Reversible excision of the wzy locus in Salmonella Typhimurium may aid recovery following phage predation

Bacteriophage (phage) are promising novel antimicrobials but a key challenge to their effective implementation is the rapid emergence of phage resistance. An improved understanding of phage-host interactions is therefore needed. The Anderson phage typing scheme differentiates closely related strains of Salmonella enterica serovar Typhimurium (S. Typhimurium) based on sensitivity to a panel of phage preparations. Switches in phage type are indicative of changes in phage sensitivity and inform on the dynamics of phage interaction with their host bacteria. We investigated the molecular basis of switches between the relatively phage sensitive S. Typhimurium DT8 and phage resistant DT30 strains that are present in the same phylogenetic clade. DT30 strains emerged from DT8 strains predominantly by deletion of a genomic region affecting the wzy locus encoding an O-antigen polymerase. The deletion site was flanked by two perfect direct repeats designated attL and attR. During broth culture in the presence of a typing phage that used O-antigen as primary receptor the Δwzy genotype increased in frequency compared with culture in the absence of phage and removal of attL prevented deletion of the wzy locus. Co-culture of S. Typhimurium DT8 with a strain lacking wzy resulted in reversion of the latter to wild type. We propose a model in which reversible deletion of the wzy locus enables recovery of S. Typhimurium DT8 following predation by phage that use O-antigen as their primary receptor. This was consistent with ancestral state reconstruction of DT8 and DT30 phylogeny that supported a model of reversible transition from DT8 to DT30 in natural populations. Importance S. Typhimurium is a major pathogen of livestock that adversely affects productivity and animal welfare and poses a risk of foodborne disease in the human population. Antibiotics are used to control Salmonella infections in livestock that contributes to the antimicrobial resistance global emergency. Viruses of bacteria (phage) are one alternative to antibiotics to control Salmonella in the food chain but their successful implementation as antimicrobials is restricted by the rapid emergence of resistance to phage. A better understanding of the outcome of phage-bacteria interactions is needed to optimise the design and implementation of phage-based antimicrobials. This study identifies a genetic mechanism that confers resistance to phage that use O-antigen as a receptor on the surface of Salmonella. The mechanism is likely to impart a fitness cost on the bacterium but importantly the mechanism has the potential to be revert to a fully fit state once phage predation ceases. A model for how the mechanism may contribute to survival and recovery following phage predation is proposed.


Introduction
The bacterial family Enterobacteriaceae include important bacterial pathogens of animals and humans including Salmonella enterica (1,2) that are also a primary concern in the emergence of antimicrobial resistance (AMR).The extensive use of antibiotics in agriculture is thought to be a key factor in the emergence of AMR (3).Phages are viruses that specifically infect and kill bacteria and as such have potential as alternatives or adjuncts to antibiotics to treat infections, prevent transmission between hosts and reduce persistence in the environment (4-7) (8)(9)(10)(11).A key barrier to the implementation of phages as antimicrobials is that bacteria that are initially sensitive to phage predation readily become resistant through spontaneous mutations affecting the expression of the bacterial cell-surface receptor used by phages to initiate infection.In our previous study, emergence of resistance within 18 hours of inoculation ranged from 10-100% of S. enterica serotype Typhimurium (S.Typhimurium) co-cultured with the single phage that were due to mutations that effected expression of lipopolysaccharide (LPS) on the cell surface (12).
Bacteria and phages exist in a continuous co-evolutionary arms race in which phage predation drives the evolution of antiphage systems (reviewed in (13,14)) and phages evolve counter-measures to evade these systems (15)(16)(17)(18)(19). Phages require receptors which are constitutively present on the cell surface of bacteria for successful attachment and entry to continue the lytic cycle or undergo lysogeny.Sensitivity to phage predation is consequently heavily influenced by the presence or absence of these receptors (20).In laboratory experiments primarily investigating Escherichia coli interactions with various model phages, resistance rapidly emerged largely due to mutations affecting expression of the primary receptor (21).However, prolonged co-evolution has been observed, for example in batch cultures of Pseudomonas fluorescens involving multiple reciprocal selective sweeps of host and phage (22)(23)(24).Frequently phages target surface receptors such as lipopolysaccharide molecules (25,26), whose mutation in many cases is accompanied by a high pleiotropic cost to fitness, limiting further co-evolution (27).
Evidence for the co-evolution and outcome of phage interactions with bacteria in nature is challenging and have therefore necessarily been largely limited to observational or correlational studies.This is in part due to the lack of large datasets linking phage sensitivity with bacterial genotypic diversity (24).A unique opportunity to investigate the dynamics of host genotype and phage sensitivity arose from the availability of a combination of phage type data and whole genome sequencing generated during surveillance of S. Typhimurium in the United Kingdom (28).Prior to the year 2014 the Anderson phage typing scheme was used to differentiate strains (phage types) of S. Typhimurium based on their sensitivity to 37 different phage preparations (29,30).These phage types are stratified into more than 300 patterns of sensitivity which include 209 definitive types (DTs: DT1 -DT209), and over 118 undefined types (Us: U210 -U327) that have patterns not included in the original scheme.
