Preference of Bacterial Rhamnosyltransferases for 6-Deoxysugars Reveals a Strategy To Deplete O-Antigens

Bacteria synthesize hundreds of bacteria-specific or “rare” sugars that are absent in mammalian cells and enriched in 6-deoxy monosaccharides such as l-rhamnose (l-Rha). Across bacteria, l-Rha is incorporated into glycans by rhamnosyltransferases (RTs) that couple nucleotide sugar substrates (donors) to target biomolecules (acceptors). Since l-Rha is required for the biosynthesis of bacterial glycans involved in survival or host infection, RTs represent potential antibiotic or antivirulence targets. However, purified RTs and their unique bacterial sugar substrates have been difficult to obtain. Here, we use synthetic nucleotide rare sugar and glycolipid analogs to examine substrate recognition by three RTs that produce cell envelope components in diverse species, including a known pathogen. We find that bacterial RTs prefer pyrimidine nucleotide-linked 6-deoxysugars, not those containing a C6-hydroxyl, as donors. While glycolipid acceptors must contain a lipid, isoprenoid chain length, and stereochemistry can vary. Based on these observations, we demonstrate that a 6-deoxysugar transition state analog inhibits an RT in vitro and reduces levels of RT-dependent O-antigen polysaccharides in Gram-negative cells. As O-antigens are virulence factors, bacteria-specific sugar transferase inhibition represents a novel strategy to prevent bacterial infections.

B acterial cell envelopes are rich in glycans that provide structural integrity and mediate intercellular interactions required for pathogenicity. 1−8 These glycans are unique, as bacteria produce sugars absent in mammalian glycomes known as "rare" or "bacteria/prokaryote-specific". 5,9−11 Among the ∼700 rare monosaccharides, L-rhamnose (L-Rha) and other 6deoxysugars that lack a C6-hydroxyl are enriched in bacteria compared to mammals. 12 Across microbes, L-Rha is required for the construction of different cell envelope glycoconjugates. 9,12−15 In the Gram-positive, Acid-fast Mycobacterium genus, which includes the pathogen Mycobacterium tuberculosis (Mtb), an α-L-Rha-(1→3)-α-D-GlcNAc linker between the peptidoglycan and arabinogalactan layers is essential for viability ( Figure 1A, lef t). 16−20 In many strains of the Gramnegative Escherichia coli, the same disaccharide is found in Oantigen (O-Ag) polysaccharides in lipopolysaccharide (LPS) ( Figure 1A, right). O-Ags are required for virulence, and their sugar sequences distinguish serotypes. 9,13,21,22 While the cell envelopes of E. coli and Mtb differ, L-Rha is incorporated by a dedicated rhamnosyltransferase (RT) called WbbL in both species.
WbbL transfers a sugar from the dTDP-β-L-Rha donor to an acceptor, lipid pyrophosphate-GlcNAc ( Figure 1B). 23,24 E. coli utilizes the lipid carrier undecaprenyl pyrophosphate (C55-PP, Und-PP), while Mtb uses decaprenyl pyrophosphate (C50-PP). Despite the presence of wbbL in diverse bacterial species, little is known about the mechanisms by which glycolipid RTs select donors among the vast pool of cellular sugar metabolites. 17,25−28 Like many O-Ag glycosyltransferases, WbbL is localized proximal to the cytoplasmic membrane, which complicates purification. 7,29,30 Additionally, discrete donor and acceptor substrates are not readily accessible, as syntheses of activated β-L-sugars 31−38 and glycolipids with long hydrophobic tails 39−43 are limited.
Since WbbL is essential for virulence (E. coli) or survival (Mtb), obtaining a better understanding of substrate preferences would have implications in the design of both antibiotic and antivirulence agents. Resulting antivirulence strategies may avoid selective pressures that drive antibiotic resistance mechanisms. 44 Here, we study three glycolipid RTs from different species, two from Gram-negatives, and one from mycobacteria. A collection of synthetic (deoxy)nucleoside diphosphate-sugar ((d)NDP-sugar) donor and glycolipid acceptor analogs are used to examine molecular recognition of substrates. These findings lead to a tactic to inhibit E. coli WbbL in vitro and in cells.
To evaluate donor recognition, we assessed binding of RT to dTDP-β-L-Rha analogs obtained by our synthetic and chemoenzymatic routes. 31,45 WbbL is predicted to be membrane associated; 18,30,46 hence, E. coli and Mtb WbbL membrane fractions were isolated. To broaden the scope of our study and obtain higher protein quantities, we identified a putative similar protein, RfbF, in the Gram-negative thermophile Thermus thermophilus using sequence-based analyses (Tables S1−S2, Figures S1−S2). 23 Table  S3), which is in the concentration range of cellular nucleotidesugars and the measured K M (∼35 μM) with crude Mtb WbbL. 30 An L-Rha-1-phosphate fragment demonstrated binding, but beyond a measurable K D (>1 mM). 51 Unlike reported RTs that glycosylate glycans, natural products or proteins, 46,52−55 RfbF did not bind nucleotide alone (dTDP) at concentrations tested ( Figures S3−S4). Stereochemical inversion at the C4-position using dTDP-6-deoxy-β-L-talose (dTDP-β-L-6dTal) led to comparable affinity as the native substrate, while altering the C2-and C4-positions with dTDPβ-L-fucose (dTDP-β-L-Fuc) produced an ∼15-fold enhancement in K D , indicating the C2-hydroxyl conformation is important for recognition. The consequence of C6-hydroxylation was assessed with dTDP-β-L-mannose (dTDP-β-L-Man) and the bacterial metabolite dTDP-α-Glc, which showed 60-and ∼75-fold increases in K D , respectively, compared to the native donor ( Figure 2C, left). Similarly, alteration of the anomeric position in dTDP-α-L-Rha resulted in no detectable binding. Hence, changes to the anomeric or C6 position may cause steric clashes in the active site.
