Structure of a nascent membrane protein as it folds on the Bam complex

Abstract

Mitochondria, chloroplasts, and Gram-negative bacteria are encased in a double layer of membranes. The outer membrane contains proteins with a β-barrel structure^1,2. β-barrels are sheets of β-strands wrapped into a cylinder with the first strand hydrogen-bonded to the last strand. Conserved multi-subunit molecular machines fold and insert these proteins into the outer membrane^3–5. One subunit of the machines is itself a β-barrel protein that plays a central role in folding other β-barrels. In Gram-negative bacteria, the β-barrel assembly machine (Bam) consists of the β-barrel protein BamA and four lipoproteins^5–8. To understand how the Bam complex accelerates folding without using exogenous energy (e.g., ATP)⁹, we trapped folding intermediates on the machine. We report here the structure of the Bam complex folding BamA itself. The BamA catalyst (BamA^M, for BamA^machine) forms an asymmetric hybrid β-barrel with the BamA substrate (BamA^S). The N-terminal edge of BamA^M has an antiparallel hydrogen-bonded interface with the C-terminal edge of BamA^S, consistent with previous crosslinking studies^10–12; the other edges of BamA^M and BamA^S are close to each other but curl inward and do not pair. Six hydrogen bonds in a membrane environment make the interface between the two proteins very stable. This stability allows folding but creates a high kinetic barrier to substrate release once folding has finished. Features at each end of the substrate overcome the barrier and promote release by stepwise exchange of hydrogen bonds. This mechanism of substrate-assisted product release explains how the Bam complex can stably associate with the substrate during folding and then turn over rapidly when folding is complete.

Structures of the Bam complex from bacteria have provided clues to the mechanism of substrate folding^13–17. A striking finding in some of these structures is that the BamA β-barrel component of the machine (Fig. 1a) is present in an open state with unpaired N- and C-terminal β-strands^13–15,17. This open seam was proposed to pair with substrate β-strands by hydrogen bonding to facilitate β-sheet formation, leading to a general model in which folded portions of the substrate undergo “budding” through the open seam into the membrane^13,18–20. Crosslinking experiments have provided support for interaction of β-barrel substrates with the edges of the open seam^10–12, but no model for folding has addressed how the Bam complex catalyzes rapid and repeated assembly of substrates into β-barrels in the absence of an exogenous source of energy^9,21,22. Paired β-strands in a membrane are so stable that the kinetic barrier to disrupt them is enormous. To explain how the folding process is catalyzed requires structures of folding intermediates trapped on the Bam complex.

Figure 1 |. Deletion of extracellular loops causes substrates to stall at different stages of folding on the Bam complex.

a, Overview of β-barrel assembly by the Bam complex in Gram-negative bacteria. b, Schematic of engineered BamA substrates that contain extracellular loop deletions (BamA^S-ΔL). The residues deleted in each substrate are indicated. For simplicity, only the β-barrel domain of the substrate is shown. c, Expression levels of 6×His-tagged substrates in strains with or without degP. α-RpoA immunoblots are provided as loading controls. d, Quantification of immunoblotting data shown in c, with expression levels for each substrate calculated as the percent that remains when degP is expressed. The plotted data represent mean ± SEM from quantification data in c and additional independent replicates (n=6 for WT, n=3 for other substrates). e, Urea extraction of 6×His-tagged BamA^S-ΔL substrates. Samples from total cell lysates (top), membrane fractions before urea incubation (middle), and membrane fractions that remained after urea incubation (bottom) were analyzed. α-LptF immunoblots are provided as loading controls. Data shown are representative of results from two biological replicates. f, In vivo photocrosslinking of 6×His-tagged full-length BamA^M(S439pBPA) to 3×FLAG-tagged substrates. The strain lacking degP is used to ensure equal expression levels for all mutants, as shown in c. POTRA domains 3, 4, and 5 (Δ172–421) are removed in the substrates to prevent them from forming Bam complexes if they complete folding. Data shown are representative of results from three biological replicates. g, Cryo-EM map of substrate-bound Bam complex colored by local resolution.

Capturing substrates on the Bam complex

We developed an approach to trap partially folded substrates on the Bam complex of Escherichia coli. To do so, we selected BamA as a substrate of the Bam complex, so that BamA within the machine (BamA^M) folds other copies of BamA. We generated a series of substrates (BamA^S), each of which lacked one of the eight extracellular loops (Fig. 1b and Extended Data Fig. 1a). We reasoned that loop deletions could prevent proper formation of β-hairpins, resulting in slowed folding. To test this possibility, we assessed susceptibility of the substrate variants to the periplasmic protease DegP to determine whether folding stalled in the periplasm before membrane insertion²³. Removal of loops within the C-terminal half of BamA^S resulted in DegP susceptibility, but removal of loops within the N-terminal half did not (Fig. 1c-d). Likewise, urea extraction experiments, which indicate whether a substrate is properly integrated in the membrane, showed that substrates lacking loops in the C-terminal half of the protein could be extracted by the denaturant, while substrates lacking N-terminal loops could not (Fig. 1e). Both experiments are consistent with the idea that the C-terminal half of the BamA^S β-barrel, but not the N-terminal half, is needed for integration into the membrane and hence is assembled early in the folding process.

To assess whether the substrates accumulate on the Bam complex during folding, we incorporated the photocrosslinkable amino acid para-benzoyl phenylalanine (pBPA) near the N-terminal β-strand of the seam in BamA^M (at residue S439). We have previously found that, upon irradiation with ultraviolet light, pBPA introduced at this position forms crosslinks to the C-terminal region of a different substrate, LptD, during its folding¹². Importantly, substrates that fold slowly form stronger crosslinks to BamA^M than wild-type substrates because they have a longer residence time. Here, we tested crosslinking from BamA^M (S439pBPA) to loop-deleted substrates; in these substrates, we also removed POTRA domains 3, 4, and 5 because, as we have shown, removal of these domains prevents BamA^S from forming BamABCDE complexes if it does complete folding²⁴. We found that loop-deleted substrates indeed crosslinked strongly to the BamA^M N-terminus (Fig. 1f). Taken together, these results show that removal of extracellular loops in BamA^S leads to accumulation of the substrates on the machine, but either outside or within the membrane, depending on the stage in the folding process at which they stall. Our results are consistent with earlier reports showing that deletions of the extracellular loops of BamA lead to impaired cell growth and lower protein levels^25,26.

Structure of a substrate-engaged Bam complex

We chose the substrate with loop 1 removed (BamA^S-ΔL1) for structural studies because it accumulates at a late stage of folding, when it has already integrated into the membrane but remains associated with the Bam complex. To stabilize the interactions between BamA^M and the folding substrate, we engineered a disulfide bond between the two proteins (Extended Data Fig. 1b-f and Extended Data Fig. 2a). The substrate used for structure determination appears capable of adopting a folded state under reducing conditions, although folding occurs slowly (Extended Data Fig. 2b). We purified the BamABCDE-BamA^S complex (Extended Data Fig. 2c-d) and determined its structure by cryo-EM to an overall resolution of 4.1 Å (Extended Data Fig. 3, Fig. 1g, and Supplementary Table 1). The cryo-EM map contains density for all five components of the Bam complex and for the substrate (Fig. 2a-b and Extended Data Fig. 4). All β-strands of the BamA^M β-barrel were resolved, as were all β-strands of the BamA^S β-barrel except β1 and β2. Extracellular loop 1, which we removed from the substrate, would normally connect these two β-strands. The POTRA domains of the substrate were also not visible in the structure.

