The origins and details of the Campylobacter isolates used in this study are given in Supplementary Table S1. Isolates were routinely cultivated on Blood Agar Base No. 2 (Oxoid) supplemented with 5% (v/v) bovine blood and incubated at 37°C under microaerobic conditions (6% v/v CO₂; 6% v/v O₂; 3.4% v/v H₂, and 80.6% v/v N₂). Growth experiments were carried out with and without 100 mM glucose supplementation in Tryptic Soy Broth (TSB) without Dextrose (Becton Dickinson and Co.) or in modified MCLMAN minimal media (Alazzam et al., 2011) without lactate and with 2 mM aspartate. Viable cell counts were determined by plating serial dilutions in phosphate buffered saline on Blood Agar Base No. 2 (Oxoid) and incubating under the conditions described above.

Campylobacter isolates (Supplementary Table S1) were screened for the ED pathway genes by PCR using oligonucleotides for edd (5′-ATAAATTGGGATGATTTTG and 5′-TCTAAACCCTGCAAAAAGCTC giving a ca 810 bp fragment) and eda (5′-GGAGAAAAAATGCAAACAA and 5′-TCTAAACCCTGCAAAAAGCTC giving a ca 600 bp fragment). Oligonucleotides for napA and napB (periplasmic nitrate reductase) were used in multiplex PCR reactions to discriminate C. jejuni subsp. doylei from C. jejuni subsp. jejuni and C. coli using the nap mpx2 primer set as described by Miller et al. (2007), which consists of internal napA primers and napB flanking primers. Chromosomal DNA was extracted with the DNeasy Blood and Tissue kit (Qiagen) and the DreamTaq Green DNA polymerase (Thermo Fisher) was applied for the PCR amplifications.

Genomic DNA was extracted from bacterial cultures of CCUG 18265, CCUG 24567, CCUG 26155, B13117, and CV1257 using the Wizard Genomic DNA Purification Kit (Promega, Southampton, UK). Whole-genome sequencing (WGS) was carried out at the Wellcome Trust Sanger Institute, UK. Illumina multiplex libraries were generated by acoustic shearing, after which up to 96 libraries were pooled in equimolar ratios and sequenced on a single flow cell lane on the Illumina HiSeq platform, producing 100 bp paired-end reads, as previously described (Cody et al., 2013). The short-read data were submitted to an automated pipeline (Jolley and Maiden, 2010; Bratcher et al., 2014), which integrates VELVET version 1.2.01 (Zerbino and Birney, 2008) and VELVETOPTIMISER version 2.2.0 (Zerbino, 2010) for de novo assembly. Draft genomes were then uploaded to the C. jejuni/coli PubMLST database (Supplementary Table S2).

Loci corresponding to glcP (CAMP2017), pgi2 (CAMP2018), glk (CAMP2019), pgl (CAMP2020), zwf (CAMP2021), edd (CAMP2022), and eda (CAMP2023) were added to the C. jejuni/coli PubMLST database (PubMLST locus names shown in parentheses), using sequences from C. jejuni subsp. doylei 269.97 (GenBank accession number NC_009707) to seed the database. The glc loci, in addition to MLST, rMLST, and C. jejuni/coli core genome scheme loci, were annotated in the study genomes using the Bacterial Isolate Genome Sequence Database (BIGSDB) ‘autotagger’ functionality implemented in PubMLST (Jolley and Maiden, 2010). The autotagger uses BLAST to search a genome for sequences similar to loci defined in the database. For sequences with ≥98% identity to existing alleles, the autotagger defined the position of the locus and assigned a unique allele number in order of discovery (Jolley and Maiden, 2010). Sequences with <98% identity to existing alleles were curated manually. The same approach was used to annotate components of the glc locus in publicly available genomes present in the C. jejuni/coli PubMLST database. A summary of the allelic data for MLST and glc loci from all ED-positive isolates was generated using the BIGSDB data export plugin. ED types were generated as for MLST sequence types: unique combinations of alleles across the glc loci, taken in gene order from glcP to eda, were assigned arbitrary numbers in order of discovery.

