Structure of an atypical periplasmic adaptor from a multidrug efflux pump of the spirochete Borrelia burgdorferi

Periplasmic adaptor proteins are essential components of bacterial tripartite multidrug efflux pumps. Here we report the 2.35 Å resolution crystal structure of the BesA adaptor from the spirochete Borrelia burgdorferi solved using selenomethionine derivatized protein. BesA shows the archetypal linear, flexible, multi‐domain architecture evident among proteobacteria and retains the lipoyl, β‐barrel and membrane‐proximal domains that interact with the periplasmic domains of the inner membrane transporter. However, it lacks the α‐hairpin domain shown to establish extensive coiled‐coil interactions with the periplasmic entrance helices of the outer membrane‐anchored TolC exit duct. This has implications for the modelling of assembled tripartite efflux pumps.


Introduction
Tripartite efflux pumps span the inner and outer membranes of Gram-negative bacteria to expel antibiotics, metals and other noxious molecules [1][2][3]. They are key to survival and the development of multidrug resistance (MDR). Efflux substrates bind to an inner membrane (IM) transporter, typically a proton antiporter (e.g. Escherichia coli AcrB), before delivery to the TolC exit duct [4]. Trimeric TolC [5], anchored in the outer membrane (OM), projects an a-helical barrel into the periplasm where it contacts tip-to-tip [6][7][8] the 70 Å periplasmic extension of the IM trimeric AcrB transporter [9]. Extensive in vivo site-specific cross-linking and multidomain docking showed that pump assembly is established and stabilised by extensive interactions of both IM and OM components with a ubiquitous third component, the periplasmic adaptor protein [7]. Structural analysis of the Pseudomonas aeruginosa adaptor MexA [7,10] and subsequently other multidrug resistance [11,12] and metal [13,14] efflux pump adaptors have revealed conserved flexible, linearly arranged domains: a membrane proximal (MP) domain, a b-barrel domain, a lipoyl domain and an a-helical hairpin. The MP, b-barrel, and lipoyl domains of AcrA contact AcrB while the full length of the 47 Å long a-hairpin establishes extensive coiled-coil interactions with the periplasmic entrance of each TolC protomer [7,15]. The adaptor multidomain structure is conserved throughout proteobacteria, but primary sequence comparison in the spirochete Borrelia burgdorferi suggested that a tripartite MDR pump, BesABC, included an adaptor that appeared to lack an a-hairpin [16]. This study reported besC, besA and besB are co-transcribed in an operon, and that insertional disruption of the besCAB operon abrogates expression of BesC, BesB and BesA and causes up to 64-fold reduction in the minimal inhibitory concentrations for several antibiotics, SDS and ethidium bromide. Bunikis et al. concluded that BesABC is a functional tripartite RND efflux pump. The absence of an adaptor hairpin in a tripartite pump would be unprecedented and would have implications for understanding the assembly of MDR efflux pumps [2]. We therefore crystallised the BesA adaptor.

Materials and methods
Expression of soluble BesA -The BesA sequence indicated a signal sequence and lipidation site (Cys 26) comparable to IM-associated periplasmic adaptors E. coli AcrA and P. aeruginosa MexA, so besA codons 27-318 from Borrelia burgdorferi B31 genomic DNA (ATCC 35210) were amplified with forward and reverse primers (5 0 -GCGCCATATGGTTGGTGATAACAAGCTAGATGAC-3 0 , 5 0 -GCGCGGATC CTTAAATATTGCTTTCGGCTGAAAGACCCTC-3 0 ), and the product digested with NdeI/BamHI and cloned into pET28 (Novagen). The resultant pET28-BesA plasmid encodes BesA in which the N-terminal signal sequence is replaced with a hexahistidine tag.
For selenomethionine incorporation, E. coli B834 (DE3) was transformed with pET28-BesA, and grown in M9 minimal medium containing 50 lg ml À1 selenomethionine. At A 600 0.6, 0.1 mM IPTG was added and the culture continued for 16 h at 18°C. Cell pellets were resuspended in 50 mM HEPES pH 8.0, 400 mM NaCl, 5% glycerol, 10 mM MgCl 2 . 1 mM TCEP was included in all buffers. Cells were broken in a cell disruptor (30 000 psi) and centrifuged (150 000Âg, 1 h at 4°C). Supernatant was incubated with IMAC resin (Biorad) and 4 mM imidazole (1 h, 4°C). Resin was washed in buffer A (25 mM HEPES pH 8.0, 500 mM NaCl, 8 mM imidazole) containing 0.1% Triton X-100, then buffer A alone. Protein was eluted in 25 mM HEPES pH 8.0, 200 mM NaCl and 250 mM imidazole and loaded onto a Superdex S75 column equilibrated in 25 mM HEPES pH 8.0, 150 mM NaCl and peak fractions concentrated to 10 mg ml À1 . To produce native BesA, E. coli C41 [17] cells bearing pET28-BesA were grown in 2xTY medium, with expression and purification as above, except the hexahistidine tag was removed with thrombin (Novagen) before loading on a Superdex S75 column. Protein eluted with a molecular weight consistent with monomeric BesA.
