A BTP1 prophage gene present in invasive non-typhoidal Salmonella determines composition and length of the O-antigen of the lipopolysaccharide

Salmonella Typhimurium isolate D23580 represents a recently identified ST313 lineage of invasive non-typhoidal Salmonellae (iNTS). One of the differences between this lineage and other non-iNTS S. Typhimurium isolates is the presence of prophage BTP1. This prophage encodes a gtrC gene, implicated in O-antigen modification. GtrCBTP1 is essential for maintaining O-antigen length in isolate D23580, since a gtrBTP1 mutant yields a short O-antigen. This phenotype can be complemented by gtrCBTP1 or very closely related gtrC genes. The short O-antigen of the gtrBTP1 mutant was also compensated by deletion of the BTP1 phage tailspike gene in the D23580 chromosome. This tailspike protein has a putative endorhamnosidase domain and thus may mediate O-antigen cleavage. Expression of the gtrCBTP1 gene is, in contrast to expression of many other gtr operons, not subject to phase variation and transcriptional analysis suggests that gtrC is produced under a variety of conditions. Additionally, GtrCBTP1 expression is necessary and sufficient to provide protection against BTP1 phage infection of an otherwise susceptible strain. These data are consistent with a model in which GtrCBTP1 mediates modification of the BTP1 phage O-antigen receptor in lysogenic D23580, and thereby prevents superinfection by itself and other phage that uses the same O-antigen co-receptor.


Introduction
Salmonella enterica subspecies enterica (S. enterica) is responsible for 98% of human infections caused by Salmonella. Over 1500 serovars of S. enterica are recognized in the Kaufman-White scheme (Grimont and Weill, 2007), which uses differences in the O-antigen of the lipopolysaccharide (LPS) and flagellar H antigen to differentiate between isolates. Over 120 serovars are recognized as having the group O4 O-antigen (Grimont and Weill, 2007), which is characterized specifically by an abequose moiety linked to the mannose with a 1-3 linkage (Reeves, 1993). This is a subgroup of the O12 positive isolates that have an O-antigen subunit backbone consisting of mannose, rhamnose and galactose. Salmonella enterica subspecies enterica serovar Tyhimurium, hereafter referred to as S. Typhimurium, is one of these O4 (previously group B) serovars.
Salmonella Typhimurium is a non-host-restricted serovar typically associated with localized gastrointestinal infection in humans. However, there is an epidemic of invasive S. Typhimurium infections in Sub-Saharan Africa predominantly associated with the novel multilocus sequence type, ST313 (Kingsley et al., 2009;Okoro et al., 2012). It is hypothesized that the immunocompromised status of the population allowed evolution of these strains since it can be traced to follow the spread of HIV and is currently not commonly found outside of Africa (MacLennan et al., 2010;Okoro et al., 2012). Sequencing of a representative ST313 isolate, S. Typhimurium D23580, identified several pseudogenes compared with other S. Typhimurium isolates. Many of these are shared with the host-restricted S. Typhi (Holt et al., 2008), which could suggest that D23580 is becoming more host restricted. An additional differentiating feature is the distinct repertoire of five prophage-like elements in the D23580 genome (Kingsley et al., 2009). One of these is the BTP1 prophage, and it was recently determined that this prophage carries glycosyltransferase (gtr) genes that constitute a putative O-antigen modification operon (Davies et al., 2013).
The O-antigen-modifying gtr operons have been found in phage genomes, on prophages within Salmonella genomes and in the context of phage remnant sequences (Vernikos and Parkhill, 2006;Davies et al., 2013). The gtr operons described to date consist of three genes: gtrA, gtrB and gtrC. Based on biochemical analysis of P22 mediated O-antigen modification (Makela, 1973) and related gtr operons in Shigella flexneri, it is hypothesized that GtrA and GtrB add the glucose to a carrier lipid and flip it to the periplasm, where the GtrC protein performs the modification while the O-antigen side chain is being assembled (reviewed in Allison and Verma, 2000). Phylogenetic mapping of the GtrC proteins showed that there are at least 10 different 'families' of gtr operons in S. enterica genomes (Davies et al., 2013), and individual strains can encode for as many as four different gtr operons. It is thought that each family is responsible for a unique modification. Furthermore, the expression of many gtr operons is under the control of epigenetic phase variation (Broadbent et al., 2010a). Thus, the gtr repertoire of an isolate with four phase-varying gtr operons could result in up to 16 individual phenotypes within a population based just on the O-antigen composition.
