Flexibility in the Periplasmic Domain of BamA is Important for Function

Summary

The β-barrel assembly machine (BAM) mediates the biogenesis of outer membrane proteins (OMPs) in Gram-negative bacteria. BamA, the central BAM subunit composed of a transmembrane β-barrel domain linked to five polypeptide transport-associated (POTRA) periplasmic domains, is thought to bind nascent OMPs and undergo conformational cycling to catalyze OMP folding and insertion. One model is that conformational flexibility between POTRA domains is part of this conformational cycling. Nuclear magnetic resonance (NMR) spectroscopy was used here to study the flexibility of the POTRA1–5 domains in solution. NMR relaxation studies defined effective rotational correlational times and together with residual dipolar coupling data showed that POTRA1–2 is flexibly linked to POTRA3–5. Mutants of BamA that restrict flexibility between POTRA2 and POTRA3 by disulfide crosslinking displayed impaired function in vivo. Together these data strongly support a model where conformational cycling of hinge motions between POTRA2 and POTRA3 in BamA is required for biological function.

Graphical Abstract

eTOC Blurb

BamA is essential for folding and insertion of b-barrel Outer Membrane Proteins in Gram-Negative bacteria. Warner et al. show that the 5 POTRA domains of BamA feature a hinge between POTRA2 and 3 and its flexibility is important for BamA in vivo function.

Introduction

BamA is a multi-domain protein found in all diderm bacteria that together with the lipoproteins BamB-E constitute the β-barrel assembly machine (BAM) complex responsible for folding and insertion of outer membrane proteins (OMPs) into the bacterial outer membrane (Voulhoux et al., 2003; Wu et al., 2005). BamA is essential for cell viability and is comprised of five N-terminal polypeptide transport-associated (POTRA) repeats that extend into the periplasm and a C-terminal β-barrel domain that spans the outer membrane (Gentle et al., 2005; Noinaj et al., 2015). The POTRA domains share a common fold consisting of two α-helices packed against a three-stranded β-sheet (Gatzeva-Topalova et al., 2008; Kim et al., 2011), whereas the transmembrane domain adopts a 16-strand β-barrel fold (Albrecht et al., 2014; Ni et al., 2014; Noinaj et al., 2013). The POTRA domains provide a scaffold for the BAM complex by interacting with BamB (Jansen et al., 2015; Kim et al., 2007; Vuong et al., 2008) and a subcomplex of the lipoproteins Bam C, D and E (Kim et al., 2007; Ricci et al., 2012; Sklar et al., 2007).

Although the mechanism of OMP folding and insertion by BAM remains unclear, BamA is the central component of the complex. BamA by itself accelerates folding and insertion of model OMPs into liposomes, albeit at substantially lower rates than those observed in vivo (Gessmann et al., 2014; Patel and Kleinschmidt, 2013; Plummer and Fleming, 2015). Reconstitution of BamA with the BAM lipoproteins increases its efficiency, suggesting that the lipoproteins enhance BamA activity (Hagan et al., 2010; Roman-Hernandez et al., 2014). Several models of OMP insertion have been proposed where the BamA transmembrane barrel plays a crucial role. As in all OMPs, the β-barrel of BamA is an antiparallel β-sheet closed into a barrel by the interaction of the first and last strands. However, this barrel “seam” is shorter in BamA with only a few hydrogen bonds and therefore likely weaker than other OMPs (Bakelar et al., 2016; Gu et al., 2016; Han et al., 2016; Noinaj et al., 2013). This observation inspired insertion models where β-hairpins from nascent OMPs are sequentially inserted in the BamA barrel seam leading to “budding” of nascent OMPs into the membrane from BamA (Noinaj et al., 2014; Noinaj et al., 2015). Alternative models suggest that the BamA barrel induces local defects in the outer membrane that facilitate insertion of nascent OMPs (Fleming, 2015; Gessmann et al., 2014; Noinaj et al., 2013; Plummer and Fleming, 2015). Recently, structures of BamACDE and BamABCDE complexes further illustrate the dynamic nature of the BamA barrel, showing conformations where the barrel seam is open, creating a “lateral gate” that may aid OMP insertion (Bakelar et al., 2016; Gu et al., 2016). This model is also supported by functional assays showing that disulfide locking of the BamA barrel lateral gate is lethal (Noinaj et al., 2014).

Whereas the BamA barrel domain appears essential for membrane insertion, the initial interaction with nascent OMPs is thought to be mediated by its periplasmic POTRA repeats, with BamD also being implicated in substrate recognition (Hagan et al., 2015). Crystallographic analyses of the first four POTRA domains (POTRA1–4) of E. coli BamA suggest that the POTRA domains bind substrate OMPs at the edge of their β-sheets in a process known as β-augmentation, thus inducing nucleation of β-strands in the nascent β-barrels (Gatzeva-Topalova et al., 2008; Kim et al., 2007). Crystal structures of BamA show that POTRA1–2 and POTRA3–5 form two subdomains (Gatzeva-Topalova et al., 2008; Gatzeva-Topalova et al., 2010; Kim et al., 2007; Noinaj et al., 2013). The relative orientations of POTRA domains within these subdomains are well conserved even when comparing structures from E. coli, Haemophilus ducrei and Neisseria gonorrhea (Figure 1) (Gatzeva-Topalova et al., 2008; Gatzeva-Topalova et al., 2010; Kim et al., 2007; Noinaj et al., 2013). Furthermore, small-angle X-ray scattering (SAXS) studies indicated that POTRA3–5 has a rigid orientation in solution (Gatzeva-Topalova et al., 2010) and a PELDOR spectroscopy study of POTRA1–2 concluded that it behaved as a single rigid unit in solution (Ward et al., 2009). However, the orientation between these subdomains varies widely in the crystal structures, apparently due to a flexible hinge between POTRA 2 and POTRA3 (Figure 1). The extremes of the observed orientations give rise to so-called “compact” and “extended” conformations for the domain (Gatzeva-Topalova et al., 2008; Kim et al., 2007). In agreement with these observations, solution SAXS data on POTRA1–5 was inconsistent with POTRA1–5 tumbling as a single rigid species in either the “compact” or “extended” form (Gatzeva-Topalova et al., 2008). Instead, it fit best to a population-weighted average of conformations with approximately 25% “compact” and 75% “extended” forms consistent with a model of two rigid subdomains, POTRA1–2 and POTRA3–5, connected by a flexible hinge. It has been proposed that conformational cycling in BamA is necessary for OMP folding and insertion (Rigel et al., 2013), and recent structures of BamA in complex with multiple BAM subunits show that the POTRA domains form a closed ring structure along with BamD, where the ring is adjacent to the membrane (Bakelar et al., 2016; Bergal et al., 2016; Gu et al., 2016; Han et al., 2016). Flexibility at the POTRA2–3 hinge would allow opening and closing of this ring structure, where the ring has been proposed to hold nascent OMPs close to the membrane prior to their insertion and folding. However, the physiological importance of flexibility around the POTRA2–3 hinge has not been directly assessed, and recent solid-state NMR studies suggested limited flexibility for the POTRA domains in BamA (Sinnige et al., 2014).

