Multivariate Analyses Revealed Distinctive Features Differentiating Human and Cattle Isolates of Shiga Toxin-Producing Escherichia coli O157 in Japan

Abstract

Genotypes of Shiga toxin-producing Escherichia coli (STEC) O157 isolated from humans and cattle were analyzed by uni- and multivariable logistic regression, and population structure methods, to gain insight into transmission and the nature of human infection. Eleven genotyping assays, including PCR typing of five virulence factors (stx₁, stx₂, stx_2c, eae, and ehxA) and a lineage-specific polymorphism assay using six markers (LSPA6), were considered in the analyses. The prevalence of the stx₁, stx₂, and stx_2c virulence factors was significantly different between human and cattle isolates. However, multivariable regression revealed that the presence of only the stx₂ gene was significantly associated with human isolates after controlling for confounding effects. LSPA6 typing demonstrated an apparent difference in the distribution of LSPA6 lineages between human and cattle isolates and a strong association between stx genotypes and LSPA6 genotypes. Population genetics tools identified three genetically distinct clusters of STEC O157. Each cluster was characterized by stx genotypes and LSPA6 genotypes. The human isolates typically comprised LSPA6 lineage I with stx₁ stx₂ strains and LSPA6 lineage I/II with stx₂ or stx₂ stx_2c strains. In contrast, the cattle isolates comprised LSPA6 lineage II strains with stx_2c or stx₂ stx_2c strains in addition to the clusters identified for the human isolates. Our analyses provide new evidence that the stx₂ gene is the most distinctive feature in human isolates compared to cattle isolates in Japan, and only a subset of the genetically diverse population isolated from cattle is involved in human illnesses. Our results may contribute to international comparisons and risk assessments of STEC O157.

INTRODUCTION

Shiga toxin-producing Escherichia coli (STEC) is one of the most common casual agents for food-borne illnesses worldwide. Of the STEC serotypes, O157 is the most abundant serogroup isolated from patients with food-borne illness. It often causes severe symptoms, such as hemolytic uremic syndrome (HUS) and thrombotic thrombocytopenic purpura (20, 31). The major reservoir of STEC O157 is thought to be cattle, although prevalence estimates for STEC O157 in beef and dairy cattle vary widely (0.2 to 48.8%) (9, 10). It is commonly believed that the pathogen is transmitted to humans through food or direct contact with cattle (7, 20). To assess the risk attributable to the STEC O157 population, it is important to understand the phenotypic and genotypic differences between human and cattle isolates. However, there are few reports that compare the population genetics of human and cattle isolates, and this is due partially to the lack of appropriate typing methods and analytical tools.

There are a number of genotyping methods used for epidemiological studies of STEC O157, such as PCR typing of several virulence factors, lineage-specific polymorphism assay using six markers (LSPA6), restriction fragment length polymorphism, pulsed-field gel electrophoresis, and variable-number tandem repeat analysis (11, 32, 34). Despite their resolution, many of these techniques do not reveal distinctive patterns or associations with human and cattle sources. Among them, stx genotyping and LSPA6 typing have been used to identify genotypic associations with bacterial origins.

The most important virulence factor in STEC O157 is a set of Shiga toxins (Stxs), comprising Stx1 and Stx2. Several variants of Stx2 have been found in STEC; in STEC O157, many of the strains carry Stx2, Stx2c, or both (27). Many previous studies have detected only the generic stx₂ gene (called general-stx₂ in this paper) and do not distinguish between variants of this gene. However, recent interest has focused on the importance of identifying variants of the stx₂ gene, such as the stx_2c gene. In this paper, we refer to the stx₂ gene that does not include all other stx₂ variants as the “stx₂” gene. Some studies reported a much higher prevalence of the stx_2c gene in cattle isolates than in human isolates (21). Other studies suggested that the stx₂ gene is more likely to be linked to human enteritis and cause severe symptoms than are other variants (13, 22). The other PCR-based method, LSPA6 typing, was developed from octamer-based genome scanning and can distinguish the human-biased lineage (LSPA6 lineage I [LI]) from the bovine-biased lineage (LSPA6 lineage II [LII]) (14, 34). Polymorphisms of six genetic loci used for LSPA6 typing would be selectively neutral and are thus more appropriate for describing bacterial divergence.

