Structural insights into the membrane-extracted dimeric form of the ATPase TraB from the Escherichia coli pKM101 conjugation system

Background Type IV secretion (T4S) systems are involved in secretion of virulence factors such as toxins or transforming molecules, or bacterial conjugation. T4S systems are composed of 12 proteins named VirB1-B11 and VirD4. Among them, three ATPases are involved in the assembly of the T4S system and/or provide energy for substrate transfer, VirB4, VirB11 and VirD4. The X-ray crystal structures of VirB11 and VirD4 have already been solved but VirB4 has proven to be reluctant to any structural investigation so far. Results Here, we have used small-angle X-ray scattering to obtain the first structural models for the membrane-extracted, dimeric form of the TraB protein, the VirB4 homolog encoded by the E. coli pKM101 plasmid, and for the monomeric soluble form of the LvhB4 protein, the VirB4 homolog of the T4S system encoded by the Legionella pneumophila lvh operon. We have obtained the low resolution structures of the full-length TraB and of its N- and C-terminal halves. From these SAXS models, we derive the internal organisation of TraB. We also show that the two TraB N- and C-terminal domains are independently involved in the dimerisation of the full-length protein. Conclusions These models provide the first structural insights into the architecture of VirB4 proteins. In particular, our results highlight the modular arrangement and functional relevance of the dimeric-membrane-bound form of TraB.


Background
Type IV secretion (T4S) systems are one of six secretion systems used to transport effector proteins or DNAs through the cell membrane of Gram-negative bacteria. These six secretions systems can be categorised into two classes. The first class of secretion systems mediates substrate transfer from the cytosol to the extracellular milieu in one step: substrates captured from the cytosol are released extracellularly without the need for a periplasmic intermediate [1]. The second class encompasses a range of specialised outer membrane (OM) secretion systems: the substrate is first transported through the inner membrane (IM) to the periplasm via the general SecA-BYEG secretion machinery and then uses specialised OM systems for extracellular release [2,3]. T4S systems belong to the first class.
T4S systems export proteins and DNA-protein complexes and fulfil a wide variety of functions, such as i-the conjugative transfer of plasmids and other mobile DNA elements to bacterial recipient cells, ii-the direct uptake of DNA from the extracellular milieu or iii-the delivery of protein or DNA substrates to eukaryotic target cells [4,5]. T4S systems are used by several plant and human pathogens for virulence. Such bacterial pathogens include Agrobacterium tumefaciens, the causative agent of crown gall disease in plants, Bordetella pertussis, the agent responsible for whooping cough in children, and Helicobacter pylori, responsible for gastric ulcers and stomach cancer [6][7][8][9]. In addition, there are intracellular bacterial pathogens utilising T4S systems for virulence, such as Brucella suis, the causative agent of brucellosis, and Legionella pneumophila, the causative agent of Legionnaires' disease [10,11].
T4S systems are generally composed of 12 protein components forming a macromolecular assembly inserted into the bacterial cell envelope [5]. These proteins are named VirB1-VirB11 and VirD4, based on the widely used nomenclature of the model system, the A. tumefaciens VirB/D4 T4S system. Three ATPases are key components of the T4S system: VirD4, VirB11 and VirB4. VirB4 proteins are the largest and the most evolutionarily conserved proteins in T4S systems [12] but their function remains unclear. Although VirB4 proteins have clearly defined Walker A and Walker B motifs characteristic of ATPases [13], until very recently no ATPase activity had been demonstrated for any VirB4 homologues [13]. However, two recent studies have shown that ATPase activities of VirB4 proteins are crucially dependent on solution conditions and on the oligomerisation state of VirB4 [14,15]. For TrwK, the VirB4 homolog encoded by the R388 conjugative plasmid system, Rabel et al. [13] initially reported that the protein exhibited no ATPase activity and was monomeric. However Arechaga et al. [14], in a subsequent study, reported an ATPase activity of TrwK in the presence of acetate ions, possibly due to a small proportion of an hexameric form of the protein. TraB, the VirB4 homolog encoded by the pKM101 conjugative plasmid system, also exhibits ATPase activity in the presence of acetate ions and is primarily hexameric under these solution conditions [15]. Interestingly, TraB partitions between the cytosol and the inner membrane, and the membrane-extracted form does not exhibit ATPase activity, even in the presence of acetate ions [15]. This membrane-extracted form of TraB was also shown to be dimeric. It was concluded that cytosolic TraB is in equilibrium between a dimeric form that binds DNA and nucleotides, but is unable to hydrolyze ATP, and an acetate-induced hexameric form able to hydrolyse ATP. TraB purified from the membrane is in the dimeric form, and is unable to transition to the hexameric form even in the presence of acetate ions [15]. Interestingly, A. tumefaciens VirB4 was also shown to form active dimers in vivo [16], strongly supporting a functional role of this dimer, besides the hexameric form.
The structure of VirB4 proteins is still unknown, as they have resisted extensive crystallisation efforts either in the hexameric or the dimeric form. Attempts at visualising acetate-induced hexameric TraB by negative stained electron microscopy or small-angle X-ray scattering have failed [15]. Recently, based on sequence similarities with TrwB (the VirD4 homolog from the plasmid R388 conjugation system), the A. tumefaciens VirB4 C-terminal domain was modelled, as an homo-hexameric ring [17] much like VirB11 and VirD4 [18]. However no structural experimental data has yet backed this model, most probably because it has been impossible so far to stabilise and isolate the hexameric form of VirB4. Here we report the low resolution structure of the membrane-extracted dimeric form of TraB, using small-angle X-ray scattering (SAXS). We also performed a SAXS analysis of the N-terminal (TraB NT ) and C-terminal (TraB CT ) domains of TraB, and of the full-length monomeric LvhB4, the VirB4 homolog from the L. pneumophila T4S system, which represents the first in vitro study of a member of the L. pneumophila lvh T4S system. Altogether, our results provide the first insights into the architecture of the highly conserved VirB4 family of proteins.

