Cryo-EM structures of LolCDE reveal the molecular mechanism of bacterial lipoprotein sorting in Escherichia coli

Bacterial lipoproteins perform a diverse array of functions including bacterial envelope biogenesis and microbe–host interactions. Lipoproteins in gram-negative bacteria are sorted to the outer membrane (OM) via the localization of lipoproteins (Lol) export pathway. The ATP-binding cassette (ABC) transporter LolCDE initiates the Lol pathway by selectively extracting and transporting lipoproteins for trafficking. Here, we report cryo-EM structures of LolCDE in apo, lipoprotein-bound, and AMPPNP-bound states at a resolution of 3.5 to 4.2 Å. Structure-based disulfide crosslinking, photo-crosslinking, and functional complementation assay verify the apo-state structure and reveal the molecular details regarding substrate selectivity and substrate entry route. Our studies snapshot 3 functional states of LolCDE in a transport cycle, providing deep insights into the mechanisms that underlie LolCDE-mediated lipoprotein sorting in E. coli.


Introduction
The outer membrane (OM), hallmark of gram-negative bacteria, lies at the frontline of interaction with environment serving as a potent permeability barrier that prevents entry of many toxic substances into the cell [1,2]. Central to OM biogenesis and physiology are the lipoproteins that peripherally anchored to the membrane via their N-terminal lipid moiety. OM lipoproteins underpin the functioning of a diverse array of machineries that are responsible for the lipopolysaccharides (LPS) export, assembly of integral OM proteins and peptidoglycan cell wall, to name a few [3][4][5][6]. Lipoproteins are therefore indispensable for the survival of gramnegative bacteria [7], bearing important implications for efforts to develop novel antimicrobial agents against multidrug-resistant bacteria [8][9][10][11][12]. In gram-negative bacteria, lipoproteins are synthesized in the cytoplasm and matured in the inner membrane (IM). In Escherichia coli, PLOS   catalytically dead LolCDE mutant with 2 mM AMPPNP (the nonhydrolysable ATP analogue). Thereby, we obtained a structure of AMPPNP-LolCDE determined at 3.6 Å resolution (S5, S6A and S6B Figs). The overall structure of apo-LolCDE resembles a dumbbell (Fig 1D). In spite of the fact that LolC and LolE share only 26.0% sequence identity (S7A Fig), they adopt a similar fold. It comprises a TMD domain of 4 transmembrane helices (TM1 to TM4), a periplasmic loop connecting TM3 and TM4 [denoted as Loop LolC (residues: Ser338-Val358) and Loop LolE (residues: Ser342-Ser373), respectively] and a periplasmic localized domain (PLD) stemming from TM1 and TM2 (Fig 1D). Superposition of LolC with LolE gives an RMSD of 2.27 Å for PLDs (over 134 Cα atoms) and 1.63 Å for TMDs (over 176 Cα atoms) (S7B and S7C Fig). The TMD domains of LolC and LolE are assembled primarily via intermolecular TM1-TM2 interactions ( Fig 1D). Strikingly, the TM1s and TM2s of both LolC and LolE are bent outwards creating a V-shaped substrate-binding cavity open to the periplasm. Both PLDs are in elevated position about 20 Å above the plane of membrane, as a result of TM1s and TM2s extending into the periplasm (Fig 1D). The feature echoes what was observed in the structures of the toxin and antibiotic ABC transporter MacB [37,38]. Moreover, the 2 PLDs interact closely with each other, manifesting a twist of approximately 20˚perpendicular to the membrane plane. The surface burial area is calculated to be 405.7 Å 2 .
Similar to LptB 2 FG, an ABC transporter that extracts LPS from the IM [39][40][41][42], the substrate-bound LolCDE adopts almost the same conformation as apo-LolCDE (with an RMSD of 1.28 Å over 1158 aligned Cα atoms) (S8A Fig). It is worth to note that in both apo-LolCDE and RcsF-LolCDE structures, the V-shaped cavity is surrounded by 2 structured motifs, Loop-LolC and Loop LolE (Fig 1E). This creates 2 intermolecular interfaces, that is, the Loop LolC -TM2 LolE and the Loop LolE -TM2 LolC interfaces, perhaps serving as entry gates for lipoprotein substrates. In the V-shaped cavity, the 3 acyl chains of RcsF straddles the TMD interface of LolC and LolE, with 1 acyl chain, R1, residing on the Loop LolC side, and the other two, R2 and R3, extending to the Loop LolE side (Fig 1E). This is in line with the 2 recently reported cryo-EM structures of the lipoprotein-bound LolCDE complexes [33,34]. Importantly, the proteinaceous part of RcsF extends from the V-shaped cavity and appears only on the Loop LolC side, suggesting that RcsF might enter the cavity laterally via the Loop LolC -TM2 LolE interface. In addition, a U-shaped loop (denoted as U-Loop) from Loop LolE dips into the V-shaped cavity, presumably buttressing the configuration of the V-shaped cavity (Fig 1E).
In stark contrast to lipoprotein binding, AMPPNP binding causes significant conformational changes of LolCDE ( S8B Fig and S1 Movie). First, binding of AMPPNP leads to a tight dimerization of the 2 LolDs, resulting in the Cα-Cα distance between Gly45 of the Walk A motif in 1 LolD and Ser147 of the signature motif in the other shortened from 16.7 Å to 6.8 Å (Fig 1F). Second, the V-shaped cavity is flattened, and its periplasmic surface area is reduced to approximately one-third ( Fig 1G). Third, in the periplasm, the 2 PLDs become more compact, with a further increase of surface area of 21.1 Å 2 , in comparison to RcsF-LolCDE ( Fig 1H). Consequently, no lipoprotein is found in the AMPPNP-LolCDE structure, and the U-Loop is squeezed out (S8C Fig). In summary, while our RcsF-LolCDE and AMPPNP-LolCDE structures overlay well with the recently published structures [33,34] (S9A and S9B Fig), our apo-LolCDE structure is strikingly different from the apo-LolCDE structure (denoted as apo-LolCDE � , PDB code: 7ARI) by Tang and colleagues [33] (S9C Fig). This prompted us to further verify the apo-LolCDE structure using biochemical approaches.

