The O-Antigen Epitope Governs Susceptibility to Colistin in Salmonella enterica.

Group D and group B Salmonella enterica serovars differ in their susceptibility to colistin with the former frequently intrinsically resistant (MIC > 2 μg/ml); however, the mechanism has not been described. Here, we show that the O-antigen epitope in group D Salmonella governs the levels of colistin susceptibility. Substitution of the rfbJ gene in a group B Salmonella with the rfbSE genes from a group D Salmonella conferred a decrease in susceptibility to colistin. The presence of dideoxyhexose, abequose, and the deoxymannose, tyvelose, differentiate the Salmonella group B and group D O antigens, respectively. We hypothesize that the subtle difference between abequose and tyvelose hinders the colistin molecule from reaching its target. Whole-genome sequencing also revealed that increased colistin susceptibility in a group D Salmonella veterinary isolate was due to a defect in the O-antigen polymerase protein, Rfc. This study shows that two different mechanisms that influence the presence and composition of O antigens affect colistin susceptibility in Salmonella enterica IMPORTANCE Some serovars of Salmonella, namely, those belonging to group D, appear to show a degree of intrinsic resistance to colistin. This observed intrinsic colistin resistance is of concern since this last-resort drug might no longer be effective for treating severe human infections with the most common Salmonella serovar, Salmonella enterica serovar Enteritidis. Here, we show that the O-antigen epitope in group D Salmonella governs the levels of colistin susceptibility. Using whole-genome sequencing, we also revealed that increased colistin susceptibility in a group D Salmonella veterinary isolate was due to a defect in the O-antigen polymerase protein, Rfc. In summary, we show that two different mechanisms that influence the presence and composition of O antigens affect colistin susceptibility in Salmonella enterica.

As well as evolved and acquired resistance to colistin, there are several species of Gram-negative bacteria that are naturally resistant to colistin; these include Proteus spp., Serratia spp., Edwardsiella tarda, and Burkholderia cepacia complex (11)(12)(13)(14). The natural resistance to colistin in these species is due to LPS that has been modified with L-Ara4N by various mechanisms, one example being the overexpression of the arnBCADTEF operon as observed in Serratia spp. (15,16).
In 2012, Agerso et al. (17) determined the colistin MIC population distribution for Salmonella enterica at the serovar level. Salmonella enterica serovar Dublin and S. enterica serovar Enteritidis were less susceptible than other Salmonella serovars originating from humans and S. enterica serovar Typhimurium of animal/meat origin. With Salmonella Dublin and Salmonella Enteritidis belonging to the same O group (O: 1,9,12), Agerso et al. suggested that the surface LPS, O antigen, contributes to colistin susceptibility. Analysis of the pmrA and pmrB DNA sequences of the isolates showed identical sequences for all except in pmrA in one isolate of a group B Salmonella and pmrB in one isolate of a group C1 Salmonella; the MICs of colistin for both strains were Ͻ1 g/ml, in which missense mutations were observed. Agerso et al. concluded that increased MICs for colistin were associated with specific serovars, particularly S. Dublin and S. Enteritidis among the isolates studied, and recommended that Salmonella inhibited by MICs of Ͼ2 g/ml for colistin be evaluated at the serovar level. More recently, a report by the European Food Safety Authority (EFSA) and ECDC (European Centre for Disease Prevention and Control) (18) reported data for 2016 on antimicrobial resistance in zoonotic and indicator bacteria, submitted by 28 EU Member States. This report showed that colistin MICs of Ͼ2 g/ml were observed for Salmonella serovars belonging to group D, which possess O:9 somatic antigens. The report highlighted that a large proportion of the colistin resistance in Salmonella in chickens appeared to be related to the occurrence of S. Enteritidis and the higher level of intrinsic resistance of this serovar. As the European Committee on Antimicrobial Susceptibility Testing (EUCAST) recommended breakpoint concentration of 2 g/ml is the same concentration as the epidemiological cutoff values applied in the analysis, the observed colistin resistance is of concern, since colistin might not be effective for treating severe human infections with the most common Salmonella serovar.
