Variation in Stress Resistance Patterns among stx Genotypes and Genetic Lineages of Shiga Toxin-Producing Escherichia coli O157

Abstract

To evaluate the relationship between bacterial genotypes and stress resistance patterns, we exposed 57 strains of Shiga toxin-producing Escherichia coli (STEC) O157 to acid, freeze-thaw, heat, osmotic, oxidative, and starvation stresses. Inactivation rates were calculated in each assay and subjected to univariate and multivariate analyses, including principal component analysis (PCA) and cluster analysis. The stx genotype was determined for each strain as was the lineage-specific polymorphism assay (LSPA6) genotype. In univariate analyses, strains of the stx₁ stx₂ genotype showed greater resistance to heat than strains of the stx₁ stx_2c genotype; moreover, strains of the stx₁ stx₂ genotype showed greater resistance to starvation than strains of the stx₂ or stx_2c genotypes. LSPA6 lineage I (LI) strains showed greater resistance to heat and starvation than LSPA6 lineage II (LII) strains. PCA revealed a general trend that a strain with greater resistance to one type of stress tended to have greater resistance to other types of stresses. In cluster analysis, STEC O157 strains were grouped into stress-resistant, stress-sensitive, and intermediate clusters. In stx genotypes, all strains of the stx₁ stx₂ genotype were grouped with the stress-resistant cluster, whereas 72.7% (8/11) of strains of the stx₁ stx_2c genotype grouped with the stress-sensitive cluster. In LI strains, 77.8% (14/18) of the strains were grouped with the stress-resistant cluster, whereas 64.7% (11/17) of LII strains were grouped with the stress-sensitive cluster. These results indicate that the genotypes of STEC O157 that are frequently associated with human illness, i.e., LI or the stx₁ stx₂ genotype, have greater multiple stress resistance than do strains of other genotypes.

INTRODUCTION

Shiga toxin-producing Escherichia coli (STEC) strains are some of the most common food-borne pathogens worldwide. Among various STEC serotypes, O157 is the most prevalent serogroup in food-borne infections, and STEC O157 infections have often been associated with severe conditions, such as hemolytic uremic syndrome (HUS) (26, 38). The low infectious dose and serious nature of STEC O157 has stimulated interest in the determinants of the survival of the organism in food and the environment. STEC O157 may encounter various environmental stresses, including nutrient depletion in the environment and changes in temperature and osmotic pressure. In the human stomach, STEC O157 must survive the low pH of gastric fluid. In order to understand the ecology of STEC O157 and establish effective control measures against the bacterium, extensive research has been done on bacterial stress responses (5, 24). Although some of these studies have shown the importance of cross-protection in stress resistance (5, 20, 33), the association between more than three types of stress has seldom been investigated. In practice, multiple stress resistance of a pathogen is important, because control measures in the food supply industry are based on the concept of employing multiple hurdles to decrease pathogen survival (14).

Recently, several studies have suggested that the bacterial populations of STEC O157 isolated from humans and cattle are genetically distinct (4, 19, 21, 45). Zhang et al. (44) and Abu-Ali et al. (1) reported that Shiga toxin production and the ability to adhere to intestinal cells are related to the genetic lineages of STEC O157. Stress resistance is an additional phenotype that may be related to genotype. Microarray and quantitative reverse transcriptase PCR analyses of STEC O157 strains have shown that several genes involved in the stress response are differentially expressed between human- and bovine-biased genotypes (7, 39). However, the difference in stress resistance among genotypes has been shown with respect to acid stress (25, 28, 34, 39). In other bacterial species, multivariate analyses, including principal component analysis (PCA) and cluster analysis, have revealed correlations between stress resistance, metabolic patterns, and their environmental niches (16, 29, 35). These methods would be useful to elucidate associations between genotype and several types of stress resistance phenotypes in STEC O157.

In this study, patterns of stress response in STEC O157 isolates were investigated in six different assays: acid, heat, freeze-thaw, high osmotic pressure, oxidative stress, and starvation. These stresses were selected because STEC O157 often encounters these stresses in food and host environments (5, 36). The relationships between stress resistance patterns and genotypes were further analyzed by univariate and multivariate methods. Genotypic traits were characterized by stx typing and a lineage-specific polymorphism assay with 6 markers (LSPA6), because these genotypic traits are known to reflect the divergence of STEC O157 (18, 19, 42). In this study, our objective was to examine the differences in stress resistance patterns among these genotypes of STEC O157.

MATERIALS AND METHODS

Bacterial strains.

