Progressive genome-wide introgression in agricultural Campylobacter coli

Hybridization between distantly related organisms can facilitate rapid adaptation to novel environments, but is potentially constrained by epistatic fitness interactions among cell components. The zoonotic pathogens Campylobacter coli and C. jejuni differ from each other by around 15% at the nucleotide level, corresponding to an average of nearly 40 amino acids per protein-coding gene. Using whole genome sequencing, we show that a single C. coli lineage, which has successfully colonized an agricultural niche, has been progressively accumulating C. jejuni DNA. Members of this lineage belong to two groups, the ST-828 and ST-1150 clonal complexes. The ST-1150 complex is less frequently isolated and has undergone a substantially greater amount of introgression leading to replacement of up to 23% of the C. coli core genome as well as import of novel DNA. By contrast, the more commonly isolated ST-828 complex bacteria have 10–11% introgressed DNA, and C. jejuni and nonagricultural C. coli lineages each have <2%. Thus, the C. coli that colonize agriculture, and consequently cause most human disease, have hybrid origin, but this cross-species exchange has so far not had a substantial impact on the gene pools of either C. jejuni or nonagricultural C. coli. These findings also indicate remarkable interchangeability of basic cellular machinery after a prolonged period of independent evolution.


Introduction
Bacterial genomes show great flexibility in their genome size and composition and can acquire genes encoding entire metabolic pathways from other organisms (Lawrence 1999;Ochman et al. 2000). Nevertheless, there are many species that are characterized by a large and stable 'core genome'. Although DNA within the core genome can be replaced in recombination events, this is almost always with homologous DNA from another member of the same species (Fraser et al. 2007). An important evolutionary question is what underlies this stability (Doolittle & Zhaxybayeva 2009). Is the core genome in each species a coadapted unit, such that equivalent genes taken from other species will not function properly? Is each gene in the core genome adapted to the specific sets of environments that the species inhabits? Or is acquisition of DNA limited by mechanisms that prevent uptake of DNA from outside the species?
Campylobacter are Gram-negative microaerophilic epsilon proteobacteria that inhabit the intestinal tracts of birds (Waldenstrom et al. 2002;Sheppard et al. 2010a) and other animals (Rosef et al. 1983) and have some capacity to survive in the nonenteric environment (Sopwith et al. 2008). They have relatively small genomes of approximately 1.6 megabases (Parkhill et al. 2000), which limits the diversity of functions available to each organism. However, in common with many other bacteria, Campylobacter shows high levels of recombination (Suerbaum et al. 2001;Snipen et al. 2012), which might compensate for small genome size by providing each organism with the ability to import genes that confer adaptations to specific environments.
Campylobacter jejuni and Campylobacter coli are among the main causes of human gastroenteritis worldwide, largely because of infection of farm animals and transmission through the food chain to retail products (Sheppard et al. 2009). Both species are associated with several agricultural hosts (Sheppard et al. 2010a). C. jejuni are usually more abundant in cattle and chickens, and C. coli dominates in pigs (Thakur et al. 2006). C. jejuni has also been isolated from many wild bird species (Waldenstrom et al. 2007;Colles et al. 2008), but little is known about the distribution of C. coli among wild hosts.
Campylobacter-like organisms were described by Theodor Escherich in 1886 (Escherich 1886), and the genus was formally named in 1963 (Sebald & Veron 1963). C. coli, identified as vibrio isolated from pig faeces (Doyle 1948), was designated as a species distinct from C. jejuni in 1973 (Veron & Chatelain 1973), and this species classification has been uncontroversial. However, by analysing the sequence from 7 housekeeping loci from a large number of strains, we found evidence for the acquisition of substantial amounts of C. jejuni DNA by one of the three C. coli clades (Sheppard et al. 2008(Sheppard et al. , 2011. Both the correctness and implications of this finding have been debated (Cohan & Koeppel 2008;Doolittle 2008;Caro-Quintero et al. 2009;Lefebure et al. 2010), and many questions remain about the patterns of introgression in the core and pan genome, and how it has influenced the evolution of these important pathogens.
Here, in addition to four previously published genomes, we chose 26 Campylobacter isolates from clinical, agricultural and nonagricultural sources (Tables S1 and S2, Supporting information), to encompass the known diversity based on analysis of seven housekeeping loci (Sheppard et al. 2009). The genomes of the isolates were sequenced using the Illumina GA and Roche 454 platforms and assembled de novo. We performed model-based analysis of evolution of the core and pan genomes to reconstruct a history of betweenspecies genetic exchange within the genus.