We then tested GDP-and UDP-sugars as ligands, since these nucleotides commonly activate cellular sugars. 56−58 GDP-β-L-Rha showed no binding to RfbF, likely due to the bulky purine ( Figure 2C, right). UDP-β-L-Rha bound with an ∼4.5-fold increase in K D relative to dTDP-β-L-Rha; however, the cellular metabolite UDP-α-GlcNAc did not bind at μM concentrations, indicating that a C2′-hydroxyl on ribose is accommodated, but additional changes in the sugar moiety are not. These observations suggested that a pyrimidine nucleotide-β-6-deoxysugar could serve as an RT donor.
In E. coli, WbbL is required to synthesize oligosaccharide units (O-units), which are flipped, polymerized, and ligated to Lipid A's core sugars in a model O-Ag (O16) pathway, as well as others 7,13,65 ( Figure 5A). Since O-Ag are virulence factors, 21 we next aimed to evaluate the cellular consequences of WbbL inhibition. As L-Rha-1-phosphate weakly bound to RfbF, we hypothesized that a reported 6-deoxy-iminosugar (2) 66,67 resembling L-Rha could block the donor site. We synthesized 67 and then titrated 2 into E. coli WbbL reactions containing the native acceptor with the donor present around cellular  concentrations. 56 The resulting half maximal inhibitory concentration (IC 50 ) was ∼3 mM ( Figure 5B). Direct binding of 2 to RfbF was not detected by ITC, indicating a K D > 1 mM ( Figure S20). 2 may instead bind the acceptor complex more tightly, as observed for other iminosugars with added nucleotide. 68 Compared to structurally related L-monosaccharides ( Figure 5B), 2 was a more potent WbbL inhibitor than L-Rha (∼53% versus 38% inhibition, respectively), likely because it acts as a transition state mimic. 69 Lack of inhibition by L-Man reinforced the importance of a C6-deoxy in binding. E. coli WbbL docking experiments suggested that 2 binds like the native sugar in the donor site, while L-Man does not ( Figure  S21).
To evaluate the cellular effect of the iminosugar, an E. coli strain expressing wbbL ( Figure S22) was grown with and without 2. Following LPS extraction, E. coli cells exposed to 2 produced >50% less LPS than untreated samples when an equivalent number of cells was analyzed ( Figures S23−S24), even though 2 was not toxic ( Figure S25). 70−72 To assess if the loss of LPS was due to diminished O-Ag, we analyzed extracted LPS by SDS-PAGE followed by silver staining and blotting for O16. As seen in Figure 5C, O-Ag levels decreased in samples containing 2 (lanes 7−9) versus untreated cells (lanes 4−6), with an ∼40% reduction calculated by densitometry ( Figure  S26). 73 However, there was not complete loss of O-Ag as in E. coli lacking functional wbbL (lane 1) compared to cells with native wbbL (lane 2). Hence, the iminosugar is not an antibacterial, but impairs WbbL-dependent O-Ag synthesis, which represents a potential antivirulence strategy to impair host infection. 3,44,70,74 In conclusion, we found that donor recognition by WbbL/ RfbF is driven by three factors: the absence of a C6-hydroxyl, and the presence of a β-anomeric center and pyrimidine nucleotide. Binding was enthalpy-driven, with negative free energy only for the canonical donor. We hypothesize that as K D increases, an unfavorable entropic contribution results from less desolvation and/or conformational freedom of non-native ligands upon binding. 52,75−82 Exclusion of activated α-sugars and C6-hydroxyl sugars prevents "incorrect" sugar incorporation into glycans, as cellular UDP-α-GlcNAc and dTDP-α-Glc concentrations (∼200 μM) are below measured K D values. 56,58,83−86 Notably, a natural product RT utilized dTDP-α-Glc as a donor, demonstrating different preferences exist. 87 WbbL proteins transferred dTDP-β-L-6dTal 88 to glycolipids, implying that bacterial polysaccharides can be engineered, as explored by others. 7,89−93 Acceptor analysis indicated that membrane fractions containing endogenous lipids could be used to assess preferences. While RTs can recognize short lipids, E. coli WecA, the first transferase in the O-Ag pathway ( Figure 5A), requires at least a C35 lipid substrate; hence, shorter lipid lengths are sufficient for later stages of assembly. 59,60,94 Based on predicted structural similarity across these RTs, it is perhaps unsurprising that consistent substrate recognition trends were observed.
In contrast to well-studied bacterial glycosyltransferases, including the peptidoglycan precursor synthase MurG, 95,96 few RT inhibitors exist. 66,67 Since MurG uses the donor UDP-α-GlcNAc, methods to modulate MurG are not transferable to WbbL. Further, while the toxin colicin M depletes O-Ag by targeting carrier lipid modification, it also affects essential biosynthetic pathways. 97 We found that 2 reduces cellular O-Ag levels and is nontoxic; however, like other iminosugars, 2 requires improvement before assessing virulence effects in a host. 66,98 Our findings will inform substrate analogue design to discover more potent inhibitors, and probe surface glycan biosynthesis mechanisms and host−pathogen interactions. ■ ASSOCIATED CONTENT
Tables S1−S8, Figures S1−S26; description and schemes of experimental methods for nucleotide sugar and glycolipid syntheses; compound characterization (NMR, mass spectrometry); methods for plasmid construction, protein expression and purification (Tables S6−S8 contain information on plasmids, primers and strains that were utilized); and additional analyses of enzymatic activities of WbbL and RfbF, confirmation of coupled products (HPLC and mass spectrometry analyses), and bioinformatic analyses (PDF)