LptE binds to and alters the physical state of LPS to catalyze its assembly at the cell surface

Significance

The outermost membrane of Gram-negative bacteria contains lipopolysaccharide (LPS), and its proper placement on the cell surface is required to establish an effective permeability barrier. The presence of LPS prevents small hydrophobic molecules from entering the cell making it difficult to develop antibiotics. LPS is synthesized inside the cell and must move across three compartments to the cell surface. The final step of LPS transport (Lpt), translocation of LPS across the outer membrane, is accomplished by a two-protein complex (LptDE) that must insert LPS into its outer leaflet. We have identified a binding site within LptE critical for the proper function of the translocon. LptE binds LPS and changes its physical state to facilitate its translocation through the outer membrane.

Abstract

The assembly of lipopolysaccharide (LPS) on the surface of Gram-negative bacterial cells is essential for their viability and is achieved by the seven-protein LPS transport (Lpt) pathway. The outer membrane (OM) lipoprotein LptE and the β-barrel membrane protein LptD form a complex that assembles LPS into the outer leaflet of the OM. We report a crystal structure of the Escherichia coli OM lipoprotein LptE at 2.34 Å. The structure reveals homology to eukaryotic LPS-binding proteins and allowed for the prediction of an LPS-binding site, which was confirmed by genetic and biophysical experiments. Specific point mutations at this site lead to defects in OM biogenesis. We show that wild-type LptE disrupts LPS–LPS interactions in vitro and that these mutations decrease the ability of LptE to disaggregate LPS. Transmission electron microscopic imaging shows that LptE can disrupt LPS aggregates even at substoichiometric concentrations. We propose a model in which LptE functions as an LPS transfer protein in the OM translocon by disaggregating LPS during transport to allow for its insertion into the OM.

The surface of most Gram-negative organisms is covered with lipopolysaccharide (LPS) (Fig. 1), a complex glycolipid containing six fatty acyl chains attached to a core oligosaccharide, to which the O-antigen polysaccharide is added (1). LPS molecules are assembled into the outer leaflet of the outer membrane (OM) to form a barrier that hinders the entry of small hydrophobic compounds, including antibiotics, into the cell (2, 3). In Escherichia coli, seven essential LPS transport (Lpt) proteins, LptABCDEFG, form a transenvelope complex that facilitates LPS transport from its site of synthesis across the periplasm and OM to the cell surface (4) (Fig. 1). The ATP-binding cassette (ABC) exporter LptBFG extracts LPS from the outer leaflet of the inner membrane (IM) in an ATP-dependent manner and delivers it to LptC, a periplasmic LPS-binding protein anchored into the IM (5–8). Subsequent rounds of ATP hydrolysis drive LPS across the periplasmic protein bridge created from homologous domains in LptC, LptA, and the N-terminal, periplasmic domain of the OM protein LptD (4, 9, 10). In this way, hydrolysis of ATP on the cytoplasmic face of the IM is capable of providing energy to extract LPS from the periplasmic face of the IM and to provide energy to push LPS unidirectionally along the periplasmic bridge. To complete the biogenesis of the LPS surface, it must be transported through the OM, so that it can be specifically assembled into the outer leaflet.

Fig. 1.

Lipopolysaccharide structure (Left) and its assembly pathway into the Gram-negative OM (Right).

Once LPS reaches the end of the bridge at the inner leaflet of the OM, it is assembled into its outer leaflet by the LptDE translocon (11, 12). The mechanism by which this two-protein complex moves an LPS molecule containing six fatty acyl chains and up to hundreds of monosaccharide units across the OM is not known. Several significant challenges must be overcome for efficient LPS transport. Insertion of LPS into the inner leaflet must be prevented or remediated, so that the asymmetric lipid bilayer of the OM can be established and maintained (13, 14). In addition, interactions between LPS molecules within the bridge must be altered as this 1D assembly passes through the OM and ultimately spreads into a 2D surface. It is believed that LptD must be involved in LPS transport because it is the only integral OM spanning protein in the Lpt pathway. It is also known that LptE forms a plug inside the barrel of LptD (11, 15), but how this two-protein complex accomplishes the goal of selective translocation into the outer leaflet is unknown. Although it has been established that LptE plays an essential role in the assembly of functional LptD (11, 16–19), its role in LPS transport is unclear.