Whole genome sequencing (WGS) for surveillance of S. Typhimurium is now routine (31) and in the years 2014 and 2015 both phage typing and whole genome sequencing was performed on over 1700 S. Typhimurium isolates for surveillance and outbreak detection (28).The Salmonella Typhimurium typing phages (STMPs) group in three main clusters based on their sequence similarity.Most of the phage are P22-like phage and likely use LPS as primary receptor, STMP8 and 18 are ES18-like phage that likely target FhuA, while STMP12 and 13 are SETP3-like with an unknown receptor (32).
Phylogenomic analysis of S. Typhimurium sequence data revealed that major lineages were characterised by distinct patterns of sensitivity to phage indicated by the prevalence of specific phage types (28).An observation that was also true for enterohemorrhagic E. coli where phage type correlated with bacterial lineages (33).Furthermore, multiple lineages of S.
Typhimurium containing multidrug resistant (MDR) strains that emerged and spread widely in livestock species over the past 50 years exhibited less sensitivity to predation by the panel of typing phages.There was also evidence for evolution to decreased sensitivity to the panel of phage during emergence and clonal expansion of the most recently emerged MDR monophasic (I,4, [5],12:i:-) clone of sequence type 34 (ST34) that emerged around 2005 in Europe (34,35).The ST34 epidemic strains were predominantly DT193 in the year 2014 but ancestral state reconstruction indicated that the common ancestor of the ST34 epidemic clone was DT120 that had greater sensitivity to the panel of phage.Decreased sensitivity to the phage was associated with the multiple acquisitions of a prophage mTmII that resulted in clonal expansion of progeny perhaps due to the presence of an as yet unidentified antiphage mechanism (28).
Here we investigate the genetic basis of changes in sensitivity to phage predation in a subpopulation of S. Typhimurium that are adapted to circulation in populations of duck and associated with occasional epidemics of human salmonellosis in the UK and Ireland (36,37).
S. Typhimurium isolates from this clade are predominantly DT8, with strains of DT30 that were characterised by a decrease in sensitivity to phage predation, sporadically distributed within the population structure of the clade.DT8 and DT30 are susceptible to STMP8, an ES18 like phage likely to target FhuA, an outer membrane protein.DT8 is also susceptible to STMP10, 11, 16, 20, 22, 23, 26 29 and 32, all of which are P22-like phages expected to target the O-antigen (32).We determine the molecular basis of the decreased sensitivity to phage predation leading to the change from DT8 to DT30.We test the hypothesis that precise excision of the wzy gene encoding an O-antigen polymerase is a mechanism of resistance and recovery following predation by phage that use O-antigen as receptor to initiate infection.Phylogenetic analysis.Phylogenetic trees from whole genome sequence data were constructed using SNPs identified in the whole genome sequences by aligning reads using BWA-MEM (39), variant calling with Freebayes (40) and SNP filtering using vcflib/vcftools (41), combined as a pipeline using Snippy v4.3.6 (42).Maximum-likelihood trees were constructed using a general time-reversible substitution model with gamma correction for amongst-site rate variation with RAxML v8.0.20 with 1000 bootstraps (43).

Bacterial
Bacterial growth condition and recombinant strain construction.Bacterial strains were routinely grown in atmospheric conditions at 37 C on LB plate.Liquid culture were grown in the same condition with the addition of shaking at 200 revolutions per minute (RPM).
Construction of recombinant strains was using the lambda red recombineering with primers GTGTAGGCTGGAGCTGCTTCG and CATATGAATATCCTCCTTAGT to amplify the cat gene or aphII gene from plasmids pKD13 or pKD14, respectively (44).Variable non-priming sequence at the 5' end of primers was used to direct allelic replacement in the chromosome (sup Table 1).This approach was used to knock out the wzy gene (replaced by aphII), insertion of the cat gene into the wzy-thrS intergenic region, the attL wzy (replaced by aphII gene) region and to insert a kanamycin resistance cassette in the neutral intergenic region adjacent to the iciA gene in strain L01157-10 (45).The Lambda red recombinase encoding plasmid, pSIM18 was used for the recombineering step.This plasmid uses a temperature sensitive replicons and promotor to activate recombination machinery proteins exo, beta, and gam when exposed to 42 C, and cure the lambda red containing plasmids at 37°C.Knock in of a wzy-tet cassette, and tet cassette alone was carried out as described previously with minor modifications (46).A modified pDOC-GG-glmS plasmid with ampicillin selection marker was used as a delivery vector.The plasmid encodes a λ-red recombinase which allows sequence specific homologous recombination.The constructs were designed to insert the cassettes in the glmS intergenic region which was previously reported as a neutral locus for S. Typhimurium fitness (46).All the restriction enzyme digestion steps were using BsaI type II restriction enzyme (NEB).The Golden gate assembly of the modified pDOC-GG was using primers in Supplementary Table 1.The tet cassette was obtained by digestion of the muABbla tet 5' plasmid.The cassettes were inserted in different S. Typhimurium strains as described in the text.
Estimation of ancestral phenotype histories.We generated various models, and tested model fit by using the likelihood ratio test (LRT) in R. We compared scaled maximum likelihood generated ancestral probabilities at bifurcating nodes between models generated using an equal rate for transition (Q) or allowing different rates for Q.We also generated models using a Markov-Chain with Monte-Carlo (MCMC) sampling approach using equal rates for Q and another with a predefined distribution of Qs.The likelihood ratio distribution asymptotically converges toward a χ 2 distribution.We could therefore determine which model was significant by using the LRT and identify where the result sits within a χ 2 probability table, using one degree of freedom for an extra parameter of different rates of Q.
