Shiga Toxin-Mediated Hemolytic Uremic Syndrome: Time to Change the Diagnostic Paradigm?

Background Hemolytic uremic syndrome (HUS) is caused by enterohemorrhagic Escherichia coli (EHEC) which possess genes encoding Shiga toxin (stx), the major virulence factor, and adhesin intimin (eae). However, the frequency of stx-negative/eae-positive E. coli in stools of HUS patients and the clinical significance of such strains are unknown. Methodology/Principal Findings Between 1996 and 2006, we sought stx-negative/eae-positive E. coli in stools of HUS patients using colony blot hybridization with the eae probe and compared the isolates to EHEC causing HUS. stx-negative/eae-positive E. coli were isolated as the only pathogens from stools of 43 (5.5%) of 787 HUS patients; additional 440 (55.9%) patients excreted EHEC. The majority (90.7%) of the stx-negative/eae-positive isolates belonged to serotypes O26:H11/NM (nonmotile), O103:H2/NM, O145:H28/NM, and O157:H7/NM, which were also the most frequent serotypes identified among EHEC. The stx-negative isolates shared non-stx virulence and fitness genes with EHEC of the corresponding serotypes and clustered with them into the same clonal complexes in multilocus sequence typing, demonstrating their close relatedness to EHEC. Conclusions/Significance At the time of microbiological analysis, ∼5% of HUS patients shed no longer the causative EHEC, but do excrete stx-negative derivatives of EHEC that lost stx during infection. In such patients, the EHEC etiology of HUS is missed using current methods detecting solely stx or Shiga toxin; this can hamper epidemiological investigations and lead to inappropriate clinical management. While maintaining the paradigm that HUS is triggered by Shiga toxin, our data demonstrate the necessity of considering genetic changes of the pathogen during infection to adapt appropriately diagnostic strategies.

Although HUS is typically caused by EHEC, stx-negative/eaepositive (stx2/eae+) E. coli strains are occasionally excreted by patients with HUS [21,22]. However, the frequency of such strains is unknown, and their origins and clinical significance are poorly understood. To answer these questions, we studied stools from HUS patients, processed so as to detect these variants. We characterized the identified isolates and compared them to EHEC associated with HUS with respect to serotypes, virulence and fitness genes, phenotypes, and multilocus sequence types. the EHEC isolates (Table 1). One additional stx2/eae+ strain belonged to serotype O121:H19 (which was also found among EHEC; Table 1), and the remaining three were nontypeable ( Table 1). Each of the 43 stx2/eae+ strains lacked all known stx alleles. The stools from which these strains were isolated contained neither stx genes as demonstrated by PCR screening of enriched primary stool cultures, nor free Stx as demonstrated by the Vero cell assay on stool filtrates, nor any classic bacterial enteric pathogens including Salmonella, Shigella, and Yersinia spp., and Campylobacter jejuni.

Comparison of phenotypes of stx2/eae+ E. coli and EHEC
The stx2/eae+ E. coli isolates shared with EHEC of the corresponding serotypes several diagnostically useful phenotypes (Table 3), but, in contrast to EHEC, their culture supernatants were not toxic to Vero cells (Table 3), the cell line that is sensitive to all Stx variants described until now. This suggests that these strains did not produce Stx encoded by stx gene(s) that might have been undetectable with our PCR protocol.

Multilocus sequence typing (MLST) analysis of stx2/ eae+ E. coli and EHEC
The phylogenetic relationships between the stx2/eae+ E. coli O26:H11/NM, O103:H2/NM, O121:H19, O145:H28/NM, and O157:H7/NM and EHEC of the same serotypes were determined by MLST analysis of randomly selected strains ( Figure 1). The stxnegative and stx-positive strains of each serotype shared the same sequence type or at least six of the seven alleles investigated [30] and clustered therefore into the same clonal complex (Figure 1). In contrast, the stx-negative strains of the five different serogroups showed no close relationship based on their allelic profiles.