Figure 1.

Orientations of BamA POTRA domains. (a) Crystal structures of BamA POTRA3–5 domains superimposed on POTRA4; (b) Crystal structures of BamA POTRA1–2 domains superimposed on POTRA2; (c) Crystal structures of BamA POTRA1–3 domains superimposed on POTRA3 displaying conformational flexibility in the hinge between POTRA 2 and POTRA3 that give rises to “compact” and “extended” conformations. The POTRA domains shown are from the following crystal structures: E coli BamA POTRA 1–4 (PDBID 2QZC) red; POTRA 1–4 (PDBID 2QDF) green; POTRA 1–4 (PDBID 3EFC) blue; POTRA 4–5 (PDBID 3OG5) gray; POTRA 4–5 (PDBID 3Q6B) magenta; BamACDE complex (PDBID 5EKQ) yellow; Neisseria gonorrhoeae BamA (PDBID 4K3B) cyan; and Haemophilus Ducreyi BamA POTRA 4-5-barrel (PDBID 4K3C) orange.

Here, the flexibility of POTRA1–5 in solution was assessed directly using NMR measurements of the backbone amides. Evidence for conformational exchange from relaxation data and residual dipolar couplings (RDCs) demonstrated two subdomains in BamA that show conformational flexibility at the POTRA2–3 hinge. The functional importance of this hinge flexibility is demonstrated in vivo by complementation assays of a BamA depletion E. coli strain with disulfide cross-linked mutants of BamA.

Results

Resonance assignment and chemical shift analysis of backbone amides in BamA POTRA1–5

The first step in the NMR studies of POTRA1–5 was resonance assignment of the backbone amides. The multi-domain character of POTRA1–5 simplified this process, because essentially complete assignments were previously obtained for POTRA1–2(Knowles et al., 2008) and POTRA4–5 (Gatzeva-Topalova et al., 2010). Since the single POTRA 3 domain by itself is not soluble, transverse relaxation-optimized spectroscopy (TROSY) based backbone assignment experiments (Cavanagh et al., 2007; Pervushin et al., 1997) were performed on a ²H, ¹³C, ¹⁵N-labeled POTRA3–5 construct leading to assignment 87% of the non-proline residues. Exchange broadening of the amide resonances was observed for multiple regions in POTRA3 preventing assignments of some residues as illustrated in Figure S1a. The assignments for POTRA3–5 were then transferred to POTRA1–5 by comparison of the 2D ¹H, ¹⁵N heteronuclear single quantum correlation (HSQC) spectra, yielding assignment of 319 of 389 of non-proline residues including 99% in POTRA1–2 and 71% in POTRA3–5. A 3D ¹⁵N NOESY-HSQC spectrum on POTRA1–5 was used to validate assignments where the observed nuclear Overhauser effects (NOEs) between amide protons were compared with NOEs predicted from the X-ray structures, as previously described (Warner et al., 2011). Figure S1b shows the unassigned residues mapped onto the model of POTRA1–5.

To help understand interdomain interactions involving POTRA3 in POTRA1–5, the ¹H, ¹⁵N chemical shifts in the POTRA1–2 or POTRA4–5 constructs were compared with those for the same residues in the POTRA1–5 construct. The changes in chemical shifts of backbone amides in POTRA2 and POTRA4 in constructs with and without POTRA3 were then used to probe for conformational changes arising from interdomain interactions. As seen in Figure 2, only small chemical shift perturbations were observed in POTRA2 in constructs with and without POTRA3 (all less than ≈0.3 ppm), whereas large differences were observed for POTRA4 (many residues > 1.5 ppm). The small chemical shift perturbations in the POTRA2 domain mean that the chemical environment was unchanged upon addition of POTRA3, indicating a limited interaction between POTRA2 and POTRA3. These data support a model with flexibility at the POTRA2–3 junction.

Figure 2.

Interdomain interactions in BamA POTRA1–5. (a) The absolute value of the ¹H/¹⁵N chemical shift perturbations (Δδ ¹H,¹⁵N ), which were obtained by comparing POTRA1–2 or POTRA4–5 with POTRA1–5. POTRA2 has small chemical shift perturbations upon addition of POTRA3 as compared with those in POTRA4. (b) The large chemical shift changes observed in the POTRA4 domain are consistent with the domain packing observed in the crystal structures 3EFC (Gatzeva-Topalova et al., 2008) and 2QDF (Kim et al., 2007). The Δδ ¹H,¹⁵N values are mapped on the spliced model for POTRA1–5 (see Methods) with perturbations between 0.1 to 0.2 ppm in yellow, >0.2 to 0.4 ppm in orange and >0.4 in red. The larger Δδ ¹H,¹⁵N values tend to cluster to the linker connecting POTRA3 and POTRA4, the α2 helix and loop L2, and sections of strands β2 and β3 and loop L4 in POTRA4. For residues with perturbations > 0.1 their heavy atoms are shown as spheres.

