Substrate recognition by a carbohydrate-binding module in the prototypical ABC transporter for lipopolysaccharide O-antigen from Escherichia coli O9a

Escherichia coli serotype O9a provides a model for export of lipopolysaccharide (LPS) O-antigen polysaccharide (O-PS) via ABC transporters. In O9a biosynthesis, a chain-terminator enzyme, WbdD, caps the nonreducing end of the glycan with a methylphosphate moiety and thereby establishes chain-length distribution. A carbohydrate-binding module (CBM) in the ABC transporter recognizes terminated glycans, ensuring that only mature O-PS is exported and incorporated into LPS. Here, we addressed two questions arising from this model. Are both residues in the binary terminator necessary for termination and export? And is a terminal methylphosphate moiety sufficient for export of heterologous glycans? To answer the first question, we uncoupled WbdD kinase and methyltransferase activities. WbdD mutants revealed that although the kinase activity is solely responsible for chain-length regulation, both activities are essential for CBM recognition and export. Consistent with this observation, a saturation transfer difference NMR experiment revealed a direct interaction between the CBM and the terminal methyl group. To determine whether methylphosphate is the sole determinant of substrate recognition by the CBM, we exploited Klebsiella pneumoniae O7, whose O-PS repeat-unit structure differs from O9a, but, as shown here, offers the second confirmed example of a terminal methylphosphate serving in substrate recognition. In vitro and in vivo experiments indicated that each CBM can bind the O-PS only with the native repeat unit, revealing that methylphosphate is essential but not sufficient for substrate recognition and export. Our findings provide important new insight into the structural determinants in a prototypical quality control system for glycan assembly and export.

Lipopolysaccharide (LPS) 5 is the major glycolipid of the Gram-negative bacterial outer membrane (1). It is almost always essential for cellular viability and is important for pathogenicity. The tripartite structure of LPS consists of the outer membrane-embedded component, lipid A (also known as endotoxin), a core oligosaccharide, and a structurally variable and immunogenic sugar polymer termed O-antigen (O-PS) (2). In human pathogens, O-PS is one of the first molecules encountered by the host, and it provides protection to the bacterium from immune processes such as the complement system (3). For resistance to serum killing, the length of the O-PS is a critical factor in promoting a successful defense, with longer O-PS providing a more effective barrier (4,5). Loss of O-PS chainlength regulation has resulted in loss of serum resistance and susceptibility to leukocytes in some Gram-negative bacteria (5). There are many different O-PS structures with the diversity driven by selective environmental factors, such as host immune response and bacteriophages that exploit O-PSs as receptors (6).
Many O-PSs are assembled by a mechanism that requires an ATP-binding cassette (ABC) transporter to export lipid-linked O-PS intermediates from the site of synthesis in the cytosol to the periplasm (7,8), where the LPS molecule is completed before being translocated to the cell surface by the Lpt (LPS transport) machinery (9, 10). The ABC transporter provides an important role in determining the chain-length distribution of the O-PS products, with two main mechanisms being described. In the simpler process exhibited by Klebsiella pneumoniae serotype O2a (11), the O-PS is polymerized to completion within the cytosol by a biosynthetic enzyme complex. The chain-length distribution is controlled by the relative activities of a complex of glycosyltransferase (GT) enzymes and the ABC transporter. The O2a ABC transporter does not possess strict O-PS specificity, and it can export polymers with diverse This research was supported by funding from the Natural Sciences and Engineering Research Council of Canada (to C. W.). The authors declare that they have no conflicts of interest with the contents of this article. This article contains Table S1 and Figs. S1-S4. 1 Both authors contributed equally to this work. 2  cro ARTICLE repeat-unit structures, but polymerization and export are obligatorily coupled (11,12). The more intricate Escherichia coli serotype O9a O-PS assembly system incorporates an additional mechanism that imposes a stricter level of control over chain length. This requires the installation of a chain-terminating residue that creates an export signal recognized by a carbohydrate-binding module (CBM) attached to the ABC transporter. In this system, polymerization and export can be temporally uncoupled in vitro (11). The structural requirements for glycan recognition by the CBM provide the focus of this study.
During O-PS biosynthesis in both systems, the 55-carbon polyisoprene carrier lipid, undecaprenol phosphate (Und-P) serves as the acceptor for the shared initiating phosphoglycosyltransferase enzyme, named WecA (2). In serotype O9a, the resulting Und-PP-GlcNAc product is then extended by serotype-specific mannosyltransferases to create the O9a O-PS backbone (Fig. 1). Central to polymerization is WbdA, a dualdomain GT polymerase that catalyzes formation of alternating pairs of ␣1-2and ␣1-3-linkages (13,14). WbdA operates in a distributive (rather that processive) manner, where the growing chain is released from the enzyme at each step (14,15). The length of the O9a O-PS is dictated by WbdD. The C terminus of WbdD contains a membrane-anchoring helix and a region that recruits WbdA into an active membrane-bound heterocomplex ( Fig. 1C) (14,16). The N terminus of WbdD possesses kinase and methyltransferase catalytic sites that add a phosphomethyl moiety to the nonreducing terminus of the O-PS, blocking further extension by WbdA (17)(18)(19). The physical separa-tion of the chain-terminating domain from the membrane by an extended coiled-coil ensures that only O-PS of a minimum length can be capped (20). In addition to operating as a molecular ruler, the coiled-coil structure facilitates multimerization of functional WbdD trimers. The relative abundance of WbdA and WbdD also affects the distribution of O-PS lengths, and this is explained by the variable geometry model that describes the outcome of changes in the stoichiometries of the WbdA-WbdD heterocomplexes (21).
Essential to this chain-length regulation strategy is a quality control mechanism on the ABC transporter protein complex, which ensures that only terminated (and chain-regulated) O-PS is exported for assembly on the cell surface. O-PS ABC transporters are composed of homodimers of the transmembrane domain protein (Wzm) and the nucleotide-binding domain protein (Wzt). In E. coli O9a and related assembly systems, Wzt possesses a C-terminal CBM, which is specific for its cognate O-PS (12,22,23). Binding of O-PS by the CBM is a prerequisite for transport, and removal or mutation of the CBM abrogates export (23). The Aquifex aeolicus Wzm-Wzt ABC transporter is currently the only solved structure for this class of exporter, and it shares similarity with E. coli O9a Wzm-Wzt (24). Although the A. aeolicus structure was solved without the cognate CBM, it has allowed the development of mechanistic models for O-PS transport. Key mechanistic predictions were tested by examining the phenotypes of the corresponding mutations in E. coli O9a Wzm-Wzt. In vitro data shows that the CBM stimulates ATP turnover by Wzm-Wzt in proteoliposomes (24), suggesting a more complex role that is not confined Amphipathic helices The coiled-coil domain acts as a molecular ruler separating WbdA, which is recruited to its interaction site at the C terminus of WbdD (indicated), from the chain-terminating catalytic sites of WbdD.

