Dynamics of a bacterial multidrug ABC transporter in the inward- and outward-facing conformations

Abstract

The study of membrane proteins remains a challenging task, and approaches to unravel their dynamics are scarce. Here, we applied hydrogen/deuterium exchange (HDX) coupled to mass spectrometry to probe the motions of a bacterial multidrug ATP-binding cassette (ABC) transporter, BmrA, in the inward-facing (resting state) and outward-facing (ATP-bound) conformations. Trypsin digestion and global or local HDX support the transition between inward- and outward-facing conformations during the catalytic cycle of BmrA. However, in the resting state, peptides from the two intracellular domains, especially ICD2, show a much faster HDX than in the closed state. This shows that these two subdomains are very flexible in this conformation. Additionally, molecular dynamics simulations suggest a large fluctuation of the Cα positions from ICD2 residues in the inward-facing conformation of a related transporter, MsbA. These results highlight the unexpected flexibility of ABC exporters in the resting state and underline the power of HDX coupled to mass spectrometry to explore conformational changes and dynamics of large membrane proteins.

Import of nutrients and export of signaling molecules or noxious compounds are essential processes for life, and selectivity for crossing membrane is usually governed by dedicated transporters. The ATP-binding cassette (ABC) is one of the largest families of transporters involved in these “checkpoints.” They are found in all living species and use ATP hydrolysis to transport a wide variety of substrates (1). The dysfunctions of several ABC transporters cause severe pathologies, such as adrenoleukodystrophy, hyperinsulinemic hypoglycemia, or cystic fibrosis (2). Other medically important ABC transporters are responsible for multidrug resistance (MDR) phenotypes. Multidrug transporters are capable of recognizing and expelling many unrelated organic compounds with quite dissimilar chemical scaffolds. Thus, although their original function is to protect healthy cells, the occurrence of these transporters in malignant tissues, pathogenic microorganisms, or parasites confers a resistance toward the curative drugs. The archetype of multidrug ABC transporters is the human MDR1 (or ABCB1); it is responsible for the failure of chemotherapeutic treatments in cancerous tissues. In bacteria, the identification of related transporters is relatively recent, compared with other families of multidrug transporters that use the proton gradient as the energy source (3). However, growing bodies of evidence support the implication of these multidrug ABC transporters in antibiotic resistance in many species, jeopardizing a successful therapy (4).

All ABC transporters share a common architecture, with two transmembrane domains (TMDs) responsible for substrate translocation and two nucleotide-binding domains (NBDs). In most exporters, these four domains are either linked into a single polypeptide or function as dimers with one NBD fused to one TMD (homo- or heterodimers). As opposed to the TMDs, the NBDs are highly conserved, with several distinctive motifs, notably the Walker A and Walker B found in many ATPases and the ABC signature specific to this family (5, 6). NBD structures have shown that during the catalytic cycle, the two NBDs interact transiently to form a sandwich dimer in a head-to-tail fashion, thereby trapping two ATP molecules at their interface (7–9). In this closed conformation, which can be stabilized either by ATPase-inactive mutations (10) or by vanadate-induced ADP trapping (11), each composite ATP-binding site is made of elements from one NBD (i.e., Walker A and B motifs), whereas the ABC signature is provided in trans by the second NBD. The overall structure of exporters is less compact than that of importers with two large intracellular domains (ICDs), ICD1 and ICD2, which protrude from the TMD to interact with the NBD. Importantly, whereas ICD1 from one monomer interacts with its own NBD, ICD2 (or ICD4 for full-length exporters) interacts mainly with the NBD of the other monomer (or other moiety) (12–14), and this has been validated by cross-linking experiments (15, 16). Snapshots of 3D structures and electron paramagnetic resonance (EPR) spectroscopy suggest that large motions occur during the catalytic cycle of ABC exporters, with the transporter oscillating between an apo, open (or inward-facing), conformation, with its two NBDs lying far apart, and an ATP-bound closed (or outward-facing) conformation, with the NBDs engaged in a tight interaction (12–14, 17). However, the open structures have raised some skepticism and were proposed to reflect nonphysiological states of the proteins, possibly induced by the presence of detergent (18, 19). Importantly, in all of the structures solved, the contacts between ICDs and NBDs are always maintained, because these two domains keep the same overall conformation and interactions. This suggests that NBD closure operates by a rigid body motion (12, 14). On the other hand, cross-linking experiments between ICDs and NBDs give support to alternative conformations for these two domains in different ABC exporters (15, 16, 20, 21).