Verification of the apo-LolCDE structure
In our apo-LolCDE structure, the Cα-Cα distance between Ala106 in PLD LolC and Ser173 in PLD LolE is only 5.4 Å, whereas the distance is increased to 48.6 Å in apo-LolCDE � [33] (Fig  2A). To validate our structure, 2 cysteines were incorporated into LolCDE to generate Lol-C A106C DE S173C mutant in order to probe the distance between LolC and LoE in LolCDE. In the absence of reducing reagents, we observed intermolecular disulfide bond between Lol-C A106C and LolE S173C almost fully formed in the purified LolC A106C DE S173C protein. Addition of β-mercaptoethanol (β-ME) disrupted the intermolecular disulfide bond (Fig 2B). We thus conclude that the arrangement of the 2 PLDs in LolCDE agrees well with our apo-LolCDE structure. Furthermore, the crosslinks remained the same, regardless of the presence of RcsF ( Fig 2B). These results further confirmed that both the apo-and RcsF-bound LolCDE adopt the same conformations as we observed.
To investigate whether conformational changes of apo-LolCDE are required for RcsF entry into the V-shaped cavity, we incorporated a photo-crosslinkable unnatural amino acid (p-benzoyl-phenylalanine (pBPA)) [43] at Leu256 of LolC (Fig 2A), whose side chain points to the Vshaped cavity. As shown in Fig 2C, LolC×RcsF adducts are detected when both RcsF and LolCDE were reconstituted in nanodisc, demonstrating that the exogenously added RcsF is able to enter the V-shaped cavity of LolC L256pBPA DE (no Cys mutations). Furthermore, we also detected the LolE-LolC×RcsF adducts in LolC A106C DE S173C sample upon UV radiation. This observation implicates that conformational change of the 2 PLDs of LolC and LolE is not required for the entry of RcsF into the V-shaped cavity ( Fig 2C). We observed, however, post-RcsF entry into the V-shaped cavity of disulfide-bonded LolC A106C DE S173C complex; it failed to release from the complex to LolA, unless the intermolecular disulfide bond is disrupted by the addition of reducing reagent (Fig 2D). Failure of RcsF release to LolA by the LolC A106C-DE S173C complex suggests that either an overall conformational change of LolCDE is required or the disulfide bond-fixed PLDs interfere with the lipoprotein transport. In line with our in vitro photo-crosslinking findings, complementation assay also showed that lolC A106C DE S173C failed to rescue the growth of the lolCDE-depleted E. coli cells unless reducing reagent is supplemented (Figs 2E and S10A). Taken together, we conclude that our apo-LolCDE structure represents the correct apo conformational state of LolCDE, which resembles closely the conformation of RcsF-LolCDE. This study also correlates well with the structural and functional studies of LptB 2 FG [39][40][41][42].