Serological investigations
Serum samples were collected during the acute phase of HUS from 12 of 17 patients who shed stx2/eae+ E. coli O157 and from 10 of 26 patients who shed stx2/eae+ non-O157 strains. All 12 patients with E. coli O157 strains developed anti-O157 lipopolysaccharide (LPS) immunoglobulin M (IgM) antibodies. Among the 10 patients with non-O157 E. coli from whom serum samples were available, only the patient who shed E. coli ONT:H7 developed anti-O157 LPS IgM, suggesting a recent infection with E. coli O157 which probably precipitated the HUS. IgM antibodies to O157 LPS were not detected in the other nine sera from patients who shed stx2/eae+ E. coli O26:H11/NM (n = 5), O103:H2 (n = 1), O145:H28/NM (n = 2), and ONT:H6 (n = 1). Each of the five patients who excreted stx2/eae+ E. coli O26 had IgM antibodies to O26 LPS; presence of anti-O103 and anti-O145 LPS antibodies in the one and two patients, respectively, who shed stx2/eae+ E. coli strains of these serogroups could not be determined because of insufficient amount of the serum samples.
Characterization of HUS patients who excreted stx2/eae+ E. coli but not other pathogens Twenty of 43 HUS patients who shed stx2/eae+ E. coli as the only pathogens were males and 23 were females. Thirty-seven patients for whom information about age was available were children between 5 months and 9 years (mean age, 31.3 months; median age, 24 months). The mean age of these patients was significantly lower than that of patients who shed EHEC strains (range, 4 months to 64 years, mean, 34.6 months; median, 27 months) (P = 0.003; Mann-Whitney U test). None of the 43 patients who excreted stx2/eae+ E. coli strains and nine (2.0%) of the 440 patients who excreted EHEC died during the acute phase of HUS.   Efa1