To shed light on structural modifications that might occur during the mechanism of ABC transporters, we used a hydrogen/deuterium exchange (HDX) approach coupled to mass spectrometry (MS). HDX has been widely used on soluble proteins to probe the accessibility of backbone amides, thus highlighting secondary structure modifications at the peptide level (22, 23). For membrane proteins, this approach is only beginning to emerge (24–26), but its usefulness has been demonstrated, for example, to pinpoint the conformational changes associated with the binding of ligands, agonists, or antagonists, to the β-2-adrenergic receptor GPCR (27). Given the importance of membrane proteins, our results emphasize the benefit of using HDX to address fundamental issues about their functioning. Here, it revealed the dynamics of a multidrug transporter homologous to MsbA, Sav1866, and to murine MDR1 (28, 29). In particular, it showed that the two ICDs of BmrA behave quite differently, depending on the conformational state considered, and are unexpectedly highly flexible in the resting state. These results extend our vision of the functioning mechanism of ABC exporters and complement the snapshots of the catalytic cycle acquired from high-resolution 3D structures.

Results

BmrA Switches Between Two Opposite Conformations.

3D structures of ABC exporters solved in the presence of detergent have suggested that large conformational changes might occur during the ATPase cycle (12, 14). To investigate this, we checked whether the presence of detergent affected the conformational changes that take place during the catalytic cycle of BmrA. In agreement with our previous results (30), the ATPase activity of BmrA was conserved in a detergent solubilized state and sensitive to vanadate inhibition (Fig. S1). Then, global HDX was performed on the protein either purified in detergent or in a membrane environment previously shown to preserve the functional state of the transporter (29). The deconvoluted masses obtained from electrospray of BmrA, either embedded in membrane (66,341 ± 3 Da) or purified in n-dodecyl-β-D-maltoside (DDM) (66,339 ± 3 Da), are in very good agreement with the theoretical mass of BmrA, possessing only one extra Ser amino acid on its N terminus (66,340 Da; Fig. S2). The other peaks of lower intensity were due to sequence heterogeneity (SI Results) and unidentified noncovalent adducts on both forms of BmrA.

After incubation of BmrA preparations in deuterated buffers, the mass of deuterated BmrA was measured. The percentage of deuterated amides was calculated, taking into account the average of total number of amide hydrogens in BmrA (585 for the native protein) and the mass increase. Curves for deuteration kinetics show the same behavior, with the exchange steadily increasing with time for BmrA in both states (Fig. 1). However, BmrA in membranes was approximately half less deuterated (13–23%) than in its solubilized form (23–47%). This lower deuteration exchange in membrane could possibly reflect a more compact conformation of the BmrA dimer. However, when the same experiment was performed after prior incubation of BmrA with vanadate in the presence of ATP-Mg (closed symbols in Fig. 1) [i.e., conditions known to promote the formation of a closed dimer (11)], a lower deuteration level was found for BmrA either in membrane (squares in Fig. 1) or in detergent (circles in Fig. 1). Thus, although HDX occurs at approximately twice the rate in detergent compared with the membrane-bound BmrA, vanadate inhibition induced a significant decrease in HDX, to a similar extent for both environments. Therefore, this shows that in either state, BmrA can switch between two different conformations: an apo one whereby deuterium exchange readily occurs and a closed one more resistant to deuterium exchange.

