Biochemical Journal

Research article

Autotransporter passenger domain secretion requires a hydrophobic cavity at the extracellular entrance of the β-domain pore

Yujia Zhai , Kai Zhang , Yanwu Huo , Yanshi Zhu , Qiangjun Zhou , Jiuwei Lu , Isobel Black , Xiaoyun Pang , Aleksander W. Roszak , Xujia Zhang , Neil W. Isaacs , Fei Sun


Whooping cough (pertussis) is a highly contagious acute respiratory illness of humans caused by the Gram-negative bacterial pathogen Bordetella pertussis. The AT (autotransporter) BrkA (Bordetella serum-resistance killing protein A) is an important B. pertussis virulence factor that confers serum resistance and mediates adherence. In the present study, we have solved the crystal structure of the BrkA β-domain at 3 Å (1 Å=0.1 nm) resolution. Special features are a hairpin-like structure formed by the external loop L4, which is observed fortuitously sitting inside the pore of the crystallographic adjacent β-domain, and a previously undiscovered hydrophobic cavity formed by patches on loop L4 and β-strands S5 and S6. This adopts a ubiquitous structure characteristic of all AT β-domains. Mutagenesis studies have demonstrated that the hairpin-like structure and hydrophobic cavity are crucial for BrkA passenger domain (virulence effector) translocation. This structure helps in understanding the molecular mechanism of AT assembly and secretion and provides a potential target for anti-pertussis drug design.

  • autotransporter
  • Bordetella pertussis
  • membrane protein
  • outer membrane translocation
  • passenger domain secretion
  • trypsin-accessibility assay


Whooping cough (pertussis) is a highly contagious acute respiratory illness of humans caused by the Gram-negative bacterial pathogen Bordetella pertussis. Although it was long considered a childhood illness, in reality it can affect all age groups. Despite global childhood immunization, the disease remains fifth among vaccine-preventable death aetiologies in children under 5 years of age [1]. Worldwide, it is estimated that approximately 20–40 million cases occur every year, with 200000–400000 childhood deaths [2]. Although ~90% of the deaths occur in the developing world, there has been an increasing trend in reported pertussis infections in countries with high vaccination coverage, especially in young infants, adolescents and adults [3]. Research efforts are thus being directed at reinforcing existing vaccination strategies and finding new preventive drugs to combat B. pertussis [4].

BrkA (Bordetella serum-resistance killing protein A) is an important B. pertussis virulence factor that confers serum resistance and mediates adherence [5]. BrkA mutants of B. pertussis have been shown to be severely defective in their ability to colonize mice and to cause lethal infection in the infant mouse model [57]. Recent studies have shown that the recombinant BrkA protein is a promising candidate antigen to improve existing Pa (acellular pertussis) vaccines in humans [8].

BrkA is expressed as a 103 kDa precursor (preBrkA) that is processed during secretion to a 42-amino-acid signal peptide, a 73 kDa passenger domain and a 30 kDa β-domain [5,911]. The translated nascent preBrkA is targeted to and translocated across the IM (inner membrane) via a classical Sec-dependent process initiated by the signal peptide [1214]. After translocation across the IM, the signal peptide is cleaved, yielding the mature precursor (proBrkA) in the periplasm. After the β-domain of proBrkA integrates into the OM (outer membrane), the passenger domain is translocated across the OM and self-cleaved at a site between Asn731 and Ala732 by an unusual catalytic dyad formed by the cyclization of asparagine [15]. After translocation across the OM, the passenger domain remains tightly associated with the bacterial surface [16]. The β-domain of BrkA shows a pore-forming ability in black lipid bilayer experiments and has 55% sequence identity with the β-domain of pertactin, another highly immunogenic virulence factor of B. pertussis [10].

BrkA belongs to the AT (autotransporter) superfamily whose members are often virulence proteins secreted by Gram-negative bacteria. All ATs contain an N-terminal cleavable signal peptide, a functional passenger domain and a C-terminal β-domain [17,18]. Although the size of the passenger domains is highly variable and often exceeds 100 kDa, all available evidence indicates that the passenger domains of classical ATs adopt a β-helix structure [1923]. The β-domains of classical ATs are relatively uniform in size (~300 amino acids) and essential for passenger domain translocation across the OM. The crystal structures of the β-domains of NalP and EspP, two classical ATs from Neisseria meningitidis and Escherichia coli strain O157:H7 respectively and EstA, a full-length AT esterase from Pseudomonas aeruginosa have been reported [2426]. All of these AT structures have a common 12-stranded β-barrel domain with a N-terminal α-helix located in the hydrophilic central pore. There are also several disease-related virulence passenger domains whose crystal structures have been determined [19,20,23,27]. However, the mechanism used to translocate the passenger domains across the OM still remains unclear. Four major models have been proposed for the AT passenger domain translocation mechanism: (i) the threading model, (ii) the hairpin model, (iii) the multimer model, and (iv) the Omp85 (outer membrane protein 85)-mediated model [28]. The threading model, whereby translocation is initiated by the insertion of the N-terminus of the passenger domain into the β-domain pore, has been questioned, as there are no specific N-terminal motifs found to initiate the insertion [29]. The hairpin model involves the formation of a hairpin-like structure at the C-terminus of the passenger domain followed by a sliding of the proximal polypeptide through the pore [18,30]. The multimer model argues that the passenger domain (which may be folded at least partially in the periplasm) is transported across the OM through a large common channel formed by an oligomerization of at least six β-domains [31]. In the Omp85-mediated model, exogenous factors such as the Omp85-containing OM complex interact with the β-domain of AT and drive the passenger domain translocation [28].