¹⁵N relaxation data as a probe of backbone flexibility in POTRA1–5

The {¹H}-¹⁵N heteronuclear NOEs, R₁, and R_1ρ values were measured for most of the assigned backbone amide residues in POTRA1–5 (Figures 3a and S2). The {¹H}-¹⁵N heteronuclear NOEs provide direct information on regions of backbone disorder in proteins (Cavanagh et al., 2007). The {¹H}-¹⁵N heteronuclear NOE data were analyzed here to determine if there were regions of POTRA1–5 with small or negative NOE values, indicative of high disorder. As seen in Figure 3a, the range of {¹H}-¹⁵N heteronuclear NOE values for the non-terminal residues was typical for proteins with stable secondary structures. Excluding the N- and C-termini, there are no stretches of small {¹H}-¹⁵N heteronuclear NOE values indicative of a highly flexible linker. In POTRA1–5, the domains are connected by very short 3–4 amino acid linkers. No heteronuclear NOE (or relaxation data) could be obtained for residues in the POTRA2–3 linker, due to unassigned residues or exchange broadening. Heteronuclear NOEs were obtained for most of the non-proline residues in the other linkers in POTRA1–5 and these showed no evidence of significant disorder for the backbone amides..

Figure 3.

¹⁵N relaxation data show that POTRA1–5 tumbles as two subdomains. (a) The {¹H}-¹⁵N heteronuclear NOEs values for individual amide residues, where the dashed line shows an NOE threshold value of 0.70. (b) The residue-specific rotational correlation times, τ_c, are shown as a function of sequence and were calculated from backbone ¹⁵N relaxation measurements as described in Methods. The average rotational correlation times, <τ_c>, were calculated for each of the individual domains and are shown as horizontal lines. The secondary structure for POTRA1–5 is shown above.

Interdomain flexibility in POTRA1–5 was further investigated by analysis of the residue-specific correlation times, τ_c, derived from the R₁, and R_1ρ experiments. The extent of the relative flexibility between domains in a protein can be estimated from analysis of the τ_c, which was approximated as the ratio of R₁ and R₂ ¹⁵N relaxation times (where R₂ was calculated from R_1ρ as described in Methods and Supplemental Material) (Kay et al., 1989). As seen in Figure S2, there is a fair amount of scatter for the R₁ and R_1ρ relaxation parameters as well as the τ_c values (Figure 3b). It is likely that the scatter is not only due to experimental uncertainty but arises from other factors including local dynamics and the non-spherical shape for POTRA1–5 (Tjandra and Bax, 1997). To help reduce any effect of local dynamics, the rotational correlation times for the individual domains (<τ_c>) in POTRA1–5 were calculated using only residues that had heteronuclear NOE values > 0.7. Using this cutoff, the individual domains have <τ_c> values of 18.3 ± 0.2, 21.1 ± 0.3, 28.0 ± 0.8, 28.4 ± 0.6 and 24.4 ± 0.6 ns for POTRA1 through POTRA5, respectively. POTRA1 and POTRA2 have the smallest <τ_c> values and POTRA3 and POTRA4 have the largest. We next used programs that calculate the hydrodynamic properties of rigid and multi-domain proteins to predict the effective rotational correlation times for POTRA1–5 when there is flexibility at one or more linkers (see Methods). The hydrodynamic parameters for rigid subdomains of POTRA1–5 were calculated using the HYDROPRO program (Ortega et al., 2011). The HYCUD program (Rezaei-Ghaleh et al., 2013) was then used to calculate the effective rotational correlation times for a model of POTRA1–5 where different rigid subdomains in POTRA1–5 were connected to each other by flexible 4-residue linker(s). Since the POTRA domains are all similarly sized, the POTRA1–5 system has a high degree of symmetry. By assuming a rigid POTRA1–3 is connected to a rigid POTRA4–5 by a flexible linker, the HYCUD program predicts <τ_c> values for POTRA1–3 that are almost twice those for POTRA4–5 which is inconsistent with the experimental data (Figure 3b). Assuming flexibility at both the POTRA2–3 and POTRA3–4 interfaces predicts similar <τ_c> values for POTRA1–2 and POTRA4–5, which is also inconsistent with the data. The best fit to the experimental <τ_c> values is when there is a flexible linker at the POTRA2–3 interface which yields <τ_c> values of 18 ± 3 and 28 ± 3 ns for rigid POTRA1–2 and rigid POTRA3–5, respectively, at 30 °C in POTRA1–5. POTRA3–5 has ≈50% larger <τ_c> than POTRA1–2, which is in excellent agreement with the experimental <τ_c> values of 19.0 ± 0.03 ns and 26.5 ± 0.2 ns for POTRA1–2 and POTRA3–5 in POTRA1–5, respectively (weighted averages of the experimental τ_c for the residues in these regions). Thus, comparison of the observed and predicted rotational correlation times support that POTRA1–2 and POTRA3–5 act as rigid independent species (on the ns timescale) connected by a short flexible linker.

Residual dipolar couplings show that POTRA1–2 and POTRA4–5 form rigid individual units within a dynamic POTRA1–5

The flexibility of domains in multi-domain proteins can also be studied using RDCs (Fischer et al., 1999). The ¹H-¹⁵N RDCs for a protein provide direct information on the orientation of individual amide bond vectors relative to a fixed molecular axis system, which is represented by a molecular alignment tensor (Tjandra and Bax, 1997). To test if individual domains have a fixed orientation in solution, an alignment tensor is calculated for each domain by analyzing the RDC data using the high-resolution structure for each domain. If the individual domains have similar values for the magnitude (|D_a|) and rhombicity (R) of their alignment tensors, then the domains are considered to have a rigid, fixed orientation (Fischer et al., 1999).

The ¹H-¹⁵N RDCs were measured on a ²H, ¹⁵N sample of POTRA1–5 in 90% H₂O/10% ²H₂O in 10 mg/ml Pf1 phage alignment medium using the ARTSY procedure as described in the Supplemental Material (Fitzkee and Bax, 2010). A total of 246 ¹H-¹⁵N backbone RDCs were obtained, but only residues in the more locally rigid parts of the protein were used in the analysis, leading to a trimmed set of 133 RDCs. A residue was considered rigid if it had an {¹H}-¹⁵N heteronuclear NOE > 0.70. Figure S3 shows this trimmed set of RDCs mapped onto the structure of POTRA1–5, illustrating that there are fewer RDCs in POTRA3 and that RDCs were observed in the helical, strand and loop regions of the structure. Next, the crystal structures of POTRA1–4 (PDB: 2QCZ(Kim et al., 2007)) and POTRA4–5 (PDB: 3Q6B(Zhang et al., 2011)) were used together with the trimmed RDC data for POTRA1–5 to calculate the D_a and R values for the alignment tensors of the individual POTRA domains RDCs in the Pf1 phage medium (see Table 1). The errors in D_a and R were estimated from analysis of the histogram plots using a bootstrap analysis (see Figure 4a and Methods). A set of independent RDC data were collected in a second alignment medium, 3% C12E5 (n-dodecyl-penta(ethylene glycol))/hexanol, using the same procedure employed for Pf1 phage, with results given in Table 1 and Figure 4b.