A total of 57 STEC O157 strains, including 27 human isolates and 30 cattle isolates, were used in this study (Table 1). The human isolates comprised 23 isolates from enteritis patients and enteritis-linked isolates and four reference strains obtained from the American Type Culture Collection. All of these strains, except the four reference strains, were isolated in Japan between 1995 and 2009. The presence of stx₁, stx₂, stx_2c, and LSPA6 lineage has been determined previously (19). The LSPA genotype was determined using octamer-based genome scanning, which uses short sequences from the E. coli genome for genotyping (17). Therefore, polymorphisms in these six loci for LSPA6 are regarded as selectively neutral (42). LSPA6 lineages are associated with stx genotypes and the genetic clades defined by single nucleotide polymorphisms (SNP clades). In previous studies, LSPA6 lineage I (LI) strains were more likely to carry stx₂ than LSPA lineage II (LII) strains (19, 43, 45). Only limited information is available about the association between LSPA6 lineage and SNP clade; however, the hypervirulent SNP clade, clade 8, belongs to LSPA6 lineage I/II (LI/II) (12, 18).

Table 1.

Genotypic characteristic and IRI values of the strains used in this study

Strain	Source	stx genotype	LSPA6 genotypea	LSPA6 lineagea	IRI valueb (stress)
					Acid	Freeze-thaw	Heat	Osmotic	Oxidative	Starvation
ESC226	Cattle	stx1	111111	I	−0.98	−2.29	0.06	−2.54	0.75	−4.70
ESC343	Cattle	stx1	111111	I	−0.91	−3.07	0.37	−3.94	0.09	−3.12
ATCC35150	Human	stx1	111111	I	−1.12	−3.19	−0.20	−3.35	−0.27	−4.80
ATCC43890	Human	stx1	111111	I	0.56	−3.15	0.58	−2.48	−0.06	−3.09
EC55	Human	stx2	111111	I	0.52	−3.03	0.49	−3.22	−0.33	−3.13
ESC361	Human	stx2	111111	I	−0.96	−3.23	0.15	−5.30	−0.25	−3.10
EC10	Human	stx1stx2	111111	I	−0.57	−3.30	−0.03	−3.65	0.11	−4.68
EC18	Human	stx1stx2	111111	I	−0.93	−3.11	0.34	−2.31	0.60	−5.95
EC28	Human	stx1stx2	111111	I	−0.97	−3.05	0.80	−2.87	0.98	−4.13
EC32	Human	stx1stx2	111111	I	−0.82	−2.84	0.87	−2.87	0.53	−4.03
EC33	Human	stx1stx2	111111	I	−0.83	−2.42	0.20	−2.97	0.92	−4.91
EC43	Human	stx1stx2	111111	I	−1.04	−3.05	−0.71	−3.79	−0.01	−5.49
EC52	Human	stx1stx2	111111	I	−0.89	−3.11	−0.33	−3.35	0.13	−4.92
EC70	Human	stx1stx2	111111	I	−1.84	−3.17	0.30	−3.65	−0.09	−4.40
EC170	Cattle	stx1stx2	111111	I	−1.08	−3.24	−0.43	−3.46	0.12	−4.57
ESC228	Cattle	stx1stx2	111111	I	−0.82	−3.17	−0.12	−2.40	0.88	−4.22
ESC342	Cattle	stx1stx2	111111	I	−1.11	−3.06	0.29	−3.51	−0.05	−4.61
ATCC43895	Human	stx1stx2	111111	I	−1.22	−3.41	−1.20	−3.71	−0.21	−5.01
EC160	Cattle	stx2	211111	I/II	−0.51	−2.83	0.13	−3.64	−0.40	−3.11
ESC225	Cattle	stx2	211111	I/II	−0.92	−2.49	−0.44	−2.04	1.00	−3.77
ESC344	Cattle	stx2	211111	I/II	−1.01	−3.18	−0.36	−3.89	0.32	−5.34
