The twelve isolates representing five distinct PFGE genotypes were analyzed using MLST 14. Three different STs belonging to three clonal complexes were found among the isolates (Figure 1). Human and chicken isolates of the same PFGE genotype also had the same ST. All isolates within each ST shared the alleles in all seven loci investigated (data not shown).

On average 1,477 non-redundant probes (92%: range from 1,408 to 1,499) remained after the data preprocessing step and generated a valid log₂ ratio (M-value). Several of the excluded probes had low signal intensities in both channels and represented probes that failed during array production. To estimate the noise level associated with the experiment and to define M-value thresholds for sequence divergence, we analyzed the M-value distribution of two self-self hybridizations of reference strain DNA prepared at two separate occasions (Figure 2A). As expected for a data set with low technical noise, these hybridizations had narrow M-value distributions (sd = 0.17 and 0.14). Only a few probes had absolute M-values of >0.5 (5 probes (0.3%) for the first hybridization, and 22 (1.3%) for the second hybridization). One single probe (0.06%) in the second hybridization had an absolute M-value of >0.75.

Next, a comparison between the sequenced strains NCTC 11168 and RM1221 allowed for estimation of the effect of sequence divergence on the M-values. We used sequence similarity searches to match each probe with the genome sequence of the RM1221 strain and to calculate the number of matching nucleotides (all probes have a 100% match in the genome of NCTC 11168). There were in total 160 probes (10%) with less than 55 matching bases, putatively measuring highly divergent genes or genomic regions absent in RM1221. Approximately 51% (811/1,601) of the probes had a perfect match in RM1221, while 21% (336/1,601) had a one-base mismatch. A reduction in the number of matching bases displayed a strong association with reduced hybridization signals, and hence a trend towards lower M-values in the comparison between RM1221 and NCTC 11168 (Figure 2B, Table 1). Interestingly, already a one-base divergence resulted in a significant change in the M-values at the global level (shift of mean value from 0.02 to -0.12; p < 10^-16, one-sided, two-sample t-test).

We further used the RM1221 vs. NCTC 11168 hybridization data to define an M-value threshold for detection of divergent genes. The effect of M-value threshold on sensitivity and specificity for detection of probes with at least one, two, three or four mismatches to the reference strain was plotted (Figure 3). We used this information, in combination with the self-self hybridization data, to define a conservative M-value threshold of less than -0.75 for divergent or absent genes. At this stringency level, none of the probes with a perfect match in the RM1221 genome were erroneously identified as divergent in comparison with NCTC 11168. On the other hand, using this M-value cutoff only 32% of the probes with ≥1 mismatch could be detected as divergent (Figure 3A). The fraction of correctly identified divergent probes increased to 93% when the analysis was restricted to probes with at least four mismatches (94% sequence identity, Figure 3D). Collectively, the analyses indicate that the oligonucleotide platform does not permit reliable genotyping at the single-base level, but can with high confidence be used to identify absent or divergent genes (≥4 mismatches, i.e. sequence identity ≤94%). At this level the probability of false-positives is low and the sensitivity to detect ≥ 4-bp mismatches is high (>93%), providing a reasonable balance between false positives and false negatives.

The isolates included in the study showed large differences in the number of probes (range from 30 to 420) that were sequence divergent compared to the reference strain. In total 29% (439/1,527) and 17% (253/1,527) of the probes were variable in more than two and three of the studied isolates, respectively. These numbers are in line with previous CGH studies 25262728293031. The variable probes represented genes distributed over the entire chromosome, and showed some local clustering (Figure 2C). Probes exhibiting M-values of >0.75 were classified as representing genes with higher copy numbers in the test isolate compared to the reference strain NCTC 11168. Genes with higher copy numbers were detected in five of the test isolates (three genes in C12, one in C36, nine in H467, six in C20, and two in H312). In isolates C12 and C20 two consecutive genes with higher copy numbers could be detected (C12: Cj0078c and Cj0079c, C20: Cj0967 and Cj0968, and Cj1419c and Cj1420c). The M-values for all probes are available through the ArrayExpress data repository (accession number E-TABM-460).