Substrate recognition by a glycan ABC transporter
to substrate recognition. Notably, constructs where the O9a CBM and nucleotide-binding domain are artificially separated retain export function in vivo (23). Similar chain termination/export strategies exist in other species and/or serotypes. For example, O-PS chain regulation is achieved by the addition of a methyl group in E. coli O8 or a sugar (3-deoxy-D-manno-oct-2-ulosonic acid; Kdo) in K. pneumoniae O12 (25). The involvement of both phosphate and methyl residues in chain-length regulation in E. coli O9a appears unnecessarily complicated; in principle, a single terminating residue should suffice. Given the pivotal role the E. coli O9a O-PS system plays as an influential prototype for bacterial glycan export, it is essential to understand the details and limitations of substrate recognition and export. This provides the goal of the current study.

WbdD kinase activity is solely responsible for arresting O9a O-PS polymerization
The modal distribution of O9a-PS chain lengths is established by the coiled-coil domain, which physically separates the WbdD termination enzymes from membrane-anchored WbdA (20). Previous work established that WbdD halts WbdA-mediated O-PS extension by installation of a phosphomethyl moiety to the nonreducing terminus, where new backbone residues are added (17,20,21). Logically, the addition of a single terminating residue to the site of O-PS extension should be sufficient for regulation of chain-length distribution. To confirm this hypothesis and rule out any unanticipated requirements for a functional methyltransferase, we exploited the observation that overexpression of a functional WbdD terminator decreases the average chain length of O-PS (21). Mathematical modeling supports the hypothesis that O-PS chain length responds to altered stoichiometry within the WbdA:WbdD heterocomplex, with the resulting changes in geometry of the complex altering access of the glycan to termination catalytic sites (21). The O-PS phenotype is readily assessed by SDS-PAGE profiles of Und-PP-linked O9a intermediates in whole-cell lysates. These intermediates are not visible in silver-stained SDS-PAGE but are readily visualized in immunoblots using O-PS-specific antibodies (11,12,26).
These experiments were performed in a manA ⌬wzm-wzt:: aphA-3 background, where O-PS production is conditionally dependent on the addition of D-mannose to the growth medium (17). This is necessary to avoid any unexpected effects on viability of the transformants, known to occur when O9a export is eliminated (22). After SDS-PAGE of whole-cell lysates, O9a intermediates were detected using O9a-specific antibodies (Fig. 2). Prior in vitro experiments verified that WbdD H132A and WbdD D351A lack methyltransferase and kinase activities, respectively (19). Each variant was expressed from a pBAD promoter by titrating the L-arabinose inducer, and the expected increase in protein production was confirmed by exploiting the His 6 tag for Western immunoblotting. As shown previously (21), increasing levels of WT WbdD resulted in an inducer-dependent reduction in the average O-PS intermediate chain-length ( Fig. 2A). The profiles of O9a intermediates in cells overexpressing the methyltransferase mutant (WbdD H132A ) were qualitatively similar to those with WT WbdD (Fig. 2B). In contrast, the kinase mutant (WbdD D351A ) had no effect on the chain-length distribution of O-PS intermediates (Fig. 2C). These results show that the addition of the activity of the kinase is sufficient to determine O-PS chain length in vivo and that the kinase and methyltransferase activities can be uncoupled.