Purification of TraB domains and LvhB4
Based on sequence homology between the C-terminal domain of TraB and the protein TrwB, the E. coli R388encoded VirD4 homolog, we previously established that TraB can be divided into two folded domains ( Figure 1A): a C-terminal domain (residues 448 to 848, TraB CT ), which is soluble, and an N-terminal domain (residues 1-442, TraB NT ), which partitions between a soluble form in the cytosol and a membrane-bound form [15]. These two domains are functional as they are each able to bind DNA and ATP [15]. Full-length TraB (TraB FL ) was subsequently found to partition between a soluble and membrane-bound form [15]. Various transmembrane domains predictors were used to screen the sequence of TraB (DAS, http://www.sbc.su.se/ miklos/DAS/; HMMTOP, http://www.enzim.hu/ hmmtop/; TMPred, http://www.ch.embnet.org/software/ TMPRED_form.html; TMHMM, http://www.cbs.dtu.dk/ services/TMHMM-2.0/; and TopPred, http://mobyle.pasteur.fr/cgi-bin/portal.py?form=toppred). The predictions were not fully consistent, except for one stretch that was predicted by two out of the 5 different predictors, between residues 254 and 271. This suggests the existence of a transmembrane (TM) segment in TraB, or of a hydrophobic patch through which TraB might be associated with the membrane.
For this study, TraB FL and TraB NT were both purified from the membrane fraction, while TraB CT , which is soluble and does not partition in the membrane, was purified from the soluble fraction. For comparison, we cloned, expressed and purified the full-length LvhB4, the VirB4 homolog from the L. pneumophila T4S system, for which no predicted TM domain was found. Indeed, LvhB4 purifies from the soluble fraction and not from the membrane fraction, demonstrating that the protein is not located in the membrane. All four proteins were purified to homogeneity using the same two-step purification strategy ( Figure 1B and "Materials & Methods"). In SDS-PAGE the proteins migrate at their expected molecular mass: 102 kDa for TraB FL , 55 kDa for TraB NT , 49 kDa for TraB CT , and 94 kDa for LvhB4 ( Figure 1B).