Fig. 1.

Global deuteration kinetics of BmrA. HDX was followed for BmrA in membranes (squares) or purified in DDM (circles), either in the resting state (open symbols) or in a vanadate-inhibited state (filled symbols). SDs for two separate experiments are shown.

To bring further evidence for the similar behavior of BmrA in detergent and membranes, limited digestion with trypsin was carried out. BmrA in the apo conformation was rapidly digested by trypsin in membranes as well as in DDM (Fig. S3). In contrast, in the closed conformation (+Vi), BmrA was quite resistant to this protease in both environments. Thus, regardless of its environment, BmrA adopts two different conformations with opposite sensitivity to trypsin.

Peptide Mapping of BmrA.

To achieve a sequence coverage as large as possible, local HDX was performed on purified BmrA because (i) a much higher deuteration level was found for the protein in detergent, and (ii) the presence of Escherichia coli proteins in the membrane fraction would hamper the unambiguous identification, or the determination of deuterium incorporation, for some BmrA peptides. Optimized conditions led to a sequence coverage of 78% using pepsin (Table S1). It was lower in the TMD (60%), with the majority of the identified peptides belonging to the hydrophilic ICDs, ICD1 and ICD2 (Fig. S4). In contrast, the NBD sequence was almost fully covered by the pepsin digestion (98%). Because deuteration induced the spreading of isotopic envelopes (i.e., the ensemble of isotopic peaks of a peptide), it often led to signal decrease and overlapping. Thus, a significant number of peptides were lost after deuteration, giving 45% and 92% coverage for TMD and NBD, respectively. Importantly, most of the two ICDs were still covered by HDX because these two subdomains are essential for the transmission of the conformational changes between the NBD and the TMD (13, 31, 32).

Comparison of the Closed Forms of BmrA Using HDX.

Besides the inhibition by vanadate, another way to promote the closed conformation in ABC transporters is to use catalytic inactive mutants (8, 9, 33). For BmrA, we previously showed that the conserved glutamate residue E504 adjacent to the Walker B motif is very likely the catalytic base for ATP hydrolysis, because mutation of this residue fully abrogated the activity (10, 34). Deuterium incorporation was monitored as a function of time and was compared for each peptide generated from the three following forms of BmrA: wild-type in the resting state (apo form), wild-type inhibited by vanadate (in the presence of ATP-Mg), and E504A mutant in the presence of ATP-Mg (Fig. S5). In some cases (e.g., in Fig. 2), spectra from the vanadate-inhibited form showed two envelopes of isotopic peaks, thus giving two different kinetics, with the minor envelope corresponding to that of the wild-type in the apo form. Presumably, this envelope originates from BmrA dimers not inhibited among the whole population, because the inhibition of ATPase activity by vanadate was not total (∼80%; Fig. S1) (see also refs. 17 and 30). In contrast, the E504A mutant showed practically only one isotopic envelope that was always similar to that of the low molecular mass observed in the vanadate-inhibited sample. Given the propensity of the E504A BmrA mutant to readily adopt the closed conformation of BmrA in the presence of ATP-Mg, and the homogeneity of HDX of its peptides, we focused on the differences between this mutant in the closed conformation (ATP-bound state) and the wild-type in the apo conformation (resting state).

Fig. 2.

Isotopic profiles of 203–215 peptide at two deuteration times. The mass spectra of peptide 203–215 from ICD2 is shown before (black peaks) or after 15 s (blue peaks) and 3,600 s (red peaks) of deuteration. This peptide was from the wild-type BmrA in the resting state (A) or vanadate-inhibited state (B) or from the E504A mutant incubated in the presence of ATP (C).

Influence of the BmrA Conformation on HDX of the ICDs and NBDs.