Results

Structure of LptE Suggests Its Role in LPS Binding.

Bioinformatic analysis demonstrates that LptE is present in all Gram-negative bacteria (SI Appendix, SI Text and Fig. S1). With the hope that structural information would help us understand its function, we sought to crystallize LptE from E. coli. Limited proteolysis (SI Appendix, SI Materials and Methods) was performed to define the boundaries of the rigid core of the protein amenable for crystallization experiments, and the construct comprising residues 19–186 gave rise to crystals which diffracted X-rays to 2.34-Å resolution. The LptE structure, along with those of its homologs available in the Protein Data Bank (PDB ID codes 2JXP, 2R76, and 3BF2), were used to devise testable models of LPS binding. LptE consists of two α-helices (α1 and α2) and four β-strands, three of which are long (β2–β4) and one significantly shorter (β1) (Fig. 2). The helices are packed against each other and against one face of the four-stranded β-sheet (Fig. 2A), as in LptE homologs from other organisms (SI Appendix, Fig. S4). The electron density for the first 15 N-terminal residues and the final 18 C-terminal residues of mature LptE are missing. It is possible that the N terminus of LptE is flexible to allow it to be connected to its membrane anchor at the N terminus, while forming a plug within the lumen of the β-barrel of LptD (11, 15). The disordered C-terminal portion of LptE is consistent with the observation that it is readily cleaved by limited proteolysis of the LptDE complex (11). In any event, genetic studies have established that this construct is functional (20).

Fig. 2.

The structure of LptE is homologous with proteins that bind negatively charged oligosaccharides, including LPS. (A) LptE exhibits a two-layer sandwich architecture composed of a four-stranded β-sheet (labeled as β1-β4) and two α-helices (labeled as α1 and α2) packed against one surface of the sheet. (B) Overlay of LptE (green) with HpaA (yellow), a protein from H. pylori that binds neuraminyllactose (PDB ID code 3BGH; rmsd of C_α atoms 3.16 Å). (C) LptE (green) exhibits structural similarity to the horseshoe crab anti-LPS factor, LALF (cyan), an LPS-binding protein from the hemocytes of Limulus (rmsd of C_α atoms 2.74 Å). The amphipathic loop (AL) in LALF, connecting its strands β2-β3, is highlighted in red and labeled, as is the structurally analogous loop connecting β2-β3 in LptE (purple). The positions of basic residues R91 (in the loop connecting β2 and β3, highlighted in purple) and K136 (in the loop connecting strands β4 and α2) is indicated by asterisks.

A search for structural homologs of LptE in the PDB reveals its similarity to proteins that bind negatively charged carbohydrates (SI Appendix, SI Text). Among the closest structural matches is a putative neuraminyllactose-binding hemagglutinin homolog, the Helicobacter pylori adhesin A (HpaA), whose ligand is a negatively charged trisaccharide 3-siallyllactose (Fig. 2B) (21, 22). The structure of LptE also bears significant resemblance to the structure of the black tiger shrimp (Penaeus monodon) and horseshoe crab (Limulus) anti-LPS factor (LALF) (Fig. 2C) (23). This protein, found in the hemolymph of arthropods, is part of a primitive host defense system against microbial invasion (24, 25). It has been proposed that these anti-LPS factors bind LPS through interactions between the negatively charged portions of the lipid A of individual LPS molecules and a region corresponding to the exposed loop connecting β strands 2 and 3 (highlighted in red in the structure of LALF and in violet in the corresponding location in LptE in Fig. 2C) (26). Synthetic peptides derived from the corresponding loop in the LALF also demonstrate LPS binding (27). Another feature of LptE that caught our attention is the region connecting the strand β4 with the helix α2, proximal to the other putative LPS-binding site, which comprises the most highly conserved segment among LptE homologs in Proteobacteria (SI Appendix, Figs. S2 and S3).

Substrate-Binding Site Mutants of lptE Exhibit LPS-Assembly Defects.

