGF29) consisted of strains positive for locus fprn33 isolated between 21 May 1996 and 9 May 1997. The other cluster (CGF30) contained strains found to be negative for locus fprn33, including all five isolates from 23 Dec 1997 and one from the prior sampling date of 9 May 1997. This suggests a shift in the clonal architecture of the O157:H7 herd strains over time. Microarray analysis of three PT87 strains showed that the phage-related OI#76/S-loop#119, in which fprn33 is located, was completely absent in fprn33-negative strain H4420 but fully intact in fprn33-positive strains E2328 and F1082 (55). PFGE typing was able to distinguish the isolates from 23 Dec 1997, one isolate from 9 May 1997, and one isolate from 24 March 1997 as a separate type, X01.0086, and delineated seven separate PFGE types among the herd strains, but as Fig. 2 shows the herd strains did not form a discrete cluster when grouped by PFGE banding pattern.

DISCUSSION

The rapid progress in bacterial genome sequencing has allowed the development of high-resolution subtyping methods that can provide inter- and intraspecies bacterial genome comparisons. This study capitalized on prior comparative genomics hybridization experiments that examined the genomes of 34 E. coli O157:H7 strains and regions of variability therein, which allowed the development of a PCR-based typing system able to group strains into phenotypically related lineage-specific types, to determine those that were epidemiologically linked, and to offer discrimination between related and nonrelated isolates (48, 55). A study by Wick et al. using a smaller set of strains found most of these same regions to be divergent and phage related (52). The CGF loci used in this study were selected to capture the diversity in the E. coli O157:H7 population, not to reproduce the dendrogram created using microarray data from the original strains used in the study by Zhang et al. (55). It is therefore interesting to see that the relationships observed among strains with the microarray data have been maintained in the CGF-based dendrogram. We therefore conclude that CGF data provide a meaningful snapshot of the genetic diversity of the E. coli O157:H7 population and the genotypes of specific strains. This CGF method relies on 23 loci that are either present or absent and thus in theory offers 2²³ (8,388,608) possible fingerprints.

On the basis of octamer-based genome scanning, E. coli O157:H7 strains have been split into two distinct clonal lineages, lineage I and lineage II, that differ in genotype and host ecology. Lineage II strains are less frequently associated with human disease due either to inefficient transmission from bovine sources or lack of virulence to humans (26). The work of Zhang and colleagues identified an intermediate lineage that shared characteristics with both octamer-based genome scanning O157:H7 lineages (55). Those intermediary strains were designated lineage I/II and corresponded to LSPA6 type 211111 (53). All strains in this study grouped according to LSPA6 lineage, and each lineage formed its own branch of the dendrogram at a similarity threshold of 60%. It is worth noting that CGF data could be used to cluster strains into meaningful groups concordant with broad subtypes, such as LSPA6 lineage, and narrow subtypes, such as PT, respectively, depending on the percent similarity threshold value that was used, while targeting none of the loci used in LSPA6 typing. We have recently published a study in which a strong relationship between PT and LSPA6 lineage genotype was shown to exist among E. coli O157:H7 strains from Canada (57). In that study we found that PT23, PT45, PT54, PT67, and PT74 were LSPA6-LII specific, that LSPA6-LII strains were significantly less likely to be isolated from humans than LSPA6-LI strains, and that these LSPA6-specific groups were phenotypically meaningful in traits such as toxin production and antimicrobial resistance. Thus, CGF appears to produce genotypically and phenotypically relevant subtyping data.