Of the 10 GtrC families identified in Davies et al., the biochemical activity has been identified for only two, family 1 GtrC (serotype O1) and family 3 GtrC (serotype O12 2). Specifically, the phage P22 family 1 gtrC operon results in the addition of a glucose residue via an α1-> 6 linkage to the galactose of the O-antigen subunit and the O1 serotype (Fukazawa and Hartman, 1964;Makela, 1973;Van der Byl and Kropinski, 2000). Recently, the factor O122 modification was attributed to a gtr operon from family 3, which mediates addition of a glucose molecule to the galactose via an α1-> 4 linkage (Bogomolnaya et al., 2008). However, since the expression of both are under the control of phase variation, the associated factor O1 and O122 modifications will not always be detectable (Broadbent et al., 2010a). To date, the activity of the remaining families of GtrC proteins has not been elucidated.
There is evidence that the Gtr-mediated modification helps prevent superinfection by altering the O-antigen (co)-receptor recognized by the phage, which provides a rationale for the presence of these genes on phage genomes. The P22 tailspike protein (TSP) contains endorhamnosidase activity to cleave the O-antigen and better gain access to the bacterial surface. This is blocked by α1-6 glucosylation of the galactose in the O-antigen subunit (O1 serotype), which is mediated by the P22 gtr operon. The rhamnosidase activity is not blocked by the chromosomally encoded family 3 gtr operon that results in α1-4 glucosylation (O12 2 serotype). In contrast, phages 9NA and KB1 are able to cleave O-antigen containing the α1-6 glucosylation but not the α1-4 glucosylation, so there is some difference in the glycanase activity of different phages. P22 infection is also blocked by acetylation of the rhamnose (Wollin et al., 1987). More recently, the phase variable expression of the O122 modification was demonstrated to directly affect the ability of phage SPC35 to infect S. Typhimurium (Kim and Ryu, 2012).
Many gtr operons have been retained in the genome of Salmonella serovars despite the degradation of the surrounding phage sequences (Davies et al., 2013). This indicates that these operons may provide some additional, direct benefit for the bacterium. Indeed, in S. flexneri, glucosylation by the products encoded in the gtrV operon increases virulence of the strain by compacting the O-antigen layer to allow the type 3 secretion system better access to epithelial cells (West et al., 2005). In S. Typhimurium, the family 3 gtr-mediated glucosylation of the O-antigen is associated with an increase in the persistence of the bacteria in the intestine (Bogomolnaya et al., 2008).
The gtr operon encoded by BTP1 is classified as a family 2 gtr operon (Davies et al., 2013); this family is associated with invasive Salmonella serovars, such as S. Typhi and S. Paratyphi A. D23580 was the first S. Typhimurium isolate in which a family 2 gtr operon was identified, which suggests that the modification performed by this operon may have contributed to the invasive capacity of this invasive non-typhoidal Salmonellae (iNTS) isolate. In this study we examined the functionality and significance of the prophage-encoded gtr BTP1 present in the iNTS strain D23580.

Results
The gtr BTP1 operon is required for full length of O-antigen in iNTS strains with BTP1 prophage To assess the role of the family 2 gtr BTP1 operon (STMMW_03911-03921), these two genes were deleted and introduced into the native BTP1 prophage in S. Typhimurium D23580 (sMV189), resulting in Δgtr BTP1 (sMV386). The LPS was analyzed on a sodium dodecyl sulfatepolyacrylamide gel electrophoresis (SDS-PAGE) gel to determine if a shift could be observed that would be indicative of addition of a sugar moiety (Davies et al., 2013). Unexpectedly, the O-antigen of this Δgtr BTP1 strain was very short (Fig. 1A, lane 2). The O-antigen of the Δgtr BTP1 strain differs from that of a mutant in the wzz gene, which is responsible for producing the long O-antigen chain length; this difference is especially evident in the range of 5-9 O-antigen subunits (Fig. 1A, lane 2 vs. 4). The Δgtr BTP1 LPS phenotype also differed from a rough strain that lacks an O-antigen (data not shown). This indicates the gtr BTP1 mutation did not affect the core LPS and O-antigen biosynthesis pathways, and the reduced O-antigen length is due specifically to the absence of gtr BTP1 . The short O-antigen phenotype was complemented by the cloned gtr BTP1 operon (sMV359) as well as by the family 2 gtr operon from S. Enteritidis PT4 (referred to as SEN-F2; sMV338); the GtrC SEN-F2 protein shares 77% identity with GtrC BTP1 . However, complementation with the family 1 phage P22 gtr operon (sMV531) only partially restored the amount of long O-antigen (Fig. 1B).