The ancestral histories across a reconstructed phylogeny of 165 DT8 isolates and 34 DT30 isolates were assessed to determine the probable phenotype of each ancestor at each node in the tree.This was then used to determine a likely history of changes from the phage sensitive to phage resistant phenotype (DT8 to DT30).This was undertaken with maximum likelihood, and MCMC approaches (47).Maximum likelihood based states at nodes were estimated with Ancestral Character Estimation (ACE) from R package ape (48) and pastML (49).The transition rate matrices (Q) were estimated from the tip data and models allowing for different rates of transition were used.MCMC approach was conducted with discrete character mapping using posterior sampled maps from SIMMAP (50).One thousand sampled stochastic character maps were constructed after a burn-in period of 1000 iterations for Q, followed by 1,000,000 Markov-Chain steps and sampling for the posterior every 1000 generations using the pre-computed distribution of Q (51).Competing statistical models using either equal rates of transition or allowing different values for Q were compared using the likelihood ratio test (LRT, above) computed via an in-house script.The more complex model was required to reside in the right hand most 5% of a χ 2 distribution with one degree of freedom to be considered significant.This was conducted using mean log-likelihoods of sampled estimated histories from MCMC analysis using both an equal rate Q, and a precomputed values for Q estimated from the tip data.Resulting data was interpreted and viewed using R package phytools (51), iTOL (52), and pastML (49).
A permutation test was used to assess if the estimated character state at a node was due to a skewed result from sample bias, with 20 permutations.Tip labels were randomly assigned to the tree through the 'sample' R command and a starting tree constructed using the maximum likelihood-based function ACE software, with a subsequent tree sampled every 100 iterations of MCMC, 100 times, resulting in a distribution of 100 trees with ancestral states for each permutation.To assess whether the permutated data was from the same distribution as that estimated from actual tip data, pairwise, Mann-Whitney-Wilcoxon tests were performed.
Bacterial genome wide association.Genome wide association was conducted to discern potential genetic polymorphisms associated with a trait of phage resistance for 34 DT30 isolates, or susceptibility to 10 phages for 164 DT8 isolates.This was undertaken through DNA of length K (k-mer) based analysis, where k-mers were identified from draft genome assemblies of each of the 196 isolate sequences using frequency based string mining algorithm FSM-lite (53).The approach adjusts the probability of association of k-mers based on phylogenetic structure.The population structure was estimated using mash and converted to a three dimensional distance matrix (54).Subsequently mixed linear model approaches were used for testing k-mer significance implemented with Sequence Element Enrichment (SEER) (55).This was done initially with no significance filtering of k-mers and then the top 1% of k-mers in the range of p-values determined a likelihood ratio test (LRT) p-value cutoff of 1x10 -3 .k-mers with a likelihood ratio test (LRT) p-value < 1x10 -1 were plotted.
Determination of sequence read mapping to the wzy locus depth.A genomic region of 6kb encompassing the wzy locus and flanking genes was extracted from reference sequence L01157-10 (RefSeq Accession no.GCF_902500305.1), and short-read Illlumina WGS data for 196 DT8 complex isolates was mapped to this region using Bowtie2 without filtering secondary mapped reads to ensure that any reads mapping to the region would be included (56).Bedtools-2.26.0 was used to extract the read depth per nucleotide (57).These were split into 250bp bins and the mean average of each section used as raw data for heatmap using R package gheatmap of ggtree (58).The read data was normalised by dividing the raw value for 250bp bins by the average read depth over the chromosome.This enabled visualisation of each wzy region in the context of each sequencing run.SNPs were identified in wzy regions using snippy v4.3.6 as previously mentioned with DT8 isolates S04527-10 and L01157-10 as references for variant calling (38).
Determination of phage sensitivity and phage type.Phage typing was carried out as described in Public Health England's phage typing protocol for S. Typhimurium using typing phages.STMP8, 10, 18, 20, 29, and 32 were a kind gift from UKHSA.Briefly, a single colony of the bacteria to be tested was incubated in 4 ml of nutrient broth with static, atmospheric conditions at 37°C for 2 hours.Subsequently a nutrient agar plate was flooded with the culture before drying.10 μL of each phage suspension at recommended titre dilution was spotted onto the plate and incubated for 16 hours.Phage typing was used to assess that strains of DT8 and DT30 were susceptible/ resistant to the corresponding phage.Constructed mutants were also subject to challenge with the typing phage to assess the phenotype.A colony picked from the clearance zone in an experiment in which strain L01157-10 was exposed to phage STMP10 was used in reversion experiments (L01157-10 Δwzy).

Determination of genotype and wzy circularisation by PCR amplification. Presence of
the wzy region was routinely tested using primer pair TGCGACTATCAGGTTACCGT and GTTAGCGTGCGGTCAAGATC that anneal in the nucA and thrS genes.qPCR was utilized to quantify the wzy -genotype in clonal populations with and without phage selection.Specific amplification of the circularised wzy locus was using primers AAGCCGAGACTCAGAGTGAC and CTCCGCCCTAATCCACATCT, and the same primers were used for the determination of the nucleotide sequence of the amplicon using TubeSeq (Eurofins).
