A detailed description of the epidemiologic and environmental investigations is the subject of a manuscript in preparation (A. Ellis, pers. comm.). A total of 532 human stool specimens were tested for Campylobacter spp. Stools from 116 persons were positive for the organism, and 11 of these were also positive for E. coli O157:H7. Of these 116 strains, 106 were submitted to NLEP for further analysis, along with 20 strains from southern Ontario not directly linked to the outbreak. Campylobacter spp. (49 isolates) obtained from animals or manure on 11 of 13 farms tested were also sent to the NLEP for further analysis (2,3). No Campylobacter organisms were isolated from the 57 water samples tested.

All 175 isolates were characterized, first by biotyping and serotyping, then by phage typing (Table 1). A subset of 83 isolates was further characterized by PFGE, while 115 isolates were subsequently tested by fla-RFLP typing. C. jejuni or C. coli were confirmed by using PCR for the hippuricase and aspartokinase genes, a strategy that also allowed the definitive identification of hippuricase-negative (hipp. neg.) C. jejuni strains. Five biotypes (I, II, III, IV, and hipp. neg.) were found among the isolates, with biotype II predominating. HS serotyping detected 14 different serotypes among the larger group of 175 isolates. Three HS serotypes were epidemiologically associated with the outbreak (Tables 1 and 2). Most outbreak-associated strains were HS serotype O:2. Phage typing was useful for further strain discrimination, yielding 22 PTs (25 if phage type variants were included) plus two isolates with atypical lytic patterns and two untypeable strains. PT 33 was most commonly associated with outbreak strains, though other phage types were also outbreak-associated. HL serotyping generated 29 types from the group of 175 strains. PFGE divided the 83 strains tested into more than 30 types when both SmaI and KpnI were used (Tables 2 and 3). Though fla-RFLP typing produced 22 different types, only 7 were epidemiologically associated with the outbreak. When combined, the results from all phenotypic and genotypic assays created a large number of distinct types (Tables 2, 3, and 4). HL serotyping allowed typing of 150 (86%) of 175 isolates tested. HS serotyping achieved 97% typeability, while phage typing and molecular typing methods typed 99% and 100% of strains tested, respectively.

The characteristics of outbreak strains were derived by correlating the results of phenotypic and genotypic assays. Only biotypes II and hipp.-neg. C. jejuni were strongly associated with the outbreak, although 13 biotype III isolates were also identified. Most (99/106; 92%) of the patient isolates epidemiologically associated with the outbreak expressed HS serotype O:2 (Table 2). The most common phage types among all isolates tested were PT 33 (100/175 isolates), PT 13 (12/175 isolates), and PT 1 (10/175 isolates). These phage types were found in isolates epidemiologically associated with the outbreak as well as those that were not, although 82/106 (77%) of patient isolates associated with the outbreak were PT 33 (Table 2). Of the 57 fla-RFLP type 1 isolates characterized, 50 (88%) were epidemiologically associated with the outbreak. PFGE SmaI types CASAI.0001, .0002, .0004, and .0011 clustered on the same branch of a dendrogram constructed with PFGE patterns from isolates obtained at the time of the outbreak (Figure 1) and were closely associated with other type characteristics connected with the outbreak, including HS serotype O:2; fla type 1; PT 33; and HL types 125, 128, and UT. Five other isolates considered epidemiologically unrelated to the outbreak had Penner type O:2, fla-RFLP type 1, PT 33, and biotype II. Four of these isolates were HL serotype 125. One strain had the PFGE outbreak type CASAI.0001, CAKNI.0001, and three of the four other strains had outbreak type CASAI.0002, CAKNI.0002. The final strain had PFGE type CASAI.0002 and CAKNI.0036, a PFGE pattern varying from CAKNI.0002 only by two minor bands at the bottom of the gel. In this context, all strains were considered outbreak type 1 strains. A second fla-RFLP type, 34, was closely associated with 11 isolates from both humans and animals, all of which were epidemiologically associated with the outbreak. All strains with fla-RFLP type 34 were HS serotype O:2, hipp. neg., and PFGE type CASAI.0003, CAKNI.0003, although four different HL serotypes and five different phage types were present (Table 4). In addition, one strain with fla-RFLP type 99 was HS serotype O:2, hipp. neg., and PFGE type CASAI.0003, CAKNI.0003. This combination of types and subtypes was considered outbreak type 2 (Table 2). As shown in Figure 2, although fla-RFLP type 99 is more similar to type 1 than type 34, it still differs from type 1 by three bands. Only seven other isolates representing a few other distinct C. jejuni types were also considered to be epidemiologically associated with the outbreak (Table 2).