A thorough understanding of how virulence factors are secreted and what elements of the β-domain are important for passenger domain translocation is of great importance for drug development, as this domain is a potential target for the design of compounds to inhibit the translocation of virulence factors across the OM. In the present study, we solved the crystal structure of the BrkA β-domain (BrkAβ) at a resolution of 3 Å (1 Å=0.1 nm). Structural analysis and mutagenesis studies revealed a hydrophobic cavity, at the extracellular entrance of the β-domain, which is crucial for passenger domain translocation. Our study gives new insights into the molecular mechanism of AT secretion and provides a potential target for anti-pertussis drug design.


Protein expression, refolding and purification

The DNA fragment encoding BrkAβ (residues Ala728–Phe1010) was constructed in expression vector pET11a (Novagen). The protein was overexpressed as inclusion bodies in E. coli strain BL21(DE3). The cells were washed in buffer A, containing 50 mM Tris/HCl (pH 8.0), 200 mM NaCl and 10 mM benzamidine, and lysed by sonication in buffer A containing 1% (v/v) Triton X-100. After centrifugation at 20000 g for 30 min, the pellet was washed with buffer A to remove the detergent. Inclusion bodies containing BrkAβ were solubilized in buffer A with the addition of 8 M urea. Refolding was initiated by rapid 10-fold dilution of the urea-solubilized protein into buffer A with 0.5% LDAO (N,N-dimethyldodecylamine-N-oxide) added. Refolding was carried out for 72 h at 37°C. The protein was subsequently dialysed against buffer C, containing 25 mM Tris/HCl (pH 8.0), 100 mM NaCl, and 0.06% LDAO. sulfopropyl-Sepharose chromatography was performed on an ÄKTA prime instrument (GE Healthcare) in buffer C, and a linear gradient up to 500 mM NaCl was used to elute the target protein. BrkAβ was purified further by ion-exchange chromatography with a Resource S column (1 ml; GE Healthcare) that was equilibrated in 20 mM Hepes (pH 7.5) and 0.06% LDAO on a Biologic Duoflow system (Bio-Rad Laboratories). Applying a very small elution gradient of NaCl from 100 mM to 250 mM, two peaks were eluted and fractionated. Fractions in the first peak were collected and purified further by applying to a Mono S column (1 ml; GE Healthcare), during which the detergent LDAO was exchanged to 0.6% C8E4 [tetra(ethylene glycol) (mono-octyl)ether]. The purified BrkAβ protein was then concentrated and used for crystallization.

SEC-MALS (size-exclusion chromatography with multi-angle light-scattering detection)

The instrumental set up used for SEC-MALS experiments consisted of an Agilent Technologies 1100 HPLC system connected in series with a DAWN® HELEOS™ II light-scattering detector (Wyatt Technology) and an Optilab rEX interferometric refractometer detector (Wyatt Technology). The purified BrkAβ protein was diluted to a concentration of ~5 mg/ml in SEC (size-exclusion chromatography) buffer 1 (20 mM Hepes, pH 7.5, 150 mM NaCl and 0.06% LDAO). Analytical SEC was performed at 16°C using a Shodex Protein 803 KW column (Shoko America) equilibrated with SEC buffer 1. A 50 μl volume of diluted protein was injected into the column and eluted at a flow rate of 0.5 ml/min. The column effluent was monitored in-line with three detectors that simultaneously monitored UV absorption, LS (light scattering) and the RI (refractive index). Detector outputs were digitized and acquired by a computer running ASTRA software (Wyatt Technology). The resulting three chromatograms were aligned after correcting the inter-detector volume delays between the UV absorption and LS and between the RI and LS detectors. The oligomeric molecular mass of BrkAβ at the eluted peak was calculated using ASTRA software as described previously [32,33]. Two further runs with selected detergents were performed in SEC buffer 2 (20 mM Hepes, pH 7.5, 150 mM NaCl and 0.02% dodecyl maltoside) and SEC buffer 3 (20 mM Hepes, pH 7.5, 150 mM NaCl and 0.6% C8E4) respectively.

Crystallization and heavy-atom soaking

The solution concentration of BrkAβ was assayed by the BCA (bicinchoninic acid) method then diluted to ~25 mg/ml with 20 mM Hepes (pH 7.5) and 0.6% C8E4. A second detergent, DHDM (2,6-dimethyl-4-heptyl-β-D-maltopyranoside), was added into the protein solution with a final concentration of 55 mM. Crystals were grown at 18°C by the hanging-drop vapour-diffusion method. The best crystals were grown from a mixture of 1 μl of protein solution and 1 μl of mother liquor {100 mM Tris, pH 8.5, 17% PEG [poly(ethylene glycol)] 2000, 200 mM LiCl and 15 mM spermidine} over a 200 μl reservoir solution (100 mM Tris/HCl pH 8.5, 28% PEG2000 and 200 mM LiCl). To screen for heavy-atom compounds bound to BrkAβ, a native gel-shift assay was performed. The heavy-atom derivatives were prepared by adding K2Pt(NO2)4 directly into the crystallization drop with a final concentration of 5 mM and soaking for 3 weeks. Crystals were collected and frozen directly in liquid nitrogen without additional cryoprotectant. Heavy-atom soaking improved the diffraction quality with smaller mosaicity and better resolution from 3.5 to 3.0 Å at the synchrotron, although no heavy atoms were found in the final electron density. We speculate that it was the dehydration effect, rather than heavy-atom binding, that improved the crystal quality.