The primary objectives of this study were to determine the mechanism of decreased susceptibility to colistin in S. Enteritidis and S. Dublin isolates and the role that the O-antigen epitope of group D Salmonella has in determining colistin susceptibility.
No differences were detected in the genes in which mutations can confer resistance to polymyxins and those related to LPS synthesis in the genomes of S. Dublin L00668-14, S. Enteritidis S02454-14, and S. Enteritidis S02576-14 isolates compared to their homologous serovars. However, compared to a nonhomologous serovar (S. Typhimurium, isolate SL1344), numerous nonsynonymous single nucleotide polymorphisms (SNPs) were observed in pmrB, pmrC, arnB, arnC, arnA, arnD, arnT, arnE, arnF, rfbB, rfbD, rfbA, rfbC, rfbI, rfbF, rfbG, rfaC, rfaF, rfaD, and rfaL, all genes in which mutations can confer polymyxin resistance (Table S1). To check whether the observed SNPs in these genes were isolate specific or serovar specific, we ran an in silico search, using "snippy," against approximately 3,000 genomes of S. Enteritidis and 200 genomes of S. Dublin, obtained from The European Bioinformatics Institute (EMBL-EBI) (https://www.ebi.ac.uk) (accession numbers in Table S2). The in silico screen showed that the SNPs observed from our genome sequence data were also detected in all of the S. Enteritidis and S. Dublin genomes screened. Therefore, we concluded that the nucleotide variants observed in theses serotypes are specific to these serotypes and not linked to colistin susceptibility differences.
Serotype conversion from O4 to O9 and O9 to O4 serotype has an effect on colistin susceptibility. In Salmonella enterica, the genes involved in the biosynthesis of the basic O antigen are located in a specific locus between the galF and gnd genes. The O-antigen unit is synthesized by sequential transfer of individual sugars from their respective dinucleotide precursors to the carrier lipid, undecaprenyl pyrophosphate (UndPP). These reactions are catalyzed by specific sugar transferases, and the last sugar in the chain confers the O-antigenic epitope characteristic of the group. In Salmonella group D serovars, the rfbSE genes encode the enzymes CDP-paratose synthase and Colistin Susceptibility in Group D Salmonella enterica ® CDP-paratose-2-epimerase, respectively. CDP-paratose synthase synthesizes paratose from its precursor, CDP-3,6-dideoxy-␣-D-mannose, and CDP-paratose-2-epimerase catalyzes the isomerization of CDP-paratose to CDP-tyvelose. Tyvelose is the last in a repeat unit of four sugars, which make up the O antigen in group D Salmonella and is the sugar that confers the O9 antigen. In Salmonella group B serovars, the rfbJ gene encodes the enzyme CDP-abequose synthase, which synthesizes abequose from its precursor, CDP-4-dehydro-3,6-dideoxy-␣-D-glucose. Abequose is the last in a repeat unit of four sugars, which make up the O antigen in group B Salmonella, and is the sugar that confers the O4 antigen (19)(20)(21). In order to test whether serovar conversion from O4 to O9 and O9 to O4 serovar, respectively, affects colistin susceptibility, we replaced the rfbJ gene in S. Typhimurium (SL1344) on the chromosome with the rfbSE genes from S. Enteritidis (NCTC 13349) and S. Dublin (CT_02021853). We also replaced the rfbSE genes on the chromosomes of S. Enteritidis (NCTC 13349) and S. Dublin (CT_02021853) with the rfbJ gene from S. Typhimurium (SL1344) (Fig. 1). To establish that the LPS pathway had not been disrupted during the construction of the rfbSE and rfbJ crossover mutants, we analyzed lipopolysaccharide profiles by silver staining.  