Nonreducing Terminal Modifications Determine the Chain Length of Polymannose O Antigens of Escherichia coli and Couple Chain Termination to Polymer Export via an ATP-binding Cassette Transporter*

The chain length of bacterial lipopolysaccharide O antigens is regulated to give a modal distribution that is critical for pathogenesis. This paper describes the process of chain length determination in the ATP-binding cassette (ABC) transporter-dependent pathway, a pathway that is widespread among Gram-negative bacteria. Escherichia coli O8 and O9/O9a polymannans are synthesized in the cytoplasm, and an ABC transporter exports the nascent polymer across the inner membrane prior to completion of the LPS molecule. The polymannan O antigens have nonreducing terminal methyl groups. The 3-O-methyl group in serotype O8 is transferred from S-adenosylmethionine by the WbdDO8 enzyme, and this modification terminates polymerization. Methyl groups are added to the O9a polymannan in a reaction dependent on preceding phosphorylation. The bifunctional WbdDO9a catalyzes both reactions, but only the kinase activity controls chain length. Chain termination occurs in a mutant lacking the ABC transporter, indicating that it precedes export. An E. coli wbdDO9a mutant accumulated O9a polymannan in the cytoplasm, indicating that WbdD activity coordinates polymannan chain termination with export across the inner membrane.

Lipopolysaccharide (LPS) 1 is a unique and abundant glycolipid found in the outer membranes of Gram-negative bacteria. LPS has three structural domains (1). The hydrophobic lipid A forms the outer leaflet of the outer membrane and is responsible for the endotoxic properties of LPS. A short core oligosaccharide extends from lipid A. In many bacteria, the core is capped with a repeating unit glycan polymer known as the O polysaccharide (O-PS; O antigen). Lipid A is structurally conserved among Gram-negative bacteria, whereas limited variability of the core oligosaccharide is often observed within species. For example, five distinct core structures have been identified in different isolates of Escherichia coli (2). In contrast, O-PS structures vary extensively within a given species because of differences in the number and type of sugars in the repeat unit and the nature of glycosidic linkages within and between repeat units. Variation of the O-PS forms the basis of the O-antigen serotyping scheme. There are ϳ170 O serotypes in E. coli (3). LPS preparations from a given isolate contain a spectrum of molecular species with different sizes. The distribution can range from lipid A core molecules devoid of O-PS to LPS molecules with greater than 100 O-PS repeat units. However, most O-PS-substituted LPS molecules in a preparation fall within a limited size range (i.e. a modal distribution). Lipid A-core and O-PS are synthesized separately at the cytoplasmic face of the inner membrane. The two component parts are subsequently ligated at the periplasmic face of the inner membrane prior to export to the cell surface (reviewed in Ref. 1). Goldman and Hunt (4) first suggested that the modal distribution was established by competition between O-PS polymerization and termination (by ligation). However, subsequent work has implicated specific components of the O-PS biosynthesis systems in regulating the modal distribution of some O-PSs (see below).
O-PSs are assembled on a 55-carbon lipid acceptor, undecaprenol phosphate (und-P), and biosynthesis is initiated by transfer of a sugar-1-phosphate residue from its nucleotide diphosphosugar precursor to und-P. Subsequent extension and processing of the und-PP-linked intermediate proceeds through one of three distinct pathways. Although one of these pathways is currently confined to a single example (5), the other two are widespread among Gram-negative bacteria. The two major pathways are termed Wzy (polymerase)-dependent and ATPbinding cassette (ABC) transporter-dependent biosynthesis, respectively. They differ by the mechanisms involved in polymerization of the repeat units and in the process of translocation of the und-PP-linked polymer or intermediates across the inner membrane.
In the Wzy-dependent mechanism (reviewed in Ref. 1), single O-PS repeat units are assembled on an und-P carrier lipid by sequential glycosyltransferase reactions. A translocase (Wzx) then mobilizes und-PP-linked repeat units to the periplasmic face of the inner membrane, where the polymerase (Wzy) assembles those repeat units into und-PP-linked polysaccharide. Chain extension occurs by transfer of the growing glycan from the und-PP carrier to the nonreducing terminus of another und-PP-linked monomer, effectively extending the chain one repeat unit at a time. The extent of polymerization is controlled in a process that is not yet understood by the chain-length regulator protein, Wzz (formerly Rol or Cld) (6 -8). Different Wzz proteins confer a characteristic modal distribution of O-PS chain length when expressed in a heterologous Wzy-dependent system (9,10). The O-PS chains of wzz mutants are composed predominantly of one or two repeat units, and the amount of each molecule is inversely proportional to its size, i.e. fully elongated LPS species are extremely rare (7,11).
