A new antibiotic traps lipopolysaccharide in its intermembrane transporter

Abstract

Gram-negative bacteria are extraordinarily difficult to kill because their cytoplasmic membrane is surrounded by an outer membrane that blocks the entry of most antibiotics. The impenetrable nature of the outer membrane is due to the presence of a large, amphipathic glycolipid called lipopolysaccharide (LPS) in its outer leaflet¹. Assembly of the outer membrane requires transport of LPS across a protein bridge that spans from the cytoplasmic membrane to the cell surface. Maintaining outer membrane integrity is essential for bacterial cell viability, and its disruption can increase susceptibility to other antibiotics^2-6. Thus, inhibitors of the seven lipopolysaccharide transport (Lpt) proteins that form this transenvelope transporter have long been sought. A new class of antibiotics that targets the LPS transport machine in Acinetobacter was recently identified. Here, using structural, biochemical and genetic approaches, we show that these antibiotics trap a substrate-bound conformation of the LPS transporter that stalls this machine. The inhibitors accomplish this by recognizing a composite binding site made up of both the Lpt transporter and its LPS substrate. Collectively, our findings identify an unusual mechanism of lipid transport inhibition, reveal a druggable conformation of the Lpt transporter and provide the foundation for extending this class of antibiotics to other Gram-negative pathogens.

Fig. 1. Macrocyclic peptides block LPS transport by binding to the inner membrane complex.

a, Schematic of the seven protein LPS transport machine. b, Structures of macrocyclic peptides that prevent growth of Acinetobacter strains. The compounds were selected from those prepared during the drug discovery–development process: compound 1 (RO7196472) was a potent hit found early in the discovery process; compound 2 (Zosurabalpin) is a clinical candidate; and compound 3 (RO7075573) was an important preclinical lead. Compound 2a, the epimer of compound 2 at the starred position, is an inactive compound that was used as a negative control. c, Cryo-EM structure of the inner membrane A. baylyi LptB₂FG complex bound to LPS and 1. The drug has 500 Ų contact with LptFG and 230 Ų contact with LPS. Postprocessing of the map was carried out using DeepEMhancer. The unsharpened map is shown as an outline to show the positioning of the detergent micelle. Inset shows a close-up view of LPS and 1. LptB, LptF, LptG, LPS and 1 are coloured tan, green, blue, yellow and purple, respectively. d, Cryo-EM structure of Acinetobacter LptB₂FG with Acinetobacter LPS and 1 bound in the lumen of the transporter in white superimposed with the structure of Acinetobacter LptB₂FG bound to E. coli LPS and 1 (LptF, LptG, LPS and 1 are coloured green, blue, yellow and purple respectively). The overall r.m.s.d. is 0.44 Å over 7,999 atoms.

Fig. 2. Compound 1 binds an intermediate transport state with LPS bound in the transporter.

a, View of the ternary complex highlighting key contacts from LptF (bolded) to both LPS and 1. LptF, LptG, LPS and 1 are coloured green, blue, yellow and purple, respectively. b, Table showing the MICs of 1 against A. baylyi containing various LptF variants. MIC values were consistent across three cultures started from individual colonies. c,d, 1 inhibits LPS transport to LptA by wild-type LptB₂FGC (c) but not by LptB₂F^E249KGC or LptB₂F^I317NGC (d). Lipopolysaccharide transport from LptB₂FGC to LptA modified with a photocrosslinkable amino acid (I36pBPA) was monitored in the presence of the indicated dose of 1 by exposing the samples to UV light after 60 min of transport, quenching by addition of SDS-loading buffer, PAGE to separate LPS-LptA adducts from LPS and western blotting against LPS. Data shown are representative of experiments conducted in biological triplicate. e, Cryo-EM structure of LptB₂FG with LPS bound in the lumen of the transporter in white superimposed with the LptB₂FG-LPS-1 structure, which is coloured as in a. The two structures have an r.m.s.d. of 0.31 Å over 8,010 atoms.