Size Determination of TraB domains and LvhB4
We investigated the oligomeric state of all four proteins in the GF sol (TraB CT and LvhB4) and GF mb (TraB NT and TraB FL ) buffer conditions (see definition of GF sol and GF mb in Materials and Methods). Table 1 summarizes the results obtained by Gel Filtration, Dynamic Light Scattering and Native-Gel electrophoresis. The calibration of the gel filtration column (see "Materials & Methods") allowed us to evaluate the apparent molecular mass of the proteins according to their elution volume. TraB FL ran as a 198.1 kDa protein, TraB NT as a 122.1 kDa protein, TraB CT as a 100 kDa protein, and LvhB4 as a 92.7 kDa protein. By comparison with the calculated molecular mass obtained from the amino acid sequence, we concluded that TraB FL , TraB NT and TraB CT were all forming dimers under the examined buffer conditions. In contrast, LvhB4 behaved as a monomer in the same conditions. DLS and blue-native PAGE (BN-PAGE) confirmed these results for TraB FL and TraB CT (Table 1). In conclusion, all TraB-derived constructs form only dimers in solution, whereas LvhB4 forms monomers.

Overall SAXS Parameters
Small-angle X-ray scattering (SAXS) studies yield information on the size and shape of macromolecules in solution, and also on the oligomerisation state of macromolecules. The overall dimensions of a protein can be evaluated by  All MM are given in kDa. "MMCalc" is the MM calculated from the amino-acid sequence, "GF" stands for gel filtration, "DLS" stands for Dynamic Light Scattering, "BN-PAGE" stands for Blue Native-PAGE. The oligomeric state is deduced from the comparison between MM Calc and the experimental MM as determined by the other approaches. "ND" stands for non-determined.
its radius of gyration, R G , while the molecular mass of the scattering particle is inferred from the forward scattering intensity, I(0), both derived from the Guinier law (see Materials and Methods). The scattering curves of all the constructs followed the Guinier law very well ( Figure 2) and did not display any sign of aggregation in solution.
We determined the molecular mass of each construct of TraB from their experimental I(0) and obtained 219 kDa for TraB FL , 106 kDa for TraB NT , and 83 kDa for TraB CT ( Table 2). After comparison with the expected molecular mass of the monomer of each TraB-derived constructs we concluded that all of them were dimers under the examined solution conditions ( Table 2). For TraB FL and TraB NT , the molecular masses inferred from the forward scattering intensity are very similar to the theoretical molecular masses calculated from their sequence. This strongly suggests that very few detergent molecules are bound to the proteins, and therefore that the contribution of the detergents to the scattering intensity can be considered as negligible. LvhB4 was confirmed to be a monomer with a molecular mass experimentally calculated to be 116 kDa, very close to the expected monomer size of 97 kDa. The I(0) determination of the molecular mass of all four proteins were consistent with our biochemical results and confirmed their oligomeric states. These results suggest that both the N-and Cterminal halves of TraB participate in the dimer interface of the full-length protein.
The radii of gyration measured on the different constructs are summarized in Table 2. Surprisingly, TraB FL and TraB NT have similar radii of gyration (58.6 ± 0.6 Å and 60.3 ± 0.7 Å respectively), in spite of the molecular mass of TraB NT being half of that of TraB FL . In contrast, the radii of gyration of TraB CT and LvhB4 are smaller (41.5 ± 0.5 Å and 37.2 ± 0.6 Å respectively). Proteins of similar radius of gyration may have very different shape and mass depending on their structure. Thus, TraB FL and TraB NT may have similar R G values but different structures. According to the SAXS results, TraB CT and consequently TraB FL are on average more compact than TraB NT . The R G of TraB NT reflects a less compact structure with a lower molecular mass, whereas, the R G of TraB FL results from a more compact structure and higher molecular mass, these two parameters counterbalancing each other to yield similar R G values.
We then calculated the pair-distance distribution function, P(r), from the SAXS curves (see "Materials & Methods"). The P(r) functions for all four constructs exhibited a bell-shaped curve with a slightly extended profile for the higher distances (data not shown), indicating a globular but somewhat elongated conformation. The comparison of the values obtained for the radius of gyration (R G ) and for the maximum dimension (D max ) for the four proteins gives an idea of their anisotropy. To obtain an estimation of the anisotropy of the protein, we calculated the ratio between R G and D max values for each construct. In the case of a sphere, this ratio is equal to 0.39, as the radius of a sphere is equal to (3/5) 1/2 R, where R is the radius of the sphere. Table 2 summarizes the values computed for all the constructs. The ratio R G /D max is 0.30 for TraB FL , 0.30 for TraB NT , 0.31 for TraB CT and 0.32 for LvhB4, significantly different than the value for a sphere, considering the error bars measured on the R G and D max . Interestingly, despite having different sizes, all four constructs exhibit the same anisotropy (ratio R G /D max of~0.3), indicating that the proteins are rather anisotropic, and thus elongated.

