A periplasmic coiled-coil interface underlying TolC recruitment and the assembly of bacterial drug efflux pumps

Identifying the Adaptor Docking Interface on TolC.

Cysteine substitutions of TolC residues were made in the wild-type TolC protein tagged with a C-terminal hexahistidine sequence (plasmid pACT7HisTolC; ref. 6). Control experiments showed that each of the His-tagged, Cys-mutated forms of TolC (and those of AcrA described below) assembled functional pumps with their wild-type partners, conferring wild-type antibiotic resistance (data not shown). The 471-residue wild-type TolC (TolC^WT, numbering of the mature protein) has no intrinsic cysteines and was used as a control. Twelve residues were substituted (see Fig. 1). D¹²¹, S¹²⁴, Q¹³⁹, S¹⁴², R¹⁵⁸, L¹⁶⁵, L¹⁶⁹, V¹⁹⁸, Q³⁵², S³⁵³, S³⁶³, and I³⁶⁹ are located in the periplasmic coiled-coils below the equatorial domain: residues D¹²¹, S¹²⁴, Q¹³⁹, and S¹⁴² are located specifically on α-helix 3 of the outer coiled-coil (denoted H3, helix numbering according to ref. 12), R¹⁵⁸, L¹⁶⁵, and L¹⁶⁹ are located on the outer coiled-coil H4, Q³⁵², S³⁵³, and S³⁶³ are on the inner coiled-coil H7, and I³⁶⁹ is on H8. The 12th residue, V¹⁹⁸, lies in the loop region of the equatorial domain.

Exponential cultures expressing one of the TolC variants or TolC^WT were subjected to three different membrane-permeable hetero-bifunctional cross-linkers. In one set of experiments, two flexible cross-linkers were reactive toward sulfhydryl- and primary amine-groups at either end, respectively, namely N-succinimidyl 3-(2-pyridyldithio)-propionate (SPDP) with a short spacer arm of 6.8 Å and sulfosuccinimidyl 6-[3′(2-pyridyldithio)-propionamido] hexanoate (LC-SPDP) with a long spacer arm of 15.6 Å. Both contain a disulfide bond in the spacer, which can be cleaved by a reducing agent. TolC and cross-linked AcrA proteins were affinity purified via the TolC-hexahistidine tag and analyzed by SDS/PAGE and immunoblotting. Each TolC variant was expressed and isolated at comparable levels throughout (Fig. 1A), but AcrA was only copurified when the TolC-S¹²⁴C, -Q¹³⁹C, -S¹⁴²C, and -S³⁶³C variants were used as bait. The same results were evident with each of the two cross-linkers (Fig. 1A Upper), although with TolC-Q¹³⁹C, more AcrA was copurified by the long LC-SPDP than the short SPDP. No AcrA cross-links were observed without cross-linker, or when the TolC^WT protein was the bait.

A second approach used a cross-linker (B4M) that is both sulfhydryl- and photo-reactive. This cross-linker attaches to the modified TolC cysteine variants, but is nonselectively photo-reactive at its other end. It is noncleavable and has a flexible, intermediate length spacer arm of 10 Å. As before, TolC variants were cross-linked in vivo and the assembled complexes were purified and analyzed by SDS/PAGE and immunoblotting. Only TolC variants TolC-S¹²⁴C, -Q¹³⁹C, -S¹⁴²C, and TolC-S³⁶³C cross-linked to AcrA (Fig. 1B), and there was an especially strong reaction with TolC-Q¹³⁹C.

These results identified the same residues as those when using the two amine-reactive cross-linkers in Fig. 1A. The positive cross-links identified by both approaches form a cluster along helix H3, extending to just under the equatorial domain, and include the tip of helix H7 (Fig. 1C). In contrast, residues facing the TolC subunit interface, along H4 and H7* (* denoting the adjacent monomer), did not cross-link to the adaptor protein.

Identifying the TolC Recruitment Interface on the AcrA Adaptor.

To determine the corresponding interaction surface for TolC on the α-helical coiled-coil of the cognate AcrA, we performed reciprocal in vivo cross-linking with SPDP and LC-SPDP using cysteine variants of AcrA. Seventeen residues were mutated to cysteine along the two helices of the AcrA α-helical hairpin (N-terminal helix, α1, encompassing residues 75–108; C-terminal helix, α2, residues 116–149, numbering of the mature protein). The orientation of the side chains of these residues divide the hairpin essentially into two faces, with one group of side chains facing the β-barrel domain (AcrA-Q⁹²C, -K¹¹⁶C, -L¹²³C, -N¹³⁰C, -T¹³⁴C, -K¹³⁷C, -A¹³⁸C, -R¹⁴⁴C), and the other group facing away (AcrA-D⁷³C, -A⁷⁵C, -A⁷⁹C, -D⁸⁷C, -L¹⁰⁰C, -R¹⁰⁴C, -Q¹¹²C, -I¹¹⁴C, -E¹¹⁸C) (see Fig. 2B).