Data collection and structure determination

Several native datasets were collected from the preliminary crystals at the ESRF (European Synchrotron Radiation Facility), Grenoble, France, but the crystals showed twinning and attempts to phase the data by MR (molecular replacement) failed. K2Pt(NO2)4-soaked crystals were subsequently used to collect SAD (single anomalous diffraction) data at 100 K at beamline BL17U, SSRF (Shanghai Synchrotron Radiation Facility), Shanghai, China. The wavelength is 1.07190 Å at the platinum absorption peak. To avoid the CCD (charge-coupled device) detector saturation at low-resolution diffraction while using strong beam intensity and recording high SNR (signal-to-noise ratio) data for high-resolution diffraction, two separate datasets were collected. One dataset was collected with a high beam attenuation (0.6 mm thickness) and a large beamstop-to-crystal distance to ensure accurate intensity detection at low resolution (to 45 Å). The second dataset was collected with small beam attenuation (0.4 mm thickness) and a short beamstop-to-crystal distance to stop diffraction lower than 15 Å and increase the SNR for high-resolution diffraction. As a membrane protein crystal, the solvent content of BrkAβ crystal is very high: 69.6% with the Matthews coefficient 4.04 Å3/Da (Table 1). As a result, the BrkAβ crystals were more sensitive to radiation damage and the total exposure time was carefully estimated using RADDOSE [34] to avoid severe radiation damage and crystal decay. Owing to one long unit cell axis of 406.6 Å, the orientation of the crystal was carefully manipulated to minimize the number of overlapping diffraction spots. All datasets were processed and merged using HKL2000 [35].

View this table:
Table 1 Data collection and refinement statistics

The corresponding parameters for the highest-resolution shell are shown in parentheses. RmergehΣi|Iih−<Ih>|/ΣhΣi<Ih>, where <Ih> is the mean intensity of the reflection h. Rwork=Σ(‖Fo|−|Fc‖)/Σ|Fo|; Rfree=Rfactor for a selected subset (5%) of the reflections that was not included in previous refinement calculations.

Although an anomalous signal was detected from the diffraction data, phasing was unsuccessful, owing to the low occupancy and high mobility of the heavy atoms. Eventually, the crystal structure of BrkAβ was determined by MR. The starting model used was the crystal structure of EspP (PDB code 2QOM) [25] with side chains truncated to alanine or glycine, except where sequence alignment suggested identity. The Matthews coefficient calculation suggested that there are possibly three or four molecules in one asymmetric unit with solvent content of 69.6 and 59.5% respectively. A self-rotation function computed by MOLREP [36] gave no significant peak for either three-fold or four-fold non-crystallographic symmetry. The best solution obtained from an MR computation, using MOLREP to search for four molecules in the asymmetric unit, exhibited severe steric clashes, invalidating the possibility of four molecules in one asymmetric unit. The molecule causing severe clashes was deleted and the remaining three (named as M1, M2 and M3; see Figure 2) used for further phasing. (Interestingly, the relative positions and orientations of these could not be obtained if MOLREP was used to search for three molecules in an asymmetric unit.)

Although the initial electron-density map had a good appearance in a number of regions, further density modification and phase refinement did not give significant improvement or decreased R-factors (Rwork/Rfree=42.9/49.1). An examination of the map showed good density for M2 and M3, but relatively poor density for M1. Molecules M2 and M3 had a shoulder-to-shoulder packing similar to that of EspP, and molecule M3 and its crystallographic symmetric neighbour M4 (see Figure 2) were packed head-to-head. Similar head-to-head packing between M1 and M2 could be obtained by rotating molecule M1 by 180°, exchanging the extracellular and intracellular ends. Working on the hypothesis that, in Nature, the crystal packing should the simplest available, we inverted the orientation of molecule M1 and used this modified MR solution for further phasing. Rigid-body refinement of this model decreased the R-factors significantly (Rwork/Rfree 35.6/40.5), proving our hypothesis. The density map was dramatically improved further by imposing the NCS (non-crystallographic symmetry) averaging combined with phase extension from 8.0 to 3.0 Å. The averaged density map showed very good connectivity, and most residues could be traced and docked into the sequence. Conserved residues between EspP and BrkAβ and large residues such as tryptophan and tyrosine were assigned and used as a reference to locate other residues during model building. Cycles of structure refinement were performed using CNS [37] with the NCS restriction. A negative B-factor (−50 Å2) was applied to sharpen the 2FoFc map that was used to rebuild the model and correct some errors. After further model building and refinement by CNS, Refmac [38] was used for a final TLS (Translation Liberation Screw-motion) refinement without the NCS restriction. The crystal structure of BrkAβ was finally refined to 3.0 Å with Rwork of 26.6%, Rfree of 32.4% and an overall FOM (figure of merit) of 0.78. The final electron-density map has good consistency with the structural model (see Supplementary Figure S1 at The average B-factor of the final refined structure is 93.1 Å2, close to the calculated Wilson B-factor of 97.3 Å2. The high B-factor is thought to be due to the high solvent content and radiation damage. No water molecules were added into the model owing to the edge resolution of 3.0 Å. There is an elongated stretch of electron density present in the central pore of the structure, which is too smeared to trace any residues (see Supplementary Figure S2 at The statistics for data collection, processing and structure refinement are summarized in Table 1.