Fig. 3. A binding pocket is created for 1 in LptB₂FGC by moving TM helices of LptC and LptF.

a, The acyl chain that is added by LpxM, highlighted in salmon, nestles between LptF helices 2, 4 and 5. Residues contacting this acyl chain are labelled. Residues that contact this acyl chain and elicited resistance in spontaneous mutation studies are bolded. LptF is shown in green, LPS in yellow and LptG in blue. b, LPS isolated from a ΔLpxM strain renders LptB₂FGC resistant to 1 (ref. ). Lipopolysaccharide transport from LptB₂FGC to LptA was measured as described in Fig. 2. Data shown are representative of experiments conducted in biological triplicate. c, Cryo-EM structure of Acinetobacter LptB₂FGC superimposed with the structure of Acinetobacter LptB₂FG in complex with LPS and 1. The LptB₂FGC structure is shown in pink, whereas the LptB₂FG-1-LPS structure is coloured as in Fig. 2a. The observed positioning of the TM helix of LptC sterically clashes with the compound 1 binding site observed in the LptB₂FG structures. The positioning of LptF helices 2–5 are also shifted relative to what was observed in the LptB₂FG structures.

Fig. 4. Macrocyclic peptide antibiotics bind an LPS-bound intermediate state in which the TM helix of LptC has moved from the lumen.

a–c, Cryo-EM structures of 1 (a), 2 (b) and 3 (c) bound to LptB₂FG. The observed positioning of the LptC TM helix from Fig. 3d is highlighted in pink. d, Transport of LPS to LptA by LptB₂FGC is not inhibited by 3 in vitro but LPS transport by LptB₂FG-ΔTM-C is inhibited by 1, 2 and 3 in vitro. Data shown are representative of experiments conducted in biological triplicate. e, 1 treatment increases the ATPase activity of LptB₂FG, LptB₂FGC and LptB₂FG-ΔTM-C in an LPS-dependent manner. This effect is reduced in the presence of the TM helix of LptC. ATP hydrolysis was monitored by measuring concentrations of inorganic phosphate. Experiments were conducted in biological triplicate and data are presented as mean values ± standard deviation. f, 2 binds Lpt in the presence of LPS and absence of LptC. The binding of radiolabelled 2 to His-tagged LptB₂FG and LptB₂FGC in the presence and absence of LPS was measured in a SPA. Data are presented as counts per minute (c.p.m.) in arbitrary units (a.u.) and are from three biological replicates. g, The cellular activity of 1-derivatives correlates to their observed binding to LptB₂FG through SPA. The ability of 1-derivatives to displace radiolabelled 2 from His-tagged LptB₂FG was measured in the presence of LPS. Active compounds 1–3 showed potent binding to LptB₂FG, whereas the inactive control compound 2a did not. 2a is the epimer of 2 at the highlighted (*) carbon; Fig. 1. Uncertainties represent the standard deviation of three biological replicates.

Fig. 5. E. coli and Acinetobacter have distinct druggable pockets at the LPS–LptFG interface.

a–f, Three different representations of the structures of either Acinetobacter LptB₂FG (a,c,e) or E. coli LptB₂FG (b,d,f) in complex with E. coli LPS. In the case of Acinetobacter, the positioning of 1 is as observed experimentally. In the case of E. coli, 1 is placed based on alignment to the Acinetobacter LptB₂FG structure. a,b, The binding site for macrocyclic peptide 1 that is present in Acinetobacter (a) is not present in E. coli (b). As highlighted, the drug has steric clashes with both LPS and helix 5 of E. coli LptF. c,d, The electrostatic surface of Acinetobacter (c) and E. coli (d) LptB₂FG, with negative surfaces shown in red and positive surfaces in blue. Note that the primary amine of the macrocyclic peptides lodge into a negative pocket in Acinetobacter LptB₂FG that does not exist in E. coli LptB₂FG. e,f, Both Acinetobacter and E. coli LptB₂FG have extra cavities formed between LPS and Lpt protein. In this Article, we have validated that drug binding to a composite surface between Acinetobacter LptB₂FG and LPS (purple pocket, e) can block LPS transport. Analogous pockets exist in other species (pink, f), providing opportunities for future drug design.

Extended Data Fig. 1. Sequences of Lpt proteins from Acinetobacter baylyi ADP1, Acinetobacter baumannii 19606 and E. coli K12.

Percent coverage and identities when compared to the Acinetobacter baylyi ADP1 sequence are shown. Alignments were made using ClustalO and then coloured in MView (https://www.ebi.ac.uk/Tools/msa/clustalo/).