ESC349	Human	stx2	211111	I/II	−1.21	−3.10	−0.26	−3.53	0.18	−4.53
ESC360	Human	stx2	211111	I/II	−0.83	−2.47	0.89	−2.51	0.74	−3.03
EC59	Human	stx2c	211111	I/II	−0.93	−3.00	−0.01	−2.32	0.80	−4.13
EC157	Cattle	stx2c	211111	I/II	−1.09	−3.17	−0.89	−3.61	−0.23	−5.09
ESC206	Cattle	stx2c	211111	I/II	−0.38	−2.75	−0.44	−2.87	0.80	−3.96
ESC219	Cattle	stx2c	211111	I/II	−0.10	−3.28	0.88	−1.96	0.79	−3.84
EC66	Human	stx1stx2c	211111	I/II	−0.93	−2.11	0.16	−2.19	0.93	−4.14
ESC138	Cattle	stx1stx2c	211111	I/II	−1.26	−2.89	0.92	−2.39	0.75	−4.14
EC1	Human	stx1stx2c	211111	I/II	−1.01	−2.95	0.49	−2.32	0.71	−4.18
EC42	Human	stx1stx2c	211111	I/II	−1.20	−3.00	0.39	−2.61	0.57	−4.16
ESC362	Human	stx1stx2c	211111	I/II	−1.12	−2.66	0.00	−2.33	0.55	−3.94
ATCC43889	Human	stx1stx2c	211111	I/II	−0.77	−2.85	0.94	−2.84	1.21	−3.70
ESC356	Human	stx2c	221212	II	−1.05	−2.92	0.11	−3.03	0.31	−4.35
EC37	Human	stx2c	222212	II	−0.23	−2.97	0.68	−1.94	0.38	−3.53
EC38	Human	stx2c	222212	II	−0.60	−2.70	0.99	−2.63	1.00	−3.16
EC169	Cattle	stx2c	222212	II	−0.83	−2.80	0.37	−3.87	−0.49	−3.06
EC183	Cattle	stx2c	222212	II	−0.91	−2.99	0.61	−4.24	−0.43	−3.03
ESC209	Cattle	stx2c	222212	II	−0.78	−3.08	0.32	−2.31	1.35	−4.07
ESC340	Cattle	stx2c	222212	II	−0.84	−2.87	0.70	−2.83	0.81	−3.04
ESC211	Cattle	stx2c	222222	II	−0.92	−2.88	0.48	−3.26	0.38	−3.09
EC181	Cattle	stx1stx2c	221222	II	−1.11	−3.09	−0.33	−3.71	−0.35	−4.72
EC164	Cattle	stx1stx2c	222222	II	−1.28	−3.26	−0.41	−3.47	−0.33	−5.04
ESC213	Cattle	stx1stx2c	222222	II	−0.34	−2.56	1.45	−1.77	1.75	−4.02
ESC214	Cattle	stx1stx2c	222222	II	−0.86	−2.65	1.34	−2.38	1.06	−3.02
ESC215	Cattle	stx1stx2c	222222	II	−0.93	−2.94	1.04	−2.45	0.99	−3.07
ESC216	Cattle	stx1stx2c	222222	II	−0.51	−2.69	1.83	−1.79	0.93	−4.32
ESC220	Cattle	stx1stx2c	222222	II	−0.65	−2.83	0.89	−1.91	0.55	−3.37
ESC222	Cattle	stx1stx2c	222222	II	−0.72	−2.78	0.99	−2.25	1.13	−4.17
ESC229	Cattle	stx1stx2c	222222	II	−0.67	−2.97	1.30	−2.55	1.18	−3.05
EC44	Human	stx2	231111	Other	−0.68	−2.65	1.14	−2.49	0.85	−3.07
EC45	Human	stx2	231111	Other	−0.70	−2.06	0.09	−2.35	1.07	−4.37
EC175	Human	stx2c	212111	Other	−1.06	−3.08	0.41	−4.36	−0.57	−3.35
ESC223	Cattle	stx2c	212211	Other	−0.91	−1.38	0.86	−3.07	0.58	−3.07
ESC231	Cattle	stx2c	221111	Other	−0.97	−2.90	0.08	−2.58	0.94	−4.83
ESC367	Cattle	stx2c	221211	Other	−0.91	−3.23	0.35	−2.57	0.77	−4.91
ESC339	Cattle	stx2c	252211	Other	−1.09	−3.12	1.09	−2.85	1.09	−4.36

^aThe LSPA6 alleles were placed in the following order: folD sfmA, Z5935 gene, yhcG, rtcB, rbsB, and arp-iclR. LSPA6 genotypes of 222222, 221222, 222212, and 221212 were regarded as LII.

^bThese values are calculated from inactivation rate. Because of log transformation, some values are positive, whereas the others are negative. Smaller IRI values correspond to greater stress resistance.

Stress resistance assays.