We next carried out a hierarchical clustering analysis using the microarray data to identify similarities among the isolates. The origin of the isolates (chicken or human) had no effect on the clustering. Instead, the isolates clustered into groups similar to those obtained by PFGE and MLST. Three major clusters were identified (Figure 4A). The first included all isolates with PFGE types A, B and C, belonging to the MLST ST-21 clonal complex. Within this cluster, the two isolates of PFGE type B clustered together, while another cluster was formed by the two type C and three type A isolates. The remaining type A isolate clustered outside this tight cluster. The microarray data indicated that this cluster of isolates is similar to the reference strain NCTC 11168 (Figure 4A). This relatedness is further supported by the MLST data; strain NCTC 11168 belongs to the same clonal complex as the type A, B and C isolates (ST-21 clonal complex), although it is of a different sequence type (ST-43). The second cluster included PFGE type D isolates (MLST clonal complex ST-677) and strain RM1221, and the third PFGE type H isolates (MLST clonal complex ST-48).

Previous CGH studies of C. jejuni have identified 18 genomic regions enriched for genes with diverging sequences 2729. We analyzed the presence of variable genes in the five groups of isolates (A, B, C, D and H) defined by PFGE and microarray clustering. A region was confirmed variable if the calculated average M-value of the group was <-0.75 for at least one of the probes in the region. Using this approach, we noted that all of the previously identified variable regions differed in at least one of the five isolate groups, and further that four variable regions (regions 1, 9, 12 and 13) showed divergence in all five groups (Figure 4B). Also region 11 was highly variable, with four isolate groups showing variability in the region. We further mined our data for additional regions that showed variability in multiple isolates. We found three additional regions, spanning genes Cj0137–Cj0145 (region 19), Cj0356c-Cj0360 (region 20) and Cj1047c-Cj1069 (region 21).

We further used the Clusters of Orthologous Groups of proteins (COG) database 33 to analyze the functional group assignments of the variable genes (M-values of <-0.75 for the corresponding probes). In all isolates genes from multiple COG categories were found variable, indicating that sequence divergence is not restricted to genes encoding specific functions (Figure 4C). Furthermore, using Fisher exact test we identified a significant overrepresentation of divergent genes in the COG category M (cell wall/membrane/envelope biogenesis) and V (defense mechanisms) in several of the isolates (Figure 4D). A strong enrichment of the same categories was observed when the analysis was restricted to genes with high sequence divergence (i.e., M-values of <-2.0). However, when the analysis was carried out using genes with moderate sequence divergence (M-values between -2.0 and -0.75), no significant enrichment of COG categories could be observed. These results suggest that the moderate sequence divergence reflects normal interstrain variability unlikely to affect protein function in any substantial way. Furthermore, the probes with moderate sequence divergence seem to be distributed over the entire length of the chromosome, while the probes with high sequence divergence seem to be more tightly clustered (Figure 2C).

Campylobacter spp. isolates (n = 90) were collected from all reported cases of domestically acquired campylobacteriosis in four Swedish regions within the scope of a study conducted in July through October 2003. During the same time period and in the same geographical areas, fresh poultry products from retail were purchased and analyzed for campylobacter. Isolates from both patients and poultry were species identified by polymerase chain reaction 35 and the C. jejuni isolates subtyped using PFGE and the restriction enzyme SmaI as earlier described 7. Isolates sharing PFGE genotype with at least one other isolate were further PFGE-genotyped using the restriction enzyme KpnI. Each unique banding pattern was assigned an identifying letter. For the current study, six pairs of C. jejuni isolates with identical SmaI and KpnI genotypes were selected to represent each of the KpnI genotypes A (two pairs), B, C, D and H (Figure 1). Each pair consisted of one chicken and one human isolate originating from the same geographical region. The two completely sequenced strains NCTC 11168 24 and RM1221 36 were also included.

All isolates were cultured on blood agar plates at 37°C for 48 h in a microaerophilic environment. Genomic DNA for MLST and microarray analyses was extracted using the DNeasy tissue kit (Qiagen, Hilden, Germany).

The seven loci used in the MLST 14 were polymerase chain reaction amplified using primers and conditions according to the C. jejuni MLST website http://pubmlst.org/campylobacter. Nucleotide sequences were obtained by sequencing of both strands using the BigDye Terminator v3.1 Cycle Sequencing kit and a 3130xl Genetic Analyzer (Applied Biosystems, Foster City, USA). Sequences were assigned allele numbers, and the sequence type (ST) and lineage (clonal complex) of each isolate was determined by interrogating the MLST database.