Table 1.

Alignment tensors for individual POTRA domains determined from RDCs for POTRA1–5 in two alignment media

In 10 mg/ml Pf1 phage
	POTRA1	POTRA2	POTRA3	POTRA4	POTRA5
Daa	−12.3 ± 0.8 12.9 ± 1.0	15.0 ± 0.9	−8.6 ± 2.0	−9.5 ± 0.7	−9.0 ± 0.7
Ra	0.58 ± 0.13 0.63 ± 0.08	0.53 ± 0.04	0.18 ± 0.22	0.18 ± 0.06	0.17 ± 0.12
N	34	41	16	21	19
In 3% C12E5/hexanol
	POTRA1	POTRA2	POTRA3	POTRA4	POTRA5
Daa	9.5 ± 1.8	9.8 ± 0.8	−5.7 ± 1.1 7.8 ± 2.0	−5.3 ± 0.9 5.1 ± 1.1	−5.2 ± 1.1 8.7 ± 2.5
Ra	0.33 ± 0.12	0.23 ± 0.18	0.61 ± 0.08 0.59 ± 0.26	0.50 ± 0.21 0.57 ± 0.22	0.42 ± 0.16 0.33 ± 0.22
N	31	40	10	22	15

^aErrors for D_a and R were estimated by bootstrapping with resampling as described in Methods. For D_a values with bimodal distribution, both positive and negative values are reported with the corresponding R values.

Figure 4.

Bootstrap analysis of alignment tensors of individual domains of POTRA1–5 calculated from RDCs collected in (a) 10 mg/ml Pf1 phage alignment medium and (b) 3% C12E5/hexanol alignment medium. Histogram plots of the distribution of D_a from the bootstrap analysis are shown for the individual POTRA domains, where the bootstrap consisted of 4,000 datasets generated with random resampling as described in Methods. The histograms were generated with bins of 0.5 Hz and fit to a Gaussian (solid lines). Bimodal distributions were observed in some of the domains, especially in the 3% C12E5/hexanol medium.

Analysis of the alignment tensors for the POTRA domains in the Pf1 phage indicates that POTRA1–2 makes up one rigid unit and POTRA4–5 is part of a second rigid unit in POTRA1–5 (Table 1). This analysis showed that POTRA1 and POTRA2 have similar D_a and R values (Table 1), where the D_a values for POTRA1 showed a bimodal distribution, due to the degeneracy of D_a (Figure 4a). This degeneracy leads to ± D_a having the same fit to the RDC data when two of the axes in the tensor have the same magnitudes, which occurs at both R = 0 and R = 2/3 (Chen and Tjandra, 2012). As seen in Table 1, the bimodal distributions generally occur for R values close to the maximum of 2/3. Although the D_a and R values for POTRA1 and POTRA2 differ by more than the errors estimated by the bootstrap analysis, the values are substantially different than those of POTRA4 and POTRA5 (Table 1). The similarity in D_a and R for POTRA4 and POTRA5 indicates the POTRA4–5 subdomain reorients as a rigid unit (Table 1 and Figure 4a). Thus, the differences in D_a and R for the POTRA1–2 and POTRA4–5 subdomains in the Pf1 alignment medium provide strong evidence for flexibility between these subdomains. The timescale (frequency) of this flexibility in the RDC analysis is less than the magnitude of the RDC themselves, which have a frequency of ≈10 Hz; therefore, corresponding to timescales faster than 100 ms (Chen and Tjandra, 2012). The alignment tensors were also determined from RDCs in 3% C12E5/hexanol alignment medium and compared to the Pf1 data had somewhat larger errors due to lower signal-to-noise in the experimental data and more of the POTRA domains had bimodal distributions (Table 1 and Figure 4b). The overall patterns are generally similar with POTRA1–2 and POTRA4–5 acting as rigid subdomains.

Complementation assay indicates that flexibility of POTRA1–5 is important for BamA function

To test the functional importance of BamA flexibility at the hinge between POTRA2 and POTRA3, single cysteine mutations were introduced in POTRA2 and POTRA3 to restrict hinge flexibility upon formation of an interdomain disulfide bond. The X-ray crystal structures of the BamA POTRA domains in “compact” or “extended” conformations were used to guide the mutagenesis (Gatzeva-Topalova et al., 2008; Kim et al., 2007). As seen in Figures 5a and 5b, respectively, the S143C-A220C mutant brings the two cysteines in proximity in the compact conformation of Bam A, whereas the V144C-D211C mutant brings two other cysteines in proximity in the extended conformation.

Figure 5.

The effects of BamA double-cysteine mutants on cell growth. Two double-cysteine mutants of BamA, S143C-A220C and V144C-D211C where designed such that disulfide bond formation would stabilize the (a) compact or (b) extended conformation of BamA, respectively. The E. coli BamA depletion strain JCM166 transformed with a plasmid encoding for wild-type BamA (black), BamA(S143C-A220C) (green), BamA(V144C-D211C) (red) or a GFP negative control (blue), were grown in the presence of arabinose and then switched at time=0 to cultures containing (c) fucose, (d) fucose + TCEP, (e) arabinose, or (f) arabinose + TCEP.

An in vivo complementation assay was then utilized to test the functional importance of restricting the flexibility at the hinge between POTRA2–3 in BamA. Since the bamA gene is essential, E. coli bamA-null mutants cannot be constructed. However, Silhavy and coworkers developed an E. coli strain, JCM166, in which the endogenous bamA gene is deleted and a bamA copy is inserted in the genome under the control of an arabinose promoter (Wu et al., 2005). These cells grow normally in arabinose containing media. However, switching to fucose-containing media represses expression of genomic bamA causing the cells to die after a few generations due to depletion of BamA. This phenotype can be complemented by expression of His-tagged-BamA (_HisBamA) from a constitutive promoter in a low copy-number plasmid pZS21 (Kim et al., 2007). This construct provides a platform to test the ability of double-cysteine mutants of BamA to complement the genomic BamA depletion phenotype.

