Pathogenic Potential to Humans of Bovine Escherichia coli O26, Scotland

This pathogen may be the next Shiga toxin–producing E. coli of concern.

E nteropathogenic Escherichia coli (EPEC) and enterohemorrhagic E. coli (EHEC) are gastrointestinal pathogens associated with asymptomatic carriage and human diseases ranging from mild diarrhea to hemorrhagic colitis and hemolytic uremic syndrome (HUS) (1). Worldwide, EHEC serogroup O157 strains are responsible for most human Shiga toxin-producing E. coli (STEC) infections (2). However, recent outbreaks in Germany (3) and the United States (4) have highlighted the increasing role of other Shiga toxin-producing serogroups (termed non-O157 strains) in causing human disease.
The pathogenesis of EHEC strains is associated with production of Shiga toxins expressed from lysogenic bacteriophages in the EHEC genome. There are 2 predominant classes of Shiga toxins (1 and 2), each encoded from 2 genes, stxAB, with the genotypes simplifi ed to stx 1 or stx 2 in this study. Whereas EHEC is regarded as an emerging zoonotic pathogen, the related EPEC strains cause diarrhea, especially in infants in developing countries (1).
EPEC and EHEC express a type III secretion (T3S) system that translocates multiple effector proteins into host cells and manipulate host innate responses, which are needed for colonization (5)(6)(7). The T3S system is central to the formation of attaching and effacing lesions on the intestinal epithelium that requires the bacterial outer membrane protein intimin (encoded by eae) and the secreted bacterial protein-the translocated intimin receptor (Tir)that is injected into the host cell. Studies of EPEC and EHEC O157 show that these pathogens trigger different actin polymerization pathways respectively involving Tir cytoskeleton coupling proteins (tccP or tccP2) (5). The formation of attaching and effacing lesions is needed for bacterial colonization by EHEC O157 and EHEC O26 in cattle. However, Shiga toxin (Stx) is the principal factor responsible for severe human illness, including HUS (8). Strains with stx 2 alone appear more strongly associated with HUS than do strains with only stx 1 (8)(9)(10). These Pathogenic Potential to Humans of Bovine Escherichia coli O26, Scotland observations have been supported by mouse (11) and primate models (12). Among non-O157 EHEC, Stx-producing E. coli O26 also causes human disease (13)(14)(15) and has been isolated from livestock (16). However, unlike EHEC serogroup O157, it may be pathogenic for both cattle and humans (17). Although the origin of human E. coli O26 infections is rarely identifi ed (18), evidence exists of person-to-person spread (19) and foodborne transmission (20). Furthermore, because EHEC O26 has been isolated from the feces of cattle and other animals (16,21,22), potential exists for direct and indirect zoonotic transmission to humans (20).
In the United Kingdom, Stx-producing human E. coli O26 infections are usually uncommon and not clinically severe. Most identifi ed EHEC infections are associated with serogroup O157, and incidence rates in Scotland are among the highest, compared with rates in countries with comparable surveillance (23). In contrast with human infection rates, farm and animal prevalence of E. coli serogroups O26 and O157 are similar in Scotland (21).
Our fi rst objective was to identify the cause of the disparity between the incidence of E. coli O157 and O26 infections among humans in Scotland by examining the natural heterogeneity among serogroup O26 strains and reexamining their prevalence in cattle. As a second objective, we used different molecular techniques to compare the relationships between E. coli O26 isolates recovered from humans and cattle.

Bacterial Isolates
All bacterial isolates used in this research are shown in Table 1

PCR for Virulence Genes
Genes encoding stx 1 , stx 2 , eae, and hlyA were detected by using multiplex PCR (24). Possession of the locus of enterocyte effacement (LEE) pathogenicity island was confi rmed by PCR screening for sepL. Genes encoding tccP and tccP2 were detected with PCR by using universal primer pair univtccP/tccP2-F and tccP-R (25). Primers and PCR conditions used in this study are listed in Table 2.

