Genomic variations leading to alterations in cell morphology of Campylobacter spp

Campylobacter jejuni, the most common cause of bacterial diarrhoeal disease, is normally helical. However, it can also adopt straight rod, elongated helical and coccoid forms. Studying how helical morphology is generated, and how it switches between its different forms, is an important objective for understanding this pathogen. Here, we aimed to determine the genetic factors involved in generating the helical shape of Campylobacter. A C. jejuni transposon (Tn) mutant library was screened for non-helical mutants with inconsistent results. Whole genome sequence variation and morphological trends within this Tn library, and in various C. jejuni wild type strains, were compared and correlated to detect genomic elements associated with helical and rod morphologies. All rod-shaped C. jejuni Tn mutants and all rod-shaped laboratory, clinical and environmental C. jejuni and Campylobacter coli contained genetic changes within the pgp1 or pgp2 genes, which encode peptidoglycan modifying enzymes. We therefore confirm the importance of Pgp1 and Pgp2 in the maintenance of helical shape and extended this to a wide range of C. jejuni and C. coli isolates. Genome sequence analysis revealed variation in the sequence and length of homopolymeric tracts found within these genes, providing a potential mechanism of phase variation of cell shape.


Results and Discussion
Identification and isolation of rod-shaped C. jejuni mutants. Tn mutant libraries are a useful tool for screening a range of phenotypes derived from random mutation. We screened a pre-existing STM Tn library constructed in C. jejuni strain M1 23 by light microscopy, with the aim of identifying bacteria with altered cell morphologies. STMs analysed in the screen were derived from 45 different signature tags, with each tag comprising 10-50 STMs. Of 1,933 Tn mutants cultured and screened individually, 89 were exclusively rod-shaped, 22 had mixed rod and helical cell morphologies and 25 showed an exclusively "short-helical" morphology, which included a mixture of helical, S-shaped, crescent and coccoid forms (Fig. 1).
The STM Tn mutants screened were generated as 1,933 independent and random mutation events but, the observed morphologies had a seemingly non-random distribution within the library: 109 of the 136 non-helical Tn mutants belonged to only seven signature tags (Fig. S1). We used linker PCR and plasmid rescue to determine the chromosomal location of the Tn insertion sites within 49 non-helical Tn mutants, and found that the pattern of non-helical morphologies did not correlate clearly with particular Tn insertion sites, suggesting that the observed cell morphologies were not the result of the primary Tn mutation event. To investigate this in more detail, we decided to test some of the genes mutated in the Tn library for their role in cell shape determination. We chose three genes for this initial analysis that were similar to genes from other bacteria that are involved in cell shape determination. These were CJM1_0159 which encodes a predicted coiled-coil region also seen in crescentin, a protein responsible for curvature in C. crescentus 21 , and CJM1_0631 and dtpT which encode large, transmembrane di-/tripeptide transporters that we predicted were either important for cell wall integrity or transported substrates necessary for PG synthesis. We mutated these genes in C. jejuni by allelic-replacement and showed that this did not consistently result in rod morphology in the WT background. Taken together, the data indicated that the Tn mutation event was not responsible for rod shape cells in the library and that therefore another source of variation may be responsible for the loss of helical cell shape.
Without a clear pattern of Tn insertion sites to account for the different morphologies within the library, we hypothesised that there was another source of variation within the non-helical STMs. Since STM morphologies were maintained throughout the cell cycle and across multiple generations, we further hypothesised this was due to a genetic mutation, as opposed to differences in genetic regulation.
Pulsed-field gel electrophoresis (PFGE), a highly discriminatory technique used for distinguishing bacterial strains 24,25 was performed to investigate whether major chromosomal rearrangements could account for the difference between rod-shaped STMs and the WT. No major chromosomal rearrangements could be detected by pulse-field gel electrophoresis.