Attempts to express correctly folded and localised BesB using N-and C-terminal his-and strep-tags as well as fusions with maltose-binding protein and GFP were repeatedly unsuccessful. A construct, in which the region up to and including the BesB N-terminal transmembrane helix was replaced with the equivalent region of AcrB, also failed to express. Though small amounts of BesC can be expressed, we have not been able to crystallise it.
Crystallisation and data collection -BesA crystallisation was conducted using sitting drop vapour diffusion at 15°C. 2 ll protein solution was mixed with either 1 ll (C2 and P2 1 crystal forms) or 2 ll (P2 1 2 1 2 1 crystal form) crystallisation reagent and equilibrated against 500 ll reagent. The P2 1 2 1  Crystal structure determination and analysis -X-ray data sets were indexed and integrated using iMosflm [18] and scaled using Scala or Aimless in the CCP4 suite [19]. Crystallographic phases for the 2.6 Å selenomethionine dataset were obtained by Singlewavelength anomalous diffraction (SAD). Selenium sites were identified using HYSS [20], and phases calculated with Phaser [21]. Density modification was performed in Parrot [22] using 4fold non-crystallographic symmetry (NCS) and an initial model of BesA built with Buccaneer [23] with four monomers in the asymmetric unit. The structure was completed with manual model-building in Coot [24] and refinement in Phenix [25] using NCS restraints between each of the three BesA domains. Upon completion of the selenoprotein model, refinement was switched to the isomorphous high-resolution data, maintaining the same 'free' reflection list. Model building and refinement employed Coot [24] and Refmac [26], with NCS restraints as before.
Structures of C2 and P2 1 crystal forms were determined by molecular replacement using Molrep [27] or Phaser [21] with a BesA monomer as a probe. Refinement was performed as above except the C2 data initially benefitted from jelly body refinement and did not make use of NCS restraints. Validation was assisted by Molprobity [28] and Procheck [29]. The final model of the high resolution C2 crystal form is the only dataset for which a reasonable atomic model could be built for the residual loop (i.e. residues 106-114). The structures have been deposited in the PDB with accession codes 4KKS, 4KKT and 4KKU.
Molecular dynamics (MD) simulations -Energy minimizations and MD simulations were performed using Amber11 [33] as previously [34]. The protein formal charge of +6e was neutralized by adding Cl À ions. The simulated system consists of 276 residues, 6 Cl À ions, and 18,589 water molecules, making a total of 60 159 atoms. MD trajectory calculations of 23 ns were performed on the Darwin cluster (http://www.hpc.cam.ac.uk/services/darwin. html).

Results and discussion
The crystal structure of the complete BesA protein was solved by SAD in three space groups, P2 1 2 1 2 1 (2.35 Å resolution), P2 1 (2.5 Å) and C2 (2.6 Å) (Table 1), in which either a monomer (C2) or four intertwined head-to-tail monomers (P2 1 and P2 1 2 1 2 1 , Fig. 1A) are present in the asymmetric unit. Given the protein is membrane localised by an N-terminal lipidation, this head-to-tail arrangement, also seen in the isolated CusB structure [14], cannot be physiologically relevant. The BesA monomer shows the archetypal, multidomain adaptor topology, with the MP, b-barrel, and lipoyl domains in linear arrangement (coloured orange, yellow and green, respectively, in Fig. 1A and B). However, the expected final a-hairpin domain is completely absent, confirming what was predicted from sequence analysis [16], leaving only a nine residue loop (Pro106 to Leu114, blue in Fig. 1B) linking the two halves of the lipoyl domain. Density for this loop is visible only in the C2 monomer (Fig. 1C) where it is stabilised by the loop of a neighbouring monomer in the crystal lattice, but high B-factors here and poorly defined density for the P2 1 2 1 2 1 or P2 1 monomers indicate the small loop is highly flexible. The BesA architecture is distinct, the other adaptor structures have a well defined a-helical hairpin up to 67 Å/90 residues long (Fig. 2) [10][11][12][13][14]. The B. burgdorferi genome reveals no other adaptors that could perform the function of BesA, and no hairpin-encoding gene in the vicinity of the besCAB operon. However we cannot exclude that a novel, unidentified factor exists that could compensate for the lack of an a-hairpin in BesA.