Farms near the town of Walkerton were considered as possible sources of bacteria causing the outbreak. Sampling of animals on 11 farms yielded a number of C. jejuni and C. coli isolates (Table 3). Isolates of outbreak type 1 were found from cattle on farm 2 and farm 14, whereas the second outbreak type was found in cattle on farm 2. A variety of different strains were obtained from other farms, although none expressed characteristics of the two major outbreak types.

Two of the strains recovered from animals on farm 7 shared some characteristics with isolates from humans epidemiologically associated with the outbreak. A patient isolate with HS serotype O:4 complex (O13), biotype II, phage type 6, fla type 91, and HL serotype 7 was similar to an animal isolate with HS serotype O:4 complex (O:13), biotype II, phage type 6, fla type 93, and HL serotype UT. Two isolates from humans associated with the outbreak had HS serotype O:4 complex, biotype II, fla type 93, HL serotype 7, and PT 13 or 71. Isolates from animals on farms 1 and 7 were similar but were considered epidemiologically unrelated to the outbreak (2,3).

Phenotypic and molecular typing methods together support the hypothesis that bacteria entered the Walkerton municipal water supply from neighboring farms and implicate farm 2 as the major source of outbreak strains. This conclusion was consistent with hydrogeologic models in which runoff from heavy rains swept Campylobacter spp. and E. coli O157:H7 bacteria from farm 2 into the vicinity of well 5, where they gained access to the well and were distributed through the town’s water supply (2,3). A few isolates indistinguishable from the outbreak strain were recovered from patients not epidemiologically associated with the outbreak, suggesting that these patients might indeed have been associated with the outbreak. The outbreak case definition would not exclude sporadic cases occurring at the same time as the outbreak. These isolates could represent cases of secondary transmission or patients having an indirect association with the outbreak that were not identified during the epidemiologic investigation.

Isolates from some patients who were epidemiologically associated with the outbreak produced molecular subtyping results that differed from the outbreak type, suggesting that these bacteria might have been acquired from a source other than well 5 or that they may have been present on farm 2 adjacent to well 5 but not detected. These organisms could have entered the water supply near well 6, though that well was not as susceptible to contamination as well 5 (3). If well 6 was involved, isolates with the types found on other farms (e.g., farm 7) near the well should have comprised a higher proportion of outbreak strains. Patients could have acquired the organisms through direct or indirect contact with animals or persons from farms or from some other common source. Strains with characteristics similar to these non-O:2 strains were often not found on farms in the Walkerton area. Though the outbreak affected many residents in this area, it may have occurred against a background of sporadic cases.

The diversity seen among the Campylobacter isolates is in striking contrast to the single E. coli type infecting Walkerton outbreak patients and in cattle on farm 2 (2,3, data not shown). Furthermore, during the New York state fair outbreak, a single Campylobacter PFGE type predominated (9). Isolates from a point source outbreak caused by tuna salad had the same HS serotype, HL serotype, and biotype (33). Routine surveillance of Campylobacter by HS serotyping and phage typing identified a single type that caused an outbreak associated with stir-fried food in the United Kingdom (34). HS serotype, ribotype, DNA profile, and PFGE all showed the same profile in isolates obtained from a 6-week continuous source waterborne outbreak in a town in Denmark (35). A damaged sewer line was implicated in this outbreak. In contrast, of 25 outbreaks investigated by Frost et al. (36), isolates with only one PT and HS serotype were found in 13 outbreaks and multiple types (up to eight) in 12 outbreaks. The diversity of HS serotypes and PFGE types encountered in Walkerton may therefore be somewhat unusual, while the diversity of HL serotypes and phage types is consistent with information in the literature. This diversity could be the result of inclusion of strains or types that were not outbreak related or from the heterogeneity of types at nearby farms. Existing data do not allow us to determine which of these hypotheses is correct.

Continuous, comprehensive databases of molecular subtyping data for Campylobacter species have not yet been developed in Canada. Whether the Walkerton outbreak types are rare types or common types in Canada is not known. This uncertainty makes interpretation of the data more difficult and highlights the need for continuous surveillance of pathogens to support the interpretation of typing and fingerprinting data.

Different methods performed quite differently for characterizing strains. Fla-RFLP typing and Penner serotyping appeared to group strains into larger clusters, which was useful for identifying outbreak-associated strains. Results from these two methods together would have allowed good predictions about whether a Campylobacter isolate should be included in the outbreak investigation. A close correspondence has previously been found for flaA RFLP types and SmaI PFGE types which, together with HS serotypes, were found to identify C. coli clonal lines having epidemiologic significance (37). HS serotype O:2 appears to be a common strain of Campylobacter (38) and is found frequently in isolates from both humans and cattle (39,40). Additional information from fla-RFLP typing may therefore be necessary for more definitive discrimination. Several isolates belonged to the HS O:4 complex, with each antigen expressed variably in individual strains. Strains expressing this complex predominated on farms 3 and 7 (Table 3) and were also found in three isolates from patients (Table 2). Only isolates from farm 7, near well 6, had the fla-RFLP types 90 and 93 in common with patient isolates. The sources of infection of these patients was not clear, although well 6 probably did not become contaminated (3).