In lieu of major chromosomal rearrangement, we performed WGS to search for polymorphisms that might explain the phenotypes. In total, 66 rod-shaped Tn mutants, 6 short-helical shaped Tn mutants and 15 Tn mutants showing mixed rod and helical morphologies were selected and sequenced alongside the WT parent and 6 helical Tn mutants as controls.
Importantly, we observed that, although the WT M1 strain is characterised by helical morphology, our WT M1 laboratory strain (believed to have been passaged only twice since original isolation) contained a mixture of helical and rod-shaped cells. This finding is consistent with the discovery of a proportion of rod-shaped bacteria within the predominantly helical targeted deletion strains already mentioned. Following this discovery, we sought to isolate pure populations of the helical and rod-shaped bacteria and noticed a clear association between cell and colony morphologies. Helical bacteria formed shiny, rounded colonies on MH agar whereas rod-shaped bacteria formed dull, flat and often larger colonies (Fig. 2). This association was confirmed by light microscopy for over 150 colonies. We hypothesise that the different colony morphologies may be due to a discrepancy in the packing of helical versus rod cells within the colony 26 , as observed in cell shape mutants of other bacterial species 18 . We used colony purification to isolate helical and rod-shaped populations from C. jejuni strains M1, NCTC11168, 81116 and 81-176 (Fig. 3), as well as from the isogenic M1 mutant CJM1_0159. A selection of helical and rod-shaped WT isolates from these four C. jejuni strains was included in the WGS analysis. We have summarized the string of methods applied and conclusions made in the process of the STM analysis, as described above, in Fig. S2.
To determine whether C. jejuni morphological changes could be observed during in vitro growth, we performed serial passages of single colonies on MH agar and screened each passage for cell shape mutants. Specifically, we identified rod-shaped bacteria in passages of helical bacteria, and for helical-shaped bacteria in passages of rod bacteria. We screened over 40,000 colonies from four passages of both helical and rod WT lineages and identified four rod-shaped mutants from the helical lineages. These rod-shaped mutants were from the first passage of isolate H2 (H2_P1_R1, where H2 = helical isolate 2, P1 = first passage and R1 = rod 1), the second passage of isolate H5 (H5_P2_R1 and H5_P2_R2) and the fourth passage of isolate H5 (H5_P4_R1). No helical-shaped revertants from the rod lineages were identified. The four rod-shaped clones, each with a helical isolate from the same lineage and passage number as a control, were analysed by WGS. This screen for cell shape variants was also used to determine the rate of helical-to-rod morphological change in C. jejuni M1, which we calculated as a frequency of 7.7 × 10 −4 , or 7.25 × 10 −5 phenotypic variants/division (Table S1).  (Table S2) and changes in the number of bases in documented phase variable regions (PVRs) ( Table S3).
Phase variation (PV) is a common source of genetic and phenotypic variation in many bacteria, including C. jejuni [27][28][29] . PV enables a stochastic 'on' and 'off ' switching of genes in bacteria, providing population diversity that may promote immune evasion or survival in varying environmental conditions 26 or if the bacterium inhabits more than one niche 30 . Regions of the bacterial genome that are particularly prone to these reversible mutations are known as PVRs and include simple sequence repeats, inverted repeats, gene duplications and methylation sites. Homopolymeric tracts (HTs) are highly susceptible to slipped-strand mispairings, which alter the length of the tracts and generate frameshift mutations during DNA replication and repair 26,31 . The frameshift mutations cause a change to the downstream amino acid sequence and almost always result in premature stop codons. In this way, PVRs are able to randomly switch genes 'on' and 'off ' and stochastically regulate gene expression 26 . Although these mutations are heritable, they are also often reversible. This is usually evident by a variety of HT lengths of a PVR within a population 27,28 or observed directly by bacterial passage 32,33 . The rate of PV is generally greater than that of spontaneous mutation and is typically calculated to 10 −4 mutations/division or more 26,28,34 . Rates of PV can differ depending on environmental conditions and/or whether the genotypic switch is from 'on-to-off ' or 'off-to-on' and these rates can vary up to 10-fold 28,[34][35][36][37] .