In vivo cross-linking experiments with SPDP and LC-SPDP were carried out as before, and the complexes were purified and subjected to immunoblot analyses (Fig. 2A). The His-tagged wild-type TolC was detected consistently in all samples, and no copurification of AcrA was observed in the negative controls without cross-linker or with the wild-type AcrA, which has no freely accessible cysteines (the sole natural cysteine is at the extreme N terminus, but it is lipid-modified and anchored in the inner membrane after removal of the signal sequence; ref. 28). AcrA-TolC SPDP cross-links were evident with AcrA-A⁷⁹C, -D⁸⁷C, -L¹⁰⁰C, -R¹⁰⁴C, -Q¹¹²C, and AcrA-E¹¹⁸C, but not the other variants. These cross-links were also obtained with the longer LC-SPDP; in addition, two AcrA variants (A⁷⁵C and I¹¹⁴C) were cross-linked only by the longer LC-SPDP (Fig. 2A). These TolC links highlight AcrA residues that have side chains facing away from the β-barrel domain and the rest of the AcrA molecule, showing that the orientation of the adaptor hairpin is important. Residue 92 in the middle of helix α1 is oriented away from the proposed interaction face and it did not form cross-links, whereas the two positive residues at the tip of α2, 114 and 118, are oriented toward TolC (Fig. 2B).

These results identify the N-terminal helix of the AcrA coiled-coil as the TolC interaction surface. The distance to the TolC lysine residues located furthest down on the TolC α-barrel (K¹³⁰, K³⁸³, and K³⁴⁵, Fig. 3A) narrows down the overall positioning of the AcrA hairpin, because A⁷³C at the bottom of helix α1 does not link to TolC, whereas A⁷⁵C slightly higher up only cross-links with the long arm cross-linker LC-SPDP, and A⁷⁹C further up the coiled-coil is isolated with both the short- and long-arm cross-linker.

Modeling the TolC–AcrA Coiled-Coil Interface in Closed and Open States.

Using the crystal structures of TolC and AcrA (Protein Data Bank ID codes 1EK9 and 2F1M, respectively), we pursued a data-driven molecular docking approach that could take advantage of the large number of mapped interactions generated by the cross-linking experiments. The approach used sets out to integrate data from two separate sets of experiments, in one case specifically anchoring cross-linkers to 12 residues along the TolC lower α-helical domain, and in the reciprocal case to 17 residues along the AcrA α-helical hairpin. The final model presented was required to be compatible with both independent sets of data.

Initially, we sought conformations of AcrA that could dock onto TolC when constrained by the positive cross-linking distances. This search used simulated annealing in the molecular dynamics module of the CNS refinement program (29) applied to the cysteine-mutant variants of AcrA and TolC, and it was started with the structures placed in 250 random initial conditions [details of the modeling approach can be found in supporting information (SI) Text ]. Tight conformational restraints were applied to produce rigid body docking, except in the equatorial region of TolC where restraints were relaxed (residues 408, 421, and 195–207) to allow flexibility. This is a legitimate approach because, though ordered in the TolC crystal structure, the equatorial domain is conformationally variable or even disordered in other crystallized examples of the TolC family (27).

Due to the relatively sparse distribution of lysine side chains on the interacting domains, the cross-linking data were found to be surprisingly restrictive with respect to allowed orientations of AcrA on TolC, and only one solution was found that satisfied all of the distance restraints (see SI Fig. 4). This solution required the loop at the tip of the AcrA hairpin to be adjacent to the TolC equatorial domain, but made only limited use of the flexibility allowed. As a result, the equatorial domain obscures parts of the TolC coiled-coils that we have suggested may be involved in direct contacts with the adaptor helical hairpin (12, 24, 25). Therefore, to explore more precisely the alignment of the helices, the residues of the equatorial domain were removed from the coordinates, and the docking analyses were extended to a rigid body search of 3,600 orientations in the program ZDOCK (30) (details of this procedure can be found in SI Text). Again, a single solution was consistent with all of the positive cross-linking data (Fig. 3). The modeled distances closest to the optimal distance for cross-linking are given in SI Table 1. This model was also found to be consistent with the negative data when orientation and exposure of partner cysteine and lysine residues were taken into account, as well as possible steric hindrance posed by the cross-linker atoms. As observed in the initial docking with the equatorial domain in place, there was more than one possible partner lysine residue for most positive cysteine cross-links. Owing to the nature of the cross-linking experiments, a range of products may be expected that are optimized for particular partner residues, giving a cluster of solutions differing by perhaps 6 Å rmsd. Modeling this as a single model will produce an apparent range of cross-linking distances. However, the docking protocols used will give a final solution that is likely to be placed with better precision (in the range of ±3 Å rmsd; ref. 31), allowing us to be confident about the likely residues at the interface.