The rfbSE genes encode the enzymes CDP-paratose synthase and CDP-paratose-2-epimerase, respectively. CDP-paratose synthase synthesizes paratose from its precursor, CDP-3,6-dideoxy-␣-D-mannose, and CDP-paratose-2epimerase catalyzes the isomerization of CDP-paratose to CDP-tyvelose. Tyvelose is the last in a repeat unit of four sugars, which make up the O antigen in group D Salmonella, and is the sugar that confers the O9 antigen. (C1) Synthesized mutation cassette containing a trimethoprim resistance gene (dhfr) (red box) flanked at the 5= end by DNA sequence homologous to rfbUVX and at the 3= end by DNA sequence homologous to rfbXESH. (C2) Synthesized mutation cassette containing a trimethoprim resistance gene (dhfr) (red box) flanked at the 5= end by DNA sequence homologous to rfbUVX and at the 3= end by DNA sequence homologous to rfbXJH. (D1) The synthesized mutation cassette (blue dashed rectangular box) recombined into the chromosome of S. Typhimurium, where the rfbJ gene has been replaced by the rfbSE genes that are highlighted in yellow. (D2) The synthesized mutation cassette (blue dashed rectangular box) recombined into the chromosome of S. Enteritidis and S. Dublin, where the rfbSE genes have been replaced by the rfbJ gene that is highlighted in green. Excision of the trimethoprim gene (dhfr) (red box) from the chromosome was conducted by using pACBSCE.
Dublin (CT_02021853), and the rfbSE and rfbJ crossover mutants showed that all gave a smooth LPS profile, and band patterns were similar, suggesting that the biosynthetic pathway had not been disrupted (Fig. 2). Serotyping of the rfbSE crossover mutants by slide agglutination confirmed successful serovar conversion from O4 to O9 with good agglutination with O9 antisera and no agglutination with O4 antisera, whereas their parent strains agglutinated only with O4 antisera ( Table 1). Serotyping of the rfbJ crossover mutants by slide agglutination confirmed successful serovar conversion from O9 to O4 with good agglutination with O4 antisera and no agglutination with O9 antisera, whereas their parent strains agglutinated only with O9 antisera (Table 1).
Compared to the parental strain, S. Typhimurium SL1344, both rfbSE (S. Enteritidis) and rfbSE (S. Dublin) crossover mutants had decreased susceptibility to colistin (MIC of 3/3.5 g/ml versus MIC of 0.85 g/ml; Table 1). Compared to the parent strains S. Enteritidis NCTC 13349 and S. Dublin CT_02021853, the respective rfbJ crossover mutants had an increased susceptibility to colistin (MIC of 1.5 g/ml versus MIC of 5.5/6 g/ml; Table 1).
Defects in O-antigen polymerase can increase colistin susceptibility in group D Salmonella. To ascertain whether the natural resistance to colistin exhibited by group Colistin Susceptibility in Group D Salmonella enterica ® D Salmonella is O antigen specific or whether other mechanisms were involved, we sequenced the genome of a colistin-susceptible S. Enteritidis (S02703-14) isolated from a chicken. Isolate S02703-14 was more susceptible to colistin (MIC ϭ 0.75 g/ml) than was the S. Enteritidis NCTC 13349 reference strain (MIC ϭ 5.5 g/ml). Following wholegenome sequencing of isolate S02703-14 and comparison with the genome of S. Enteritidis (NCTC 13349), five nonsynonomous single nucleotide polymorphisms, one nonsynonomous multinucleotide polymorphism (MNP), and one deletion mutation were found (Table S3). The deletion mutation occurred in the rfc gene which encodes the O-antigen polymerase Rfc, responsible for linking the O-antigen tetrasaccharide units into long chains (21). The deletion occurred at position 1818500 on the chromosome of isolate S02703-14 where deletion of a cytosine (C) in the codon for serine (TCT) at amino acid position 152 in the Rfc protein caused