Figure 2 |. The N-terminal β-strand of BamA^M interacts with the C-terminal β-strand of the substrate by β-sheet augmentation.

a, Side view of the cryo-EM reconstruction of the Bam complex bound to the substrate (BamA^S-ΔL1). BamA^M is colored in green, BamB in silver, BamC in wheat, BamD in pink, BamE in cyan, and the substrate in blue. This color scheme is maintained in figures showing the cryo-EM structure. Approximate boundaries of the outer membrane (OM) are shown. BamC and POTRA domains 1 and 2 of BamA^M are shown at a lower contour level than the rest of the structure. b, Top-down view turned 90 degrees from view in a. P2, P3, and P4 refer to the POTRA domains of BamA^M that are visible in this view. c, Side view of the atomic model presented in the same orientation as in a. d, Overlay of BamA^M from substrate-bound complex (green) and from a substrate-free, lateral-open complex (orange, PDB ID: 5EKQ). e, Enlarged view of boxed region from c. The cryo-EM map is shown in gray. f, Interactions between β1 of BamA^M and β16 of the substrate shown in stick representation. Hydrogen bonds between the β-strands are shown as dashed black lines, and the bond lengths (Å) are labelled.

Using previously published X-ray crystallographic structures^15,27, we built atomic models into our cryo-EM map (Fig. 2c). The overall architecture of the Bam complex in its substrate-bound state is similar to that seen in substrate-free states^14–17, with the organization of the periplasmic components most closely resembling that of substrate-free structures containing a laterally-open form of BamA (Extended Data Fig. 5). Structures of the individual components are likewise similar to those already published (Extended Data Fig. 6). There are, however, notable conformational differences between the N-terminal segment of the BamA^M β-barrel in our substrate-engaged complex and the corresponding segment in any of the substrate-free, lateral-open structures (Fig. 2d). Specifically, the first and second β-strands of BamA^M in the substrate-engaged β-barrel bend outward, separating them more from the C-terminal β-strand than they are in the substrate-free β-barrel. Smaller deviations in β-strands 3, 4, and 5 allow the larger changes in β1 and β2, but all remaining β-strands superimposed well on previous structures of BamA (Fig. 2d and Extended Data Fig. 6).

Like the BamA^M β-barrel, the membrane-embedded BamA^S-ΔL1 β-barrel has a lateral opening. The location of BamA^S in the membrane is consistent with our biochemical experiments (Fig. 1). Density was absent for β1 and β2 of the substrate, presumably because deletion of loop 1, which would normally connect these two β-strands, prevented proper formation of the corresponding β-hairpin (Extended Data Fig. 7). Additionally, the POTRA domains of the substrate could not be resolved. However, the rest of the β-barrel closely resembled the folded form of BamA, suggesting that the substrate in the structure is an on-pathway intermediate (Extended Data Fig. 7). The N-terminal β-strand of BamA^M and the C-terminal β-strand of BamA^S are paired in an interaction that involves six hydrogen bonds (Fig. 2e-f), creating a BamA^M-BamA^S hybrid barrel with a continuous lumen.

Asymmetric interface between BamA^M and BamA^S

Whereas the C-terminal β-strand of BamA^S is captured by the N-terminal β-strand of BamA^M, the β-strands at the other ends of BamA^S and BamA^M are not hydrogen bonded (Fig. 3a-c). Instead, these ends curve inwards, so that the barrels make contact along their exterior surfaces that would normally contact membrane lipids. Given that we used a disulfide bond to tether the C-terminal region BamA^M to the N-terminal region of BamA^S to stabilize the complex during purification, it was possible that the asymmetric interface we observe in our structure reflected a non-native conformation of the complex. To address this, we determined another structure with the same substrate but using a cysteine tether introduced between the other ends of BamA^M and BamA^S (Extended Data Fig. 8). This 6.5 Å structure is remarkably similar to our 4.1 Å structure (Extended Data Fig. 8i), indicating that using cysteine tethers does not alter the overall conformation of the complex.

Figure 3 |. The hybrid barrel is asymmetric with the C-terminal β-strand of BamA^M and the N-terminal β-strand of the substrate unpaired in the membrane.

a, Side view turned 180 degrees from view in Fig. 2c to show interactions between the C-terminus of BamA^M and the N-terminus of the substrate. b, Enlarged view of boxed region from a. The cryo-EM map is shown in gray. This view shows the “seal” formed by the interaction of these ends of BamA^M and the substrate. c, Top-down view of a slice through the cryo-EM map, as indicated in a. The N- and C-termini of BamA^M and the substrate and approximate positions of L3, L5, and L8 within the substrate are indicated. Approximate positions of pBPA substitutions are indicated with stars. d, In vivo photocrosslinking of Strep-tagged BamA^S-ΔL substrates containing pBPA at positions T467, Y531, M741, or F804 and deletion of POTRA domains 3–5. Immunoblotting was performed using α-His, α-Strep, α-LPS, and α-FLAG antibodies to detect BamD (loading control), the substrates, substrate-bound LPS, and BamA^M, respectively. For α-FLAG immunoblot, a longer exposure (top, to detect crosslinks) and a shorter exposure (bottom) are shown. Data shown are representative of results from two biological replicates. e, Schematic of substrates in which two extracellular loops are removed. The residues removed are as given in Fig. 1b. f, Expression levels of substrates containing two loop deletions in strains with or without degP. α-RpoA immunoblot is provided as a loading control. g, Quantification of immunoblotting data shown in f, performed as in Fig. 1d. The plotted data represent mean ± SEM from quantification of data in f and additional independent replicates (n=3 for each substrate).

The asymmetry of the hybrid β-barrel implies that side chains within BamA^S that are oriented outward should be exposed to lipopolysaccharide (LPS) molecules in the membrane except near its N-terminus, which should be inaccessible to LPS. We tested this implication by incorporating pBPA at four sites oriented toward the exterior of BamA^S (Fig. 3c and Extended Data Fig. 9a-b), including at a site on the surface that curves inward (T467) in the hybrid β-barrel. After in vivo photocrosslinking, we purified the Bam-BamA^S complex and analyzed crosslinks from BamA^S to both LPS and BamA^M (Fig. 3d). The results showed that the T467pBPA substitution in BamA^S-ΔL1 yielded a crosslink to BamA^M, whereas the other substitutions (Y531pBPA, M741pBPA, F804pBPA) yielded crosslinks to LPS. These results agree with the finding from our structure that most of the substrate’s exterior surface faces the membrane, whereas the N-terminal β-strands interact with BamA^M.

β-strands 1 and 2 of the substrate are not visible in our structure. To determine their location, we substituted F428 and F440 in BamA^S-ΔL1 with pBPA. We observed crosslinks from the substrate to BamA^M (Extended Data Fig. 9c-d). Based on where β-strand 3 is located in the structure, and because the periplasmic loop from β-strand 3 to β-strand 2 is not long enough to reach the accessory Bam subunits, we infer that β-strands 1 and 2 likely interact with the BamA^M β-barrel or POTRA domain 5 of BamA^M.

Although there is an unpaired edge at one end each of BamA^M and of BamA^S, inward curvature ensures that these edges face into the lumen of the hybrid barrel, where they are solvated by water. The seal created by the hydrophobic interaction interface between BamA^M and BamA^S prevents entry of lipid molecules into the lumen (Fig. 3b-c). Early intermediates may not be able to curve around to form this interface, and indeed, we did not detect crosslinks to LPS using pBPA substitutions within BamA^S-ΔL5 or BamA^S-ΔL8 (Fig. 3d). This finding, combined with our results above, supports the notion that the polypeptide chain takes different routes during early and late folding rather than “budding” continuously from the BamA^M β-barrel and into the membrane^{11,20,23,28,29}. Sequential membrane insertion of hairpins would imply that the nonpolar side of any hairpin would always face lipid (Fig. 3c-d). Specifying the mechanism of early folding will require visualizing an early intermediate.

