Identification of host-associated alleles by multilocus sequence typing of Campylobacter coli strains from food animals

Campylobacter is a leading cause of human bacterial gastroenteritis in the industrialized world (Altekruse et al., 1999; CDC, 2004; Friedman et al., 2000; Miller & Mandrell, 2005), with a reported incidence in the United States of 12·6 cases per 100 000 persons in 2003 (CDC, 2004). Campylobacter was second only to Salmonella enterica in the number of laboratory-confirmed cases of bacterial food poisoning in the USA and surpasses Salmonella in many other nations (Miller & Mandrell, 2005). The majority (>90 %) of campylobacterioses are caused by Campylobacter jejuni, with the remainder caused primarily by Campylobacter coli (Gillespie et al., 2002). Although C. coli is responsible for fewer food-borne illnesses than C. jejuni, the impact of C. coli is still substantial; more than 25 000 cases of food-borne illness in the UK were estimated to be caused by C. coli in the year 2000 (Tam et al., 2003).

The increasing recognition of C. coli as a food-borne human pathogen has led to the development of two novel multilocus sequence typing (MLST) systems for C. coli epidemiological studies (Dingle et al., 2005; Miller et al., 2005a). Both systems, termed CcMLST1 (Dingle et al., 2005) and CcMLST2 (Miller et al., 2005a), utilize the same seven housekeeping loci and allele end points as the C. jejuni MLST method described by Dingle et al. (2001), differing only in the primer sets used for the amplification. In conjunction with the publication of these systems, the C. jejuni MLST database () (Jolley et al., 2004), maintained at the University of Oxford, was expanded to include C. coli typing and strain information. Although C. coli has been shown to be less variable than C. jejuni (Dingle et al., 2005; Duim et al., 1999), substantial diversity was apparent across the seven C. coli MLST loci (Dingle et al., 2005; Miller et al., 2005a), indicating that CcMLST1 or CcMLST2 would be suitable for typing C. coli strains. With a sample set of 56 C. coli strains, primarily from chicken and swine, we previously identified 37 sequence types (STs) (Miller et al., 2005a). Similarly, Dingle et al. (2005) identified 34 STs among 68 human clinical, swine, chicken and Penner reference strains, although only 6 STs were observed in the chicken and swine strains (3 each).

The level of diversity across the MLST loci is important for the potential identification of alleles or STs that may correlate with animal host, geographical location, or other factors related to the source of the strain. A large number of MLST alleles have been identified in C. jejuni; however, many of these are unique in the database and, thus, cannot be used currently to trace strains back to a food source. Of the 155 major C. jejuni MLST alleles (represented at least 5 times in the database), only 65 showed any association with a host; the majority (54/65 alleles) were associated with avian hosts, generally wild birds (especially starlings), and only 11 alleles were associated strongly with a food source, namely chicken. There were no C. jejuni alleles with significant association to other animal sources, including cattle, swine and sheep. Nevertheless, it does appear that certain C. jejuni ST clonal complexes are associated with animal source(s). Colles et al. (2003) reported that the ST-45 complex was associated with turkeys and chickens, and the ST-42 and ST-61 complexes were associated with sheep. Also, Manning et al. (2003) identified an association between swine and members of the ST-403 complex.

Neither C. coli MLST study reported host association at either the allele or the ST level. This may be due to a level of genetic diversity in this species that is insufficient for the identification of host-associated alleles or STs. Another possibility is that C. coli is sufficiently variable, but that the sample sets used to validate the C. coli MLST systems were not representative of the true diversity. It is also conceivable that other potential MLST loci, not used in either method, contain host-associated alleles, and that the current methods, while suitable for typing C. coli strains, cannot be used for source tracking.

To address some of these issues, we employed MLST to determine the STs of 488 C. coli strains from food animals (cattle, chickens, swine and turkeys), collected at different times and from different geographical locations. Host-associated alleles and STs were identified among the 488 C. coli strains. Our findings have implications regarding the potential use of these host-associated alleles and STs in the source tracking of C. coli.