The propensity of C. jejuni to undergo PV 29 may be exacerbated by the absence of a functional mismatch repair (MMR) system in this species 38 . MMR systems have been found to protect bacterial genomes from slipped-strand mispairing [39][40][41] and it has been postulated that the lack of this repair mechanism in C. jejuni makes its genome more susceptible to PV 38 . Indeed, MMR mutants in other species show increased mutation rates 42,43 . Alternatively, higher rates of PV in C. jejuni could be due to inaccuracies of the C. jejuni DNA polymerase, as has also been suggested for H. pylori 44 .
In C. jejuni, PVRs have been identified in capsule, LOS and flagellin glycosylation loci as well as a range of loci of unknown function 4,27,28,45 . Most identified PVRs in C. jejuni are within polyG:C tracts of seven or more nucleotides in length 26,27,45 . However, PV in C. jejuni has also been observed in polyA:T tracts and in HTs as short as two nucleotides 27,32 , demonstrating that this phenomenon is not limited by nucleotide or tract length. However, investigations into variable polyA:T tracts are limited by their density within the C. jejuni genome, which is roughly 70% AT rich 27 . Considering all these complexities of PV analysis, it is reasonable to hypothesise that PV in C. jejuni may occur in more regions than currently recorded.
The WGS demonstrated the fluidity and susceptibility to mutation of the C. jejuni genome. There were numerous SNPs and INDELs that were distinct from the GenBank reference genome of the same strain (CP001900.1). Interestingly, genetic differences were observed in isolates from the same laboratory stock, the same isolates following in vitro passage and from isolates subjected to genetic manipulation. Many mutations identified throughout the C. jejuni genomes fell within HTs of a range of lengths and nucleotide sequence. As previously described, slipped strand mispairing at runs of Cs or Gs has been described in C. jejuni 27 . Although rarely classified as a mechanism of PV in Campylobacter, we present evidence here that runs of As or Ts are also susceptible to length variation, suggesting that these regions are particularly susceptible to mutation and that PV in Campylobacter should be re-evaluated. However, whether these mutations occur naturally within C. jejuni populations, due to an error prone DNA polymerase, a lack of DNA repair mechanisms and/or the instability of HTs, or whether the mutations emerge from stresses induced by the laboratory environment and genetic manipulation remains to be determined. For example, differences within the STM library could be derived from differences in growth or transformation micro-environments while the STM library was being made. Collectively, the presence of these mutations warns against assumptions that isolates within any C. jejuni population are genetically identical (also addressed in refs [46][47][48][49] or that any site-directed genotypic changes to C. jejuni are responsible for observed phenotypic changes. To overcome this uncertainty during phenotypic characterisation, the precedent should be maintained that C. jejuni isolates should be assessed by genome sequencing, by appropriate complementation analyses of mutants, and/or by performing laboratory manipulations and physiological assays on a panel of isolates. Mutations in pgp1 or pgp2 were present in all rod-shaped C. jejuni laboratory isolates. Every rod-shaped Tn mutant from the M1 STM library and every rod-shaped isolate from WT strains M1, 81116, 81-176 and NCTC11168 had a SNP or INDEL within the PG peptidase-encoding genes pgp1 or pgp2 ( Fig. 4. and Table 1). Rod-shaped bacteria isolated from the mixed rod-and-helical Tn mutants also contained a SNP or INDEL within pgp1 or pgp2 (as determined by PCR and Sanger sequencing) ( Fig. 4. and Table 1). No mutations in either of these genes were observed in any of the helical isolates. This strict correlation between the rod-shaped morphology and mutations in these genes was confirmed by site-directed mutagenesis (SDM) of a selection of rod STMs and rod WT isolates. SDM was performed to lock one of the variable homopolymeric tracts in pgp1 and pgp2 into its 'on' length. The rod STM isolate adopted a helical morphology after the incorporation of pgp1_ ON (Fig. S3) and all three tested rod STMs adopted a helical morphology after the incorporation of pgp2_ON (Fig. S4).