The interaction between the N-terminal β-strand of the BamA^M β-barrel with the C-terminal edge of the BamA^S β-barrel has direct implications for the folding mechanism. By binding the exposed N-terminal edge of BamA^M as an extended β-strand, the C-terminus of the substrate forms a new edge to template binding of the subsequent β-strand in a process known as β-strand augmentation^30–32. Therefore, folding will be directional, proceeding from the C-terminus to the N-terminus of the substrate. Based on crosslinking experiments, a sequential, C- to N-terminal folding model was proposed for the mitochondrial homolog Sam50¹⁰. We probed directional folding by generating six BamA^S variants in which two extracellular loops had been removed (Fig. 3e). In each construct, removal of the loop closer to the C-terminus determined the overall susceptibility to protease digestion, which reports on membrane integration (Fig. 3f-g). These differences in proteolytic susceptibility allow us to conclude that the region including loop 8 folds before that including loop 5, and so forth. This finding establishes C- to N-terminal directional folding and is consistent with sequential β-strand augmentation.

Mechanism of substrate release

The six hydrogen bonds at the interface between the N-terminal β-strand of the BamA^M β-barrel and the C-terminal β-strand of the BamA^S β-barrel (Fig. 2e-f) prevent premature product release, but the strength of that interaction raises the question of how release occurs at all. In a membrane, disrupting a hydrogen bond costs approximately 4 kcal/mol^33,34. Thus, the kinetic barrier to disrupting all six hydrogen bonds at once would be insurmountably high, even though the process is thermodynamically favorable.

Two unexpected features of the structure suggest a mechanism for substrate β-barrel closure and release. First, the edges of the two barrels at the folding end of the substrate do not engage in hydrogen bonding but instead turn inward toward the water-filled lumen of the hybrid barrel. Thus, no polar bonds to the machine need to be broken at that edge. Second, the register of the hydrogen bonds between the N-terminal β-strand of the BamA^M β-barrel and the C-terminal β-strand of the BamA^S β-barrel is not the same as it is between the N- and C-terminal β-strands of a folded BamA β-barrel^13,35. The hydrogen-bonded residues on the BamA^M side are largely the same as those at the seam of a closed β-barrel, but they interact with different residues in the C-terminal β-strand of the trapped substrate. In our structure, BamA^M splays open at the extracellular end of its N-terminal β-strand (Fig. 2d), facilitating pairing with residues in the C-terminal β-strand of BamA^S that precede those paired with the N-terminal β-strand in the fully folded structure (compare Figs. 2f and 4a). This pairing creates an overhang at the end of the BamA^S C-terminal β-strand (Fig. 4b-c) by steric clash with the loop between β-strands 14 and 15 of BamA^S. These observations suggest a stepwise model for how release occurs (Fig. 4d). When the N-terminal β-strand of BamA^S adds to the substrate β-barrel, it approaches the C-terminal overhang (Fig. 4d). The N- and C-termini of the substrate begin to form hydrogen bonds, sequentially disrupting bonds between the substrate and the machine. Unlike concerted disruption, sequential displacement is energetically feasible. As hydrogen bonds form to close the substrate β-barrel, the C-terminal β-strand of the substrate peels away stepwise from its interaction with the N-terminal β-strand of the BamA^M β-barrel. The accompanying relaxation of strain in the splayed N-terminal hairpin might accelerate the strand exchange.

Figure 4 |. Release from BamA^M requires structural features of the substrate N- and C-termini that allow stepwise hydrogen bond exchange.

a, Interactions between β1 and β16 in the folded form of BamA (PDB ID: 4N75). Residues G807 to W810, which form an overhang in Bam-bound BamA^S (see below) are colored in brown. b, Interaction between BamA^M and the substrate (left). The view is from the interior of the two β-barrels. Residues within the substrate C-terminal overhang are in brown. The final two residues of the substrate could not be resolved, and their expected location is indicated with a dashed brown line (right). c, View of BamA^M-substrate interactions turned 180 degrees from view in b and similar to the view in Fig. 2e. With the exception of the C-terminal β-strand (β16) and overhang of the substrate, proteins are shown in semi-transparent surface representation. Residues in BamA^M β1 and substrate β16 are labelled. d, Model of substrate release involving pairing between the C-terminal overhang and N-terminal β-strand of the substrate. The substrate C-terminal overhang is in brown, and residues that interact with BamA^M in the Bam-associated substrate are in yellow. e, Alignment of the C-terminal regions of BamA and other Omp85 superfamily members. The invariant glycine is highlighted in yellow. f, Expression levels of 6×His-tagged substrates in strains with or without degP. α-RpoA immunoblots are provided as loading controls. g, Quantification of immunoblotting data shown in f, performed as in Fig. 1d. The plotted data represent mean ± SEM from quantification of data in f and additional independent replicates (n=3 for each substrate). h, In vivo photocrosslinking of 6×His-tagged full-length BamA^M(S439pBPA) to 3×FLAG-tagged substrates with POTRA domains 3–5 removed. Data shown are representative of results from three biological replicates.

We tested the implication of the strand-exchange mechanism as follows. First, we ruled out a special role for an invariant glycine present in BamA (G807 in E. coli) and homologs in mitochondria and chloroplasts, just at the boundary of the C-terminal overhang in our structure (Fig. 4e). Changing G807 to valine has been reported to cause outer membrane defects³⁶. We made the G807V mutation to determine whether it prevents release of an otherwise wild-type BamA^S but found no defects in its assembly (Fig. 4f-h), as judged by DegP sensitivity and by crosslinking. These results indicate that the G807V substitution alters BamA function rather than assembly.

Second, we made substrate variants with proline substitutions at positions 807, 808 or 809 in the overhanging C-terminal segment. Proline distorts the conformation and hydrogen bonding in a β-sheet. DegP did not degrade these proline-substituted substrates (Fig. 4f-g), implying that they undergo membrane integration. In vivo photocrosslinking to BamA^M (Fig. 4h) showed that these substrates accumulate on the machine in a membrane-integrated state. Moreover, deleting the C-terminal residues (809–810) also caused late-stage assembly defects (Fig. 4f-h). These results suggest that forming hydrogen bonds between the C-terminal overhang and N-terminal β-strand of BamA^S facilitates release of BamA^S from BamA^M.

Discussion

The structure reported here supports the central idea of models in which substrates pass through the side of the BamA β-barrel into the membrane (Extended Data Fig. 10)^13,18–20. It is also consistent with crosslinking studies showing that the C-terminal β-strand of the folding substrate interacts strongly with the N-terminal edge of the seam of the BamA^M β-barrel^10–12. Moreover, its two unexpected features—the unpaired edges at one junction and the hydrogen-bond register and overhang at the other—resolve problems posed by models postulating hydrogen-bonded arrays at both edges of the machine. Sequential replacement of each substrate-machine hydrogen bond by a substrate-substrate hydrogen bond yields a succession of rapid steps rather than a single, slow one. We believe that the β-barrel closure and release mechanism described here is general: other Bam substrates could have C-terminal overhangs (Supplementary Table 2) that, when bound to the machine, would protrude inside the hybrid barrel to engage their own N-termini. With this information, it may be possible to design antibiotics that bind to features of BamA and inhibit β-barrel folding and release^37–40.

Methods

SDS-PAGE and immunoblotting.

Homemade Tris-HCl 4–20% polyacrylamide gradient gels or Mini-PROTEAN TGX 7.5% precast gels (BioRad) were used with Tris-glycine running buffer. 2× SDS sample buffer refers to a mixture containing 125 mM Tris (pH 6.8), 4% (wt/vol) SDS, 30% (vol/vol) glycerol, 0.005% bromophenol blue, and 5% (vol/vol) β-mercaptoethanol. SDS-PAGE analyses were performed at 200 V for 45 to 60 min. To analyze purified protein complexes for cryo-EM, SDS-PAGE was performed and followed by staining with Coomassie Brilliant Blue (Alfa Aesar). Coomassie-stained SDS-PAGE gels were imaged using the “Gel” feature of an Azure Biosystems C400 imager. For Western blotting, proteins were transferred onto Immun-Blot PVDF membranes (Bio-Rad) and then incubated with appropriate antibodies. All HRP conjugates were visualized with the Amersham ECL Prime Western blotting detection reagent (GE Healthcare). Western blots were imaged using the “lowest” sensitivity setting of the “Chemi” feature of an Azure Biosystems C400 imager. Uncropped immunoblots are available in Supplementary Figure 1.

