Inter-laboratory evaluation of three flagellin PCR/RFLP methods for typing Campylobacter jejuni and C. coli: the CAMPYNET experience


Clare S. Harrington, Danish Veterinary Institute, Bülowsvej 27, DK-1790 Copenhagen V, Denmark (e-mail:


Aims: To compare typeability, discriminatory ability, and inter-laboratory reproducibility of three flagellin PCR/RFLP (fla typing) methods previously described for Campylobacter.

Methods and Results: The sample set (n = 100) was diverse, including both C. jejuni (n = 85) and C. coli (n = 15). Two of the three flaA typing methods amplified flaA alone, whereas one, a multiplex assay, amplified flaB in addition to flaA. DdeI restriction enzyme was employed for all methods, but HinfI was also investigated. 98–100% typeability was obtained for flaA-based methods, but only 93% for the multiplex assay, due to inconsistent amplification of a non-specific product. In addition, there appeared to be selective amplification of flaA over flaB. More DdeI types were generated using a longer flaA PCR amplicon, whilst additional use of HinfI increased the number of types by ca 25%. Inter-laboratory reproducibility for both flaA-based methods was defined at 100%.

Conclusions:Fla typing requires standardization with respect to PCR primers and restriction enzymes. This study identified an assay, employing the full flaA gene and DdeI digestion, as an appropriate method on which to standardize. 100% inter-laboratory reproducibility was demonstrated using that method.

Significance and Impact of the Study: This work should facilitate progress towards inter-laboratory standardization of fla typing.


Flagellin (fla) gene typing is one of an increasing number of genotyping methods used during epidemiological investigations of the two most commonly isolated thermotolerant Campylobacter species, C. jejuni and C. coli (Newell et al. 2000; Wassenaar and Newell 2000). These species are known to be a major cause of acute human enteritis in the developed world (Friedman et al. 2000). DNA-based methods have become widely used for typing Campylobacter spp., as they offer higher typeability, greater discriminatory power, and are available on a potentially universal scale compared with phenotyping methods, such as serotyping, for which standardized reagents are not commercially obtainable (Newell et al. 2000). Available genotyping methods have been reviewed recently, and include pulsed-field gel electrophoresis (PFGE), randomly amplified polymorphic DNA (RAPD) fingerprinting, amplified fragment length polymorphism (AFLP) profiling, ribotyping, and a proprietary, automated variation of ribotyping, referred to as RiboPrinting (Newell et al. 2000; Wassenaar and Newell 2000). A multi-locus sequence typing (MLST) method, based on sequence analysis of regions of seven housekeeping genes, has also been described (Dingle et al. 2001). Whilst each of these methods has potential advantages and disadvantages, fla typing has become very widely used because of its speed and simplicity. The technique is based on PCR amplification of part(s) of the flagellin gene locus, followed by restriction enzyme digestion to generate simple restriction fragment length polymorphism (RFLP) fingerprints (Newell et al. 2000; Wassenaar and Newell 2000).

The flagellin gene locus for C. jejuni and C. coli consists of two highly homologous, tandem genes, flaA and flaB (Nuijten et al. 1990), such that either (or both) gene(s) may act as target(s) for the initial PCR. Three of the more commonly used flagellin gene typing assays (Table 1) were developed by Nachamkin et al. (1993, 1996), Santesteban et al. (1996), and Ayling et al. (1996). The latter method is based on co-amplification of flaA and flaB using a multiplex primer set, whereas the first two methods are based on amplification of the flaA gene alone (Table 1). DdeI is used as primary restriction enzyme for all three assays, whilst HinfI has been used to enhance discrimination in the latter two methods.

Table 1.  Properties of fla typing methods investigated during this study
 Primer (orientation)*Primer targetAnneal site†Amplified geneAmplicon size (bp)†Typeability (%)Number of profiles§D-index using Dde
  1. *fwd, forward; rev, reverse.

  2. †Anneal site and amplicon sizes are based on the 1731 bp open reading frame of the flaA and flaB sequences for Campylobacter jejuni strain 81116 (GenBank Accession J05635; Nuijten et al. 1990).

  3. ‡10 bp of the Method 3 product is contributed by the cloning sequence of the forward primer.

  4. §D-index was calculated according to the method of Hunter and Gaston (1988), using one isolate from each outbreak set. ND, not determined.