Mutations in pgp1 and pgp2 were in a wide array of base locations (Fig. 4) and their predicted effects on translation included single amino acid changes and 4 bp truncations -all of which were sufficient to cause a predicted loss of function of these proteins. Function was tested by analysis of the cell wall muropeptides of rod-shaped and helical isolates, using High-Performance Liquid Chromatography (HPLC) (Fig. S5, and Table S4). These analyses demonstrated that rod-shaped pgp1 mutants in both the M1 and 81116 backgrounds have similar muropeptide profiles, and that these are distinct from the muropeptide profiles from their respective helical WT cells. Likewise, muropeptide profiles of rod-shaped pgp2 mutants in the M1 and 81116 backgrounds were similar to each other and distinct from WT cells. The muropeptide compositions are also similar to those of pgp1 mutants, pgp2 mutants or WT in the 81-176 background 9,10 , except that the putative dipeptide (peak 3) and putative acetylated tetra-tetrapeptide (peak 12) fractions are greater in the 81116 and M1 pgp1 mutants than in the 81-176 pgp1 mutant. These profiles also expand our understanding of muropeptide compositions of helical versus rod-shaped bacterial species. Combined, these data demonstrate that many different mutations can affect the functionality of Pgp1 and Pgp2. Interestingly, the most frequent site of mutation within pgp1 was an HT consisting of 8 As towards the 3′ end of the gene. This HT had 7 or 9 As in many of the rod-shaped isolates in the M1, 81116 and 81-176 backgrounds, resulting in a truncation that is predicted to remove approximately 14% of the WT protein. The rate of morphological change is greater than that of the spontaneous mutation (Table S1), so we speculated that there may be phase variable control of cell shape.
To further investigate this hypothesis, we conducted another experiment looking for rod-shaped bacteria in passages of helical bacteria, and for helical-shaped bacteria in passages of rod bacteria. In this experiment, the starting helical isolate was M1cam (Table S5) and the rod isolate was M1 'R2' containing the HT 7A (Table S5). 112 spontaneous rod derivatives were identified from 375,000 screened colonies. Over 525,000 colonies were screened looking for helical derivatives from the rod-shaped starting bacteria without detection of changes in colony morphology. Next we performed WGS on genomic DNA obtained from each of these colonies and mapped the reads to the M1cam (CP012149) and identified the SNPs and INDELS. We discovered that the helical to rod isolates R'50, R'82 and R'88 all had normal CJM1cam_0872 (pgp1), and all had a frameshift caused by a 2-base insertion in the same position in CJM1cam_1302 (pgp2) (Fig. 4, Table 1). All other helical to rod isolates had normal CJM1cam_1302 (pgp2), and all had a frameshift caused by a single base deletion in the same position in CJM1cam_0872 (pgp1) (Fig. 4, Table 1). This screen was also used to determine the rate of helical-to-rod morphological change in C. jejuni M1, under these experimental conditions, which we calculated as a frequency of 2.1 × 10 −4 (Table S1). This frequency is similar to the frequency that we calculated with a different experimental procedure (Table S1), and the differences are likely to be due to differences in the experimental procedure, and the estimation of the number of colonies screened. These data, combined with the data from the earlier experiments, demonstrate that the helical-to-rod morphological switch occurs repeatedly in standard laboratory growth conditions. It remains possible that the rod-to-helical switch is also occurring but at a lower rate. Alternatively, different environmental pressures may be necessary to observe the rod-to-helical switch.

