Genome-wide insights into population structure and host specificity of Campylobacter jejuni

The zoonotic pathogen Campylobacter jejuni is among the leading causes of foodborne diseases worldwide. While C. jejuni colonises many wild animals and livestock, persistence mechanisms enabling the bacterium to adapt to host species' guts are not fully understood. In order to identify putative determinants influencing host preferences of distinct lineages, bootstrapping based on stratified random sampling combined with a k-mer-based genome-wide association was conducted on 490 genomes from diverse origins in Germany and Canada. We show a strong association of both the core and the accessory genome characteristics with distinct host animal species, indicating multiple adaptive trajectories defining the evolution of C. jejuni lifestyle preferences in different ecosystems. Here, we demonstrate that adaptation towards a specific host niche ecology is most likely a long evolutionary and multifactorial process, expressed by gene absence or presence and allele variations of core genes. Several host-specific allelic variants from different phylogenetic backgrounds, including dnaE, rpoB, ftsX or pycB play important roles for genome maintenance and metabolic pathways. Thus, variants of genes important for C. jejuni to cope with specific ecological niches or hosts may be useful markers for both surveillance and future pathogen intervention strategies.

For the original sample sources of the C. jejuni genomes investigated here, the relative proportion and absolute distribution for each of the BAPS clusters are visualised in Fig. 2a and supplementary Figure S1a. We identified a close phylogenetic relationship between genomes of BAPS cluster 5 representing the origin chicken with those of BAPS cluster 15 representing waterborne environmental C. jejuni (Fig. 1).
The genomes of BAPS cluster 15 and those of BAPS clusters with genomes from less than 15 isolates were not analysed with respect to their lifestyle preference and were therefore used as a control group in our study.
The lifestyle preference of each major BAPS cluster was determined and subjected to an internal assessment: As shown in Table S2, our assignments are generally concordant with lifestyle preferences reported for frequently occurring lineages such as CC-353, CC-354, CC-443, CC-464 and CC-52 (chicken), CC-42 and CC-61 (cattle) and CC-403 (pig). We also identified the probable lifestyle classification for the CC-22 lineage (cattle) and for isolates belonging to ST-2274 (chicken) ( Table S1 and Table S2). Of note, the C. jejuni genomes associated with CC-21, CC-45 and CC-48 fulfilled the criteria for host-generalist lineages (Table S2).
Our core genome-based phylogenetic analysis further revealed that cattle-related BAPS cluster 4 lineages (including CC-42 and CC-22) were more closely related to host-generalist lineages of BAPS cluster 6 with CC-21 than to other cattle-related lineages, for instance those of BAPS cluster 10 ( Fig. 1). This also holds true for the chicken-related phylogenetic background ( Fig. 1): While chicken-related BAPS cluster 1 was found being more closely related to BAPS cluster 6 of host-generalist lineage, BAPS cluster 5 showed less phylogenetic distance to BAPS cluster 8 (host-generalist lineage). These findings clearly reject the hypothesis of a common evolutionary background for host-specific lineages with respect to the host species represented here.
Minimum spanning trees based on MLST utilising BAPS cluster classification and lifestyle preferences are shown in the supplementary material ( Figure S1). Finally, the accessory genome profiles of all genomes were visualised by t-SNE plots in Fig. 2b-d including sample origin, BAPS cluster and lifestyle preference. As expected, the overall population structure derived from the core genome is mirrored in the accessory genome content. Each BAPS cluster carries its unique set of accessory genes (Fig. 2c) confirming the population structure based on BAPS.
Also, C. jejuni genomes belonging to different BAPS clusters while sharing a particular lifestyle preference differ with respect to their accessory gene content (Fig. 2d). This observation is supported, for instance, by the accessory gene content identified for the cattle-specific BAPS clusters 4 and 10 (CC-42 and CC-61) and the chicken-specific BAPS clusters 1, 5 and 9 (CC-354, CC-692, CC-257, etc.) (Figs. 2c,d). Overall, BAPS clusters with a host-generalist lifestyle preference appear to have a broader gene pool within the accessory genome content than strains identified as host-specific.