Nachamkin et al. (1993, 1996)A1 (fwd)flaA4–29flaA172598410·973
A2 (rev)flaA1705–28
Santesteban et al. (1996)Pg50 (fwd)flaA1–18flaA1457100370·970
RAA19 (rev)flaA/B1432–57
Ayling et al. (1996)Cj 431 (fwd)fla10–32flaA/B1499‡93NDND
Cj 432 (fwd)flaB10–30
Cj 433 (rev)flaA/B1470–98

The widespread use of fla gene typing is attributable to its good performance in most of the criteria listed for effective evaluation of molecular typing systems (Struelens et al. 1996). The basic requirements of high typeability, reproducibility, and relatively high discriminatory power are fulfilled, as are various convenience criteria, such as accessibility, rapidity, cost-effectiveness, and ease of use. One further criterion, of increasing significance in recent years, and yet to be evaluated for fla gene typing, is that of inter-laboratory reproducibility (Meinersmann 1998). The need for this has been driven by the existence of increasing numbers of geographically disparate, multi-centre networks for comparison of molecular subtyping data, such as the ‘PulseNet’ initiative, established by the Centers for Disease Control and Prevention (CDC) in the United States (Swaminathan et al. 2000; website, Such networks have been made possible through the development of commercially available computer software packages, enabling the electronic transfer and numerical analysis of large volumes of molecular data (de Boer et al. 2000).

The current study was performed as part of the CAMPYNET network; a project financed by the European Union to co-ordinate the molecular subtyping of C. jejuni and C. coli (Newell 1998; website, The main aims of the study were to compare the relative typeability, inter-laboratory reproducibility, and discriminatory abilities of the three-flagellin gene PCR/RFLP protocols detailed above. Two different PCR protocols were performed in each of the three participating institutes, such that each protocol was run at two different institutes. The full CAMPYNET isolate set (n = 100) was used for comparison of PCR typeability; whereas a subset of isolates (n = 50) was used for inter-laboratory evaluation of RFLP profile reproducibility.

Materials and methods

Bacterial isolates and culture

The CAMPYNET sample set (n = 100) included both C. jejuni (n = 85) and C. coli (n = 15). It was derived from diverse ecological sources, and from nine different countries throughout Europe. The majority of isolates (n = 78) were sporadic, but epidemiologically related isolates (n = 22), representing seven different groups (three human outbreaks; four poultry flocks), were also included. Details are given in Table 2. Isolates were kept at −70°C or −80°C (Institute 2), as frozen stock cultures containing 10% (v/v) glycerol. They were grown on blood agar base no. 2 (Oxoid, Basingstoke, UK) containing 5% (v/v) defibrinated cattle, horse, or sheep blood at Institutes 1, 2, and 3 respectively. Plates were incubated under microaerobic conditions at 37°C, for 48–72 h.

Table 2.  Details of the CAMPYNET strain set
CNET strain number(s)SpeciesCountry of originSourceEpidemiological group*
  1. *No known epidemiological connection unless otherwise stated.

001–003/004Campylobactor jejuniDenmarkHuman/water1 (Outbreak 1)
005–008Campylobactor jejuniUK (Scotland)Human2 (Outbreak 2)
009–012Campylobactor jejuniUK (Northern Ireland)Human3 (Outbreak 3)
014–015Campylobactor jejuniNetherlandsChicken4 (Flock 1)
016–018Campylobactor jejuniNetherlandsChicken5 (Flock 2)
019–021Campylobactor coliNetherlandsChicken6 (Flock 3)
022–023Campylobactor jejuniDenmarkChicken7 (Flock 4)
024–028Campylobactor jejuniFinlandWild bird
029–032Campylobactor jejuniNetherlandsHuman
033–036Campylobactor jejuniUKHuman
038–040Campylobactor jejuniBelgiumHuman
041–044Campylobactor jejuniSwedenHuman
045–048Campylobactor jejuniDenmarkHuman
049Campylobactor jejuniFranceHuman
051Campylobactor coliFranceHuman
052, 053Campylobactor jejuniNetherlandsCattle
055Campylobactor jejuniUK (Northern Ireland)Cattle
056, 057Campylobactor jejuniDenmarkCattle
058, 059Campylobactor jejuniUKCattle
060Campylobactor jejuniUKSheep
062Campylobactor coliNetherlandsChicken
063Campylobactor coliUK (Northern Ireland)Chicken
064Campylobactor coliDenmarkChicken
065Campylobactor jejuniDenmarkChicken
066, 067Campylobactor coliNetherlandsPig
068, 069Campylobactor coliUK (Northern Ireland)Pig
070–072Campylobactor coliDenmarkPig
073Campylobactor jejuniNetherlandsCattle
075Campylobactor jejuniUK (Northern Ireland)Cattle
076–078Campylobactor jejuniNetherlandsChicken
082Campylobactor coliUK (Northern Ireland)Chicken
083–085Campylobactor jejuniUK (Northern Ireland)Chicken
086–090Campylobactor jejuniUKChicken
091–092Campylobactor jejuniDenmarkChicken
093Campylobactor jejuniDenmarkHuman
094–096Campylobactor jejuniDenmarkChicken
099–101/103Campylobactor jejuniUKHuman/waterOutbreak?
104–107Campylobactor jejuniUKSheep
109–113Campylobactor jejuniSwedenDog

DNA extraction

Deoxyribonucleic acid was extracted using Easy-DNA (Invitrogen Life Technologies, formerly Gibco BRL, Paisley, UK) or PureGene (Flowgen, Ashby-de-la-Zouch, UK) commercial kits at Institutes 1 and 3 respectively. Institute 2 used a phenol–chloroform purification method, as described previously (Harrington et al. 1997), except that the final RNase treatment of extracted DNA was omitted.