Plasmids, strains, and oligonucleotides.

Plasmids, strains, and oligonucleotides used in this study are reported in Supplementary Tables 3, 4, and 5, respectively.

Analysis of cellular BamA levels.

Derivatives of a plasmid containing 6×His-BamA cloned into the pZS21 vector (pSK476) were generated to contain single loop deletions (pDT175-pDT182), double loop deletions (pDT260-pDT265), or C-terminal overhang mutations (pDT521, pDT536-pDT538, pDT550). MC4100 cells⁴¹ and MC4100 degP::cam cells²³ were transformed with these plasmids. The resulting strains were grown in LB supplemented with 50 μg/mL kanamycin (for MC4100) or 50 μg/mL kanamycin and 30 μg/mL chloramphenicol (for MC4100 degP::cam) (37°C, 220 rpm). Once an OD600 of ~1.0 was reached, the cells from a 1 mL sample were collected by centrifugation (5,000 x g, 10 min, 4°C). The resulting cell pellets were resuspended in 80 μL of a 1:1 mixture of 2× SDS sample buffer and buffer containing 20 mM Tris-HCl pH 8.0 and 150 mM NaCl. After boiling for 10 min, the samples were applied to SDS-PAGE and analyzed via Western blotting. 6×His-BamA was detected by using a penta-His (HRP) antibody (Qiagen). RpoA was detected using a mouse anti-RpoA (E. coli RNA Polymerase α) primary antibody (BioLegend, clone 4RA2) followed by a sheep anti-mouse (HRP) secondary antibody (GE Healthcare).

Membrane extraction with urea.

Membrane extraction was performed in a manner as previously described⁴². MC4100 degP::cam strains were generated harboring the pZS21 vector encoding 6×His-BamA or derivatives containing single or double loop deletions. These strains were grown overnight in LB supplemented with 50 μg/mL kanamycin and 30 μg/mL chloramphenicol (37°C, 220 rpm). These cultures were used to inoculate 100 mL of LB with the same additives via 1:100 dilutions. The resulting cultures were grown (37°C, 220 rpm) until an OD600 of ~1.0 was reached. The cells were harvested via centrifugation (5,000 x g, 10 min, 4°C) and resuspended in 4 mL of resuspension solution (50 mM Tris-HCl pH 8.0, 10 μg/mL DNase). Cells were lysed via sonication (on and off at 10 second intervals for a total of 3 min on and 3 min off). The resulting lysate was centrifuged to pellet cell debris (5,000 x g, 10 min, 4°C). The supernatant was used to prepare normalized cell lysate suspensions to an A280 value of ~24 (assayed by using NanoDrop). An aliquot of each normalized lysate sample was removed, mixed 1:1 with 2× SDS sample buffer (containing β-mercaptoethanol), and boiled for 10 min. For each sample, 800 μL of the remaining lysate was transferred to 70.1 Ti ultracentrifuge tubes (Beckman Coulter) along with an additional 5 mL of 50 mM Tris-HCl pH 8.0. Membrane fractions were then isolated via ultracentrifugation (100,000 x g, 25 min, 4°C) using an Optima XE-90 ultracentrifuge (Beckman Coulter). The supernatant was discarded, and the membrane pellet was resuspended in 800 μL of 6 M urea. The resuspensions were incubated at 37°C for 1 hour. An aliquot of each sample, representative of total membrane protein content, was removed, mixed 1:1 with 2× SDS sample buffer (containing β-mercaptoethanol), and boiled for 10 min. The post-wash membrane pellet was isolated from the remainder of each sample via ultracentrifugation (100,000 x g, 45 min, 20°C) after addition of 5 mL of 50 mM Tris-HCl pH 8.0. The pellets were resuspended in 500 μL of water. An aliquot of each sample was removed, mixed 1:1 with 2× SDS sample buffer (containing β-mercaptoethanol), and boiled for 10 min. The samples were applied to SDS-PAGE and analyzed via Western blotting. 6×His-BamA was detected by using a penta-His (HRP) antibody (Qiagen). LptF was detected using LptF antiserum followed by a donkey anti-rabbit (HRP) secondary antibody (GE Healthcare). The source of rabbit anti-LptF antiserum has been previously reported⁴³.

In vivo photocrosslinking of BamA to substrates.

Photocrosslinking experiments are based on techniques as previously described^44,45, with modifications. The plasmid pSup-BpaRS-6TRN encodes an orthogonal tRNA and aminoacyl-tRNA synthetase to incorporate the unnatural amino acid pBPA at amber (TAG) stop codons⁴⁶. This plasmid contains a chloramphenicol resistance cassette, so is not compatible with the MC4100 degP::cam strain. To circumvent this, we generated a variant of the pSup-BpaRS-6TRN plasmid that instead contains a spectinomycin resistance cassette. In brief, pSup-BpaRS-6TRN minus the chloramphenicol resistance cassette was amplified, and the spectinomycin resistance cassette was amplified from the pCDFDuet vector (EMD Millipore). The pSup-BpaRS-6TRN backbone and the spectinomycin resistance cassette were joined via Gibson assembly⁴⁷ to generate pSup-BpaRS-6TRN(spec^R) (pDT504).

6×His-BamA^M containing pBPA substitutions were cloned into the pZS21 vector to generate pJL77. 3×FLAG-tagged substrates were cloned into the pTrc99a vector to generate pDT209 (WT BamA^S); pDT201-pDT208 (single loop deletions); pDT268, pDT270, and pDT272 (double loop deletions); and pDT526, pDT558-pDT560, and pDT562 (C-terminal overhang mutants). The 3×FLAG-tagged substrates also contain deletion of POTRA domains 3, 4, and 5 (Δ172–421) to avoid the possibility of these substrates forming Bam complexes if they complete folding. The deletion does not prevent folding of an otherwise wild-type BamA²⁴, and allows us to assess BamA^S mutants in terms of their ability to fold rather than function within a Bam complex. In these constructs, POTRA domains 1 and 2 are retained since deletion of all five POTRA domains prevents an otherwise wild-type BamA from folding properly.

MC4100 degP::cam strains were generated, each harboring pSup-BpaRS-6TRN(spec^R), pJL77, and one of the substrate-encoding plasmids. These strains were grown overnight in LB supplemented with 50 μg/mL carbenicillin, 50 μg/mL kanamycin, 50 μg/mL spectinomycin, and 0.2% (w/v) glucose (37°C, 220 rpm). These overnight cultures were diluted 1:100 into 100 mL of fresh LB containing the same additives minus glucose but supplemented with 0.8 mM H-p-Bz-Phe-OH (pBPA, Bachem), and were grown (37°C, 220 rpm) to OD600 ~0.35. After normalization by optical density, each culture was split in half, with one half used directly for irradiation with UV light at 365 nm for 10 min (on ice). Photocrosslinking was performed by using a UVP Blak-Ray B-100AP high-intensity UV lamp with a 100 W spot bulb. All cells were then pelleted by centrifugation (5,000 x g, 10 min, 4°C).