GtrC BTP1 does not require GtrA/B for activity
Comparative genome analysis showed that family 2 gtr operons differed from other gtr operons due to the presence of a truncation of the gtrB gene (Davies et al., 2013). Analysis of specifically the gtr BTP1 family 2 operon identi-fied additional differences. First, gtr BTP1 has a complete gtrB gene deletion, unlike the C-terminal gtrB truncation found in other family 2 gtr operons (Davies et al., 2013). Second, gtrA BTP1 (referred to further as gtrA*) has a premature stop codon at nt 276 of the coding sequence, compared with one at nt 390 in the S. Typhi and S. Enteritidis family 2 gtr operons.
The mutations associated with the gtr BTP1 operon suggest that the GtrA/B proteins are no longer necessary for the activity of GtrC. Indeed, the Δgtr BTP1 short O-antigen phenotype can be complemented by providing just GtrC BTP1 in trans (Fig. 2, lane 3). However, another possibility is that GtrA/B complementation occurs from one of the other chromosomal gtr operons. This is conceivable since the amino acid sequence identity of GtrA and GtrB between different gtr families is high, and the D23580 chromosome codes for family 3 and family 4 gtr operons in addition to gtr BTP1 (Davies et al., 2013). To address the necessity of GtrA/B for GtrC BTP1 functionality, a D23580 derivative with deletion of all three different gtr operons was constructed ('basal' D23580 strain) (sMV688). In this strain with an empty vector, the O-antigen is short (Fig. 2, lane 4). Expression of long O-antigen is restored by complementation with just GtrC BTP1 (Fig. 2, lane 5). This shows that GtrC BTP1 -dependent O-antigen modification does not require any GtrA or GtrB proteins. Furthermore, this  indicates a mechanistic difference between modification by family 2 GtrC proteins and GtrC proteins from families with conserved gtrAB genes.
Identifying key features of the GtrC BTP1 protein Alignment of the family 2 GtrC proteins with members of GtrC family 1 or 3, which have documented glucosylation activity, shows the former have a 33 amino acid conserved region toward the N-terminus that is not shared with the GtrC proteins from the other families (data not shown). BLAST analysis showed that this region aligned with the N-terminus of the S. Typhimurium O-antigen acetyltransferase OafA, and also contains the RXXR motif that is required for Oac-dependent acetylation of the O-antigen in S. flexneri (Thanweer and Verma, 2012) (Fig. 3A). A site-directed mutant in the RXXR motif was generated to determine if this region was also important for GtrC BTP1 activity; the R71A/R74A mutation of the GtrC BTP1 protein (sMV648) failed to complement the short O-antigen phenotype of the Δgtr BTP1 strain (Fig. 3B, lane 4). This mutant GtrC protein is still expressed and can be detected in the membrane at levels similar to wild-type GtrC (Fig. 3C). These data show that the RXXR motif is required for GtrC BTP1 activity, in the same way as it is required for Oac acetylation in S. flexneri.
GtrC BTP1 distinguishes itself from S. flexneri Oac, first, by being in the context of a (remnant) gtr operon. Second, topological analysis demonstrated that Oac has 10 transmembrane domains and a C-terminal cytoplasmic domain (Thanweer and Verma, 2012), whereas GtrC BTP1 is predicted to have 11 transmembrane domains and a 273 amino acid long C-terminus in the periplasm (Fig. 4A). To determine the localization of the RXXR motif and the C-terminal domain, GtrC BTP1 protein topology was analyzed with PhoA protein fusions. The PhoA enzyme is only active in the periplasm, so the location of loops between transmembrane domains can be determined by assessing the absence or presence of PhoA enzymatic activity in strains containing these fusions. Five different fusions were created (vectors pMV406-410) and PhoA activity  was investigated in both Escherichia coli and S. Typhimurium. A fusion at amino acid 37, at the end of the first predicted transmembrane domain, was PhoA-positive, whereas fusions at amino acids 56 and 76 were PhoAnegative; the latter two amino acids flank the RXXR loop, indicating those essential amino acids are located in the cytoplasm (Fig. 4B). In agreement with the topology prediction, PhoA fusions at amino acids 370 (at the beginning of the predicted C-terminal periplasmic region) and 640 (at the C-terminal end) resulted in PhoA-positive colonies, indicating periplasmic localization. GtrC BTP1 activity furthermore requires at least the final 238 amino acids of the predicted 273 amino acid periplasmic C-terminus, as the O-antigen of a strain with gtrC BTP1 truncated at amino acid 402 (sMV548) resembles that of the Δgtr BTP1 strain (sMV386) (Fig. 4C). These data are consistent with the long C-terminus of GtrC BTP1 being located in the periplasm and having an essential role for the biological function of the protein.
Taken together, the data described earlier suggest that the family 2 gtr operon encodes an O-antigen acetyltransferase. First, there is no requirement for GtrAB proteins that are necessary for glucose modification, and second, the RXXR motif is essential, similar to the role of this motif in S. flexneri O-antigen acetyltransferase Oac.