RcsF-LolCDE interactions and lipoprotein substrate selectivity
In our RcsF-LolCDE structure, 3 acyl chains (R1, R2, and R3) and the N-terminal 14 residues of the mature RcsF are well resolved (the first Cys residue of a mature lipoprotein named +1 position) (Fig 3A). Specifically, R1 sits between TM1 LolC and TM2 LolE (Fig 3B) by making hydrophobic interactions with residues Val44, Val47, and Met48 of TM1 LolC and Met267 of TM2 LolE (Fig 3C). In the opposite side, R2 and R3, surrounded by TM2 LolC , TM1 LolE , and Loop LolE (Fig 3B), are in close contact with residues Met266 and Met267 of TM2 LolC , Val43, and Phe51 of TM1 LolE , as well as Met261, Ile265, and Ile268 of TM2 LolE (Fig 3C). To probe the functional significance of these contacts, we made point mutations for the abovementioned residues by introduction of hydrophilic residues. We then tested these mutants one by one. Each LolCDE mutant is successfully expressed and assembled giving the right size exclusion Zoom-in view showing 2 amino acids in 2 PLDs, Ala106 and Ser173, which were replaced with cysteines in (B to E). Leu256 LolC shown in purple spheres was substituted with pBPA for in vitro photo-crosslinking in (C). (B) Coomassie-stained SDS-PAGE gel assessing disulfide bond formation of LolC A106C DE S173C and RcsF-LolC A106C DE S173C . The samples of lanes 2 through 4 were supplemented with SDS loading dye without β-ME, and the samples of lanes 6 through 8 were supplemented with SDS loading dye with β-ME. Note that RcsF migrates slower after addition of reducing agent. (C) In vitro photo-crosslinking. LolC L256pBPA DE proteins with or without 2 cysteine mutations were reconstituted with RcsF in nanodisc. The LolC×RcsF and the LolE-LolC×RcsF adducts were detected by immunoblotting. (D) The in vitro lipoprotein transport assays. To break intermolecular disulfide bond, the nanodisc-embedded RcsF-LolC A106C DE S173C protein was incubated with TCEP prior to the addition of LolA  chromatography profile (S11A Fig). These mutations (the sole exception being M267D of TM2 LolE ), however, all have their substrate entry severely affected, as inferred from no detectable LolC×RcsF adducts by photo-crosslinking ( Fig 3D). Consistent with this, complementation assays also showed that these lolCDE mutants supplement failed to fully restore the growth of the lolCDE-depleted E. coli to a varied degree (Figs 3E and S10B). These findings highlight the functional importance of 3 acyl chain-interacting residues in LolCDE and, in turn, corroborate our structural observations.
Of the 14 visible residues of RcsF in the RcsF-LolCDE structure, only 3 residues, Cys+1, Ser +2, and Met+3, are enclosed by the V-shaped cavity of LolCDE (S13A Fig). The remaining residues adopt a stretched loop conformation protruding from the cavity on the Loop LolC side ( Fig 3B). First, we tested the functional importance of Met+3-interacting residues in LolC. As shown in Fig 3G, we found that mutations (M48D, F51D, L55D, and V260D of LolC) did not affect the complex assembly (S11B Fig). However, they severely interfered with the entry of RcsF and failed to rescue the growth of the lolCDE-depleted E. coli (Figs 3H and S10C). Next, we tested whether our structural observations agree well with the Lol avoidance signal hypothesis. In line with the hypothesis, we found that S21D (+2 position), S21E (+2 position), as well as a series of hydrophobic-to-hydrophilic mutations at the +3 position of RcsF either abolished or severely affected RcsF entry into the V-shaped cavity of LolCDE ( Fig 3I).
Taken together, our structural observations and functional studies reveal a bipartite binding mode of a lipoprotein to LolCDE. The 3 acyl chains and the first 3 residues of a mature lipoprotein dictate its binding affinity to LolCDE and substrate selectivity, respectively, thereby providing a structural explanation for the Lol avoidance signal hypothesis.