In the ABC transporter-dependent pathway, O-PS chains are  elongated on the und-PP-linked intermediate by processive  glycosyl transfer onto the nonreducing end of nascent polymer  (reviewed in Ref. 1). Polymerization occurs within the cytoplasm, and an ABC transporter is required for export of the nascent polymer to the periplasmic face of the inner membrane, where it is ligated to lipid A-core (12). This system requires neither a Wzy polymerase enzyme nor a Wzz chainlength regulator. However, the resulting O-PSs still exhibit a modal chain-length distribution. Recent structural analysis has identified novel residues at the nonreducing end of some O-PSs synthesized by the ABC transporter-dependent pathway (Ref. 13 and the references therein). Here we describe for the first time the role played by these nonreducing terminal modifications in O-PS chain termination.
The polymannan O-PSs of E. coli O8, O9, and O9a provide models for the ABC transporter-dependent pathway. These polymers share structural features and conserved biosynthesis steps, reflecting genes common to the O-PS biosynthesis genetic loci (Fig. 1). The polymannans have nonreducing terminal O-methyl groups (13,14). Identical O-PS structures and gene clusters are found in Klebsiella pneumoniae O3 (E. coli O9) and K. pneumoniae O5 (E. coli O8) as a result of lateral gene transfer (15). Structural and biosynthetic data for the E. coli and K. pneumoniae analogs are interchangeable. The mechanistic details of assembling the E. coli O8 and O9a O-PSs on the lipid carrier have not been fully elucidated, but a working biosynthetic model has been proposed based on the available structural and genetic evidence (13,16). O-PS synthesis begins with the conserved formation of a "primer" by transfer of a GlcNAc-1-phosphate residue to und-P. This step is catalyzed by the product of the wecA gene (17). WecA activity is not confined to O-PS biosynthesis, and the structural gene is located outside of the O-PS biosynthesis gene cluster (18). Using the primed lipid intermediate as an acceptor, an adaptor disaccharide is assembled by the mannosyltransferases WbdC and WbdB (16). Adaptor synthesis commits the lipid intermediate to the O-PS biosynthetic pathway. The remaining polymer containing the repeating unit is assembled by the processive activity of one or both of the WbdA and WbdB mannosyltransferases. The E. coli O9 and O9a structural variants arise from mutations affecting the WbdA mannosyltransferase, and a single amino acid change is sufficient to alter the structure of the O-PS repeat unit ( Fig. 1) and convert serotype O9 to O9a (19). The primer, adaptor, and repeat-unit domains are clearly identified in structural studies of the LPSs (13). In this report we demonstrate that the WbdD proteins control the chain length of the E. coli O8 and O9a polymannans by modifying the nonreducing end of nascent und-PP-linked polymer. Additionally, we provide evidence that these terminal modifications couple polymerization/termination with the export of nascent O-PS across the inner membrane.
Construction of manA Mutants-Mutants defective in mannose-6phosphate isomerase (manA) were constructed by transferring the manA4 allele from the Hfr donor E. coli CWG102 into CWG291 (O8) and CWG28 (O9a). Hfr-mediated chromosomal transfer was performed by plate mating and manA derivatives (CWG636, O8 and CWG634, O9a) were obtained by selecting for a Tn10 insertion (Tc r ) linked to the manA4 allele (Table I). The ManA Ϫ phenotype was confirmed by the lack of acid production on MacConkey-mannose agar and by showing that O-PS synthesis was dependent on the inclusion of mannose in the growth media.
DNA Methods-Custom-made oligonucleotides were obtained from Sigma-Genosys and used in PCR amplification under conditions optimal for each primer pair. Template DNA was obtained by resuspending bacterial cells from a single colony directly into the PCRs. Pwo DNA polymerase (Roche Applied Science) was used if amplification products were to be cloned, and Platinum Taq (Invitrogen) was used for screening of mutants and ligation products. PCR products and restriction fragments were purified either from agarose gels with the Ultraclean15 DNA purification kit (MOBIO Laboratories) or directly from the PCR with the Qiaquick PCR purification kit (Qiagen). Plasmid DNA was purified with the GeneElute plasmid purification kit (Sigma). Restriction endonuclease digestions, DNA modification, and DNA ligation were all performed using standard methods as recommended by the enzyme manufacturers. DNA sequencing was performed by the Guelph Molecular Supercenter (University of Guelph, Guelph, Canada).