Substrate recognition by a glycan ABC transporter
membrane, an E. coli O9a manA ⌬wbdD mutant was transformed with plasmids expressing the WbdD mutant variants. Export activity was monitored by silver-stained LPS profiles in whole-cell lysates separated by SDS-PAGE (Fig. 3A). Whereas plasmids expressing WT WbdD restored the production of O-PS-containing LPS visible in silver-stained gels, only O-PSdeficient LPS was evident in cells expressing either WbdD H132A or WbdD D351A . However, as expected, small amounts of Und-PP-linked O-PS were still detected by Western immunoblotting in cells expressing WbdD H132A or WbdD D351A . Whereas the WbdD H132A variant produced a banding pattern resembling the O-PS chain-length profile of the WT LPS, the WbdD D351A O-PS was of a higher molecular weight, as anticipated from the lack of chain-length regulation described above.
To confirm export status, immunofluorescence microscopy was performed to determine whether O-PS was present on the cell surface of the transformed mutants. O-PS was probed with a primary antiserum directed against the O9a antigen and a secondary antibody conjugated with FITC (Fig. 3B). Images of the parent O9a strain and the manA ⌬wbdD mutant transformed with the WT wbdD gene displayed robust signal from surface-exposed O9a O-PS. In contrast, manA ⌬wbdD cells transformed with vector or either mutant wbdD gene generated no immunofluorescence, consistent with the data in Fig. 3A. Brightfield images show that cells accumulating intracellular O-PS intermediates possess morphological defects (typically elongated cells), as reported previously (22). Taken together, these results indicated that the ABC transporter could not export either the nonmethylated O-PS or the nonterminated (nonphosphorylated and nonmethylated) O-PS.

The carbohydrate-binding module on Wzt is unable to recognize nonmethylated O9a O-PS
Previous work demonstrated that the C-terminal domain of Wzt functions as a CBM (23). In that study, an in vitro pulldown approach revealed the CBM binds its cognate O-PS, but not O-PS lacking the terminal methylphosphate or O-PS from E. coli O8 possessing a polymannose structure (with a repeatunit structure different from that of O9a) capped with a terminal methyl group. To determine whether the requirement for O-PS methylation for export is dictated by the CBM-binding step, purified LPSs lacking either terminal phosphomethyl or methyl modifications were tested using the same pulldown approach. LPS molecules with chemically distinct nonreducing termini were generated using a strategy exploited previously to generate nonterminated K. pneumoniae O12 LPS (12). This

Substrate recognition by a glycan ABC transporter
method takes advantage of the ability of the K. pneumoniae O2a ABC transporter to export heterologous O-PS intermediates. This transporter lacks a CBM and has no specificity for either the repeat-unit structure or terminal modifications of O-PS substrates (11,27). It can therefore export LPS molecules with varying terminal modifications. LPS species were purified from E. coli O9a ⌬wbdD transformed with plasmids to express the K. pneumoniae O2a transporter and appropriate WbdD variants.
The pulldown experiments follow retention (or not) of LPS by His 6 -tagged CBM using nickel magnetic beads (Fig. 4). In the control experiments with WT O9a LPS, a considerable amount of LPS eluted in fractions containing CBM as reported previously (23). However, when CBM was incubated with either nonmethylated or nonmethylated/nonphosphorylated LPS, no binding was evident, indicating that the methyl group is essential for recognition by the CBM, and the phosphate group alone is insufficient. The inability of the CBM to bind improperly terminated O-PS is entirely consistent with the transport defects described above. These data support the notion that the impaired transport reflects altered substrate recognition but cannot exclude additional defects in later transport stages resulting from a terminally exposed phosphate.