One attractive hypothesis is that residues in these two loops are involved in LPS binding. Because basic residues are crucial in LPS interactions with LALF, we made single and pairwise replacements of basic residues at position 91 and 136 in LptE (highlighted by asterisks in Fig. 2C) with acidic ones. Mutants producing either LptE(R91D) or LptE(K136D) did not show significant OM permeability defects against a battery of antibiotics. However, cells producing LptE(R91D,K136D) became sensitive to vancomycin and rifampin (Fig. 3A). This phenotype is characteristic of defective OM biogenesis, and we reasoned it could arise either from defective assembly of the LptDE translocon or its impaired function in LPS translocation (–30). Previously characterized lptE mutations disrupt the OM-permeability barrier simply because they interfere with the assembly of oxidized LptD (LptD_OX) (16). Levels of LptD_OX were equivalent to wild-type in the LptE(R91D) mutant, but we did observe a modest assembly defect caused by the K136D substitution (Fig. 3B). To determine whether this assembly defect is responsible for the lptE(R91D,K136D) phenotypes, the following comparisons were done. LptD is translocated from the cytoplasm via a SecB-dependent pathway; a secA827::IS1 allele interferes with this translocation and results in drastically lowered LptD_OX abundance (31). We note that the LptD_OX assembly defect caused by secA827::IS1 is much more severe than that caused by the K136D mutation (Fig. 3A). However, as evidenced by antibiotic sensitivity, the OM barrier disruption caused by lptE(R91D,K136D) is much greater than that caused by the secA827::IS1 mutation. Accordingly, the antibiotic-sensitivity of the LptE(R91D,K136D)-producing mutant strain cannot be attributed to low LptD_OX levels, because a secA827::IS1 strain produces an even lower level of LptD_OX but maintains full resistance to these antibiotics. We therefore conclude that these lptE mutations primarily affect the function of the OM translocon, rather than its assembly.

Fig. 3.

Changing both R91 and K136 in LptE perturbs the OM barrier. Sensitivity to rifampin and vancomycin was determined for each strain by plating serial dilutions of cultures (A). Level of OM-assembled LptD_OX were determined in lptE haploid strains expressing the denoted plasmid-borne lptE alleles (B). secA alleles linked to leuA:: Tn 10 were introduced in the plptE⁺-expressing strain.

Positively Charged Residues at the Extracellular Side of LptE Mediate LptE-LPS Interactions.

The combination of structural and genetic evidence led us to hypothesize that LptE interacts with LPS at R91 and K136 as a requirement for translocation to the cell surface. Solubilized LPS has the propensity to form aggregates making biochemical characterization challenging (32). Therefore, we performed binding experiments on LPS-coated surfaces. We immobilized LPS onto carboxymethylated dextran (CM3) chips and used surface plasmon resonance (SPR) to monitor interactions (Fig. 4A) (33). When chips derivatized with LPS are subjected to injections of polymyxin B, a cyclic LPS-binding peptide (34), it remained on the surface as indicated by the change in mass adhered to the chip (measured in resonance units), whereas RNase A, known to recognize other negatively charged sugars, is not retained by LPS. The square-shaped signal observed for RNase A arises from mass entering and leaving the flow cell without interacting with the surface. Upon injection of LptE, the sensorgram shows an initial increase followed by rapid decrease of signal below the final signal of immobilized LPS. Multiple injections of the same concentration of LptE converge to a signal intensity corresponding to the level before application of LPS, indicating that LptE removes LPS from the chip surface (Fig. 4A). Therefore, LptE binds to and solubilizes LPS adhered to a CM3 chip. Because the structure of LptE shows homology to the eukaryotic protein LALF, which binds and transports LPS (23, 35), we wondered whether LALF would show similar binding interactions with LPS in our SPR assay. In fact, LALF is also able to extract LPS from the chip surface (Fig. 4A).

Fig. 4.