Pulsed-fi eld Gel Electrophoresis
We performed pulsed-fi eld gel electrophoresis (PFGE) on 187 isolates (33 from humans and 154 from cattle) ( Table 1) by using the method of Willshaw et al. (26), standardized by the Scottish E. coli O157/VTEC Reference Laboratory (SERL; Edinburgh, UK) and the Laboratory of Gastrointestinal Pathogens (Colindale, UK). A phage λ ladder (48.5 kb) was used as a DNA size standard, and gels were run with a linearly ramped switch time of 5-50 s applied for 38 h at a voltage of 5.4 V/cm and an included angle of 120°. PFGE profi les were analyzed by using BioNumerics (version 3.0) software (Applied Maths, Kortrijk, Belgium). The degree of similarity between profi les was determined by the Jaccard coeffi cient, and dendrograms were generated by using the unweighted pair group method with arithmetic mean with 1.3% tolerance and 1% optimization settings.

Multilocus Sequence Typing
We performed multilocus sequence typing (MLST) on 63 isolates (30 from humans and 33 from cattle) ( Table   440 Emerging  Table 2). Sequence data for espA and the LEE1 promoter region were compared by using ClustalX (www.clustal.org). Each unique sequence was given a different allele number assigned in the order in which they were discovered.

Data Management and Statistical Analyses
Prevalence of E. coli O26 E. coli O26 isolated from cattle are more heterogeneous than E. coli O157 with respect to the presence of stx and eae ( Figure 1). Therefore, we determined the prevalence of E. coli O26 for 4 different groups with potential differences in virulence: 1) all E. coli O26, similar to Pearce et al. (21); 2) stx 1 +; 3) stx 1 +stx 2 +; and 4) stx 2 +eae+. Farm-level and fecal pat-level prevalences were calculated by using the method of Pearce et al. (28). SAS version 9.1.3 (SAS Institute, Cary, NC, USA) was used to fi t generalized linear mixed models and to generate bootstrap-based estimates of key parameters. Excel 2000 (Microsoft Corporation, Redmond, WA, USA) was used to implement a Latin hypercube sampling to convert results from generalized linear mixed models into prevalence taking into account random effects (29).

PFGE Analysis
Differences between isolates from humans and cattle were evaluated by assigning isolates to statistically distinct groups. The methods used are described in (30). Group membership was determined by estimating an optimum cutoff value using dendrogram-based distances. Statistical support for each group was evaluated by using the cophenetic correlation. The statistical signifi cance of the cophenetic correlation was evaluated by using a Mantel test. Program zt (31) was used with 1,000 simulations, and the probability of the observed cophenetic correlation occurring by chance was <5% for groups of isolates regarded as statistically distinct.
A subset of 41 isolates from Scotland (8 from humans and 33 from cattle) were included in the MLST and PFGE analyses. The PFGE data for 3 isolates from humans were missing and therefore could not be included in the comparison. Agreement between these 2 techniques was examined.

Nonmetric Multidimensional Scaling
To identify patterns in data associated with the clinical severity of infection in humans, we used nonmetric multidimensional scaling (NMS) to analyze all genotype characteristics measured. PC-ORD software version 6.03 (MjM Software Design, Gleneden Beach, OR, USA) was used. The main matrix was created from the following isolate data: MLST type (ST, ST complex), tccP and tccP2, Stx status, hlyA, eae, LEE, and sequence upstream region of LEE1. A second matrix that used isolate data and an indicator of clinical severity was also created. Isolates were separated into 2 groups: 1) HUS and those associated with HUS and 2) non-HUS, including those from diarrhea, bloody diarrhea, and hemorrhagic colitis. The grouping was chosen because HUS has clearly identifi able clinical characteristics and is a reliable indicator of severe illness.
We used NMS with a Euclidian distance measure after standardizing each variable by division by its standard deviation. The dimensionality of the dataset was determined by plotting an inverse measure of fi t ("stress") to the number of dimensions. Optimal dimensionality was based on the number of dimensions with the lowest stress. A 3-dimensional solution was shown to be optimal. We performed several NMS runs to ensure that the solution was stable and probably represented a confi guration with the best possible fi t. For each NMS run, we used 500 iterations with random starting coordinates. The overall signifi cance of the difference between HUS and non-HUS groups was assessed by using multiresponse permutation procedures (32).