Extended Data Fig. 2. Expression, purification and cryo-EM data processing for LptB₂FG in complex with LPS and 1.

a. Representative size-exclusion chromatogram of Acinetobacter baylyi LptB₂FG. b. SDS–PAGE gel showing peak fractions from the size-exclusion chromatography of the complex in a and used in subsequent structural experiments described in this figure. Four independent purifications of this construct gave similar size-exclusion and SDS–PAGE results. c. Representative (n = 12,402) micrographs of LptB₂FG in complex with 1 and LPS embedded in vitreous ice. d. Scheme of three-dimensional classification and refinement of cryo-EM particle images. The initial models used for 3D classification were generated by ab initio reconstruction in cryoSPARC. e. Representative selected two-dimensional class averages of cryo-EM particle images. (f-g) Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC from the overall (f) and locally refined (g) structures of 1-bound LptB₂FG. The resolution was determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 3.0 Å. h. LPS and 1 fit in the final locally refined, sharpened map. Post processing of the map was done using DeepEMhancer.

Extended Data Fig. 3. Cryo-EM data processing and analysis for Acinetobacter LptB₂FG with Acinetobacter LPS in the presence and absence of 1.

a. Structures of E. coli and Acinetobacter LPS. The parts that were not modelled in the structures presented in this work are coloured in red. b. Scheme of three-dimensional classification and refinement of cryo-EM particle images for Acinetobacter LptB₂FG in the presence of Acinetobacter LPS and 1. The initial models used for 3D classification were generated by ab initio reconstruction in cryoSPARC. c. Representative selected two-dimensional class averages of Acinetobacter LptB₂FG in the presence of Acinetobacter LPS and 1. (d-e) Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC from the overall (d) and locally refined (e) structures of Acinetobacter LptB₂FG in the presence of Acinetobacter LPS and 1. The resolution was determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 3.15 Å. f. 3D classification without alignments of the dataset detailed in (b–e) revealed conformational flexibility in the lysine of 1. There is a major population (purple box in b, transparent surface in f, state 1) in which said lysine is in contact with the 2’-phosphate of LPS and a minor population (pink box in b, solid surface in f, state 2) where the lysine is oriented towards the carboxylate of the branched KDO from LPS. This KDO is conserved in E. coli and Acinetobacter LPS, as shown in panel a. (g-h). There is an ordered DDM molecule in the Acinetobacter LptB₂FG + 1 structure obtained with E. coli LPS, but not in the structure obtained using Acinetobacter LPS. LptF, LptG, LPS and 1 are coloured green, blue, yellow and purple. The structure obtained using E. coli LPS is shown in g with the DDM coloured violet. The Acinetobacter LPS acyl chain that overlaps with the location of the DDM highlighted in salmon (h).

Extended Data Fig. 4. Cryo-EM data processing and analysis for drug-free Acinetobacter baylyi LptB₂FG.

(a-d). Processing for the structure of Acinetobacter baylyi LptB₂FG in the presence of E. coli LPS. a Scheme of three-dimensional classification and refinement of cryo-EM particle images. The initial models used for 3D classification were generated by ab initio reconstruction in cryoSPARC. b. Representative selected two-dimensional class averages of cryo-EM particle images. (c-d) Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC from the overall (c) and locally refined (d) structures of drug-free LptB₂FG. The resolution was determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 3.0 Å. (e-h) Processing for the structure of Acinetobacter baylyi LptB₂FG in the presence of Acinetobacter baylyi LPS. e Scheme of three-dimensional classification and refinement of cryo-EM particle images. The initial models used for 3D classification were generated by ab initio reconstruction in cryoSPARC. f. Representative selected two-dimensional class averages of cryo-EM particle images. (g-h) Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC from the overall (g) and locally refined (h) structures of drug-free LptB₂FG. The resolution was determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 3.1 Å. i. Cryo-EM structure of Acinetobacter LptB₂FG with Acinetobacter LPS bound in the lumen of the transporter in white superimposed with the structure of Acinetobacter LptB₂FG bound to 1 and Acinetobacter LPS (LptF, LptG and LPS are coloured green, blue and yellow, respectively). The overall r.m.s.d. is 0.37 over 7714 atoms. j. Cryo-EM structure of Acinetobacter LptB₂FG with Acinetobacter LPS bound in the lumen of the transporter in white superimposed with the structure of Acinetobacter LptB₂FG bound to E. coli LPS (LptF, LptG and LPS are coloured green, blue and yellow, respectively). The overall r.m.s.d. is 0.51 over 8138 atoms. (k-l). LptB₂F(R55G)G and LptB₂F(R30A)G have comparable ATPase hydrolysis rates to wild-type LptB₂FG (k) but are unable to transport LPS to LptA (l). ATPase rates were measured using liposomes containing LptB₂FG complexes and LPS and the presented data represent averages and standard deviations of results determined using three different proteoliposomes preparations for each LptF variant. Inorganic phosphate release was measured using a molybdate assay. LPS transport from LptB₂FG to LptA^I36pBPA in the presence of LptC-ΔTM was measured by detecting UV-dependent LptA-LPS crosslinks by LPS immunoblotting. Data shown is representative of experiments conducted in biological triplicate.