DISCUSSION
Stxs produced by EHEC are considered the major precipitants of the microvascular endothelial injury that underlies HUS [1,12]. These toxins and their encoding genes are also the major targets exploited in the laboratory diagnosis of EHEC infections [2,31]. Our finding of stx-negative E. coli strains that are closely related to EHEC as the only putative bacterial pathogens in stools of ,5% of patients with HUS during a long-term study sheds therefore new light into microbiological, diagnostic, epidemiological, and clinical aspects of this disorder. Several lines of evidence support the hypothesis that these stxnegative strains represent derivatives of original infecting EHEC that lost the ability to express Stx (EHEC-LST), in these cases because of stx-bacteriophage excision during infection. First, the spectrum of serotypes of the stx-negative isolates is similar to that of EHEC strains isolated from HUS patients ( The absence of serogroup O111, which was found in 2.5% of EHEC-excreting patients, among the HUS-associated EHEC-LST isolates (Table 1) is probably because stx 1 , which is the most prevalent stx in EHEC O111 [33] is encoded within a defective prophage, which has been immobilized in the EHEC genome [34], preventing the stx loss by phage excision. Also, the ratio between SF and NSF EHEC-LST O157 isolates (88% vs. 12%) ( Table 1), while in contrast to that observed between SF and NSF EHEC O157 isolated from HUS (26% vs. 74%) ( Table 1), is proportional to the greater frequency with which the stx loss occurs in SF EHEC O157:NM [21,22]. Second, the EHEC-LST isolates share with EHEC of the corresponding serotypes non-stx virulence and fitness genes and belong to the same MLST clonal complexes. This demonstrates a common phylogeny and conservation of variable genome regions in the two groups of organisms. Third, the possibility of stx loss from EHEC O26:H11/NM and SF EHEC O157:NM during the course of HUS has been proposed in our previous study based on closely related molecular characteristics of stx-positive and stx-negative isolates from the initial and follow-up stools, respectively [21]; the stx loss has been confirmed in vitro [16]. Fourth, the genomic loci where stx-converting bacteriophages integrate into the genomes of EHEC O26:H11/ NM [16] and O157:H7/NM [14,15] were unoccupied in all stx2/ eae+ isolates of these serotypes indicating that the stx loss resulted from the excision of stx-converting phages. Altogether, these data strongly suggest that the stx2/eae+ E. coli strains isolated from HUS patients were EHEC-LST.
Although the design of the study does not allow to determine whether the stx2/eae+ E. coli strains could be primary pathogens that triggered the HUS in patients from whom they were isolated, this seems to be unlikely, taking into account the paradigm that HUS is caused by Stx [1,2]. However, because they have unoccupied stx-bacteriophage integration sites, the stx2/eae+ E. coli O26:H11/NM and O157:H7/NM strains can be transduced with stx-harboring phages and converted thus to EHEC, at least in vitro [16]. Whether such an event can occur during infection and whether it could trigger HUS remains to be established. Moreover, conditions favoring lysogenic conversion or stx loss in vivo are poorly understood [18].
The ratio between EHEC-LST (5.5%) and EHEC (55.9%) isolated in our study is ,1:10. The finding that every 10 th patient with HUS, a condition that was most probably triggered by an EHEC infection, does not shed EHEC, but rather excretes EHEC-LST when stool is subjected to appropriate microbiological analysis, has important practical implications. First, the stx loss in an EHEC strain during infection can mislead epidemiological investigations because an stx-negative strain would not be, based on currently used criteria, considered to be epidemiologically related to stx-positive strains, even though of the same serotype. Therefore, the awareness of the possibility that a patient with HUS can excrete, in lieu of the original infecting EHEC, EHEC-LST which shares non-stx molecular characteristics with EHEC of the corresponding serotype, can assist epidemiologists to link correctly epidemiologically related cases, to identify the source of the infection and to trace modes of transmission. In such studies it is necessary to bear in mind that the loss of stx-harboring bacteriophages can alter pulsed-field gel electrophoresis patterns of the strains [15,16], so that the epidemiologically related stxpositive and stx-negative strains can more or less differ in their fingerprints [16]. Second, microbiological identification of patients infected with EHEC O157:H7 early in illness is strongly associated with a good nephrologic outcome [1], probably because such an expeditious diagnosis prompts early volume expansion [35]. In patients excreting an EHEC-LST, the microbiological diagnosis may be delayed or the EHEC etiology of the disease is missed using tests that rely solely on the detection of stx genes or Stx production [31]. However, this concern might or might not be appropriate, depending on when in the course of the disease EHEC-LST replace the EHEC that almost certainly preceded the EHEC-LST. Whereas underdiagnosing EHEC-LST might not be clinically critical in patients with overt HUS who had already developed microvascular injury, the information about the presence of EHEC-LST in the stool (which may indicate continuing presence of the original EHEC in an amount undetectable by PCR stx screening) is of an important diagnostic value in patients with diarrhea, especially those with bloody diarrhea which often precedes HUS [1,2,5]. In such patients, a prompt diagnosis of EHEC-LST infection should alert the treating physician that the patient could develop HUS and should be monitored assiduously, receive isotonic volume expansion [35], and not be given antibiotics [36,37] or antimotility agents [1,38]. Clearly, we need better, and broader, microbiologic procedures to detect, in addition to stx or Stx, also non-stx/Stx EHEC targets.
How should EHEC-LST be detected? eae, which is present in the majority of EHEC isolated from HUS patients in Europe [2,19] and the United States [7,8], appears to be a quite appropriate additional diagnostic target. Specifically, based on our finding that eae-negative EHEC account for ,4% of HUSassociated EHEC isolates in Germany [19] and assuming, based on the data from the present study, that stx loss occurs in ,10% of the infecting EHEC, the using eae as a target to identify EHEC-LST would miss only ,0.4% of such strains. This proportion of missed pathogens might be higher in regions where eae-negative EHEC account for a higher proportion of HUS isolates [10,39]. In this case, the gene encoding the Stx-producing E. coli autoagglutinating adhesin (saa) [39], which is present in the majority of eaenegative EHEC associated with HUS [39], might be a suitable alternative to search for EHEC-LST. Optimal detection algorithms, and non-stx loci, depend on geographic and temporalspecific epidemiological trends, and for this reason it is necessary to continue microbiological surveillance. In this regard, it is critical to not abandon culture in favor of non-culture methodology, but to apply both modalities simultaneously to all stool specimens. In consideration of current epidemiology and etiology [1,2], we believe that optimal diagnosis for both EHEC and EHEC-LST should consist of plating on sorbitol MacConkey (SMAC) agar (detection of NSF E. coli O157:H7/NM) and enterohemolysin agar (detection of the most frequent non-O157 serotypes based on the enterohemolytic phenotype) (Table 3) [9,24], Stx or stx gene testing, and targeting eae (or saa) and sfpA (the latter for a specific detection of SF E. coli O157:NM) ( Table 2) [23]. In culturenegative patients for whom serum samples are available, detection of antibodies against LPSs of the most frequent E. coli serogroups associated with HUS (both O157 and non-O157) can be an alternative approach to detect infection with both EHEC and EHEC-LST.
The application of stx/Stx-independent diagnostic strategies to identify EHEC-LST in HUS patients appears to be appropriate to consider for several reasons. First, stool samples from such patients are frequently collected only after HUS develops, i.e. $1 week after the onset of the prodromal diarrhea [1], as was true also for the majority of patients in our study. At this point in illness, EHEC might have been cleared [40] or stx might have been lost from infecting organisms [21]. In the latter case, using a non-stx diagnostic target such as eae or saa can still identify, with a high probability, the causative agent. Moreover, such diagnostic approaches would permit a prospective systematic clinical study to determine if stx loss by the infecting EHEC during the course of HUS might result in a less severe acute disease and/or a decreased rate of late sequelae.
Our study has two major limitations. First, although we systematically sought stx2/eae+ E. coli strains in stools that were negative for EHEC, as indicated by negative result of PCR screening for stx genes, we did not seek such strains in stool samples that contained EHEC. Stools that were stx-positive in PCR screening were only analyzed for stx-positive colonies. This approach rendered it impossible to determine if some of the patients had both stx2/eae+ E. coli strains and EHEC in the same stool, a situation that would indicate progressive loss of stx by the infecting EHEC population. Although poly-isolate analyses in other studies where a panel of markers were used including stx nucleic acid hybridization did not detect such a mixed population [41,42], in these studies, stools were collected early in illness, and only five colonies were studied. Further studies targeting systematically both stx and eae (or other loci) in sequential stools from HUS patients are needed to determine the dynamic of stx loss during infection and to identify factors that could influence this process, such as serotype of the infecting EHEC, patient-related and/or environmental factors. Second, we do not have sufficient data to determine if antibiotics played a role in stx loss, as suggested in several experimental studies [14,17].
In conclusion, at the time of microbiological analysis, an appreciable subset of patients with HUS shed no longer EHEC sensu stricto, but do excrete EHEC that lost stx. Diagnostic strategies need to be formulated to detect such pathogens and treating physicians should be immediately informed. Patients who shed such strains should be considered as potentially infected by EHEC and managed accordingly, at least until more data about the clinical significance of the EHEC-LST emerge. While not changing the paradigm that Stxs are the critical virulence determinant of EHEC responsible for HUS, our data do demonstrate the necessity of taking into account possible genetic changes of the pathogens during infection when developing appropriate diagnostic strategies and interpreting results of microbiological analyses.