Quantification of wzy genotype.The relative and absolute quantification of wzy and Δwzy genotypes was determined by qPCR using an Applied Biosystems™ StepOnePlus™.The strain to be tested was grown for 16-18 hours at 37°C in LB broth with shaking atmospheric conditions.This was sub-cultured in 10 mL of LB broth with 10 4 cells per mL using a guide of an OD 600 of 3.5 equalling 10 9 cells per mL before challenge with 3.34x10 4 or 3.34x10 5 pfu/mL with STMP10.Then at 2-hour intervals for 8 hours, and a sample 24 hours post inoculum, 400 µL of culture was taken and stored at -20°C.Once all samples had been frozen genomes were extracted using Promega® Maxwell.L01157-10 was used as a positive control for growth, L01157-10 wzy -for control of growth of a resistant mutant under phage pressure, and a pure lysate with 3.44x10 5 phage to control for any effect of phage DNA increasing the Ct values.qPCR was undertaken using extracted genomic DNA with rpoD as a housekeeping control gene, and wzy assessed using primer pairs wzy_qPCR_absent_F, wzy_qPCR_absent_R and wzy_qPCR_present_F, wzy_qPCR_present_R.A limit of detection for wzy -was established to the nearest log 10 through 1 in 10 dilution of culture.
Transfer of the wzy locus during co-culture.Co-culture and PCR amplification of the wzy locus from suspected revertant colonies was used to assess the reversion of wzy.A donor strain of S. Typhimurium L01157-10 wzy + was constructed by insertion of a cat gene conferring resistance to chloramphenicol into the intergenic region between wzy and attR wzy .
A recipient strain of S. Typhimurium L01157-10 Δwzy picked from the otherwise cleared plaque formed by phage STMP32 in an agar overlay assay and determined to have the deletion by analysis of whole genome sequence was further engineered by inserting an aphII gene conferring resistance to kanamycin in the 5' intergenic region of iciA and subsequent selection of a nalidixic acid resistant variant by culture on LB agar containing 100 mg/l nalidixic acid.Two genetically unlinked selectable markers were used in the recipient to discount the possibility of transfer from the recipient strain to the donor.We reasoned that potential transfer of the aphII gene by transduction and selection for nalidixic acid resistant variants of L01157-10 wzy + are both rare events and therefore the frequency of both events occurring together is extremely unlikely.Presence of the wzy region was tested using primer pair TGCGACTATCAGGTTACCGT and GTTAGCGTGCGGTCAAGATC that anneal in the nucA and thrS genes, presence of wzy was tested using primers GCCTGAAGATTTTGGCGCAT, TGCGCTGACTTTGTTTCCTG that both anneal within the wzy gene, and presence of the cat cassette with ACAAACGGCATGATGAACCT and GCACAAGTTTTATCCGGCCT that both anneal within the cat gene.Six revertants were subject to WGS with Illumina HiSeq® 2500 to check the genotypes of the mutants and reversion of wzy within the same chromosomal location.

Results
The S. Typhimurium DT30 strains are sporadically distributed within the DT8 clade.
Isolates typed as DT8 and DT30 are predominantly isolated from ducks and form a distinct clade within the S. Typhimurium population structure (28,36,38).Of 2213 S. Typhimurium strains isolated from ducks between 1992 and 2016, 68% (n=1509) were DT8 and 23% (n=352) were DT30.To further investigate the phylogenetic relationship of DT8 and DT30 strains we established a convenience collection of 176 whole genome sequences of S.
Typhimurium isolated between the years 1993 and 2013 during surveillance by the Animal and Plant Health Agency (APHA), in the UK (Supplementary Table 1).These whole genome sequences were supplemented with 41 isolates from human infection and two genomes described previously (59).A maximum likelihood phylogenetic tree based on sequence variation in the core genome of 200 isolates (Figure 1A) indicated that strains of DT30 strains were distributed widely on the phylogenetic tree of the clade among DT8 with little evidence of clonal expansion.
The wzy locus is polymorphic in the DT8-DT30 clade and is flanked by two direct perfect repeats.To test the hypothesis that genome sequence polymorphisms account for decreased sensitivity to phage predation of DT30 compared to DT8 strains, a genome-wide association study (GWAS) was undertaken to identify k-mers associated with these two phagetypes.2,617 k-mers that had a significant (likelihood ratio test, LRT p<0.1) association with the DT8 or DT30 trait after adjusting for the phylogenetic signal, mapped to four loci in the DT8 L01157-10 chromosome.A minority of k-mers (n=602) had an LRT p-value ranging from 0.1 and 0.01 and mapped to yjbR (n=200) encoding an uncharacterised DNA-binding protein encoding gene, fucO (n=201) encoding lactaldehyde reductase or gss (n=201) encoding a bifunctional glutathionylspermidine synthetase/amidase.The majority of k-mers, (n=2,015) mapped to wzy gene which encodes an O-antigen polymerase and had a likelihood ratio of association p-value < 0.0001 (Figure 1B).Quantification of whole genome sequence reads of the S. Typhimurium DT8 and DT30 strains that aligned to sequence within and flanking the wzy locus of S. Typhimurium DT8 strain L01157-10 whole genome sequence indicated a region of zero mapped reads in 14 of 34 DT30 isolates, consistent with a deletion of this region (Figure 1A).A further 20 strains that were typed as DT30 had sequence coverage across the wzy locus.Of these, three DT30 strains with sequence coverage had mutations resulting in a premature nonsense mutation in wzy and likely to abrogate function of the o-antigen polymerase, two SNPs and a small indel.Another strain had a nonsynonymous SNP predicted to result in a lysine to asparagine substitution with unknown impact on function (Figure 1A).