PCR amplification and restriction digestion of the flagellin gene(s)

Three fla typing protocols were investigated during the course of this study. For comparative purposes, primer anneal sites for each PCR assay are listed in Table 1, alongside their anticipated gene target(s) and expected amplicon sizes. The given nucleotide positions and amplicon sizes may differ slightly from previously published values, as a result of small (up to 12 bp) inter-strain variations in the length of the flaA gene (Harrington et al. 1997). PCR reactions for each method were performed according to published protocols, with minor modifications where necessary, as listed below. Aliquots of each PCR reaction (typically 5–10 μl) were checked for the presence of the expected amplicon by electrophoresis on 1–2% agarose, according to standard protocols (Sambrook et al. 1989). Subsequent restriction digestions of PCR products with DdeI or HinfI were as recommended by the enzyme vendors. Enzymes were purchased from Invitrogen Life Technologies, excepting DdeI and HinfI used at Institute 1, which were from Roche Diagnostics (formerly Boehringer Mannheim, Lewes, UK) and New England BioLabs (Hitchin, UK) respectively. Reaction buffers for restriction digestion were used at 1X final concentration, and incubation was at 37°C for 3 h or 16 h. Variations with respect to the volume of PCR product used, total reaction volume, and amount of digested product used for subsequent analysis by gel electrophoresis are listed below.

Method 1 (Institutes 1 and 2). Primers were as described by Nachamkin et al. (1993), and experimental procedures as modified by Nachamkin et al. (1996). Anneal sites for the forward (A1, 5′-GGA TTT CGT ATT AAC ACA AAT GGT GC-3′) and reverse (A2, 5′-CTG TAG TAA TCT TAA AAC ATT TTG-3′) primers correspond to the 5′-most and 3′-most ends of the flaA gene, with an anticipated amplicon size of 1725 bp. Both institutes used purified DNA as template (50–100 ng per 100 μl reaction) in place of heat-treated cell lysates. A reduced anneal temperature of 45°C was used for isolates giving weak or no product at the published anneal temperature of 55°C. All other PCR reaction conditions were as published, including the source of Taq polymerase (Invitrogen Life Technologies). PCR product (5 μl) was digested using 2·5 U (or 1 U) of DdeI restriction enzyme in a final reaction volume of 30 μl (or 20 μl) at Institute 1 (and 2) respectively. Half the restricted sample was subjected to electrophoresis at Institute 1, whereas the full sample was loaded at Institute 2.

Method 2 (Institutes 1 and 3). This protocol was as described by Santesteban et al. (1996), based on the primers initially described by Alm et al. (1993b). There is a 1-base misprint in the primer sequence published by Santesteban et al. (1996) for the mixed synthesis reverse primer, RAA19, which should read 5′-GCA CC(CT) TTA AG(AT) GT(AG) GTT ACA CCT GC-3′. This has anneal sites 274 bp upstream from the 3′-ends of both flaA and flaB, but only the flaA gene is amplified in conjunction with the forward primer (pg50, 5′-ATG GGA TTT CGT ATT AAC-3′), as the latter is specific for the 5′-most region of the flaA gene alone. Anticipated amplicon size is 1457 bp. Purified DNA (50–100 ng per reaction) was used at both institutes, irrespective of whether or not isolates were DNase-negative. Institutes 1 and 3 performed 35 and 25 PCR cycles, and used final reaction volumes of 100 and 50 μl respectively. All other PCR reaction conditions were as published except that Institute 3 used Taq polymerase from Perkin Elmer (Cambridge, UK). Separate digestions were performed with DdeI and HinfI, using 4 μl PCR product and 10 U of enzyme in a final reaction volume of 30 μl at Institute 1, and 17·5 μl of product with 5 U enzyme in a final volume of 20 μl at Institute 3. Half the sample was subjected to electrophoresis.