JCM166 cells transformed with pZS21-_HisBamA(wt) grow normally in fucose because the plasmid-encoded _HisBamA complements the depletion of genome-encoded BamA (Figure 5c, black line). Conversely, cells transformed with a control pZS21 plasmid encoding GFP instead of BamA (pZS21-GFP), die after a few generations in fucose (Figure 5c, blue line). Cells transformed with pZS21-_HisBamA(S143C-A220C) or pZS21-_HisBamA(V144C-D211C) have impaired growth in fucose indicating that their in vivo activity is compromised (green and red lines respectively, in Figure 5c). However, the double-cysteine mutants are indistinguishable from wild-type BamA in supporting growth in fucose when the media is supplemented with the reducing agent Tris(2-carboxyethyl)phosphine (TCEP) (Figure 5d). These results indicate that disulfide bond formation between the introduced cysteines restricts BamA flexibility and impairs its in vivo activity. Several control experiments further support this conclusion.

First, whereas all cells grow normally in arabinose with or without TCEP (Figures 5e and 5f), JCM166 cells died after a few generations in fucose supplemented with TCEP (Figure 5d, blue line). This demonstrates that TCEP rescue of the growth phenotype of double cysteine mutants is not due to derepression of the endogenous bamA. Second, the impaired growth of double-cysteine containing mutants of BamA is not due to lower expression or misfolding. Figure S4 shows that _HisBamA(wt), _HisBamA(S143C-A220C) and _HisBamA(V144C-D211C) are all expressed at similar level and are efficiently assembled into BAM complexes with compositions similar to that for _HisBamA(wt). This indicates that the double cysteine BamA mutants are membrane targeted and do not have folding defects that prevent binding to the other four BAM subunits. Third, Figure S5 shows that BamA single cysteine variants _HisBamA(S143C), _HisBamA(A220C), _HisBamA(V144C) and _HisBamA(D211C) complement the endogenous BamA depletion. This suggests that the growth phenotype of BamA double cysteine mutants is not due to disulfide crosslinking of the engineered cysteines with other periplasmic or membrane proteins and further illustrates that the mutations do not cause BamA folding defects.

Direct observation of disulfide bond formation between the engineered cysteines in _HisBamA(S143C-A220C) and _HisBamA(V144C-D211C) is difficult because the cysteines are too close in the sequence to give altered electrophoretic mobility upon disulfide formation (see below). However, cysteine labeling with maleimide-biotin indicates that the cysteines engineered in _HisBamA(S143C-A220C) and _HisBamA(V144C-D211C) are engaged in disulfide bonds (Figure 6a). The sulfhydryls in _HisBamA double cysteine mutants expressed in JCM166 cells grown in the presence of arabinose and TCEP remain reduced and thus are efficiently labeled with maleimide-biotin, which can be visualized with streptavidin-HRP (TCEP + in Figure 6a). Conversely, when the cells are grown in arabinose without TCEP (TCEP – in Figure 6a), the labeling efficiency is severely reduced indicating that the cysteines are mostly engaged in disulfides and thus unavailable for labeling (arabinose allows expression of the endogenous bamA gene and thus cell growth). As controls, the single cysteine variant, _HisBamA(V144C), which cannot form the intramolecular disulfide is efficiently labeled in both the presence and absence of TCEP, while a cysless _HisBamA variant shows no labeling at all. Furthermore, as shown in Figure 6b, a Western blot performed under non-reducing conditions (− DTT in Figure 6b) of double cysteine _HisBamA mutants or a cysless _HisBamA variant expressed in JCM166 cells grown in the presence of arabinose (no TCEP), shows no bands with a molecular weight higher than BamA. These experiments directly show that the double cysteine _HisBamA mutants do not form intermolecular disulfides with other periplasmic or outer membrane proteins. In summary, when expressed under normal non-reducing media (no TCEP), most of the cysteines in _HisBamA(S143C-A220C) and _HisBamA(V144C-D211C) are forming intramolecular disulfide bonds (Figure 6a and Figure 6b, -DTT). Thus, we conclude that the disulfides are intramolecular.

Figure 6.

BamA double-cysteine mutants form intramolecular disulfides in vivo. (a) The E. coli BamA depletion strain JCM166 expressing His-tagged BamA cysteine free (Cysless), the double cysteine mutants (V144C-D211C and S143C-A220C) and, where indicated, a single cysteine mutant (V144C), were grown in the presence of arabinose. (a) The cells were cultured with (+) or without (−) TCEP and when the OD reached ≈0.6, the cells were treated with biotin-maleimide to label available cysteines. The _HISBamA mutants were purified by Ni-NTA and analyzed by Western blotting probed with BamA antibodies (α-BamA) or Streptavidin-HRP (Strep-HRP) to reveal biotin labeling. (b) The cells were cultured without TCEP and when the OD reached ≈0.6, the cells were treated with NEM to block reactive cysteines. The _HISBamA mutants were purified by Ni-NTA and analyzed by Western blotting probed with BamA antibodies (α-BamA) under reducing (+DTT) or non-reducing (−DTT) conditions.

Taken together, the complementation and cysteine labeling experiments show that flexibility of BamA at the hinge between POTRA2 and POTRA3 is important for full in vivo activity and that disulfides between these two POTRA domains lead to impaired growth.

Discussion

A number of complementary NMR experiments were performed on POTRA1–5, all showing that the five POTRA domains of BamA display a unique pattern of flexibility, where both POTRA1–2 and POTRA3–5 are rigid on the nanosecond timescale but are connected by a flexible linker. The ¹⁵N relaxation data were used to calculate residue-specific rotational correlations times and showed that on a fast (nanosecond) timescale POTRA1–2 and POTRA3–5 behave as rigid subdomains connected by a short flexible linker. In addition, a significant number of the residues in POTRA3 could not be assigned due to exchange broadened resonances (Figure 7), indicating microsecond to millisecond conformational exchange processes in this region. Exchange broadening was also observed for many residues in POTRA3 in the POTRA3–5 construct, indicating there is conformational dynamics for these regions even in the smaller construct (Figure S1a). This conclusion is supported by the X-ray structures for POTRA3 in the POTRA1–4 construct (Kim et al., 2007), where, as seen in Figure 7b, the long helix α2 in POTRA3 is partially unfolded. The electron density could not be fit for L2 loop which connects to helix α1 in the “compact” X-ray structure, presumably due to conformational disorder in the crystal (see Figure 7b).