Low Resolution Shapes from Ab Initio Modeling
The overall shapes of the four proteins were computed from the SAXS profiles using the program GASBOR [19]. For all TraB derived constructs (TraB FL ; TraB NT and TraB CT ), we used an imposed 2-fold symmetry axis (referred as P2 in GASBOR) for generating the reconstructed models, in agreement with the oligomerisation state inferred from the biochemical data and from the forward scattering intensity I(0). Similar shapes were also obtained without imposing any symmetry (referred as P1 in GASBOR, data not shown). For LvhB4, no symmetry was imposed to reconstruct the 3D-volume. Several independent calculations provided highly reproducible results, with very similar models, and fit to the data of similar quality. The average shapes calculated from repeated, multiple modelling processes (data not shown) give each time a shape similar to the best individual model, defined by the lowest ChiExp value. Figure 3 shows that the fit to the experimental data of these shapes for all four proteins is very good, as confirmed by the value of ChiExp ( Table 2). The very reproducible shapes together with the very low ChiExp, obtained when fitting the data ( Table 2) give good confidence that the inferred model is not an artifact due to the SAXS intrinsic degeneracy and that these shapes are reliable. We decided to use the best model for each construct as representative of their low resolution structure in solution. Overall, the shapes of the four proteins appear to be globular but rather elongated (Figure 4), as expected from the R G /D max ratio. TraB FL presents two symmetrical lobes with two distinctive protruding ends, with each lobe being potentially attributable to one TraB FL monomer in the dimeric assembly ( Figure 4A). On the other hand, the shapes of both TraB NT ( Figure 4B) and TraB CT ( Figure 4C) appeared to be rather flat and elongated with few protruding extensions. The symmetrical 2-fold axis separating the two monomers is however clearly visible in the reconstructed volumes, even without imposing any symmetry (data not shown). Interestingly, the low-resolution structure of LvhB4 ( Figure 4D) shows an asymmetrical and compact, yet rather extended conformation.

Superposition and comparison of the TraB derived models
TraB NT and TraB CT represent the N-and C-terminal halves of TraB FL , respectively (see Figure 1 and "Materials & Methods"). Thus, given that all the three TraB-derived constructs are dimeric, the sum of the reconstructed volumes of both TraB NT and TraB CT should give a value similar to the volume of TraB FL . We used the program CRYSOL [20] to evaluate the volume of the reconstructed models (Table 2). We found that the sum of the volumes of TraB NT and TraB CT gives a value very close to the volume of TraB FL (represented here by the ratio of the volumes being close to 1.0, see table 2). We then manually   superimposed the shapes of TraB NT and TraB CT onto the shape of TraB FL . Several respective orientations of TraB NT and TraB CT were tested and for only one orientation the TraB NT and TraB CT models fit well together into the TraB FL model without any clash ( Figure 5A). In the proposed TraB FL model, the longest dimension of TraB NT and TraB CT are 45°apart, with the longest dimension of TraB NT coinciding with that of the TraB FL dimer. In this model, the more compact structure of TraB CT lays onto the more elongated structure of TraB NT (Figure 5B and 5C). Two schemes for the location of the TraB monomer and its two domains can be inferred from these SAXS envelopes of TraB FL and TraB fragments ( Figure 5D and 5E). In one scheme (dimer 1, Figure 5D), each TraB monomer is positioned on either side of an axis perpendicular to the long axis of the TraB dimer. In the other scheme (dimer 2, Figure 5E), each TraB monomer is positioned on each side of the long TraB dimer axis.