Recombination events in Campylobacter jejuni lineages.
Recombination events that show more differences between taxa than expected by mutation-driven evolutionary processes alone were illustrated in Fig. 3.
In-silico identification of host-specific factors. After identifying significant k-mers using a consensus GWAS approach, the k-mers were mapped to an annotated reference genome in order to identify coding sequences (CDS) of the genome known to promote a particular lifestyle preference of C. jejuni 25 . A visualization of the resulting genes with corresponding p-values and frequencies for the matching k-mers are provided in supplementary Figure S2.
CDS identified by k-mers in the genomes of C. jejuni isolates with lifestyle preferences in pig and cattle showed a denser distribution around the expected allele frequency than the results obtained for the genomes representing chicken-or host-generalist lineages ( Figure S2). Population structure of C. jejuni based on the core genome alignment with BAPS clusters and clonal complexes colour-coded in the inner ring; Lifestyle preferences of the genomes coded in the second -ring; and country of genome origin described in the outer ring. The leaves are coloured by the origin of each sample.   www.nature.com/scientificreports/ The genes identified by our analysis included accessory genes present in a limited number of genomic backgrounds and allelic variants of the core genome content. We identified several variants of core genes supporting specific lifestyle preferences in C. jejuni. To further evaluate the putative host-specific importance of the allelic variants identified, genes under consideration have been checked for non-synonymous base changes by comparing their predicted amino acid sequences. Several of these predicted aa sequences can be linked to particular lifestyle preferences of C. jejuni isolates. Details for all loci and aa sequence variants identified are provided in the Tables S3 (cattle), S4 (chicken), S5 (pig) and S6 (host-generalists).
Accessory genes and allelic variants of the core genome associated with C. jejuni lineages assigned as pig-specific. In the genomes belonging to BAPS cluster 11 (CC-403) we identified 21,681 k-mers which are significantly associated with the host pig. These k-mers mapped to 49 accessory genes and 78 allelic variants of the core genome (Table S5). Considering the accessory genes, 14 were exclusively found within C. jejuni genomes from pig hosts. (Table 1). Three accessory genes (A6J90_06670, A6J90_06675, A6J90_02350) belonged to transcription units encoding type II restriction modification systems (RM systems), while a further gene encodes the restriction subunit (R) of the host specificity determinant (hsdR; A6J90_08990) of a type I RM system. Additional 8/14 genes were annotated as hypothetical or putative proteins without any specific functional information available in NCBI GenBank (17.06.2020).
Considering the k-mer results for genes belonging to the core genome, nucleotide changes leading to actual effects with respect to host adaptation capabilities of certain lineages are difficult to pinpoint. Here, we noted alterations for the predicted aa sequences associated with the capability of C. jejuni to synthesize vitamins and enzyme co-factors such as TenI and Dxs (Fig. 4a). In addition, the predicted aa sequence for Cj1484 was found to be altered (Fig. 4a).
Accessory genes and allelic variants of the core genome associated with C. jejuni lineages assigned as cattle-specific. We further identified 66,491 k-mers for the cattle-associated genomes matching to 71 accessory genes and to 136 core gene variants (Table S3). According to our GWAS analysis, a particular accessory gene content which is representative for the lineages in both cattle-associated BAPS clusters (4 and 10) was not identified. However, 16 accessory genes were identified by k-mers significantly associated with CC-61 (BAPS cluster 10; Table S3). These genes belonged to a region of 9.9 kb size in C. jejuni (NCTC13261_01705 up to NCTC13261_01720). That particular locus contains 16 open reading frames encoding a HicA-HicB toxin/ antitoxin system inhibiting the transfer of mRNA in case of nutrient limitation, a protein known to be involved in extracytoplasmatic stress response (YafQ) and regulatory protein RepA for plasmid DNA repair (Table S3). www.nature.com/scientificreports/ Within the core genome we identified a 9.7 kb locus of 9 adjacent genes ( Table 2) that encode for a ribosomal complex. While the allelic variants (non-synonymous substitutions) dnaE and ffh (Fig. 4b) were identified as cattle-specific, identical variants of arsC, aroF, uraH, rplS, trmD, rimM and rpsP were identified in host-generalist BAPS cluster 8, too. However, for the genes uraH, arsC, rplS and rpsP, detected SNPs lead to synonymous changes only, indicating their biological importance as conserved housekeeping genes within the C. jejuni lineages investigated here.