Cloning and Overexpression of the wbdD Gene Products-The genes encoding WbdD O8 and WbdD O9a were cloned on DNA fragments amplified by PCR from the chromosomes of E. coli 2775 and E69, respectively. Restriction sites designed in the primers were used to clone the amplified fragments in pBAD24 (22) to form pWQ53 (O8) and pWQ52 (O9a). The primers for wbdD O8 were 5Ј-CCCGGAATTCACCATGGGTTCGT-CGTTTTATCG-3Ј (EcoRI site underlined) and 5Ј-CAATACGTCGACT-TATTTCTCCGTATTTTTATTTTTTAATTCG-3Ј (SalI). Those for wbd-D O9a were 5Ј-GATCGAATTCACCATGACTAAAGACTTAAACACGCT-GGT-3Ј (EcoRI) and 5Ј-GTTAGGTACCTTTATTTTTCGTTAGTTTGAG-AT-3Ј (KpnI). The WbdD O9a protein encoded by pWQ52 contained a C-terminal FLAG peptide fusion. To construct the FLAG-fusion the primer pairs, 5Ј-CCGGGGTACCGACTACAAGGACGACGACGACA-A-3Ј (KpnI) and 5Ј-AATTTTCTGCAGTTACTTGTCGTCGTCGTCCT-T-3Ј (PstI), were annealed, and extended with the Klenow fragment of DNA polymerase (Invitrogen). The KpnI and PstI restriction sites designed into the primers were used to insert the resulting FLAG-encoding fragment in-frame at the 3Ј end of the wbdD O9a open reading frame.
Construction of Chromosomal Insertion Mutations by Allelic Exchange-The wbdD O9a gene was inactivated by removing a 0.5-kb SmaI fragment from the middle of the open reading frame cloned in pWQ52 and replacing it with a SmaI fragment containing the aacC1 gene from pUC-Gm (23) to form pWQ54. The aacC1 gene cassette is nonpolar when inserted in the same orientation as wbdD O9a . A fragment containing the inactivated wbdD O9a gene was isolated from pWQ54 and cloned into the temperature-sensitive suicide delivery vector pKO3 (24) to generate pWQ55. To transfer the wbdD O9a ::aacC1 mutation into the E. coli O9a chromosome, CWG634(pWQ55) transformants were plated onto LB (glucose and chloramphenicol) agar and incubated at 45°C. Colonies from the initial selection, potentially containing integrants, were pooled and subjected to further selection for plasmid excision on LB (glucose, sucrose, and gentamycin) agar at 30°C.
For construction of a wzm-wzt chromosomal mutation, a fragment containing both wzm-wzt and flanking DNA was amplified by PCR from E. coli CWG28 chromosomal DNA using the primers, 5Ј-TTCCGGTAC-CATGACGAGGCGTCGTTTATCG-3Ј (KpnI site underlined) and 5Ј-CT-CAGGTACCTTGGTAGCTAGTAAAGGACGAC-3Ј (KpnI). The vector pBluescriptSKϩ was modified by excising a HincII-EcoRV fragment from the multiple cloning site. KpnI sites designed into the PCR primers were used to clone the wzm-wzt fragment into the KpnI site of the modified pBluescriptSKϩ derivative to form pWQ56. A HincII-EcoRV fragment spanning the wzm and wzt genes was replaced with a SmaI fragment containing the aphA-3 gene from pYA3265 to form pWQ57. A KpnI fragment containing the disrupted wzm-wzt genes was inserted into the KpnI site of the suicide delivery vector pRE112 to form pWQ58. The use of pRE112 derivatives for chromosomal gene replacement is described elsewhere (25). Derivatives of CWG634 containing the ⌬wzm-wzt::aphA-3 mutation were selected on LB (glucose, sucrose, kanamycin, and streptomycin) agar at 30°C. Chromosomal insertion mutations in strains CWG635 (CWG634 wbdD::aacC1; Gm r ) and CWG638 (CWG634 ⌬wzm-wzt::aphA-3; Km r ) were confirmed by analysis of PCR products amplified using primers flanking the targeted region.
Preparation of Cell Envelope Fractions-Cultures (200 ml) were grown to mid-exponential phase (A 600 nm ϭ 0.6). The cells were collected by centrifugation at 5,000 ϫ g for 10 min, washed once in 10 mM HEPES, pH 7.5, and resuspended in 25 ml of assay buffer (50 mM HEPES, 20 mM MgCl 2 , 2 mM dithiothreitol, pH 7.5). The cells were lysed by ultrasonication with intermittent cooling on ice. The cell lysate was cleared by centrifugation at 20,000 ϫ g for 15 min. The membranes were separated from the cleared lysate by centrifugation at 100,000 ϫ g for 1 h, resuspended in 0.75 ml of assay buffer, and stored at Ϫ75°C. . The reactions were incubated at 30°C for 1 h and stopped by the addition of 25 l of SDS-PAGE sample buffer (312 mM Tris-HCl, pH 6.8, 10% (w/v) SDS, 0.0125% (w/v) bromphenol blue, 50% (v/v) glycerol, 1.8 M ␤-mercaptoethanol). The samples were heated in a boiling water bath for 5 min, and 20-l aliquots were separated by SDS-PAGE. SDS-PAGE gels containing products labeled with [ 14 C]mannose and 32 P were visualized using a PhosphorImager (Bio-Rad Molecular Imager FX). Products labeled with [ 3 H]methyl residues were visualized by fluorography on Kodak BioMax MR 1 film and using Amplify (Amersham Biosciences) as the fluorographic reagent.