Direct interaction between the nonreducing terminal methylphosphate and the CBM
During the characterization of the O9a CBM, residues in a putative binding pocket were identified that abrogated O-PS binding in vitro or had adverse effects on export in vivo (23). A saturation transfer difference (STD) NMR experiment (28) was used to provide more insight into the interaction and to formally verify the direct role of the methyl group in Wzt O9a -C: O9a binding. Similar experiments have been performed previously to precisely define the interaction sites of antibodies with glycans by "epitope mapping" (e.g. see Refs. 29 -31). The method relies on the application of precise radiofrequency pulses to selectively saturate the NMR signals of the protein; this saturation is then transferred to ligands that transiently bind the protein, attenuating the NMR signals of ligand protons at and near the binding site. To better observe small (Ͼ5%) attenuations, STD spectra are collected as the difference of an "on-resonance" spectrum (i.e. saturation pulses applied on the protein signals) and an "off-resonance" reference spectrum (i.e. saturation pulses applied far away from any NMR signals). STD spectra therefore display only signals corresponding to ligand protons that are at and near the protein-binding site. To address complications arising from saturation in the presence of larger ligands (30), a saturation transfer double difference spectrum (STDD) (32) is obtained by subtracting an STD spectrum obtained from a sample without protein from the proteinligand STD spectra.
Serotype O9a O-PS was purified, and its structure was confirmed by comparison against the reported NMR data (13, 33) (data not shown). For STD NMR experiments, two samples were prepared. The first sample contained an approximately 100-fold excess of O9a polysaccharide mixed with Wzt O9a -C, whereas the second sample contained O9a polysaccharide without protein. Saturation times were tested at 1, 2, 3, and 6 s (Fig. 5A). It was determined that 3 s was sufficient for magnetization transfer, and data were collected. The most intense signal from the obtained STDD spectrum (aside from a differencing artifact arising from the Tris buffer) was a doublet at ␦ H 3.63 ppm with coupling constant J ϭ 11 Hz (Fig. 5B), which precisely corresponds to the chemical shift of the methyl protons on the nonreducing terminal methylphosphate group from O9a O-PS (33). These findings offer clear and direct data for the interaction of the nonreducing terminal methyl group with Wzt O9a -C and are entirely consistent with its requirement for Wzt O9a -Cbinding in LPS pulldown experiments as well as its requirement for display of the O9a polysaccharide on the cell surface.

Identification of a terminal methylphosphate moiety in the K. pneumoniae O7 O-PS
The data above reveal the importance of the methylphosphate group for substrate recognition and export but do not reveal whether the terminal modification is sufficient by itself for these processes. To address this question, we examined the ability of the O9a CBM to bind an O-PS possessing a terminal methylphosphate group but a different repeat-unit structure. This was provided by K. pneumoniae O7, whose detailed structure has not been reported previously. Examination of the K. pneumoniae O7 wb* locus (Fig. S1)

Substrate recognition by a glycan ABC transporter
tent with chain-terminating ABC transporter-dependent O-PS biosynthesis. Wzt O7 possesses a proposed CBM domain sharing 34%/50% identity/similarity (E ϭ 7e Ϫ26 ) with the CBM of Wzt O9a (Fig. 6A). A tertiary structure prediction of Wzt O7 -C using Phyre 2 (34) indicates structural homology between the C-terminal domains (Fig. 6B). Furthermore, residues equivalent to those essential for Wzt O9a  The previously reported structural study of O7 O-PS (32) was performed without the use of NMR spectroscopy and did not exclude the possibility that the anomeric configuration of one of the sugar residues could be reversed. Additionally, the possibility of terminal modification of O-PS chain was not investigated. We used 1 H, 13 C, and 31 P NMR spectroscopy, including two-dimensional 1 H, 1 H correlation spectroscopy (COSY), total correlation spectroscopy (TOCSY), NOE spectroscopy (NOESY), heteronuclear single-quantum coherence (HSQC), heteronuclear multiple bond correlation (HMBC), and HMQC-TOCSY experiments to unequivocally determine the structure of the repeat unit and the nonreducing terminus of the O-PS. NMR spectra of O-PS contained the signals for four major spin systems, which were assigned to a Ribf residue (A) and three Rhap residues (B, C, and D) (Table 1, Fig. 7A, and Figs. S2-S4). The ␤-anomeric configuration of Rib was confirmed by the position of the Rib C-1 signal at ␦ 108.2 (compare published data (36) of ␦ 103.1 and 108.0 for methyl glycosides of ␣and ␤-Ribf, respectively). The ␣-configuration of all three Rha residues were inferred from the H-5 and C-5 chemical shifts at ␦ H 3.76 -3.90 and ␦ C 70.4 -70.7 (compare published data (37) of ␦ H 3.86, ␦ C 69.1 for ␣-Rhap, ␦ H 3.39, ␦ C 72.8 for ␤-Rhap). These data, together with the sequence of the sugar residues determined by HMBC and NOESY (Fig. S2) experiments, confirmed the structure of O7 O-PS reported earlier (Fig. 7A).
In addition to the signals for sugar residues, the NMR spectra contained a characteristic sharp signal at ␦ H /␦ C 3.63/54.3 with J H,P ϭ 11 Hz, which was previously assigned to a methylphosphate group at the nonreducing terminus of E. coli O9a O-PS (33). The presence of a methylphosphate group was confirmed by correlation between methyl protons and phosphate at ␦ H /␦ P 3.64/1.75 observed in the 1 H, 31 P HMBC spectrum (Fig. 7D). Furthermore, a 1 H, 31 P HMQC-TOCSY experiment showed correlation between phosphate and H-1 to H-5 of a minor ␣-Rhap spin system denoted as B. The selective one-dimensional TOCSY (Fig. 7B) and NOESY experiments (not shown) confirmed the assignment of B H-1 to H-6 signals. A relatively low-field position of the signal for B C-3, together with strong HMBC correlation between phosphate and B H-3, demonstrate that the methylphosphate group is attached at position O3 of B. The linkage between B and the next sugar residue could not be determined unambiguously, although the HMBC correlation at ␦ H /␦ C 5.09/79.3 indicates that B is linked to another Rha residue. Thus, the NMR spectroscopic analysis confirmed the structure of the O7 repeat unit reported earlier (35) and established the second known example of a methylphosphate O-PS terminal modification.