Comparison between TraB FL and LvhB4 and orientation of the TraB monomers
VirB4 proteins are the most conserved components (amino-acid sequence wise) of the T4S systems. It was then logical to think that VirB4 proteins could have a similar shape, albeit with different oligomeric states. We thus hypothesised that if TraB FL and LvhB4 monomers (34% identity) share a similar overall tertiary structure, the latter could help us localise TraB FL monomers in the dimeric model. The ratio between the volumes of two LvhB4 monomers and one dimeric TraB FL is of the same order as the ratio of their molecular masses, with LvhB4 slightly smaller than TraB FL monomers ( Table 2). Therefore, we tried to manually fit two LvhB4 monomers into the envelope of the TraB FL dimer. Figure 6 shows the results obtained with two possible orientations: orientation 1 ( Figure 6A and 6B) with each of the LvhB4 monomer being on each side of an axis perpendicular to the TraB FL dimer longest axis; and orientation 2 ( Figure 6C and 6D) with each LvhB4 monomer being on each side of the longest axis of the TraB FL dimer. In Figure 6B and 6D, we have superimposed the schematic diagrams of Figure 5D and 5E, respectively. In both orientations, there remains empty spaces in the TraB FL dimer, not filled by two LvhB4 monomers. In the case of orientation 1, this empty space is localized in between the two monomers. This is not in agreement with the evidence of a stable TraB FL dimer and with the fact that both the N-and C-terminal domains have been also isolated as dimers. Conversely, in the case of orientation 2, this empty space is localized at the N-terminal ends of LvhB4. This is consistent with the shorter N-terminal domain of LvhB4 compared to TraB and subsequently with the smaller volume of the LvhB4 model compared to the corresponding volume of the TraB FL monomer (taken as half of the observed TraB FL dimer volume; Table 2). In addition, the contact areas between the two N-terminal domains and the two C-terminal domains are larger, allowing formation of stable TraB FL , TraB NT and TraB CT dimers (see cartoons in Figure 6D). In conclusion, we favor orientation 2 as best describing the possible arrangement of the TraB FL monomers into the dimeric model.
Comparison between TraB CT model and the homology-based structure of At-VirB4 As mentioned previously, Middleton et al. [17] have modeled the structure of the C-terminal domain of the A. tumefaciens VirB4 (At-VirB4-Cter) based on the sequence homology with the VirD4 protein TrwB from the R388 conjugation machinery. Since TraB and At-VirB4 are highly homologous, we used this At-VirB4-Cter model and tried to fit two of them into the TraB CT dimeric shape. Indeed, as shown in Figure 7 (panel A), we could manually fit the two At-VirB4-Cter models into the TraB CT dimer volume. The two At-VirB4-Cter subunits are arranged according to the two-fold symmetry axis of the TraB CT dimer. As shown in Figure 7A, this docking identifies the C-terminal end of TraB as participating in the dimer interface. Similarly, we superimposed the At-VirB4-Cter model with the SAXS envelope of LvhB4 (Figure 7, panel B). From this comparison we propose that the C-terminal domain of LvhB4 is possibly localised at the wider end of the curved shape, opposite to the narrow end. The empty space in the superimposition would consequently be the N-terminal domain of LvhB4. This orientation fits well with the models presented in Figure 6C and 6D, showing the

Discussion
The SAXS experiments reported here confirm that TraB and its N-and C-terminal domains are dimeric in the acetate-free solution conditions under which the experiments were conducted, indicating that both domains participate in the dimer interface. The structures of the two domains revealed elongated shapes, which in the full-length protein come together at a 45°angle. The superposition of the LvhB4 structure could resolve the ambiguity as to where the TraB monomer lies, and favored dimer 2 ( Figure 5E), where the monomer would extend along the long axis of the dimer structure. Indeed, in the superimpositions of the two LvhB4 monomers onto the TraB FL dimer structure presented in Figure 6B and 6D, only the one aligning the LvhB4 monomers along the long axis of the TraB structure gives rise to an extended dimer interface. The configuration in Figure 5D (Dimer 1) would be expected to yield a less stable dimer than is observed. Also, a more extended conformation of TraB is consistent with our observation that TraB is susceptible to proteolysis and that limited proteolysis of TraB very rapidly yields TraB CT (data not shown).
VirB4 is a family of very conserved proteins that are essential components of T4S systems [12] However, recent biochemical studies have revealed that this family of proteins is more diverse than originally expected [13]. For example, their oligomerisation state appears different depending on the system under investigation and the conditions under which they are studied. TraB has been shown to be in equilibrium between two oligomeric states, dimer and hexamer, dependent on the solution conditions, namely the presence or absence of acetate ion [15]. TrwK appears to transition between a major monomeric form and minor hexameric form [14]. The VirB4 homolog encoded by the cag pathogenicity island in H. pylori appears to be monomeric [13] and we show here that, under the solution conditions examined, the VirB4 homolog from L. pneumophila LvhB4 is monomeric. Hexamer formation appears to be required for ATP-hydrolyzing activity: indeed only hexameric forms of VirB4 homologs have been shown to exhibit ATPase activity [14,15]. So far, only sparse information has been gathered about the function of the dimeric TraB. We recently reported its DNA and nucleotide binding activities [15], while A. tumefaciens VirB4 was shown to direct dimer formation when fused to the N-terminal portion of the cI repressor protein [16]. The different subcellular localisation of TraB together with the recent characterization of a degenerated nucleotide binding site in its N-terminal domain [15] are features also observed in the SecA translocase, perhaps suggesting an evolutionary relationship between the two protein families [21,22]. Finally, TraB in the context of the entire T4S machinery might interact with different partners. Indeed, its close association with TraA (the VirB3 homolog encoded by the pKM101 plasmid) or its documented protein-protein interactions with other T4S system components could induce the conformational changes necessary to reshape the active site or radically change its cell environment more specifically, in order to stabilise an active dimeric membrane-bound form.