In Vitro Incorporation of Radiolabeled Substrates into Membrane
Analytical Methods-The protein concentrations of membrane preparations were determined with the Bio-Rad D C protein assay kit using bovine serum albumin as the standard.
Whole LPS was prepared for SDS-PAGE by proteinase K treatment of whole cell lysates (26). Prior to loading, the samples were heated at 100°C for 5 min. SDS-PAGE was performed using 12% acrylamide gels (3.3%C) in Tris-glycine buffer (27). LPS was visualized by silver staining (28).
Preparation of Anti-O9a Antibody-New Zealand White rabbits were immunized with formalin-killed whole cells of E. coli CWG28. To prepare O9a-specific antibodies, immune serum was adsorbed with whole cells of both E. coli CWG291 and F470.
Immunofluorescence Microscopy-Immunofluorescence microscopy was used to visualize the O9a antigen in fixed cells with and without a permeabilizing step. Bacteria were grown in M9-glucose (0.2% w/v) broth at 37°C to an A 600 nm of 0.3. Mannose (0.4% w/v) was added to the cultures, and incubation was continued for 2 h to allow accumulation of O-PS. Cells (1 A 600 nm ) were collected by centrifugation at 4500 ϫ g, resuspended in 1 ml of formaldehyde solution (5% (v/v) in PBS), and incubated at 4°C for 16 h.  (13). The repeating unit is enclosed by square brackets. The exact terminal structure is known for K. pneumoniae O5 (E. coli O8) but not for K. pneumoniae O3 (E. coli O9/O9a). The O9a polymannan is a variant of O9 with one less mannose residue in the repeat unit and arises from a single amino acid change in the WbdA protein (19). Man, D-mannose; Me, methyl residue. B, organization of the gene clusters encoding enzymes required for polymannan biosynthesis. Identical clusters are found in E. coli and K. pneumoniae. Genes encoding proteins with identical functions in both serotypes are indicated only once for both gene clusters. The genes for mannosyltransferases are indicated in gray. The wbdD and wbdA genes differ considerably between the serotypes. suspension (0.01 ml) was applied to the well of a glass slide coated with poly-L-lysine and incubated at room temperature for 10 min. Bacterial cells were made permeable by a modification of a method described elsewhere (29). The fixed cell sample was incubated at room temperature for 15 min in 10 l of lysozyme solution (0.5 mg/ml in 25 mM Tris-HCl, 10 mM EDTA, pH 8.0) and then at room temperature for 15 min in Triton X-100 (0.1% v/v in PBS). Antibody labeling of the O9a O-PS was performed by treating the slides at room temperature for 15 min with bovine serum albumin solution (1% w/v in PBS), washing extensively in PBS, and incubating with anti-O9a antiserum (1:100 in bovine serum albumin solution) at 37°C for 30 min. The slides were washed extensively in PBS and treated with rhodamine red-conjugated goat anti-rabbit antibody (1:50 in bovine serum albumin solution; Jackson Immunoresearch). The labeled slides were washed in PBS and mounted in Vectashield (Vector Laboratories). Bacteria were viewed on a Zeiss Axiovert 200 microscope using a 100ϫ objective lens, and the images were processed using Openlab software (Improvision).

Conserved Motifs in the WbdD Proteins Implicate Them in the Addition of Nonreducing Terminal Residues to the O8 and O9a
Polymannans-The predicted WbdD proteins from E. coli serotypes O8 and O9a differ in size at 48,630 Da (O8) and 81,731 Da (O9a), respectively. Although these proteins share only limited overall similarity, they do contain the same conserved motifs (Fig. 2). WbdD O8 and WbdD O9a share a region of sequence near their C termini that exhibits 34% identity and 53% similarity over 113 amino acids (Fig. 2). These domains are predicted to have a high probability (Ͼ98%) of forming coiled-coil structures as determined by the COILS algorithm (30). Each exhibits the characteristic heptad repeat motif (31) with hydrophobic residues at positions a and d and charged or polar residues at e and g. Residues a and d provide the interface between interacting ␣-helices, and different modes of interaction are possible. An additional putative coiled-coil domain (probability Ͼ95%) is found in WbdD O9a only (Fig. 2).