The assembled sequence of S. Typhimurium DT8 strain L01157-10 has 2816 bp between the stop codon of nucA and the start codon of thrS compared to 259 bp in the assembled whole genome sequence of all 14 strains with the deletion at the wzy locus.Therefore, the deletion was 2558 bp.The genome sequence flanking the wzy gene contained two 113 bp identical direct repeats that coincided with the deletion boundaries observed in the DT30 strains in which the deletion at the wzy locus contained a single copy of the repeat sequence.The direct repeats flanking the wzy gene were therefore designated attL wzy (nucA / wzy intergenic) and attR wzy (wzy / thrS intergenic) and attB wzy (Δwzy) due to their resemblance to att repeat sequences that flank prophage in the bacterial chromosome and that mediate excision (60).
Together these observations were consistent with a site-specific mechanism for deletion of this region (Figure 1C).

Deletion of the wzy gene results in resistance to typing phage.
To investigate the role of wzy in the sensitivity to predation by typing phages, a series of genetically engineered variants of DT8 strain L01157-10 in which the wzy gene was either deleted or replaced in an alternative genomic location for complementation were constructed.Strain L01157-10 Δwzy::aphII in which wzy was deleted and replaced by the aph gene was resistant to phage STMP32 tested DT8-lysing phage (Figure 2).Reintroduction of the wzy gene into the 3' intergenic region of the glmS gene, linked to a tet gene to aid selection, resulted in return of sensitivity to phage STMP32 while the tet gene alone did not result in restored sensitivity (Figure 2).The phage sensitivity phenotype of the L01157-10 Δwzy::aph strain was therefore the same as DT30 strains.However, it is known that mutation other than those in wzy can result in changes in LPS expression that can also affect sensitivity to phage predation.We therefore tested if insertion of the wzy gene linked to a tet gene could also complement the phage sensitivity phenotype of strain S03645-11 DT30 that was resistant to typing phage STMP32 (Figure 2).Consistent with mutation of wzy in this strain being responsible for the loss of sensitivity to typing phage and the DT30 phenotype strain S03645-11 wzy, tet was sensitive to the typing phage.But introduction of tet alone resulted in no change to sensitivity (Figure 2).We therefore concluded that deletion of the wzy gene and flanking sequence was a common mechanism for the switch from DT8 to DT30 phage sensitivity phenotype.
The Δwzy genotype variants are present in cultures of S. Typhimurium DT8 that increase in frequency during phage predation.Since the wzy locus was flanked by attL wzy and attR wzy direct repeats was consistent with a site-specific mechanism of deletion we tested the hypothesis that excision of the locus occurs spontaneously during culture by amplifying by PCR across the wzy locus.PCR of genomic DNA prepared from stationary phase cultures of two DT8 strains L01157-10 and S04527-10 using oligonucleotide primers that flanked the wzy locus including the attL wzy and attR wzy repeats, resulted in two amplicons of 2.8 kb and 150 bp consistent with a mixture of the wild type wzy locus and the deletion variant.In contrast, a naturally occurring variant of L01157-10 in which the wzy locus had been deleted, picked from resistant bacteria in the zone of clearance due to lysis by phage STMP10 placed over a lawn of L01157-10 (L01157-10 Δwzy), resulted in only the 150 bp amplicon consistent with deletion of the wzy locus (Figure 3A).Using qPCR with the same oligonucleotide primers we estimated that in stationary phase LB broth culture of L01157-10, the ratio of wzy : Δwzy was approximately 500:1.To investigate the impact of phage predation on the ratio of wzy : Δwzy, we determined the change in frequency of each genotype during culture of strain L01157-10 in the presence and absence of phage STMP10.In the absence of phage STMP10, the frequency of each genotype was stable during 24 hours of culture in LB (Figure 3B).In the presence of phage STMP10, the frequency of the wild type wzy locus decreased and the frequency of the Δwzy genotype increased, consistent with selection of the deletion variant in response to phage predation.The genotype of a strain in which wzy was deleted and replaced by the aph gene remained constant in the presence of phage STMP10.Together these data suggest that the wzy locus is spontaneously excised during culture and the deletion variant is maintained at a constant frequency.Further, upon predation by a phage that uses LPS as a receptor, the frequency of the deletion variant increases relative to the wild type resulting in a change in ratio of wzy : Δwzy to approximately 30:1.

Deletion of the wzy locus is through an attL-dependent excision mechanism.
To investigate whether the direct repeats attL wzy and attR wzy were required for excision of wzy locus, strain L01157-10 recombinants in which either attL wzy or attR wzy were replaced by the aphII gene were constructed by allelic replacement.An L01157-10 ΔattR wzy strain had a growth defect during culture in LB as indicated by small colonies on LB agar, potentially due to the deletion affecting the thrS promotor region, and therefore was not investigated further.
However, an L01157-10 ΔattL wzy strain had a similar growth characteristic as the wild-type parent strain.PCR amplification across the wzy locus of strain L01157-10 resulted in a 2.8 kb band and a 150 bp amplicons but only the larger amplicon in L01157-10 ΔattL wzy (Figure 4A).This was consistent with attL wzy being essential for deletion of the wzy locus.Prophage excision that is also dependent on attL and attR repeats results in the formation of a circular molecular of the phage genome (60).To investigate if the wzy locus also formed a circular molecule, we PCR amplified genomic DNA prepared from strain L01157-10 using oligonucleotide primers that annealed within the excised region of the wzy locus in an outward facing orientation.This reaction resulted in a 0.5 kb fragment consistent with a circularisation of the wzy locus.The nucleotide sequence of the amplicon indicated that the amplicon spanned the left and right ends of the excised wzy locus consistent with a circular form of the wzy locus.Taken together, the observation suggested that loss of wzy is driven by excision of the region using the flanking attL repeat and likely the attR repeat (Figure 4B).