Additional non-synonymous, cattle-specific allelic variants were also identified on independent positions within the genome, including the alleles Cj0495 (Fig. 4b), dsbI and Cj1233 (Table 2). Accessory genes and allelic variants of the core genome associated with C. jejuni lineages assigned as chicken-specific. In comparison to the lineages associated with cattle, pig or even the hostgeneralists, chicken-associated lineages showed the broadest phylogenetic diversification in our study, mirrored by multiple lineages and CCs ( Fig. 1), including enhanced divergence within a specific CC (CC-353 or CC-1034). Accordingly, this particular heterogeneity resulted in less host-specific signatures. The 5 712 chicken-associated k-mers identified by our GWAS analysis cover 17 accessory genes and 25 core gene variants (Table S4). A gene for a TraG-like protein of the type IV secretion system 26 was detected among the accessory genomes in 59/90 chicken-associated genomes (Table 3). TraG-like proteins are known to play a crucial role in the conjugative transfer of plasmids 27 . Additionally, two genes for putative proteins of unknown function are carried by 66 and 68 of the chicken associated strains, respectively (Table 3).
Like the cattle-associated lineages, chicken-associated genomes carry host-adapted allelic variants ( Table 3). The allele encoding a specific aa variant of rpoB was identified in most of the genomes in all three Table 1. Selected accessory genes and allelic variants of the C. jejuni core genome content pig-associated. a Locus tag for accessory genes based on C. jejuni reference genome CP022076.1 (NCBI accession). Locus tags for allelic variants of the core genome refer to C. jejuni strain NCTC11168 (NCBI accession: AL111168.1). b Position of core genes in the reference strain BfR-CA-14430. c Clusters of orthologous groups (http:// clovr. org/ docs/ clust ers-of-ortho logous-groups-cogs/). d Number of genomes assigned to a particular lifestyle carrying the gene or allelic variant (pig, cattle, chicken, host generalists, others). e A indicates that a gene belongs to the accessory genome content of C. jejuni, while V indicates a specific allelic variant of the core genome content.    www.nature.com/scientificreports/ www.nature.com/scientificreports/ chicken-associated BAPS clusters (Table 3, Fig. 4c). The gene variant encoding FlgB (Fig. 4c) is identical in BAPS clusters 1 and 5 (chicken) and the host-generalist BAPS cluster 2 (CC-21). Furthermore, a very closely related aa variant was identified in BAPS cluster 9 (chicken) as well. Additionally, the same allelic variant of the pycB gene is carried by most genomes of BAPS clusters 1 and 9 (Table 3).

Independent adaptation of host-generalist lineages.
Considering the core genome phylogeny of the C. jejuni strains presented here, the host-generalist lineages of BAPS cluster 8 appear to have evolved from independent genomic backgrounds, while other host-generalist lineages, for instance those of BAPS clusters 2, 3 and 6, appeared to be linked to each other (Fig. 1). In total, we have identified 37 339 k-mers which were mapped to 33 accessory genes and 87 core gene variants (Table S6). Accessory gene content exclusively associated with all host-generalist lineages was not identified by use of GWAS. A multitude of different allelic variants assigned to the core genome were identified for BAPS cluster 8 when compared with the genomes of the more closely related lineages of clusters 2, 3 and 6 ( Table 4). Notably we also identified closely related variants for different core genes shared by all host-generalist lineages. These included ftsX, a gene involved in cell division, arsC, an arsenate reductase, further ribosomal genes (rplS and rpsP) and Cj0459c, known as a nicking endonuclease and purine-specific ribonuclease 28 (Table 4). While the amino acid sequence encoded by ftsX shows a particular host-generalist-associated variant (Fig. 4d), the amino acid sequence determined by arsC, rplS, rpsP and Cj0459c are conserved in the C. jejuni population. Hence, k-mers identified for these CDS were associated with synonymous changes only. BAPS clusters 2, 3 and 6 harbour identical allelic variants for dnaE and ffh (Fig. 4b).