The methylphosphate group is necessary but not sufficient for CBM binding and transport
To examine whether the nonreducing terminal modification is the sole contributor to Wzt O9a -C O-PS recognition, we performed a pulldown experiment using the E. coli O9a CBM with K. pneumoniae O7 LPS (Fig. 8, A and B). No O7 LPS was detected in elution fractions, indicating that the CBM did not bind O7 LPS. To validate these results, the reciprocal experiments were done using purified Wzt O7 -C. Evident from the silver stain, the O7 CBM was only able to bind its cognate LPS (Fig. 8, C and D). Finally, in vivo mutant complementation experiments were performed in E. coli O9a mutants lacking a functional O-PS ABC transporter (manA ⌬wzm-wzt:: aphA-3) (Fig. 9). Expression of the K. pneumoniae O7 ABC

Substrate recognition by a glycan ABC transporter
transporter in this mutant could not restore translocation of the O-PS across the inner membrane, as evidenced by the absence of banding on silver-stained SDS-polyacrylamide gels, consistent with results from in vitro pulldown experiments. Because the O7 and O9a LPS structures contain the same nonreducing terminal modification, the lack of O-PS binding and export must be due to differences in the mainchain polymer.

Discussion
The E. coli O9a O-PS biosynthesis and export system represents an elegant model for regulating polymer chain length distribution. By combining a molecular ruler with polymerizationtermination and exporter-recognition mechanisms, the cell ensures an appropriate O-PS length distribution to optimize survival within the host. This system has become an important and influential prototype for similar strategies involved in the assembly of a variety of O-PS and other cell-surface glycans from Gram-negative and Gram-positive bacteria. The experimental platform for investigating substrate engagement by the E. coli O9a transporter therefore has broad relevance in microbial glycobiology.
Comparable systems have been identified in E. coli O8 and K. pneumoniae O12 and in S-layer glycoprotein assembly in Geobacillus stearothermophilus NRS2004/3a, Geobacillus tepidamans GS5-97 T , and Aneurinibacillus thermoaerophilus L420-91 T (27). In most examples, a single terminal residue (and presumed termination-export signal) is used. Examples include an O-methyl group on E. coli O8 O-PS and on the S-layer glycans from G. stearothermophilus NRS2004/3a (38) and A. thermophilus L420-91 T (39), as well as a Kdo on K. pneumoniae O12 (25). An exception is the addition of GlcNAc and N-acetylmuramic acid residues onto separate positions of the final rhamnose in G. tepidimans GS5-97 T (40), but only one of these residues is positioned to directly impede chain extension. The overall strategy for the biosynthesis of the terminally methylated D-rhamnose polymer in G. stearothermophilus NRS2004/3a is remarkably similar to that used for the methylated D-mannose E. coli O8 antigen (41). In contrast, E. coli O9a utilizes a binary terminator, consisting of a phosphate and an O-methyl group (18,25). An identical structure terminates the related mannans produced by E. coli O9, K. pneumoniae O3a/O3b/O3c (42), and Hafnia alvei PCM1223 (33). This reflects lateral gene transfer of the locus, followed by diversification of the repeat-unit structure by mutations in wbdA. K. pneumoniae O7 uses the same methylphosphate terminating moiety to cap a different repeat-unit structure.
Despite the similar termination chemistry, the K. pneumoniae O7-biosynthesis gene cluster does not encode a homolog of the E. coli O9a WbdD protein. Instead, a candidate methyltransferase domain is part of a large multidomain predicted protein (Orf6) that also contains two candidate glycosyltransferases (belonging to family GT2) and a coiled-coil region (Fig.  S1). This multidomain format is reminiscent of the K. pneumoniae O12 WbbB protein, where the terminating CMP-Kdodependent glycosyltransferase is separated by a coiled-coil domain from the glycosyltransferase modules that polymerize the repeat-unit glycan (43,44). The kinase involved in the O7 termination reaction has not been identified, but, by analogy to WbdD, we anticipate this domain being located in proximity to the methyltransferase. Orf6 does possess an extensive region (47 kDa) downstream of the methyltransferase that could not be assigned by a homology search, and this may provide kinase activity. Because the identity of the kinase is not central to this  C (residues 303-455). A, pairwise sequence alignment was performed by Blast (A). The indicated secondary structure elements are from the Wzt O9a -C crystal structure (Protein Data Bank entry 2R5O) and a Phyre 2 model of Wzt O7 -C. The residues coloured red are important for O-PS binding by Wzt O9a -C (23). B, cartoon representation overlay of Wzt O7 -C model (orange) with Wzt O9a -C crystal structure (blue). The structural overlay was generated using PyMOL (Schrödinger, LLC).