Reagents.
All chemicals were purchased from Sigma-Aldrich Chemicals or Fisher Scientific. PCR enzymes and reagents were purchased from New England Biolabs or Epicentre. DNA sequencing chemicals and capillaries were purchased from Applied Biosystems. Sequencing and PCR oligonucleotides were purchased from Qiagen.

C. coli strain sources and isolation methods.
C. coli strains (488) from cattle (feedlot and dairy), chickens, swine and turkeys were obtained for this study (for a complete listing of strains and strain information see the Supplementary Table S1 available with the online journal). The strains were obtained from the Agricultural Research Service (ARS) - Russell Research Center (RRC; 333 strains, including 54 from cattle, 120 from chicken and 159 from swine), the ARS - National Animal Disease Center (NADC; 27 strains from turkeys) and North Carolina State University (NCSU; 128 strains, including 9 from cattle, 9 from chicken, 26 from swine and 84 from turkeys).

Feedlot cattle samples.
The RRC Campylobacter strains used in this study were collected as part of the 1999 National Animal Health Monitoring System (NAHMS) study of the health and management of USA feedlot cattle (). The study focused on feedlots with capacity of 1000 head or more, as such facilities held over 80 % of USA cattle as of February, 1999. The cattle sampled were a mean age of 18 months old. Faecal samples were obtained from 73 feedlots in 11 major USA feedlot states: California, Colorado, Idaho, Iowa, Kansas, Nebraska, New Mexico, Oklahoma, South Dakota, Texas and Washington. Samples were collected during a one year period (October 1999 through September 2000) from pen floors. For each feedlot, 25 samples were collected from 3 pens. Feedlots were sampled twice during the one year collection period. Samples were shipped for overnight delivery to the USDAARSBEARRU laboratory in Athens, GA, USA, for isolation of Campylobacter.

For Campylobacter isolation, faecal samples were diluted 1 : 4 and 1 : 40 in PBS. One hundred microlitre aliquots of each dilution were spread uniformly on duplicate Campy-Cefex plates (Stern et al., 1992). The plates were placed in zip-top bags and incubated microaerobically (5 % O₂, 10 % CO₂, 85 % N₂) for 48 h at 42 °C. Campylobacter was presumptively identified from microscope wet mounts of cells using phase-contrast optics at x100 magnification.

For the present study, a total of 54 strains from a pool of 448 strains, comprising 43·5 % (448/1029) of all the Feedlot 1999 Campylobacter strains, were randomly selected from among the frozen stock Feedlot 1999 culture collection. Strains were derived from each of the 11 states that were surveyed, representing 67·1 % (49/73) of all feedlots sampled.

Swine samples.
RRC swine strains were from faecal samples from animals (1 week2·5 years old) and were obtained as described above. Swine isolates from eastern North Carolina (NCSU) were obtained from finishing hogs (mean age 6 months) by direct platings on CCDA (Oxoid) as described for isolations from turkey faecal samples (Smith et al., 2004).

Broiler chickens.
Whole broiler carcasses (RRC; 56 weeks old at sampling) were bagged aseptically and rinsed for 1 min with 400 ml sterile PBS. Campylobacter was isolated from the carcass rinse following selective enrichment as described by Musgrove et al. (2003). Broiler samples from eastern North Carolina (NCSU) were obtained from 46-week-old birds as described for isolation from turkeys (Smith et al., 2004).

Turkey samples.
NCSU Campylobacter strains were from 318-week-old birds and were isolated by direct platings on CCDA as described by Smith et al. (2004). NADC Campylobacter strains were isolated from cloacal swabs of market-weight turkeys as described by Wesley et al. (2005).

DNA purification.
Campylobacter genomic DNA from RRC strains was prepared using the PUREGENE yeast and Gram-positive bacteria genomic DNA purification kit from Gentra Systems using the protocol recommended by the manufacturer (protocol 01110). Genomic DNA from NCSU strains was prepared using the Qiagen DNeasy tissue kit using the protocol recommended by the manufacturer. Genomic DNA from NADC strains was prepared as described previously (Miller et al., 2005b).