Fig. 3A shows the general position of the docked AcrA adaptor molecules on the TolC trimer. This model positions the hairpin of the adaptor alongside the intramolecular groove formed between the lower TolC α-helices H7/H8/H3 and directs the adaptor lipoyl and β-barrel domains outwards from the TolC trimer. As described above, the vertical register of AcrA is defined by the cross-links to the lowest lysine residues (shown as cyan spheres) available on TolC. The blue equatorial elements in Fig. 3 A and B represent their crystal conformation, which would sterically hinder the placement of AcrA in this register. However, by applying a simple rigid body rotation of 20° around TolC glycine residues 187 and 405, the equatorial domain is easily accommodated around the tip of the AcrA hairpin (gold elements in Fig. 3 A and B rotated about the residues indicated by green spheres). The displacement of the equatorial domain envisaged here allows significant improvements in electrostatic and H-bonding interactions (3- and 4-fold improvement in calculated energies, respectively) compared with the original docking that allowed only limited flexibility for the domain. This is particularly clear in the region of the loop at the tip of the AcrA hairpin, where several hydrophobic residues are buried under the equatorial domain. Formation of the docked complex buries 2,874 Ų of solvent exposed surface, excluding the equatorial domain elements. Once the equatorial domain is modeled back in position over the tip of the AcrA hairpin, this surface is increased to 3,452 Ų, making this the best docked solution. Analysis of the helix–helix packing at the AcrA–TolC interface shows contact angles similar to those observed in other proteins (32). As shown in Fig. 3B, helix α1 of AcrA aligns in an approximately antiparallel fashion to TolC helix H3 and also makes parallel contacts with helix H8, whereas the upper region of helix α2 makes a parallel contact with H7.

The surface representation in Fig. 3B illustrates how, in general, this solution gives a close fit between the interacting faces of AcrA and TolC, but Fig. 3C shows a region adjacent to H7/H8 of the TolC inner coiled-coil where there is a space between the surfaces. This appears to be related to the presence of several alanine residues (shown as magenta atoms) in the helix α2 of AcrA, and these small residues limit the number of side chain atoms available to give effective packing at the TolC surface. A possible explanation of this is that the TolC coordinates used for the AcrA docking are those of the closed form and the observed space allows the movement of the TolC inner coiled-coil during transition to the open state. Consistent with such a key role, a number of these residues, particularly those directed toward TolC, are conserved in the AcrA protein family (18). To test this possibility, we simulated the opening in the docked complex by forcing movement of the lower region of the TolC inner coiled-coil (from residue Ala³⁴⁷ to Ala³⁸², Fig. 3C gold elements) to the previously modeled open state of TolC (12, 26). Not only was there room for this movement, but also the more favorable shape complementarity in this area for the open state complex gave an improvement of 20% in the calculated binding energy (31).

Bacterial Strains and Expression Plasmids.

Cross-linking of TolCHis-cysteine variants was carried out in E. coli C41ΔtolC, a colicin E1-resistant isolate of (DE3) lysogen E. coli C41 (39) in which the loss of TolC was confirmed by Western blotting. Native AcrAB proteins were expressed from the chromosome. TolC wild-type and cysteine variants (expressed from pACT7HisTolC; ref. 6) were tagged with hexahistidine replacing the C-terminal 35 residues without attenuating function. Experiments using AcrA cysteine variants were carried out in the triple knockout E. coli MCΔtolCΔacrAB (6), a derivative of E. coli MC1061 converted to a (DE3) lysogen (Novagen, Merck Biosciences, Darmstadt, Germany). Wild-type and cysteine variants of AcrA were expressed from pACT7AcrA (6), His-tagged TolC was expressed from pLGTolCHis (made by subcloning tolC from pACT7HisTolC into the pLG339 BamHI site, using BglII and BamHI). AcrB was expressed from pETAcrB [acrB was PCR amplified from the E. coli MC1061 chromosome and cloned into pET11 (Novagen) using NdeI and BamHI]. In every case, proteins were synthesized at basal levels without T7 promotor induction.