Strains were grown in tryptic soy broth (TSB; Becton, Dickinson and Company, NJ) at 37°C (Incubator MIR-262; Sanyo, Osaka, Japan) for 20 h. For acid stress, the culture of each strain was diluted 100-fold into 10 ml of minimal E glucose medium (EG medium) (41) acidified with hydrogen chloride (Kanto Chemical Co., Inc., Tokyo, Japan) to pH 2.5. The broth was incubated at 37°C, and viable cell counts were performed at 0, 2, 6, 8, and 12 h. To enumerate the viable cells, the inoculated broth was serially diluted with phosphate-buffered saline (PBS; Nissui Pharmaceutical Co., Ltd., Tokyo, Japan) and pour-plated onto tryptone soya agar (TSA; Oxoid Ltd., Hampshire, United Kingdom). All plates were incubated at 37°C, and colonies were counted after 48 h. For heat stress, the culture of each strain was diluted 100-fold in 1.5 ml of TSB, and the broth was submerged in a water bath (Thermominder EX; Taitec Cooperation, Saitama, Japan) at 52°C. Viable cell counts were performed at 0, 2, 4, and 6 h as described above. For freeze-thaw stress, the culture of each strain was diluted 100-fold in 10 ml of TSB, and the broth was subsequently frozen at −20°C (Medicool MPR-411 FRS; Sanyo). One freeze-thaw cycle consisted of freezing at −20°C for 22.5 h followed by thawing at 37°C for 1.5 h. Freeze-thaw cycles were repeated seven times for each sample. Viable cell counts were performed at days 0, 1, 2, 4, and 7 as described above. For high osmotic pressure stress, the culture of each strain was diluted 100-fold in 10 ml of a solution of 20% (wt/vol) NaCl (Wako Pure Chemical Industries Ltd., Tokyo, Japan). The solution was incubated at 37°C, and viable cell counts were performed at days 0, 1, 2, 4, and 7 as described above. For oxidative stress, the culture of each strain was diluted 100-fold in 10 ml of PBS containing 1 mM H₂O₂ (Wako). The solution was incubated at 37°C, and viable cell counts were performed at 0, 2, 4, 6, 8, and 12 h as described above. For starvation stress, cells were harvested by centrifugation at 4,000 × g for 20 min and washed twice with 10 ml of distilled water (DW) (Nihon Millipore Ltd., Tokyo, Japan). Thereafter, the cells were resuspended in 10 ml of DW at 10³ CFU/ml. The suspension was then incubated at 25°C. Viable cell counts were performed at days 0, 7, 14, 21, and 28 as described above. DW was used, because the population of STEC decreases very slowly in other nutrient-depleted media, such as phosphate-buffered saline (11), and it is not appropriate to measure an inactivation rate. Therefore, this assay evaluated the bacterial response against starvation in the environment as well as low osmotic pressure.

Calculation and validation of inactivation rate index.

We used the raw count data to construct an inactivation rate index (IRI) representing, for each strain, the rate of reduction in the bacterial count caused by each stress assay. Given an initial concentration (C₀) of bacteria in an aliquot, we assumed the subsequent concentration undergoes a constant rate of log reduction, so that at sampling time t_i the log concentration is log C_i = log C₀ − αt_i, where α represents the rate of inactivation or hazard rate and i is 1, 2, 3, etc. Even if the inactivation curve is not exponential, α is still a meaningful phenotypic measure, because it represents the average rate of inactivation. We are not attempting here to produce an accurate inactivation model so have avoided multiparameter models such as the Weibull (30). The observed count at sampling time t_i is obtained from a plated volume D_i (the volume of dilution), and thus the expected count is μ_i = D_iC_i. We assume then that the observed count has a Poisson distribution with mean μ_i, for which log μ_i = log D_iC_i = log D_i + log C_i = log D_i + log C₀ − αt_i.

Thus, α may be estimated as the negative slope in a Poisson regression model with log D_i as an offset. This was fitted using the glm function of R 2.11.2 (32) to obtain α values for each stress-strain combination, with smaller IRI values corresponding to greater stress resistance. Graphical inspection of α values showed that they were highly skewed, and thus they were log transformed before they were submitted for further analysis to give a reasonably symmetrical, approximately normal, distribution. This transformation can be further justified, in a wider context, by noting that modeling of the log hazard is a common approach in survival analysis (6), corresponding to an assumption of proportional hazards. Thus, the IRI used in this study was log α, the log of the average inactivation rate.

Univariate analyses of different STEC O157 genotypes and their resistance to stress.

The IRI values of stx genotypes or LSPA6 lineages were compared using multiple Student's t tests with Tukey's adjusted P value.

Multivariate analyses of IRI values in all stress resistance assays.