Mutagenesis, trypsin accessibility assay and Western blot analysis

The DNA fragments of BrkAΔ-PC and its 12 mutants (BrkAΔ-1–12) were cloned (see Figure 4A and Table 2) by the overlapping PCR method from the plasmid pD06935 containing the full-length BrkA [9] and inserted into the vector pEXS-DH [32]. E. coli C41(DE3) cells transformed with an empty vector or the vectors containing the above genes were grown to a D600 of 0.4 and induced by 0.02 mM IPTG (isopropyl β-D-thiogalactopyranoside) for 2 h at 37°C. Then, 2 ml of culture was harvested by centrifugation at 2000 g for 10 min and resuspended in 410 μl of PBS. For 200 μl of cells, 4 μl of 10 mg/ml trypsin was added. After incubation at 37°C for 10 min, the cells were put on ice immediately and 20 μl of FBS (fetal bovine serum) was added to stop digestion. The cells were pelleted by centrifugation at 2000 g for 5min at 4°C, washed three times with PBS containing 10% FBS and once in PBS alone. The washed cells were finally resuspended in 5× SDS loading buffer and boiled for 10 min. As a control, another 200 μl aliquot of cells was simultaneously processed in the same manner without added trypsin. Samples were resolved by SDS/PAGE (12% gels) and transferred on to nitrocellulose membranes (Millipore) for 70 min at 300 mA at 4°C. The non-secreted and secreted proteins were detected by anti-His antibody (Sigma) and horseradish peroxidase-conjugated goat anti-mouse secondary antibody (ZSGB-Bio) with a working dilution of 1:3000 and 1:2500 respectively. The membranes were incubated in luminol reagent (Santa Cruz Biotechnology) and exposed to film.

View this table:
Table 2 Description of mutant BrkA constructs

Bacterial membrane extraction and separation of OM and IM

The OM of E. coli expressing BrkAΔ-PC or its mutants was prepared using the methods of Filip et al. [39] and Kumar et al. [40]. Briefly, E. coli strain C41(DE3) cells transformed with an empty vector, or the vectors containing the above genes, were grown at 37°C in 500 ml of 2YT [1.6% (w/v) tryptone/1% (w/v) yeast extract/0.5% NaCl]/ampicillin medium. When the D600 of the culture reached ~0.8, IPTG was added to a final concentration of 0.02 mM and the culture was incubated further at 37°C for 2 h. Cells were harvested by centrifugation at 6300 g for 15 min (Beckman JA10 rotor), washed with ice-cold PBS containing 10% (v/v) glycerol and 1 mM PMSF, and resuspended in 20 ml of TE buffer (20 mM Tris/HCl, pH 8.0, and 10 mM EDTA) with protease inhibitors (Roche) added. After sonication on ice, the lysate was centrifuged at 8600 g for 10 min at 4°C (Beckman JA25.5 rotor). The supernatant was collected and centrifuged at 20000 g for 20 min at 4°C (Beckman JA25.5 rotor). The supernatant was collected again and ultracentrifuged at 35000 rev./min for 1 h at 4°C (Beckman Ti 70 rotor). The pellet, which is the total membrane fraction, was homogenized with 10 ml of TE buffer. Then the sample was ultracentrifuged again at 35000 rev./min for 35 min at 4°C (Beckman Ti 70 rotor). The pellet was homogenized with 4 ml of TE buffer, and the total protein concentration was determined using the BCA method. The protein concentration was adjusted to 5 mg/ml and solubilized by sarkosyl with a final concentration of 1% (w/v) at 4°C for 1 h. In this step, the IM could be resolved into the solubilized supernatant. After ultracentrifugation at 35000 rev./min for 1 h at 4°C (Beckman Ti 70 rotor), the sarkosyl-insoluble OM fraction was homogenized in 2 ml of TE buffer. The total protein concentrations of both the IM and OM fractions were determined using the BCA method and adjusted to 1 mg/ml using TE buffer. Samples were resolved by SDS/PAGE (12% gels) and subjected to Western blot analysis.


Overall structure of BrkAβ

Purified refolded BrkAβ was crystallized in space group C2221, and its crystal structure was determined by MR and refined at 3.0 Å resolution with Rwork 26.6% and Rfree 32.4% (Table 1). The overall structure of BrkAβ (residues Ala728–Phe1010) is a 12-stranded antiparallel β-barrel with a height of ~70 Å and an ellipsoidal cross section of ~19 Å×27 Å (Figure 1A). The interior of the β-barrel is packed with 37 hydrophilic and 34 hydrophobic residues. The exterior of the β-barrel is composed of a hydrophobic belt that is lined by two girdles of aromatic residues at the top and bottom (Figure 1A). The width of the hydrophobic belt (~30 Å) is complementary to the non-polar portion of the OM (Figure 1A). The β-barrel has a shear number of 14, and the β-strands are tilted at an angle of ~45° to the barrel axis (Figure 1B). The β-strands (S1–S12) range in length from 11 to 22 residues and are connected by short periplasmic turns (T1–T5) and extracellular loops (L1–L6) of various lengths (Figure 1). The longest loop (L4 or L5) contains 20 residues, whereas the shortest loop (L6) contains only two residues. L1 and L5 cannot be completely built because of the poor electron density, indicating their structural flexibility. In L4, two antiparallel β-strands form a small β-sheet and protrude into the extracellular space. A short loop before the first strand of the β-barrel is inserted into the central pore from the periplasmic side. Although an elongated stretch of electron density is present in the pore, it is too smeared to trace any residues before Asp745 (Figure 1B and see Supplementary Figure S2). Since black lipid bilayer experiments indicated the processing of the passenger domain of BrkA occurs at the site between Asn731 and Ala732 [10,11,16], i.e. before Asp745, it is further believed that the passenger domain is cleaved inside the central pore of the BrkA β-barrel.

Figure 1 Overall structure and topology of BrkAβ

The β-strands (S1–S12) of the barrel are coloured green, extracellular loops (L1–L6) and periplasmic turns (T1–T5) are coloured pink, the small β-sheet of L4 is coloured blue, untraced residues of L1 and L5 are coloured yellow, and the segment inserted into the pore is coloured cyan. All of the strands, loops and turns are labelled. (A) Cartoon representation of BrkAβ overall structure produced using Molscript [45] and Bobscript [46]. Side chains of the aromatic girdle are shown in gold sticks. (B) Topology diagram of BrkAβ. The residues are numbered according to the full-length mature BrkA protein. The residues facing the barrel pore are highlighted by lime squares and those facing the lipid bilayer by light green squares.