Isolates and sequencing
Isolates were chosen from multilocus sequence-typed collections to represent known diversity among C. jejuni and the three major C. coli clades, including nonagricultural strains for each lineage. These were cultured and genomic DNA was sequenced using Roche GS-FLX or Illumina Genome Analysers (see Appendix S1). Details of all the isolates used, including four complete C. jejuni genomes (Parkhill et al. 2001;Fouts et al. 2005;Pearson et al. 2007) from the NCBI database (accession numbers: NC_009839; NC_008787; NC_003912; NC_002163), are included in Tables S1 and S2 (Supporting information).
Genetic relationships between C. coli and C. jejuni A schematic diagram of the genomics analysis pipeline is given in Fig. S1 (Supporting information). The Bacterial Isolate Genome Sequence Database (BIGSdb) (Jolley & Maiden 2010) was used to store contiguous sequences and whole genome data from Genbank. Locus names and reference sequences were defined based upon the finished genome of isolate NCTC11168 (Cabello et al. 1997;Parkhill et al. 2001;Gundogdu et al. 2007). The presence of a preliminary set of orthologs was defined by identifying reciprocal best hits to 11 168 loci, with at least 70% nucleotide identity and 50% difference in alignment length using the BLAST algorithm. The analysis of orthology was made for every genome, and the core genome, consisting of genes ubiquitous among isolates of the genus, was defined.
Gene orthologs were aligned on a gene-by-gene basis using MUSCLE (Edgar 2004) and then concatenated into contiguous sequence for each isolate genome including gaps for missing nucleotides (or entire genes). A phylogeny of whole genome alignments (1.53 Mbp) was reconstructed using MEGA (Kumar et al. 2008) version 3.1 with the Kimura 2-parameter model and neighbourjoining clustering.
The ancestry of individual nucleotides was estimated using the model-based clustering algorithm implemented in the software STRUCTURE (Falush et al. 2003). A file describing all the 239 543 nucleotide substitutions and the position of the polymorphic sites was constructed from the gene-by-gene alignment file, for loci present in all the genomes. STRUCTURE was run for 100 000 iterations following a 20 000 iteration burn-in. Genes were ordered by the amount of introgression in the 828 complex, identified with STRUCTURE. For the 13 most introgressed genes, also present in the analysis of Lefebure et al. (2010), individual neighbour-joining trees were constructed with and without our strains (Fig. S8, Supporting information).
Pairwise alignments of genomes were generated using progressive MAUVE version 2.3.1 (Darling et al. 2004(Darling et al. , 2010 with the default parameter settings and analysed using a Bayesian change point model as described previously . This model assumes that the level of nucleotide divergence between the two genomes follows a stepwise constant function and uses a reversible-jump MCMC to reconstruct this function. Histograms were then built to show the distribution of the level of divergence along the genomes.