Campylobacter jejuni AROS v1.0 oligonucleotide probe set was purchased from Operon Biotechnologies (Cologne, Germany). The set consisted of 70-mer oligonucleotides representing 1,546 open reading frames (ORFs) from strain NCTC 11168, 51 ORFs from the virulence plasmid pVir from strain 81–176, and 4 ORFs from plasmid pCJ01 from strain 21190. The oligonucleotides were printed in triplicates on CodeLink Activated slides (GE Healthcare, Uppsala, Sweden) and processed according to the manufacturer's instructions. For probe annotation the version 1.3.2 (dated May 16, 2007) of the OMAD database 37 was used. Additional details on the microarray production are available through the ArrayExpress microarray data repository (accession number A-MEXP-925) 38.

Test DNA extracted from the twelve C. jejuni isolates was co-hybridized with reference DNA extracted from strain NCTC 11168 in one hybridization per isolate. Additional hybridizations were performed to compare strains NCTC 11168 and RM1221, and to compare two separate target preparations (culturing, DNA extraction, labeling and hybridization) of the NCTC 11168 strain to establish the magnitude of technical noise in the experimental setup. Both sets of additional hybridizations were carried out in a dye-swap manner.

For labeling, 3 μg genomic DNA in 21 μL water was sonicated for 30 s. Fluorescent Cy3 or Cy5 dyes were incorporated in a mixture of 15 μg random octamers, 40 U of Klenow enzyme (Invitrogen, Paisley, UK), 6 nmol of each dATP, dGTP and dCTP, 3 nmol dUTP, and 50 nmol Cy3-/Cy5-dUTP (GE Healthcare, Uppsala, Sweden) in 1 mM Tris pH 8.0, 100 μM EDTA. The mixture was incubated at 37°C for 2 h after which 5 μL 0.5 M EDTA was added. Unincorporated dye was removed using the Microcon-30 columns (Millipore AB, Solna, Sweden) and the dye incorporation efficiency measured using a Nanodrop ND-1000 spectrophotometer (Nanodrop, Rockland, USA). Reference DNA was combined with an equal amount of reciprocally labeled test DNA, dried down using a speed vac, resuspended in 100 μL hybridization buffer (5 × SSC, 50% formamide, 0.1% SDS, 0.1 μg/μL tRNA), and denatured for 2 min at 95°C. The samples were cooled on ice, transferred to the microarrays and hybridized for 16 h at 42°C. The arrays were then washed once for 5 min with 2 × SSC, 0.1% SDS at 42°C, once for 5 min with 0.1 × SSC, 0.1% SDS at room temperature, and four times for one minute with 0.1 × SSC at room temperature. All washing steps were done under agitation. The slides were dried by brief centrifugation at low speed.

The microarrays were scanned using a GenePix 4000B scanner (Molecular Devices, Sunnyvale, USA). Features were identified and fluorescence intensities extracted using the irregular feature-finding approach implemented in GenePix Pro 5.1 (Molecular Devices). Further analysis was carried out in the R environment for statistical computing 39 using the aroma 40, Bioconductor 41 and kth-packages 42. No subtraction of the local background was carried out, as this was found to slightly increase the variability between replicated features. A feature was considered unreliable and removed if: a) the feature contained less than 55 pixels, or if for both channels b) 10% or more of the pixels were below the signal intensity of the local background plus two standard deviations of the background, c) the signal-to-noise ratio was below 3, d) the signal was saturated, or e) the intensity was below the mean signal of negative controls (probes with random sequence). The design of the probe set is based on the genome sequence of the NCTC 11168 strain, and hence absent or sequence divergent genes in the test isolate (labeled in Cy5) compared to the reference strain (Cy3) show negative log₂ (Cy5/Cy3) ratio values (M-values). Data normalization was carried out in a block-wise manner assuming equal sums of the two channels using a non-divergent set of probes. These probes were obtained after removal of 20% of the probes with the most negative M-values. After normalization, replicates of each probe were averaged, discarding probes that had only one available measurement. The microarray dataset is available through ArrayExpress (accession number E-TABM-460) 38.