The N-terminal regions of the WbdD proteins share significant similarity with AdoMet-dependent methyltransferases. The WbdD proteins share the most similarity with a consensus motif from the UbiE family of methyltransferases that are involved in ubiquinone/menaquinone biosynthesis (Fig. 2). Three conserved amino acid sequence motifs (I, II, and III) have been identified in most methyltransferases, and they are implicated in AdoMet binding and methyl transfer (32). Motifs I, II, and III were identified in the deduced amino acid sequence of WbdD O8 . The WbdD O9a protein contained motifs I and II, but motif III was not evident (Fig. 2).
The WbdD O9a protein contains an additional region sharing similarity with the catalytic domain consensus motif from ser- ine/threonine protein kinases (Ref. 33 and Fig. 2). The invariant aspartate (Asp 351 ) and asparagine (Asn 356 ) residues of the kinase catalytic loop (motif VIb) and the highly conserved DFG (Asp 369 -Phe 370 -Gly 371 ) sequence (motif VII) were identified in WbdD O9a (Fig. 2). The absence of a comparable kinase domain in WbdD O8 is reflected in its smaller size compared with WbdD O9a .
Previous structural analyses of the E. coli O8 PS and the structurally identical Klebsiella O5 PS have shown the presence of O-methyl substitutions at the nonreducing ends of these polymers (13,14). The similarity shared by WbdD O8 and known methyltransferases is consistent with the hypothesis that WbdD modifies the nonreducing end of the growing und-PPlinked O-PS. In the K. pneumoniae O5 (E. coli O8) polymer, the precise linkage of the 3-O-methyl group was identified by twodimensional NMR experiments (13). However, the terminal linkage of the identified methyl group in the K. pneumoniae O3 (E. coli O9) was not conclusively established. Given that WbdD O9a contains putative methyltransferase and kinase domains, the terminal modification may be more complex, potentially involving both phosphorylation and methylation.

Overexpression of WbdD Decreases O-PS Chain Length-
The SDS-PAGE profiles of LPSs from E. coli CWG636 (serotype O8) and CWG634 (O9a) exhibit clusters of high molecular weight O-substituted LPS molecules, reflecting modal O chain-length distributions (Fig. 3A, lanes 1 and 3). In initial experiments designed to address the role of the wbdD gene products in establishing modality, the relevant genes were cloned behind the arabinose-inducible promoter of pBAD24 and overexpressed in their hosts. After induction of wbdD O9a expression, the SDS-PAGE profile of serotype O9a LPS from E. coli CWG634 (pWQ52) changed, such that most LPS molecules contained short chain O-PS (Fig. 3A, lane 4). The amount of a given LPS species (reflected by band intensity on SDS-PAGE) diminished as the number of repeating units increased. In the corresponding experiments performed in the E. coli O8 background, induction of wbdD O8 expression in E. coli CWG636 (pWQ53) also resulted in shorter O-PS chains (Fig. 3A, lane 2). These data support the notion that WbdD may function as an O-PS chain-length regulator. Interestingly, O-PS chain length was not reduced by overexpressing the heterologous plasmidencoded WbdD proteins (data not shown). The WbdD proteins are therefore specific for a given serotype.
The altered LPS SDS-PAGE profile resulting from WbdD overexpression could reflect either a direct effect on O-PS chain length or an alteration favoring preferential selection of short O-PS chains for ligation onto lipid A-core. To resolve these possibilities, E. coli CWG28 (O9a) membrane fractions were used to determine the effect of wbdD O9a overexpression on the incorporation of [ 14 C]mannose from GDP-[ 14 C]mannose into polymannan in vitro. The membranes from the cells overexpressing WbdD O9a showed a significant (2.3-fold) reduction in [ 14 C]mannose incorporation compared with membranes isolated from noninduced culture (Fig. 3B). und-PP-linked O-PS To analyze the effect that overexpressing the WbdD proteins had on O-PS synthesis, strains containing either pWQ52 or pWQ53 were grown for 16 h in 5 ml of LB containing 0.4% (w/v) glucose to repress WbdD expression. The cultures were then diluted 1:100 into media containing 0.1% (w/v) L-arabinose (or no arabinose), and the cultures were grown to an A 600 nm ϭ 0.5. A, silverstained SDS-PAGE of E. coli O8 and O9a LPS. The samples were prepared from transformants of E. coli CWG634 (O9a manA) and CWG636 (O8 manA) containing plasmids pWQ52 (wbdD O9a ϩ ) and pWQ53 (wbdD O8 ϩ ), respectively. Cultures were grown in the presence of 0.2% (w/v) D-mannose, providing permissive conditions for polymannan synthesis in the manA background. In the absence of arabinose (lanes 1 and 3), wbdD expression is noninduced, and the profile is identical to wild-type E. coli prototypes for the O8 and  (Fig. 4B). In the absence of added AdoMet, the labeled products were aberrantly large. A substantial reduction in chain length was evident in the presence of 1 M AdoMet, and a small additional reduction occurred when the amount of AdoMet was increased to 10 M. Beyond that, no further reduction in polymannan chain length was detected. The concentration-dependent effect of AdoMet provided further support for the critical role of WbdD and 3-O-methylation in the regulation of chain length in serotype O8 polymannan.