Event based analysis of C. coli clade 1 evolution
Multiple sequence alignment of the contigs for each genome was performed using progressive MAUVE version 2.3.1 (Darling et al. 2004(Darling et al. , 2010 with the default parameter settings. The progressive MAUVE backbone output file was used to assign regions of each genome as either core ('backbone') segments, conserved among all of the genomes, or accessory ('variable') segments absent from at least one alignment. Briefly, the multiple genome alignment was automatically analysed to identify conserved segments using a homology hidden Markov model (Treangen et al. 2009). Regions where the posterior probability of sequence homology was >90% using a model trained on 80% identity and tuned to the sequence composition of Campylobacter were considered to be homologous. Nonhomologous regions create alignment gaps, and those alignment gaps were used to delineate a 'backbone' of conserved segments among each pair of genomes by simply calling any region with >20 nucleotides inserted or deleted in one genome as nonbackbone (indels > 20 nt). Pairwise backbone predictions were merged into multigenome backbone predictions using the previously described methods (Treangen et al. 2009). Using this technique, the amount of core and accessory genome was determined for all the isolates and for C. jejuni and C. coli individually and the three C. coli clades separately and for subsets of isolates (Fig. S9, Supporting information).
Additionally, a gene-by-gene alignment was extracted from BIGSdb (Jolley & Maiden 2010) for genes present in all the C. coli clade 1 genomes. A genealogy for these alignments was estimated using CLONALFRAME, a modelbased approach to determining microevolution in bacteria . This programme differentiates mutation and recombination events on each branch of the tree based on the density of polymorphisms. Clusters of polymorphisms are likely to have arisen from recombination and scattered ones from mutation. Run on the C. coli clade 1 alignment, CLONAL-FRAME, estimated that recombination introduced polymorphism at an average of 8% of affected sites. This value is higher than the genetic diversity within C. coli clade 1 and thus corresponds to imports from the other clades and species, as well as back-recombination events replacing previously introgressed DNA. The programme was run with 50 000 burn-in iterations followed by 50 000 sampling iterations. The consensus tree represents combined data from three independent runs with 75% consensus required for inference of relatedness. For each branch on the CLONALFRAME genealogy, a list of homologous recombination events was extracted. Recombination events were defined as sequences of length >50 bp with a probability of recombination 75% over the length reaching 95% in at least one site.
To investigate the acquisition and loss of nonhomologous DNA, we used the model-based Bayesian method implemented in GENOPLAST (Didelot et al. 2009). This model allows the rates of gain and loss of genetic elements to be investigated over time in individual lineages. A multiple alignment was produced for C. coli clade 1 genomes (excluding isolate 16) using progressive MAUVE (Darling et al. 2004(Darling et al. , 2010, and the conserved orthologous segments and repeat elements defined the core genome. Large gaps ( 500 bp) in the alignment, where one or more genomes contain a sequence absent in the other genomes, identify the position of imported DNA in the accessory genome. A binary matrix of presence/absence of genetic features of length 50 bp was constructed using the bbFilter script distributed with MAUVE. GENOPLAST was run with default parameters using this matrix and the genealogy inferred by CLONALFRAME as input.
The origin of homologous and nonhomologous recombination events in C. coli clade 1 genomes was determined using the BLAST algorithm (Altschul et al. 1990). A list of events identified with CLONALFRAME and GENOPLAST was extracted, and sequences were compared to a library database of all the C. jejuni and C. coli clade 2 and 3 genomes. The origin of the events was assigned based on the similarity (S) to a library sequence. Specifically, the E-value that measures the reliability of the S score was calculated for all BLAST matches ( 70% identity, 50% alignment), and the event was inferred to have originated in the species/clade containing homologous sequence with the lowest E-value.
In addition to determining the clonal genealogy and the origin of recombination, CLONALFRAME analysis information was used to investigate the impact of homologous recombination with C. jejuni on sequence divergence in C. coli clade 1 using sequence variation at 51 ribosomal protein (rps) subunit loci (see Appendix S1).

Functional analysis of introgression
The proportion of nucleotide substitutions that changed the amino acid sequence in homologous sequence was investigated for recombinant C. jejuni genes found in C. coli clade 1. Genome comparison was made between an example C. jejuni (isolate 4) and an unintrogressed C. coli (isolate 23). These strains shared 1081 genes, defined as homologous sequence with >70% identity over >50% of the gene alignment; 584 of these genes were involved in recombination in at least one C. coli clade 1 isolate and 497 were not. The number of nonsynonymous differences (N), number of synonymous differences (S) and ratio of nonsynonymous to synonymous mutations (dn/ds) were determined from gene-by-gene alignments of recombining and nonrecombining genes using MEGA software version 3.1 (Kumar et al. 2008). If recombinant sequence is removed by the action of selection against divergent amino acid sequence, then there will be a greater than expected number of synonymous substitutions in recombined genes and a lower dn/ds ratio.
To investigate the relationship between the rate of genetic import and the functional category of genes, the number of genes involved in homologous recombination was determined for each cluster of orthologous groups (COG) category. For each COG category, the number and total length (bp) of imports were determined. This allowed the determination of the rate of imports per nucleotide and the proportion of genes from each COG involved in recombination. Some genes are present in two or more COG categories and were counted once for each COG. A second analysis of the function of recombined genes was carried out by determining the genes that were found only in C. jejuni and C. coli clade 1 and were absent in unintrogressed C. coli.
Genome comparison was carried out to identify differences in the gene content of the C. jejuni and C. coli genomes. By organizing C. jejuni genes absent from unintrogressed C. coli into functional categories, groups of genes of related function were identified. A second comparative analysis identified genes that were found only in C. jejuni and introgressed C. coli genomes.