Genomic analyses were carried out using the hierarchical gene-by-gene approach implemented in BIGSDB, which allows users to compare isolates at varying numbers of loci, depending on the level of resolution required (Jolley and Maiden, 2010; Maiden et al., 2013). Species assignments of all ED-positive isolates, and C. coli clade membership, were confirmed using rMLST (Jolley et al., 2012; Jansen van Rensburg et al., 2016). A maximum likelihood tree based on concatenated nucleotide sequences of the glc genes was reconstructed in MEGA version 6.06 (Tamura et al., 2013) using the General Time Reversible model with gamma-distributed rates with 500 bootstrap replicates. The resulting phylogenetic tree was annotated online using the Interactive Tree of Life version 3 (Letunic and Bork, 2016). Relationships among ED-positive isolates were established based on wgMLST analyses carried out using the GENOME COMPARATOR module implemented in the C. jejuni/coli PubMLST database (Jolley and Maiden, 2010). Following the exclusion of genomes in which ≥1% of the 1,343 C. jejuni/coli core genome scheme loci were incomplete (i.e., at the ends of contigs), relationships among C. jejuni and C. coli isolates were evaluated using wgMLST. Isolates were compared to the reference genomes of C. jejuni NCTC11168 (GenBank accession number AL111168) (Parkhill et al., 2000; Gundogdu et al., 2007) or C. coli 15-537360 (CP006702) (Pearson et al., 2013), using the default GENOME COMPARATOR settings. Distance matrices generated by GENOME COMPARATOR were visualized as networks using the NeighborNet algorithm (Bryant and Moulton, 2004) in SplitsTree version 4.13.1 (Huson and Bryant, 2006). Further analyses were carried out to compare the isolates B13117 and CV1257. The genome sequences of these isolates were annotated using Prokka version 1.0 (Seemann, 2014). The number of shared and unique coding sequences was estimated with Roary version 3.6.0 (Page et al., 2015) using the default settings. The capsule gene regions of these isolates were compared using ACT (Carver et al., 2005).

BIOLOG^TM phenotype microarrays were set up according to the manufacturer’s description for C. jejuni. Briefly, individual isolates were cultivated overnight on Blood Agar Base No. 2 supplemented with 5% v/v bovine blood at 37°C in a microaerobic atmosphere. Cells were harvested from plates with 1.2x IF-0a solution (Biolog, Inc.), gently resuspended and adjusted to an optical density (600 nm) of 0.8. For each strain, 8 ml cell suspension was mixed with 12 ml 1.2x IF-0a (Biolog, Inc.), 0.24 ml Dye mix D (Biolog, Inc.), 6 mg ml^-1 BSA, 1.26 mg ml^-1 NaHCO₃, and 1.76 ml water. This mixture (0.1 ml) was added to the wells of PM1 phenotype microarray plates (Biolog), and the initial absorbance at 590 nm was read in an ELISA plate reader. Plates were incubated at 37°C in a microaerobic atmosphere and the absorbance read again following 24 and 48 h of incubation.

Following 7 days of incubation at 37°C in a microaerobic atmosphere, C. coli cultures in glucose-free TSB with and without supplementation with 100 mM glucose were homogenized carefully by pipetting. Biomass was harvested from 5 ml culture by centrifugation and the pellets incubated overnight at 50°C for evaporation of water content. The dry weight of biomass was determined as the mean of four replicates.

Campylobacter coli B13117 and CV1257 were cultivated at 37°C under microaerobic conditions in TSB with and without supplementation with 100 mM glucose. Three independent cultures for each strain and condition were grown; following 24 h of incubation, cells were harvested from 5 ml culture by centrifugation for 1 min at 8,000 × g at 0°C. The supernatants were filter sterilized (0.2 μm pore size) and stored at -20°C until analysis. A procedure based on the methyl chloroformate derivatisation protocol described by Smart et al. (2010) was used for the analysis of mixtures containing known metabolites with and without complexation. All samples were analyzed in a randomized order. The system was controlled by ChemStation (Agilent technologies). Raw data was converted to netCDF format using Chemstation (Agilent), before the data was imported into Matlab R2014b (Mathworks, Inc., Natick, MA, USA) and processed using PARAFAC2 (Bro et al., 1999; Kiers et al., 1999) to obtain relative concentrations for each peak. PARAFAC2 was applied using in-house algorithms.