Prevalence in Cattle
Of the 338 farms visited, 68 were positive for E. coli O26 (mean prevalence 20.1%, exact binomial 95% CI 16.0-24.8); these 68 farms were evenly distributed across Scotland (Figure 2, panel A). Farms were classifi ed by their isolates' virulence groups: stx-, stx 1 +, and stx 1 +stx 2 +. No farm had stx 2 + only strains. Farms with stxisolates clustered in southern Scotland. Farms that were stx 1 + were fairly evenly distributed, but those with stx 1 +stx 2 + isolates were found mainly in the northeast (Figure 2, panel B). Analysis of the farms grouped according to their Scottish animal health district (21) signifi cantly supported this observation (p = 0.005). No seasonal differences were observed (p = 0.59).

PFGE Genotyping
PFGE profi les of the 187 E. coli O26 isolates from 154 cattle and 33 humans were compared (online Technical Appendix Figure 1, wwnc.cdc.gov/EID/pdfs/11-1236-Techapp.pdf). This comparison identifi ed 4 groups comprising 5, 19, 142, and 16 isolates, which all included E. coli O26 isolates from cattle and humans. The 4 groups are characterized by statistically signifi cant cophenetic correlations ranging from 0.807 to 0.945. Isolates from groups 2 and 4 were mainly stx-, whereas groups 1 and 3 were mainly stx+. Five isolates were not assigned to a group. All of the stx+ isolates from outside Scotland were in group 2, along with the 5 stx+ isolates from Scotland. The remaining isolates from Scotland were in group 1 (2 isolates), 2 (3 isolates), and 4 (1 isolates).
We compared PFGE profi les of the 41 E. coli O26 isolates from 33 cattle and 8 humans. PFGE grouping correlated strongly with MLST ST (online Technical Appendix Figure 2). PFGE was more discriminative than MLST because each ST was represented by >1 PFGE pattern.

MLST Analysis
Irrespective of source, most isolates had a close genetic background (Table 4). Ninety percent of all isolates were either ST21 or ST29, both belonging to ST complex 29 and differing by 1 nt in the adk locus. Most isolates from continental Europe were ST21 (74%) and stx 2 + (53%). Considerably fewer isolates from humans in Scotland were ST21 (36%) and stx 2 + (18%). Other differences between Scotland and continental Europe were related to the relative absence of tccP2 in isolates from Scotland and the presence of different alleles upstream of LEE (Table  4). Isolates from cattle were genotypically more similar to those from humans from Scotland than from continental Europe (Table 4), except for the proportion of isolates that were ST21 and stx 2 +. This similarity may be explained by the fact that the isolates analyzed from cattle (33 [13%] of 249) were not a random sample of all isolates from cattle but were selected on their virulence properties (i.e., stx+ and eae+) to assess their potential to cause human infection rather than to provide a direct comparison with isolates from humans in Scotland.

Disease Severity Association
Before NMS analysis, 3 isolates were excluded because of missing data. ST complex, LEE, tccP, and eae were removed because no variation existed in the data ( Table  4). The fi nal matrix comprised 5 variables and 27 isolates. The fi nal NMS solution was 3-dimensional and explained 90.8% of the variation in clinical severity and explained more variation than expected by chance (Monte Carlo, p = 0.008). Final stress for the 3-dimensional solution was 8.53 with no real risk for drawing false associations (32), and the fi nal instability was 0 with 53 iterations.
The results of the NMS models are shown in joint graphs (Figure 3, panel A) and 3-dimensional ordination graphs ( Figure 3, panel B) of the distance between sample units, which approximates dissimilarity in clinical severity. Where appropriate, R package version 2.12.1 (33) was used to add 80% confi dence ellipsoids to discriminate between groups of interest. Variables used in the NMS analysis are shown as vectors; the direction indicates positive and negative correlation (Figure 3, panel A). Although some   (21) and this study resulted from use of different statistical models. Pearce et al. (21) aimed to provide national prevalence estimates; thus, weighted estimates of mean prevalence were generated that accounted for the fractions of the national herd found in different Animal Health Districts (AHDs). By contrast, we reported the mean of the sample collected in a stratified fashion across the AHDs. Because there were only relatively small differences in mean prevalence in the different AHDs, the effect on the 2 means is negligible and the effect on the standard errors relatively small. overlap exists, the difference between HUS and non-HUS groups was statistically signifi cant (multiresponse permutation procedures, p = 0.007) (Figure 3, panel B). HUS was associated with ST21 and stx 2 , whereas non-HUS cases were associated with ST29 and the absence of stx 2 .
The hlyA gene was not associated with either condition because 69% of non-HUS isolates possess hlyA and 67% of HUS isolates possess hlyA (Figure 3, panel A).