In this study, our aim was to further our understanding of the attribution of cattle STEC O157 isolates to human infection by using population genetics-based analyses. First, odds ratios of genotypes for human STEC O157 infection were estimated by multivariable logistic regression analysis comparing the distribution of virulence factors in human and cattle isolates while controlling for confounding effects. Five important virulence factors, including the stx₁, stx₂, stx_2c, eae, and ehxA genes, and LSPA6 genotypes were used as genotypic traits throughout our study. Second, the bacterial population structure and any association with isolate origin were evaluated. A number of analyses of population structure have been implemented, using techniques that provide insights that would not be apparent by conventional analyses (16, 23). Our results revealed the most distinctive features that characterize human STEC O157 isolates and showed significant differences between the populations of human and cattle isolates.

MATERIALS AND METHODS

Bacterial strains.

A total of 144 non-sorbitol-fermenting STEC O157 strains were used in this study. They comprised 78 human isolates and 66 cattle isolates. The human isolates comprised 73 isolates from enteritis patients and enteritis-linked isolates and 5 reference strains from the American Type Culture Collection. All of these strains, except the five reference strains, were isolated in various regions of Japan from 1995 to 2009. Until use, all strains were stored at −80°C in Trypticase soya broth (Oxoid Ltd., Hampshire, United Kingdom) with 10% dimethyl sulfoxide (DMSO; Sigma Aldrich, MO) added.

Virulence factor profiling and LSPA6 typing.

For DNA extraction, the strains were grown in 10 ml of Luria-Bertani broth (Becton, Dickinson and Company, NJ) overnight at 37°C. Genomic DNA was then extracted as previously described (1). Five virulence factors, including the stx₁, stx₂, stx_2c, eae, and ehxA genes, were detected by PCR as previously described (28, 29, 33). LSPA6 typing was conducted by the method described by Ziebell et al. (38). The LSPA6 alleles were placed in the following order: folD-sfmA, Z5935 gene, yhcG, rtcB, rbsB, and arp-iclR. LSPA6 genotypes 222222, 221222, 222212, and 221212 were regarded as LII.

Uni- and multivariable logistic regression analyses.

To elucidate the associations between genotypes and origins of STEC O157, uni- and multivariable logistic regression models were constructed using the glm function in R version 2.11.1 (25). The origin of the strains, human or cattle, was considered a binary-outcome variable that was coded as 1 if the isolate was isolated from humans and 0 if it was isolated from cattle. The presence (coded as 1) and absence (coded as 0) of five virulence factors and LSPA6 alleles were used as explanatory variables. For the LSPA6 alleles, the LI-specific allele type was coded as 1, and all other allele types were coded as 0. In multivariable analysis, explanatory variables were selected through stepwise regression based on the Akaike information criterion using the stepAIC function of R.

Population structure analysis.

To investigate the population structure of STEC O157, the 11 locus-specific test results (i.e., the presence or absence of five virulence factors and the six loci from the LSPA6 typing) were analyzed. First, rarefaction analysis was performed using the Analytic Rarefaction program, version 1.3 (http://www.uga.edu/strata/software/). This analysis is used to compare genetic diversity of two populations of different sample sizes by calculating the expected numbers of variants in a range of sample sizes, with confidence limits (8). A steeper slope of the rarefaction curve indicates a higher degree of diversity. Second, the pairwise fixation index (Fst) was estimated using the ARLEQUIN program, version 3.5.1.2 (6), to evaluate statistical evidence for differentiation between the human and cattle populations of STEC O157. Third, to determine whether our samples could be grouped into genetic clusters and to infer the number of clusters that best fit the data, the Bayesian clustering method implemented in the STRUCTURE program (24) was used. This analysis uses departures from the Hardy-Weinberg equilibrium to detect population structure and begins by classifying individuals of unknown origin into a predefined number of populations (K_s). A Bayesian approach is used to infer the value of K that provides the best fit to the data, as measured by the estimated log probability of the data [ln P(D)]. Markov chain Monte Carlo searches consisted of 10,000 “burn-in” steps followed by 100,000 iterations. K values of 1 to 10 were evaluated with 10 replicate runs each, under the admixture model with correlated allele frequencies. A graph of the results was produced using the programs CLUMPP and DISTRUCT (12, 26).