The key ED pathway genes were originally identified in the genome sequence of the human blood isolate C. jejuni subsp. doylei 269.97 (Miller, 2008). Since C. jejuni subsp. doylei is overrepresented in human cases of Campylobacter bacteremia (Lastovica, 2006; Parker et al., 2007) and given the millimolar concentrations of glucose in blood, we initially wanted to test the hypothesis that the ED pathway might be specifically associated with the growth or survival of Campylobacter found in blood. Therefore, a range of Campylobacter isolates was screened for the ED pathway using PCR amplification of the key genes (Figure 1C). This screen included five clinical isolates of C. jejuni subsp. doylei (CCUG 18265, CCUG 24567, CCUG 26152, CCUG 26155, CCUG 36506), a C. coli bacteremia isolate (B13117), an in-house C. coli isolate of unknown origin (CV1257) and 10 C. jejuni subsp. jejuni bacteremia isolates (S38543, B10540, B345, B15426, B130, B17648, B19978, B3-11, B5682, B10053) (see Supplementary Table S1 for details). The screen revealed the presence of 6-phosphogluconate dehydratase (edd) and/or aldolase (eda) genes, encoding the key ED pathway enzymes, in three out of five clinical C. jejuni subsp. doylei isolates (CCUG24567, CCUG18265, CCUG26155), both C. coli isolates (B13117, CV1257), but not in any of the examined C. jejuni subsp. jejuni bacteremia isolates (Figure 1C and results not shown). These findings therefore do not support an association between blood culture isolates and the key genes of the glc locus. Analysis of WGS data for CCUG 24567, CCUG 18265, CCUG 26155, B13117, and CV1257 (Supplementary Table S2) confirmed that all isolates carried the complete glc locus.

The availability of assembled WGS data in the C. jejuni/coli PubMLST database facilitated large-scale in silico searches for additional ED-positive isolates. We found that 113 out of 6,184 isolates with genomic data contained one or more components of the glc locus. The glc locus did not assemble into a single contig in six isolates, all of which corresponded to draft genomes. Breaks occurred within the genes of interest, likely due to misassembly or lack of coverage; these isolates were excluded from further analyses. The complete glc locus was present on a single contig in the remaining 107 isolates (1.7% of the total) (Supplementary Table S2), which were collected in the UK (n = 76), Finland (n = 30), and the USA (n = 1).

Based on ribosomal multilocus sequence typing (rMLST) (Jolley et al., 2012), 70 and 37 isolates corresponded to C. jejuni and C. coli, respectively. At the population level, C. coli isolates segregate into three groups known as clades 1, 2, and 3 (Sheppard et al., 2008; Sheppard et al., 2010); the ED-positive C. coli isolates were all assigned to clade 1, except OXC7653, which belonged to clade 3. With respect to sources of isolation, C. jejuni isolates were predominantly from rats (51.4%) and wild birds (42.9%). In contrast, human disease samples accounted for 43.2% of C. coli isolates, while the remainder were from a diverse range of sources, including animals, food, and the environment (Table 1).

The seven genes comprising the glc locus were annotated as described in Experimental Procedures, using the autotagger functionality implemented in PubMLST (Jolley and Maiden, 2010). Unique alleles identified for each gene were assigned arbitrary numbers in order of discovery. Between 12 (eda) and 22 (pgi2) alleles per glc locus gene were present among C. jejuni subsp. doylei 269.97, the five ED-positive isolates sequenced for this study, and those from the C. jejuni/coli PubMLST database (Table 2). Gene lengths ranged from 624 bp (eda) to 1803 bp (edd), and allele lengths were variable for all loci except edd, eda, and pgl (Table 2). Unique combinations of glc locus alleles, taken in gene order from glcP to eda, were summarized as ED types, which were assigned arbitrary numbers in order of discovery. Overall, 46 ED types were identified, all of which were species-specific: 36 were present in C. jejuni and 10 in C. coli. C. jejuni isolates were evenly distributed across ED types, with the exception of 18 closely related rat isolates that had been isolated from four different farms between 2011 and 2012 and which all carried ED type 2 (ED allelic profile 3-4-4-5-4-5-5). Interestingly, these findings with C. jejuni contrasted with a more limited allelic diversity observed for C. coli isolates, amongst which ED type 1 (3-5-7-9-5-9-9) was predominant, accounting for 27/37 (73%) of PubMLST isolates and CV1257. Phylogenetic analysis of concatenated glc gene sequences indicated that ED types segregated into three groups, with C. jejuni ED types occurring in all three (Figure 2). There was little diversity among C. jejuni sequences from groups I and III, which primarily corresponded to isolates obtained from wild birds or rats, respectively. Sequences belonging to group II were more diverse and included the known C. jejuni subsp. doylei isolates. With the exception of clade 3 isolate OXC7653, C. coli sequences were only found in group III and were highly homogeneous (Figure 2).