Features of the substrate-binding cavity of LolCDE
A prominent feature of the substrate-binding cavity of LolCDE is its negatively charged nature. This property is conferred by 6 acidic residues Glu54, Glu255, Glu263, and Asp352 of LolC together with Asp264 and Asp364 of LolE ( Fig 4A). Photo-crosslinking and complementation assays both showed that only 2 of the 6 mutations, E263Q LolC and D264N LolE , abolished RcsF entry (Figs 4B and S11) and failed to rescue the growth of the lolCDE-depleted E. coli strain (Figs 4C and S10D). A close examination of the RcsF-LolCDE structure reveals that the side chains of Glu263 LolC and Asp264 LolE clamp Cys+1 of RcsF from opposite directions and are located right in the center of the cavity (Figs 4D and S13B). Importantly, the distance between the 2 negatively charged groups is only 8.0 Å (Fig 4D). Representative single-point mutations (E263A, E263S, E263F, E263D, and E263K) of LolC abolished RcsF entry (Figs 4E and S11D) and failed to rescue the growth of the lolCDE-depleted E. coli (Figs 4F and S10E). However, for single-point mutations (D264A, D264F, D264E, and D264K) of LolE, only mutations D264F and D264K appeared to be lethal (Figs 4F and S10E), but they all appear to severely interfere with RcsF entry (Figs 4E and S11D). In accordance with these findings, sequence alignment analysis pinpoints Glu263 that is identical in all LolC homologues, with Asp264 highly conserved among LolE homologues (S14 and S15 Figs). The negatively charged nature of the cavity, in particular, the 2 negatively charged residues, Glu263 LolC and Asp264 LolE , which Residues shown as stick model were substituted with Asp for functional assays. (D) Photo-crosslinking assessing the importance of residues of LolCDE that interact with 3 acyl chains of RcsF in (C). rcsF were coexpressed with lolCDE mutants. (E) Complementation assay for the lolCDE mutants in (C). Protein leaky expression levels of the lolC and lolE mutants were detected by western blotting (bottom). (F) Zoom-in view of the hydrophobic interactions between Met+3 of RcsF and residues of LolCDE. (G and H) photo-crosslinking (G) and complementation assays (H) assessing the functional importance of hydrophobic interactions between Met+3 and residues of LolCDE. (I) UV-dependent crosslinks between LolCDE and RcsF variants were detected by immunoblotting. Data shown in (D and E) and (G to I) are representatives of 3 replicates. https://doi.org/10.1371/journal.pbio.3001823.g003 precisely clamp Cys+1 of RcsF, may also play a role in precluding the entry of phospholipids and LPS into the lipoprotein-binding cavity from the IM, for that the negatively charged PO 4 − groups of phospholipids and LPS in the membrane are similarly positioned as Cys+1 of a lipoprotein [40,[44][45][46][47][48] (S13C Fig).
A second distinct feature of the substrate-binding cavity is that the U-Loop from the Loop LolE plugs into the TM1 LolE -TM2 LolC interface, making hydrophobic contacts with residues from both TM1 LolE and TM2 LolC (Fig 5A). By contrast, in the AMPPNP-LolCDE structure, the substrate-binding cavity is compressed with the U-Loop completely squeezed out (S8C Fig). As the sequences of both LolC homologues and LolE homologues are highly conserved (S14 and S15 Figs), and both LolC and LolE adopt a rather similar fold (S7B and S7C Fig), we speculated that the U-Loop, which is absent in Loop LolC , may participate in lipoprotein entry. To test this hypothesis, we deleted 6 residues of the U-Loop (residues Ser363-Ile368 of LolE) and tested the entry of RcsF into the V-shaped cavity. As shown in Fig 5, the deletion mutant abolished RcsF entry. By contrast, a 7-residue deletion (residues Val349-Ala355 of LolC) in Loop LolC showed no noticeable effects on RcsF entry (Figs 5B and 5C and S10F and S11E). Furthermore, any hydrophobic-to-hydrophilic mutations (I365D, Y366D, F367D, and I368D) in the U-Loop also caused LolCDE loss-of-function to varied degrees (Figs 5D and 5E and S10 and S11). These results argue for the U-Loop to play a role in maintaining the configuration of the V-shaped cavity.

A single path for lipoprotein entry into the substrate-binding cavity
As mentioned above, the substrate-binding cavity of apo-LolCDE has 2 intermolecular interfaces, the Loop LolC -TM2 LolE interface (denoted as Interface I) and the Loop LolE -TM2 LolC interface (denoted as Interface II) (Fig 6A), which can both serve as potential gates for lipoprotein entry. To figure out the exact entry route, we introduced intermolecular disulfide bonds to block 1 gate and probed the entry of RcsF via the other. In each case, however, we obtained only partial formation of disulfide bonds in the LolCDE samples; we therefore proceed focusing on the crosslinking of RscF and LolC-LolE, those crosslinks that do contain intermolecular disulfide bonds between LolC and LolE. A potential disulfide bond between E255C LolC and S362C LolE in LolCDE was first introduced in hope to prevent RcsF entry via the Interface II, and SDS-PAGE analysis revealed that, though not complete, disulfide bonds were formed in the LolC L256pBPA+E255C DE S362C complexes (S13D Fig). After reconstitution into nanodisc together with Flag-tagged RcsF, the LolE-LolC×RcsF adducts were clearly detected under nonreducing condition upon UV radiation (Figs 6B and S16 and S1 Data). This demonstrated that formation of intermolecular disulfide bonds in the LolCDE complexes blocks the Interface II; however, the RcsF entry into the V-shaped cavity is unaffected. This strongly suggests that Interface I serves as the substrate entry route. Similarly, disulfide bond between L350C LolC and R263C LolE was introduced in hope to block Interface I. Again, intermolecular disulfide bonds formation was incomplete in the LolC L256pBPA+L350C DE R263C complexes (S13D Fig), and no LolE-LolC×RcsF crosslinks were detected after in vitro photo-crosslinking ( Fig 6B). Absence of LolE-LolC×RcsF crosslinks strongly implicates that RcsF is unable to enter the cavity with Interface I blocked. Base on the results, we propose that lipoproteins enter the cavity via Interface I rather than Interface II of LolCDE. These findings correlate well with our previous claim that the U-Loop in the Loop LolE takes part in buttressing the substrate-binding cavity in an outward conformation, and with the structural observation that the proteinaceous portion of RcsF is only located on the Loop LolC side in the RcsF-LolCDE structure. Results from our experiments are self-consistent and in line with structural observations.