Results
Our initial analysis focussed on the core genome. The NCTC11168 isolate has a 1.6 Mb genome (Parkhill et al. 2000), and 0.96 Mb was aligned in all our isolates using MAUVE. Based on locus designations for NCTC11168 (1623 genes), there were 542 genes with orthologues in all the isolates (genes with at least 70% nucleotide identity and a minimum of 50% alignment length). Under the same criteria, there were 819 genes (50%) common to all C. jejuni isolates and 928, 1084 and 1078 common to C. coli clades 1, 2 and 3, respectively.
We first constructed a neighbour-joining tree based on average genetic distances amongst isolates (Fig. 1A). On the tree, C. jejuni, C. coli clade 2 and C. coli clade 3 isolates each formed discrete clusters. However, isolates previously designated as C. coli clade 1 were found in three places on the tree. The ST-828 and ST-1150 clonal complexes, which account for the great majority of strains found in agriculture and human disease (Sheppard et al. 2010b), formed discrete clusters separate from the two environmental C. coli clade 1 isolates.

Evidence for introgression
Three lines of evidence show that the large genetic distances among C. coli clade 1 isolates, illustrated by the neighbour-joining tree (Fig. 1A), are a consequence of the import of DNA from C. jejuni rather than accumulation of mutations during a prolonged period of separate evolution. The first used the linkage model of STRUCTURE (Falush et al. 2003) that reconstructs ancestral populations from DNA polymorphism data. When run assuming two ancestral populations, the inferred ancestral sources corresponded to C. jejuni and C. coli. The human and agricultural C. jejuni isolates had between 0.4% and 1.7% inferred C. coli ancestry, consistent with a low level of import. Excluding isolates from the ST-828 and ST-1150 clonal complexes, the C. coli isolates showed a comparable amount of inferred C. jejuni ancestry that ranged from 0.2% to 1.2%. The ST-828 and ST-1150 clonal complexes showed substantially more evidence for DNA import from C. jejuni ranging from 9.7% to 11.2% and 20.4% to 22.5%, respectively, spread throughout the genome (Fig. 1B).
The second line of evidence for introgression into C. coli clade 1 is provided by pairwise comparison of nucleotide differences between genomes. In the absence of gene flow, isolates from the two species should have a unimodal distribution of divergence levels reflecting accumulation of mutations throughout the genome. This pattern was observed for comparisons between C. jejuni and unintrogressed C. coli isolates, with modes between 10% and 12% (Fig. 1C). Comparisons with the ST-828 and ST-1150 clonal complex isolates showed a bimodal distribution with similar modes at 10-12% but also earlier modes at <2%. The low nucleotide divergence was consistent with recent recombination with C. jejuni. Combined with evidence for shared polymorphism found using STRUCTURE, these patterns of divergence are consistent with recent gene flow from C. jejuni. The imported DNA has greater nucleotide identity to the agricultural C. jejuni isolates than the environmental C. jejuni isolates (Fig. S2, Supporting information).
The third line of evidence is provided by constructing maximum likelihood trees separately for loci according to whether they have any C. jejuni ancestry according to CLONALFRAME. This analysis was performed for the 51 ribosomal protein subunit (rps) loci in our genomes (Jolley et al. 2012). When analysis was limited to genes where there was no C. jejuni-like sequence, the clade 1 strains clustered together (Fig. 2), consistent with their shared common ancestry as was found previously for seven MLST loci (Sheppard et al. 2008). Furthermore, as in this previous analyses, the branching pattern positioned clade 2 as a sister taxa to clade 1. In contrast, on the tree for rps loci that showed evidence of interspecies recombination, C. coli clade 1 isolates were scattered around the branches joining the two species in the tree (Fig. 2). This analysis suggests that C. coli clade 1 is a real clade and that the presence of clade 1 isolates on three different parts of the whole genome neighbour joining (Fig. 1A) is an artefact of the substantial effect of interspecies recombination on genetic distances.