A hairpin-like structure formed by L4 inserts into an adjacent pore for crystal packing

In the crystal structure of BrkAβ, there are three molecules per asymmetric unit. They adopt almost identical conformations and superimpose well with the RMSD (root mean square deviation) less than 0.8 Å (see Supplementary Figure S3 at Within one asymmetric unit, two molecules (M1 and M2) form a pseudo-dimer (D1), whereas the third (M3) forms an identical pseudo-dimer (D2) with a molecule from an adjacent asymmetric unit (M4) (Figure 2). Within such a pseudo-dimer, the small β-sheet of L4 from one monomer inserts into the hydrophobic cavity of the other monomer and forms a hairpin-like structure located exactly over the extracellular entrance of the β-domain pore. The side chains of Tyr898, Ala900, Leu904, Val906 and Val908 on the small β-sheet are involved in hydrophobic interactions with the hydrophobic patches on S5 (Tyr845 and Ile847) and S6 (Val855, Ala857 and Tyr859) of the second molecule. Although there are tight hydrophobic interactions within these pseudo-dimers, SEC-MALS demonstrated that BrkAβ exists as a monomer under different detergent conditions, even for the detergent C8E4 that was used for crystallization (see Supplementary Figure S4 at That suggests the dimer formation observed results from the crystal packing. The monomeric state of BrkAβ is in accordance with other reported AT β-domains [24,41,42]. However, the fact that L4 could be inserted into the adjacent monomer at least indicates that the hairpin structural motif could be accommodated properly within the central pore, which has been suspected for a long time.

Figure 2 Crystal packing results in a hairpin-like structure above the central pore

Crystal packing of BrkAβ in space group C2221. Molecules in one asymmetric unit are labelled and shown in blue (M1, M2 and M3) and their neighbours are in red (M4, M5 and M6). The two types of pseudo-dimers (D1 and D2) are indicated by blue and pink squares respectively. The detailed hydrophobic interactions within D2 are shown in the top-right and bottom-right insets. The bottom-right inset is rotated vertically anticlockwise by 90° with respect to the top-right one. Side chains of the hydrophobic residues involved in the D2 interaction interface are shown as sticks, coloured light magenta for M3 and light yellow for M4 and labelled accordingly. The left-hand panel for crystal packing was made by using UCSF Chimera [47] and the insets were produced using PyMOL (

Structural comparison with other AT β-domains

Before the present study, there were only three available structures for AT β-domains: Nalpβ from N. meningitidis (PDB code 1UYN) [24], EspPβ from E. coli strain O157:H7 (PDB code 2QOM) [25] and EstAβ from Ps. aeruginosa (PDB code 3KVN) [26]. Although BrkAβ shares low sequence homology with them (<25% identity), its overall structure is similar to Nalpβ, EspPβ and EstAβ with Cα RMSDs of 2.3, 1.7 and 2.6 Å respectively (see Supplementary Figure S5 at On L4 of BrkAβ, the small β-sheet (residues 896–908) contains alternating basic and hydrophobic residues. Those hydrophobic residues (Tyr898, Ala900, Leu904, Val906 and Val908) are oriented into the interior of the pore and form a hydrophobic cavity with the opposite extracellular hydrophobic residues (Tyr768, Tyr805, Tyr845, Ile847, Val855, Ala857 and Tyr859) on S2, S3, S5 and S6 (Figures 3A and 3B). Interestingly, similar hydrophobic regions are also found in other AT β-domains of Nalpβ, EspPβ and EstAβ (Figures 3C, 3D and 3E), but never reported before, presumably because of a lack of structures to be compared. Although the residues forming the hydrophobic regions are not conserved among these AT β-domains, the common hydrophobic cavity feature in their extracellular entrance of the pore suggests a relationship with the mechanism of AT secretion.

Figure 3 Extracellular hydrophobic cavities of AT β-domains

(A) The extracellular hydrophobic cavity in BrkAβ is shown with an electrostatic potential surface. (B) The labelled hydrophobic residues forming the cavity are shown as white sticks and the basic residues of L4 are shown as slate sticks. S2, S3, S5 and S6, and L4 are labelled accordingly. For comparison, electrostatic potential surfaces of Nalpβ (C), EspPβ (D) and EstAβ (E) respectively are shown in the same orientation as BrkAβ. The electrostatic potential was calculated by DelPhi [48] and all structures were produced using PyMOL (

Hydrophobic patches of L4, S5 and S6 are required for passenger domain translocation of BrkA

It has been reported that the cleaved passenger domain of BrkA remains tightly associated with the bacterial surface after secretion and is not detected in B. pertussis culture supernatants. As a result, both the unprocessed precursor and the cleaved passenger domain can be detected by anti-BrkA serum in the whole-cell lysates [16]. Furthermore, the surface-associated passenger domain could be accessed and digested by trypsin added to the culture. This kind of trypsin-accessibility assay has been widely used to study the secretion/translocation state of prokaryotic proteins [9,43,44].