Strain characterization and identification.
Campylobacter BAX PCR was performed on RRC strains according to the manufacturer's instructions (DuPont Qualicon). The PCR primer sequences were designed from regions of the cadF virulence gene unique to C. jejuni and C. coli (Konkel et al., 1999). For C. coli, the primers were cc-32-237 (5'-ACT CGG ATG TAA AAT ATA CAA ATT CTA CTC TT-3') and cc-30-rc682 (5'-TTT TTC TTC AAA GGC TGG ATT GAT ATC TAC-3'), whereas for C. jejuni, the primers were cj-30-560 (5'-AAA GGA AAA AGC TGT AGA AGA AGT TGC TGA-3') and cj-30-rc682 (5'-TTT TTC TTG AAA AGT TGG ATT TAT AGT AGT-3'). The amplification conditions were: 15 s at 94 °C, 2 min at 65 °C, 1 min at 72 °C (30 cycles). The amplicons were characterized by electrophoresis using 2 % (w/v) agarose gels; the expected amplification products are 560 bp for C. coli and 182 bp for C. jejuni.

NCSU strains were speciated using C. jejuni-specific hipO primers and C. coli-specific ceu primers as described previously (Smith et al., 2004). Additional speciation, using both hipO and lpxA primer sets, was performed on C. coli strains containing both C. jejuni and C. coli MLST alleles (e.g. strains comprising ST-1121). The hipO gene was amplified using the hipF (5'-ATG ATG GCT TCT TCG GAT AG-3') and hipR (5'-GCT CCT ATG CTT ACA ACT GC-3') primer set. For lpxA, multiplex PCR reactions were performed using the following primers: cj0121 (5'-ACA ACT TGG TGA CGA TGT TGT A-3'), cc0120 (5'-AGA CAA ATA AGA GAG AAT CAG-3') and KK2 (5'-CAA TCA TGW GCN ATA TGR CAA TAN GCC-3') (Klena et al., 2004). PCR reactions were performed on an MJ Research Tetrad thermocycler with the following settings: 30 s at 94 °C, 30 s at 50 °C, 2 min at 72 °C (35 cycles). The expected lpxA amplification product sizes are 391 bp for C. coli and 331 bp for C. jejuni; the expected hipO amplification product size is 176 bp.

NADC strains were speciated using a multiplex PCR method as described by Cloak and Fratamico (2002). Additionally, all NADC strains were typed by PFGE of SmaI-digested genomic DNA. All NADC strains had unique PFGE patterns except strains NADC 9203 and NADC 9519 (which were obtained from different farms).

Additional subtyping was performed on certain strains by amplifying and sequencing portions of the kpsM and kpsT capsular genes, and porA, which encodes the major outer-membrane protein. The kpsM primer set was kpsMF (5'-TTA GRG ART TAA ARA CTA GAT TTG G-3') and kpsMR (5'-TAA TTY TCA AAA AAA TTA AAT CTT AAA AG-3'), the kpsT primer set was kpsTF (5'-CTT ATC CTT TAT TTA RTG GYG GAA G-3') and kpsTR (5'-CTA TKC CYT CAT CTA CAT CAT CAT AAA C-3'), and the porA primer set was mompL (5'-ACT AGT TAA AMT TAG TTT AGT WGC AGC-3') and mompR (5'-TCT TTW TYA CCR TAG TAT AAA CCA CC-3'). Amplification conditions were identical to those used for lpxA.

MLST amplification and sequencing.
The primer sets used for amplification were: aspAF1/aspAR1, atpAF/atpAR, glnAF/glnAR, gltAF/gltAR, glyAF/glyAR, pgmF1/pgmR1 and tktF1/tktR (Miller et al., 2005a). Amplification conditions were as described previously (Miller et al., 2005a). Twenty-five tkt loci could not be amplified using the tktF1/tktR primer set. To sequence these loci, two new primers, tktFN (5'-AAC GGT TTT AAG TTA TCA T-3') and tktRN (5'-TTT TGT GAC TTC CTT CAA GC-3'), were designed based on a CLUSTALW alignment of the sequences of 462 C. coli tkt loci. The majority of the amplicons were sequenced by Agencourt Bioscience. Additional cycle sequencing reactions were performed as described previously (Miller et al., 2005a).