Lipoprotein entry is ATP-independent, but its transport to LolA requires ATP hydrolysis
To investigate which step of the lipoprotein transport cycle requires ATP, we performed in vitro lipoprotein transfer assays under different conditions. First, we probed how ATP affects the entry of RcsF into the V-shaped cavity by photo-crosslinking. Apparently, addition of ATP, regardless of incubation time, did not increase the amounts of the LolC×RcsF adducts ( Fig 6C). The relatively reduced LolC×RcsF adducts upon addition of ATP may result from the formation of the ATP-bound LolCDE complexes that had precluded the entry of RcsF as indicated by the AMPPNP-LolCDE structure. Importantly, a catalytically dead mutant LolC-D E171Q E allowed the entry of RcsF (Fig 6C), strongly supporting the notion that lipoprotein entry is ATP-independent. Next, we incorporated pBPA at Trp70 that is located at the interior of LolA (S13E Fig) and performed lipoprotein transfer assays ( Fig 6D). Clearly, LolA (W70pBPA) was able to crosslink to RcsF only in the presence of ATP and Mg 2+ (Fig 6E). Conditions that interfere with ATP hydrolysis, e.g., addition of EDTA, VO 4 − or nonhydrolysable AMPPNP, as well as the catalytically dead mutant LolCD E171Q E protein, all failed to produce LolA×RcsF adducts (Fig 6E). However, it appears that the ATPase activity of LolCDE decreases with the addition of lipoproteins, which can be stimulated by LolA (S8E Fig and S1 Data). Taken together, our studies demonstrate that lipoprotein transfer to LolA from LolCDE requires ATP hydrolysis, and efficient substrate release increases the ATPase activity of LolCDE.

Discussion
On the basis of the structural and functional results reported here, we proposed a working model for lipoprotein transport by LolCDE (Fig 7). First, apo-LolCDE embedded in the inner membrane adopts an outward-facing conformation. The open conformation of apo-LolCDE allows lipoprotein entry into the substrate cavity via the Loop LolC -TM2 LolE interface (i.e., the Interface I) in an ATP-independent manner. The bound lipoprotein has its first 3 residues encompassed by the substrate-binding cavity, the remainder bulging into the periplasm on the Loop LolC side. Second, ATP binding causes a tight dimerization of LolD and closure of Vshaped cavity, pushing the bound lipoprotein out of the cavity. The final step involves ATP hydrolysis followed by conformational changes that enable lipoprotein transfer from LolCDE to LolA. Upon release of the LolA-lipoprotein complex and ADP, LolCDE returns to its apostate primed for the next transport cycle. While LolCDE, a type VII ABC transporter, differs from the type VI ABC transporter LptB 2 FG in both substrates and transmembrane topologies [39,40,42], they do share similarities in a number of aspects. First, lipoproteins enter and lodge into the substrate-binding cavity requiring no input of energy. Rather, substrate delivery to LolA downstream in the transport cycle is driven by the energy from ATP hydrolysis. Second, substrate binding does not cause any conformational changes of the ABC transporter. Third, substrate entry into the binding cavity proceeds via one particular side of the TMD interface of the ABC transporter. The combined structural evidences bear strong implications that underlines a mechanotransmission mechanism utilized by LolCDE for substrate extrusion and delivery, similar to the type VII ABC transporter MacB [39,40,42] (S17 Fig). In this regard, it is interesting to note that the remarkable resemblance of the overall architectures of apo-and nucleotide-bound LolCDE and MacB, the only exception being that apo-MacB lacks the V-shaped substrate-binding cavity and a central channel through which substrates take path [37,38].
Collectively, here, we report 3 LolCDE structures, apo, RcsF-, and AMPPNP-bound states. While the latter two correlate well with the previously published structures reported by Tang and colleagues and Sharma and colleagues, the former deviates significantly from the apo-LolCDE � structure reported by Tang and colleagues [33,34]. The functional analysis confirms our apo-LolCDE structure, highlighting that lipoproteins enter and bind in the cavity of apo-LolCDE devoid of apparent conformational changes in an ATP-independent manner. Furthermore, our high-quality structure of RcsF-LolCDE reveals a bipartite binding mode of a lipoprotein with LolCDE, as well as a functional role of the U-loop in maintaining the configuration of the V-shaped cavity. In particular, we identified a single path of lipoprotein-entry into the LolCDE. Taken together, our results provide deep insights into the mechanisms underlying LolCDE-mediated lipoprotein sorting in E. coli. Since lipoprotein sorting is essential for the survival of gram-negative bacteria [10], the RcsF-LolCDE structure we presented here also promises to guide novel antibiotic therapies fighting against drug resistant pathogens.