LptE binds and solubilizes surface bound LPS. (A) LPS (0.1 mM) was loaded onto CM3 SPR chips saturated with polylysine for 200 s (black). Multiple, consecutive injections of 60 μM LptE remove LPS from the surface (red); 40 μM polymyxin B (orange) binds to LPS, whereas the same concentration of RNase A (cyan) is not retained. The LptE structural homolog LALF extracts LPS (4 μM) (blue). (B) Concentration dependence of LptE binding and extraction activity; 20 μM LptE (orange), 40 μM LptE (cyan), 60 μM LptE (black) with SD for every 20th measurement point, 80 μM LptE (blue), 100 μM LptE (red). (C) One hundred-second injections of 60 μM wild-type LptE (black) rapidly removes LPS from the chip, whereas 60 μM LptE (R91D, K136D) (red) does not. Single amino acid substitutions show their contribution to the effect, with 60 μM LptE (R91D) in blue and 60 μM LptE (K136D) in cyan. (D) LptE (10 μM) without (red) and with (cyan) incubation of 1 mM LPS before injection over LPS chip. The black line represents injected 1 mM LPS over an existing LPS surface. All experiments were performed at 25 °C for a 30-s injection time if not indicated otherwise.

We analyzed the concentration dependence of LptE’s interaction kinetics with LPS-coated surfaces. Two discrete interaction steps that are responsible for the removal of LPS become visible (Fig. 4B). At lower concentrations, LptE is first retained on the surface before extraction of LPS starts, whereas at higher concentrations, the binding and extraction steps are of comparable kinetics and are thus indistinguishable. Hence, LPS solubilization is a two-step process, where LptE molecules bind to the LPS aggregate until a critical concentration of LptE is achieved, at which point the second step of LPS extraction begins.

The single and double mutant LptE proteins exhibit defects in their ability to solubilize LPS. LptE(K136D) extracts LPS from the surface less effectively than wild-type LptE (wtLptE), whereas LptE(R91D) is even less efficient and the double mutant LptE(R91D,K136D) extracts none (Fig. 4D), just like RNase A (Fig. 4A). Increasing the concentration of either LptE(R91D) or LptE(K136D) does allow some extraction, whereas the double mutant is still incapable of extracting LPS (SI Appendix, Fig. S6). These observations are consistent with the fact that only the double mutant shows a phenotype in vivo.

To determine whether LptE(R91D,K136D) is deficient in binding, or only in its ability to extract LPS, we immobilized LptE, both wild type and the mutant on an SPR chip and passed the same concentration of LPS over the surface. The mutant and wild-type LptE bind LPS comparably (SI Appendix, Fig. S8). This supports our interpretation that binding and disaggregation are separate processes and that the lptE(R91D,K136D) mutations affect the latter. Taken together, these data suggest that amino acids R91 and K136 are part of an LPS-interaction site on LptE that acts to solubilize LPS by altering LPS–LPS interactions.

LptE Disrupts Large LPS Aggregates.

The fact that LptE can facilitate LPS extraction from the surface of the chip implies that interactions between LptE and LPS are more favorable than intermolecular interactions between LPS molecules. It is not surprising that LPS-binding proteins would bind LPS so tightly given the extreme sensitivity of the limulus amebocyte lysate endotoxin detection assay (36). Given that LptE can extract LPS from the chip surface, we wondered whether the reverse was also true (i.e., whether it can prevent LPS from sticking to the surface). Consistent with expectation, if a high concentration (1 mM) of LPS in solution is passed over an LPS-coated chip, an increase in signal is seen (Fig. 4C), suggesting that additional LPS molecules can adhere to the LPS surface. If 10 μM LptE is preincubated with 1 mM LPS before injection, no solubilization of LPS from the surface by LptE happens. Remarkably, we also see no increase in LPS binding to the surface, even though it is present in a 100-fold excess relative to LptE (Fig. 4C). Substoichiometric amounts of LptE are capable of preventing the interaction of LPS with LPS. One interpretation of these results is that LptE alters the structure of the LPS aggregate, even at low concentration.