Patients
During routine diagnostic work between 1996 and 2006, we sought stx2/eae+ E. coli and EHEC strains in stools (one stool per patient) from 787 epidemiologically unrelated patients with HUS. The patients were hospitalized in 23 pediatric nephrology centers in Germany and Austria described previously [5], with an extended period for patient enrollment until December 2006. Stools from patients who excreted stx2/eae+ E. coli strains and EHEC strains were collected between 5 and 14 days (median 9 days) and between 5 and 13 days (median 8 days), respectively, after the onset of prodromal diarrhea. The difference in the time of stool collection was not significant (P = 0.24, Mann-Whitney U test). HUS was defined as microangiopathic hemolytic anemia (hematocrit ,30%, with evidence of the destruction of erythrocytes on a peripheral-blood smear), thrombocytopenia (platelet count ,150,000/mm 3 ), and renal insufficiency (serum creatinine concentration greater than the upper limit of the normal range for age) [36].

Serological investigation
IgM antibodies against the O157 and O26 LPS antigens were sought in sera from acute phase of HUS using an immunoblot [21].

Statistical analysis
The non-parametric Mann-Whitney U (Wilcoxon) two-sample test for independent sample groups and OpenStat2 Software (http:// www.statpages.org/miller/openstat/) were used for statistical analysis. P values ,0.05 were considered significant.