Method 3 (Institutes 2 and 3). Primers for the multiplex PCR reaction were as described by Ayling et al. (1996). Forward primers anneal to the 5′-most end of flaA (Cj 431) or flaB (Cj 432), and both contain a non-specific 10 bp cloning sequence, depicted in lower case in the following sequences: Cj 431, 5′-aaaggatccgCGT ATT AAC ACA AAT GTT GCA GC-3′; Cj 432, 5′-aaaggatccgAGG ATA AAC ACC AAC ATC GGT-3′. The reverse primer, Cj 433, anneals 233 bp upstream from the 3′-end of both flaA and flaB, and has the following sequence: 5′-GAT TTG TTA TAG CAG TTT CTG CTA TAT CC-3′. Expected products are a 1499 bp amplicon from both flaA and flaB, 10 bp of which is contributed by the cloning sequence. PCR reaction conditions were modified from those described by Ayling et al. (1996). Briefly, each 50 μl reaction contained 1 μmol l−1 final concentration of each primer, 500 μmol l−1 final concentration of each dNTP, 3 mmol l−1 MgCl2, 2·5 U Taq polymerase (Invitrogen Life Technologies at Institute 2; Perkin Elmer at Institute 3), and 1X concentration of the supplied Taq polymerase buffer in place of Saiki buffer. Purified DNA (50–100 ng per reaction) was used as template at both institutes. Glycerol at 10% (v/v) final concentration was included when problems were experienced with non-specific products. Thermal cycling parameters were as published, except that a 1·5 min hold step at 58°C was omitted at Institute 3, prior to the final extension of 5 min at 72°C. PCR product (12 μl or 17 μl at Institutes 2 and 3 respectively) was digested with 10 U of DdeI in a final reaction volume of 20 μl. Half the restricted sample was subjected to electrophoresis.

Gel electrophoresis and image acquisition

The same molecular weight standard, a 100-bp ladder (G210A, Promega, Southampton, UK), was used for normalization of gel images at each institute. This was loaded (325, 260 or 870 ng per lane at Institutes 1, 2 and 3 respectively) in every third or fourth lane across the gel, including (usually) the outermost two lanes used. Under the electrophoresis conditions specified below, the 100-bp band of the marker ran within 1–2 cm of the gel bottom.

Institute 1. Hoefer HE 33 Minigel systems (Pharmacia Biosystems, Allerød, Denmark) were used with 1X TBE as running buffer (1X TBE is 0·089 mol l−1 Tris, 0·089 mol l−1 boric acid, 0·002 mol l−1 disodium EDTA, pH 8·0). Agarose (Seakem; FMC BioProducts, Rockland, ME, USA) was prepared by boiling in 1X TBE (40 ml per gel). Gels were poured in casting trays (7 cm wide by 10 cm long) using a 12-well comb, and were allowed to set for at least 30 min prior to use. Digested PCR products were run on 2·5% agarose at 90 V for 90 min. Gels were stained by immersion for 20 min in 1X TBE containing 4 μg ml−1 ethidium bromide. Digital images were captured under UV transillumination using a GFS-1000 gel documentation system (Techtum Lab AB, Umeå, Sweden).

Institute 2. Hybaid Midigel systems (Ashford, UK) were used with 1X TBE as running buffer. Gels (100-ml volume) were prepared as above in casting trays of 11·5 cm width by 14 cm length, using 20-well combs. SeaKem agarose (2%) was used for digested products generated using Method 1, with a run of 90 V for 3–4 h. Method 3 restriction products were run (90 V for 3–4 h) on 2% NuSieve GTG agarose (FMC BioProducts) combined with 0·7% SeaKem agarose. Gels were stained by immersion for 20–30 min in water containing 0·5 μg ml−1 ethidium bromide. Photographs were taken under UV transillumination using type 667 Polaroid film. These were digitized using a desktop digital scanner (HP Scanjet 4300C, Palo Alto, CA, USA).

Institute 3. Wide Mini-Sub Cell GT systems (Bio-Rad, Hemel Hempstead, UK) were used with 1X TAE as running buffer (1X TAE is 0·04 mol l−1 Tris–acetate, 0·001 mol l−1 disodium EDTA, pH 8·0). Gels were prepared in 1X TAE (100 ml volume) and were poured in casting trays of 15 cm width by 10·2 cm length, using 20-well combs. Digested PCR products were run at 120 V for ca 70 min using 2% NuSieve GTG agarose combined with 0·5% standard grade agarose. Gels were stained by immersion for 20–40 min in 1X TAE containing 0·5–0·75 μg ml−1 ethidium bromide. Digital images were captured under UV transillumination using a Gel Doc 1000 system (Bio-Rad).

Data analysis

Computer analysis of the images was performed using GelCompar version 4·1 and BioNumerics version 2·0 (Applied Maths, Kortrijk, Belgium). Normalized data were imported centrally into a single BioNumerics database for subsequent assignment of bands and data manipulations. GelCompar data from Institute 2 was imported using BNEXPORT software and procedures recommended by the manufacturer (Applied Maths), whereas BioNumerics data from Institute 3 was merged with the central database by copying files directly into appropriate folders. Band positions were assigned manually, and cluster analysis was performed using the band-based Dice coefficient with unweighed pair-group method using average linkages (UPGMA) clustering.