a frameshift (TCT ¡ TT ϭ S152fs). The deletion mutation was confirmed by PCR and DNA sequencing. To establish the role of the deletion mutation found in rfc, upon susceptibility to colistin, the observed nucleotide deletion (TCT ¡ TT) was introduced onto the chromosome of S. Enteritidis (NCTC 13349) by site-directed mutagenesis. The introduction of this mutation increased the susceptibility of colistin in the rfc frameshift mutant, compared to the parent strain S. Enteritidis NCTC 13349, from 5.5 to 0.8 g/ml. Poor agglutination was observed with O9 antisera (Table 1), and the LPS profile of the rfc frameshift mutant exhibited a semirough appearance (Fig. 3) with a lack of long-chain O-antigen bands observed. To determine whether the rfc frameshift mutation was responsible for the observed phenotype, we complemented both the rfc frameshift mutant and the original colistinsensitive S. Enteritidis isolate S02703-14, with wild-type rfc cloned into a pWSK30 plasmid. In both complemented strains, the susceptibility to colistin decreased from 0.75 and 0.8 g/ml, respectively, to 6 g/ml (Table 1), and good agglutination was observed with O9 antisera. The LPS profiles reverted to those observed for the parental strains (Table 1 and Fig. 3).

DISCUSSION
In the majority of bacterial species studied, natural resistance to polymyxins has been linked to the constitutive expression of the arnBCADTEF operon and/or the eptB gene. Overexpression of these genes causes the addition of phosphoethanolamine (pEtN) and/or 4-amino-4-deoxy-L-arabinose (L-Ara4N) cationic groups to the LPS, which increases the charge of the LPS and subsequently decreases polymyxin binding (22)(23)(24). Gram-negative bacteria that are known to be naturally resistant to polymyxins include Proteus spp., Providencia spp., Morganella morganii, Serratia spp., Edwardsiella tarda, and Burkholderia cepacia complex.
To date, the mechanism by which certain serovars of Salmonella enterica are intrinsically resistant to colistin has not been described. In this study, we show that intrinsic colistin resistance in two group D Salmonella serovars is attributable to the O-antigen epitope governing their antigenic specificity. The genes involved in O-antigen biosynthesis are located in the rfb gene cluster (25). The difference between the O-antigen structures of the group D and B Salmonella is the presence of CDPabequose synthase in group B versus CDP-paratose synthase in group D plus an additional enzyme, CDP-paratose-2-epimerase, which converts CDP-paratose to CDPtyvelose (26). The CDP-abequose synthase in group B is encoded by the rfbJ gene, and the CDP-paratose synthase and CDP-tyvelose are encoded by rfbS and rfbE, respectively, in group D serovars (26). In this study, we replaced the rfbJ gene in S. Typhimurium (SL1344), a group B Salmonella, with the rfbS and rfbE genes from two group D Salmonella serovars, S. Enteritidis and S. Dublin, and showed that this conferred colistin resistance to the group B Salmonella. Our data showed a decrease in susceptibility to colistin in both mutants and a change in serotype from an O4 serotype to an O9 serotype with no evidence of any other interference in LPS production as evidenced by the LPS profiles. We also conducted the reciprocal experiments and replaced the rfbSE genes in two group D Salmonella serovars, S. Enteritidis and S. Dublin, with the rfbJ gene from S. Typhimurium (SL1344). From these reciprocal experiments, our data showed an increase in susceptibility to colistin in both mutants and a change in serotype from an O9 serotype to an O4 serotype with no evidence of any other interference in LPS production as evidenced by the LPS profiles.