CGF was able not only to offer groupings of similar strains but also to provide discrimination within the groups, both of which are traits essential for a typing system (8). CGF analysis of our data set generated 49 distinct fingerprints among 54 unrelated strains and three highly related fingerprints among the 25 strains from a closed herd of cattle. Strains isolated from the same feedlot or farm have been shown to be highly clonal, with a few dominant subtypes frequently detected and other subtypes occasionally or sporadically detected (13, 16, 27). CGF analysis revealed two dominant subtypes (CGF29 and CGF30) and three sporadic subtypes (CGF28, CGF33, and CGF35) among the herd strains. The most common dominant subtype (CGF29) was exhibited by 15 of 20 herd strains isolated from 21 May 1996 through 9 May 1997, while five strains obtained during this sampling period each showed a one-locus difference with respect to CGF29. One of these variants, strain H2704 (isolated on 9 May 1997) and all five strains isolated in the subsequent sampling period (23 December 1997) shared the same fingerprint (CGF30) consisting of an absent locus fprn33 and comprised a second dominant subtype, suggesting a clonal shift in the dominant subtype of the herd strains. Locus fprn33 encodes a putative transcriptional regulator in the phage-related OI#76/S-loop#119, which also contains genes encoding hypothetical and tail fiber proteins. Microarray data showed that this entire genomic region was lost in a PT87 strain negative for fprn33 (55). It therefore seems that some loci are very stable over time and allow the differentiation of strains that diverged in the distant past, such as those specific to lineage, while others are highly unstable and allow the differentiation of closely related strains, such as those of the ADRI cattle herd. This is consistent with other findings that have found stability in genetic elements, such as the presence of stx_2c being strongly associated with lineage II strains (56), and the fact that some phage-related genomic regions may prevent the incorporation of other genetic elements through changes in phage receptors or immunity to superinfection by other bacteriophages (14, 22). As many of the CGF loci are phage related, and some PTs have been shown to be lineage specific (57), such phage-related genetic elements may play a role in the stability of clusters identified by CGF. The PT87 herd isolates did not share a CGF fingerprint with other strains, and two dominant and three sporadic subtypes could be distinguished within them, suggesting that the method may be useful in separating and typing E. coli O157:H7 outbreak isolates.

While PFGE is currently the “gold standard” of typing and has been shown to differentiate between strains of the same PT, it requires highly trained staff and extensive standardization to generate accurate results that can be compared between laboratories (2, 30, 35, 39). Figure 2 demonstrates that the tolerance levels required to compare PFGE patterns across multiple gels create clusters of identical strains that may appear slightly different to the naked eye. Moreover, temporally and geographically unrelated strains are occasionally given the same typing designation by PFGE (36). This phenomenon was evident in our data where strains TS-97 and F5 were indistinguishable by PFGE despite being isolated over 6 years apart, whereas CGF typing identified five loci where these two strains differed. Conversely, three lineage II strains isolated from the province of Alberta over 4 years (AA619-2, 12491, and AA995-2) were shown to be genetically similar and clustered together with CGF, while PFGE typing did not group the strains and placed the lineage II strains AA995-2 and 12491 among those of lineage I. This suggests that while PFGE is more appropriate for identifying the source of an outbreak, phage typing appears to give more taxonomically relevant data. CGF combines the benefits of both, being able to group strains into taxonomically meaningful subtypes as well as discriminating between and within the groupings. As has been shown, PFGE type X01.0002 in this study was found to contain unrelated strains of various PTs, making it unsuitable for higher-level discrimination among E. coli O157:H7 strains. Additionally, when the strains from this study were clustered by PFGE pattern, although no PFGE type was observed in more than one lineage, the lineage designations were no longer specific to a grouping of strains. The low specificity of PFGE has led to the recommendation that PCR-based methods be used over PFGE in typing studies (8, 44).