For protein purification, pellets were resuspended in 5 mL TBS (20 mM Tris-HCl pH 8.0, 300 mM NaCl) supplemented with 1% (w/v) Anzergent 3–14 (Anatrace), 20 mM imidazole (pH 8.0), 100 μg/mL lysozyme (Sigma-Aldrich), 1 mM PMSF (Sigma-Aldrich), and 50 μg/mL DNase I (Sigma-Aldrich). Cells were lysed via sonication (on and off at 10 second intervals for a total of 1 min and 30 sec on and 1 min and 30 sec off). The resulting lysate was centrifuged (10,000 x g, 10 min, 4°C). An aliquot of the supernatant was taken for analysis of 3×FLAG-BamA^S expression levels via SDS-PAGE and immunoblotting using a monoclonal anti-FLAG M2-peroxidase (HRP) mouse antibody (Sigma-Aldrich). The remainder of the supernatant was incubated with Ni-nitrilotriacetic acid (NTA) resin on a rocking platform (1 hour, 4°C). After removal of unbound proteins, the resin was washed twice with 50 CV (column volumes) TBS containing 0.02% Anzergent 3–14 and 20 mM imidazole (pH 8.0). Samples were eluted with 5 CV TBS containing 0.02% Anzergent 3–14 and 200 mM imidazole (pH 8.0). Eluates were supplemented with 10% TCA by volume and incubated on ice for 20 min. Precipitated proteins were pelleted via centrifugation (21,130 x g, 10 min, 4°C). All samples were resuspended in a 1:1 mixture of 2× SDS loading dye (containing β-mercaptoethanol) and 1 M Tris pH 8.0. After boiling for 20 min, each sample was analyzed by SDS-PAGE and Western blotting. 6×His-BamA^M was detected by using a penta-His (HRP) antibody (Qiagen). 3×FLAG-BamA^S that was pulled down with 6×His-BamA^M was detected by using a monoclonal anti-FLAG M2-peroxidase (HRP) antibody (Sigma-Aldrich).

In vivo photocrosslinking of BamA to LPS.

6×His-BamA containing pBPA substitutions were cloned into the pZS21 vector to generate pDT411, pDT416, and pDT421-pDT430. MC4100 strains were generated, each harboring pSup-BpaRS-6TRN (containing a chloramphenicol resistance cassette) and one of the plasmids encoding pBPA-substituted BamA. These strains were grown overnight in LB supplemented with 50 μg/mL kanamycin and 30 μg/mL chloramphenicol. These overnight cultures were diluted 1:100 into 100 mL of fresh LB containing the same additives supplemented with 0.8 mM H-p-Bz-Phe-OH (pBPA, Bachem), and were grown (37°C, 220 rpm) to OD600 ~0.35. The protocols for UV irradiation and protein purification are the same as described above for in vivo photocrosslinking of BamA^M to substrates.

In vivo photocrosslinking of substrates to LPS or BamA^M, followed by DSP crosslinking to the Bam complex.

In order to observe crosslinking from substrates to lipopolysaccharide or to BamA^M, a separate crosslinking protocol was devised to ensure that only substrates stalled on the Bam complex were assessed. In brief, we crosslinked the exterior surface of substrates to their surroundings (using incorporated pBPA) and then crosslinked stalled substrates to the Bam complex (using an amine-to-amine crosslinker). This was followed by purification of Bam-substrate complexes and release of the substrate with reducing agent (which breaks the amine-to-amine crosslinks), allowing assessment of UV-dependent crosslinks by SDS-PAGE and Western blotting.

The pJH114 plasmid⁴⁸ was modified to encode 3×FLAG-BamA^M and BamD-8×His (while eliminating the 8×His tag on BamE), generating pDT340. 2×Strep-tagged BamA^S substrates containing deletion of POTRA domains 3–5, deletion of an extracellular loop (L1, L3, L5, or L8), and a pBPA substitution (at T467, Y531, M741, or F804) were cloned into the pCDF vector to generate pDT436-pDT439 (T467pBPA), pDT451-pDT454 (Y531pBPA), pDT471-pDT474 (M741pBPA), and pDT476-pDT479 (F804pBPA). BL21(DE3) strains were generated harboring pSup-BpaRS-6TRN, pDT340, and one of the substrate-encoding plasmids.

These strains were grown overnight in LB supplemented with 50 μg/mL carbenicillin, 50 μg/mL spectinomycin, 30 μg/mL chloramphenicol, and 0.2% (w/v) glucose (28°C, 220 rpm). These overnight cultures were diluted 1:100 into 100 mL of fresh LB containing the same additives, and were grown (37°C, 220 rpm) to OD600 ~0.35. At this point, isopropyl β-D-1-thiogalactopyranoside (IPTG, VWR) and 0.8 mM H-p-Bz-Phe-OH (pBPA, Bachem) were added to final concentrations of 0.1 mM and 0.8 mM, respectively. The strains were grown for an additional 90 minutes. After normalization by optical density, each culture was split in half, with one half used directly for irradiation with UV light at 365 nm for 10 min (on ice). All cells were then pelleted by centrifugation (5,000 x g, 10 min, 4°C). Pellets were resuspended in 20 mL PBS (20 mM NaH₂PO₄ pH 7.2, 150 mM NaCl) and the amine-to-amine crosslinker dithiobis(succinimidyl propionate) (DSP, Thermo Fisher Scientific) was added to a final concentration of 0.5 mM. After incubation on a rocking platform (30 to 60 min, room temperature), the crosslinking reaction was quenched via addition of Tris-HCl to a final concentration of 20 mM. Cells were centrifuged (5,000 x g, 10 min, 4°C) and the pellets were frozen at −80°C prior to subsequent purification.

Protein purification was performed similarly as described for testing site-specific in vivo crosslinking without DSP crosslinking, with a few differences. The Ni-NTA wash step was performed using 50 CV TBS containing 0.1% Triton X-100 (Sigma-Aldrich), 0.1% SDS (J. T. Baker), and 40 mM imidazole (pH 8.0). The Ni-NTA elution step was performed using 5 CV TBS containing 200 mM imidazole (pH 8.0) and no detergent. Instead of using TCA precipitation, each sample was concentrated using an Amicon Ultra 0.5 mL 10 kDa molecular-weight cutoff centrifugal concentrator (EMD Millipore). All samples were supplemented with an equal volume of 2× SDS loading dye (containing β-mercaptoethanol). After boiling for 10 min, each sample was analyzed by SDS-PAGE and Western blotting. 8×His-BamD was detected by using a penta-His (HRP) antibody (Qiagen). 2×Strep-BamA^S substrates were detected using a monoclonal Strep-Tag II (HRP) antibody (EMD Millipore). 3×FLAG-BamA^M was detected by using a monoclonal anti-FLAG M2-peroxidase (HRP) antibody (Sigma-Aldrich). Lipopolysaccharide was detected using a mouse monoclonal anti-lipopolysaccharide core primary antibody (Hycult Biotech, clone WN1 222–5) followed by a sheep anti-mouse (HRP) conjugate secondary antibody (GE Healthcare).

Assessing cysteine-to-cysteine crosslinking.

6×His-BamA^M containing a cysteine substitution was cloned into the pZS21 vector to generate pDT511-pDT513 (C-proximal cysteine mutant) and pDT400 (N-proximal cysteine mutant). 3×FLAG-tagged substrates (BamA^S containing deletion of POTRA domains 3–5, deletion of loop 1, and a cysteine substitution) were cloned into the pTrc99a vector to generate pDT514-pDT516 (N-proximal cysteine mutant) and pDT481 and pDT486-pDT490 (C-proximal cysteine mutant). Substrates containing the above but with POTRA domains 3–5 intact were cloned to generate pDT566-pDT568. MC4100 cells were transformed with one BamA^M-encoding plasmid and one substrate-encoding plasmid. The resulting strains were grown overnight in LB supplemented with 50 μg/mL carbenicillin, 50 μg/mL kanamycin, and 0.2% (w/v) glucose (37°C, 220 rpm). These overnight cultures were diluted 1:100 into 100 mL of fresh LB containing the same additives minus glucose, and were grown (37°C, 220 rpm) to OD600 ~0.5. Cells were then collected by centrifugation (4,200 x g, 10 min, 4°C). Cell pellets were resuspended in PBS (20 mM NaH₂PO₄ pH 7.2, 150 mM NaCl). TCEP-HCl (VWR) was then added at a final concentration of 2 mM, and cells were incubated on a rocking platform (20 min, room temperature). Cells were then centrifuged (5,000 x g, 10 min, 4°C) and again resuspended in PBS. To test for cysteine crosslinking, 1,2-bis(maleimido)ethane (BMOE, Thermo Fisher) was added at a final concentration of 0.5 mM. After incubation on a rocking platform (40 min, room temperature), the crosslinking reaction was quenched via addition of L-cysteine hydrochloride monohydrate (Alfa Aesar) to a final concentration of 10 mM. Cells were centrifuged (5,000 x g, 10 min, 4°C) and the pellets were frozen at −80°C prior to subsequent purification. To test disulfide bond formation in the absence of crosslinker, cells were frozen after the initial centrifugation step. In each sample, 6×His-BamA^M was purified and crosslinked adducts were detected by Western blotting as described for site-specific in vivo photocrosslinking experiments. β-mercaptoethanol was not added to samples used to assess disulfide formation in the absence of crosslinker.