We also show that increased susceptibility to colistin in a group D Salmonella serovar, S. Enteritidis, was due to a defect in the O-antigen polymerase. Whole-genome sequencing of a colistin-sensitive S. Enteritidis isolate revealed a frameshift mutation in the rfc gene which encodes the O-antigen polymerase. The O-antigen polymerase (Rfc) is a membrane protein that is responsible for extension of the O antigen by the addition of repeating units (27). We hypothesize that the truncation of the Rfc protein caused by the frameshift mutation will lead to inefficient assembly and polymerization of the Colistin Susceptibility in Group D Salmonella enterica ® O-antigen subunits, therefore leading to a rough LPS profile and a more permeable cell membrane. To test this hypothesis we constructed the frameshift mutation in rfc on the chromosome of a colistin-resistant S. Enteritidis strain. Our data showed that the S. Enteritidis mutant was more susceptible to colistin, reflecting the original colistinsusceptible S. Enteritidis isolate S02703-14. The rough LPS profile exhibited by the rfc frameshift mutant compared to the type strain suggests that a more complex molecular mechanism may be involved. However, the whole-genome sequence data for the original isolate carrying the rfc frameshift mutation (S. Enteritidis S02703-14) does not indicate any other changes, and data from the rfc complement experiments suggest that the rfc mutation confers the rough LPS effect observed.
In this study, we have shown that the O-antigen epitope, whether it be due to inefficient assembly or substitution for another type, can govern the level of susceptibility to colistin in S. enterica. The immune-dominant sugars in the O antigens of Salmonella group B and group D strains are abequose and tyvelose, respectively (26). Although very similar in structure, differing only in the position of a hydroxyl group, the chemistry of the O antigen is an important factor. We hypothesize that this subtle difference between abequose and tyvelose is the cause of the reduced susceptibility to colistin observed in the strains and isolates in our study. One explanation is that tyvelose may hinder the colistin molecule from reaching its target, as the LPS component of the bacterial outer membrane is the initial cellular target of polymyxins (28).
Recent reports and studies have shown that higher levels of resistance to colistin were observed for S. Enteritidis than for other Salmonella serovars (17,18). This is of concern since the last-resort drug colistin might not be effective for treating severe human infections with the most common Salmonella serovar. More of a concern would be if this characteristic were to become transmissible and shared between colistinsusceptible serovars.

MATERIALS AND METHODS
Bacterial strains. Four veterinary Salmonella group D isolates were obtained from the Animal and Plant Health Agency (APHA), Surrey, United Kingdom (Table 1). All other quality control and reference strains used in this study are described in Table 1. Isolates were stored at 4°C and cultured on cation-adjusted Mueller-Hinton (MH) agar (Oxoid, Basingstoke, Hampshire, UK) and incubated overnight at 37°C.
Whole-genome sequencing (WGS), assembly, and annotation. Whole-genome sequencing was conducted by the Beijing Genomics Institute (BGI), using paired-end sequencing performed on the Illumina HiSeq 4000 platform. DNA extraction from each isolate was conducted using the Bioline Isolate II genomic DNA kit (Bioline, UK). The trimmed read FASTQ files for each isolate were converted using FASTQ Groomer in order to be consistent with Sanger FASTA format. The paired reads were joined by "FASTQ interlacer" using sequence identifiers. The reads were then assembled by de novo assembly using Velvet Optimiser. All operations were conducted using a Galaxy installation on CLIMB (Cloud Infrastructure for Microbial Bioinformatics). The reference genomes used were S. Typhimurium (SL1344), S. Enteritidis (NCTC 13349), and S. Dublin (CT_02021853). Annotation was performed using RAST (http:// rast.nmpdr.org/rast.cgi).