A Phosphorylation Process Determines Chain Length of Nascent E. coli O9a Polymannan-The presence of putative kinase and methyltransferase domains in the WbdD O9a protein suggested a role for both phosphorylation and methylation in the synthesis of the O9a O-PS. To establish whether und-PP-linked E. coli O9a O-PS was methylated, membranes from E. coli CWG634 were incubated with S-Ado-[ 3 H]Met in various combinations with unlabeled ATP and GDP-mannose. Labeling of high molecular weight products occurred only in those reactions containing both GDP-mannose and ATP (Fig. 5A, lane 4) but not in reactions in which GDP-mannose and/or ATP was omitted (Fig. 5A, lanes 1-3), indicating that methylation was dependent on both polymer synthesis and phosphorylation. The products were also examined from corresponding reactions containing GDP-[ 14 C]mannose as the source of label (Fig. 5B). Polymannan chain length was aberrantly large in the absence of AdoMet and ATP (Fig. 5B, lane 1). Chain length was reduced in reactions containing ATP plus AdoMet and ATP alone but not in those supplemented with only AdoMet (Fig. 5B, lanes  2-4). The chain length was reduced in an ATP concentrationdependent manner, with effects becoming apparent at a concentration of 0.02 mM (Fig. 5C). Further reductions were evident as the concentrations of ATP increased to 0.2 and 2 mM. These results explained the inability to establish direct connectivity between a specific mannose residue and the methyl group during NMR structural analyses of the Klebsiella O3 polymannan (13).
The nature of the precursors for WbdD-mediated chain termination (AdoMet and ATP) and the evident effects in in vitro reactions suggested that these reactions occurred in the cytoplasm prior to export of nascent polymannan. To unequivocally verify that this was the case and provide further insight into the order of events, the effect of ATP on chain length was assessed in CWG638(⌬wzm-wzt::aphA-3). The absence of the ABC transporter ensured that the products examined were und-PP-linked glycans and that only those events preceding export would be examined. The ATP concentration-dependent reduction in chain length of in vitro reaction products synthesized by membranes from the export-deficient mutant was indistinguishable from that seen with membranes from the export-proficient parent (Fig. 5, compare C and D). These results indicated that chain-length determination preceded export and that chain length was not simply controlled by the rate of export of und-PP-O-PS through the ABC transporter.
To unequivocally confirm the involvement of WbdD O9a in chain-length determination, a chromosomal wbdD O9a ::aacC1 mutation (E. coli CWG635) was constructed by allelic exchange. As shown in Fig. 6A, membranes from E. coli CWG635 were still able to synthesize O9a polymannan in vitro, although the chain length was increased relative to that made by the parent. However, the polymeric product was not labeled by either S-Ado-[ 3 H]Met or [ 32 P]ATP (Fig. 6, B and C), nor was the chain length responsive to increasing ATP concentration in the reaction (Fig. 6D).
Terminal Modification of the O9a O-PS by WbdD Is Essential for Its Export from the Inner Membrane-Growth of the chromosomal wbdD O9a mutant (E. coli CWG635) was impaired in the presence of mannose, i.e. permissive conditions for O9a polymannan synthesis in vivo (Fig. 7B). Complementation of the mutation with a plasmid carrying wbdD O9a restored normal growth properties (data not shown), indicating that the impairment was due only to the wbdD O9a defect and did not involve an unanticipated second-site mutation or any polar effects of the chromosomal insertion on downstream genes. To overcome the growth defect in E. coli CWG635 and examine O-PS synthesis, E. coli CWG634 and CWG635(wbdD O9a ::aacC1) were grown to early exponential phase in minimal-glucose media, and mannose was then added to activate polymannan synthesis. The LPS profiles were then assessed by SDS-PAGE analysis of whole cell lysates (Fig. 7A). The parent strain, E. coli CWG634, produced typical high molecular weight O9a LPS. In contrast, CWG635(wbdD O9a ::aacC1) did not exhibit lipid A-core-linked O9a O-PS under permissive conditions. When a functional wbdD O9a gene was supplied in trans on plasmid pWQ52, expression of the high molecular weight O9a LPS was restored (Fig. 7A). FIG. 7. Phenotypic analysis of E. coli CWG635 (wbdD O9a ::aacC1). A, examination of the capability of E. coli CWG634, CWG635 (wbdD O9a ::aacC1), and CWG635 transformed with pWQ52 (expressing WbdD O9a ) to synthesize O9a antigen-substituted LPS. Cultures were grown to mid-exponential phase (OD 600 ϭ 0.4) in minimal medium containing 0.2% (w/v) glucose (i.e. conditions that were not permissive for O9a-polymannan synthesis, because of the manA mutations in each strain). At this point, each culture received mannose (final concentration, 0.4%) to allow O9a synthesis to proceed. After 2 h of further incubation, samples were taken, and whole cell lysates were prepared for silverstained SDS-PAGE. B, growth impairment in E. coli CWG635 (wbdD O9a ::aacC1) under conditions permissive for O9a-polymannan synthesis. The growth of E. coli CWG635 in M9-glucose (0.2% w/v) was followed after the addition of either mannose (0.4% (w/v) final concentration) or an equivalent volume of water. The growth curve was performed in triplicate, and the range of variation for individual time points was within Ϯ10%. C, in vitro synthesis of O9a polymannan by membranes prepared from E. coli CWG635 (wbdD O9a ::aacC1) and its parent, CWG634.