Figure 7.

The unassigned residues in POTRA3 cluster near the POTRA2–3 interface. The unassigned residues in POTRA3 are highlighted in red on the structure of POTRA3 in the (a) extended and (b) compact forms of POTRA1–4. These residues are mainly in the L2 loop and α2 helix as well as the L3 loop and portions of the β2 and β3 strand. The grey dashed ovals highlight areas of structural differences for POTRA3 in the two crystal structures and the dashed red line in (b) indicates zero occupancy for the residues 201–214.

The ¹⁵N relaxation data also support flexibility in POTRA1–5 by showing that POTRA1–2 is conformationally distinct from POTRA3–5. As seen in Figure 3b, the average residue specific correlation time, <τ_c,>, for POTRA3–5 is 26.5 ± 0.2 ns, which is ≈50% larger than the 19.0 ± 0.03 ns observed for POTRA1–2, consistent with the two subdomains reorienting at different rates. Since POTRA1–2 and POTRA4–5 are approximately the same size, the large difference in their <τ_c,> indicates that POTRA3 affects the reorientation POTRA4–5 more than POTRA1–2, and is consistent with POTRA3 associating with POTRA4–5. This interpretation is further supported by the results obtained from the HYDROPRO and HYCUD programs that predicted <τc,> values for POTRA1–2 and POTRA3–5 of 18 and 28 ns, respectively, where again the <τ_c,> for POTRA3–5 is ≈50% larger than that for POTRA1–2 (see Results).

Analysis of chemical shift data for the amides in POTRA1–5 also supports the model of flexibility at the POTRA2–3 interface. As seen in Figure 2, the small chemical shift changes at the POTRA2–3 interface strongly support the absence of interdomain interaction at the interface. This conclusion is consistent with a previous proposal for flexibility at this interface, which was based on the dearth of interdomain interactions observed at POTRA2–3 compared to the other interfaces in the X-ray structure of POTRA1–4 (Gatzeva-Topalova et al., 2010).

NMR RDC studies are often used to probe rigidity/flexibility in multi-domain proteins (Fischer et al., 1999; Tolman et al., 2001). A standard approach is to calculate the magnitudes, |D_a|, and rhombicities, R, of alignment tensors of individual domains where different values for different domains indicate that these domains are not behaving as a single rigid unit. The D_a and R values for individual domains in POTRA1–5 are given in Table 1 and indicate that POTRA1–2 and POTRA4–5 act as rigid independent units. One complication in this type of analysis is that, even in a rigid structure, small differences in the crystal and solution structure can lead to significant changes in the predicted RDC values, introducing so-called “structural noise” into the analysis (Zweckstetter and Bax, 2002). To illustrate, Figures S6 and S7 show the fits of the measured RDC data with those predicted from the crystal structure for the individual domains in the Pf1 and C12E5/hexanol alignment media, respectively. In Pf1 medium, POTRA1, POTRA2, POTRA4 and POTRA5 show good fits of the observed and predicted RDCs (RMS errors between 2.0 and 3.0). The RMS error for POTRA3 is twice that of the other domains, likely arising from exchange broadened peaks and lower signal-to-noise. The RMS errors are generally larger for the C12E5 medium due to lower signal-to-noise for the data in this medium and thus somewhat larger errors in the RDCs. Thus, the D_a and R values in the POTRA domains were only qualitatively analyzed here to probe flexibility in POTRA1–5. Since determination of the D_a values is less susceptible to errors arising from structural noise than R values (Zweckstetter and Bax, 2002), the former were the primarily focus in the analysis of RDC data for POTRA1–5. As seen in Table 1 and Figure 4, the POTRA1 and POTRA2 have similar D_a values as do POTRA3, POTRA4 and POTRA5 (in both alignment medium). However the D_a values for POTRA1/2 differ substantially from POTRA3/4/5. Note that POTRA3 has a fewer RDCs in both alignment media due to exchange broadening of resonances and therefore fewer assignments (Figures S1 and S3), thus it is harder to make as strong conclusions for POTRA3 as the other domains. In conclusion, the RDC data here are consistent with POTRA1–2 and POTRA4–5 acting as rigid subdomains in POTRA1–5 and indicate that at least a majority of POTRA3 is reorienting with the POTRA4–5 subdomain.

Rigidity for POTRA4–5 fragment is consistent with all the previous RDC, NOE, SAXS and X-ray data; however, there are mixed data concerning the rigidity of POTRA1–2. A 2008 NMR study of POTRA1–2 suggested interdomain flexibility based on a lack of observed NOEs between the domains (Knowles et al., 2008). However, a subsequent electron paramagnetic resonance (EPR) study concluded that POTRA1 and POTRA2 have a fixed orientation in solution (Ward et al., 2009). Analysis of multiple X-ray structures containing POTRA1–2 shows low root-mean-square deviation for superimposition of POTRA1–2 (Figure 1b) and supports single conformation for these domains. Thus, the RDC analysis performed here, along with previous EPR data and crystallographic data support the conclusion that POTRA1–2 exists as a rigid subdomain in POTRA1–5.

The NMR studies here demonstrate flexibility between the rigid POTRA1–2 and POTRA4–5 domains, which could arise from flexibility at the interface between POTRA2–3, the interface between POTRA3–4, or within POTRA3 itself. Multiple types of NMR data here support that the flexibility is between POTRA2–3. First, the chemical shifts of the backbone amides of POTRA2 were similar in the presence or absence of POTRA3, whereas this was not true for the POTRA3–4 interface (Figure 2). Chemical shifts are very sensitive to local conformation and these data strongly support there is little or no interaction between POTRA2 and POTRA3. Second, the average residue specific rotational correlation times differ between POTRA1–2 and POTRA3–5 with these times being ≈50% larger for POTRA3–5 compared to POTRA1–2 (Figure 3b). Since each POTRA domain is the same size, this again supports that POTRA3 is reorienting with the POTRA4–5 subdomain. Lastly, the D_a values for POTRA3 are more similar to those for POTRA4 and POTRA5 than those in POTRA1 and POTRA2, in two different alignment media (Table 1). This indicates that the regions of POTRA3 where resonances assignments could be made are reorienting with POTRA4–5. A detailed analysis of the POTRA domain interfaces previously reported (Gatzeva-Topalova et al., 2008) describes several stabilizing interactions between POTRA1 and 2 as well as POTRA3 and 4 whereas the POTRA2–3 interface lacks significant stabilization. This provides a structural rationale for the dynamics observed in the NMR experiments presented here.