Assessing heat modifiability of substrates.

2×Strep-BamA^S containing deletion of POTRA domains 3–5 and the C690S, C700S, and T467C mutations was cloned into the pTrc99a vector to generate pDT534. Another plasmid containing the same mutations, but with deletion of loop 1 (Δ430–439) was cloned to generate pDT535. The equivalent constructs with the E800C mutation were cloned to generate pDT509 (loop 1 intact) and pDT510 (loop 1 deleted). MC4100 cells were transformed with pDT534, pDT535, pDT509, or pDT510. The resulting strains were grown overnight in LB supplemented with 50 μg/mL carbenicillin and 0.2% (w/v) glucose (37°C, 220 rpm). These overnight cultures were diluted 1:100 into 5 mL of fresh LB containing the same additives minus glucose, and were grown (37°C, 220 rpm) to OD600 ~0.4. Cells were then collected by centrifugation (4,200 x g, 10 min, 4°C) and resuspended in 2× SDS loading dye (containing β-mercaptoethanol). Each sample was divided into two, with one half boiled for 5 min and the other half remaining unboiled. All samples were analyzed by SDS-PAGE and Western blotting. 2×Strep-BamA^S was detected using a monoclonal Strep-Tag II (HRP) antibody (EMD Millipore).

Expression and crosslinking of substrate-bound Bam complex for cryo-EM.

The plasmid pJH114 was modified such that BamA^M contained the C690S and C700S substitutions, in addition F804C, generating pDT517. 2×Strep-BamA^S containing deletion of POTRA domains 3–5, deletion of loop 1, the C690S and C700S substitutions, and T467C, was cloned into the pBAD33 vector to generate pDT518. BL21(DE3) cells harboring pDT517 and pDT518 were grown overnight (37°C, 220 rpm) in LB supplemented with 50 μg/mL carbenicillin, 30 μg/mL chloramphenicol, and 0.2% glucose. This overnight culture was diluted 1:100 into 3 × 1.5 L of LB supplemented with 50 μg/mL carbenicillin and 30 μg/mL chloramphenicol. The resulting cultures were grown (37°C, 220 rpm) until an OD600 of ~0.7 was reached. At this point, the temperature was turned down to 30°C, and cells were allowed to continue shaking. After 20 min, protein expression was induced with isopropyl β-D-1-thiogalactopyranoside (IPTG, VWR) and L-(+)-arabinose (Alfa Aesar) at final concentrations of 0.2 mM and 0.1% (w/v), respectively. After three hours of additional shaking, cells were harvested via centrifugation (4,200 x g, 10 min, 4°C). Cell pellets were resuspended in PBS (20 mM NaH₂PO₄ pH 7.2, 150 mM NaCl) and incubated with 0.3 mM copper(II) sulfate (Acros Organics) and 0.3 mM 1,10-phenanthroline (Sigma-Aldrich) for 30 minutes. Cells were centrifuged (5,000 x g, 10 min, 4°C) and the pellets were frozen at −80°C prior to subsequent purification.

For the low resolution cryo-EM structure, the purification was performed similarly, but using plasmids pDT397 and pDT500 containing the S439C and E800C mutations, respectively. Cells were treated with BMOE as described above rather than with copper(II) sulfate and 1,10-phenanthroline.

Purification of substrate-bound Bam complex for cryo-EM.

Cell pellets expressing Bam-substrate complex as described above were thawed and resuspended in buffer containing 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 100 μg/mL lysozyme (Sigma-Aldrich), 1 mM PMSF (Sigma-Aldrich), 50 μg/mL DNase I (Sigma-Aldrich), and 2.5 mM MgCl₂ (Sigma-Aldrich). Cells were lysed using an Emulsiflex C3 (Avestin) at a pressure of 10,000 to 15,000 psi. After lysis, cell debris was removed via centrifugation (5,000 x g, 10 min, 4°C). Membrane fractions were isolated via ultracentrifugation using a 45 Ti rotor (Beckman Coulter) (37,000 rpm, 45 min, 4°C) and an Optima XE-90 ultracentrifuge (Beckman Coulter). The membrane pellet was resuspended in buffer containing 20 mM Tris-HCl pH 8.0, 150 mM NaCl, 100 μg/mL lysozyme, and 1 mM PMSF. Membrane fractions were solubilized via incubation with 0.75% n-dodecyl-β-D-maltopyranoside (DDM, Anatrace) and 0.5% glyco-diosgenin (GDN, Anatrace) on a rocking platform (2 hours, 4°C). Unsolubilized material was then isolated via ultracentrifugation in a 70 Ti rotor (Beckman Coulter) (37,000 rpm, 30 min, 4°C). The supernatant, consisting of solubilized membrane proteins, was removed and supplemented with imidazole (pH 8.0) to a final concentration of 5 mM.

The supernatant was incubated with Ni-nitrilotriacetic acid (NTA) resin (Qiagen) that had been pre-washed with 10 CV buffer W1 (20 mM Tris-HCl pH 8.0, 150 mM NaCl, 10 mM imidazole pH 8.0, 0.02% GDN). After batch binding on a rocking platform (1 hour, 4°C), the resin was washed with 10 CV buffer W1. Elution was performed via addition of 5 CV buffer E1 (20 mM Tris-HCl pH 8.0, 150 mM NaCl, 200 mM imidazole pH 8.0, 0.02% GDN).

The Ni-NTA eluate was immediately incubated with Strep-Tactin XT Superflow resin (IBA Lifesciences) that had been pre-washed with 10 CV buffer W2 (100 mM Tris-HCl pH 8.0, 150 mM NaCl, 0.02% GDN). After batch binding on a rocking platform (1 hour, 4°C), the resin was washed with 15 CV buffer W2. Elution was performed via addition of 9 CV buffer E2 (100 mM Tris-HCl pH 8.0, 150 mM NaCl, 1 mM EDTA, 50 mM D-biotin, 0.02% GDN).

The Strep resin eluate was concentrated using an Amicon Ultra 4 mL 100 kDa molecular-weight cutoff centrifugal concentrator (EMD Millipore). The sample was then applied to an ÄKTA Pure (GE Healthcare Life Sciences) for purification via size-exclusion chromatography using a Superdex 200 Increase 10/300 GL column. The protein was eluted in buffer containing 20 mM Tris-HCl pH 8.0, 150 mM NaCl, and 0.02% GDN. After elution, protein corresponding to the center peaks of the chromatogram was concentrated to 5 mg/mL using an Amicon Ultra 0.5 mL 100 kDa molecular-weight cutoff centrifugal concentrator (EMD Millipore). Regarding the purifications performed to obtain both the 4.1 Å and 6.5 Å structures, a final yield of approximately 0.1 mg of complex per liter of bacterial culture could be obtained.

Electron microscopy data collection.

Purified substrate-bound Bam complex as described above was applied to glow-discharged Quantifoil R 2/1 holey carbon 400-mesh copper grids (Quantifoil). Grids were blotted for 4–5 s at 100% humidity with the blot force set to 16, and flash frozen by liquid nitrogen-cooled liquid ethane using a Thermo Fisher Scientific Vitrobot Mark IV (Thermo Fisher Scientific). The grid was then loaded onto a Titan Krios G3i electron cryo-microscope (Thermo Fisher) operated at 300 kV accelerating voltage. Image stacks (movies) were recorded on a Gatan Bioquantum K3 Imaging Filter (Gatan, USA) using the super-resolution counting mode and the calibrated magnification of 58717× using SerialEM⁴⁹. The slit of the energy filter was set to 25 eV, with a defocus range between 1.1 and 2.8 μm. The electron dose rate was 17 e−/physical pixel per second, and the subframe time was set to 0.06 second. A total exposure time of 3 second resulted in 50 subframes per image stack. The total electron dose was 70 e−/Ų (~1.4 e−/Ų per subframe). The multi-shot scheme in SerialEM was used for data collection, with settings of 4 holes per stage move, and 5 shots per hole, which greatly sped up the data collection. The data collection for both structures (4.1 Å and 6.5 Å) was performed in the same manner.