Core and pan genome evolution
Having established that up to 23% of the C. coli clade 1 genome is of C. jejuni origin, we investigated evolution within the clade and the sequence of events responsible for introgression. An alignment of C. coli clade 1 genomes was constructed (excluding one strain with low genome coverage) and a tree of clonal relationships was estimated using CLONALFRAME . The analysis showed that the ST-828 and ST-1150  (Parkhill et al. 2000), and each nucleotide is coloured according to genetic ancestry to C. jejuni (■) or C. coli (□). (C) Histograms of nucleotide divergence between C. jejuni (isolate 4) and genomes from C. coli clade 1 (red), 2 (yellow), 3 (green) and C. jejuni (blue). Pairwise comparisons between C. jejuni and un-recombined clade 1 (isolate 23 in the example), and clade 2 and 3 genomes (upper panel) show a unimodal distribution with modes between 10 and 12%. Comparison of C. jejuni with C. coli clade 1 isolates from the ST-828 and ST1150 complexes (middle panel) has a bimodal distribution with similar modes at 10-12% but with an earlier mode at <2%. The nucleotide divergence of the earlier mode is similar to comparison between two C. jejuni genomes (bottom panel).
clonal complexes are more closely related to one another than to an environmental isolate (isolate 23) (Fig. 3A). Most of the C. jejuni DNA found in the ST-828 complex was also found in the ST-1150 complex ( Fig. 1B and Fig. S3, Supporting information), implying that this genetic material was imported by the common ancestor(s) of both complexes. Subsequent to the divergence of the two complexes, the ST-1150 complex has acquired substantially more C. jejuni DNA than the ST-828 complex although import is ongoing in both complexes.
In addition to recombination of homologous sequence, our approach allows us to investigate the evolution of the pan genome, which occurs via acquisition and loss of genes. A multiple genome alignment was constructed for C. coli clade 1 isolates using MAUVE (Darling et al. 2004), and GENOPLAST (Didelot et al. 2009) was applied to the alignment blocks to identify those that were gained or lost on particular branches of the tree (Fig. 3C). The origin of imported DNA was inferred by BLAST comparison of the sequences to reference genomes from C. jejuni and C. coli clade 2 and 3. An equivalent analysis of origin was performed for homologous imports (Fig. 3C). In total, 438 nonhomologous and 2237 homologous recombination events were inferred, although for methodological reasons, both homologous and nonhomologous events were only identified reliably towards the tips of the tree. In both cases, approximately 50% of events were of C. jejuni origin although for homologous recombination, the proportion is higher for many of the short branches. This demonstrates that a high magnitude of introgression has occurred in both the core and pan genomes.

Effects of introgression on the core genome
The availability of unintrogressed C. coli genomes provides insight into where introgression has occurred and its genomic effects. Data in a variety of bacterial species have suggested that recombination is homology dependent (Cohan 2002;Fraser et al. 2007). We found that recombination was rarer in areas of the genome where there was high divergence between C. jejuni and the unintrogressed C. coli (Fig. 4). However, the observed degree of homology dependence was several orders of magnitude weaker than in other species where mismatch repair mechanisms prevent the integration of most sequences that contain even small numbers of nucleotide differences (Cohan 2002;Fraser et al. 2007). Based on a conservative threshold for identifying interspecies recombination events, CLONALFRAME analysis implied that 9 (95% credibility regions 5-16) times as  Fig. 2 The effect of interspecific recombination on tree clade structure. Maximum likelihood trees, based on the Tamura-Nei model, of 30 Campylobacter jejuni and C. coli genomes are based on concatenated sequences of ribosomal protein (rps) subunit loci genes that show (A) 35 genes with no evidence of homologous recombination and (B) 16 with evidence of recombination in at least one isolate using CLONALFRAME. Isolates belonging to C. jejuni are shown in blue, and those belonging to C. coli clade 1 are indicated in red, clade 2 in yellow, and clade 3 in green. The trees are drawn to scale, with branch lengths measured in the number of substitutions per site. The scale bar represents a genetic distance of 0.01. Trefoil clade structure is resolved in nonrecombining genes.
many substitutions were introduced on average by interspecies recombination as by mutation or intraspecies recombination. Within the most divergent regions of the genome (approximately 20% nucleotide divergence), the rate of interspecies exchange is approximately half the genome-wide average, but this still implies nucleotides are at least four times more likely to be changed by cross-species recombination than by new mutation or within-species recombination. This level of recombination would lead to progressive species convergence if maintained throughout the genome over time (Sheppard et al. 2008).
Comparing a unintrogressed C. coli clade 1 isolate (isolate 23) with a C. jejuni genome (isolate 27) shows that there are on average 38 nonsynonymous substitutions per gene. This is approximately an order of magnitude more than in pairwise comparisons within unintrogressed Campylobacter populations, which range from 1.4 for C. coli clade 2 to 5.5 for C. jejuni, comparable to 2.9 nonsynonymous substitutions per gene between human and chimpanzee, for example (Consortium(TCSaA) (2005). We found that introgression has taken place at equivalent rates in regions of both high and low nonsynonymous differentiations between the species (Fig. S4, Supporting information) and that there are on average 9.8 protein coding differences per gene between an unintrogressed clade 1 isolate (strain 23) and a member of the ST-1150 complex (strain 25). There was also no evidence for large differences in the rate of introgression between broadly defined functional categories (Fig. S5, Supporting information). Thus, introgression has greatly increased overall genetic diversity across the genome in C. coli clade 1 and introduced thousands of changes that have potential functional significance. (B) Recombination and mutation events on each branch of the genealogy, for an example, 1046 bp gene (metC2). Crosses indicate substitutions by either mutation or recombination. Recombination events inferred with high posterior probability are coloured according to whether the imported DNA is more similar to genomes from Campylobacter jejuni (blue) or C. coli clades 2 and 3 (yellow). (C) Total number of inferred homologous and nonhomologous imports on each branch from each species across the genome. Isolate 16 was excluded from this analysis because of poor genome coverage.