To gain insight into patterns of stress resistance, IRI values from all of the six stress resistance assays were analyzed by PCA and cluster analysis. PCA is useful for identifying a trend in a multivariate data set or a correlation between variables (40). The transformation of PCA consolidates the information of a data set into a few new variables, principal components (PCs). PCAs were computed from the following equation: Z_i = a_iacidx_acid + a_ifreezex_freeze + a_iheatx_heat + a_iosmoticx_osmotic + a_ioxidativex_oxidative + a_istarvationx_starvation, where Z_i refers to a PC score in one strain in the ith PC, a refers to a factor loading, and x refers to the IRI value of the strain (Table 1) in a stress resistance assay. The a values are calculated from the correlation matrix of the observed data. First, a₁ values are computed to maximize the variance of Z₁, constrained so that the sum of the squares of factor loadings is one. This is termed the first PC, and the maximized variance the first eigenvalue. Then, a₂ is calculated to maximize the variance of Z₂ subject to orthogonality to Z₁. The computation can be continued until the sixth (the number of the variables) PC is calculated. However, PCA is usually stopped when the cumulative variance reaches 70% or when the next eigenvalue becomes less than 1.0 (13, 27). Factor loadings for each variable and PC scores for each PC have a specific interpretation. In our data set, PCs described stress resistance patterns of STEC O157. For each PC, factor loadings of each variable represent the contribution of the variable to that particular pattern. On the other hand, PC scores represent relative values of each strain among all of the strains. Therefore, the interpretations for PC scores are made according to the loadings in each PC. PCA is also useful for graphic representation. In a PCA biplot, each strain is plotted according to the first and second PC scores to show the largest variation in the data set. In the biplot, strains that have similar stress resistance patterns will be closer to each other. PCA was performed using the princomp function of R with default settings.

To see a general trend of multiple stress resistance, cluster analysis was performed using the partitioning-around-medoids (PAM) clustering method (15) in the library “cluster” in R with default settings. This method was used to inform grouping of STEC O157 strains on the basis of similarities in their IRI values for the six stress resistance assays. The number of clusters was evaluated from 2 to 10 by the silhouette width (15) and visual inspection of PCA biplots. The stress resistance pattern of each cluster was characterized by mean IRI values, and mean IRI values were compared using multiple Student's t tests with adjusted P values. From these results, the number of clusters for subsequent analysis was decided, where the majority of strains can be defined by the mean IRI values (e.g., stress resistant, susceptible, etc.). Associations between resistance patterns and stx genotypes or LSPA6 lineages were explored by the distribution of the IRI-based clusters among genotypes.

RESULTS

Calculation and validation of IRI values.

IRI values were obtained from the results of the six stress resistance assays (Table 1). Although IRI values were calculated from the inactivation rate, positive and negative values resulted from subsequent log transformation. These transformed IRI values gave an approximately normal distribution and thus were appropriate for multivariate analyses. The range of IRI values varied among the assays; therefore, to exclude the effect of the ranges of IRI values among the assays, IRI values were standardized before multivariate analyses. For standardization, within each assay, the mean was subtracted from each IRI value and then divided by the standard deviation (31).

Univariate analyses of STEC O157 genotypes and their stress resistance.

Associations between the source or genotype of STEC O157 and the IRI values from each stress resistance assay were evaluated. No significant differences in IRI values between human and cattle isolates of STEC O157 were observed. However, in comparisons between IRI values and genotypes, significant associations were observed in the heat and starvation assays. Strains of the stx₁ stx₂ genotype (strains with both stx₁ and stx₂ genes) showed significantly greater resistance to heat than strains of the stx₁ stx_2c genotype (P = 0.011). Strains of the stx₁ stx₂ genotype also showed greater resistance to the starvation assay than strains of the stx₂ or stx_2c genotypes (stx₁ stx₂ strains versus stx₂ strains, P = 0.019; stx₁ stx₂ strains versus stx_2c strains, P = 0.011). LI and LI/II strains showed significantly greater resistance to heat than LII strains (LI strains versus LII strains, P = 0.005; LI/II strains versus LII strains, P = 0.015). In the starvation assay, LI strains showed greater resistance than LII strains (P = 0.024).

Multivariate analyses of IRI values in all stress resistance assays.