Cloning, expression, and purification of the LolCDE, RcsF-LolCDE, and LolCD E171Q E complexes
The 3 gene fragments containing lolC with NcoI/XbaI, lolD with XbaI/KpnI, and lolE with KpnI/HindIII were amplified individually from E. coli K-12 MG1655 genomic DNA by PCR. To obtain the RcsF-LolCDE complex, the gene fragment containing rcsF was amplified from E. coli K-12 MG1655 genomic DNA and ligated into pET28a vector. Plasmids pBAD22-LolCDE and pET28a-RcsF were then cotransformed into E. coli BL21 (DE3) for overexpression. Expression and purification of the RcsF-LolCDE and the catalytically dead LolCD E171Q E complexes followed a similar protocol for the LolCDE complex as described above.

Nanodisc reconstitution
POPG (Avanti Polar Lipids) was solubilized in chloroform and dried under nitrogen to form a thin lipid film. The lipid film was hydrated and resuspended at a concentration of 25 mM POPG in 250 mM sodium cholate. LolCDE, LolCD E171Q E, or RcsF-LolCDE complex, MSP1D1 membrane scaffold protein [49], and POPG were mixed at a molar ratio of 1:2.4:80 in a buffer containing 15 mM sodium cholate and incubated for 30 min at 4˚C. Detergents were removed by incubation with 0.8 mg/ml Bio-Beads SM2 (Bio-Rad) overnight at 4˚C. Nanodisc-embedded LolCDE, LolCD E171Q E, and RcsF-LolCDE complexes were further purified using a Superose6 increase 10/300GL column (GE Healthcare) in a buffer containing 20 mM Tris-HCl (pH 8.0) and 150 mM NaCl.

Cryo-EM specimen preparation, data acquisition, and processing
To prepare samples for cryo-EM analysis, 3 μl of nanodisc-embedded LolCDE, LolCD E171Q E, or RcsF-LolCDE complex at a concentration of 0.6 mg/ml was applied to glow-discharged Quantifoil holey carbon grids (2/2, 300 mesh). For AMPPNP/Mg 2+ trapping, the samples were incubated with 2 mM AMPPNP (Sigma) and 2 mM MgCl 2 for 30 min at 4˚C before applying the samples to cryo-EM grids. Grids were blotted for 5 s with 75% relative humidity and plunge-frozen in liquid ethane cooled by liquid nitrogen using automatic plunging device EMGP (Leica). Cryo-EM data set of LolCDE and RcsF-LolCDE complexes were collected in a 300-KV Titan Krios G2 microscope (Thermal Fisher) equipped with a direct detector K2 camera (Gatan) and GIF quantum energy filter (energy width of 20 eV). The data set of AMPPNPbound LolCD E171Q E was collected in a 200-KV Talos Arctica microscope (Thermal Fisher) equipped with direct detector K2 camera (Gatan) and GIF quantum energy filter (energy width of 20 eV) using beam-image shift data collection method [50]. Micrographs of LolCDE, RcsF-LolCDE and LolCD E171Q E were recorded at a defocus range of −1.8 to −2.2 μm with a pixel size of 1.36 Å, 1.04 Å, and 1.0 Å, respectively. The beam-induced motion of each micrograph stack was corrected by MotionCor2 [51], and the defocus parameter of each micrograph was determined by CTFFIND4 [52]. For LolCDE data set, 134,138 particles were picked from 300 micrographs using the Laplacian-of-Gaussian method in Relion-3.0. Five good classes were selected after 2D classification and were used as the references for automatic particle picking from all micrographs. A total of 2,666,570 particles were picked and subjected to 2D classification. After 3 rounds of 2D classification, 976,646 particles were kept for further data processing. The initial model for 3D classification was generated from selected particles using the SGD initial model generation program in Relion-3.0. After 3D classification, 503,068 particles were selected and subjected to 3D refinement, CTF refinement, and particle polishing, yielded a 4.0-Å EM map in Relion-3.0. For RcsF-LolCDE and AMPPNP-LolCD E171Q E data sets, a similar process procedure was applied in Relion-3.0 [53]. Then, 225,933 and 277,296 good particles were transferred to CryoSPARC-2.14 [54]. The 3.5-Å and 3.6-Å maps for RcsF-LolCDE and AMPPNP-LolCD E171Q E were, respectively, generated after NU-refinement, particle subtraction, and local refinement with a soft mask on protein parts.
To obtain the apo-LolCDE map from LolCDE dataset, 503,068 particles that yielded 4.0-Å map were subjected to further multireference 3D classification with references generated from this map by multiply soft masks that includes lipoprotein densities or not. The references were initial low-pass filtered to 15 Å and an angular sampling of 3.7˚was combined with local angular searches in the classifications. After several rounds of 3D multireference classification, 135,391 particles yielded a 4.2-Å apo-LolCDE map with no lipoprotein densities in the cavity. The other 367,675 particles yielded a 4.1-Å lipoprotein-LolCDE map with strong densities of lipoproteins in the cavity. To avoid reference bias, the reclassified maps were all reconstructed with the alignment file of the previous 4.0-Å map. For validation, we generated 2 references that one included no densities of lipoprotein and the other included the densities of proteinaceous parts but not the 3 acyl chains of lipoprotein, then reconstructed reclassified particles after classification with the 2 references to generate 2 maps. The densities of proteinaceous parts and the 3 acyl chains of lipoprotein are all still in 1 map and not in the other one (S2G Fig). In the same way, we also did classifications with 2 references that one of them lacked the densities of Loop LolE or partial TM2 LolE . The maps generated from the 2 different classifications both have the densities lacked in references (S2H and S2I Fig). Taken together, this supported that there was no reference bias in our reclassification method and the map of apo-LolCDE was correct.