gtrC BTP1 expression is not phase variable
Expression of most gtr operons across the gtr families, except family 4, is under the control of phase variation associated with a signature sequence in the regulatory region (Broadbent et al., 2010a;Davies et al., 2013). The sequence upstream of gtrA* in BTP1, however, contains a 365 nucleotide sequence that disrupts this signature sequence. To assess whether the insertion disrupts phase variation, a strain carrying a fusion of this region from gtr BTP1 to lacZ was analyzed. This strain gave rise to colonies with a uniform Lac+ (ON) phenotype (data not shown), reflecting a lack of phase variation.
Two transcriptional start sites (TSS) of gtr BTP1 were identified in strain D23580 by differential RNA-seq analysis. One TSS was upstream of gtrA*, and the second upstream of gtrC BTP1 (Fig. 5A). The former is the same TSS previously identified for the gtr LT2_I operon (STM0559-7) in S. Typhimurium strain LT2 that is phase variable (Broadbent et al., 2010b). However, as shown earlier, in strain D23580, transcription from this promoter region is not controlled by phase variation. The second TSS is in a region created by the gtrB deletion in D23580 and presumably would not be present in other gtr operons. Transcriptomic data from RNA-seq experiments (Fig. 5B) indicate B. Absolute gene expression of the gtr BTP1 locus. The heat map shows the levels of gene expression of the gtr BTP1 genes as TPM values, derived as described (Kroger et al., 2013). that both gtrA* and gtrC BTP1 are usually expressed, and therefore a degree of O-antigen modification is likely to occur, under a wide range of growth conditions. These ranged from standard laboratory growth conditions to those designed to mimic in vivo infections (Kroger et al., 2013; see Experimental procedures section for further details). Future work is needed to determine the relative contribution of the two promoters for gtrC BTP1 expression.
The Δgtr BTP1 effect on O-antigen length requires the BTP1 phage tailspike protein Analysis of the sequence adjacent to the gtr BTP1 operon identified gene STMMW_03901, which is convergently transcribed (see Fig. 5A). STMMW_03901 encodes for a TSP with 70% identity to that of the well-studied Salmonella phage P22 protein. The phage P22 TSP has endorhamnosidase activity that allows cleavage of the O-antigen. Specific amino acids important for this enzymatic activity of P22 TSP (namely, Glu 359, Asp 390 and Asp 392) are present in BTP1 TSP Steinbacher et al., 1996). Thus, the shorter O-antigen of the Δgtr BTP1 strain could be due to BTP1 TSP-dependent cleavage of full length O-antigen.
This hypothesis was tested by deleting the TSP gene in the D23580 Δgtr BTP1 mutant (Fig. 6). When the Δgtr BTP1 mutation is combined with the TSP deletion, the same full length O-antigen is seen as in the wild-type strain (Fig. 6, lane 3). This correlation between O-antigen length, depending on the presence of gtrC BTP1 and the TSP with its associated rhamnosidase activity, held true for two other iNTS strains (5597 and A38596) that are phylogenetic relatives of D23580 and harbor the BTP1 prophage (Kingsley et al., 2009;Okoro et al., 2012) (Fig. 6, lanes  4-9). These data lead us to conclude that in wild-type D23580, GtrC BTP1 acts to protect the O-antigen from cleavage by the phage BTP1 tailspike-encoded rhamnosidase activity.

Significance of GtrC BTP1 in phage-bacteria interactions
The findings described earlier are consistent with a model in which prophage-encoded GtrC BTP1 mediates O-antigen modification, possibly via acetylation. Modification of a receptor is a common phage strategy to protect a lysogen from superinfection by phages that use the same (co)receptor, in this case the O-antigen. Phage P22 is known to bind O-antigen as a receptor via its TSP and the BTP1 TSP sequence is homologous to that of P22 (Steinbacher et al., 1997), suggesting that BTP1 may also use the O-antigen as a (co)-receptor. Furthermore, our data indicate that unmodified, full-length O-antigen on the cell surface can be cleaved by the BTP1 phage TSP similar to P22 (Steinbacher et al., 1997), which can facilitate phage penetration.
In this model, the short O-antigen of the Δgtr BTP1 strain is a result of BTP1 phage-dependent O-antigen cleavage and, therefore, free phage BTP1 particles should be present in cultures of strain D23580. Indeed, plaques are observed upon spotting supernatant of a culture of strain D23580 on a lawn of 'basal' LT2, which is a mutant that lacks both native gtr operons (Fig. 7A). No plaques were observed with the supernatant of a D23580 mutant that has the BTP1 TSP gene deleted from the prophage, indicating these plaques are derived from BTP1 phage and not from other prophage present in this strain (Fig. 7A) (Kingsley et al., 2009). Furthermore, culture supernatants from the D23580-related iNTS strains A130 and A38596 also showed plaque formation, indicative of free BTP1 phage (data not shown).