All IRI values in the six stress resistance assays were included in the multivariate analyses. First, in PCA, the first and second PCs accounted for 45.8% and 22.1%, respectively, of the variance within the multivariate data set. Because most of the variation (67.9%) was contained in the first and second PC and no other PCs possessed eigenvalues of >1.0, we focused on these PCs. All of the factor loadings on the first PC were negative values, indicating that the first PC described multiple stress resistance, with higher scores associated with greater resistance to the six stresses, and lower scores associated with susceptibility to them (Table 2). On the other hand, higher scores in the second PC were associated with resistance to the acid, heat, and starvation stresses and susceptibility to the osmotic, oxidative, and freeze-thaw stresses. Pairwise comparisons of PC scores in the first PC and stx genotypes or LSPA6 lineages showed higher scores in the strains of the stx₁ stx₂ genotype (P = 0.0099) and LI strains (P = 0.0088) than in the strains of the stx₁ stx_2c genotype and LII strains, respectively (Table 3). This result indicates the general trend of greater multiple stress resistance of the strains of the stx₁ stx₂ genotype and LI strains than those of the stx₁ stx_2c genotype and LII strains, respectively. In PC scores in the second PC, there was no significant difference among genotypes.

Table 2.

Principal component analysis of inactivation rate index values in six stress resistance assays

Variable	Factor loadinga in:
	1st PC	2nd PC
Acid	−0.30	−0.43
Freeze-thaw	−0.36	0.12
Heat	−0.48	−0.21
Osmotic	−0.48	0.37
Oxidative	−0.47	0.44
Starvation	−0.31	−0.65
Proportion of variance	0.46	0.22
Cumulative proportion	0.46	0.68

^aFactor loading indicates the relative weight of the individual stress resistance assay on each principal component (PC).

Table 3.

Differences in PC scores between stx genotypes and LSPA6 lineage

Genotype	PC score (mean ± SD) in:
	1st PCa	2nd PCb
stx genotype
stx1	0.33 ± 1.51	−0.67 ± 1.87
stx2	0.07 ± 1.78	−0.59 ± 1.67
stx2c	−0.14 ± 1.38	−0.25 ± 1.16
stx1stx2	1.21 ± 1.37c	0.59 ± 0.66
stx1stx2c	−1.11 ± 1.98c	0.26 ± 0.62
stx2stx2c	−0.49 ± 0.64	0.59 ± 0.34
LSPA6 lineage
I	0.99 ± 1.41d	−0.07 ± 1.49
I/II	−0.06 ± 1.44	0.34 ± 0.87
II	−0.75 ± 1.79d	−0.32 ± 0.92
Other	−0.60 ± 1.46	0.23 ± 1.28

^aHigher PC scores indicate greater resistance to the six stresses.

^bHigher PC scores indicate greater resistance to acid, heat, and starvation stresses and lesser resistance to freeze-thaw, high osmotic pressure, and oxidataive stresses.

^cSignificant (P = 0.0099) difference in mean PC scores in a pairwise t test with adjusted P values.

^dSignificant (P = 0.0088) difference in mean PC scores in a pairwise t test with adjusted P values.

In cluster analysis, the average silhouette width was the largest when the number of clusters was six (see Fig. S1 in the supplemental material); however, the number was not decisive, because its standard deviation was large. According to the comparison of IRI values among clusters, most of the strains (>85%) were grouped to stress-resistant or -sensitive clusters when the numbers of clusters were two and three (see Table S1 in the supplemental material). When the number of clusters was two, each strain was grouped essentially according to its PC score in the first PC (see Fig. S2 in the supplemental material). Therefore, in order to describe the diversity of stress resistance pattern, three clusters were considered appropriate for the analysis. When the number of clusters was predefined as three, 31, 7, and 19 strains were grouped into clusters 1, 2, and 3, respectively (Fig. 1). To characterize stress resistance patterns, mean IRI values were compared among these clusters (Table 4). The mean IRI value of cluster 1 was significantly lower than that of cluster 3 in all of the six assays. It suggested that cluster 1 is a group relatively resistant to the stresses in this study and that cluster 3 is a stress-susceptible group. Cluster 2 can be characterized as a group of intermediate or variable stress resistance. The mean IRI values for cluster 2 in the acid, freeze-thaw, and heat stress assays were between those for clusters 1 and 3. In the osmotic and oxidative stress assays, strains in cluster 2 showed the greatest stress resistance; however, they showed the most susceptibility to starvation stress (Table 4). Among these clusters, genotypes of STEC O157 were not distributed evenly (Fig. 2). All strains of the stx₁ stx₂ genotype were grouped to cluster 1, whereas most of the strains (72.7%) of the stx₁ stx_2c genotype were grouped to cluster 3. Three of four strains (75.0%) of the stx₂ stx_2c genotype were grouped to cluster 1. The difference in distribution of clusters was more apparent for the LSPA6 lineage. The most common cluster for LI strains was cluster 1; on the other hand, LII was the most common in cluster 3. LI/II strains showed an intermediate pattern between LI and LII strains. In the strains carrying only one type of stx and strains of atypical LSPA6 lineage, none of the clusters held the majority.