Relationships among ED-positive isolates were examined using whole-genome multilocus sequence typing (wgMLST). Seven isolates were excluded from these analyses as 1–16.5% of loci belonging to the C. jejuni/coli core genome scheme were incomplete (i.e., at ends of contigs). With respect to C. jejuni, wild bird, rat, and known C. jejuni subsp. doylei isolates largely occupied distinct parts of the network that correlated with membership of the three ED nucleotide sequence groups (Figures 2 and 3). Members of these groups were genetically diverse and were separated by an average of 1,294 loci, with the exception of small clusters of closely related isolates, particularly among those obtained from farm-associated rats (Figure 3A). Isolates belonging to these clusters were typically separated by fewer than 100 loci and the majority carried ED type 2. Although C. coli isolates were also genetically diverse, they were separated by shorter distances, averaging 973 loci (Figure 3B). While the majority carried ED type 1, these isolates did not group by source or ED type.

The presence of the ED pathway genes within the subset of Campylobacter isolates studied here indicates that these isolates should be able to utilize glucose. To confirm this, a Biolog^TM phenotypic microarray was initially used to examine the potential metabolism of various carbohydrates by the ED-positive isolates C. coli B13117, C. coli CV1257 and C. jejuni subsp. doylei 269.97 in comparison to the widely used lab strain C. jejuni NCTC11168, which does not carry the ED pathway. With the Biolog^TM system, substrate uptake and metabolism is detected as stimulation of bacterial respiration and quantitated colorimetrically via tetrazolium dye reduction (Bochner et al., 2001). This revealed a correlation between the presence of the ED pathway and metabolism of glucose, as glucose was seen to stimulate respiration of the ED-positive C. jejuni subsp. doylei 269.97, C. coli B13117 and C. coli CV1257, while no signal was observed for C. jejuni NCTC11168. In contrast, L-fucose was seen to stimulate respiration of all four isolates regardless of the presence of the ED pathway (Supplementary Figure S1).

To investigate if glucose is able to support growth of C. coli isolates carrying the ED pathway, the growth of four ED-positive C. coli isolates representing three different sequence types (B13117, CV1257, Dg172, and Dg349) was examined in comparison to an ED-negative strain (OXC6725; PubMLST id 18282). Of these isolates, CV1257, Dg172, and Dg349 all had ED type 1 (the same as in the isolates used by Vorwerk et al., 2015), while B13117 carried ED type 45, which only differed by a single synonymous substitution in pgi2 (471: T- > C). The isolates were grown in minimal media (MCLMAN) without and with glucose supplementation as the main carbon source; a low concentration of L-aspartate (2 mM) was included as a nitrogen source as this was found to greatly aid growth. Following overnight incubation at 37°C under microaerobic conditions, the viable cell count of each of the ED-positive isolates had increased significantly with glucose (by ∼3 logs), while no such increase in cell density was seen for either the ED-negative strain or any of the cultures without glucose supplementation (Figure 4). These data show a clear correlation between possession of the ED pathway in these C. coli isolates and their growth on glucose.

In order to examine the physiological impact of glucose metabolism by C. coli, the growth of ED-positive isolates was further examined in the rich medium tryptic soy broth (TSB) without and with glucose supplementation. Here, a range of carbon sources (mainly amino-acids and peptides), are available in addition to the added glucose. At 37°C in a microaerobic atmosphere, the exponential growth rates of B13117 and CV1257 were seen to be unaffected by the availability of glucose (Supplementary Figure S2). After 12 h of incubation, however, the viability of the un-supplemented cultures declined significantly. In contrast, viable cell numbers were sustained or slightly increased in the glucose-supplemented cultures (Supplementary Figure S2), hence, suggesting glucose might support extended viability and stationary survival of these ED-positive isolates, at least up to 24 h.

Extending the growth experiments with prolonged incubation time over several days and including four ED-positive C. coli isolates (B13117, CV1257, Dg172, and Dg349) and an ED-negative control C. coli strain (OXC7218; PubMLST id 22709), revealed that glucose did indeed support significant extended survival of the ED-positive isolates (Figure 5). When cultivated with glucose at 37°C in a microaerobic atmosphere, the ED-positive isolates B13117, CV1257, Dg172, and Dg349 displayed at least 10 times higher viable cell numbers after 4 and 7 days as compared to cultures without glucose, while no such effect of glucose could be seen with the ED-negative strain OXC7218 (Figure 5). These data show a major impact of glucose on the late stationary phase survival of ED-positive C. coli isolates under otherwise nutrient rich conditions.