The same is true for several other genes such as dxs, cysM and pckA ( Table 4) that are broadly distributed across the C. jejuni genome and are involved in multiple metabolic pathways. Additionally, genes involved in transcriptional pathways such as rpoD and substrate transport functions like ybiT (Fig. 4d) were identified.

Discussion
We show how the recently emerging research field of bacterial GWAS was able to identify genetic signatures that possibly play important roles for the host-specificity of Campylobacter. For each of the lifestyle preferences of C. jejuni investigated, we identified a broad set of allelic variants being associated with particular host-specific lineages from distantly related BAPS clusters, providing evidence for host-adaptive genetic signatures 29 . We also extended the scheme of lifestyle preferences based on MLST to a whole genome level by applying BAPS and identified 15 distinct phylogenetic clusters. The efficiency of the proposed approach to identify lifestyle preferences by assigning host-specific or host-generalist C. jejuni lineages was verified by performing a comparison of the predicted lifestyles. For instance, CC-42 or CC-61 (cattle), CC-354 or CC-692 (chicken) and CC-403 (mammalian/pig) lifestyle assignments were verified with previously published reports on these C. jejuni lineages 8,30,31 . Additionally, putative novel lifestyle preferences of distinct lineages, i.e. CC-22 (cattle-specific) and ST-2274 (chicken-specific), were identified using the definition described above.
C. jejuni isolates assigned to either chicken or host-generalist lineages showed a diverse population structure, as reported before 32 . Contrarily, we found C. jejuni genomes identified as cattle-specific (CC-42 and CC-61) or pig-specific (CC-403) were less diverse and more clonal. Previous studies assumed that the tight clonal structure of the cattle-associated lineages CC-42 and CC-61 resulted from a more recent onset of the colonization of cattle by C. jejuni and therefore may reflect a bottleneck event in its evolution 24,29 . A similar host-adaptation process is possibly indicated by the limited diversity of CC-403 (pig-specific) assigned to BAPS cluster 11 in our study. Table 3. Selected accessory genes and allelic variants of the core genome content associated with the host chicken. a Locus tags for accessory genes based on C. jejuni reference strain NCTC13265 genome LR134498.1 (NCBI accession), while locus tags for allelic variants of the core genome refer to C. jejuni strain NCTC11168 (NCBI accession: AL111168.1). b Position of core genes in the reference strain BfR-CA-14430. c Clusters of orthologous groups (http:// clovr. org/ docs/ clust ers-of-ortho logous-groups-cogs/). d Number of genomes assigned to a particular lifestyle carrying the gene or allelic variant (pig, cattle, chicken, host generalists, others). e A indicates that a gene belongs to the accessory genome content of C. jejuni and V indicates a specific allelic variant of the core genome content. www.nature.com/scientificreports/ Genetic variation is known to be a pre-requisite to evolutionary change 33 . Since 2016, bacterial GWAS has advanced as a suitable method to identify genetic alterations associated with a phenotypical traits in large WGS datasets 34,35 , including studies on C. jejuni [13][14][15][16] . Acting like a "sieve", genetic selection allows only a subset of mutations to persist and become an observable difference between genomes 33 . Allelic variants of C. jejuni core genes, independently acquired by different phylogenetic lineages leading to changes of known or predicted amino acid sequences, likely reflect adaptation to a particular ecological niche and/or host 36,37 . We have identified allelic variants of core genes which were clearly associated with the host species pig, cattle and chicken, even among distantly related BAPS clusters [BAPS 4 and 10 (cattle); BAPS 1, 5 and 9 (chicken)]. Further allelic variants (e.g. ftsX in CC-45 and CC-21) were identified as putative markers for host-generalist lineages. This observation is supported by the lack of notable recombination between CC-45 and CC-21 30 , indicating that these variants occurred independently of phylogenetic background and geographic origin. Therefore, mutant selection leading to homoplasy would be the most reasonable assumption. More research on the subject, including isolates covering a broader time span is needed to gain further insight into the bacterial evolution of C. jejuni.