Solid-state NMR has been previously used to probe the structure and dynamics of BamA. Baldus and coworkers performed solid-state NMR on a precipitate of full-length BamA and observed fast (nanosecond to microsecond) dynamics for the 5 POTRA domains (Renault et al., 2011). However, their subsequent study of this system in liposomes indicated that the POTRA domains are rigid with no evidence of fast motions (Sinnige et al., 2014). The latter study differs from the solution NMR results here, where POTRA1–2 and POTRA4–5 form rigid subdomains and multiple lines of evidence show flexibility at the POTRA2–3 interface. One possibility for this difference is that including the β-barrel transmembrane domain of BamA at the C-terminus of POTRA5 directly rigidifies the other POTRA domains. However, this is unlikely given the POTRA domains do not have any long-range interactions due to its “beads-on-a-string” type structure. Another possibility is that one or more of the POTRA domains are interacting with lipids in the bilayer, which leads to the rigidity observed in the solid-state NMR studies. It is unlikely that lipid interactions in POTRA5 or POTRA4 would reduce the flexibility at the POTRA2–3 interface, given the non-globular structure for POTRA1–5. However, lipid-protein interactions directly with POTRA1 or POTRA2 could lead to reduced flexibility and therefore to the differences observed between the solid-state and solution NMR experiments (Fleming et al., 2016). Another possibility is that molecular crowding on the bilayer surface reduces flexibility. Very high protein-to-lipid ratios are required for adequate signal-to-noise in the solid-state NMR experiments (Sinnige et al., 2014); thus, the POTRA domains of neighboring BamA proteins in the bilayer could be bumping into one another leading to a more rigidified structure than what is observed in solution, as recently reported in a molecular dynamics study of BamA embedded in a the outer membrane (Fleming et al., 2016). Solution NMR studies have recently been performed on a BamA construct in a nanodisc membrane mimic (Morgado et al., 2015). This system consisted of the full-length β-barrel domain of BamA but only had the POTRA5 or POTRA4–5 periplasmic domains. It was found that POTRA5 does not form a folded structure in the absence of POTRA4, which is consistent with previous studies on isolated POTRA5 and POTRA4–5 in solution (Sinnige et al., 2015). It would be interesting to measure the <τ_c,> values for the individual POTRA domains in the full-length BamA in a nanodisc to see if there is flexibility at POTRA2–3 interface, as was observed here, in the presence of bilayer.

While the presence of interdomain dynamics has been demonstrated for many multi-domain proteins (Bertini et al., 2004; Blackledge, 2010; Diaz-Espinoza et al., 2007; Goto et al., 2001; Maciejewski et al., 2011; Ryabov and Fushman, 2006), the functional significance of these motions is often difficult to establish experimentally. Thus, here we complemented the NMR studies showing flexibility with an in vivo study of BamA mutants to probe the functional consequences of disrupting this flexibility.

Cysteine residues were introduced in POTRA2 and POTRA3 such that, upon disulfide bond formation, the flexibility of the BamA POTRA2–3 interface would be restricted. Two sets of double-cysteine mutants were designed to allow disulfide formation based on the compact and extended conformations observed in the crystal structures. Both mutants displayed impaired activity in vivo as judged by complementation assays (Figure 5). Although these assays have the advantage of testing activity in the full cellular environment under physiological conditions, it is difficult to quantitatively correlate the observed growth phenotype with the specific activity of BamA. For example, a mutation in the BamA promoter that reduces expression at least five fold does not display a growth phenotype (Aoki et al., 2008). This means that the mutant with approximately one fifth of the steady-state levels of BamA grows just like wild-type in standard rich media. Therefore, the reduction in growth rates observed for the double-cysteine mutants of BamA are likely due to substantial reductions in BamA specific activity.

The E. coli periplasmic environment is not highly oxidizing under normal growth conditions, often leading to incomplete disulfide bond formation of double-cysteine mutants. For example, ≈50% disulfide crosslinking efficiency was observed between cysteines introduced in the first and last strands of the BamA β-barrel (Noinaj et al., 2014) or between cysteines engineered in BamA and BamB (Jansen et al., 2015) or BamA and BamD (Bergal et al., 2016). The data here showed efficient cysteine labeling with maleimide-biotin for the BamA double-cysteine mutants in cells grown in the presence of TCEP and markedly reduced labeling in its absence (Figure 6a). This indicates that under normal growth conditions (no TCEP) the cysteines are mostly engaged in disulfide bonds that the complementary experiments showed to be intramolecular disulfides (Figure 6b). Nevertheless, the clear reduced-growth phenotype of the double-cysteine mutants was reproducible and importantly, completely reversed when the cells were grown in the presence of TCEP (Figure 5). These experiments demonstrate that intramolecular disulfides form between POTRA2 and POTRA3 in the double-cysteine mutants, substantially reducing the BamA activity and leading to reduction of growth rate, whereas reduction of disulfides with TCEP restores BamA activity.