The role of the hydrophobic cavity in the passenger domain translocation of BrkA could be studied by expressing BrkA and a series of mutants with modified hydrophobic patches (Table 2) in E. coli (a similar Gram-negative bacterium to B. pertussis with the same virulence factor secretion mechanism) and using the trypsin-accessibility approach to investigate whether the mutants have decreased or destroyed passenger domain translocation activity. However, overexpression of the full-length BrkA protein in E. coli is lethal [16], although it has been shown that the upstream 31–39-residue region preceding BrkAβ is essential and sufficient for passenger domain translocation [9]. Accordingly, we cloned a new construct BrkAΔ-PC for positive control (similar to the BrkA deletion construct E in [9]), in which Gln60–Asp544 in the BrkA passenger domain were deleted and a His6 tag was inserted between Gly59 and Ala545 to produce a 57 kDa precursor (preBrkA) with an unprocessed signal peptide, a 52 kDa mature full-length protein (BrkA) and a 22 kDa passenger domain (Figure 4A). As shown in Figure 4(B), for BrkAΔ-PC without trypsin treatment, both the 52 kDa full-length protein and the 22 kDa cleaved passenger domain can be detected by an anti-His antibody. Treating the whole cells with trypsin resulted in the complete digestion of the 22 kDa moiety, indicating that BrkAΔ-PC could successfully secrete its passenger domain to the bacterial surface in a manner similar to the wild-type BrkA. Because it will normally take several minutes to complete the passenger domain secretion process [29], the 52 kDa full-length protein represents the intermediates inaccessible to trypsin, which is located in the periplasm or only targeted to the OM without passenger domain translocation. After trypsin treatment, the amount of the 52 kDa band is decreased, which is believed to be due to further secretion occurring during trypsin treatment. The membrane localization of expressed BrkAΔ-PC was visualized by immunoelectron microscopy (see Supplementary Figure S6 at and confirmed by statistical nanogold counting (see Supplementary Table S1 at All of the above demonstrates that BrkAΔ-PC is a rational positive control to study the passenger domain translocation mechanism using the trypsin-accessibility assay.

Figure 4 Construction of BrkAΔ-PC, and trypsin-accessibility and membrane localization assays

(A) BrkA domain organization and the construct design of BrkAΔ-PC. SP, signal peptide. The inserted His6 tag is depicted as a grey square. Both the signal peptide-processing site and the passenger domain cleavage site are indicated by black arrows. All other labels are self-explanatory. (B) Results of trypsin-accessibility assay for BrkAΔ-PC and its mutants. Cells were processed in the presence (+) or absence (−) of trypsin as described in the Experimental section. (C) Detection of BrkAΔ-PC and its mutants in the OM. (D) Detection of BrkAΔ-PC and its mutants in the IM. Descriptions of all of the mutants in (B), (C) and (D) are found in Table 2. The band positions for the precursor (PreBrkA), the mature full-length protein (BrkA) and the cleaved passenger domain (passenger) are indicated by black arrows and labelled accordingly in (B), (C) and (D). Molecular masses are indicated in kDa.

To determine the effect of the small β-sheet of L4 on the translocation of the passenger domain, we made a truncation mutant BrkAΔ-1 in which Lys896–Val908 were deleted from BrkAΔ-PC (Figure 3B and Table 2). Although the 22 kDa passenger domain was detected for BrkAΔ-1, the ratio of its amount to that of the 52 kDa moiety is much lower than that of BrkAΔ-PC (Figure 4B), indicating that the translocation efficiency of BrkAΔ-1 passenger domain was largely reduced. To investigate whether the decrease in translocation efficiency would result from the removal of an extracellular loop, another truncation mutant BrkAΔ-2 was made in which Asp951–Gly963 of L5 were deleted from BrkAΔ-PC (Figure 1B and Table 2). No effect on passenger domain translocation of this mutant could be found (Figure 4B), which is consistent with that of EspPβ1 (βL5) [25]. These data indicate the small β-sheet of L4 is required for efficient translocation of the passenger domain.

To examine whether the hydrophobic or basic residues on the small β-sheet in L4 (Figure 3B) contribute to the translocation effect, we made three mutants from BrkAΔ-PC. One mutant, BrkAΔ-3, which contains three point mutations, R899D, R905D and R907D (Figure 3B and Table 2), can successfully translocate its passenger domain to the bacterial surface with no distinct difference from BrkAΔ-PC (Figure 4B). However, no cleaved passenger domain can be detected in the other two mutants, BrkAΔ-4 and BrkAΔ-5 (Figure 4B), in which the hydrophobic residues (Tyr898, Ala900, Leu904, Val906 and Val908) in the small β-sheet were each mutated to aspartate and lysine respectively (Figure 3B and Table 2). These data suggest that it is the hydrophobic surface and not the positive electrostatic surface of the small β-sheet that is essential for the translocation of the passenger domain.

To assess further the role of the extracellular hydrophobic cavity (Figure 3A) of BrkAβ for passenger domain translocation, mutations of other hydrophobic residues on the cavity were made from BrkAΔ-PC. Located in the central region of the cavity, the hydrophobic residues on S5 (Tyr845 and Ile847) and S6 (Val855, Ala857 and Tyr859) were each replaced by aspartate in BrkAΔ-6 and BrkAΔ-7 respectively (Figure 3B and Table 2). The severe translocation inability of those mutants (Figure 4B) revealed that the hydrophobicity of the cavity is required for translocation. To test this hypothesis further, more hydrophobic residues of the cavity were each mutated to aspartate simultaneously so that the cavity hydrophobicity could be reduced further. As expected, for those mutants BrkAΔ-10, BrkAΔ-11 and BrkAΔ-12 (Figure 3B and Table 2), no cleaved passenger domain can be detected at all (Figure 4B). In contrast, two single mutations, BrkAΔ-8 and BrkAΔ-9 (Figure 3B and Table 2), where only one selected hydrophobic residue on S2 or S3 in the cavity is mutated to an acidic residue, did not affect passenger domain translocation significantly (Figure 4B), possibly because the two residues are located in the peripheral region of the hydrophobic cavity and the single mutation does not affect its overall hydrophobicity. These data show that the hydrophobic surface of the cavity formed by at least Tyr845, Ile847, Val855, Ala857 and Tyr859 on S5 and S6 is essential for the translocation of the passenger domain.