Assignment of allele numbers, STs and clonal complexes.
Allele sequences were obtained using the Perl program MLSTparser (Miller et al., 2005a). All allelic sequences were queried against the C. jejuni/C. coli MLST database (). Sequences identical to existing alleles were assigned that allele number, whereas novel sequences were entered into the MLST database and assigned new allele numbers (these sequences are indexed by locus and allele number). STs were also determined with MLSTparser, and novel STs were assigned new ST numbers through the MLST database.

STs were grouped into clonal complexes using the program eBURST () (Feil et al., 2004). Clonal complexes were defined as groups of two or more independent strains that shared identical alleles at five or more loci; each complex was named after the putative founder ST (e.g. the ST-1161 complex). Variable sites were identified using MEGA version 2.1 (Kumar et al., 2001). A minimum spanning tree displaying possible evolutionary relationships between the STs was created with BioNumerics (version 4.0; Applied Maths) using default parameters.

Sequence typing of food animal strains
A total of 488 C. coli strains derived from four different species of food animals were typed. These strains included 63 from feedlot and dairy cattle across multiple geographical locations within the USA between October 1999 and January 2004, 185 from swine across multiple geographical locations within the USA, 129 either from live chickens on farms in eastern North Carolina between June 2002 and July 2003 or from broiler carcasses representing multiple geographical locations within the USA between 1999 and 2001, and 111 from turkeys on farms in eastern North Carolina or in the proximity of Ames, IA between June 2001 and September 2004. All strains were typed using the method of Miller et al. (2005a). A list of all 149 identified STs is presented in Fig. 1. A complete list of strains and strain information (e.g. source, animal, isolation location) is presented in Supplementary Table S1.

(30K): Fig. 1. Minimum spanning tree depicting the clustering of 146 identified STs composed primarily of C. coli alleles. The tree was created using BioNumerics (version 4.0; Applied Maths). Each ST is represented by a circle. The size of each circle is proportional to the number of strains that comprise that ST. Circles representing isolates from more than one food animal type are sectioned in proportion to the animal type composition. Thick, short lines connect single locus variants, thin, longer lines connect double locus variants and dashed lines represent three or more allele differences. The ST at the centre of the major cluster represents ST-828. Host-associated branches are designated C1 through C7.

Diversity of C. coli alleles and STs
Sequence typing of these 488 C. coli strains identified 119 alleles across the seven MLST loci; 54 (45 %) of these alleles were novel. In many instances, these novel alleles differed from alleles identified previously () by only one nucleotide substitution and could have been the result of amplification errors; therefore, to ensure that these differences were not due to PCR errors, each novel allele was reamplified and sequenced. Sequences for each amplification and reamplification matched 100 % for all novel alleles. Due in part to the large number of novel alleles, 125 of the 149 identified STs were also novel. Substantial diversity was seen among the swine, chicken and turkey strains (Table 1). However, the cattle strains appeared very clonal; 52/63 (83 %) of these strains possessed a single ST, ST-1068. The 47 RRC ST-1068 cattle strains were subtyped further by sequencing portions of the highly variable kpsM and kpsT capsular genes, and the major outer-membrane protein gene, porA. All 47 strains had identical kpsM, kpsT and porA sequences.

Table 1. C. coli STs and their distribution by host Boldface entries represent novel alleles or STs. Alleles identified previously in C. jejuni or alleles with a high similarity to existing C. jejuni alleles are underlined.