The recent reports of BAM complex X-ray structures (Bakelar et al., 2016; Bergal et al., 2016; Chen et al., 2016; Gu et al., 2016) allow us to hypothesize on the role that BamA flexibility plays in BAM function. As seen in Figure 8, BamD interacts with the POTRA5 and the POTRA1–2 domains of BamA forming a ring in the periplasmic side of the BAM complex. This ring is adjacent to the membrane and is thought to bind nascent OMPs (Bakelar et al., 2016; Bergal et al., 2016; Gu et al., 2016; Han et al., 2016). Flexibility at the hinge between POTRA2 and POTRA3 (labeled with an arrow in Figure 8) could then regulate access to the ring and/or modulate its dimensions. This may allow the BAM complex to accommodate nascent OMPs of different sizes. BamB interacts primarily with POTRA3, but it is positioned right at the hinge between POTRA2 and POTRA3 and has been previously proposed to modulate BamA flexibility (Jansen et al., 2012, 2015). We hypothesize that the hinge at POTRA2–3 allows POTRA1–2 to switch between binding of BamD (shown in Figure 8) and/or binding to BamB; therefore changing the conformation of the BAM periplasmic ring. This conformational change could then be part of the BamA-catalyzed cycle of OMP folding and insertion, and would rationalize the phenotype observed here in the double-cysteine mutants that restrict POTRA2–3 hinge flexibility.

Figure 8.

Structure of the BAM complex. (a) Side view, parallel to the membrane plane, of the BAM complex with BamA in green, BamB in yellow, BamC in light cyan, BamD in orange and BamE in blue. POTRA domains 1–3 are labeled P1-P3 and the hinge between POTRA2 and POTRA3 is indicated with an arrow. (b) Bottom view of the BAM complex, looking perpendicular to the membrane plane from the periplasmic side. POTRA domains 1–5 are labeled P1-P5 and colors and labels are as in (a).

Materials and Methods

Preparation of isotopically labeled BamA POTRA constructs

Plasmids pMS487 (BamA POTRA4–5; BamA_264–424), pMS488 (BamA POTRA3–5; BamA_172–426), pMS679, or (BamA POTRA1–5; BamA_25–424) were transformed into E. coli Rosetta (DE3) cells (Novagen). Expression and purification of ²H, ¹³C, ¹⁵N-labeled proteins was performed as described in Supplemental Information.

NMR experiments on BamA POTRA constructs

NMR spectra for the backbone (¹H^N, ¹⁵N^H, ¹³C^α, ¹³C^β) resonance assignments were made as described in Supplementary Material. The amide ¹H-¹⁵N RDCs on POTRA1–5 were measured in two different alignment media, Pf1 phage and 3% n-dodecyl-penta(ethylene glycol)/n-hexanol (C12E5/hexanol) using the amide RDCs by TROSY method (ARTSY) as described in Supplemental Information (Fitzkee and Bax, 2010).

The data for determining {¹H}-¹⁵N heteronuclear NOEs and the ¹⁵N R₁ and R_1ρ relaxations rates were collected at 800 MHz on a 0.5 mM ²H, ¹³C, ¹⁵N-labeled sample of POTRA1–5 as described in Supplemental Information using standard pulse sequences (Cavanagh et al., 2007). R₂ values were estimated from R₁ and R_1ρ, neglecting relaxation from chemical exchange, using protocol described in Supplemental Information. Residue-specific τ_c values were calculated from the ¹⁵N R₁ and R₂ values for residues using Equation 1, which is a simplification of Equation 8 in (Kay et al., 1989).

$${\tau }_c=\frac{1}{4\pi {\nu }_N}\sqrt{6\frac{R_2}{R_1}-7}$$ Eq. 1

The HYDROPRO program (Ortega et al., 2011) was used to calculate hydrodynamics parameters for the various models of POTRA1–5 using coordinates for POTRA1–5 spliced from X-ray structures of POTRA1–4 (PDBID 3EFC) and POTRA 4–5 (PDBID 3OG5). The HYCUD program (Rezaei-Ghaleh et al., 2013) used these hydrodynamic parameters to predict the effective rotational correlation times for the POTRA1–2 and POTRA3–5 subdomains in POTRA1–5 at 30 °C, where these two subdomains were connected by a 4 amino acid unstructured region. These correlation times were determined using HYCUD where the structural input consisted of a set of 4 different orientations of the POTRA1–2 and POTRA3–5 subdomains (Rezaei-Ghaleh et al., 2013) which were generated from the spliced model (Gatzeva-Topalova et al., 2010). These structures spanned the full range of orientations previously used in modeling the SAXS data on POTRA1–5 (Gatzeva-Topalova et al., 2010).

The PyMOL program was used for visualization of the X-ray structures and preparation of the figures (The PyMOL Molecular Graphics System, Version 1.4, Schrödinger, LLC.).

Calculation of D_a and R from RDC data

The D_a and R for the individual POTRA domains were determined with singular-value decomposition (SVD) in XPLOR-NIH version 2.25 (Schwieters et al., 2006) using the experimental RDCs and crystal coordinates from either POTRA 1–4(PDBID 3EFC) or POTRA 4–5(PDBID 3OG5). Errors were estimated by bootstrapping with resampling as described in Supplemental Information.

In vivo complementation assay

Chemically competent E. coli JCM-166 cells were transformed with plasmid pZS21 encoding BamA WT, BamA(S143C-A220C), BamA(V144C-D211C) or the control protein GFP, and the cells were then plated on LB agar supplemented with 50 µg/mL kanamycin and 0.1% arabinose. One colony from each plate was used to inoculate separate 5 mL cultures of LB with supplemented with 50 µg/mL kanamycin and 0.05% arabinose and incubated overnight at 37 °C. The cells were pelleted and washed with 5 mL of LB to remove residual arabinose, resuspended in LB media and used to inoculate 3 mL cultures pre-warmed to 37 °C such that the OD₆₀₀ was 0.025. The cultures were supplemented with 50 µg/mL kanamycin and either 0.05% arabinose, 0.05% arabinose + 5 mM TCEP, 0.05% fucose, or 0.05% fucose + 5 mM TCEP. The OD600 was taken every hour. When the OD₆₀₀ reached 0.5–0.6, cells were diluted to an OD₆₀₀ of 0.025 in 3 mL with the same media. All experiments were carried out in duplicate.

Biotin-maleimide Labeling and Western Blotting

Detailed protocols for these experiments are included in the Supplemental Information.

Highlights.

• BamA with a transmembrane β-barrel and soluble POTRA motifs mediates OMP biogenesis

• NMR relaxation studies show POTRA1–2 is flexibly linked to POTRA3–5 in BamA

• Residual dipolar couplings indicate POTRA1–2 and POTRA4–5 behave as rigid species

• Disulfides that restrict POTRA2–3 flexibility impair in vivo function of BamA