The major differences in HDX between both forms can be visualized on the “heat map” presented in Fig. S6. The regions in red or orange correspond to peptides with more than 45% and 30% HDX difference, respectively, after 1 h deuteration. They clearly belong to ICD1, ICD2, and the NBD, particularly the Walker A motif and the ABC signature, which are known to be involved in nucleotide binding. To highlight the HDX data, two BmrA models (28) based on the X-ray structures of homologous bacterial ABC transporters were used (Fig. 3). For the apo conformation of BmrA, the open structure of EcMsbA was used as a template, whereas the closed conformation of BmrA (mimicked by the ATP-bound E504A mutant) was modeled on the Sav1866 structure (13, 14). In agreement with the closed model, the regions participating in either NBD dimerization or direct interaction with ATP/Mg show a rather low level of HDX. This is the case, for instance, for the peptide 372–383, which entirely covers the Walker A motif (HDX of 29%), or for the peptide 479–492 encompassing the ABC signature (26%). In contrast, the former peptide became much more rapidly deuterated in the apo conformation with a high deuteration level of the Walker A motif (82%), consistent with the presence of a large accessible loop, whereas the peptide covering the ABC signature motif also showed a substantial increase in its exchange rate (69%). The lowest value of HDX obtained for this second peptide suggests that in the open conformation, the region likely conserved an α-helical structure less prone to HDX than a loop. Thus, the exchange rates for these two peptides are in agreement with the open model used to depict the apo conformation (Fig. 3). For other peptides of the NBDs known to be involved in ATP binding or NBD dimerization, similar trends were observed, with a faster exchange rate in the apo conformation compared with the ATP-bound conformation (e.g., peptide 501–509 including most of the Walker B motif). Other regions of the NBDs not involved in either process gave similar HDX in both conformations with, for instance, a high deuteration level consistent with a loop structure (84%) for peptide 439–447.

Fig. 3.

HDX of peptides using BmrA 3D models. 3D models of BmrA in the open conformation (Left) or in the closed conformation (Right). One subunit was drawn in wheat color and the other in light gray. Identified peptides are drawn with rainbow colors (in one monomer and in all ICDs) according to their percentage of deuterium exchange after 3,600 s (scale of exchange shown at right). Selected regions are also shown with arrows.

A different picture emerges when one focuses on ICDs. In contrast to the two models suggesting that these domains have a similar conformation in the two states, open and closed, peptides from both ICDs showed a large difference in the HDX in these two conformations. Thus, ICD2 was highly deuterated in the apo conformation, with 82% and 79% for peptides 203–215 and 216–236, respectively, whereas in the closed conformation the HDX rate significantly decreased to 33% and 34%, respectively. Likewise, but to a lower extent, regions 104–114 and 115–132 from ICD1 were also fairly deuterated in the apo conformation (66% and 81%, respectively). This value dropped to 27% in the closed state for the former peptide, whereas it remained highly deuterated for the latter (69%). Because the solvent accessibility of amide hydrogens from both ICDs is essentially unaffected upon the dimerization of the two NBDs (e.g., for peptides 104–114 and 203–215) (Fig. S7), the much faster rate of HDX observed in the resting state for these peptides must reflect a modification of conformation and/or flexibility of each ICD. It must be noted that in the closed state, some residues of ICD1 interact with ICD2 within the same subunit (see ref. 13) and thus will help to restrict the motion of the two ICDs.

To further assess the flexibility of these two domains in the 3D structures of the related transporters used to model BmrA, molecular dynamics simulations of ∼100 ns were performed on EcMsbA and Sav1866, along with that of StMsbA that was included for comparison. After reaching a plateau for the rmsd of the Cα for each conformation, the fluctuation of the position of the Cα atoms were calculated for the last 20 ns of the trajectories, and the results are shown in Fig. S8. Clearly, a minor fluctuation was observed in the closed conformations of either Sav1866 or StMsbA, regardless of the Cα considered (always below 2.5 Å). In contrast, some basal fluctuations occurred throughout the sequence of EcMsbA and were much more pronounced (above 4 Å) for four areas, including in particular ICD2. This suggests that this latter domain must be quite flexible in this apo conformation.