Locus tag a Gene Predicted function BfR-CA-14430 b COG c COG c description
For each of the CC-42, CC-22 and CC-61 cattle-associated lineages in BAPS cluster 4 and 10, a different set of specific accessory genes was identified. This may reflect independent colonisation events of that particular host in the evolutionary history of Campylobacter 38 . In BAPS cluster 10 we have identified genes associated with a HicA-HicB toxin/antitoxin system, which is suspected to inhibit the bacterial mRNA transfer in case of limited nutrient availability [39][40][41] .
Sharing the same host does not necessarily mean ample opportunities for DNA transfer with the host, since the preferred (sub-)niche of these CCs within the gut of cattle may differ, as it has been assumed for hostgeneralist lineages previously 30 . Furthermore, structure and composition of the gut microbiome may play a role, however little is known about the microbiome ecology and the putative lineage-specific differences among C. jejuni with respect to virulence-associated strategies such as attachment to host cell tissue 42,43 .
We identified a putative cattle-specific allelic variant of DNA polymerase III subunit alpha encoded by dnaE, in which mutations have been shown to increase the overall mutation rate of E. coli 44,45 . Since an increased mutation rate is well known as a factor influencing niche adaptation 29 , the dnaE variant may promote the host-specialization processes. In addition, we found cattle-specific changes of the gene encoding Ffh, a signal recognition particle protein (SRP). Ffh initiates the co-translational targeting of membrane and secretory proteins to the cytoplasmic bacterial membrane 46 , indicating adaptation of transport processes. In E. coli, the SRP system plays an important role in membrane protein biosynthesis, and previous research also indicated that Ffh is involved in the regulation of membrane protein translation 47 . Notably, a GTPase (FlhF) possessing an active domain most similar to Ffh, was found to be involved in flagellar gene regulation and biosynthesis in C. jejuni 48 . Again, the lack of corresponding recombination patterns indicated that niche-specific environmental pressure induced the predicted amino acid change of Ffh independently in distantly related lineages as we demonstrated in Fig. 3. Indeed, ffh has already been described as a homoplasic gene on a nucleotide level in cattle-associated C. jejuni genomes by a recent study 24 .
Most of the CC-403 and ST-1942 (pig-associated) C. jejuni in BAPS cluster 11 carry a unique set of genes encoding restriction modification (RM) systems (RM I and RM II) that may contribute to lineage-specific barriers shielding the bacteria from intrusion of foreign DNA, a phenomenon reported before [49][50][51] . As well, the frequency Table 4. Selected allelic variants of the core genome content associated with host-generalists. a Locus tags for allelic variants of the core genome refer to C. jejuni strain NCTC11168 (NCBI accession: AL111168.1). b Position of core genes in the reference strain BfR-CA-14430. c Clusters of orthologous groups (http:// clovr. org/ docs/ clust ers-of-ortho logous-groups-cogs/). d Number of genomes assigned to a particular lifestyle carrying the gene or allelic variant (pig, cattle, chicken, host generalists, others). e A indicates that a gene belongs to the accessory genome content of C. jejuni. Variant (V) indicates a specific allelic variant of the core genome content. www.nature.com/scientificreports/ and pattern of intra-lineage recombination events was unique to CC-403 and its related STs, as noted before 52 . However, due to the limited number of pig-associated clades, particular differentiation between a lineage or host specific association is challenging. While amino acid variants encoded by the tenl gene is thought to affect the thiamine metabolism and may serve as markers for cattle-specific niche adaption 24 , in this study we identified pig-specific variations as well. The amino acid changes associated with the allelic variant encoding final aromatase (TenI) needed in thiamine