Fig 1.

The results of cluster analysis in biplots by principal component analysis (PCA). Distances between plots represent the similarities between strains. Each stx genotype (A) and LSPA6 lineage (B) of STEC O157 corresponds to a color, as shown. Plots located inside the circle belong to each cluster: red, cluster 1; green, cluster 2; blue, cluster 3. The distance between plots shows the similarity of their stress resistance patterns. PCA shows that higher values on the x axis indicate greater multiple stress resistance to six stresses and higher values on the y axis indicate greater resistance to acid, heat, and starvation stresses and lower resistance to freeze-thaw, osmotic pressure, and oxidative stresses.

Table 4.

Difference in IRI values between clusters in each stress resistance assay

Cluster	Mean ± SD of IRI values (stress)a
	Acid	Freeze-thaw	Heat	Osmotic	Oxidative	Starvation
1	−1.02 ± 0.24 AB	−3.00 ± 0.26 B	0.02 ± 0.53 B	−3.03 ± 0.56 B	0.41 ± 0.45 B	−4.55 ± 0.57 BC
2	−0.67 ± 0.55 A	−3.00 ± 0.15	0.36 ± 0.17 A	−4.08 ± 0.66 B	−0.34 ± 0.22 B	−3.13 ± 0.10 B
3	−0.61 ± 0.37 B	−2.68 ± 0.44 B	0.90 ± 0.43 AB	−2.35 ± 0.36 B	0.91 ± 0.39 B	−3.53 ± 0.51 C

^aA, values with the same letter designation in each assay have a statistically significant difference (P < 0.05) in mean IRI values in a pairwise t test with adjusted P values; B and C, values with the same letter designation in each assay have a statistically significant difference (P < 0.01) in mean IRI values in a pairwise t test with adjusted P values.

Fig 2.

Associations between clusters and stx genotypes (A) and LSPA6 lineages (B) of STEC O157. Cluster 1 and cluster 3 were characterized as stress-resistant and stress-susceptible groups, respectively. Stress resistances of strains in cluster 2 varied depending on the stress.

DISCUSSION

In this study, we showed an association between genotypes of STEC O157 and stress resistance in six different assays by using univariate and multivariate analyses. Because only a subset of the genotypes is likely to be involved in human infection (9, 19), our results provide an insight into how selective pressures may affect the transmission of STEC O157 from cattle to humans via food and the environment.

Significant associations were found between genotypic traits of STEC O157 and their stress responses, although no significant associations were observed between the strain sources and their stress responses. Because human and cattle isolates of STEC O157 have been reported to share a subpopulation (9, 18, 19), it is potentially more revealing to examine the distribution of phenotypes of the bacterium among the genotypes rather than the sources (19). When IRI values were compared among stx genotypes or LSPA6 lineages, significant differences were observed in the heat and starvation stresses. Interestingly, the stx₁ stx₂ genotype and LI and LI/II strains, which showed greater resistance to these stresses, have been shown to be more frequently associated with human infection (19, 43, 45). This suggests that responses of STEC O157 to these two stresses play more important roles in their ecology than other stresses do. In E. coli, heat and starvation stresses induce general stress responses and affect various components of the bacterial cell via heat shock protein and rpoS regulons, respectively (5, 8, 37). The broad spectrum of effects on bacterial metabolism and cell components might be reflected by these results; however, rpoS sequences of the strains used in this study were highly homogeneous, and there was no apparent correlation when this was investigated by uni- and multivariate analyses (see Fig. S3 in the supplemental material). It is likely that variations in other stress-related genes are responsible for the diversity in the stress response among strains.

To characterize stress resistance patterns of the strains, all IRI values were subjected to multivariate analyses, including PCA and cluster analysis. These analyses revealed the influence of all six stresses, including four stresses that did not yield significant results in the univariate analyses. In PCA, the factor loading values for all of the six assays in the first PC were in the same direction (negative values). This finding implies that the largest contribution of variation to the stress resistance pattern is in multiple stress resistance, that is, a general trend that an STEC O157 strain that is resistant to one stress is also resistant to other stresses. This finding is in agreement with the results of Benito et al. (2), who showed a similar trend in resistance to hydrostatic pressure, mild-heat, acid, oxidative, and osmotic stresses for E. coli O157 strains.