To assess the possibility that the failure to detect the cleaved passenger domain in the whole-cell lysates of the above mutants is due to an attenuated interaction between the β-domain and the cleaved passenger domain that is eventually released to cell culture, the medium supernatants of cultured bacteria expressing BrkAΔ-PC and all its mutants were analysed by dot blotting. No cleaved passenger domain could be detected from the bacterial culture (results not shown).

Another concern might be whether those mutations affect the membrane localization of the expressed protein. Using the same protocol for BrkAΔ-PC (see Supplementary Figure S6), we took hundreds of electron micrographs to visualize the immunolabelled sections of bacteria that expressed the mutants. Compared with the bacteria containing only the empty vector, there are significant nanogolds localized on the bacterial membrane for all of the mutants, although there are also enough nanogolds counted in the plasma (results not shown). This suggests that enough protein for all BrkAΔ-PC mutants folds correctly and is efficiently localized on to the membrane. It should be noted that it is not unusual to find the BrkA protein in the plasma of E. coli (see Supplementary Figure S6) because overexpression of BrkA will yield abundant inclusion bodies in the plasma and the BrkAβ protein in the present study was actually purified and refolded from inclusion body.

Although the immunolabelling electron microscopy demonstrated the membrane localization of BrkAΔ-PC and its mutants, it could be argued that defects in passenger domain secretion might be caused by a failure of the mutants to be correctly targeted to the OM, rather than a dysfunction of the β-domain translocation ability. To clarify, the bacterial membranes were extracted and separated into the IM and OM fractions according to the previously reported and widely used protocol [39,40]. As shown in Figure 4(C), with the exception of BrkAΔ-5, the 52 kDa mature full-length protein of all other mutants could be clearly detected in the OM fraction, and, for most of them, the precursor proteins (57 kDa moiety) could not be detected at all. The cleaved passenger domain could also be clearly identified for those BrkA proteins (BrkAΔ-PC, BrkAΔ-2, BrkAΔ-3, BrkAΔ-8 and BrkAΔ-9) that have significant passenger domain translocation activity. Furthermore, all of the BrkA proteins could be found from the IM fraction (Figure 4D) and most of them exhibited two bands, one of which (57 kDa moiety) represents the precursor (preBrkA) with the signal peptide unprocessed and the other (52 kDa moiety) representing the mature full-length protein (BrkA) that has been translocated across the IM via the Sec-pathway and has not been targeted to the OM. Interestingly, although BrkAΔ-5 could not be found in the OM fraction, it can be clearly detected as the precursor form in the IM fraction (Figure 4D), which is consistent with the observation that only the precursor band could be found in the trypsin-accessibility experiments (Figure 4B).

Taking all of the above data together, we can conclude that, except for BrkAΔ-5, all the BrkA proteins in the present study can be correctly localized into the OM and the undetected cleaved passenger domain in the trypsin accessibility assay is due to the failure of the BrkAβ translocation ability, which is caused by the destruction of the hydrophobic cavity formed by L4, S5 and S6. For BrkAΔ-5, we speculate that changing the hydrophobic residues of L4 to basic ones (BrkAΔ-5, Table 2), not to acidic ones (BrkAΔ-4, Table 2), will alter the tertiary structure of BrkA significantly, preventing its correct translocation across the IM via a Sec-dependent pathway.

Implications for the secretion mechanism of ATs

We have described the crystal structure of BrkAβ and assessed the essential role of the hydrophobic cavity, formed by the hydrophobic patches of L4 (Tyr898, Ala900, Leu904, Val906 and Val908), S5 (Tyr845 and Ile847) and S6 (Val855, Ala857 and Tyr859), for the passenger domain translocation. Our experiments have shown that disruption of any one of these hydrophobic patches will result in a remarkable decrease in translocation efficiency. After a comparison with other available structures of the classical AT β-domains, we found that the hydrophobic cavity formed by S5, S6 and L4 exhibits a ubiquitous structural characteristic of the extracellular part of the β-domain, despite the low sequence homology (Figures 3A, 3C, 3D and 3E). This structural conservation might play a role in the secretion mechanism of AT passenger domain.

As addressed above, there are four proposed models for the AT passenger domain translocation: (i) the threading model, (ii) the hairpin model [18], (iii) the multimer model [31], and (iv) the Omp85-mediated model [28]. The lack of specific N-terminal motifs to target the pore makes the threading model unlikely [28,29]. The multimer model can also be challenged as a general mechanism, since most AT β-domains, as well as the BrkAβ in the present study, were reported in the monomeric state [24,41,42], and it is difficult for β-barrels with hydrophobic exteriors to form a hydrophilic channel [28]. Recent studies have provided evidence to support the Omp85-mediated model [29] and the hairpin model [30]. Ieva et al. [29] reported that the C-terminal polypeptide segment of the passenger domain is embedded into the β-domain pore in the periplasmic space before targeting the OM. Although this suggests that the passenger domains might be translocated across the OM by a concerted mechanism via exogenous factors such as the Omp85 complex, their data do not exclude the hairpin model. Recently, Junker et al. [30] introduced pairs of cysteine residues into the sequence of the pertactin passenger domain and showed that the C-terminus of the domain firstly emerges across the OM before forming a stable folded form. In the context of the hairpin model, this suggests that the C-terminus of the passenger domain forms a stable core first and thereby provides a template of β-strands to promote fast folding and translocation of the proximal N-terminal portions through the pore [30].