To examine whether the GtrC BTP1 O-antigen modification affects phage susceptibility, spot titer tests were carried out with the BTP1-and P22-containing supernatants. Plaque formation by these phages was tested against the basal LT2 strain either with or without GtrC BTP1 , and against wild-type D23580 and the Δgtr BTP1 mutant. As expected, no plaque formation occurs on a D23580 lawn, consistent with an active phage repressor system. Interestingly, some P22 plaques did form on the Δgtr BTP1 mutant, suggesting that P22 is a heteroimmune phage (right column, Fig. 7B). Importantly, GtrC BTP1 activity blocked infection by both phages BTP1 and P22 (Fig. 7B). Specifically, BTP1 and P22 formed plaques on the basal LT2 strain containing only empty vector (Fig. 7B). This plaque formation, however, is absent when spotting BTP1 phage on this strain containing pgtrC BTP1 (sMV816). Similarly, P22 infection of LT2 is strongly inhibited by pgtrC BTP1 (Fig. 7B). P22 also does not infect wild-type D23580 but can infect the Δgtr BTP1 D23580 mutant. These data illustrate the protective nature of the modification of the O-antigen by GtrC BTP1 .
Importantly, both the BTP1 and P22 phages mediated cleavage of the full length O-antigen present on basal LT2 (Fig. 7C, lanes 2 and 4). When the GtrC BTP1 protein was present, O-antigen was not cleaved from the surface of the bacteria and full length LPS was maintained (Fig. 7C,  lanes 6 and 8). No O-antigen cleavage was observed with supernatant from the Δgtr BTP1 ΔTSP strain that lacks both gtr BTP1 and TSP (Fig. 7C, lanes 3 and 7). Thus, GtrC BTP1 activity protects the O-antigen from BTP1 TSP mediated O-antigen cleavage.
Another role of phage receptor modification could be the prevention of readsorption of phage released by lysogenic kin. To address whether GtrC BTP1 -mediated O-antigen modification can prevent readsorption, phage titers from wild-type D23580 and the Δgtr BTP1 mutant were compared A. Supernatant from D23580 cultures, containing BTP1 phage, or the ΔgtrΔTSP mutant were spotted on lawns of basal LT2 + pLAC22. B. Supernatant from cultures contacting BTP1 or P22 phage were spotted at equal titer (about 3.5 × 10 5 pfu ml −1 ) on lawns of wild-type D23580, the Δgtr BTP1 mutant, basal LT2 + pLAC22 or basal LT2 + pgtrC BTP1 . C. Cultures of basal LT2 containing either empty vector or pgtrC BTP1 were incubated with supernatant of cultures from strains (as indicated along top of gel) to investigate O-antigen cleavage from the surface. LPS analysis performed as described in Fig. 1.   Fig. 8. Modification by GtrC BTP1 prevents readsorption of phage. A. BTP1 titer is reduced in a strain with a deletion of gtr BTP1 . Supernatants containing BTP1 phage, obtained from a D23580 culture or the Δgtr BTP1 mutant, were spotted on lawns of basal LT2. Dilutions from the original supernatant are given. B. BTP1 and P22 phage titers are affected by adsorption with strains having or lacking GtrC BTP1 O-antigen modification. BTP1 and P22 phage were incubated for 5 minutes with the indicated strains (top row shows phage titer pre-adsorption). Bacteria and phage bound to their surface were removed by pelleting and the free, unadsorbed phage numbers determined by spotting the supernatant on a lawn of basal LT2. The same dilution of phage for all samples is shown. (Fig 8A). Phage concentrations produced by overnight cultures of D23580 were 5.8 × 10 7 pfu/ml, but the Δgtr BTP1 mutant only produced 1.4 × 10 6 pful/ml. The lower titers in the absence of GtrC BTP1 could be due to phage particles rebinding to the surface of the bacteria that express an un-modified O-antigen. To further test adsorption directly, BTP1 phage-containing supernatant was incubated with the LT2 strain that lacks O-antigen modification, and with this strain expressing GtrC BTP1 (Fig. 8B). Adsorption with the unmodified strain resulted in a greater decrease in titer than adsorption with the strain expressing modified O-antigen. This is consistent with the conclusion that modification can reduce readsorption of phage. Less adsorption to the LT2 strain with modified O-antigen was also observed with P22 phage, consistent with the lower level of infection observed on strains with GtrC BTP1 -mediated O-antigen modification (Fig. 7). Together, our data support a model in which BTP1 phage-encoded GtrC modifies the BTP1 O-antigen (co)-receptor to reduce superinfection by heteroimmune phages that use the same O-antigen moiety as BTP1 as co-receptor, and to limit readsorption by kin.