Δwzy deletion reversion to wzy + during co-culture with a wild-type strain.We next investigated if Δwzy can revert to wild type wzy + during co-culture of a Δwzy recipient strain with a donor strain encoding a wild type wzy gene.A donor L01157-10 strain was constructed in which the cat gene conferring resistance to chloramphenicol was inserted between the 5' end of wzy and attR wzy (Figure 5A, L01157-10 wzy, cat) so that transfer of the wzy locus could be selected by culture on chloramphenicol.A recipient L01157-10 Δwzy strain that was spontaneously resistant to nalidixic acid and with an aph gene conferring resistance to kanamycin inserted in the 5' region of the iciA gene by allelic exchange was used/constructed (L01157-10 Δwzy, aph, nal r , Figure 5B).A mixture of donor and recipient strain cultured in LB for 24 hours resulted in chloramphenicol, kanamycin and nalidixic acid triple resistant L01157-10 at a frequency of 1x10 -9 per CFU.Culture of the donor and recipient strain in LB supplemented with 0.5 μg/ml, that is known to induce prophage activation, resulted in an increase in frequency by 100-fold to 1x10 -7 per CFU (Figure 5B).

Ancestral state reconstruction of DT8 / DT30 clade is consistent with bidirectional
switching between states.Our laboratory data was consistent with a bidirectional excision and reversion of the wzy locus.To investigate the dynamics of phage sensitivity changes in natural populations of S. Typhimurium, we investigated the dynamics of DT8 and DT30 phenotype by estimating ancestral state at each bifurcating ancestral node and along each branch using two complementary methods, a Markov-Chain with Monte-Carlo sampling (MCMC) and a maximum likelihood approach.To account for varying rates of change in state we tested transition-rate matrices (Q matrices) to assess significance by testing model fit using the likelihood ratio test.For MCMC derived ancestral history probabilities, a Q allowing for different rates was significant over that with an equal rate Q, with χ 2 with p=1.88x10 -15 .The model also had the greatest log likelihood of other models (scaled ML: equal rate Q = -143.6473,different rate Q = -90.9481,MCMC: equal rate Q = -140.3482,different rate Q = -185.748),and was therefore accepted as the most likely model out of all tested.To investigate whether the highest likelihood model where phage type transition rates were unequal was significant due to increased sampling of DT8 strains, rather than based on phylogenetic structure, permutation tests were undertaken.Twenty permutations were undertaken to assess if the probability distributions of each node being DT30 from each permutation were significantly different to the real data.A total of 18 permutations out of 20 were significantly different from the data (Supplementary Figure 1A).The Probability based ancestral estimation for the model with the greatest support indicated that ancestors were most likely to be DT8 (Figure 6A).A mean of 39.659 changes in phenotype state between DT8 and DT30 was estimated.The mean posterior values for Q were: In the complementary approach, maximum likelihood estimation of the probability distribution at each node indicated maximum marginal posterior probabilities similar that estimated using the MCMC approach (Figure 6B).A node graph indicated that most ancestral nodes were predicted to be DT8, but there were multiple predictions of transitions from DT8 to DT30 and back to DT8 (Figure 6C).A total of 45 changes in phage resistance phenotype were predicted, consisting of 26 switches from DT8 to DT30 and 19 switches from DT30 to DT8.The complementary approaches produced models that were highly correlated as indicated by an R 2 of 0.675 for the probability of outcomes for each common ancestor being DT30 from scaled maximum likelihood ancestral states and stochastic mapping (Supplementary Figure 1B).Together these analyses are consistent with multiple state changes from DT8 to DT30, back to DT8 consistent with a reversable mechanism producing the resistant DT30 phenotype potentially occurring throughout the population.

Discussion
Co-evolution of bacteria and phage is well-documented in vitro (22) and there is mounting evidence that phage predation impacts population structure (28,33).Phage typing has been used historically to differentiate between closely related strains of S. Typhimurium and S.
Enteritidis based on sensitivity to a panel of phage preparations (61).Investigation of phage type in the context of phylogenetic relationship of strains is one approach to understand the impact of phage predation on bacterial populations and the mechanisms that ensure fitness and survival.In a subclade of S. Typhimurium containing strains adapted to circulation in ducks, APHA surveillance data and our phylogenetic analysis of a subset of genomes indicated that around 80% were of DT8 and 20% DT30 that were sporadically distributed within the phylogenetic structure.The DT30 strains were characterised by a decrease in sensitivity to phage that use O-antigen as their primary receptor.In S. Typhimurium, less than 50-75% of isolates within epidemic clades have the dominant phage type (28) indicating frequent switching in phage sensitivity profile.In the case of the currently dominant pandemic multidrug resistant S. Typhimurium ST34 clone (35,62), of 723 strains isolated from human infection in the UK in 2014, around 70% were of the dominant phage type DT193 (28).The remaining were of 14 different phage types.In this case the ancestor was predicted to be DT120, the second most common phage type in 2014.The switch from DT120 to DT193 was due to acquisition of a prophage, mTmII and resulted in a decrease in sensitivity phage and clonal expansion (28).Decreased sensitivity to phage predation may be a common characteristic of broad host range livestock associated S. Typhimurium, perhaps due to a fitness advantage from diminished phage predation (28).In contrast, the DT8/DT30 clade contained strains of DT30 distributed within the population of DT8, with little evidence of clonal expansion of the type with reduced phage sensitivity.