Image processing and 3D reconstruction.

The movie frames were motion-corrected and dose-weighted by MotionCor2⁵⁰ and CTF parameters were estimated by CTFFIND4⁵¹. Particle picking was carried out using crYOLO⁵² giving 2,054,956 initial particles. Following successive rounds of 2D and 3D classification within Relion⁵³, 516,419 particles were then “polished” through the Bayesian polishing process⁵⁴. These particles were then subjected to further 2D and 3D classification within CryoSPARC⁵⁵, after which 223,353 particles were selected (see classification flowchart in Extended Data Figure 3) and subjected to non-uniform refinement in CryoSPARC, which led to the final reconstruction at 4.1 Å resolution. Masking strategies did not yield an improved structure. Maps used for figures were filtered according to local resolution with B-factor sharpening within CryoSPARC.

For the low resolution (6.5 Å) structure, the movie frames were motion-corrected and dose-weighted by MotionCor2⁵⁰ and CTF parameters were estimated by CTFFIND4⁵¹. Particle picking was carried out using crYOLO⁵² giving 690,143 initial particles. Following successive rounds of 2D and 3D classification within Relion⁵³, 233,064 particles were then “polished” through the Bayesian polishing process⁵⁴. Refinement in Relion led to the final reconstruction at 6.5 Å resolution. Masking strategies did not yield an improved structure. Maps used for figures were filtered according to local resolution with b-factor sharpening within Relion. Structural biology applications used in this project (other than CryoSPARC) were compiled and configured by SBGrid⁵⁶.

Model building, refinement, and validation.

The atomic model was generated using available structures of the Bam complex components. For BamA^MBDE, an initial model was generated by rigid-body fitting these components into the 4.1 Å EM map using UCSF Chimera⁵⁷. The BamA^M POTRA domains and BamE were obtained from PDB ID: 5D0O. The BamA^M β-barrel and BamD were obtained from PDB ID: 5D0Q. BamB was obtained from PDB ID: 2YH3. For the substrate, a homology model generated with I-TASSER was used^58–60.

The initial modeling was followed by manual adjustments using Coot⁶¹. All selenomethiones were replaced with methionines. The improved model was then refined in real space against the cryo-EM map using real space refinement in PHENIX⁶² with secondary structure restraints. Iterative rounds of manual and automated refinement in Coot and PHENIX, respectively, generated a model that included all components except for BamC and the POTRA domains 1 and 2 of BamA^M. Due to the lower resolution of these components, the atomic models were generated by docking previously determined structures into our map (from 5D0Q for BamC and 5D0O for the POTRA domains) without further refinement. For the entirety of POTRA domains 1 and 2 of BamA^M (residues 1–171) and the C-terminal domain of BamC (residues 89–209), only the main chain atoms are included in the model. For the substrate, the two POTRA domains could not be resolved, so they are not included in the model. The final model was visually inspected for general fit to the map, and geometry was further evaluated using MolProbity⁶³. Cryo-EM data collection, refinement, and validation statistics are summarized in Supplementary Table 1. Figures depicting the structure were prepared in Chimera or PyMOL (Schrödinger, https://www.pymol.org).

Extended Data

Extended Data Figure 1 |. Capturing the BamA^S-ΔL1 substrate on the Bam complex using disulfide bond formation.

a, Topological map of the β-barrel domain of BamA with extracellular loops labelled. b, Residues near the C-terminus of the β-barrel domain of BamA^M that were substituted with cysteine (yellow sticks) shown on the structure of the β-barrel domain of BamA (green, PDB ID: 5D0O). The N- and C-terminal β-strands are labelled and shown in salmon. c, As in b, but showing residues near the N-terminus of the β-barrel domain of the substrate (BamA^S-ΔL1, blue) that were substituted with cysteine. The positions of these cysteine substitutions were selected based on reports that the C-terminal region of BamA^M is in proximity to the N-terminal region of the substrate during folding^10–12. d, Disulfide bond formation between 6×His-tagged BamA^M and 3×FLAG-tagged BamA^S-ΔP345-ΔL1 containing cysteine substitutions shown in b and c. The presence of a Bam-substrate disulfide bond is detected as a high molecular weight adduct on the α-FLAG (middle) immunoblot. The adducts generated via disulfide bond formation were not present in an amount high enough for detection with the α-His antibody but could be detected with the α-FLAG antibody. α-FLAG immunoblot of total cell lysates (bottom) shows that the substrates containing different cysteine substitutions are expressed at similar levels. In this experiment, no oxidizing agent (e.g., copper(II) sulfate and 1,10-phenanthroline) was added. e, As in d, but without any cysteine introduced into BamA^M. Disulfide bond formation is not observed between BamA^M and the substrates when no cysteine is introduced into BamA^M. f, As in d, but with substrates containing all five POTRA domains. Disulfide bond formation between BamA^M and substrates containing all POTRA domains indicates that stalling of substrates in d is not due to deletion of POTRA domains 3–5. Data shown in d-f are representative of results from two biological replicates.

Extended Data Figure 2 |. Expression and purification of substrate-bound Bam complex for cryo-EM.

a, Constructs used for protein expression. Mutations and affinity tags introduced into each protein are indicated. SP represents signal peptide; P1 to P5 represent the POTRA domains of BamA. The substrates also contain deletion of POTRA domains 3, 4, and 5 (Δ172–421) to avoid the possibility of these substrates forming Bam complexes if they complete folding. The deletion does not prevent folding of an otherwise wild-type BamA (see below). b, Assessment of heat modifiability of substrate used for cryo-EM (ΔL1, right half of blot) that contains the T467C mutation and a substrate that contains loop 1 but is otherwise identical (WT, left half of blot). The samples in the unboiled lanes can be used to calculate the fraction of each substrate that is folded in vivo (indicated below the blot). BamA^S-ΔL1 accumulates on the Bam complex because of the deletion of loop 1. c, Representative size-exclusion chromatogram of the substrate-bound Bam complex in which disulfide bond formation between the N-terminus of the substrate (BamA^S-ΔL1) and the C-terminus of BamA^M was induced with the oxidizing agent copper sulfate/1,10-phenanthroline (CuP). d, SDS-PAGE gels showing peak fractions from size-exclusion chromatography of the complex in c. The left gel shows the complex without addition of βME, while the right gel shows the complex after addition of βME to break the disulfide bond. Data shown in b-d are representative of results from two biological replicates.

Extended Data Figure 3 |. Cryo-EM data processing and analysis for substrate-bound Bam complex.

a, Representative cryo-EM micrograph of the substrate-bound Bam complex embedded in vitreous ice. b, Selected two-dimensional class averages of cryo-EM particle images. c, Scheme of three-dimensional classification and refinement of cryo-EM particle images. d, Gold-standard Fourier shell correlation (FSC) curves calculated with different masks in cryoSPARC. The resolutions were determined at FSC = 0.143 (horizontal blue line). The final corrected mask gave an overall resolution of 4.1 Å. e, Distribution of orientations over azimuth and elevation angles for particles included in the calculation of the final map. f, Cryo-EM map colored by local resolution as shown in Figure 1g, but at a lower contour level.

Extended Data Figure 4 |. Fit of the atomic model into the cryo-EM map.