Conclusions
The work presented here provides the first structural glimpse of a protein which is crucial to type IV secretion but has until now resisted X-ray crystallography or EM structural characterisation. It uncovers a modular structure that comes together in an extended dimer interface where the domains appear to "hug" each other.
The dimers corresponding to each domain could easily be put together in the envelope of the full length protein  Figure 5D being superimposed. (D): same as C but for the cartoon representation of TraB in Figure 5E being superimposed. Each panel shows two perpendicular views of the model. The two LvhB4 monomers were manually fitted into the TraB FL model. The black cross and the dashed-line both indicate the P2 symmetry axis of the dimeric models. and the structural model for LvhB4 helps suggest a potential model for the full-length protein. Intriguingly, the predicted TM segment locates within the N-terminal domain, not at the boundaries of the domain structure. This raises topological issues that can be resolved by a model invoking an orientation of the TraB NT domain facing the cytosolic side of the inner membrane, while the TraB CT domain would lie in the cytoplasm. This would be consistent with TraB being only superficially associated with the membrane, and therefore being able to partition between the membrane and the cytoplasm. It is also consistent with the dimeric structure proposed here. Docking of the TrwB protein (a potential structural homolog of the C-terminal domain of VirB4 proteins) within the envelope of the TraB CT provides further structural details. Finally, the dimeric model of TraB observed here suggests that there might be structural rearrangements required to fit the VirB4 dimeric structure into the 14-fold symmetrical core complex recently unravelled by the high resolution EM structure of the VirB7-VirB9-VirB10 complex [23] and confirmed by the subsequent crystal structure of its outer membrane-inserting part [24]. Further studies will seek to elucidate the crystal structure of a VirB4 protein and also to visualize a complex of VirB4 bound to the core complex.

Cloning of TraB domains and LvhB4
Cloning of the full-length traB gene (traB FL , amino acids 1-866; Figure 1), the region encoding the N-terminal domain (traB NT , amino acids 1-442; Figure 1) and the C-terminal domain (traB CT , amino acids 448-848; Figure 1), together with the full-length lvhB4 gene (lvhB4 amino acids 1-826; Legionella pneumophila strain JR32) was as described in Durand et al. [15]. All four constructs allow the expression of N-terminally His 6 -tagged recombinant proteins, referred to thereafter as TraB FL , TraB NT , TraB CT , or LvhB4. After DNA sequencing (MWG Biotech) to check that the sequences did not contained any mutation, the four plasmids were transformed by heat-shock in chemically competent BL21 star (DE3) cells (Invitrogen), for large scale production of the recombinant proteins.