In cluster analysis, STEC O157 strains were grouped into three clusters according to their stress resistance patterns. These clusters were characterized by mean IRI values (Table 4). Multiple comparisons elucidated that strains in cluster 1 showed greater resistance in all of the six assays than strains in cluster 3 did. These results corroborate the first PC in PCA, which showed that the most discriminative factor is multiple stress resistance. Interestingly, the clusters were not distributed evenly among genotypes of STEC O157. This suggests that the stress resistance patterns are associated with the phylogeny of STEC O157 to some extent. In contrast to the predominance of cluster 1 in LI strains, LII strains were predominantly comprised of cluster 3. LI/II strains showed an intermediate distribution among the clusters (Fig. 2), consistent with the intermediate genetic properties in this lineage (18, 45). Similar trends were observed in strains carrying two different types of stx, i.e., stx₁ stx₂, stx₁ stx_2c, and stx₂ stx_2c genotypes. This could be attributable to the strong association between the stx genotype and the LSPA6 lineage (12, 19, 43, 45). In a previous study, most of the strains (>85%) of the stx₁ stx₂, stx₂ stx_2c, and stx₁ stx_2c genotypes were shown to belong to LI, LI/II, and LII, respectively (19). On the other hand, strains carrying only one type of stx showed a more diverse stress resistance pattern. This result is in concordance with the result of a study that showed that these strains are genetically more diverse than strains carrying two types of stx (19). The second PC showed that the stress responses to acid, heat, and starvation stress tend to be similar to each other and likewise osmotic, oxidative, and freeze-thaw stresses. Although we cannot infer whether or not this is the result of overlapping genetic responses, these stress responses may have a similar role in the ecology of STEC O157.

An important implication of the clustering result is that human-biased genotypes, such as LI strains and strains of the stx₁ stx₂ or stx₂ stx_2c genotype were more likely to be stress resistant than the other genotypes. In food and the environment, pathogens encounter various types of stress. Those stressors could explain a pressure for a subset of genotypes to survive and consequently cause human infections. In Listeria monocytogenes and lactic acid bacteria, relationships between stress resistance patterns and their environmental niches have been shown (3, 29, 46). It is plausible that a similar selective pressure exists for STEC O157. Previously, differences in acid resistance among STEC O157 strains from different sources and of different genotypes have been reported (25, 28, 39). In our study, however, no significant difference of IRI values in acid resistance assays was observed among the genotypes. Further studies are required to draw any conclusion regarding the acid resistance in STEC O157. In other studies on stress resistance of STEC O157, the results of Benito et al. (2) are concordant with our results, while the results of Malone et al. (22) are not. In order to compare the results across studies, additional work is needed to confirm the genotypes under investigation. In further studies on the stress resistance of STEC O157, the stx genotype and LSPA6 lineage would be useful, because differences in distribution of these genotypes among sources have been identified in several population genetics analyses (18, 19).

The other important explanation for genetic divergence is the presence of virulence factors. Higher levels of stx expression and adherence to colonic cells from cattle were observed in LI strains (44). Interestingly, simultaneous upregulation of genes relating to a DNA damage repair system, SOS response, and stx in LI strains was observed in a microarray study (7). These associations between regulation of stress response and virulence properties may enhance bacterial survival in the environment, as well as pathogenesis. On the other hand, Dowd and Ishizaki (7) showed that genes encoding heat and cold shock proteins were differentially expressed between LI and LII strains of STEC O157. Heat and cold shock proteins influence various cell components and regulatory genes (8, 10) and may therefore be candidates for mediators of multiple stress resistance and, possibly, virulence properties. In this study, the association between stress resistance and a highly virulent genotype, clade 8, defined by single nucleotide polymorphisms (23), remains unclear. Several studies showed that strains of clade 8 belong to LI/II (12, 18). In this study, LI/II strains showed intermediate stress resistance in cluster analysis (Fig. 2).

In conclusion, our study revealed that (i) in our collection of STEC O157 isolates, heat and starvation stresses appeared to be more important for characterizing the bacterial population than were other stresses; (ii) there was a positive correlation among the six types of stress resistance; and (iii) some genotypes showed an association with multiple stress resistance. It is of interest that human-biased genotypes, such as LI strains or strains of the stx₁ stx₂ or stx₂ stx_2c genotype, showed greater stress resistance than strains of other genotypes did. However, the practical importance and mechanisms of stress resistance need further investigation. Our results also showed the importance of using STEC O157 strains from different genotypes in survival studies.