PFGE data correlated well with HS serotyping and fla-RFLP data. A group of closely related PFGE patterns was associated with the outbreak. PFGE was more discriminatory than fla-RFLP typing and HS serotyping, and during the outbreak, additional information had to be collected to associate all five SmaI PFGE types with the outbreak. This limited the utility of PFGE for identifying outbreak strains until epidemiologic data were available. Although fla-RFLP typing had a lower apparent discriminatory power compared with PFGE, it was more useful for organizing strains into epidemiologically relevant groups. Close examination of PFGE patterns CASAI.0001, .0002, .0004, and .0011 indicated an underlying similarity, with changes that suggested duplication, insertion, or deletion of DNA from a common ancestral type (Figure 1). Strains with patterns CASAI.0001, .0004, and .0011 all shared a common KpnI restriction pattern (CAKNI.0001). Differences in these patterns are consistent with changes within restriction sites or rearrangements. PFGE pattern CASAI.0002, however, was associated with patterns CAKNI.0002 and CASAI.0003. CAKNI.0002 differs from CAKNI.0001 only in the position of the top band, which is higher in CAKNI.0001 (data not shown). Patterns CASAI.0001, .0004, and .0011 differed from CASAI.0002 by the position of a single larger band in each pattern (Figure 1). These differences are more consistent with the addition of DNA through insertion of exogenous material or duplication of chromosomal loci. Recombination appears to occur frequently within Campylobacter species and, with genomic rearrangement, contributes to the genomic instability characteristic of certain strains (39–41). The events causing the PFGE changes seen in these closely related patterns remain to be determined. That such changes may occur at relatively high frequency is suggested by the discriminatory power of PFGE compared with other typing methods.

Phage typing was useful in defining the outbreak strains in early stages of the investigation due to the speed with which results could be obtained. However, the second most common outbreak type isolated, defined by HS serotype O:2, hipp. neg. biotype, fla-RFLP type 34, PFGE CASAI.0003, had several phage types (13, 14, 28, 33 var., 71). Phage types varied independently of the other characteristics measured (Table 4), giving this typing method a higher apparent discriminatory power than HS serotyping, fla typing, or PFGE. Including isolates into the outbreak on the basis of phage type alone,without accompanying epidemiologic data, would have been difficult. This factor may limit the utility of phage typing for detection of outbreaks, though at least one outbreak has been identified on the basis of phage typing and HS serotyping. PT 33 was, however, an effective marker for the most prevalent outbreak type.

HL serotypes 4, [4,125], 100, 112, [112, 125], 125, [125,128], and 128 were almost exclusively associated with outbreak isolates. Though it would have been difficult to identify outbreak strains on the basis of HL serotype alone, this method did help confirm the link between outbreak strains in humans and isolates from farm 2. The 125 and 128 serotypes have been seen infrequently by NLEP and are more “unique” markers than either HS serotype O:2 or phage type 33. Serotype HL 5 was associated with the HS O:17 complex in isolates from human patients and was not found in isolates from any of the farms. The source of these isolates was not determined, though O:17 strains have previously been recovered from poultry (24). HL serotype 7 was associated with the HS O:4 complex discussed earlier, suggesting that the associations between HL and HS serotypes noted previously (23) may not be random. Within the outbreak strain, however, changes in HL serotype appeared to occur more frequently than, and independently from, other type characteristics. HL typing would not have been of use in the identification of the two Walkerton outbreak strains if used in the absence of epidemiologic information. That two HL serotypes could be found, namely types [4, 125], [112, 125], and [125, 128], all of which included HL 125, was interesting. Further characterization of these complex HL serotypes could provide useful laboratory-based epidemiologic information.

In summary, two Campylobacter jejuni strains were associated with the Walkerton outbreak through the use of different typing and subtyping methods in combination with epidemiologic data. These methods were useful for defining the scope of the outbreak, for identifying the source of strains, and for tracing the route by which bacteria infected humans. The bacteriologic findings fully support the results of the epidemiologic and hydrogeologic investigations (2,3), which suggest that bacteria from cattle manure were able to enter groundwater after heavy rains and contaminate a well serving the town of Walkerton, subsequently infecting those consuming the water. Some investigators think that adult beef cattle represent a limited threat to water supplies and subsequent transmission of Campylobacter to humans (42). However, recent investigations suggest that the environment, as well as cattle and other farm animals, may play an important role in human infection with these organisms (38,39). Studies of the contribution of cattle feedlots and other farm operations to Campylobacter contamination of surface waters and watersheds, as well as subsequent human infections, would provide useful information for farm management practices and the protection and management of water resources.