Two strains from cattle, RRC026 and RRC050, possessed STs identified previously in C. jejuni, ST-21 and ST-38, respectively (Table 1 and Supplementary Table S1). Also, one turkey strain, NCSU064, possessed the novel ST-1162 ST composed entirely of C. jejuni alleles (Table 1 and Supplementary Table S1). Finally, the cattle strain RRC043 possessed the novel ST-1244 ST, consisting of six alleles identified previously in C. jejuni or with a high similarity to existing C. jejuni alleles, as well as the C. coli allele uncA17. Even though these strains had been classified originally as C. coli, results typical for C. jejuni were obtained for these three strains with the hipO and lpxA (Klena et al., 2004) primer sets (data not shown). In addition to the alleles within the ST-21, ST-38, ST-1162 and ST-1244 STs, four other divergent alleles were identified. These four alleles, aspA103, aspA117, aspA120 and tkt67, were related phylogenetically to the aspA and tkt alleles of C. jejuni. Ninety-seven C. coli MLST alleles remained after subtraction of aspA103, aspA117, aspA120, tkt67 and the alleles comprising STs 21, 38, 1162 and 1244. Within this new sample set, 158 polymorphic sites were present across the 7 MLST loci (ranging from 10 sites in the 10 glnA alleles to 41 sites in the 16 tkt alleles).

To assess both the diversity and the relatedness of STs within the complete sample set, a minimum spanning tree for 145 of the 149 STs was constructed (Fig. 1); STs 21, 38 and 1162, previously identified only in C. jejuni, and ST-1244 were omitted. The majority of identified STs were members of a single, genetically related, cluster (I; Table 1), with ST-828 as the central ST (Fig. 1); however, several STs, many notably from turkey strains, formed a smaller, secondary, cluster (II; Table 1). All STs in this secondary cluster contained aspA103. As with RRC026 (ST-21), RRC050 (ST-38), NCSU064 (ST-1162) and RRC043 (ST-1244), the species designations of strains comprising the divergent STs in the secondary cluster were examined further with additional PCR primer sets. Amplification using cadF (Konkel et al., 1999), hipO and lpxA (Klena et al., 2004) primer sets was in agreement with the previously determined C. coli status of these strains.

Using eBURST with a group definition of five identical loci, we identified two clonal lineages, with founder STs, ST-828 or ST-1161 (data not shown). The majority of C. coli STs (142/145, 98 %) were members of one of these 2 lineages, with 128 STs belonging to one clonal lineage with founder type ST-828, and 14 STs belonging to a second, aspA103-associated lineage with founder type ST-1161. These two identified clonal lineages correlate well with the two minimum spanning tree clusters described above (Fig. 1).

Host association of C. coli alleles and STs
For each MLST locus there was one allele defined as common (aspA33, glnA39, gltA30, glyA82, pgm104, tkt43 and uncA17; Table 2). Common alleles were found in strains from all four animal sources; generally, such alleles were identified in <50 % of the strains from a particular host (with the exception of pgm104, where 51 % of the pgm104 alleles were found in swine strains). Additionally, the common alleles were numerically the most prevalent allele at each locus. For instance, aspA33 and glnA39 represented 380/485 (78 %) and 318/485 (66 %) of aspA and glnA alleles, respectively.

Table 2. Host association of C. coli MLST alleles Alleles represented <5 times in the sample set were excluded from the table. Associated alleles are those found approx. >90 % in one particular host. Common alleles are those found in all hosts, where no host represents >50 % of the alleles. Alleles that demonstrate 100 % host association with a single food-animal source are in boldface.

While common alleles were identified at each MLST locus, analysis of the typing data revealed that a substantial number of major C. coli MLST alleles were strongly host associated. Several C. coli alleles were identified only once or twice, and were therefore excluded from further analysis of potential host association. Table 2 lists the host association of 46 major C. coli MLST alleles (an incidence amongst the samples of n>5 being chosen arbitrarily). Host-associated alleles were defined in this study as those identified at approximately >90 % in strains from one particular host (e.g. 97 % of the glnA38 alleles were identified in strains from swine). Using this criterion, swine-, chicken- and turkey-associated alleles were identified (e.g. aspA32, gltA65 and gltA103, respectively); however, no cattle-associated alleles were identified in the sample set. Also, certain alleles were found only in strains from chickens or turkeys (e.g. uncA41). Although they represented <90 % of the alleles in respect of each host, they represented greater than 90 % when combined; therefore, these alleles were classified as poultry-associated. Each locus had at least one swine-associated and one turkey-associated allele. Even though poultry-associated alleles were present at all loci except aspA, only one allele associated with chicken-derived strains was identified (gltA65). For 16 of the 46 major MLST alleles, there was complete association (100 %) between allele and host. This number would increase to 25 with the inclusion of alleles associated with a poultry host (e.g. aspA103 and glyA140).