Production and Purification of Recombinant Proteins
E. coli strain BL21 star/DE3 (Invitrogen) containing one of the recombinant plasmids was grown at 37°C in Terrific Broth supplemented with 100 μg/ml of Ampicillin (Sigma-Aldrich), until the culture reached an A 600 nm of 1.2. Cultures were then shifted to 16°C for 1 h, before isopropyl-β-D-thiogalactopyranoside (IPTG) was added to a final concentration of 1 mM and growth continued for 15 h at 16°C. Cells were harvested by centrifugation, resuspended in 20 mM TrisHCl (pH7.5) and store at -20°C.
All subsequent steps were carried out at 4°C. TraB CT and LvhB4 were purified from cytoplasmic extracts as follow. The cells were defrosted and one tablet of Protease inhibitor cocktail EDTA free (Roche) was added, together with 300 mM NaCl and 1 mM β-mercaptoethanol (βME). After cells were broken by two rounds through an EmulsiFlex-C5 homogeniser and DNA fragmentation by sonication, the lysate was clarified by centrifugation at 18,000 r.p.m. for 45 min in a Sorvall SS-34 rotor. The clarified lysate was loaded onto a HisTrapHP 5 ml column (GE Healthcare) equilibrated in buffer A sol (20 mM TrisHCl/pH7.5, 300 mM NaCl, 1 mM βME) plus 4% of buffer B sol (20 mM TrisHCl/pH7.5, 300 mM NaCl, 1 mM βME, 500 mM Imidazole). The column was then washed with 100 ml of buffer A sol plus 8% buffer B sol . Finally the proteins still bound to the column were eluted in a gradient from 4% to 100% of buffer B sol in 100 ml. Eluted fractions containing either TraB CT or LvhB4 were pooled and concentrated in less than 4 ml before being loaded onto a HiPrep 16/60 Sephacryl S-300 HR column (Amersham) equilibrated in buffer GF sol (20 mM Tris/HCl pH 7.5, 50 mM NaCl, 1 mM βME). The proteins TraB CT and LvhB4 both eluted as a single peak. Fractions under this peak were pooled.
TraB FL and TraB NT were purified from membrane extracts as followed. The cells were defrosted and one tablet of protease inhibitor cocktail EDTA free (Roche) was added, together with 50 mM NaCl and 1 mM βME. After cells were broken by two rounds through an EmulsiFlex-C5 homogeniser and DNA fragmentation by sonication, unbroken cells were removed by centrifugation at 14,000 r.p.m. for 10 min in a Sorvall SS-34 rotor. Total membranes were pelleted by ultracentrifugation (45 min at 100,000 g, 4°C) and resuspended in buffer EB (20 mM Tris-HCl/pH 7.5, 50 mM NaCl, 1 mM βME, 1% (v/v) Triton ® X-100) supplemented with one tablet of protease inhibitor cocktail EDTA free (Roche). Membrane-embedded proteins were extracted during 1 h at 4°C. The membrane extract was further clarified by ultracentrifugation (30 min at 100 000 g, 4°C). Triton ® X-100 was only used for extraction, then it was replaced by the hydrogenated Triton ® X-100 (H) (Calbiochem) that does not absorb in UV. We further used a concentration of 0.01% Triton ® X-100 (H) (0.16 mM) since it was below the CMC of the detergent (0.2-0.9 mM), thus avoiding the formation of detergent micelles. The cleared extract was loaded onto a HisTrapHP 5 ml column (GE Healthcare) equilibrated in buffer A mb (20 mM TrisHCl/pH7.5, 300 mM NaCl, 1 mM βME, 0.01% Triton ® X-100 (H) ) plus 4% of buffer B mb (20 mM TrisHCl/pH7.5, 300 mM NaCl, 1 mM βME, 0.01% Triton ® X-100 (H) , 500 mM Imidazole). The column was then washed with 100 ml of buffer A mb plus 8% buffer B mb . Finally the proteins still bound to the column were eluted in a gradient from 4% to 100% of buffer B mb in 100 ml. Eluted fractions containing either His TraB FL or His TraB NT were pooled and concentrated in less than 4 ml before being loaded onto a HiPrep 16/60 Sephacryl S-300 HR column (Amersham) equilibrated in buffer GF mb (20 mM Tris/HCl pH 7.5, 50 mM NaCl, 1 mM βME, 0.01% Triton ® X-100 (H) ). The proteins TraB FL and TraB NT both eluted as a single peak. Fractions under this peak were pooled. Apparent molecular mass of proteins eluted from the gel filtration column was deduced from a calibration carried out with low and high molecular mass calibration kits (Amersham Biosciences). Determination of protein concentration was carried out by either using the theoretical absorption coefficients at 280 nm, which were obtained using the program Prot-Param on the EXPASY server (available on the World Wide Web at http://www.expasy.ch/tools), or with the Bio-Rad protein assay reagent (Bio-Rad).