Discussion

The goal of this study was to get insights into the dynamics of a multidrug ABC transporter in two different conformations: the resting state, in which the transporter adopts presumably an open conformation (the extent of which remains to be determined), and the closed state, in which the two NBDs dimerize in a transient, ATP-bound, conformation. Overall, global HDX and trypsin digestion show that either in membrane or in detergent, BmrA can switch between two very different conformations during its catalytic cycle: a closed form rather resistant to trypsin digestion and HDX, and another one in the resting state quite sensitive to trypsin and more susceptible to HDX. Thus, our results support the postulate that multidrug ABC transporters can cycle between an open, inward-facing, conformation in the resting state and a closed, outward-facing, ATP-bound conformation. Similar conclusions were drawn for the catalytic cycle of MsbA from cross-linking experiments (35) or from EPR spectroscopy (17, 36, 37). On the basis of 3D structures of ABC exporters, the transition between these two opposite conformations was proposed to operate according to a rigid body motion (14). This entails that the same overall structure of the two monomers (or moieties for full-length transporters) would be maintained throughout the catalytic cycle. However, local HDX obtained for BmrA clearly challenges this assumption and suggests that in the resting state, BmrA samples different conformations. However, these different conformations seem to be restricted to both ICDs. Indeed, apart from the peptides involved either in the interactions between the two NBDs or with ATP, no major changes are observed for the other peptides from either the NBD or the transmembrane helices, although limited information is available for the latter. Therefore, the interaction between the TMD and the NBD that is secured by the two ICDs in the closed state must become loose in the resting state owing to a greater flexibility of these ICDs. This would allow some reorientation of the whole NBD compared with the TMD and would explain some unexpected cross-links previously observed for BmrA. It was found that disulfide bonds were formed between ICD1 and the Q-loop of the NBD in the resting state only (20). However, in all structures of exporters, either closed or open, ICD2 is sandwiched between ICD1 and the Q-loop, thereby impeding the possible interaction between these two protein segments (12–14). Hence, some reorganization of BmrA domains would be required to allow the cross-link between ICD1 and the Q-loop (20). Other unpredicted cross-links questioning the conformation observed in 3D structures were also reported for two other ABC transporters, TAP1/TAP2 and Yor1p (15, 21). These results lend support to additional conformations of these transporters, with different orientations of the NBD relative to ICDs or of the two ICDs with respect to each other. Other studies performed on homodimeric ABC transporters, LmrA and MsbA, using either solid-state NMR or EPR spectroscopy, reached similar conclusions of alternative conformations (or greater flexibility) of the NBD in the resting state, as opposed to the TMD (36, 38). In addition to the NBD flexibility, EPR experiments also support a large conformational entropy of residues from ICD1 in the resting state, as evidenced by the large distance distribution of the spin label probe tethered to residue 115 of MsbA (36). Unfortunately, the flexibility of ICD2 could not be tested because the tethering of a spin probe to cysteine residues in this domain was detrimental to MsbA activity (37). For the cystic fibrosis transmembrane conductance regulator (CFTR), although cross-link patterns support a global architecture similar to that found in Sav1866 or MDR1, a considerable flexibility at the interface of NBDs and ICDs was proposed to explain the formation of cross-links between these two domains, even with long cross-linker reagent (up to 24.7 Å) (16). In addition, a recent report using a genetic screen with the yeast ABC multidrug transporter PDR5 supports the idea of a noncanonical coupling between the NBD and the TMD as being mainly achieved through ICD1 and not through ICD2 (39). Whether this is a peculiar property of PDR5 or reflects an alternative conformation of the ICDs remains to be investigated. Interestingly, a molecular dynamics study of Sav1866 indicated that in a closed conformation and among residues whose Cα show the largest displacement between the ATP and ADP+Pi bound states, some belong to ICD2 (residues 221–222) and to ICD1 (residues 110–113) (40). Here, molecular dynamics simulations suggest that in the apo conformation of EcMsbA, large fluctuations in the positions of the Cα atoms of ICD2 would take place. Conformational flexibility, and in particular of α-helices, has been reported for some membrane proteins, such as the lactose permease (41, 42), the osmoprotectant symporter ProP (43), and the glycerol facilitator (44), and this property likely pertains to the transport of their respective substrates (45). For multidrug ABC transporters, the flexibility of ICDs might not only be involved in the transport mechanism but possibly also permits the polyspecificity of drug recognition, provided that this flexibility could be propagated to the drug-binding site(s).