The conserved MP, b-barrel, and lipoyl domains of BesA have low RMSDs compared to counterparts in other MDR [7,[10][11][12] and heavy metal efflux (HME) pump adaptors [13,14] (Supplemental Table 1). This conforms to the view they play a key role in pump assembly and function, particularly in binding to the large periplasmic extension of the IM transporters. As would be expected in this case, primary sequence analysis indicates the cognate interacting transporter domains of BesB and other transporters are likewise structurally conserved (Supplemental Table 2) [16]. In both the data-based assembly model of AcrAB-TolC [7] and the in vitro co-crystal structure of an adaptor-transporter (CusBA) subcomplex [35] tripartite pump assembly is facilitated by the flexibility of the adaptor's serially-linked domains [11,13], which was confirmed by MD analysis of the closely related MexA [36]. This 'hinge' flexibility at the inter-domain linkers is also evident in the BesA variant space groups which show movement of the lipoyl domain, relative to the other domains (Fig. 3A). We further analysed BesA interdomain dynamics by performing a 23 ns MD simulation (Fig. 3B) of a single monomer in aqueous solution. Two major motions are observed 'hinged' at the interdomain linkers, i.e. independent movement of both the lipoyl domain (Fig. 3B, left) and the MP domain (Fig. 3B, right) relative to the b-barrel domain.
This conserved organisation, domain structure and inter-domain flexibility indicate a common adaptor function in proteobacteria Values in parenthesis are for the highest resolution shell. and spirochetes. The absence of an a-hairpin in the BesA adaptor therefore has significant implications for tripartite efflux pump assembly. Our data-based model of the E. coli AcrAB-TolC pump suggests the IM transporter and OM channel contact each other via their juxtaposed apical a-helices, mediated by extensive interactions of these two components with the four adaptor domains (Fig. 4, left) [7]. Based on this model our predicted assembly of the Borrelia pump indicates interaction between the exit duct and transporter would be essential to provide a continuous tunnel for substrate to bypass the periplasm and exit the cell (Fig. 4, right). In this consistent view of assembly the adaptor lipoyl domain would be well positioned to con-tact the exit duct (Fig. 4), as it would also be in the predicted HME CusBAC assembly [35,37] in which the adaptor sits further up the IM transporter than is calculated for drug efflux adaptors AcrA/MexA. This suggests a key role for the lipoyl domain in the spirochete Be-sABC pump not only in interacting with the transporter TolC-docking domains, as in the E. coli/Pseudomonas pumps [7], but also in stabilising TolC interaction. Otherwise, there would be very little contact between adaptor and OM exit duct (Fig. 4, right). The BesA architecture is incompatible with any model in which adaptor hairpins are suggested to form homotypic interactions to create a tunnel that contacts TolC tip-to-tip (Supplemental Fig. 1, left) [38], as in such a model the predicted assembly is reliant on the hairpin to form a sealed channel between separated IM and OM components   (Supplemental Fig. 1, right). Indeed, this model (Supplemental Fig. 1, left) is inconsistent with in vivo cross-linking data showing AcrB and TolC interact [6,8], in vivo cross-linking that shows the full length of the AcrA hairpin interacts with the periplasmic entrance of TolC [7,15] and the CusBA crystal structure in which interactions between CusB adaptors are mediated not through the a-hairpin domains but through the b-barrel and lipoyl domains [3,35,37]. The high resolution structure of BesA establishes unequivocally an important divergence from the paradigm adaptor structure. Primary sequences of other putative spirochete adaptors suggest the BesA architecture may not be unique, as adaptor sequences lacking the a-hairpin coding region can be identified in Borrelia, Treponema and Spirochaeta (Supplemental Fig. 2). Indeed, comparable primary sequences are also evident rarely in proteobacteria (e.g. Legionella). This indicates that BesABC and most likely other similar pumps are assembled without the extensive coiled-coil interactions shown to be central to the otherwise closely related tripartite efflux pumps of E. coli and P. aeruginosa. It is therefore possible that despite close structural similarity among components different pumps require distinct component interactions.