PCR typeability

This was determined using the full (n = 100) CAMPYNET isolate set, with each PCR method run at two different institutes. The same results were obtained between paired institutes. A significant number of isolates gave a weak product (n = 10) or failed to generate a product (n = 2) using the published anneal temperature of 55°C for Method 1. Reducing the anneal temperature to 45°C allowed generation of a strong product for all but two of these isolates, CNET 034 (C. jejuni) and CNET 063 (C. coli), thereby increasing typeability to 98%. The specificity of PCR for the remainder of the CAMPYNET sample set (n = 88) was not affected by this modification. Method 2 gave 100% typeability; all isolates generated strong products under the PCR assay conditions described. The situation was more complex with respect to Method 3. The majority of isolates (n = 97) gave a strong product, whereas three isolates (CNET 085, 086 and 089; all C. jejuni) failed to generate a product. However, the method was inconsistent with respect to amplification of a 900 bp non-specific product. The published protocol (Ayling et al. 1996) required inclusion of glycerol in the PCR assay at 10% (v/v) final concentration, without which the majority of isolates generated the non-specific band. Inclusion of glycerol, as published, suppressed the non-specific band for most isolates, but did not completely eradicate it. Four isolates (CNET 024, 088, 105 and 106; all C. jejuni) still gave two products, hence the PCR typeability for this method was reduced to 93% overall.

DdeI profile features and reproducibility for each method: evaluation of data quality

For each method, profiles were identical within epidemiologically related sets, but were distinct between sets (Fig. 1). Bands were reliably assigned (all methods) to 100 bp at Institutes 1 and 2, despite use of different image acquisition systems (Fig. 1). Image resolution was not so clear at Institute 3, for which 140 bp was the limit of dependable band detection (data not shown). Method 1 generated three to seven bands (usually five or six) over the size range 100–1200 bp, and typically contained one extra band of ca 200–280 compared with Method 2 (Figs 1 and 2). Sometimes, the only difference was that one band was shifted to a higher molecular size for Method 1, as seen for group 5 isolates (Fig. 1a,b). These differences allowed improved discrimination by Method 1 of certain isolates giving visually similar profiles using Method 2 (e.g. groups 1 and 3; groups 2, 5 and 7). An additional band of ca 120 bp was typically generated by Method 3 compared with Method 2, but group 7 profiles also contained a second additional band at ca 800 bp (Fig. 1b,c). The sum of Method 3 band sizes therefore exceeded the size of the initial PCR amplicon for these isolates, but otherwise, the initial amplicon size (Table 1) was not exceeded for any method. Occasional partial digest bands (indicated by arrows) were easily differentiated from ‘real’ bands for Methods 1 and 2 (Fig. 1a,b), but were harder to diagnose for Method 3, because of their higher intensity (Fig. 1c), especially at Institute 3 (data not shown). For these reasons, in addition to the problems experienced with a non-specific band during PCR, results from Method 3 were not further investigated in this study.

Figure 1.

Polymerase chain reaction (PCR)/restriction fragment length polymorphism (RFLP) (DdeI) profiles for isolates (n = 22) representing seven different epidemiological groups. (a), Method 1; (b), Method 2; and (c), Method 3. The dendrogram is based on cluster analysis of Method 1 profiles using the Dice coefficient with unweighed pair-group method using average linkages (UPGMA) clustering (1% optimization, 1% tolerance). Epidemiological groups are indicated in parentheses next to the appropriate branch on the dendrogram. The molecular weight scale is given in base pairs above the gel images for each method. Arrows indicate PCR/RFLP bands that are generated through partial digestion

Figure 2.

Polymerase chain reaction (PCR)/restriction fragment length polymorphism (RFLP) profiles and assigned profile types for isolates representing the full (n = 100) CAMPYNET isolate set. (a), Method 1 (Dde I); (b), Method 2 (DdeI); and (c), Method 2 (HinfI). An asterisk (*) by the isolate number indicates Campylobactor coli. The dendrogram is based on cluster analysis of Method 1 profiles using the Dice coefficient with unweighed pair-group method using average linkages (UPGMA) clustering (1% optimization, 1% tolerance). Epidemiological sets are indicated next to the appropriate branch on the dendrogram. The molecular weight scale is given in base pairs above the gel images for each method. Arrows indicate PCR/RFLP bands that are generated through partial digestion

Establishing parameters used for numerical cluster analysis of DdeI profiles

Optimal parameters for cluster analysis were defined as the most stringent settings giving 100% similarity within epidemiologically related sets (Table 2), but still allowing differentiation between these sets. The UPGMA clustering method was used with the Dice (band-based) similarity coefficient, and data were compared over a size range of 120–1500 bp. Using data from a single laboratory, parameters were defined as 1% optimization and 1% position tolerance. Figure 1 shows the dendrogram obtained using Method 1 profiles derived from each of the seven epidemiologically related groups, all of which were differentiated. Seven distinct groups were also obtained using Method 2 profiles (dendrogram not shown).