The reason for the difference in apparent fitness of DT30 investigated in this study and DT193 studied previously, as assessed by their relative level of clonal expansion ( 28) is likely to be the result the distinct mechanism of decreased sensitivity to phage.In this study, the wzy locus was deleted in 41% of DT30 strains, accounting for the change in sensitivity to phage using O-antigen as receptor and the change in phage type.Mutations in the coding sequence of wzy was also present in other DT30 strains.The wzy gene encodes an O-antigen polymerase required for the elaboration of long chain O-antigen on the surface that forms a barrier ( 63) that contributes to resistance to lysozyme (64), oxidative stress (65), colicins (66) and bile acids (67), and also affects the interaction with the complement system (68).It is therefore not surprising that mutation of wzy results in attenuation of virulence in mice, albeit not as attenuated as mutations in waaG, waaI, waaJ, wbaP and waaL, other genes involved in LPS biosynthesis that would also be expected to result in resistance to phage that use Oantigen as a receptor (69).In contrast, mTmII lysogeny responsible for the switch from DT120 to DT193 and decreased sensitivity to typing phage by an as yet unknown mechanism appears to have little or no fitness cost.
Genetic and phenotypic heterogeneity is a recurring theme in the interaction of bacteria with the host, environment and bacteriophage.The most common mechanism that generates heterogeneity in an otherwise clonal population is phase variation that results from slip strand mis-pairing, sequence inversion or differential methylation (reviewed in (70)).In bacterial pathogens heterogeneity generated from phase variation is important for functions including evasion of cross-immunity (71), persistence in the host (72), biofilm formation and detachment (73), bistable expression of energetically expensive type III secretion systems (74), and protection against bacteriophage and exogenous DNA (12,75).The role of heterogeneity in defence against bacteriophage is particularly evident for genes involved in modification of LPS, a common receptor for bacteriophage.For example, it has been proposed that phase variation of the lic2A gene of Haemophilus influenzae encodes a glycosyltransferase that modifies LPS by the addition of galactose moiety is involved in a herd-resistance-like phenomenon resulting from reduced phage titres affecting predator-prey dynamics (76).S. Typhimurium also encodes multiple glycosyl transferase genes that modify LPS and an O-antigen chain length modulating gene whose expression are also under phase variation that contribute to resistance to bacteriophage (75,77,78).We describe a potentially novel form of phase variation in which heterogeneity is generated by reversible deletion of the wzy gene required to elaborate long chain O-antigen of LPS.
The mechanism for excision of the wzy locus is not known but we establish that it requires at least attL, one of a pair of 113 bp direct repeats flanking the wzy locus.The presence of attL and attR suggest that the mechanism may be analogous to other excision and integration systems that employ integrase and in some case with an excisionase (60).However, direct repeat sequences flanking the wzy locus are unlike the att repeat sequences of phage that are inverted repeats and often form attB and attC sequences with limited sequence similarity except for short (3-10bp) sequences where homologous recombination occurs.The wzy locus repeats bare some resemblance to the FRT and loxP sites of the site-specific recombinase systems from Saccharomyces cerevisiae and P1 phage, respectively, that can operate to excise or integrate sequence based on direct repeat sequences, catalysed by a site-specific recombinase.However, FRT and LoxP sequences are also distinct from attL and attR, each being only 34bp and composed of two symmetrical 13bp recognition sequences with a variable central sequence.In comparison, the wzy locus repeats are 113bp and lack symmetry within the sequence.We were unable to identify site-specific recombinase target sequences with a similar length and characteristics to that of the attL and attR.The relatively long and identical repeats may alternatively support the possibility that excision and integration was mediated by RecA dependent or independent recombination.RecA dependent recombination occurs between repeat sequences of greater than the minimal efficient processing segment (MEP) of 23-27bp (RecA/RecBC) and 44-90bp (RecA/RecF) with little increase in frequency above around 100bp (79).The greater the distance between repeats or decrease in sequence identity of the repeat sequence decreases the efficiency of recombination, and our observations are therefore consistent with such a mechanism.A RecA independent mechanism involving crossover between replicating sister strands, may also play a role (80), although it is not clear if this is capable of forming a circular excised molecular as observed for the wzy locus.
The potential for reversion of S. Typhimurium strains with the wzy gene deletion back to wild type by transfer of wzy from bystander S. Typhimurium is a critical observation that identifies this as mechanism that is potentially analogous to classical phase variation (70).The mechanism of wzy locus transfer is not known but the frequency increases in the presence of mitomycin C that is known to induce the SOS response and activate prophage (81).Prophage mediated genetic transfer has been observed in S. Typhimurium that was induced by the antibiotic carbadox ( 82) and prophage-like gene transfer agent (GTAs) in several bacterial species, that are thought to be cryptic prophage that retain the ability to package host DNA and carry out generalised transduction (83).Circularisation of the wzy locus in the donor strain may increase the frequency of transfer by protecting the molecule from endonucleases.