RESULTS

Presence patterns of virulence factors and relation to LSPA6 lineages.

Prevalence estimates for the five virulence factors, including the stx₁, stx₂, stx_2c, eae, and ehxA genes, and the combination of stx virulence factors (stx genotypes) are shown in Table 1. Most of the isolates from humans (97.4%) and cattle (95.5%) were positive for general-stx₂, and this difference was not significant (P = 0.66). However, when the distinction was made between variants of the stx₂ gene, the prevalence of the stx₂ and stx_2c genes was markedly different between human and cattle isolates. The prevalence in human isolates was significantly higher for the stx₂ gene (P < 0.00001) and lower for the stx_2c gene (P < 0.00001) than in cattle isolates. All of the isolates carried the eae gene. Most of the isolates carried the ehxA gene, and there was no significant difference in prevalence between human and cattle isolates. For the combination of stx genes, there was also an apparent difference in the distribution between human and cattle isolates. The prevalence of the stx₁ stx₂ genotype (positive for both stx₁ and stx₂ genes) in human isolates was higher than that in cattle isolates (P < 0.00001), whereas the prevalence of the stx_2c and stx₁ stx_2c genotypes in human isolates was lower than that in cattle isolates (P < 0.00001).

Table 1.

Distribution of five virulence factors and stx types among STEC O157 isolates

Genotype	No. (%) of positive strains from:b		Total no. (%) of positive strains
	Humans	Cattle
Virulence factors
stx1	53 (68.0)c	27 (40.9)c	80 (55.6)
stx2	69 (88.5)c	18 (27.3)c	87 (60.4)
stx2c	18 (23.1)c	49 (74.2)c	67 (46.5)
eae	78 (100)	66 (100)	144 (100)
ehxA	76 (97.4)	64 (97.0)	140 (97.2)
stx genotypesa
stx1	2 (2.6)	3 (4.5)	5 (3.5)
stx2	8 (10.3)	3 (4.5)	11 (7.6)
stx2c	6 (7.7)c	32 (48.5)c	38 (26.4)
stx1stx2	50 (64.1)c	11 (16.7)c	61 (42.4)
stx1stx2c	1 (1.3)c	13 (19.7)c	14 (9.7)
stx2stx2c	11 (14.1)	4 (6.1)	15 (10.4)

^aCombinations of stx genes.

^bNumber (%) of positive strains among the isolates from the same origin.

^cSignificant difference in prevalence between human and cattle isolates (P < 0.001).

LSPA6 genotypes and lineages are shown in Table 2. In human isolates, the predominant LSPA6 lineage was LI, and most of the other isolates belonged to LSPA6 lineage I/II (LI/II). Other isolates, except one isolate of LII, belonged to undefined lineages. These strains differ by only one allele from LI or LI/II strains. In contrast, LSPA6 genotypes were more evenly distributed among cattle isolates. The most predominant lineage was LII (36.7%), but the prevalence of LI/II (33.3%) was similar. A higher prevalence (P < 0.00001) of LI and lower prevalence (P < 0.00001) of LII was observed for human isolates compared to prevalences of cattle isolates.

Table 2.