Numerical cluster analysis of DdeI profiles; discriminatory power

Cluster analysis was performed on a set of 98 (Method 1; excludes two untypeable isolates) and 100 (Method 2) profiles using the parameters defined above. Method 1 data fell into 40 phenons at the 100% similarity level, each of which was designated a distinct DdeI profile type, reflecting the order of branching in the dendrogram (Fig. 2). Method 2 DdeI data (Fig. 2b) and HinfI data (Fig. 2c) are also shown, together with the profile types assigned for all three data sets (Fig. 2). Clustering of Method 2 data yielded 36 phenons at a 93·3% cut-off level for identity (data not shown). Whilst most groups clustered at 100% similarity, Sant_05 and 11 profile groups were defined at 93·8% and 93·3% respectively (data not shown). Of the two isolates untypeable by Method 1, one (CNET 034) generated a distinct Method 2 profile type (Sant_36), whereas the other (CNET 063) fell into an existing profile type (Sant_05). Neither dendrogram distinguished overall between C. jejuni and C. coli, with some profile types (Nach_06, 26; Sant_05, 11-A, 16, 21) shared between species (Fig. 2). One profile type for each method (Nach_13 and Sant_11) contained two visually similar, but distinct profiles, which could not be differentiated under the clustering parameters used. These were designated as subtypes A and B (Fig. 2). Thus, in total, Methods 1 and 2 generated 41 and 37 profile types respectively (Table 1). The distribution of isolates among types differed slightly between methods, with more unique profiles generated by Method 1, and with fewer isolates in profile groups compared with Method 2 (Fig. 2). A slightly higher Simpson discriminatory index was obtained for Method 1 (0·973) compared with Method 2 (0·970), as calculated using the method of Hunter and Gaston (1988).

Enhanced discrimination using HinfI (Method 2)

Fourteen distinct HinfI profiles were obtained, each containing two to four bands within the size range 100–1200 bp (Fig. 2c). Profiles were assigned visually, as so few bands were available for analysis. As with DdeI profiles, partial digest bands were sometimes a problem, especially at higher molecular sizes (indicated by arrows on Fig. 2c). The most common profile (Sant_H_01) accounted for 55% (47 of 85) of isolates, when considering only one isolate from each epidemiologically related group. Six DdeI Method 2 profile groups (Sant_01, 05, 11-A, 16, 19, 28) were further subdivided using HinfI, increasing the number of types from 37 (DdeI alone) to 46 combined DdeI and HinfI types (Fig. 2c).

Inter-laboratory reproducibility

This was assessed for Methods 1 and 2 using a subset of 50 isolates, including three isolates from each of five epidemiologically related sets. Method 1 data were compared from 120 to 1500 bp, whereas Method 2 data were compared from 140 to 1500 bp. Clustering parameters were as defined above. For Method 1, all 50 replicate pairs were found to identify at 100% similarity, including triplets of isolates from epidemiologically related sets. A comparison of the 35 replicate sporadic pairs alone also resulted in identification at 100% similarity for each replicate pair (Fig. 3). Clustering was not so straightforward with the 50 replicate pairs for Method 2 profiles (dendrogram not shown). Three replicate pairs (CNET 039, 040, 100) failed to identify at 100%, and one profile group (Sant_20), previously containing isolates at 100% similarity, was split into two branches related at 96·1%, on inclusion of data from the second institute. Increasing the optimization setting to 2% allowed identification of all replicate pairs at 100%, except for CNET 040 from one institute, which then misidentified as Sant_19 instead of Sant_20. It was later found that profiles from CNET 039, 40, 100 had been run outside the outermost markers on their gels at one institute, and were therefore not accurately normalized. Removing these profiles from the Method 2 data set allowed identification of all isolates within their appropriate groups, and identification at 100% of all remaining replicate pairs (data not shown).

Figure 3.

Inter-laboratory reproducibility of DdeI profiles (Method 1) for a subset (n = 35) of sporadic isolates. The dendrogram is based on cluster analysis of replicate pairs using the Dice coefficient with unweighed pair-group method using average linkages (UPGMA) clustering (1% optimization, 1% tolerance). The molecular weight scale is given in base pairs above the gel images. Isolate numbers with and without an appendage of _1 were run at Institute 2 and Institute 1 respectively


PCR Typeability

Methods 1, 2 and 3 gave 98, 100 and 93% typeability respectively (Table 1). Typeability has been reported previously at 100% for all three methods (Nachamkin et al. 1993, 1996; Ayling et al. 1996; Santesteban et al. 1996). However, the CAMPYNET sample set was more diverse than sample groups used in these previous studies, which may account for the apparent discrepancies. Typeability levels in the current study were determined at two different laboratories for each method, and were found to be the same between laboratories, indicating that different DNA extraction methods had no apparent effect on typeability. Other investigators have advocated time-saving template preparations such as simple cell-suspension boiling (Nachamkin et al. 1993) or PCR direct from frozen culture storage beads (Santesteban et al. 1996; Owen and Leeton 1999). The former method allows for ca 85% typeability (Nachamkin et al. 1993), with failed reactions attributable to residual DNase activity, not fully inactivated by boiling (Santesteban et al. 1996). Similar results were found for the bead method, with untypeable isolates made typeable using extracted DNA (Santesteban et al. 1996). During the current study, boiled lysates were not stable when stored at −20°C, and for this reason, as well as the failure rate seen for the boiling method, extracted DNA was the template of choice.