Extended Data Fig. 5. Loss of function mutations in LpxM render cells resistant to 1.

5 distinct mutations in LpxM were observed to provide resistance to 1. The location of the mutated residues relative to the active site (active-site residues are bolded) is shown.

Extended Data Fig. 6. Cryo-EM data processing and analysis for LptB₂FGC.

a. Representative size-exclusion chromatogram of Acinetobacter baylyi LptB₂FGC. b. SDS–PAGE gel showing peak fractions from the size-exclusion chromatography of the complex in a and used for the structural experiments. Three independent purifications of this construct gave similar size-exclusion and SDS–PAGE results. c. Representative (n = 18,862) micrograph of LptB₂FGC embedded in vitreous ice. d. Scheme of three-dimensional classification and refinement of cryo-EM particle images. The initial models used for 3D classification were generated by ab initio reconstruction in cryoSPARC. e. Representative selected two-dimensional class averages of cryo-EM particle images. (f-g) Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC from the overall (f) and locally refined (g) structures of LptB₂FGC. The resolution was determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 3.6 Å. h. Electron density of the LptB₂FGC structure with LptF coloured in green, LptG in blue and LptC in pink. The lipopolysaccharide positioning from our LptB₂FG structure is overlayed with the LptB₂FGC electron density. Post processing of the map was carried out using DeepEMHancer.

Extended Data Fig. 7. Cryo-EM data processing and analysis for 2- and 3-bound LptB₂FG.

a. Scheme of three-dimensional classification and refinement of cryo-EM particle images for 2-bound LptB₂FG. The initial models used for 3D classification were generated by ab initio reconstruction in cryoSPARC. b. Representative selected two-dimensional class averages of 2-bound LptB₂FG. (c-d) Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC from the overall (c) and locally refined (d) structures of 2-bound LptB₂FG. The resolution was determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 3.25 Å. e. Scheme of three-dimensional classification and refinement of cryo-EM particle images for 3-bound LptB₂FG. The initial models used for 3D classification were generated by ab initio reconstruction in cryoSPARC. f. Representative selected two-dimensional class averages of 3-bound LptB₂FG. (g-h) Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC from the overall (f) and locally refined (g) structures of 3-bound LptB₂FG. The resolution was determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 3.1 Å.

Extended Data Fig. 8. Biochemical determinants of compound binding.

a) 2 treatment increases the ATPase activity of LptB₂FG, LptB₂FGC and LptB₂FG-ΔTM-C but only in the presence of LPS. b) 3 treatment increases the ATPase activity of LptB₂FG, LptB₂FG-ΔTM-C but only in the presence of LPS. 3 does not affect the ATPase of LptB₂FGC. c) LPS isolated from a ΔLpxM strain renders LptB₂FGC resistant to 1. For panels a–c, ATPase rates were measured using liposomes containing the indicated Lpt complex indicated LPS variant and the presented data represent averages and standard deviations of results determined using three different proteoliposomes preparations for each condition variant. Inorganic phosphate release was measured using a molybdate assay. d) 2 does not inhibits LPS transport to LptA by wild-type E. coli LptB₂FGC. Lipopolysaccharide transport from LptB₂FGC to LptA modified with a photocrosslinkable amino acid (I36pBPA) was monitored in the presence of the indicated dose of 2 by exposing the samples to UV light after 60 min of transport, quenching by addition of SDS-loading buffer, PAGE to separate LPS-LptA adducts from LPS and western blotting against LPS. Data shown is representative of experiments conducted in biological triplicate.