The ability of membranes from E. coli CWG635-(wbdD O9a ::aacC1) to synthesize O9a polymannan was clearly established by PAGE analysis of the in vitro products reported in the preceding text. Direct comparative analysis of the mannosyltransferase activity in membranes, assayed in the absence of ATP, indicated that the wbdD O9a membranes showed only slightly less activity than those of the parent (Fig. 7C). The reason for this slight reduction are unclear but could be related to minor polarity issues resulting from the insertion mutation or reduced amounts of mannosyltransferases because of altered stability of the mRNA in the mutant. Regardless, the absence of O-PS-substituted LPS in E. coli CWG635 (wbdD O9a ::aacC1) could only be explained by either an unexpected inability to form polymannan in vivo or an inability to export the polymer for ligation to lipid A-core at the periplasmic face of the membrane. To address these possibilities, the location of polymannan was determined by immunofluorescence microscopy (Fig. 8) using antibody specific for the O9a antigen. The parent, E. coli CWG634, showed strong immunofluorescence (Fig. 8A) at the cell periphery as expected for surface exposed LPS. In contrast, E. coli CWG635 (wbdD O9a ::aacC1) showed no surface immunofluorescence (Fig. 8B), a phenotype shared with E. coli CWG638 (⌬wzm-wzt::aphA-3) (Fig. 8D). E. coli CWG638 also showed evidence of cell elongation and filamentation. To confirm that the ⌬wzm-wzt and wbdD O9a mutants synthesized intracellular O9a polymannan, the cells were permeabilized, and diffuse internal immunofluorescence was detected throughout the cytoplasm (Fig. 8, C and E). Thus the inability of wbdD O9a mutants to make O-PS-substituted LPS is due to a defect in export.

DISCUSSION
The regulation of O-PS chain length is critical for virulence of Gram-negative bacteria. Efficient chain-termination processes are required for the establishment of an appropriate modal chain length distribution and for ensuring that most lipid Acore is capped with O-PS. These characteristics dictate resistance to complement-mediated serum killing and virulence (35). The involvement of Wzz (formerly Rol or Cld) in regulating the chain length of O-PS synthesized by Wzy-dependent pathways is well accepted. Mutations in wzz make Salmonella enterica serovar Typhimurium (11), E. coli O75 (36), and Shigella dysenteriae serotype 2a (37) sensitive to complement and avirulent. Yersinia enterocolitica O:8 mutants with unregulated chain length are sensitive to killing by polymorphonuclear leukocytes and are attenuated for virulence (38). In S. dysenteriae 2a, Wzz creates precise modal O-PS chain length that is critical for the surface exposure of IcsA and therefore influences actin-based motility and intercell invasion (37, 39 -42). Although the exact role of O-PS chain length may vary in different Gram-negative pathogens, it is anticipated that Wzz function is important in other bacterial pathogens whose O-PS is synthesized in a Wzy-dependent pathway. In the Wzy-dependent pathway, the polymerization of O-PS occurs at the periplasmic face of the inner membrane, and products of varying lengths are efficiently ligated to lipid A-core. Thus the influence of Wzz on O-PS chain length is readily apparent in SDS-PAGE profiles of LPS from a culture.