HDX coupled to MS is commonly used to probe the dynamics of soluble proteins, but this technique remains artful in the case of membrane proteins, especially because detergent often reduces the sequence coverage. Up to now, the only ABC transporter for which this technique was used was the soluble NBD1 of CFTR (46). However, our results on BmrA emphasize how useful this technique is, especially to study membrane proteins with large extramembranous domains. This includes, for example, secondary transporters of the resistance-nodulation-cell division family, such as AcrB (47), or primary transporters of the P-type family like the Ca²⁺-ATPase (48). The use of HDX coupled to MS to unravel different conformations of these membrane proteins should shed new light on their transport mechanism.

Materials and Methods

Site-Directed Mutagenesis and Purification of BmrA.

The E504A BmrA mutant was constructed as previously described (10). The E504A mutant and wild-type BmrA were overexpressed as previously described (49), and BmrA-enriched membranes were either frozen in liquid nitrogen and kept at −80 °C until used or solubilized in 1% DDM. After solubilization, BmrA was purified by Ni-affinity chromatography as previously described (29).

HDX of BmrA.

The HDX reaction was initiated by a 10× dilution into deuterated buffer containing 50 mM NaCl and 0.05% DDM. Time course of the HDX was followed over a 1-h period by sequential withdrawing of 120 μL deuterated samples, which were immediately added to 26 μL of quenching buffer (8 M guanidium chloride, 500 mM glycine HCl, pH 2.2), rapidly mixed, and flash-frozen in liquid nitrogen.

HPLC Peptide Separation.

After protease digestion in solution or on-column in an ice bath at 0 °C, peptides were loaded onto a peptide MicroTrap (Michrom Bioresources) column and washed with 0.03% TFA in water (HPLC solution A). They were then separated on a reversed-phase C12 column (1 × 50 mm, Jupiter; Phenomenex) using a linear gradient of 15–45% (vol/vol) solution B (CH₃CN 95% and TFA 0.03%) during 26 min. The column was connected to the electrospray source of mass spectrometers until 30% (vol/vol) B and then disconnected to avoid pollution of the electrospray source by the detergent. The valves, trap cartridge and column were cooled to 0 °C by immersion in an ice bath.

Mass Spectrometric Analyses of Peptides.

The tandem MS (mapping) analyses were performed on an ion trap mass spectrometer (Esquire 3000+; Bruker Daltonics) to enable the identification of the peptides after their separation on HPLC. Accurate mass measurements and the analysis of the local kinetics of deuteration were done on a time-of-flight (TOF) mass spectrometer (6210; Agilent Technologies) equipped with an electrospray source. Data were processed with MassHunter Qualitative Analysis, and the deconvolution and calculation of the average masses were carried out in Magtran (50).

Electrospray Mass Spectrometry of the Full-Length Protein.

The molecular mass of the intact protein was measured with the same electrospray TOF mass spectrometer coupled with HPLC. Protein was loaded onto a protein MicroTrap (Michrom Bioresources) column and washed with 0.03% TFA in water (HPLC solution A). The protein and detergent were then separated on a reversed-phase C18 column (1 × 50 mm, Jupiter; Phenomenex) using a linear gradient of 5–60% (vol/vol) solution B (95% CH₃CN, 4.97% water, and 0.03% TFA) during 3 min.