The results of the present study do not exclude either the Omp85-mediated model or the hairpin model, but provide more implications for these two mechanisms. Figure 4 shows that, although some mutants (BrkAΔ-4, BrkAΔ-10, BrkAΔ-11 and BrkAΔ-12) totally lose their passenger domain translocation activity, they could also be efficiently targeted into the OM, suggesting that altering the extracellular hydrophobic cavity of the β-domain does not affect its OM localization. If the Omp85-mediated model is correct, although alteration of the hydrophobic cavity might inhibit the C-terminal segment of the passenger domain from being correctly inserted into the β-domain pore, the inability to form a stable intermediate does not influence its interaction with an Omp85 complex and its integration into the OM. However, the question of why the passenger domain could not be translocated across the OM when the protein has been correctly loaded on to the OM remains. In contrast, our observations are more compatible with the hairpin model. First, although the hairpin-like structural motif observed in the present study results from crystal packing, it at least suggests that this kind of structural motif could be properly accommodated within the pore, which has been argued for many years. Secondly, following the work by Junker et al. [30] showing that the C-terminus is the first portion of the pertactin passenger domain to cross the OM before forming a stable structure, we speculate that the extracellular hydrophobic cavity might provide a favourable environment to anchor the hairpin-like hydrophobic structural motif and thereby promote fast and efficient folding of subsequently secreted N-terminal portions of the passenger domain. There remains the question of why the mutant BrkAΔ-4, in which the hydrophobic residues of L4 were replaced by acidic residues, can inhibit translocation more efficiently than BrkAΔ-1, an L4 deletion mutant. The reason might be that more electrostatic charges in the former mutant influence the overall hydrophobicity of the cavity more severely than a deletion of the loop. Presumably, the passenger domain secretion mechanism of AT might derive from both the Omp85 model and the hairpin model, as indicated previously by Bernstein (in Figure 3b of [28]). In the periplasm, the passenger domain C-terminal segment is initially incorporated into the β-domain pore and forms a hairpin-like structure that is protected by periplasmic chaperone proteins. Following this, the β-domain pore is integrated on to the OM with the aid of Omp85. After integrating into the OM, the hairpin-like structural motif will provide a template of β-strands to initiate passenger domain translocation whose completion could perhaps be driven by the β-helix folding of the passenger domain on the surface.

Many ATs have been shown, or proposed, to be virulence factors essential for the pathogenesis of disease-causing bacteria. Understanding how virulence factors are secreted and the elucidation of the atomic structures of ATs are of great importance. Inhibition of the virulence factor translocation may provide a possible therapeutic approach to block colonization by limiting the expression of functional virulence factors on the pathogens. The identified hydrophobic cavity formed by L4 (Tyr898, Ala900, Leu904, Val906 and Val908), S5 (Tyr845 and Ile847) and S6 (Val855, Ala857 and Tyr859) of BrkAβ provides such a potential target for anti-pertussis drug design and gives a new direction in the study of AT translocation mechanisms.


Neil Isaacs and Xujia Zhang initiated the project. Isobel Black, Yanshi Zhu and Yanwu Huo cloned and purified the protein and obtained the first crystals. Yujia Zhai and Yanwu Huo performed the crystallization optimization and screening. Yanshi Zhu and Aleksander Roszak collected and processed diffraction data at the ESRF. Yujia Zhai and Qiangjun Zhou collected diffraction data at Shanghai. Yujia Zhai and Kai Zhang processed the data and solved the crystal structure. Yujia Zhai performed all of the mutagenesis and biochemical studies with the help of Xiaoyun Pang. Jiuwei Lu finished the immunolabelling experiments. Yujia Zhai and Fei Sun analysed the data and wrote the paper. Neil Isaacs, Yanshi Zhu and Aleksander Roszak contributed to the editing and revising of the paper before submission. Fei Sun and Neil Isaacs supervised the whole project.


This work was supported by the National Science Foundation of China [grant numbers 30970569 and 30721003], the ‘973’ project of the Chinese Ministry of Science and Technology [grant numbers 2006CB911001, 2006CB806506 and 2009CB918803] and a Partnering China Award from the Biotechnology and Biological Sciences Research Council.


We thank Professor Rachel Fernandez from the University of British Columbia for supplying us with the plasmid pD06935 containing the full-length BrkA gene. We are grateful to Jun Ma and Yingzhi Xu (F.S.'s group) for their help on cloning mutants and Western blot experiments. We also thank Xudong Zhao, Ruimin Zheng and Su Liu (Core Facilities for Protein Sciences, Chinese Academy of Sciences) for their group's help on maintaining the instruments in our laboratory. We acknowledge the ESRF, Grenoble, France, for the provision of synchrotron radiation facilities and thank Dr Gianluca Cioci for assistance in using beamline ID29.


  • The crystal structure co-ordinates of BrkA β-domain have been deposited in the PDB with accession number 3QQ2.

Abbreviations: AT, autotransporter; BCA, bicinchoninic acid; BrkA, Bordetella serum-resistance killing protein A; BrkAβ, BrkA β-domain; C8E4, tetra-(ethylene glycol) (mono-octyl)ether; D, pseudo-dimer; ESRF, European Synchrotron Radiation Facility; FBS, fetal bovine serum; IM, inner membrane; IPTG, isopropyl β-D-thiogalactopyranoside; L, loop; LDAO, N,N-dimethyldodecylamine-N-oxide; LS, light scattering; M, molecule; MR, molecular replacement; NCS, non-crystallographic symmetry; OM, outer membrane; Omp85, outer membrane protein 85; PEG, poly(ethylene glycol); RI, refractive index; RMSD, root mean square deviation; S, β-strand; SEC, size-exclusion chromatography; SEC-MALS, SEC with multi-angle light-scattering detection; SNR, signal-to-noise ratio; T, periplasmic turn


View Abstract