The majority of C. coli STs also showed an association with a host (Table 1). Of the 25 major (n>5) STs identified in this study, 19 (73 %) were found only in strains from a particular animal type. Of these, eight STs were associated with swine strains, three STs were associated with chicken strains and seven STs were associated with turkey strains. Three STs (ST-1119, ST-1121 and ST-1170) were found only in strains from poultry and three STs (ST-828, ST-1068 and ST-1104) were found only in cattle and swine strains. Minimum spanning tree analysis of the C. coli STs also identified several, genetically related, host-associated branches in the primary cluster (C1C7; Fig. 1). Five of these branches (C1C5), originating in the tree from single locus variants of ST-828, were strongly associated with swine, while two branches (C6 and C7) were strongly associated with poultry.

MLST typing of 488 C. coli strains from 4 different types of food animals revealed a substantial level of diversity in alleles and STs, particularly among the poultry and swine strains. Two, genetically related, clusters were identified (Fig. 1). The majority of identified STs were members of cluster I, whereas the secondary cluster, cluster II, was defined by the presence of aspA103, an allele found previously to be phylogenetically related to C. jejuni aspA alleles (Miller et al., 2005a). The detection of aspA103 in these C. coli STs suggests lateral transfer of this allele from C. jejuni to C. coli, with subsequent divergence in C. coli. Evidence for such a lateral transfer has been described by Meinersmann et al. (2003) and Schouls et al. (2003). In contrast to the large amount of diversity detected among the poultry and swine strains, the cattle strains were relatively clonal; 83 % of the cattle isolates possessed the ST-1068 ST in addition to identical kpsM, kpsT and porA alleles. A substantial number of kpsM, kpsT and porA alleles exist in C. coli (data not shown), suggesting that the cattle strains containing ST-1068 either are the same strain or are members of a group of highly related strains. The RRC cattle strains typed in this study represent 26 feedlots in 11 different states. The presence of a highly related or identical group of C. coli strains, isolated from different regions in the USA over a 4 year period, in cattle raises some interesting questions. C. coli strains containing ST-1068 were also found, although at a lower incidence, in samples from swine; therefore, it is possible that C. coli strains containing ST-1068 are particularly suited to colonization and growth in large animals. Additional experiments, using a larger ST-1068 sample set from cattle and swine, will be necessary to address these issues.

A noteworthy result from this study was that, even though common C. coli MLST alleles were identified at each locus, several host-associated alleles and STs were identified. The majority of major alleles (defined as alleles represented 5 times in the sample set) and STs were associated with a particular animal type (turkeys, chickens or swine). In some instances, the allele association was very pronounced with associations approaching or equalling 100 % (e.g. aspA32, gltA44). Each locus had at least one host-associated MLST allele and also one common allele. The composition of STs was also host associated. With few exceptions, strains from a particular animal type possessed STs composed of at least one relevant host-associated allele. In some cases, e.g. ST-1128, ST-1144, ST-1161, the ST was composed of all host-associated alleles, swine-associated alleles in the first two examples and turkey/poultry-associated alleles in the latter example.

Several host-associated alleles identified in earlier studies (Dingle et al., 2005; Miller et al., 2005a) were also identified here. For instance, pgm113 and uncA41 were detected only in poultry-derived strains here and in the earlier studies. Overall, our results were consistent with those obtained already (Miller et al., 2005a; Dingle et al., 2005), with a few exceptions. Specifically, allele aspA53, which in this study and in the study by Dingle et al. (2005) was found exclusively in swine strains, was detected in swine as well as chicken strains in our earlier study (Miller et al., 2005a). Inclusion of the alleles from earlier studies (Dingle et al., 2005; Miller et al., 2005a) changes the aspA53 allele frequencies slightly, but does not change the host association. Another deviation from previously described allelic associations was tkt35, which in this study was found in 46/50 (92 %) swine strains, but was rarely detected in strains from swine (and was instead more frequently detected in strains from chickens) in an earlier study that analysed C. coli strains from food animals in the UK (Dingle et al., 2005). It is not clear whether this represents a fundamental difference between USA and UK C. coli strains. If so, alleles such as tkt35 would be very useful for source tracking because it is important to be able to compare human clinical strains to animal and environmental strains from the same geographical location and not to the species as a whole.