Substrate recognition by a glycan ABC transporter
study and is not required for data interpretation, it was not pursued here. We demonstrated that the WbdD kinase domain was sufficient for chain length determination. However, the quality control mechanism conferred by the CBM shows an additional dependence on methyltransferase activity. STDD NMR experiments demonstrated that Wzt O9a -C interacts directly with the methyl substituent at the nonreducing terminus, and this likely provides the most prominent hydrophobic interaction involved in O9a polysaccharide recognition by Wzt. The O9a CBM crystal structure has been solved, and the binding pocket was characterized using site-directed mutagenesis in conjugation with transport and CBM binding experiments (23). Several residues were necessary for substrate binding. Aromatic residues (Trp-286 and Tyr-404) presumably contribute via -stacking with the monosaccharide rings, whereas polar residues (Asn-348, Arg-402, and Asp-405) may interact with sugar hydroxyl moieties or the terminal unit. Because these are conserved between Wzt O9a and Wzt O7 , additional side chains distinct to each homologue must provide the noncovalent interactions critical for serotype specificity. However, these are not sufficient for robust productive binding in the absence of any terminal modification, as seen here and in previous experiments using in vitrosynthesized O9a glycan with or without complete terminal modification (23). STDD NMR experiments implicated the terminal methyl group as an interaction group with the CBM. Unfortunately, these experiments could not identify additional hydrophobic interactions arising from the main polymer, due to inefficient magnetization transfer. Furthermore, STD approaches are not suitable for identification of electrostatic or polar interactions. Interactions contributed by the main-chain polymer may be predominantly polar, arising from hydroxyl groups, and these are not detected because of the fast exchange of hydroxyl protons with solvent. It is likely that the phosphate moiety on the O9a glycan provides contact points within the binding pocket (e.g. through electrostatic interaction with the critical Arg-402 residue) (23) and therefore still contributes to recognition of the fully modified glycan. The extensive binding pocket likely accommodates main-chain sugars that contribute to the binding interface.
The O9a and O7 O-PS structures contain identical termination groups on the C-3 position of the nonreducing terminal sugar of their given polysaccharides. Because Wzt O9a -C did not interact with O7 LPS during pulldown experiments, we conclude that one or more monosaccharide units provide functional groups that contribute to direct interactions with the CBM. Cocrystal structures with terminated O-PS analogues are essential to conclusively verify structural features related to O-PS specificity, but, to date, we have been unable to obtain structures of CBM:O-PS complexes for a detailed view of the interactions.
Here, we focus solely on the role of the terminal modification in the substrate recognition process. How this interfaces with substrate engagement by the transporter proper remains to be

Substrate recognition by a glycan ABC transporter
resolved. The ultimate goal of this and related studies is to recapitulate the export process with purified proteins and substrates in proteoliposomes, and, in this context, resolution of the chemical determinants for export is a critical step toward this objective. However, why E. coli O9a has evolved and maintainedamorecomplexstrategy,comparedwiththesimplemethylation process in the related E. coli O8, remains unanswered.

Strains, plasmids, and growth conditions
The bacterial strains and plasmids used in this study are described in Table 2. Cultures were grown in lysogeny broth (LB) at 37°C with aeration and were supplemented with 100 g ml Ϫ1 of ampicillin, D-glucose (0.4% w/v), D-mannose (0.4% w/v), or L-arabinose (0.0002-0.2% (w/v)) where appropriate, unless otherwise specified.