With respect to untypeable isolates in the current study, it seems most likely that the reverse primers were at fault when no product was amplified (n = 2 and 3 for Methods 1 and 3 respectively), as all three flaA-specific forward primers share significant overlap, whilst the reverse primers target different sites (Table 1). The typeability for Method 1 was determined after reducing the anneal temperature of the PCR reaction from 55 to 45°C. This was to allow for improved amplification of weak products obtained from 10 of the isolates. Weak products (22 of 65) were also described in the original publication of this method, and were compensated for by ethanol precipitation prior to digestion (Nachamkin et al. 1993). The lower typeability of Method 3 (93%) was partly as a consequence of inconsistent amplification of a non-specific band of 900 bp. Inclusion of glycerol helped suppress this band, but did not completely eradicate it. It has since been determined (unpublished data) that the 900 bp band is amplified by the two ‘forward’ primers described by Ayling et al. (1996), and partial sequence analysis (156 bp) has shown it to have ca 97% homology to a putative NTPase gene (Cj 0581) detected in the genome sequence of C. jejuni NCTC 11168 (Parkhill et al. 2000; website, The problem with the non-specific band might be alleviated by increasing the anneal temperature used during PCR amplification, or by synthesis of primers without the 5′-terminal cloning sequences. However, such work was beyond the scope of this study.

Quality of data

Whilst each institute used a standardized PCR protocol, the procedures used for subsequent restriction digestion, electrophoresis and image capture were different, and this is reflected in the quality of the data obtained. Bands were clearly visible down to 100 bp at two institutes (Fig. 1) but could be detected only down to 140 bp at the remaining institute (data not shown). Loss of resolution at low-molecular mass was later found to be as a consequence of faulty UV transillumination equipment. Attention to image capture is essential for subsequent data manipulations. Unless the 100 bp marker and sample bands of that size can be seen on the final digitized image, then significant loss of data will occur, as one to four bands occur between 100 and 200 bp for all three methods tested (Fig. 1). Incomplete digestion of product also complicated band allocation. In general, partial digest bands (indicated by arrows in Fig. 1 and Fig. 2c) were easily identified, since their intensity was below that of the next smallest fragment, in contrast to the general trend that band intensity increased with size. Partial digest bands were especially a problem at Institute 3, where bands determined as ‘real’ or ‘partial’ at other institutes, could sometimes be seen with equal intensity (data not shown). Higher volumes of PCR product were digested at the latter institute, and the most likely explanation for partial digestion is inhibition of the restriction digest reaction by components carried over from PCR. Such problems have not been directly noted in other studies, but previously published profiles show a mixture of faint and intense bands, especially at higher molecular sizes (Ayling et al. 1996; Camarda et al. 2000; Shreeve et al. 2000). Purification of the amplicon prior to digestion would alleviate these problems, as would a reduction in the quantity of PCR product used. In either case, optimization of restriction digestion is clearly essential to generating high quality data.

Discriminatory power (DdeI) in comparison with other genotyping methods

The Simpson discrimination (D)-index represents the probability of distinguishing two strains chosen at random from a population of unrelated strains (Hunter and Gaston 1988). A slightly higher D-index was identified for Method 1 (0·973) compared with Method 2 (0·970), because of the higher number of profiles obtained (41 compared with 37), and to the lower numbers of isolates distributed amongst groups of two or more (Fig. 2). Previous studies have reported discrimination levels of 0·92 (Method 3; Koenraad et al. 1995) and 0·960 (Method 1; Nielsen et al. 2000). The latter study also determined the D-index for automated HaeIII-ribotyping (0·945), PFGE using SmaI (0·974), and fluorescent RAPD profiling (0·984). These levels reflected 40, 50, and 56 types respectively, from 90 isolates, compared with 40 types generated by fla typing. A similar study (n = 50 isolates) identified 26, 31, 38 or 41 types by automated PstI-ribotyping, fla typing, PFGE or AFLP respectively (de Boer et al. 2000). Although greater numbers of profiles are generated by PFGE, AFLP and RAPD, the discriminatory levels of these methods are similar to that of fla typing (all >0·95), and fla typing remains the simplest, most widely accessible, and most cost-effective method.

Enhanced discrimination using a second enzyme

This study found that HinfI alone was not very discriminatory, generating only 14 distinct HinfI profiles, one of which (Sant_H_01) represented >50% of the isolates (Fig. 2c). Low discrimination by HinfI has been reported previously (Owen et al. 1994; Ayling et al. 1996; Santesteban et al. 1996). Despite this, HinfI was found here to subdivide six Method 2 DdeI profiles, resulting in 46 combined DdeI/HinfI types, compared with 37 DdeI types (Fig. 2). This represents a 25% increase in number of types. Higher levels (ca 50%) of enhanced discrimination by HinfI have been reported, but this may reflect the more conserved nature of the sample group used during that study (Santesteban et al. 1996). When a single enzyme is considered, DdeI has repeatedly been confirmed as more discriminatory than HinfI, PstI or EcoRI; whilst AluI has been found to generate bands too small to be practical for analysis (Wassenaar and Newell 2000; Petersen and Newell 2001). Thus, DdeI is confirmed as the most appropriate first choice of enzyme.