The presence of host-associated MLST alleles and STs suggests that source tracking may be possible in C. coli. Few host-associated alleles showed a perfect association with a host, and several of these alleles were relatively minor alleles in the sample set; therefore, single host-associated alleles should not be used for source tracking. Promising candidates for source tracking would be STs composed of several host-associated alleles. However, such STs are relatively uncommon; only four STs (ST-1128, ST-1144, ST-1150 and ST-1161) contained seven host-associated alleles. A minority (26 %) of the host-associated STs contained four or more host-associated alleles, but STs containing only one host-associated allele could be used to source track strains if the association of the one allele is extremely strong (e.g. aspA32, gltA44 and tkt169). Most STs in Table 1 are consistent with their allelic composition, with respect to host-associated alleles (i.e. alleles identified at >90 % in strains from one particular host), and could be used potentially for effective source tracking. While six of the host-associated STs contained no host-associated alleles, 110/133 (83 %) of the STs associated with one animal contained, in addition to common alleles, alleles only associated with that host. Of the remaining 17 STs, 11 contained either pgm113 or tkt47, indicating that some alleles classified as host associated in this study may not be robust indicators of source. Further analysis of additional C. coli strains, from multiple sources and geographical regions, will be necessary to determine the fitness of the host-associated alleles identified in this study for source tracking. Geographical differences in host association (such as the putative difference in tkt35 host association between UK and USA strains) may be identified. Also, as more C. coli strains are typed by MLST, alleles termed minor in this study (those with an incidence n<5) may be determined to be host associated. Experiments to determine whether C. coli host-associated MLST alleles, identified in strains from food animals, can be used to identify the potential source of C. coli clinical strains, are in progress.

It is not clear why C. coli strains contain host-associated alleles whereas the majority of C. jejuni strains do not. Some host-associated C. jejuni alleles do exist (e.g. gltA22, glyA48, pgm17 and uncA23, mostly associated with avian hosts) but not to the same extent as those described for C. coli in this study. Similarly, host-associated STs exist in C. jejuni, but these STs represent only a small fraction of the 1135 C. jejuni STs described currently (). One possibility is that the C. jejuni MLST dataset is not sufficiently diverse, geographically. Eighty-four per cent of all typed C. jejuni strains originated from the UK, Curacao and the Netherlands. A similarly geographically restricted sample set of C. coli strains (i.e. strains originating primarily in the USA) have been typed; however, host-associated alleles were found in this sample set. Therefore, it is possible that C. jejuni strains from the UK are more clonal and that analysis of additional strains from other geographical locations (e.g. N. America and Asia) may assist in identifying host-associated C. jejuni alleles. This will be especially important if geographical differences in STs are observed, similar to those identified potentially with tkt35. Also, typing additional strains may identify underrepresented host-associated alleles in the existing dataset. One such example is uncA79, identified once previously in a chicken strain (Miller et al., 2005a), but 41 times among chicken and turkey strains in the present study. Another possibility is that the current C. jejuni MLST loci, while satisfactory for typing C. jejuni strains, may not be ideal for identifying host-associated alleles. Therefore, other loci may need to be sequenced to determine if host-associated alleles exist in C. jejuni. As such, the alleles identified in this study may become less robust source-determining markers as more C. coli strains are typed. Although that appears unlikely, typing more C. coli strains from other geographical locations and food sources will enhance the data described here as well as identifying other source-tracking markers that will improve epidemiological studies of C. coli.

This work was supported by the United States Department of Agriculture, ARS CRIS project 5325-42000-041. Collection and characterization of the strains from North Carolina was supported in part by USDA NRI-2003-02017. This publication made use of the Campylobacter MLST website () developed by Keith Jolley and sited at the University of Oxford (Jolley et al., 2004). The development of this site has been funded by the Wellcome Trust.