Discussion
The BTP1 prophage is one of the distinguishing genetic features of the iNTS S. Typhimurium strain D23580 (Kingsley et al., 2009). Previously, a Salmonella-wide genome analysis for the occurrence of O-antigen modifying gtr operons determined that this prophage encoded an uncharacterized gtr operon (Davies et al., 2013). Here we analyzed the operon and its significance for this strain. This gtr operon has one full-length protein: the GtrC BTP1 protein.
We determined that GtrC BTP1 has hallmarks consistent with acetyltransferase activity. Expression of GtrC BTP1 occurs under a wide range of growth conditions that are relevant to infection, and, unlike many other gtr operons, expression of gtr BTP1 is not subject to phase variation. We show that the GtrC BTP1 -mediated modification plays an important role in the bacterium-phage interaction by protecting the bacterium from infection by both BTP1 and the related P22 phage.
Previously characterized gtrABC operons mediate the glucosylation of the O-antigen, but have different receiving moieties and linkages (Allison and Verma, 2000). In contrast, our data suggest that the gtr BTP1 locus mediates acetylation of the O-antigen, and we propose that GtrC BTP1 is the protein responsible for the acetylation of the rhamnose residue in the O-2 and O-3 positions that was recently identified in biochemical analysis of the O-antigen of strain D23580 (Micoli et al., 2014). First, GtrC BTP1 contains an acetyltransferase domain from the acyl_transf_3 superfamily within its first 360 amino acids. Second, we showed that a RXXR protein motif that is essential for the Oac acetyltransferase activity in S. flexneri (Thanweer and Verma, 2012) is present in GtrC BTP1 and also is essential for its activity (Fig. 3A and B). GtrA and GtrB have roles in transferring a glucose moiety across the inner membrane; the absence of gtrB and truncation of gtrA, and the fact they are not required for this modification (Fig. 2), provides further evidence that the activity of GtrC BTP1 does not involve glucosylation.
All the data are consistent with the conclusion that the family 2 GtrC BTP1 mediates acetylation of the rhamnose in the O-antigen. Based on high amino acid identity among the GtrC proteins of this family (Davies et al., 2013) and the ability of the S. Enteritidis family 2 gtr operon to complement the Δgtr BTP1 short O-antigen phenotype, it is highly likely that all family 2 GtrC proteins have acetyltransferase activity. In support of this conclusion, analysis of the O-antigen of defined strains of S. Typhi confirmed that acetylation of the rhamnose requires the S. Typhi family 2 gtr operon (manuscript in preparation). This S. Typhi GtrC has 77% identity with GtrC BTP1 and 99% identity with the S. Enteritidis family 2 GtrC protein, which complements the short O-antigen phenotype of our Δgtr BTP1 mutant. Thus there is strong support for the conclusion that all family 2 GtrC proteins carry out acetylation of the rhamnose in the O-antigen of these strains.
GtrC BTP1 and other O-antigen acetyltransferases, such as Oac in S. flexneri and OafA in S. Typhimurium (Slauch et al., 1996;Thanweer et al., 2008;Thanweer and Verma, 2012), share some predicted structural features. As noted earlier, all three acetyltransferase proteins contain the RXXR motif, which is necessary for the O-antigen modification by Oac and GtrC BTP1 . A cytoplasmic location of this motif has been experimentally confirmed for Oac (Thanweer et al., 2008) and GtrC BTP1 (Fig. 4), and is predicted for OafA (Thanweer and Verma, 2012). The three acetyltransferases are also membrane proteins with multiple membrane spanning regions, ranging from 9 to 11 transmembrane domains. One of the significant differences between GtrC BTP1 , OafA and Oac is the length and location of the C-terminal soluble domain. GtrC BTP1 and OafA contain a large (270 aa for GtrC BTP1 , predicted 255 aa for OafA) periplasmic tail that is necessary for activity (Fig. 4C and Hauser et al., 2011). In contrast, Oac has a relatively short (22 aa), cytoplasmically located C-terminal end. The role and relevance of the location of the C-termini remain to be elucidated, but it suggests that Salmonella GtrC acetyltransferase enzymes have an additional activity or different mechanisms of activity than Oac from S. flexneri.