As has been shown, every CGF profile was specific to a PT; excluding the PT87 herd strains there were four instances where CGF was incapable of discriminating between two strains. The question of whether seemingly unrelated strains, such as F30 and AA1002-1, both of CGF23 and isolated nearly 7 years apart, are actually similar or if CGF simply lacks the resolution to properly discern them was tested. Microarray data from the work of Zhang et al. (55) were examined where the strains 71074 and 09601Fe046.1, both of CGF52, had been previously analyzed. It was found that the genomic content of strains 71074 and 09601Fe046.1 differed from each other by 0.96% when clusters of two or more open reading frames were examined; the next most closely related strain, Zap0046, differed from 71074 and 09601Fe046.1 by 1.14% and 1.53%, respectively (data not shown). The CGF52 strains 71074 and 09601Fe046.1 were thus found to be extremely similar by microarray analysis and grouped together when the microarray data were hierarchically clustered among 31 E. coli O157:H7 strains of various PTs and lineage. Therefore, although the epidemiological data may not show an obvious linkage between strains of the same CGF type, the genomes of the strains examined showed that they were clearly related. The resolution of CGF is therefore such that confidence can be taken in the relatedness of strains grouped together, even in the absence of epidemiological data, a task for which PFGE proved unsuitable in the analysis of our strain collection.

CGF offers high reproducibility among replicates, as the analysis involves only the presence or absence of single bands, with no judgment of band size or pattern. Such ease of data acquisition allows immediate and accurate interlaboratory reproducibility and exchange. CGF does not require expensive sequencing equipment or analysis software, such as is required for MLST, MLVA, or PFGE, which lends itself to immediate implementation by any PCR-capable laboratory; the loci may also be easily transferable to higher-throughput platforms, such as single-tube suspension microarrays.

Further, the 23-locus fingerprint offers a glimpse of the genetic structure of the E. coli O157:H7 isolate being typed, not just anonymous bands generated from enzyme recognition sites, the sequences of a few loci, or the number of repeats of a small segment of DNA. Because the presence of specific genomic islands has been linked to increased virulence of strains of several bacterial species (17, 23), and as every locus in the fingerprint targets a region of a given genomic island, it is possible that a strain could be identified as “more” or “less” pathogenic to humans based on its CGF subtype.

Recently, Manning et al. (28) used single-nucleotide polymorphism (SNP) analysis to identify a unique group (clade 8) of hypervirulent E. coli O157:H7 strains in the United States. In our study CGF was shown to distinguish among E. coli O157:H7 strains with a high and low propensity for human infection, so it would be interesting to see if this hypervirulent clade of the organism could also be identified with this genotyping approach. Unfortunately, restrictions in the international transfer of pathogenic microorganisms have made it difficult to develop a common set of reference E. coli O157:H7 strains for use in genotyping studies. However, the availability of nucleotide sequence data for an increasing number of E. coli O157:H7 strains in GenBank may allow such a typing set to be available in silico.

This study presents the successful development of a high-resolution molecular typing method that exploits data obtained through comparative genomics-based population studies. A similar methodology has recently been described for typing Streptococcus pneumoniae (11), Campylobacter jejuni (49), and Escherichia coli strains (47), suggesting that this approach is likely to be useful in the development of next-generation genotyping methods. Subsequent iterations of VAP analysis targeting additional loci will be derived from new E. coli O157:H7 sequence data as they become available. There are 16 sequencing projects currently under way on this pathogen available at the NCBI “genomes in progress” web page. At the very least, insight into the presence or absence of the 23 loci is a starting point for further genetic analysis.

Conclusion.

CGF was shown to be superior to both phage typing and the current typing standard, PFGE, as it offers comparable resolution between typed strains and much greater specificity. It also benefits from being a simple and fast PCR-based assay generating binary results and one that requires minimal training for both the actual testing and the data analysis, simplifying interlaboratory comparisons. Since CGF is a molecular typing method developed from whole-genome comparisons and O157:H7 lineage-specific sequences, it targets regions found to be most variable among strains, and thus most amenable for use in typing. Furthermore, the subtype groupings given by CGF appear to be phenotypically meaningful, while those given by PFGE appear not to be. In the future, as new sequence data become available, additional variable loci should be evaluated for their use in high-resolution typing of this pathogen. CGF could be an alternative or adjunct to PFGE typing, and its performance should be evaluated against other emerging molecular typing methods, such as MLVA.