Distribution of LSPA6 genotypes among STEC O157 isolates

LSPA6 lineage	LSPA6 genotype	No. (%) of isolates
		Origin		stx genotype
		Human	Cattle	stx1	stx2	stx2c	stx1stx2	stx1stx2c	stx2stx2c
I	111111	54 (69.2)a	13 (19.7)a	5 (100)	3 (27.3)	0	59 (96.7)	0	0
I/II	211111	17 (21.8)	22 (33.3)	0	5 (45.5)	16 (42.1)	1 (1.6)	2 (14.3)	15 (100)
II	221212	1 (1.3)	6 (9.1)	0	0	6 (15.8)	0	1 (7.1)	0
	221222	0	2 (3.0)	0	0	0	0	2 (14.3)	0
	222212	0	5 (7.6)	0	0	7 (18.4)	0	0	0
	222222	0	11 (16.7)	0	0	2 (5.3)	0	9 (64.3)	0
Subtotalb		1 (1.3)a	24 (36.4)a	0	0	14 (51.9)	0	12 (44.4)	0
Other	111211	0	1 (1.5)	0	0	0	1 (1.6)	0	0
	212111	1 (1.3)	0	0	0	1 (2.6)	0	0	0
	212211	0	1 (1.5)	0	0	1 (2.6)	0	0	0
	221111	0	1 (1.5)	0	0	1 (2.6)	0	0	0
	221211	0	1 (1.5)	0	0	1 (2.6)	0	0	0
	221213	0	1 (1.5)	0	0	1 (2.6)	0	0	0
	231111	3 (3.9)	0	0	3 (27.3)	0	0	0	0
	241222	0	1 (1.5)	0	0	1 (2.6)	0	0	0
	252211	0	1 (1.5)	0	0	1 (2.6)	0	0	0
Total		77 (100)	90 (100)	5 (100)	11 (100)	38 (100)	2 (100)	14 (100)	15 (100)

^aSignificant difference in prevalence between human and cattle isolates (P < 0.001).

^bSubtotal values represent LSPA6 lineage II results.

When the associations of stx genotypes and LSPA6 lineage were evaluated, they were found to be strongly correlated (Table 2). LI strains carried only the stx₁, stx₂, or stx₁ and stx₂ genes, whereas LII strains did not have the stx₂ gene and comprised only stx_2c and stx₁ stx_2c strains. On the other hand, LI/II strains carried all the stx genotypes studied for this work, with the exception being the stx₁ genotype.

Uni- and multivariable regression analysis for five virulence factors and LSPA6 alleles.

To elucidate the distinctive features differentiating human and cattle isolates, uni- and multivariable logistic regression analyses were performed using a data set of five virulence factors and six loci of LSPA6 alleles. Results of univariable logistic regression suggested strong associations between genotypes and origins of STEC O157 (Table 3). However, when these relationships were adjusted for confounding effects, only the stx₂ variable was significantly related to human isolates (P < 0.00001), and the folD-sfmA variable was weakly associated with human isolates (P = 0.09).

Table 3.

Results of univariate and multivariate logistic regression analysis differentiating human and bovine isolates of STEC O157

Variable	Univariable analysis		Multivariable analysis
	OR (95% CI)a	P value	OR (95% CI)	P value
stx1	3.06 (1.56–6.14)	0.001	NDc
stx2	20.44 (8.83–52.03)	<0.00001	12.70 (4.82–36.76)	<0.00001
stx2c	0.10 (0.05–0.22)	<0.00001	ND
eae	NAb		NA
ehxA	1.19 (0.14–10.12)	0.87	ND
folD-sfmA	8.36 (3.99–18.42)	<0.00001	2.33 (0.86–6.06)	0.09
Z5935	9.41 (3.81–26.94)	<0.00001	ND
yhcG	9.38 (2.98–41.56)	5.80 × 10−4	ND
rtcB	20.83 (6.85–90.99)	<0.00001	ND
rbsB	NA		NA
arp-iclR	16.25 (5.31–71.13)	1.34 × 10−5	ND

^aOR, odds ratio; 95% CI, 95% confidence interval. Odds ratios above 1 indicate that the presence of virulence factors or the LSPA6 lineage I-specific allele is positively associated with human isolates.

^bNA, not available. Odds ratio could not be calculated because all of the strains used in this study carry the eae gene, and none of the bovine isolates showed the LI-specific allele type in the rbsB loci. In the rbsB locus, the prevalence between human and bovine isolates was significantly different by Fisher's exact test (P < 0.00001).

^cND, not determined. These variables were eliminated from the model through backward elimination.

Population structure analysis.