Model building and refinement
To build the model of the RcsF-LolCDE complex, the crystal structures of the PLD of LolC from E. coli (PDB 5NAA), the PLD of LolE from Acinetobacter baumannii (PDB ID: 5UDF), and LolD from Aquifexaeolicus (PDB ID: 2PCL) were fitted into the cryo-EM map in UCSF Chimera [55]. After changing all amino acids to the correct E. coli sequence and manually building the TMDs of LolC and LolE in Coot [56], the initial model was refined using real space refinement in PHENIX and manually adjusted in coot. For building the model of AMPPNP-LolCD E171Q E, the previous structure of RcsF-LolCDE was used as the starting model and fitted into the 3.6-Å map by model morphing in coot [57]. Two AMPPNP molecules and Mg 2+ are manually fitted into the 2 additional densities at the LolD molecules. The model was refined using real space refinement in PHENIX, subsequently. The atomic model of the 4.2-Å apo LolCDE was built using the structure of RcsF-LolCDE as a starting model, and the RcsF atoms that lacked relevant density were manually deleted. The structural model was refined using real space refinement in PHENIX [58]. Cryo-EM data collection and refinement statistics are summarized in S1 Table.

ATPase activity assay
The ATPase activity assays were performed using Malachite Green Phosphate Assay Kits (Bioassay Systems). The phosphate standards and blank control for colorimetric detection were prepared according to the manufacturer's instructions. Approximately 0.2 μM LolCDE, RcsF-LolCDE, or LolCDE mutant protein in nanodisc were incubated with 2 mM ATP and 2 mM MgCl 2 at room temperature for 15 min directly or incubated with LolA at room temperature for 10 min before incubated with ATP. Since the ATP concentration should be lower than 0.25 mM, the mixture was diluted 1:10 in buffer containing 20 mM Tris-HCl (pH 8.0), 150 mM NaCl. Each 80 μL sample solution was transferred into a separate well of the plate, and 20 μL of the working reagent was added. The mixtures were incubated for 30 min at room temperature for color development. The absorbance at 620 nm was measured. The ATPase activities of all samples were determined using the mean value of the samples according to linear regression for the standards. All experiments were repeated at least 3 times.

Detection of RcsF-LolCDE interactions by photo-crosslinking
E. coli BL21 (DE3) strains were cotransformed with pSup-BpaRS-6TRN22 and pBAD22-LolCDE (or lolCDE mutant), and pET28a-RcsF with a C-terminal Flag tag (or rcsF mutant). An amber (TAG) codon was introduced at L256 LolC in pBAD22-LolCDE. Cells were grown at 37˚C in LB containing 30 μg/ml ampicillin sodium, 15 μg/ml kanamycin, and 15 μg/ml chloramphenicol in dark. When the cultures reached OD 600~1.0, 0.5 mM pBPA was added into LB. After 1 h, protein expression was induced by the addition of 0.1 mM IPTG and 0.05% Larabinose at 18˚C for 14 h. Purification of RcsF-LolC L256pBPA DE complex was performed using a similar protocol for LolCDE. Protein samples were concentrated to 0.6 mg/ml using a 100-kDa cutoff spin concentrator (Millipore), then either used for western blot analysis directly or exposed to UV light (365 nm, 100 W; Thermo Fisher Scientific) for 10 min before western blot analysis.

In vitro photo-crosslinking assay
pET28a-RcsF with a C-terminal Flag-tag was transformed into E. coli BL21 (DE3) strain for RcsF expression. The bacterial cells were grown at 37˚C in LB with 35 μg/ml kanamycin until the optical density of the culture reached 1.0 at 600 nm. Protein expression was induced with 0.1 mM IPTG at 25˚C for 12 h. Purification of RcsF was performed using a similar protocol for LolCDE, with slight modifications. The supernatants were incubated with anti-Flag affinity gel (YESEN) and eluted with wash buffer containing 100 μg/ml 3×Flag-tag peptide (YESEN). The protein samples were concentrated to 0.4 mg/ml using a 30-kDa cutoff spin concentrator. LolC L256pBPA DE or cysteine mutations were purified using a similar protocol for LolCDE. RcsF, LolC L256pBPA DE or cysteine mutations, MSP1D1 membrane scaffold protein and POPG were mixed at a molar ratio of 1:1:2.4:80. The nanodisc-embedded complexes were applied to a Superose6 increase 10/300GL column preequilibrated in buffer containing 20 mM Tris-HCl (pH 8.0) and 150 mM NaCl. Peak fractions were combined and concentrated to 0.6 mg/ml, then either used for western blotting directly or exposed to UV light for 10 min before western blotting. For western blotting analysis, samples were mixed with SDS loading buffer and analyzed by SDS-PAGE and western blotting with anti-Flag antibody as described above, but β-ME was not added to samples.