Fig. 4 Homology dependence of recombination between
Campylobacter jejuni and C. coli. The distribution of divergence between C. jejuni (isolate 4) and unintrogressed C. coli clade 1 (isolate 23) for the recombinant genes (white) and the nonrecombinant genes (black). Recombination is rarer in areas of the genome where there is high divergence between species but the effect is slight with a large overlap between the two distributions. Recombination occurs between genes at all levels of divergence.
C. jejuni genes absent from unintrogressed C. coli into functional categories, large groups of genes of related function were identified (Table S3, Supporting information). Particularly notable are: (i) solute transporters of different families, which may reflect differential nutrient utilization in the two species; (ii) specific cytochromes c and their associated biogenesis proteins, which may be related to the use of specialized respiratory substrates; and (iii) TonB-dependent outer membrane proteins potentially involved in iron uptake. The list also includes genes in diverse functional categories including those involved in core cellular function. Detailed characterization of the functional consequences of these differences requires further investigation. Thirty-one genes were identified only in C. jejuni and introgressed C. coli based on BLAST similarity (Table 1). Several genes were from a region (Cj0480c-Cj0490) that has recently been shown to be involved in the transport and metabolism of L-fucose (Muraoka & Zhang 2011). Most sugars cannot be used as growth substrates for Campylobacter, due to lack of 6-phosphofructokinase (PFK) (Velayudhan & Kelly 2002), so these genes presumably allow conversion of fucose to triose phosphate or another intermediate that bypasses PFK for entry into central metabolic pathways. Fucose is a major component of host glycoproteins, particularly intestinal mucin, and obtaining these genes from C. jejuni could allow C. coli to utilize fucose and provide an advantage in the gut (Stahl et al. 2011). The list of genes present only in C. jejuni and introgressed C. coli also includes key genes associated with flagella. Phylogenetic trees of the fucose-associated genes show some isolates have sequences with high nucleotide identity to those found in C. jejuni, but other sequences are entirely distinct, suggesting that the genes were present in the C. coli pan genome prior to introgression and have been lost by unintrogressed strains (Fig. S6, Supporting information). Intriguingly, several distinct genotypes exist among the C. jejuni-like sequences consistent either with rapid diversification or with introgression on multiple occasions, suggestive of recent selection at that locus (Falush 2009).

Discussion
Extensive recent recombination between species has a number of consequences for the patterns of DNA sequence diversity observed. First, there should be regions of the genome where individuals from the two species have highly similar sequence, reflecting recent common ancestry of DNA that has been imported from one species to the other. Second, patterns of inheritance for some stretches of the genome will be inconsistent with the consensus species tree. Third, recombination may elevate diversity within introgressed populations by introducing substitutions that were fixed during species divergence.
Here, we have sought to systematically investigate genome-wide patterns of relatedness between C. jejuni and C. coli by searching for each of these types of signal within homologous parts of the genome. First, in pairwise comparison of sequence diversity between C. jejuni and different C. coli isolates, C. coli that dominate in agriculture (ST-828 and ST-1150 complexes) had large fractions of the genome with low divergence from C. jejuni-typical of that found between two C. jejuni isolates. Importantly, isolates from nonagricultural C. coli lineages did not have comparable low divergence regions. Second, we used the linkage model of STRUC-TURE to identify sections of the genome with ancestry in both species. Consistent with the pairwise sequence analysis, there was very little evidence for introgression amongst C. jejuni and nonagricultural C. coli isolates but a high degree of introgression in agricultural C. coli. Third, we observed the diversifying effect of recombination directly by reconstructing clonal relationships and the specific imports that took place during the evolution of the agricultural C. coli clonal complexes, using CLONALFRAME. We found large numbers of imports that introduced changes at approximately 12% of sites, consistent with an origin in C. jejuni. This finding also implies that introgression is ongoing in both the ST-828 and ST-1150 clonal complexes.