Multilocus sequence typing and plasmid and resistance gene identification. Multilocus sequence typing (MLST) for each isolate was conducted using the MLST finder tool (Centre of Genetic Epidemiology [CGE]; https://cge.cbs.dtu.dk/services/MLST/). Plasmid identification was performed using the Plasmid finder tool (CGE) (https://cge.cbs.dtu.dk/services/PlasmidFinder) which identified the number of plasmids and gave the Inc group and replicon type for each plasmid. Known resistance genes present in each genome were identified using the Resfinder tool (CGE) (https://cge.cbs.dtu.dk/services/ ResFinder) and the Comprehensive Antibiotic Resistance Gene Database (CARD) (http://arpcard .mcmaster.ca). Resfinder provided a list of plasmid-mediated resistance genes present in each genome, and CARD showed chromosomal genes with mutations and plasmid-mediated resistance genes. CARD also showed the sequences of all of the resistance genes detected. The sequences of chromosomal genes with mutations were analyzed by alignment of the gene sequence to the reference genome. The Artemis navigator (http://www.sanger.ac.uk/science/tools/artemis) was then used to locate the resistance genes that had been found. The amino acid sequence for each resistance gene in each isolate was aligned to the same gene's amino acid sequence in the reference genome using Clustal Omega (www.ebi.ac.uk/ Tools/msa/clustalo/).
Site-directed mutagenesis. A modified version of the method described by Kim et al. (31) was used for the construction of the rfbSE and rfbJ crossover chromosomal mutants and the rfc frameshift mutant. For the rfbSE and rfbJ crossover chromosomal mutants, mutation cassettes were synthesized by Integrated DNA Technologies, Inc. (USA). Each cassette contained a trimethoprim resistance gene (dhfr) flanked at the 5= end by sequence homologous to rfbUVX and at the 3= end by sequence homologous to rfbXESH for the rfbSE crossover mutant and rfbXJH for the rfbJ crossover mutant (Fig. 1). The respective mutation cassettes were inserted into the chromosome of S. Typhimurium (SL1344), S. Enteritidis NCTC 13349, and S. Dublin CT_02021853 using pSIM18 hygromycin selection (32). Recombinants were selected by plating onto LB agar with 50 g/ml of trimethoprim. For the rfc frameshift mutation, a mutation cassette was synthesized by Integrated DNA Technologies, Inc. (USA) which contained a kanamycin resistance gene (aph) flanked by sequence homologous to rfc which contained the nucleotide deletion giving rise to the frameshift at amino acid position 152. Excision of the trimethoprim and kanamycin selection cassettes from the chromosome was performed with pACBSCE (33). Mutant candidates were screened by PCR for loss of the selection marker and sequenced to confirm correct insertion of the desired mutation.
The rfc frameshift mutant and colistin-sensitive S. Enteritidis (S02703-14) were complemented with a wild-type rfc gene (amplified from S. Enteritidis NCTC 13349) on a pWSK30 plasmid.
LPS extraction. LPS was isolated as follows. Salmonella isolates were grown overnight at 37°C, and the following day the equivalent of 1 ml of a culture with an optical density at 600 nm (OD 600 ) of 1 was centrifuged and the cell pellet was resuspended in 100 l of lysing buffer (1 M Tris [pH 6.8], 2% sodium dodecyl sulfate [SDS], and 4% 2-mercaptoethanol). The suspension was then boiled for 10 min and centrifuged, and the supernatant was retained. Proteinase K was added to 0.25 mg/ml, and the sample was incubated at 60°C for 1 h. Finally, the LPS preparation was heated at 98°C for 10 min and stored at Ϫ20°C. LPS was resolved on 10% Bis-Tris SDS-polyacrylamide gels and visualized by silver staining using the SilverQuest kit (Thermo Fisher).
Serotyping. Serotyping of the Salmonella isolates in this study was conducted by slide agglutination assays using monovalent somatic O antiserum group B, factor 4 and group D, factor 9 (Pro-Lab Diagnostics). Two separate loopfuls of normal saline (0.85% sodium chloride) were placed on a clean glass slide. A small part of a Salmonella colony from an overnight culture plate was mixed thoroughly with both drops of normal saline on the slide to obtain a smooth suspension. One loopful of antisera was added to one of the bacterial suspension drops on the slide; to the other (control), one loopful of normal saline was added. The antiserum was mixed with the bacterial suspension using a sterile loop. The slide was gently tilted back and forth for 1 min and observed for agglutination under normal lighting conditions.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only.