In vitro lipoprotein transfer assay
To purify LolA(W70pBPA) for lipoprotein transfer assay, plasmids pBAD22-LolA (W70pBPA) with a C-terminal His-tag and pSup-BpaRS-6TRN22 were cotransformed into E. coli BL21 (DE3) strain. The bacterial cells were grown at 37˚C in LB with 100 μg/ml ampicillin sodium and 15 μg/ml chloramphenicol in dark. When the cultures reached OD600~0.8, 0.5 mM pBPA was added into LB. After 1 h, protein expression was induced with 0.05% L-arabinose at 25˚C for 12 h. After centrifugation, cells were resuspended in buffer containing 20 mM Tris-HCl (pH 8.0), 150 mM NaCl, 20 mM imidazole and lysed by sonication. The supernatants were collected after centrifugation at 18,000 rpm for 1 h at 4˚C and incubated with preequilibrated Ni-NTA agarose beads for 1 h at 4˚C. After rinsing the Ni-NTA agarose beads with wash buffer containing 20 mM Tris-HCl (pH 8.0), 150 mM NaCl, and 50 mM imidazole, the LolA (W70pBPA) protein was eluted using wash buffer containing 300 mM imidazole. The protein samples were concentrated to 0.5 mg/ml using a 30-kDa cutoff spin concentrator.

Construction of the lolCDE-depleted E. coli strain
The lolCDE-depleted strain was constructed based on E. coli strain BW25113 with deletion of lpp gene and integration of a terminator and an arabinose-inducible araBAD into the chromosome prior to the lolCDE locus, using CRISPR-Cas9 system [59]. InAU : Pleasecheckandconfirmthatthee brief, a 20-nt single guide RNA (sgRNA) targeting lpp gene was introduced in pTargetF plasmid. The BW25113 strain carrying pCas plasmid was cotransformed with the pTargetF-lpp and a donor DNA fragment that including 300 nt from each side of lpp gene. The lpp-deleted strain BW25113-DL was selected by PCR and confirmed by sequencing. As the same strategy, a 20-nt sgRNA targeting the lolCDE gene was introduced in pTargetF plasmid. An rrnB_T1-rrnB_T2 terminator was inserted in front of ParaBAD, and the donor DNA fragment carrying rrnB_T1-rrnB_T2-Para-BAD and 500 nt homologous nucleotides was cotransformed into BW25113-DL carrying pCas and pTarget-lolCDE. The lolCDE-depleted strain BW25113-DLDL was selected by PCR and confirmed by sequencing.

Complementation assay
The lpp-deleted and lolCDE-depleted BW25113-DLDL cells harboring either the pQLink-lolCDE plasmid [60] or a plasmid encoding a LolCDE mutant protein were plated on LB-agar plates with 100 ug/ml ampicillin and 0.2% L-arabinose. Single colony was picked up and inoculated into 5 mL LB with 100 ug/ml ampicillin and 0.05% L-arabinose. When cells grew to a density of OD600~1.0, they were collected by centrifugation and washed with 5 mL LB medium twice to remove L-arabinose. The washed cells were resuspended in different volumes of LB medium to ensure that each sample has the same starting cell density of OD600~0.5 and was subsequently serially diluted (1:10, 1:100, 1:1,000, 1:10,000, and 1:100,000), and 2 μl of each dilution was spotted on LB plates containing 100 ug/ml ampicillin with or without 0.2% L-arabinose. The BW25113-DLDL cells transformed with the empty plasmid pQLinkN were used as a negative control, while the cells harboring the wild-type pQLink-lolCDE plasmid was used as a positive control.
Western blot was performed to compare the protein expression levels of the wild-type lolCDE and its mutants. LolC and LolE were tagged with Strep and Flag, respectively. Anti-Strep II tag antibody (1:3,000) and anti-Flag antibody (1:3,000) were used to detect LolC and LolE with HRP-conjugated anti rabbit antibody (1:5,000), respectively. Complementation assays were performed in triplicate and a representative result is shown.   (XLSX) S1 Raw Images. Uncropped images of SDS-PAGE gels and western blots of Figs 2B-2E and 2D and 2E and 3G-3I and 4B and 4C and 4E and 4F and 5B-5E and 6B and 6C and 6E and S1A-S1D and S11A-S11F and S12A-S12F and S13D and S13F.