The bacterial pathogens is not expected to be confined to those with Wzy-dependent O-PS biosynthesis, but the differences in the synthetic mechanism and location of polymerization dictate that a fundamentally different process must be involved for chain-length determination in the ABC transporter-dependent pathway. The data reported here give the first insight into the mechanism involved. Addition of a novel nonreducing terminal residue provides a simple process to terminate the action of processive glycosyltransferase activity. However, the process cannot be simply stochastic or random but must incorporate an ability to regulate (i.e. terminate when the chain length reaches the appropriate modal value). In in vitro experiments with the E. coli O8 and O9a polymannans, this can be achieved by titrating the amount of AdoMet or ATP precursors. This may be a factor in vivo. However, overexpression of WbdD in the parental background causes a reduction in chain length, and so cellular concentration (availability) of the terminating enzyme is also important in vivo. Chain termination is linked to ABC transporter-mediated export, and thus the chain-terminating enzyme is not the sole component involved in the process. Given the need for speed and efficiency in a rapidly growing bacterial culture, a coordinated multienzyme complex seems logical. The presence of a conserved C-terminal domain with the predicted propensity to form coiled-coils offers one possible avenue to establish protein-protein interactions in such a complex. Paired coiled-coils provide a prevalent mechanism (in 3-5% of cellular proteins) for intermolecular and intramolecular interactions. They are involved in important cellular events including cytoskeleton formation and motility and as receptors for molecular recognition in eukaryotes (reviewed in Ref. 43). In bacteria, coiled-coil domains are important for interactions within multienzyme complexes such as those involved in type III protein secretion systems (44 -46). In the case of the O8/O9/O9a biosynthesis systems, WbdD proteins are the only dedicated O-PS biosynthesis components with predicted coiled-coil motifs. Therefore, the coiled-coil motifs are either involved in interactions between WbdD proteins to form a dimer or higher order structure or between WbdD and another presently unknown participant.
The data reported here cover two closely related O-serotypes in E. coli (and identical structures in K. pneumoniae). However, novel nonreducing terminal residues are found in other O-PS. For example, terminal 3-deoxy-D-manno-oct-2-ulosonic acid (Kdo) residues with different linkages are present in the O4 and O12 serotypes of Klebsiella spp. (13). Bordetella bronchiseptica O antigen has a unique 2,3,4-triamino-2,3,4-trideoxy-␣galacturonamide derivative at the nonreducing terminus (47). Terminal modifications may be a common feature in the Bordetellae as Bordetella hinzii O antigen is terminated with a 4-O-methylated GalNAc3NAcAN residue (48). In the case of B. bronchiseptica, the O-antigen biosynthesis locus also contains homologs of wzm and wzt (49), but the importance of the terminal residues has not been examined. Rhizobium etli CE3 O antigen is terminated by 2,3-di-or 2,3,4-tri-O-methylfucose (50). The O antigen biosynthesis genes map to two loci with one adjacent to the wzm-wzt homologs (51-53), but the O antigen can apparently be made in the absence of the terminal methylated residue (53). 2-O-Methyl groups are found in Vibrio cholerae O1 (54). Although V. cholerae mutants lacking this residue are not impaired in the ability to synthesize O antigen, they do lose the epitope associated with seroconversion from type Ogawa to Inaba (55).
An obvious question is whether the phenomenon described applies to other cell surface polysaccharides formed by ABC transporter-dependent pathways. Many Gram-negative bacteria have capsules formed by this type of pathway, and the E. coli group 2 capsules provide prototypes (reviewed in Refs. 56 and 57). The capsules are distinguished from O antigens by the absence of terminal lipid A-core molecules. Furthermore, capsules have a pathway for translocation to the cell surface thatisdistinctfromLPS (56,58).CapsulesfromABCtransporterdependent pathways are polymerized by processive glycosyltransferases, but the mechanisms involved in termination and potential chain-length regulation are currently unknown. Structural analyses have focused on the repeat unit structure and a conserved phosphatidyl-Kdo substituent at the reducing terminus (59). There is currently no structural evidence for nonreducing terminal modifications in the capsules. Biosynthesis of B. subtilis cell wall teichoic acids also follows an ABC transporter-dependent pathway (60), but, as is the case with the capsules, structural studies specifically aimed at elucidating the details of the nonreducing terminal structures have not been performed. An ABC transporter-dependent system is also involved in the assembly of S-layer glycoproteins of the Grampositive bacterium Geobacillus stearothermophilus (61). S-layers are regular crystalline proteinaceous arrays found on many bacterial surfaces (62), and some, like those in G. stearothermophilus, are modified with long glycan chains. The polymer in strain N2004/3a is a polyrhamnan of 15 trisaccharide repeat units and is terminated with 2-O-methylrhamnose (63). The structure also shows a reducing terminal domain, equivalent to the adaptor in the E. coli O8 and O9a antigens, that links the glycan to serine and threonine residues in the protein. The gene cluster for biosynthesis of this glycan encodes an ABC transporter and a predicted protein with a methyltransferase domain (61). The parallels between this system and ABC transporter dependent O-PS biosynthesis are striking and suggest that the chain regulation system that is described for the first time here may be more widespread in bacteria.