Rarefaction analysis demonstrated that the population of cattle isolates exhibited higher diversity than that of human isolates, as indicated by the steeper slope of the rarefaction curve (Fig. 1). This difference in genetic diversity was further tested by computing the pairwise Fst. The pairwise Fst of the population for human and cattle isolates showed significant (0.35; P < 0.001) differentiation, providing further evidence that these two populations are genetically distinct. For the clustering method implemented in STRUCTURE, a K estimate of 3 populations provided the best fit to our data. A confluent stacked bar plot, when K was equal to 3, is shown in Fig. 2. Typical genotypes for cluster 1 (Fig. 2, shown in red), 2 (green), and 3 (blue) were as follows: LI with the stx₁ stx₂ genotype (cluster 1), LI/II with the stx_2c or stx₂ stx_2c genotype (cluster 2), and LSPA6 genotype 222222 and its close relatives (LII) with the stx_2c or stx₁ stx_2c genotype (cluster 3). Almost all the human isolates were assigned to cluster 1 or 2, whereas the population of the cattle isolates was distributed among all three clusters. Furthermore, when the pairwise Fst was estimated among the same clusters, cluster 1 and 2 showed significant differences between the human and cattle populations (Fst of 0.19 and P value of 0.009 in cluster 1; Fst of 0.08 and P value of 0.045 in cluster 2). When these analyses were conducted with only the isolates from Japan, the statistical significance was the same. An exception was the multivariable logistic regression, in which the yhcG and rbsB variables were selected in the final model, as was the stx₂ gene, but the results were not significant (P > 0.05).

Fig. 1.

Rarefaction curves of Shiga toxin-producing Escherichia coli populations by isolate origin. The y axis represents the number of genotypes from 11 locus-specific tests, including virulence factors and LSPA6 alleles. Broken lines show upper and lower 95% confidence limits.

Fig. 2.

Population structure of Shiga toxin-producing Escherichia coli O157 from humans and cattle. The graph was produced using CLUMPP and DISTRUCT programs. The number of clusters (K) was predefined as three. Each haplotype is represented by a thin vertical line. Confluent stacked bar plots show the probabilities (y axis) that each of the individual 144 STEC O157 isolates (x axis) belongs to each of three Markov chain Monte Carlo (MCMC) model-derived clusters. Each color represents a population, and the color of an individual haplotype represents its proportional membership in the different clusters. Populations are color coded as follows: red, cluster 1; green, cluster 2; blue, cluster 3.

DISCUSSION

In this study, we revealed differences in the populations of STEC O157 genotypes isolated from humans and cattle in Japan using PCR typing of virulence factors and LSPA6 genotyping. Our results demonstrated that the most distinct feature of human isolates was the predominance of the stx₂ gene, whereas other stx virulence factors were less important. In addition, a strong correlation between isolate origin, stx genotype, and LSPA6 genotype was observed. These genotypic patterns were distributed differently between human and bovine isolates. Population structure analyses supported the hypothesis that the population of cattle isolates was more diverse, with only a subset of the population being linked to human disease (2, 7, 17). The importance of the stx₂ gene and the bacterial population structure adds to our understanding of the molecular epidemiology of STEC O157 and may advance global understanding of bacterial population genetics.

Our results clearly showed that the most abundant stx genotype among human isolates was the stx₁ stx₂ genotype, followed by the stx₂ and stx₂ stx_2c genotypes. STEC O157 strains carrying the stx₁ and stx_2c genes or the stx_2c gene were less likely to be involved in human illness, although they would be regarded as general-stx₂ positive. Therefore, future investigations would overestimate the risk associated with cattle isolates if stx₂ and stx_2c genes were not discriminated. However, this distribution of the stx genotypes in Japanese STEC O157 strains differs from those in Germany and Finland (3, 5), in which STEC O157 strains carrying the stx₂ and the stx₂ and stx_2c genes are predominant in clinical isolates.

The distribution of LSPA6 genotypes and their association with stx genotypes indicated a more apparent difference in the distribution of LSPA6 genotypes in Japan than that observed in North America (30, 38). Ziebell et al. (38) also showed a higher prevalence of the stx_2c gene in LII and LI/II strains than in LI strains. Our results are consistent with that study and provide further evidence of strong relationships between LSPA6 genotypes and stx genotypes. LI and LII strains did not carry the stx_2c and stx₂ genes, respectively. On the contrary, LI/II strains carried a wide range of stx virulence factors. This result is concordant with a previous report suggesting that LI/II was an intermediate genotype of LI and LII (36).