DNA methods
KOD Hot Start DNA polymerase (Novagen) was used for cloning DNA fragments. Oligonucleotide primers were obtained from Sigma and are listed in Table S1. Amplified prod-ucts were purified using a GeneJET purification kit (Thermo Fisher Scientific), and restriction endonuclease digests were performed using enzymes from New England Biolabs. Restriction sites used are indicated in Table 1. Plasmids were confirmed by restriction digests and by DNA sequencing performed by the Advanced Analysis Centre, University of Guelph.

Genomic DNA sequencing
Whole-genome shotgun sequence data for K. pneumoniae 264(1) was generated using the Solexa paired-end sequencing platform (Illumina) and genomic DNA prepared using the Illumina sample preparation kit. De novo assembly of sequencing reads (ϳ100ϫ coverage) was performed with Velvet (45). The

Substrate recognition by a glycan ABC transporter
sequence of the gene cluster for O7 O-PS biosynthesis has been deposited in NCBI (GenBank TM accession no. MN173773).

SDS-PAGE and immunoblotting
To prepare whole-cell lysates for SDS-PAGE and Western immunoblotting, bacterial cells (1 A 600 nm unit equivalent) were collected by centrifugation, solubilized in SDS-PAGE loading buffer, and heated to 100°C for 10 min prior to electrophoresis. For LPS analysis, the whole-cell lysates were treated with proteinase K (Invitrogen) for 1 h at 55°C (46). SDS-PAGE was performed using 12% acrylamide resolving gels in Tris-glycine buffer (47). LPS was visualized by staining with a silver nitrate solution (48), and protein was detected using SimplyBlue Saf-eStain (Invitrogen).
For immunoblot analyses, material resolved by SDS-PAGE was transferred to nitrocellulose membranes (Protran; PerkinElmer Life Sciences). O9a antigen was detected using polyclonal antibodies raised in rabbits (17). Alkaline phosphatase-conjugated goat anti-rabbit was used as a secondary antibody (Cedar Lane Laboratories). Hexahistidinetagged proteins were detected using mouse-anti-His 5 mAb (Qiagen) and alkaline phosphatase-conjugated goat anti-mouse secondary antibody (CedarLane Laboratories). Immunoblots were developed with nitro blue tetrazolium and 5-bromo-4-chloro-3indoyl phosphate (Roche Applied Science).

Immunofluorescence microscopy
Overnight cultures grown in LB, 0.4% D-glucose were subcultured 1:100 into fresh LB, glucose. After 4 h, cells were resuspended in fresh medium containing 0.4% D-mannose and 0.02% L-arabinose to facilitate WbdD synthesis, and growth was continued for an additional 1 h before cells were harvested. Microscopy was performed on formaldehyde-fixed whole cells as described previously (17), using rabbit-anti-O9a primary antibody and FITC-conjugated goat anti-rabbit secondary antibody (Thermo Fisher Scientific).

Wzt-C purification
Wzt O9a -C was purified as described previously (23). Briefly, E. coli Top10 harboring pWQ424 was grown at 37°C with aeration to mid-exponential growth phase and induced with 0.3% L-arabinose for an additional 2 h. Cells were collected by centrifugation, resuspended in buffer A (25 mM Tris, pH 7.5, containing 100 mM NaCl), and lysed by sonication. Cleared lysate was applied to 1-ml columns of nickel affinity resin. Resin was washed with buffer A containing 20 mM imidazole, and protein was eluted using buffer A containing 200 mM imidazole. Wzt O7 -C was purified from E. coli Top10 harboring pWQ1013, using the same growth and induction conditions. Wzt O7 -C was also purified by nickel-affinity chromatography, but buffer B (25 mM Bis-Tris, pH 7.0, 250 mM NaCl) was used.

Substrate recognition by a glycan ABC transporter
nose (0.4%) was added to initiate O-PS expression, and the cultures were grown for an additional 2 h. LPS was purified using the hot water phenol method (49). After dialysis of the extraction mixture against water to remove the phenol, proteins and nucleic acids were precipitated using cold, aqueous TCA and dialyzed until neutral. To separate O7 LPS from capsule, LPS was sedimented by ultracentrifugation for 18 h at 100,000 ϫ g in 20 mM NaOAc, pH 7.0.
O-PS for NMR analysis was isolated by acid hydrolysis of LPS at 100°C in 2% (v/v) acetic acid, followed by centrifugation at 13,000 ϫ g to remove lipid precipitate. The carbohydrate-containing supernatant was fractionated on a Sephadex G-50 superfine column (2.5 ϫ 75 cm) in 50 mM pyridinium acetate buffer (pH 4.5) at a flow rate of 0.6 ml min Ϫ1 . Elution was monitored with a Smartline 2300 refractive index detector (Knauer). The O9a O-PS-containing fraction was acidic (potentially due to contaminating capsule; data not shown), so the polysaccharide mixture was applied to a DEAE Fast Flow (GE Healthcare) column and eluted with a 5-500 mM sodium phosphate gradient at pH 6.3 monitored using absorbance spectroscopy at a wavelength of 208 nm (50). The O9a polysaccharide-containing fractions were pooled and buffer-exchanged by chromatography on a Sephadex G-50 superfine column equilibrated with water.