A scenario of Campylobacter evolution
We have reconstructed an evolutionary scenario (Fig. 5) within which to interpret patterns of diversity in both core and pan genomes. C. coli split from C. jejuni and subsequently diversified into three clades. Bacteria from both species show evidence of recombination within species and clade (for C. coli) from a version of the four gamete test applied to the two most common MLST alleles in each clade population (Sheppard et al. 2010c). This has contributed to generating the 'star-like' phylogenies amongst the C. coli clade 2 and 3 isolates in our sample (Fig. 1). However, genetic exchange between bacteria from different species and lineages has been rare enough to facilitate their progressive divergence, which has reached approximately 12% between C. jejuni and C. coli and around 4% between the three C. coli lineages.
More recently, within C. coli clade 1, a lineage arose that started to import substantial quantities of C. jejuni DNA. This lineage has given rise to two clonal complexes, the ST-828 and ST-1150 complexes. These two clonal complexes currently make up the great majority of typed agricultural isolates but a small proportion of nonagricultural ones, as shown by the epidemiologi- Table 1 Campylobacter jejuni genes with homologous sequence (70% BLAST similarity, >50% of the gene) present among introgressed C. coli clade 1 genomes and hypotheses about their potential function Gene Product Description-Hypothesis Transport and metabolism of L-fucose Cj0480c Transcriptional regulator This is divergently transcribed from the other genes in this unit and is likely to be regulating the rightward reading genes. Cj0480 is an IclR family regulator (Gundogdu et al. 2007) and could be acting as a repressor or activator inducing expression of catabolic and transport genes in response to L-fucose Cj0481 (annotated as dapA) Putative dihydropicolinate (DHP) synthase DHP, also present in Bacillus spores, is an intermediate of a variant of the lysine biosynthesis pathway from aspartate that catalyses the condensation of aspartate semi-aldehyde with pyruvate to form DHP. Cj0480 may catalyse a related lyase reaction, as there is another gene Cj0806 that could be the 'real' dapA; it may be involved as a lyase in a step of fucose catabolism? Cj0482/0483 (uxaA) Putative altronate or D-galactarate (sugar) hydrolase Could be a pseudogene because the N-terminus encoded in Cj0482 and the C-terminus in Cj0483 is separated by a stop codon. There are examples where such genes are expressed. Possible fucose hydrolase? Cj0484 Major facilitator superfamily transport protein Probably, a substrate-proton symporter to import a substrate driven by the pmf. It has some similarity to phthalate (aromatic) family transporters (Gundogdu et al. 2007). However, it is not possible to say what the substrate is likely to be from sequence data Cj0485 Dehydrogenase/oxidoreductase, FabG family. This is possibly an alcohol dehydrogenase

Cj0486
Probable L-fucose transporter This is a sugar transporter of the major facilitator superfamily, with significant similarity to the L-fucose-proton symporter of E. coli and other bacteria (Gundogdu et al. 2007). This is essential for L-fucose utilization in some C. jejuni strains (Muraoka & Zhang 2011) Cj0489/Cj0490 Putative aldehyde dehydrogenase Potentially involved in a step of fucose catabolism? Zinc uptake system Cj0263 Zinc transporter ZupT There may be a connection between the zinc uptake system genes in supplying zinc for the activity of the protease. A number of proteins contain the Cj1589 domain, so it is difficult to predict the function but there may be a zinc connection with Cj0263.

Cj0620
Zinc-dependent protease Cj1589 Zinc-dependent hydrolase, possibly a beta-lactamase or glyoxalase II Flagellin-associated genes Cj1339 (FlaA) Flagellin protein Flagellin-associated proteins that could be involved in niche colonization. The presence of AcpP2 and AcpS could suggest O-linked glycosylation of flagellin proteins being important. cally defined sample shown in Table 2. The rate of acquisition of DNA has been substantially higher for the ST-1150 complex than for the ST-828 complex. Because of the great preponderance of agricultural strains in almost all sample collections, several of our C. coli clade 2 and 3 isolates that we obtained for genome sequencing come from agriculture (Table S1, Supporting information) but these isolates have not undergone substantial introgression.
Changes in patterns of gene flow are substantially harder to demonstrate in taxa where divergence is less complete or where there are no unintrogressed ancestral clades preserved. Among bacteria in particular, divergence may be uneven across the genome because of 'fragmented speciation' (Retchless & Lawrence 2010). In this model, gene flow ceases first in some parts of the genome, for example, in regions responsible for adaptive divergence. This acts as a barrier to subsequent recombination, and progressive diversification occurs at other loci around the genome. This model is not directly applicable here because isolates found in all three C. coli clades had similar high divergence with C. jejuni across the genome (Fig. 1C), which implies that after speciation, there was an extended period of divergence with low levels of gene flow. Nevertheless, it is of interest to investigate whether the high rate of genetic differentiation between C. jejuni and C. coli has acted as a barrier to recent recombination within the two agricultural C. coli lineages. Recombination was twice as rare in regions with 20% nucleotide divergence than in regions with 10% (Fig. 4). However, even among divergent regions, the rate of recombination was sufficient to promote progressive species convergence at current levels of DNA exchange.