Induction of the SOS response also increases expression of RecA and RecN that may be involved in homologous recombination at the attB site of the recipient and the attC site of the circularised wzy locus (84).However, it also remains possible that an integrase is responsible for this genetic event and this too has the potential to be up-regulated as part of the SOS response.
The observation that wzy deletion results in resistance to phage that use LPS as a receptor and the potential for subsequent reversion raises the possibility that this is a mechanism for recovery following phage predation.We determined that in populations of S. Typhimurium DT8 cultured in the absence of phage, approximately 1 in 500 had deleted the wzy locus, providing a pool of resistant variants that in the case of predation by phage using LPS as a receptor will survive the initial attack.However, wzy mutants have a fitness cost in the host and potentially in the environment and therefore reversion is key to this being an evolutionarily stable mechanism.To enable reversion, it is essential that S. Typhimurium with an intact wzy locus is also present following phage predation.Importantly, we determined that following culture with a phage that used LPS as a receptor, although the frequency of Δwzy variants in the population increased to 1 in 30, still around 97% of surviving S. Typhimurium had an intact wzy locus.These were likely resistant to the phage due to mutations in other LPS biosynthesis genes such a single base insertion in the pgm gene that resulted in a frame shift predicted to truncate this gene (12).The potential for wzy + and wzy -variants to survive in the host or environment for long enough to allow for the predating phage to disperse is not known.During infections of the host, strains of S. Typhimurium with mutations in various LPS biosynthesis genes were able to colonise the Peyer's patches, liver and spleen of mice following oral incoulation, but at lower levels compared to the wild-type strain.It is also noteworthy that mutation of wzy had the least impact of six genes that were tested on colonisation (69) suggesting that although Δwzy variants may be present in only a relatively small proportion of the surviving population following phage predation, the wzy + proportion may increase over time.The ability of strains harbouring mutations in various LPS biosynthesis genes to persist in the intestine is not known and this is an important factor that will determine the opportunity for reversion to occur.Another important factor is the dynamics of clearance of phage from the intestine once all phage sensitive strains have been killed and this is also currently not known.
Changes in O-antigen length due to phase variation of genes encoding chain length determination protein opvAB has also been described in S. Typhimurium and may represent a complementary system to that the wzy deletion (78).Phase variation results in a subpopulation of Salmonella which have opvAB on and opvAB off allowing defence against certain phage that utilise LPS as a receptor, that was also recognised to come with an associated cost to virulence.In our study we did not investigate opvAB expression and therefore cannot compare the relative impact on evading phage predation with that of wzy deletion.It remains an open question as to how these two systems complement each other in evading phage predation.
The wzy deletion mechanism described is likely to be applicable to Salmonella enterica serovars of serogroups B1, D1 and A and at least some subspecies II serovars that encode an alternative wzy (rfc) encoded outside of the rfb locus (85,86).Serogroup B1, D1 and A serovars are thought to have acquired the wzy in the rfc locus and the ancestral wzy in the rfb locus has been deleted leaving a sequence remnant (69, 86).Although only a subset of S.
enterica serovars have the wzy (rfc) gene, these represent serotypes that account for the vast majority of human and livestock infections.For example, five of the top ten serovars accounting for non-typhoidal infections in people in the UK in 2022 are of serogroups B1 or D1 including Enteritidis, Typhimurium, Agona, Paratyphi B (Java) and Saintpaul, that together account for over half of all infections.Typhoidal Salmonella is a major cause of morbidity and mortality in South Asia, Africa and Latin America and S. Typhi and S.
Paratyphi A that cause these infections both encode wzy in the rfc locus.Previous report of deletion of the wzy locus in strains of S. Enteritidis indicate that the deletion mechanism described in our study are likely to occur in other serovars with the wzy gene in the rfc locus (87).
In this study we investigated the deletion of wzy, the selection of a wzy -subpopulation during predation by phage that use LPS as a receptor and presented evidence of subsequent reversion to wild type.In this context we propose a working model in which phage predation results in the killing of sensitive S. Typhimurium and the survival of a subpopulation that have either deleted the wzy gene or have mutations in other LPS biosynthesis genes but retain the wzy locus (Figure 7).We propose that during infection or in the environment following isolates and whole genome sequencing.163 DT8 and 34 DT30 isolates were isolated during surveillance and obtained from the Animal and Plant Health protection Agency (APHA), and Public Health England (PHE) United Kingdom (now UKHSA).Whole genome sequence data for these isolates was produced by Illumina® Mi-Seq.Three DT8 clade long read reference genomes were (S04527-10 DT8, L01157-10 DT8, S03645-11 DT30) generated by Pacific Biosciences SMRT sequencing technology (38).
These changes of state included an average of 25.754 switches from DT8 to DT30, and 13.905 for DT30 to DT8, which corresponds to the posterior values for rates of change between the two states.

Figure 3 .
Figure 3. Determination of wzy locus genotype during culture and selection in the

Figure 4 .
Figure 4. Dependency of wzy locus deletion and formation of a circularised wzy locus molecule on attL wzy repeat sequence.(A) PCR amplification of the wzy locus from S.

Figure 6 .
Figure 6.Ancestral state reconstruction of phage type for S. Typhimurium strains

Figure 7 .
Figure 7.A model for how reversible deletion of the wzy may contribute to resistance to