For these O-antigen acetyltranferases, it is unclear from where the acetyl moiety is derived, and where or when the moiety is transferred to the growing O-antigen chain. Given the requirement for both a cytoplasmic and a periplasmic domain, and the presence of multiple transmembrane domains, one possibility is that family 2 GtrC proteins function as a channel for the acetyl moiety in addition to carrying out the enzymatic transfer reaction. In this model, the RXXR cytoplasmic loop might be involved in binding the acetyl group for transfer to the periplasm. Alternatively, modification of the O-antigen subunit could occur in the cytoplasm as has been proposed for OafA (Slauch et al., 1996). The C-terminal domain may mediate necessary protein-protein interactions, potentially with the O-antigen assembly complex. Further work on the structure/function relationship of Gtr proteins is needed to elucidate these and many other aspects of gtr-mediated O-antigen modification.
Acetylation of the O-antigen, and specifically of the rhamnose moiety, has been previously described in context of lysogenic conversion by phages A3 and A4 (Wollin et al., 1987). This modification inhibited P22 adsorption and it was inferred that TSP-mediated O-antigen cleavage was inhibited (Wollin et al., 1987). Similarly, our data indicate that P22 and BTP1 phagemediated infection is inhibited by GtrC BTP1 activity, likely because the acetyl modification blocks the TSP endorhamnosidase activity. This was evident from the short O-antigen phenotype in the absence of GtrC BTP1 , and the dependency on the TSP for this phenotype. Effective protection against phage infection may also be facilitated by a high level of acetylation, with an estimated 80% of O-antigen subunits containing the modification in D23580 (Micoli et al., 2014). There are two contributing factors to this high degree of modification. First, GtrC BTP1 expression was not subject to phase variation, and thus all cells in a population will have modified O-antigen. This is in contrast to gtr P22 -mediated glucosylation, for which the mixed population generated by phase variation was directly shown to affect the dynamics of interactions of a population of cells with phage (Kim and Ryu, 2012). Second, the fact that gtrA*C appears to be expressed in a wide range of growth conditions suggests that the O-antigen will usually be in the modified state (Fig. 5B). Thus, a benefit of the BTP1 prophage for the bacterium is that it confers effective protection on the entire bacterial population against infection by BTP1 and both homo-and heteroimmune phages that use the same O-antigen moiety as co-receptor.
There are several aspects of the BTP1 phage that could potentially contribute to the virulence of iNTS BTP1 lysogens. One is the high titer of BTP1 prophage (an average of 7 × 10 7 pfu ml −1 ) we found in overnight cultures, even in the absence of phage-inducing conditions. Phageassociated lysis could produce bacterial debris, and this, in turn, could affect the development of the immune response to the iNTS infection. Additionally, the BTP1encoded family 2 gtr BTP1 operon is not under the control of phase variation, whereas other family 2 operons are phase variable, including those from S. Enteritidis and S. Typhi and the majority of gtr operons (Broadbent et al., 2010a). Thus, D23580 does not have the option for phase-variable immune evasion from antibodies developed to the acetylated O-antigen, which could be significant. However, whether acetyl modification affects the course of infection has yet to be determined. The family 2 gtr operons are associated with many invasive serovars of S. enterica, and it will be important to determine whether, as we propose, O-antigen acetylation contributes to the typical course of infection, including systemic disease, as reported for the iNTS isolates.

Molecular biology techniques
Standard molecular biology techniques were used (Sambrook et al., 1987). DNA was obtained using either genomic DNA isolated as described (Ausubel et al., 1989) or Qiagen miniprep kit for plasmids. PCRs were performed using either Merck's KOD Hot Start (Billerica, MA, USA) or Promega's GoTaq Flexi polymerases (Madison, WI, USA). Restriction enzymes and T4 ligase for cloning were purchased from NEB (Ipswich, MA, USA). All vectors and plasmids generated for this study are listed in Table 2. The site-directed mutagenesis was performed on constructs in the pLAC22 vector (Warren et al., 2000) with primers oMV954 (GAATTTTACAAGAGAGCAATACTAGCAATATTC CCAGCT) and oMV955 (AGCTGGGAATATTGCTAGTATTG CTCTCTTGTAAAATTC) using Agilent's QuikChange Lightning kit (Santa Clara, CA, USA). Bold regions indicate codons changed to introduce the mutations.