Clinical and environmental C. jejuni and C. coli isolates contain length variants of the 8-A tract
in pgp1. Next, we examined whether HT length polymorphisms in pgp1 and pgp2 also existed in a range of C. jejuni isolates from various environmental and clinical sources. We investigated the presence and allelic variances of pgp1 and pgp2 in 859 genomes of C. jejuni and C. coli. The genomes were from a wide range of isolates: 192 from clinical, agricultural and wild bird sources 48 , 319 from multiple stages of poultry processing, including farms, abattoirs and retail chicken meat 50 and 348 from clinical human cases 51 . Analysis of these genomes revealed that all isolates contained pgp1 and pgp2, suggesting that these genes are core to C. jejuni and C. coli. Furthermore, although no length variation was detected at the 4-A HT in pgp2 (where variation had been observed in the C. jejuni Tn mutant library), 29 genomes (3% of isolates) contained length variation at the 8-A HT in pgp1 ( Table 2).
The morphology of a selection of these C. jejuni and C. coli isolates was investigated by light microscopy. All isolates with 7-A or 9-A length polymorphisms at the 8-A HT in pgp1 were rod-shaped. Furthermore, a large proportion of isolates with the 7-A or 9-A HTs were from clinical sources (Fig. 5). Additionally, a recent report 22 described the complete genome sequence of C. jejuni RM1285, a rod-shaped morphological variant, an environmental isolate derived from the exudate of a commercial chicken breast acquired from a retail store. RM1285 was reported to have a single A deletion at the 1,187 th nucleotide of pgp1, that resulted in a 62-amino acid truncation of the Pgp1 protein 22 . We have identified that the A deletion occurs in the 8-A HT described above. Together, these data suggest that HT length variation in pgp1 may not simply be an artefact of laboratory conditions or manipulation and can affect the cell shape of C. jejuni and C. coli in the host and natural environment. However, it should be noted that none of our isolates, and we presume RM1285 22 , were screened directly from the primary source and each isolate had undergone a passage in the laboratory, therefore we cannot rule out the possibility that the change in the HT occurred during this growth step.

Conclusions
Cell shape is a defining characteristic of bacterial species and members of a particular species normally retain a constant morphology for generation after generation. Yet functional questions pertaining to how bacteria have particular shapes, and what evolutionary advantage being a particular shape confers, have been largely neglected until recently.
In this study, an STM library screen was performed to identify random mutants of C. jejuni with morphological changes. None of the genes disrupted by Tns were found to be responsible for helical cell shape. Instead, through extensive genome sequence analysis, we found that the observed morphological changes in the C. jejuni M1 STM library were explained by background mutations (SNPs or INDELS) in either of the carboxypeptidase genes pgp1 or pgp2, independent of the Tn insertions.
Reviewing the method used to generate the STM library, there were three stages in which background mutations could have emerged. The first was the growth of the WT population used to extract the genomic DNA, subsequently used in in vitro transposition. The second stage was the growth of the WT population used for natural transformation with the transposed DNA. The third stage was the growth of the transformants, prior to making frozen stocks. In any of these stages, a random mutational event could have occurred and proliferated to generate a non-helical STM. Considering the density of rod STMs within certain tags, with many STMs containing the same genomic variation, it is likely the rod-causing mutations in signature tags 50, 51, 53, 54, 55 and 57 occurred in the first or second stage. Mutations generated in these early stages would allow a single change to propagate throughout an entire tag or group of tags generated on the same day.
The discovery that both pgp1 and pgp2 are present in every genome from a collection of laboratory, environmental, wild bird, poultry and clinical C. jejuni and C. coli isolates strongly suggests pgp1 and pgp2 are core to these two helical Campylobacter species. Furthermore, the novel discovery that the 8-A HT in pgp1 (AT PPVR3) is variable within this collection of C. jejuni and C. coli isolates, the recent study by Gunther et al. 22 , and the fact that indels in pgp1 consistently correlates to a rod morphology when lengthened or shortened, emphasises the role of stochastic, heritable mutation in regulating Campylobacter cell shape in a range of environmental and host conditions.