We used transmission electron microscopy (TEM) to test this hypothesis. On a carbon-covered grid, pure LPS forms extended tubular filaments that are about 17 nm wide. Compared with E. coli OM that have a diameter of 13 nm, these filaments could represent a bilayer structure (37). LPS has a distinctly low critical micelle concentration (CMC) (subpicomolar) compared with other glycolipids, which are often in the millimolar range, and therefore aggregates readily (Fig. 5A). Imaging the solutions of LPS:LptE (at a 4:1 ratio) reveals that the filaments disappear for wild-type LptE (Fig. 5C) but not for LptE(R91D,K136D) (Fig. 5B). TEM images of LptE without LPS confirm that the structures seen are not protein aggregates (SI Appendix, Fig. S7). LptE therefore disrupts strong LPS–LPS interactions and affects its aggregation propensity. Free LPS is not able to spontaneously insert into phospholipid vesicles (38, 39). Because of its low CMC, there is a barrier to insertion into phospholipids that presumably involves monomerization. The disaggregation of LPS by LptE observed in TEM and surface plasmon resonance could help to insert LPS into the OM.

Fig. 5.

TEM images of LPS aggregates before and after 3 h of LptE incubation at room temperature. (A–C) LPS (200 μM), 200 μM LPS with 50 μM LptE (R91D/K136D) (B), or 200 μM LPS with 50 μM LptE wild type (C). All negatively stained with uranyl formate and imaged at a nominal magnification of 49,000×. (Scale bar: 10 nm.)

Discussion

Our results establish that LptE binds LPS specifically and that this is crucial to LptE’s function in vivo. Despite the relatively low conservation of its amino acid sequence, LptE homologs are widespread among Gram-negative organisms and are predicted to fold into the same 3D structure. The striking resemblance of the structures of LptE and the anti-LPS factors from arthropods, known to bind LPS through the loop connecting β strands 2 and 3, enabled us to probe LPS binding in that region. We constructed mutant LptE proteins that lack exposed positive charges in the corresponding loops of LptE, and the resulting mutant proteins exhibit OM defects in vivo, indicative of a compromised LPS-assembly pathway. In vitro binding experiments established that LptE binds LPS, is able to interfere with LPS–LPS interactions, and affects LPS’s aggregation state. The in vivo effect of rationally designed lptE mutations described above was mirrored in SPR and TEM experiments, where proteins lacking the positively charged residues R91 or K136 in the putative carbohydrate-binding site exhibited significantly diminished ability to extract and disaggregate LPS. As such, LptE’s ability to alter the aggregation state of LPS is key to proper OM biogenesis. The architecture and substrate-binding sites of LptE and LPS-binding proteins from the hemolymph of arthropods point to a convergent evolution of this fold for LPS binding.

The crystal structure of E. coli LptE reported here, combined with homology structural models and synteny analyses, allowed us to establish that certain structural features, in particular the loops containing basic residues, are conserved across LptE homologs (SI Appendix, Fig. S5). This conservation is not evident if one simply compares distant LptE proteins at the level of amino acid sequence (SI Appendix, Fig. S1). These conserved loops resemble a region of the OM β-barrel protein FhuA; its crystal structure of FhuA bound to LPS reveals that binding is mediated by a loop bearing positive charges (26, 40). These authors point to the fact that the FhuA LPS-binding motif bears significant resemblance to eukaryotic LPS-binding proteins. We think it is noteworthy that many of these proteins, in particular LALF, bactericidal permeability increasing protein (BPI) (41), and LPS-binding protein (LBP) (42), possess functions that extend beyond the mere binding of LPS. It has been hypothesized that the eukaryotic structural homolog LALF participates in not only the binding of LPS but also in its transport to and insertion into phospholipids membranes as part of the Limulus host immune response (23, 35). LBP and BPI are elongated in shape and contain a duplicated fold similar to that of LptE. They bind LPS through an analogous exposed loop at a tip that connects β-strands and exhibits conserved stretches of alternating cationic and hydrophobic residues (26). LBP monomerizes LPS aggregates either to present it to CD14 or, at high LPS concentrations, to insert it into membranes to activate cells independent of CD14 (43, 44). Our SPR data reveal that LptE not only binds LPS but is capable of disrupting the interactions between LPS molecules. The ability to remove immobilized LPS off the surface of the SPR chip has been observed before in SPR experiments involving lipid transfer proteins in an analogous experimental setup (45, 46). These proteins transfer glycolipids between lipid membranes. Accordingly, LptE might be able to disaggregate and transfer LPS into the OM in vivo.