In the extended incubation experiments, it was observed that glucose stimulated biofilm formation by some ED-positive isolates of C. coli. Substantial floating biofilm (flocs or a pellicle), i.e., biofilm unattached to a surface, was observed in the glucose-supplemented late stationary cultures of B13117 and Dg349, and this aggregation was in fact so viscous that it was hardly dispersible by simple pipetting (Figure 6A). However, extensive biofilm formation was not observed in either the glucose-free cultures of B13117 and Dg349 or any of the late stationary cultures of the two other ED-positive isolates CV1257 and Dg172 in the presence of glucose. The ED-negative strain OXC7218 displayed a low level of aggregation regardless of the glucose availability (Figure 6A). Note that the biofilm produced by C. coli B13117 and Dg349 is floating, i.e., unattached to a solid surface, and therefore it was not possible to quantify using traditional dye staining assays for staining of attached biofilms. Consequently, the contribution of the biofilm to the total cell biomass was quantified using the dry weight of culture aliquots following 7 days of incubation with and without glucose (Figure 6B). As expected, this showed a significant increase in the biomass of all the ED-positive isolates upon glucose supplementation, while the biomass of the ED-negative strain was unaltered regardless of glucose availability. However, the biomass of B13117 and Dg349 grown with glucose was clearly much greater than that of CV1257 and Dg172 with glucose (Figure 6B, arrows), which is in agreement with the visually observed high biofilm production by these isolates (Figure 6A, arrows).

Campylobacter coli B13117 and CV1257 were subjected to an exometabolomics “footprint” analysis to compare the extracellular metabolic profiles of these ED-positive isolates while growing in TSB with and without glucose. Changes in the media composition after growth, reflecting the metabolic activity of the bacteria (i.e., production and/or consumption of metabolites and changes in growth substrates) were evaluated using a validated method based on gas chromatography–mass spectrometry (GC–MS) analysis of methyl chloroformate derivatised samples (Smart et al., 2010) which primarily identifies amino-acids and organic acids. In the principal component analysis (PCA) shown in Figure 7A, it is clear that the footprints of B13117 and CV1257 are highly similar when the isolates are cultivated in TSB without glucose, as these samples group closely together in the same region of the PCA plot (Figure 6A). However, upon glucose supplementation, the samples from the two isolates group in opposite quadrants of the plot (Figure 6A), indicating a substantial but very dissimilar response to glucose. A heat map, representing the relative concentrations of the compounds detected by GC–MS, is shown in Figure 7B. The largely similar extracellular metabolic profiles of B13117 and CV1257, when cultivated in TSB without glucose, are clear from the overall pattern seen in the heat map profile. After growth with glucose a clear difference between the two isolates in their responses in terms of metabolite concentrations was observed (Figure 7B). Using known reference standards, some of these extracellular compounds could be identified. From the results, it seems that glucose stimulates catabolism more strongly in CV1257, since the extracellular concentration of several metabolites both from and closely related to the citric-acid cycle (e.g., succinate, pyruvate, oxaloacetate, lactate) was found to increase when this strain was cultivated with glucose as compared to a glucose-free medium, while the amino-acids isoleucine, phenylalanine, and methionine decreased (Figure 7B). The metabolic profile of B13117 also displayed a clear shift upon cultivation with glucose, with certain amino-acids from the growth medium again decreasing. However, far fewer glucose-dependent extracellular metabolite alterations of B13117 were identified with the GC–MS analysis, and in particular there was little change in citric-acid cycle related organic acids (Figure 7B). Given the strong tendency of the B13117 strain, but not CV1257, to produce biofilm upon cultivation with glucose (Figure 6), the distinct metabolic profile compared to CV1257 is consistent with glucose carbon being preferentially directed into polysaccharide formation in the former isolate.

In addition to metabolic differences in the B13117 and CV1257 isolates, we also searched for possible differences in gene content that might be related to polysaccharide formation, in particular in the capsule biosynthesis loci. A total of 1,996 unique coding sequences were identified between B13117 and CV1257, 1,618 of which were present in both isolates. Among the remaining coding sequences, 204 and 174 were specific to B13117 and CV1257, respectively. Differences in the capsule biosynthesis loci accounted for 35 (9.3%) of the isolate-specific genes, with the region being approximately 5.6 kb longer in B13117 (Supplementary Figure S3). Although the flanking genes shared >90% sequence identity, the intervening sequences were distinct (Supplementary Figure S3). Some of the additional genes in this region in B13117 have predicted functions in polysaccharide biosynthesis (Supplementary Table S3). With respect to the remaining isolate-specific genes, 174 corresponded to hypothetical proteins.