The atomic model (in stick representation) shown with the corresponding portion of the cryo-EM map (shown in gray mesh) for selected regions in a, the BamA^M β-barrel domain; b, the BamA^M POTRA domains; c, BamB; d, BamC; e, BamD; f, BamE; g, the substrate; and h, the BamA^M-substrate interaction. Side-chain densities are visible, and individual β-strands can be resolved. Images were prepared in UCSF Chimera using a 2 Å carve radius.

Extended Data Figure 5 |. Comparison of the architecture of the substrate-bound Bam complex to that of substrate-free Bam complexes.

a-c, Alignments of substrate-bound Bam complex (all components shown in green) with published Bam complex structures (shown in different colors). Alignments are performed using the β-barrel domains of BamA.

Extended Data Figure 6 |. Conformational differences between Bam complex components from substrate-bound and substrate-free complexes.

Alignments of atomic models of a, BamA^M; b, BamD; and c, BamE from our substrate-bound complex (colored as in Figure 2) with the corresponding components from a substrate-free, lateral-open Bam complex (light orange, PDB ID: 5EKQ). d-f, as in a-c, but using a different substrate-free, lateral-open Bam complex (blue, PDB ID: 5D0Q). Alignments of atomic models of g, BamA^M; h, BamB; i, BamD; and j, BamE from our substrate-bound complex with the corresponding components from a substrate-free, lateral-closed Bam complex (dark red, PDB ID: 5D0O). Alignments of BamA^M are performed using the β-barrel domains. Alignments of BamB are not shown for complexes in which BamB is absent. Alignments of BamC are not shown since BamC in our structure was obtained by docking BamC as a rigid body from PDB ID: 5D0Q. The RMSD value for each alignment, obtained with PyMOL, is indicated.

Extended Data Figure 7 |. The BamA^S-ΔL1 substrate accumulates on the Bam complex in a largely folded state.

Alignment of the atomic model of the BamA^S-ΔL1 substrate (blue) with that of folded wild-type BamA (cyan, PDB ID: 4N75). The first two β-strands of the substrate in our structure could not be resolved (β1 and β2, labelled on the structure of wild-type BamA). The similarity between the trapped substrate and folded wild-type BamA suggests that the former represents an on-pathway folding intermediate.

Extended Data Figure 8 |. Low resolution cryo-EM structure of a substrate-bound Bam complex with an alternative cysteine crosslink.

a, Residue near the N-terminus of the β-barrel domain of BamA^M that was substituted with cysteine (yellow stick) shown on the structure of the β-barrel domain of BamA (green, PDB ID: 5D0O). The N- and C-terminal β-strands are labelled and shown in salmon. b, As in a, but showing residues near the C-terminus of the β-barrel domain of the substrate (BamA^S-ΔL1, blue) that were substituted with cysteine. c, Cysteine crosslinking between 6×His-tagged BamA^M and 3×FLAG-tagged BamA^S-ΔP345-ΔL1 containing cysteine substitutions shown in a and b. The crosslinker 1,2-bis(maleimido)ethane (BMOE), which has an 8.0 Å spacer arm, was used. The presence of a Bam-substrate crosslink is detected as a high molecular weight adduct on α-His (top) and α-FLAG (middle) immunoblots. α-FLAG immunoblot of total cell lysates (bottom) shows that the substrates containing different cysteine substitutions are expressed at similar levels. Based on these results, E800C was selected for use within BamA^S-ΔL1 for cryo-EM, while S439C was introduced into BamA^M. d, Assessment of heat modifiability of substrate used for cryo-EM (ΔL1, right half of blot) that contains the E800C mutation and a substrate that contains loop 1 but is otherwise identical (WT, left half of blot). The samples in the unboiled lanes can be used to calculate the fraction of each substrate that is folded in vivo (indicated below the blot). The substrate containing the E800C mutation retains the ability to fold in vivo. e, Representative size-exclusion chromatogram of the substrate-bound Bam complex in which cysteine crosslink formation between the C-terminus of the substrate (BamA^S-ΔL1) and the N-terminus of BamA^M was induced with BMOE. f, SDS-PAGE gels showing peak fractions from size-exclusion chromatography. The left gel shows the complex from e with addition of βME, which cannot break the crosslink formed by BMOE. The right gel shows the complex crosslinked with dithiobismaleimidoethane (DTME), which is similar to BMOE, but has a 13.3 Å spacer arm and is cleavable in the presence of βME to visualize the individual components of the complex. g, Cryo-EM map obtained using the BamA^M(S439C) and BamA^S-ΔP345-ΔL1(E800C) cysteine pair crosslinked with BMOE. The two images show the same view at different contour levels. BamA^M and the substrate are labelled. h, Gold-standard Fourier shell correlation (FSC) curves for low resolution cryo-EM structure calculated in Relion. The final corrected mask gave an overall resolution of 6.5 Å at FSC = 0.143 (horizontal blue line). i, Comparison of the 4.1 Å structure to the 6.5 Å structure containing the alternative cysteine pair in BamA^M and the substrate. In the left panel, a top-down slice through the 4.1 Å cryo-EM map is shown (similar to the slice in Fig. 3c) and is overlaid with the atomic model generated from that map. In the right panel, a similar slice through the 6.5 Å cryo-EM map is shown and is overlaid with the atomic model generated from the 4.1 Å map after fitting of the atomic model into the 6.5 Å map as a rigid body. The dashed oval represents the approximate location of the cysteine tether introduced into each complex. The use of different cysteine pairs yielded similar cryo-EM structures, showing that the presence of the disulfide bond in the main (4.1 Å) structure did not introduce a non-native conformation into the complex. Data shown in c-f are representative of results from two biological replicates.

Extended Data Figure 9 |. In vivo photocrosslinking with wild-type BamA and BamA^S-ΔL1.

a, Residues in BamA substituted with the photocrosslinkable amino acid pBPA (yellow sticks) shown on the structure of the β-barrel domain of BamA (blue, PDB ID: 5D0O). These pBPA substitutions were used to observe crosslinks to LPS. The N- and C-termini of BamA are shown in salmon. All highlighted residues have side chains oriented outward towards the membrane environment. b, In vivo photocrosslinking of BamA to LPS. In a and b, the pBPA substitutions that were subsequently used to test crosslinking of stalled substrates to LPS (in Figure 3d) are indicated in green. c, Residues in β-strands 1 and 2 of BamA^S-ΔL1 substituted with pBPA shown on the structure of the β-barrel domain of BamA. Colors are as described in a. These positions were selected since β-strands 1 and 2 of the substrate are not visible in the cryo-EM structure (Extended Data Figure 8). d, In vivo photocrosslinking of Strep-tagged BamA^S-ΔL1 containing pBPA at positions F428 or F440 and deletion of POTRA domains 3–5. Immunoblotting was performed using α-His, α-Strep, and α-FLAG antibodies to detect BamD (loading control), BamA^S-ΔL1, and BamA^M, respectively. For α-FLAG immunoblot, a longer exposure (top, to detect crosslinks) and a shorter exposure (bottom) are shown. Although β-strands 1 and 2 of the substrate are not visible in the cryo-EM structure, the in vivo photocrosslinking results show that residues within these β-strands are in proximity to BamA^M. The experiment in b was performed once, and the pBPA substitutions indicated in green were tested again in Figure 3d. Data shown in d are representative of results from two biological replicates.

Extended Data Figure 10 |. Proposed model of β-barrel assembly by the Bam complex.

a, The substrate is recruited to the Bam complex. BamA^M is in a closed state prior to substrate-induced opening of its lateral gate. For simplicity, only BamA^M, the substrate, and BamD are shown. b-d, The C-terminus of the substrate interacts with the exposed N-terminal edge of BamA^M, and β-strands or β-hairpins of the substrate are added sequentially from the C- to N-terminus. Early folding may occur within the interior of the BamA^M β-barrel¹², and folded portions of the substrate may then be released outward. Full membrane integration could occur after a substantial amount of folding. Steps b and c represent intermediate stages to d, which corresponds to the cryo-EM structure of the substrate-bound Bam complex. e, The substrate is released into the membrane environment once its N- and C-terminal ends are joined.