biosynthesis were extensive and may indicate functional alterations or even loss-of-function. Further research to characterise this gene would be useful for potential agrifood intervention strategies. Since industrial diets for pigs are generally supplemented with thiamine 53 , reduction or even shutting-off the metabolic pathway might conserve energy and seems therefore beneficial for pig-specialized C. jejuni lineages. In addition, we identified a pig-specific variant of the putative thiamine-dependent synthase encoded by dxs, again underlining the general importance of specific alterations of the thiamine pathway for host adaptation of C. jejuni lineages. The majority of the accessory genome assigned in this study as chicken-specific included, among others, genes for a putative conjugative transfer protein (TraG-like), which is commonly linked to a type IV secretion system essential for DNA transfer in bacterial conjugation 54,55 . These findings are in concordance with the recombination analysis for the chicken-specific lineages (e.g. CC-257 or CC-354), which indicated multiple horizontal gene transfer events. With respect to k-mers that indicate sequence alterations of the core genomes and lead to aa variants of the respective proteins, we noted significant k-mers mapping to the gene encoding PycB, the second subunit of the anaplerotic and glucogenic pyruvate carboxylase in C. jejuni 56 . This finding indicates a specific adaptation of a basal metabolic pathway in C. jejuni. In addition, we detected significant k-mers associated with a rpoB variant, a housekeeping gene used for investigating genetic relatedness within the Campylobacter genus 57 . Interestingly, several different mutations of rpoB enhance growth at 42.2 °C compared to the wildtype in E. coli 58 . Since the body temperature of poultry is commonly between 39 and 43 °C 59 , the rpoB variant might contribute to temperature-induced adaptive changes in C. jejuni.

Locus tag a Gene Predicted function BfR-CA-14430 b COG c COG c description
The large host-generalist lineages belonging to either BAPS clusters 2, 3, 6 (CC-21/CC-48/CC-206) or BAPS cluster 8 (CC-45) showed clear differences concerning their accessory gene content, an observation confirmed by earlier results from Yahara et al., who tracked these lineages from the chicken flock through the meat production chain as well as in clinical samples of human origin 14 . Here, we have provided evidence that accessory gene patterns were mostly BAPS clusters-specific, irrespective of the sample origin (e.g. animal, human clinical or environment). Host-generalist BAPS clusters appear to possess a larger pool of accessory genes, possibly indicating a repertoire of genes promoting survival in different hosts and environments 60,61 . This idea is supported by our recombination analysis, showing that host-generalist lineages are prone to DNA exchange, thus, natural transformation and recombination between host-generalist lineages enhances adaptive possibilities needed to survive in different hosts.
Variation of predicted aa sequences possibly associated with a host-generalist lifestyle of specific C. jejuni lineages were, for instance, identified for the cell division protein encoded by ftsX. Recent work by Riedel et al. showed that ftsX transcription is downregulated in Campylobacter lari after exposure to heat stress 62 , possibly indicating certain allelic variants may differ with respect to their stress response. As mentioned earlier, allelic variants may have evolved individually in both lineages (CC-45 and CC-21/CC-48), since the recombination analysis suggests a limited number of recombination events between BAPS clusters 8 and 2, 3 and 6.
Distinct host-specific factors, such as body temperature, the structure and composition of the gut microbiota, mucosal structures and immune system shape the adaptation strategies of C. jejuni lineages. Focusing fundamental science research in these areas will enhance the opportunity to exploit this foodborne pathogen's ability to thrive in niche environments with targeted intervention strategies in the future.