The association between these genotypic traits and the source of the isolates was examined by multivariable logistic regression in order to control for confounding effects, a feature that has seldom been considered in previous studies. In the univariable regression, significant differences were observed for all stx virulence factors and LSPA6 alleles between the human and cattle isolates. Surprisingly, however, when these variables were incorporated into a multivariable regression analysis, only the stx₂ variable was significantly associated with the source, due to confounding effects between variables. For example, the apparent significant association between human isolates and the stx₁ gene revealed by univariable analysis was due to the strong correlation between the stx₁ and stx₂ genes, and simple stratification by the stx₂ status reveals no independent association between the stx₁ gene and human isolates. These results provide an interesting insight, because the stx₂ gene is the most dominant stx gene internationally, regardless of the variation in stx genotypes described above (3, 5). Other epidemiological studies showed that STEC O157 strains with the stx₂ gene are more likely to be associated with severe symptoms in humans (5, 13). It is plausible that the stx₂ gene, identified as the most distinct feature in human isolates in this study, plays the most important role in disease development as well as transmission and human infection.

Population genetics approaches clarified some characteristics of the STEC O157 population. Rarefaction analysis and pairwise Fst showed that the bacterial population of cattle isolates was more diverse and different in structure than that of the human isolates. The Bayesian clustering method identified three distinct clusters (Fig. 2). Each cluster was characterized by certain LSPA6 genotypes and stx genotypes, providing further evidence for the strong correlation between LSPA6 and stx genotypes. In addition to LSPA6 typing, more robust typing methods, such as comparative genomic fingerprinting and microarray comparative genomic hybridization, have previously supported three genetically distinct populations in STEC O157 (14, 15, 38). Furthermore, pairwise Fst tests indicate the possibility that clusters 1 and 2 may be further differentiated by isolate origin, although the small sample size made it difficult to draw firm conclusions. Therefore, significant differences between the human and cattle populations might become more apparent if other genetic markers are implemented in the model.

From a public health perspective, it is of interest that the human and cattle populations shared only clusters 1 and 2 and that cluster 3 was a rare genotype in human isolates. Phenotypic differences between these clusters may be a target for better control of STEC O157. The difference may be explained by recent reports describing the importance of genotypic and phenotypic heterogeneity of STEC O157 for virulence and adaptation to the host environment. Lowe et al. (18) and Zhang et al. (37) showed higher levels of adherence to intestinal cells and Shiga toxin production in LI strains than in LII strains. For stx genotypes, many researchers have found a correlation between the stx₂ gene and disease severity in humans (3, 5, 19). As our results suggested, if both stx genotypes and LSPA6 genotypes are taken into consideration, more informative observations can be made. The other phenotypic traits that may explain the genotypic difference include stress resistance; a microarray analysis of LI and LII strains revealed that there were differential expressions of several genes related to stress resistance, including heat or cold shock protein genes (4). In addition to investigations of various phenotypes, information on isolates from food and asymptomatic human carriers could explain how and where the selective pressure works. As Nakamura et al. (21) and Yokoyama et al. (35) reported, isolates from asymptomatic human carriers seemed to exhibit stx genotypes that were intermediate between human and cattle isolates. Further studies are required to confirm the validity of this observation.

In conclusion, our analyses of genotypes in Japanese STEC O157 strains demonstrated that (i) the stx₂ gene is the most distinctive feature of human isolates compared to cattle isolates, (ii) the stx genotype and LSPA6 lineage are strongly correlated, and (iii) STEC O157 consists of three major genetically distinct populations, and their distribution differed significantly between human and cattle isolates. Based on those results, we suggest a hypothetical model of association between genotypes and origins of STEC O157 (Fig. 3). This model shows three genetically distinct subpopulations of STEC O157. Each subpopulation is characterized mainly by certain stx genotypes and LSPA6 genotypes. The population in the cattle reservoir contains all the various genotypes. However, only a subset of the population is involved in human illness, which explains the different composition of genotypes in humans and cattle. STEC isolates from a broader range of host and geographic regions would be needed to confirm and complement the hypothetical model.

Fig. 3.

Hypothetical model of association between genotypes and origins of Shiga toxin-producing Escherichia coli O157.