Dynamic Light Scattering (DLS)
Dynamic light scattering experiments were performed with a DynaPro-801 (Protein Solutions) at room temperature. All samples were filtered prior to the measurements (Millex syringe filters, 0.22 μm; Millipore Corp.). Diffusion coefficients were inferred from the analysis of the decay of the scattered intensity autocorrelation function. The hydrodynamic radius and the molecular mass (MM) of proteins in solution were both deduced from translational diffusion coefficients. All calculations were performed using the software provided by the manufacturer (Dynamics V5.25.44).

SAXS Experiments
SAXS experiments were performed in two different locations. TraB FL and TraB NT were analysed on beamline X33 [25] at EMBL-Hamburg on storage ring DORIS III of the Deutsches Elektronen Synchrotron (DESY) using a MAR 345 image plate detector. The scattering patterns from solutions of TraB FL at protein concentrations of 3, 5, 7.5, 10, and 13.5 mg/ml, and for TraB NT at protein concentrations of 1.3, 2.1, 4.9, and 8.2 mg/ml were measured in buffer GF mb . At a sample detector distance of 2.7 m and wavelength (l) of 1.5 Å, the scattering vectors, q ranging from 0.0093 Å -1 to 0.50 Å -1 was covered (q = 4πsinθ/l, where 2θ is the scattering angle). According to radiation damage tests, one frame of 2 min exposure time was recorded for every sample. The data were normalised to the intensity of the transmitted beam and radially averaged, and the scattering of the buffer was subtracted, as absolutely no trace of the presence of micelles was detected from the buffer scattering curve. The difference curves were scaled for protein concentration and extrapolated to yield the final composite scattering curves. Molecular mass calibration was made with BSA.
TraB CT and LvhB4 were analysed at the European Synchrotron Radiation Facility (Grenoble, France) on beamline ID02 as described previously [26]. The scattering patterns from solutions of TraB CT at protein concentrations of 2.1, 3.7, 4.3, 6.1, and 8.2 mg/ml, and for LvhB4 at protein concentrations of 2, 2.9, 4.6, 6, 7.3, and 8.9 mg/ml were measured in buffer GF sol . The wavelength was 1.0 Å. The sample-to-detector distances were set at 1.0 m (TraB CT ) and 1.5 m (LvhB4), resulting in scattering vectors, q ranging from 0.011Å -1 to 0.50 Å -1 and from 0.010Å -1 to 0.37 Å -1 respectively. All experiments were performed at 20°C. Absolute calibration was made with water.

SAXS Data Evaluation
All steps for data processing were performed using the program package PRIMUS [27]. The experimental SAXS data for all samples were linear in a Guinier plot of the low q region, indicating that the proteins did not undergo aggregation. The radius of gyration R G was derived by the Guinier approximation I(q) = I(0) exp (-q 2 R G 2 /3) for qR G < 1.0. The radii of gyration R G , calculated for different protein concentrations, displayed a slight concentration dependence arising from particle interferences in solution. Interference-free SAXS profiles were estimated by extrapolating the measured scattering curves to infinite dilution. The molecular masses of the solutes were inferred from I(0) values, the forward scattering intensity, which is proportional to the molecular mass of the protein according to relationship MM~I(0)/ c, where c is the protein concentration. The intensity I (0) was experimentally inferred from the intercept of the linear fit in the Guinier plot Ln[I(q)] versus q 2 at low q values (qR G < 1.0). The program GNOM [28] was used to compute the pair-distance distribution functions, P(r). This approach also features the maximum dimension of the macromolecule, Dmax.

Ab Initio Modeling
The overall shapes of the entire assemblies were restored from the experimental data using the program GASBOR [19]. The scattering profiles were fitted on the spectrum of each protein up to q = 0.37Å -1 . GASBOR searches a chain-compatible spatial distribution of an exact number of dummy residues, centred on the C α atoms of the protein amino acid residues. We used both symmetry operations P1 and P2 proposed by the program GASBOR. At least 10 low resolution models obtained from different runs were averaged using the program DAMAVER [29] to construct the average model representing the general structural features of each reconstruction.