Binding assays
In vitro binding assays were performed as described previously (12,23). 1-ml reaction volumes of buffer B (25 mM Bis-Tris, pH 7.0, 250 mM NaCl) containing 200 g of LPS and 200 g of Wzt-C were incubated on a rotary shaker for 30 min at room temperature. This mixture was added to 50 l of Pure-Proteome nickel magnetic beads (Millipore), which were preequilibrated with buffer B, and incubated for 30 min at room temperature on a rotary shaker. The beads were collected with a magnet and washed three times with 500 l of buffer B. Protein was eluted stepwise from the beads using three washes with 100 l of buffer B containing 500 mM imidazole.

STDD NMR
NMR studies were performed at the University of Guelph Advanced Analysis Centre. The chemical shifts are referenced to 3-trimethylsilylpropanoate-2,2,3,3-d 4 (␦ H 0, ␦ C Ϫ1.6) added as an internal standard. The Bruker TopSpin 3.2 program was used to acquire and process the NMR data.
STD NMR samples contained 22.2 mg/ml (estimated 2.2 mM) O9a polysaccharide and 0.022 mM Wzt O9a -C (omitted for the "double difference" sample) in 25 mM Tris, pH 7.4, 100 mM NaCl. Deuterium oxide was added to 10% (v/v). NMR spectra were collected on a Bruker AVANCE III 600-MHz spectrometer equipped with a 5-mm TCI cryoprobe. The sample temperature was regulated at 298.2 Ϯ 1 K. STD spectra (28) were acquired using the Bruker pulse sequence stddiffesgp, which suppresses the water resonance via an excitation sculpting scheme (51). Wzt O9a -C was saturated with a train of Gaussianshaped pulses of 25 ms each (calibrated to the correct power level for a 90°Gaussian pulse), applied at 0.67 ppm for the "onresonance" spectrum and at Ϫ11.66 ppm for the "off-resonance" spectrum. Four different irradiation times of 1, 2, 3, and 6 s were employed, and the total prescan delay (i.e. including the irradiation time) was kept constant at 7 s. The acquisition time was 1.7 s. STD data were obtained by subtracting the on-resonance FID from the off-resonance FID using the stdsplit macro provided by the Bruker software program TopSpin.
The frequency-selective pulse employed in the excitation sculpting scheme is not infinitely narrow and can cause undesired attenuation of nearby ligand resonances (e.g. anomeric protons). In the default implementation, the frequency-selective pulse duration is 2 ms, which resulted in the undesired suppression of the anomeric proton resonance at 5.04 ppm to 6% of its nominal intensity. For the experiments presented in this paper, the frequency-selective pulse duration was increased to 5 ms; the reduction in the corresponding pulse bandwidth restored the intensity of the 5.04-ppm anomeric proton to 80% of the nominal value.

Substrate recognition by a glycan ABC transporter
When performed on larger ligands, the STD experiment can result in partial saturation of the ligand. To correct for this, a set of "ligand-only" STD experiments were performed on a second sample not containing Wzt O9a -C (but otherwise identical to the sample described above), and the resulting "ligand-only" STD data were subtracted from the STD data of O9a in the presence of Wzt O9a -C to produce the STDD spectra reported (32).

Elucidation of the K. pneumoniae O7 O-PS structure
O-PS samples were lyophilized twice in 99.9% D 2 O for deuterium exchange and then analyzed in 99.9% D 2 O at 30°C. Data were collected using a Bruker Avance III 600-MHz spectrometer equipped with a 5-mm TCI cryoprobe, with the exception of 31 P-based NMR experiments, performed on a Bruker 400-MHz Avance III spectrometer equipped with a 5-mm broadband Prodigy cryoprobe. Mixing times of 100 and 200 ms were used in TOCSY and NOESY experiments, respectively. The HMBC experiment was optimized for the J H,C coupling constant of 8 Hz and a J H,P coupling constant of 11 Hz. Mixing times of 100 and 80 ms were used for selective TOCSY and 1 H, 31 P HMQC-TOCSY, respectively.