Generation of mutant strains
Allelic replacement was used to introduce mutations to the chromosome (Datsenko and Wanner, 2000). Antibiotic resistance cassettes were obtained from Tn10 for Tc-resistance (TcR) and pKD13 for KmR. To generate multiple mutations in BTP1 phage and iNTS Salmonella O-antigen 271 a single strain, these resistance cassettes were occasionally removed (as indicated in Table 1) before the next mutation was introduced. Since the pKD13 KmR cassette is flanked by flippase recognition target (FRT) sites, the introduction of pCP20 (Cherepanov and Wackernagel, 1995) or its pCP20-TcR derivative allowed for the expression of Flp recombinase to produce unmarked mutant strains. When the Tn10 TcR cassette was used, lambda-red recombination was used to introduce an oligonucleotide homologous to the region surrounding the TcR cassette. This combined with counter- selection (Bochner et al., 1980) produced scarless mutations of the targeted genes. For example, to obtain strain Δgtr BTP1 (sMV386), which contains a deletion of the BTP1 prophageencoded gtrA* and the gtrC genes in D23580, the sequence from 101 nt upstream of the start codon of the pseudogene STMMW_03921 (gtrA*) to 15 nt downstream of the stop codon of STMMW_03911 (gtrC BTP1 ) was first replaced with a tetracycline cassette (amplified using oMV754/oMV755) (Table S1). This sequence was, in turn, replaced with oMV758 using the counter-selection method described. These genes are located on the phage genome between the phage TSP (TSP gene, STMMW_03901) and the phagehost genome junction.

LPS extraction and visualization
Crude LPS extracts were prepared as described (Davies et al., 2013). After running the samples on TSDS-PAGE (Tricine-SDS-PAGE) gels, two methods were used for visualization: either silver stain as described previously (Davies et al., 2013) or Western blot. For Western blots, the samples were transferred to polyvinylidene fluoride (PVDF) and blocked with phosphate buffered saline with 0.1% Tween-20 (PBS-T) and 5% milk. The O4 serum from Statens Serum Institute (Denmark) was used as primary antibody followed by goat anti-rabbit IgG-HRP from Sigma (A0545). Millipore's Luminato Western HRP substrate was used for detection.

Membrane preparation and FLAG visualization
Vectors encoding for FLAG-tagged versions of the GtrC BTP1 protein were expressed in Δgtr BTP1 . Mid-exponential phase cultures were induced with 100 μM ITPG and expression allowed to occur for 30 minutes. Cells were then pelleted. Whole cell lysate (WCL) samples were immediately boiled in SDS buffer (0.1 m Tris-HCl, pH 6.8, 4% β-mercaptoethanol, 4% SDS, 20% glycerol). For crude membrane preparations, the samples were resuspended in 1 ml PBS and subjected to a freeze/thaw cycle (−80°C/RT) three times. Unbroken cells were pelleted by centrifugation for 2 min (16,060× g) and the supernatant transferred to a new Eppendorf. Membranes were pelleted by centrifugation at 4°C for 1 hour (16,060× g). The pellet was then resuspended in SDS buffer and boiled for analysis by Western. For protein detection, SDS-PAGE gels were used and then transferred to PVDF. Blots were blocked with 5% milk PBS-T and the primary antibody was Sigma's Monoclonal ANTI-FLAG M2 antibody produced in mouse (F3615). Secondary antibody was goat anti-mouse IgG-HRP (Sigma; A0168).
Preparation of phage-containing supernatants BTP1 phage was prepared by pelleting overnight cultures of D23580, removing the supernatant and treating with a few drops of chloroform to kill any remaining bacteria. This produced phage titers with an average of 7 × 10 7 pfu ml −1 . To generate P22 supernatant for the experiments, a phage lysate (50 μl at 10 4 pfu ml −1 of wild-type P22) was added to a 10 ml culture of exponential phase basal LT2 and growth allowed to continue overnight. Bacteria were again pelleted, the supernatant removed and treated with chloroform. This produced a supernatant containing 3.5 × 10 10 pfu ml −1 .

Top agar phage assays
To create the lawn of bacteria, 100 μl of an overnight culture was mixed with 3.5 ml of molten 0.7% agar, poured onto the top of an LB agar plate and allowed to dry. Five microliters of phage-containing supernatants or dilutions thereof were spotted onto the bacterial lawns. The agar was touched with the pipette tip to indicate where drops were placed. The drops were allowed to dry before incubation overnight at 37°C.

Verifying O-antigen cleavage by BTP1 phage
To investigate O-antigen cleavage, phage-susceptible strains were grown to overnight, pelleted and then resuspended in the phage-containing supernatants containing chloramphenicol (100 μg ml −1 ) to prevent further bacterial growth and lysis from the phage. Samples were incubated with the phage for 1 hour at 37°C before bacteria were pelleted and LPS prepared as described.
Adsorption assay BTP1 and P22 were diluted to a titer of 3.5 × 10 7 pfu ml −1 . Bacterial cells (approximately 3 × 10 8 cfu ml −1 ) were resuspended in the diluted phage (multiplicity of infection of 1:10) and incubated at room temperature for 5 minutes. Bacteria were then pelleted and the supernatant was removed. Dilutions of the supernatant were used in the spot titer assay on a lawn of basal LT2 to determine number of phage remaining in the supernatant.