Together, our work demonstrates the susceptibility of C. jejuni to transforming into a rod-shaped bacterium. Helical cell shape is maintained by a delicate balance of PG lengths and crosslinking 9,10 , which can be disrupted by a single point mutation at numerous locations in the PG carboxypeptidase genes pgp1 or pgp2. Yet despite a propensity of C. jejuni to undergo mutations of these genes, the helical morphology remains the dominant form in laboratory and natural environments -reinforcing the importance of helical cell shape to the success of this bacterium. While the possibility of phase variation suggests that the rod C. jejuni morphology may have an advantage in certain environments or host associations, various observations from human sources 3,52,53 and infection models 54,55 suggest that the ability to cause human campylobacteriosis is most often dependent on a helical Campylobacter, a shape that confers the greatest motility (Fig. 6) and enables the colonisation of the cecal crypts.
The PG peptidases necessary for maintaining helical cell shape may offer useful antimicrobial targets to reduce the pathogenicity of C. jejuni or decrease its prevalence within the food chain. These targets are broad-spectrum in the sense that endo-and carboxypeptidases are common throughout the bacterial kingdom 56,57 , but many PG peptidases are redundant 58,59 and therefore detrimental effects might be compensated in other species and it may also be possible to design narrow-spectrum options that specifically target Pgp1 or Pgp2. Intriguingly, Liu et al. 60  reported a bacterial cell shape-determining inhibitor of H. pylori Csd4 (homologue of Pgp1 in C. jejuni) that causes significant cell straightening of H. pylori and a diminished, yet observable, effect on the morphology of C. jejuni.

Methods
Bacterial strains, media and growth conditions. C. jejuni strains were routinely cultured on Mueller Hinton (MH) agar (Oxoid) supplemented with 5% defibrinated horse blood (Oxoid) and 5 μ g/ml trimethoprim (Tp). Defined and Tn mutants and complemented strains were selected on 10 μ g/ml chloramphenicol (Cm) or 50 μ g/ml kanamycin (Km), as appropriate. Campylobacter spp. bacterial cultures were grown in standard microaerophilic conditions (5% CO 2 , 5% H 2 , 85% N 2 , 5% O 2 ) at 42 °C in a MACS VA500 variable atmosphere work station (Don Whitley Scientific). Electrocompetent Escherichia coli and C. jejuni used in cloning were prepared and transformed as previously described 61 . Bacterial strains and plasmids used in this study are detailed in Table S5.
Signature-tagged transposon mutant library. Generation of the C. jejuni M1 STM library was described in Grant et al. 23 . Briefly, mariner-based Tns containing a chloramphenicol acetyl-transferase (cat) cassette and a unique 40 bp DNA tag ('signature tag') were used in an in vitro transposition of WT C. jejuni M1 DNA. Transposed DNA was cloned into WT C. jejuni M1 by natural transformation. Chloramphenicol-resistant Tn mutant colonies were selected after 60-72 h growth on MH agar, subcultured on fresh agar and stored at − 80 °C. Fifty mutants were recovered from each transformation. Mutants were labelled first by their signature tag (1-95). and then by a mutant number (1-50).
Colony purification by colony morphology. Agar or broth cultures of C. jejuni were serially diluted in MH broth, spread onto fresh MH agar and incubated in standard growth conditions. Colonies were inspected by eye after 24-48 h growth and either quantified or isolated based on colony morphology. Helical colonies displayed a shiny and rounded morphology, whereas rod colonies displayed a grey, flat and often larger morphology. Cell morphology was confirmed by transferring bacteria from a colony to a glass slide and viewing cells by light microscopy.