Our observations of the unusual LPS-disaggregation behavior of LptE in vitro may help explain many of the phenomena that the OM translocon facilitates. We have previously proposed that LPS is transported in a continuous stream from the outer leaflet of the IM to the cell surface along the periplasmic bridge formed by LptC, LptA, and LptD and that ATP hydrolysis in the cytoplasm provides energy to drive this string of LPS molecules in a unidirectional fashion against a concentration gradient (9). The OM translocon must take individual LPS molecules from this 1D stream and facilitate their passage into the 2D LPS assembly in the outer leaflet. The major barrier (bottleneck) in this process is clearly transport of the amphipathic LPS molecule through the hydrophobic interior of the OM. We have shown that LptE’s interactions with LPS are sufficient to facilitate removal of LPS from an LPS aggregate on an SPR chip. We have also shown the converse; in the absence of LptE, added LPS will insert into LPS aggregates on an SPR chip (Fig. 4). Taken together, these experiments imply that LptE–LPS interactions are more stable than LPS–LPS interactions on the surface of an SPR chip. We have also shown that LptE can alter the aggregate state of LPS in solution (Fig. 5), apparently disrupting LPS–LPS interactions. In the OM translocon, a substoichiometric amount of LptE must facilitate LPS transfer through a narrow pore into the outer leaflet while preventing its insertion into the inner leaflet of the OM. We propose a model wherein LptE acts as a transfer protein to facilitate LPS movement into the OM by providing more favorable interactions between itself and LPS than between aggregated LPS molecules. Eventually, LptE must release LPS into the outer leaflet. LptE’s strong interactions with LptD could be responsible for this final step of LPS transport; by engulfing LptE, LptD facilitates the release of LPS from LptE. If this is so, then LptE is a catalyst for LPS assembly into an asymmetric LPS membrane. Localizing LptE to the inner leaflet away from the cell surface prevents this catalyst in prokaryotic systems from degrading the outer leaflet as similar proteins, such as LALF, are presumed to do in eukaryotic immune systems.

Materials and Methods

Limited Proteolysis of LptE to Identify the Fragment for Crystallization.

LptE [(50 μL, at 0.5 mg/mL in Tris-buffered saline (TBS)] was incubated with 5 μL of trypsin or subtilisin at concentrations 200, 40, 8, 1.6, and 0.32 μg/mL for 60 min at 37 °C. The reactions were quenched by the addition of phenylmethylsulfonyl fluoride to 4 mM, and the resulting fragments were analyzed by SDS/PAGE, electrospray ionization mass spectrometry, and Edman sequencing. Mass spectrometry of the tryptic digest revealed the presence of a major peak at 18,029.98 Da, which closely corresponded to the calculated mass of LptE(25–186) (18,029.5 Da), whereas the sample digested by subtilisin exhibited a strong peak at 18,865.95 Da, which closely corresponded to the calculated mass of LptE(20–188) (18,865.4 Da).

Expression and Purification of LptE for Crystallization.

The expression of the crystallization construct, cell disruption, and nickel-nitrilotriacetic acid (Ni-NTA) chromatography were performed as described above. Following the Ni-NTA chromatography, LptE was concentrated to 10 mg/mL and incubated with thrombin (EMD Biosciences), 1 unit per milligram LptE, in TBS additionally containing 2 mM CaCl₂ overnight at 25 °C. The next day, the protein was loaded onto Superdex75 10/300 (GE Healthcare) running in 20 mM Hepes/NaOH (pH 7.0), 100 mM NaCl. LptE eluted from the column as a monomer, and fractions containing pure protein were pooled and reductively methylated (47). The methylation was quenched by the addition of Tris⋅HCl (pH 8.5) to a final concentration of 20 mM, and its buffer was exchanged to water by concentrating using a 10-kDa cutoff Amicon Ultra centrifugal filter to 8 mg/mL.

Selenomethionine-labeled LptE was expressed in the same bacterial strain by using metabolic inhibition in M9 minimal medium supplemented with carbenicillin, glucose, and all amino acids, but with selenomethionine instead of methionine, as described in detail in refs. 48, 49. All subsequent purification and crystallization steps were identical to those for the nonlabeled protein.