Discussion
Although Stx-producing E. coli O26 is a major cause of HUS in continental Europe, human infection with this pathogen is uncommon in Scotland and has resulted in less severe illness. In contrast, E. coli O157 is a major cause of human infections and HUS (1). Previous reports of a high prevalence of E. coli O26 in cattle (21) do not appear to be matched by a high level of infection in humans. Our fi rst objective was to determine whether this discrepancy could *MLST, multilocus sequence typing; ST, sequence type; ND, not determined; stx, Shiga toxin; +, stx gene present; -, stx gene absent. LEE1 encodes the first operon of the locus of enterocyte effacement (LEE) and the 644-bp region sequenced includes the promoter for this operon amplified using the defined LEE primer pair in Table 2. †The ST and ST complex were assigned in accordance with the E. coli MLST database (http://mlst.ucc.ie/mlst/dbs/Ecoli). ‡espA encodes for the surface-associated protein, espA. The allele numbers at each loci were assigned in the order in which they were discovered. §Six different sequences were discovered for the region upstream of LEE1 in the E. coli O26 isolates. Approximately 644 bp of sequence data were determined, and all sequence variation between the E. coli O26 alleles occurs within a 91-bp region upstream of LEE1 in the region where regulators act in E. coli O157:H7. The allele numbers at each locus were assigned in the order in which they were discovered. ¶eae, gene that encodes intimin. #sepL, gene confirming the presence of LEE pathogenicity island. **hlyA, enterohemolysin. † †tccP/tccP2, genes encoding Tir-cytoskeleton coupling protein and Tir-cytoskeleton coupling protein 2, which are used in actin polymerization and subsequent attaching and effacing lesion formation.
be explained by the natural heterogeneity in strains by examining virulence determinants associated with serious human disease in strains recovered from cattle in Scotland.
The overall percentage of farms on which E. coli shedding was detected and the overall proportion of fecal pats positive were comparable for both organisms, but for E. coli O26, the farm and pat-level prevalence becomes smaller as the carriage of the virulence determinants eae and stx are taken into account (Figure 4). In contrast, the prevalence of virulent strains of E. coli O157 (stx+eae+) is not smaller. Thus, because there are fewer virulent strains at farm and animal levels, the risk for human illness is expected to be much lower for E. coli O26 than for E. coli O157 in Scotland.
Our second objective was to determine the genetic relationship of strains from different sources. PFGE characterization demonstrated that PFGE profi les of isolates from humans differed from those from cattle. This result was expected in the absence of a demonstrable epidemiologic link. However, human strains are widely distributed among groupings of isolates from cattle. Although an exact interpretation of relatedness of isolates from humans to animals is not possible, the isolates cannot be regarded as distinct populations. Further characterization by MLST confi rmed that the genetic backbone of bovine and human E. coli O26 strains in Scotland match and that isolates from cattle in Scotland are closely related to highly pathogenic isolates from humans in Germany and Italy.
Characterization of human strains from Scotland and continental Europe and of bovine strains from Scotland showed that most E. coli O26 strains were ST29 and were likely to be capable of expressing a T3S system because they were eae and sepL positive. Although less virulent strains generally are present in cattle in Scotland, the unexpected fi nding of this study was that the genetic background of many of the strains shows a close relationship to strains causing severe human infections in continental Europe.
The use of NMS enables identifi cation of an association between HUS and ST21 strains and confi rms the role of stx 2 alone in severe human disease. As previously suggested, the hlyA gene was not associated with HUS (8). Our results indicate that, in addition to stx 2 , tccP2 is linked to the severity of clinical disease. Therefore, the risk to humans in Scotland from isolates from cattle might be even lower for E. coli O26 if ST21 and the carriage of tccP2 are considered.