DNA sequencing. Sanger sequencing was performed by Source BioScience LifeSciences. WGS was performed at the Wellcome Trust Sanger Institute. Isolates were sequenced as multiplex libraries with 100 or 150 bp paired-end reads using next-generation Illumina HiSeq or MiSeq sequencing technology, respectively. For the sequencing of M1cam helical to rod isolates, Sequencing libraries were prepared using the NEBNext Ultra II DNA library prep kit (New England Biolabs). 250 ng DNA was sheared to 400 bp fragments in microTUBE screw-cap tubes in a M220 focused-ultrasonicator (Covaris). Following DNA library preparation, the library size was determined with a Bioanalyzer 2100 (Agilent), quantified using the Qubit dsDNA BR kit (Life Technologies), pooled in equal quantities, and analysed with the NEBNext library quant kit (New England Biolabs). The pooled library was subjected to 150 bp paired-end sequencing (Genomics core facility at Cancer Research UK). The read files were demultiplexed using the demuxFQ tool developed at Cancer Research UK. For WGS analysis, de novo   Analysis of homopolymeric tracts. The length of HTs in PVRs was assessed using the HT length of each sequencing read spanning a defined region. To be counted, the sequencing read must cover the entire region, as determined by the presence of nucleotides on either side of the HT. The distribution of HT lengths at each PVR was converted into a percentage of total reads and compiled based on whether the HT length was in or out of frame with the annotated gene sequence ('on' or 'off ' , respectively).  Table S6.

Recombinant
Site-directed mutagenesis of pgp1 and pgp2 was performed by amplifying the gene of interest with primers incorporating nucleotide change (s) (pgp1_1 to 6 or pgp2_1 to 6, respectively). Using overlapping sequences, an antibiotic resistance cassette (amplified using primers pgp1_7 and 8 or pgp2_7 and 8, respectively) was incorporated between the mutated gene and the downstream region (amplified using primers pgp1_9 and 10 or pgp2_9 Motility for all isolates decreased with increasing agar concentration until isolates were effectively non-motile through 1.0% agar (all isolates measured 1 mm diameter, roughly equivalent to the original pipette stab). Statistical differences at each agar concentration were determined using a Mann-Whitney test (*p < 0.0001). Data shown is mean and SD (n = 15). (b) Relative motility of rod isolates compared to the average motility of helical isolates decreases with increasing agar concentrations. Data shown is mean and SD (n = 3). All helical (H2, H3, H4) and rod (R2, R3, R4) isolates were also analysed by genome sequencing. and 10). PCR-amplified fragments with overlapping ends were annealed using a Gibson Assembly ® Cloning Kit (New England Biolabs) according to the manufacturer's instructions and ligated into the pMiniT vector (New England Biolabs) prior to transformation of E. coli. Purified plasmid DNA was used to naturally transform C. jejuni. The correct genomic rearrangements in the resulting C. jejuni mutants were confirmed by PCR and sequencing using the primers pgp1_ck1 to ck3 and pgp2_ck1 to ck3. Bacterial strains and plasmids used in this study are detailed in Table S5. Primers used in this study are listed in Table S6.
Plasmid rescue. Plasmid rescue was used to assess the chromosomal location of Tns within C. jejuni M1 STMs. The technique was based on the method described in Grant et al. 23 but was amended to make use of vector cloning. BglII-digested STM genomic DNA was ligated to BamHI-digested and dephosphorylated pUC19. Ligations were transformed into E. coli DH5α (Thermo Scientific) according to manufacturer's instructions and transferred onto LB agar supplemented with chloramphenicol (Cm). Colonies present after O/N incubation were used to inoculate 5 ml LB broth with Cm. After O/N incubaction of LB cultures, plasmid DNA was extracted and analysed by Sanger sequencing using primers directed out of the Tn (AJG227 and CC1, Table S6).