Enhanced discrimination using flaA in combination with flaB

It has been suggested that analysis of flaB in addition to flaA might provide additional discrimination (Alm et al. 1993b; Ayling et al. 1996). We aimed to investigate this possibility by comparing the multiplex PCR assay of Method 3 with the flaA-based assay of Method 2, both of which amplify an almost identical region of the gene(s) (Table 1). However, this comparison was not possible, as a consequence both of a non-specific PCR product, and of difficulties in band assignment for the multiplex assay, whose profiles were a mixture of faint and intense bands. A previous study attributed differences in band intensity to selective amplification of flaA over flaB during multiplex PCR (Petersen and Newell 2001). In the current study, flaB products were often weaker than flaA products when the genes were amplified separately (data not shown), also suggesting selective amplification of flaA over fla B. Separate amplification and restriction digestion of the two genes would overcome this, but would also add to the time and cost of the method. Three studies have reported an increase in discrimination of only ca 10% using flaB in addition to flaA (Alm et al. 1993b; Mohran et al. 1996; Petersen and Newell 2001), such that, overall, the benefits of analysing flaB in addition to flaA appear minimal, especially if separate PCR reactions are required.

Inter-laboratory reproducibility

Inter-laboratory reproducibility, by means of cluster analysis following electronic transfer of digitized, normalized profiles, has not been investigated previously for fla typing. The current study assessed this for Methods 1 and 2, using a subset of 50 replicate isolates. For Method 1, all replicate pairs were found to identify at 100% similarity (Fig. 3), indicating high inter-laboratory reproducibility, despite pronounced methodological variations between the two different institutes involved. For Method 2 data, all replicate pairs identified at 100% similarity, provided that three profiles from one institute were excluded. These profiles had not been adequately normalized, having been run outside the outermost markers on their gels. This shows that it is essential to run markers at regular intervals across the gel, including the outermost two lanes used, where the effects of ‘smiling’ and ‘frowning’ are at their most extreme.

Stability of flaA typing compared with other genotyping methods

The consequences of genetic instability on fla typing and other genotyping methods have been reviewed recently for Campylobacter (Wassenaar and Newell 2000; Wassenaar et al. 2000). There is evidence for recombination in the flagellin locus of mutant strains (Alm et al. 1993a; Wassenaar et al. 1995), and in natural Campylobacter populations (Harrington et al. 1997; Meinersmann and Hiett 2000). However, other genetic loci also appear to be subject to recombination (exchange of alleles), as indicated by MLST data on housekeeping genes in C. jejuni (Dingle et al. 2001; Suerbaum et al. 2001). Even so, a recent study found fla typing to be stable, along with AFLP and MLST, whereas PFGE types showed considerable genomic rearrangement (de Boer et al. 2002). Instability of PFGE profiles has been observed previously, both in vitro (On 1998) and in vivo (Wassenaar et al. 1998; Hänninen et al. 1999). Whilst considerations of genetic instability cannot be overcome, the practical solution is to use a combination of typing methods (Wassenaar and Newell 2000; de Boer et al. 2002). Fla typing presents a useful first choice method, as it is cheaper, more widely available, less time consuming, and less labour intensive than other methods.


The widespread use of flagellin gene typing is testimony both to its accessibility and its ease of use. However, global variations in procedure with respect to both primers and restriction enzymes used, do not allow for direct comparison of profiles generated in different laboratories. There is consequently a need for standardization of methods (Wassenaar and Newell 2000), which the current study aimed to address. Increased discrimination was found using the full, as opposed to a shorter segment of the flaA gene, whilst DdeI, the most commonly used enzyme for fla typing, was found to be far more discriminatory than HinfI. Our recommendation is therefore to use the full flaA gene for analysis, followed by restriction digestion with DdeI. The method of Nachamkin et al. (1993, 1996) fulfils these criteria, although modification of the reverse primer may be necessary to increase typeability from 98 to 100%, and to obviate the need for reduced anneal temperature during PCR for a minority (10%) of isolates. In this study, the method resulted in a D-index similar to those reported previously for PFGE and ribotyping, and allowed us to achieve 100% inter-laboratory reproducibility, provided that adequate attention was paid to generating high quality data. The method is therefore robust, and is suited to multi-centre projects, although ideally a second typing method should also be applied. It is our hope that this study will allow progress towards standardization of fla typing between laboratories.


This study was facilitated and funded in part by the European Commission project SMT-PL97-9508.