Creation of TolC and AcrA Cysteine Variants.

Cysteine substitutions were made by site-directed mutagenesis (QuikChange, Stratagene, La Jolla, CA) using double-stranded plasmid template (pACT7HisTolC and pACT7AcrA; ref. 6) and long mutagenic primers. Mutations were confirmed by sequencing (Geneservice, Cambridge, U.K.). Western blots confirmed that all cysteine variants were expressed at levels comparable to wild type. They all assembled functional pumps that conferred wild-type antibiotic resistance, assayed as novobiocin resistance on solid growth medium.

In Vivo Cross-Linking Using SPDP and LC-SPDP.

Cultures were grown to OD₆₀₀ 0.8 with selection. Three aliquots of each were washed twice, concentrated 7.5-fold with cross-linking buffer (20 mM Na-phosphate, pH 7.5/150 mM NaCl/1 mM EDTA) and incubated for 30 min with either DMSO only, 0.2 mM SPDP, or 0.2 mM LC-SPDP (Pierce Biotechnology, Rockford, IL) each in DMSO. After quenching with 2.5 mM Tris·HCl (pH 7.4), cells were harvested and washed twice with cross-link buffer. Cell membranes were solubilized in 8 M urea, 1% Triton X-100, 20 mM Na-phosphate (pH 7.5). Complexes were affinity-purified from solubilized membranes by incubating for 1 h with 50 μl Ni-NTA resin (Qiagen, Crawley, U.K.) and washed, first with 10 mM imidazole, 8 M urea, 1% Triton X-100, 20 mM Na-phosphate (pH 7.0), then with 20 mM imidazole, 1% Triton X-100, 20 mM Na-phosphate (pH 7.0), 0.5 M NaCl, 0.1% SDS, and third with 30 mM imidazole, 1% Triton X-100, 20 mM Na-phosphate (pH 7.5), 150 mM NaCl, 20% glycerol. All washes were carried out three times each. Proteins were eluted with 8 M urea, 50 mM Tris·HCl, 2% SDS, 0.4 M imidazole (pH 6.8), and cleaved from the complex by reducing the cross-linker with 100 mM DTT (30 min at 37°C), resolved by SDS/10% PAGE, and immunoblotted with polyclonal rabbit antisera raised against purified TolC, AcrA, or AcrB. A chemoluminescence kit (Pierce or Upstate, Charlottesville, VA) was used for immunodetection.

In Vivo Cross-Linking Using Benzophenone-Maleimide (B4M).

Cells were treated as before, but with the sulfhydryl- to photo-activatable cross-linker B4M at a concentration of 0.2 mM (Molecular Probes, Eugene, OR). Excess B4M was removed by washing twice with Tris-buffered saline (pH 7.4). Reactions were irradiated at 300–400 nm with a 250 W UV lamp (at a 15-cm distance for 5 min), and quenched with 10 mM DTT. Cells were harvested, membranes were solubilized, and complexes were affinity-purified as above. Uncleaved samples were resolved by SDS/7% PAGE and immunoblotted as above.

Modeling of the AcrA-TolC Complex.

Two methods were used. Initially, rigid body docking was combined with simulated annealing in the refinement program CNS 1.1 (29). Cross-links were defined as ambiguous restraints between the Cα-atoms of mutated cysteine residues and of all lysines on the partner. For the short cross-linking, the range of distances allowed was 6.8 ± 2.5 Å, whereas for the longer cross-linking, it was in the range of 15.6 ± 4.5 Å. Docking started from 250 random initial conditions. Models fulfilling the cross-linking distance constraints were then postrefined using the local docking program Hex 4.5 (31). In the second method, the residues in the TolC equatorial domain were initially removed from the PDB file and then remodeled after the docking step using Coot (40). TolC and AcrA were docked using ZDOCK searching >3,600 orientations (30). Models satisfying the distance constraints were then refined using RDOCK (41). For comparison purposes, energies of all models were calculated in Hex 4.5 (31), and this was also used for studies on the local interactions after modeling the open state. Buried surface areas were calculated in CNS with a 1.4-Å solvent probe radius. Helix packing was analyzed with the Helix Pair Program (32). Full details of the modeling procedures are given in SI Text. Figures of the TolC–AcrA models were generated by using PyMOL (www.pymol.org).