In the absence of stx genes, isolates from cattle feces are less of a threat to human health than E. coli O157 strains, but there are E. coli O26 in Scotland of ST21 that are LEE positive and carry tccP2. These need only to acquire the stx 2 -encoding bacteriophage and they could prove to be highly pathogenic because E. coli O26 is a highly dynamic group capable of acquiring stx-encoding phages (34,35).
Zhang et al. (14) described a shift in stx genotype from stx 1 to stx 2 : before 1994, only the stx 1 gene was identifi ed in their STEC O26 isolate collection, but after 1997, stx 2 was the only genotype identifi ed in 71% of the isolates. The trigger for this shift was not established.
At the time of our study, there was no correlation between farms in Scotland with E. coli O157 and E. coli O26 (p = 0.28) or those with E. coli O26 stx 1 + stx 2 + and E. coli O157 (p = 0.10). However, release of free stx-encoding bacteriophages into the intestinal environment (36,37)  greater pathogenic potential for humans, and possession of stx 2 may alter the bovine immune response and lead to longer shedding and to larger quantities of the shed bacterium, resulting in greater exposure to humans. Changes in virulence characteristics of strains may also affect pathogen transmission in cattle (38) and thus pose a greater threat to humans from an increase in bacterial prevalence. We suggest that limited human exposure in Scotland to E. coli O26 with virulence genes explains the recorded difference in the number of human infections attributed to E. coli O26 or E. coli O157. In addition, the more virulent forms of E. coli O26 (stx 2 +eae+) in cattle in Scotland were clustered in the sparsely populated northeast region. However, exposure routes and exposure doses for E. coli O26 and serogroup O157 may differ, although no evidence confi rms this alternative explanation for the human infection rate differences. Although cattle are hypothesized as a potential reservoir of E. coli O26 (18), little is known about the transmission routes to humans and the only evidence linking human infection with cattle is 2 HUS cases in Austria indirectly linked to consumption of unpasteurized milk (20).
The approach to laboratory diagnosis of non-O157 E. coli gastrointestinal infection in humans has changed in recent years. At the time of this study, testing of samples from humans for non-O157 E. coli was conducted by a small number of clinical diagnostic laboratories by using polyvalent antiserum. E. coli O26 isolates were forwarded to the SERL for virulence typing by PCR. SERL identifi ed additional E. coli O26 isolates by testing feces from patients with suspected STEC infection but who had negative routine culture results. However, since 2006-2007, non-O157 STEC detection in submitted fecal samples has been conducted solely by SERL by using PCR followed by culture and characterization of STEC. Since 2006-2007, the reported incidence of E. coli O26 in Scotland has increased from 1.6% of all STEC identifi ed at SERL to 6% in 2010-2011 ( Figure 5). The virulence patterns of E. coli O26 strains isolated from humans in Scotland also appear to have altered. The stx 1 + E. coli O26 isolates previously predominated, but in 2010-2011, >50% of E. coli O26 isolates possessed both stx 1 and stx 2 , and for the fi rst time, 1 isolate from human feces was solely positive for stx 2 .
In 2010, three particularly severe sporadic cases of HUS associated with E. coli O26 in children were reported in Scotland (39). The characteristics of 2 of the E. coli O26 strains included stx 1 and stx 2 (39  was reported in Scotland in a calendar year, reinforcing Bettelheim's (40) warning that we ignore these strains at our peril. Worldwide, reports of human outbreaks of non-O157 EHEC, including E. coli O26, are common (2). In addition, the recent outbreak of E. coli O104:H4 in Germany highlights the potential for Shiga toxin prophage acquisition into different E. coli genetic backgrounds with serious consequences for human health (3).