Comparison with other studies of introgression in Campylobacter
Our results contrast with those of a recent study (Lefebure et al. 2010) that analysed a larger number of genomes but failed to find evidence for substantial introgression (except in a single isolate). The disagreement reflects differences in sampling and methodology rather than biology. All but one of the C. coli isolates analysed by Fig. 5 A scenario for the evolution of Campylobacter jejuni and C. coli. These species diverged followed by the split of C. coli clades 1, 2 and 3. Recombination from C. jejuni to C. coli clade 1 began at some point before R1, and subsequent clonal expansion of introgressed lineages (828 and 1150 clonal complexes) at R1 and R2 led to the dominance of hybrid lineages in agriculture, human disease and currently available isolate collections. Clade 2 (C2) and 3 (C3) and clade 1 (C1*) populations from wild bird and environmental reservoirs (e.g. represented by isolates 16 and 23) remained unintrogressed. The cross-sectional area and diameter of the lineage 'trunks' are based on the abundance of isolates in the PubMLST database and the length of trunks is arbitrarily defined.  (Lefebure et al. 2010). The single exception was a member of the ST-1150 complex, for which introgression was inferred at 9.6% of genes. We estimated 23% introgression for the same isolate and at least 8% introgression for the ST-828 complex isolates. Moreover, the strains from Lefebure et al. (2010) are intermingled with the strains analysed here in a NJ tree constructed using whole genome sequences (Fig. S7, Supporting information).
The substantial underestimation of introgression in the previous study (Lefebure et al. 2010) reflects the absence of nonagricultural strains from their sample and the use of a single methodology that is sensitive to sampling. Specifically, Lefebure et al. (2010) looked for patterns of ancestry inconsistent with the species tree by using gene-by-gene phylogenies to identify loci where a minority fraction of C. coli isolates clustered closer to C. jejuni than to other C. coli. This approach systematically misses introgression shared by a majority of isolates within the ST-828 complex. Fig. S8 (Supporting information) shows neighbour-joining trees for the 13 genes that showed the highest levels of introgression in our analysis (>70% in each case), constructed with and without the isolates from this study. For most of the genes, the C. coli isolates from the study of Lefebure et al. (2010) contain sequences similar to those found in C. jejuni, while in every case, the nonagricultural C. coli isolates in our sample harbour entirely distinct sequences. Even in the cases where a handful of the Lefebure et al. (2010) ST-828 complex isolates do have the distinct-and presumably ancestral-C. coli sequence, the method they employed did not detect introgression because these isolates make up only a minority of their isolates. As a result, they inferred no introgression for these 15 genes, which is clearly an incorrect conclusion based on visual inspection of the neighbour-joining trees constructed using the combined sample (Fig. S8, Supporting information).
A further limitation of the method employed by Lefebure et al. (2010) is that because phylogenies were constructed only for entire genes, they were likely to miss imports of short gene fragments. The linkage model of STRUCTURE uses a hidden Markov model, which can detect any tract long enough to introduce polymorphisms at several sites that are characteristic of the other species, and we observed many such imports that were much shorter than entire genes (Fig. 3).
Ecological and evolutionary implications of introgression 2003). There are many features that might make livestock such a habitat for bacteria that have evolved in wild hosts. Furthermore, hopeful bacterial monsters can repair some of the most harmful disruptions to interactions of adaptive genes by subsequent homologous recombination of their interaction partners. Sequencing of larger numbers of isolates will allow more detailed characterization of this ongoing adaptive process and further develop our understanding of bacterial gene networks.

Supporting information
Additional supporting information may be found in the online version of this article.            Appendix S1 Supplementary Methods.