Linker PCR. Linker PCR was an alternative method used to assess the chromosomal location of Tns within C. jejuni M1 STMs. Linker oligonucleotides (254 and 256, Table S6) were annealed in 1x annealing buffer (100 mM Tris (pH 8), 500 mM NaCl, 10 mM EDTA) at 95 °C for 3 min and cooled at 21 °C for 1 h. The annealed Linker oligonucleotides were ligated to RsaI-digested STM genomic DNA. Ligated Linker-STM DNA fragments were amplified in a PCR using a Tn-specific primer and a Linker-specific primer (CAT15 and 258, respectively, Table S6). The PCR product was resolved by gel electrophoresis and analysed by Sanger sequencing using primers directed out of the Tn (AJG227 and CC1, Table S6).
Pulsed-field gel electrophoresis. Major chromosomal patterns were compared between WT and STM C. jejuni M1 DNA using PFGE. The RE SacII was chosen for DNA fragmentation as SalI, SmaI and KpnI (also used for PFGE of Campylobacter species) 24,25,66 all had recognition sequences within the Tn 55 . The PFGE method used was based on those described in Rivoal et al. and Ribot et al. 24,25 . WT and STM C. jejuni M1 agar cultures, washed and diluted in PBS to an OD 600 nm of 0.6-0.8, were preserved and lysed within plugs made with 2% low-melt agarose (Promega) in TE pH 8. Genomic DNA within the agarose plugs was digested with SacII for 5 h at 21 °C, washed in 0.5x TBE and resolved by gel electrophoresis in 1% agarose in 0.5x TBE. Electrophoresis was performed according to manufacturer's instructions at 6 V/cm for 20 h with a ramped pulse of 5-50 s using CHEF-DR ® II Pulsed Field Electrophoresis Systems (Bio-Rad) and connected to a LTD 20 cooling system (Hybaid). Gels were stained with 0.3x SYBR ® Safe (Invitrogen) and imaged on a GelDoc TM XR + (Bio-Rad) with Image Lab 3.0 software (Bio-Rad).
Muropeptide analysis. Peptidoglycan (PG) purification and digestion protocols were adapted from those described in Glauner 67 , Li et al. 68 and Frirdich et al. 10 . HPLC of purified and muramidase-digested C. jejuni PG was performed in the same manner and using the same instrumentation as described in Christie et al. 69 .
Calculations of mutation frequency and rate. The helical-to-rod switch observed in C. jejuni populations was quantified by mutation frequency and mutation rate. Mutation frequency, f, was calculated as the proportion of rod-shaped mutants, P R , that emerged during exponential growth of spiral isolates 70 . This value was incorporated into Drake's calculation for a constant rate of mutation 71 : 2 1 where μ is the mutation rate and N is the population size at time 1 or 2. We applied these equations to colony counts of helical and rod bacteria, plated on MH agar from liquid cultures of helical C. jejuni M1 undergoing exponential growth.
Light Microscopy. Light microscopy was performed on a Nikon Eclipse E200 light microscope under a 100x objective lens.
Scanning electron microscopy. To prepare samples for SEM, C. jejuni overnight (O/N) agar cultures were resuspended in 1 ml ddH 2 O and pelleted by centrifugation (9,300 × g) in microcentrifuge tubes. Bacteria were washed a total of four times before being fixed in 4% paraformaldehyde-1% glutaraldehyde in 0.1 M PBS. Fixed cells were stored at 4 °C prior to microscopy.
Motility assay. The motility of C. jejuni was quantified using motility agar made with 0.4%, 0.6%, 0.8% and 1.0% (w/v) select agar (Sigma) in MH broth. Motility agar was used to fill 6-well plates (7 ml of agar per well) 20 min prior to use. C. jejuni isolates were transferred via pipette tip from 12 h lawn growth (on MH agar plates) into each well of the motility agar. For each strain to be tested, three replicate 6-well plates were incubated for each motility agar concentration. Motility was measured as the diameter of the halo of motility after 12 h incubation.
Sequencing data. Genome sequencing data has been deposited in the European Nucleotide Archive (http://www.ebi.ac.uk/ena), study accession PRJEB16677; ERS and ERR accession numbers are provided in Table S7.