Material and methods
Strain selection and genome sequencing. A uniform stratified random collection comprising 324 C. jejuni isolates obtained from samples of four different species, including human (n = 96), chicken (n = 102), cattle (n = 98) and pig (n = 28). The original samples were collected in 16 different federal states in Germany, between 2010 and 2019. Isolates from healthy and diseased animals as well as human clinical isolates were included (Table S1). The animal-derived isolates were provided by the National Reference Laboratory for Campylobacter at the German Federal Institute for Risk Assessment (BFR) and the Institute of Microbiology and Epizootics (IMT) at Freie Universität Berlin, while the human-derived isolates were provided by the National Reference Centre for Salmonella and other Bacterial Enterics at the Robert Koch Institute (RKI). C. jejuni is rarely isolated from porcine, therefore porcine-derived isolates were limited. In order to limit spatial and temporal effects, the set of genomes investigated here was complemented by whole genome data of further 166 isolates from a Canadian study which included C. jejuni from cattle (n = 39), chicken (n = 12), human clinical cases (n = 40), environmental (n = 54) and other animal (n = 21) origins 16 . The original purpose of the Canadian study was to identify diagnostic markers which can be used for rapid screening approaches to detect C. jejuni subtypes 16 . The complete list of all 490 genomes, including available metadata such as sample origin/source and baseline typing data such as ST is provided in Table S1. Detailed protocols used for whole genome sequencing (WGS) are provided as supplementary material. Illumina raw read data sequenced for this study is available at the National Center for Biotechnology Information (NCBI) under Bioproject ID PRJNA648048. Furthermore we included the strain BfR-CA-14430, available at NCBI under the accessory numbers CP043763.1 and CP043764.1, already published as a representative C. jejuni genome by the zoonosis monitoring program of Germany 63 .
Assembly and annotation. The Illumina paired-end reads were adapter-trimmed by Flexbar v.3.0.3 64  www.nature.com/scientificreports/ settings. All genomes were annotated by Prokka v1.13 67 employing a customized database which consist of 137 complete annotated reference genomes provided by NCBI as described before 63 .
Pan-genome and phylogenetic analyses. Open reading frames (ORFs) predicted by Prokka were subsequently used as input for Roary v3.12.0 70 to calculate the pan-genome size and core genome alignment using default settings. The resulting alignment was used to calculate a maximum likelihood-based phylogeny with RAxML v.8.2.10 71 with 100 bootstraps under the assumption of the gtr-gamma DNA substitution model 72 . Clon-alFrameML v1.11 73 was used to correct for recombination events and phylogenetic groups were identified with Bayesian Analysis of Population Structure (BAPS). Here, we used BAPS with hierarchical clustering that was implemented in the R packages RhierBAPS v1.0.1 74 . Grouping of the accessory genome was further analysed by t-distributed stochastic neighbour embedding (t-SNE) 75 .
Recombination analysis. BratNextGen 76 was used to reconstruct putative recombination events based on the analysis of the core genome alignment of our selection comprising 490 C. jejuni genomes. Parameter estimation was performed based on 20 iterations and significant recombinations (p-value ≤ 0.05) were obtained using permutation testing with 100 permutations executed in parallel.

Genome-wide association study (GWAS).
In order to perform an in-depth analysis of genomic altera- . K-mers significantly representing distinct isolate origins (human, cattle, chicken or pig) were further mapped by bwa v0.7.17 77 against selected reference genomes from this study set in order to identify putative origin-specific factors, genes and consecutive gene loci. In order to reduce the false positive rate of the GWAS and account for highly unbalanced groups, we employed a bootstrapping approach. Further details can be found in the supplementary material.
The consequential set of genes was further analysed considering functional annotations and metabolic pathways using EggNog v.4.5.1 78,79 . C. jejuni lifestyle classification. In order to facilitate statistical comparison, we adapted a definition from Shepard et al. 30 and defined a set of closely-related C. jejuni lineages as host-specific if ≥ 50% genomes building the respective BAPS cluster were associated with isolates from a specific animal origin (e.g. cattle, chicken) while each of the other isolate origins contributed less than 10% in the BAPS cluster. Potential host-generalist lineages were assumed when more than 25% of the clustering genomes represented in the corresponding BAPS cluster were from C. jejuni of human clinical cases while at least two further animal origins account for more than 10% of the remaining genomes, respectively.