Conceived and designed the experiments: JA MSPS. Performed the experiments: JA. Analyzed the data: JA. Wrote the paper: JA HdW FMA MSPS. Contributed to the writing of the paper and discussions of the analysis of the simulations: FMA.
The authors have declared that no competing interests exist.
ABC transporters are a large family of membrane proteins involved in a variety of cellular processes, including multidrug and tumor resistance and ion channel regulation. Advances in the structural and functional understanding of ABC transporters have revealed that hydrolysis at the two canonical nucleotide-binding sites (NBSs) is co-operative and non-simultaneous. A conserved core architecture of bacterial and eukaryotic ABC exporters has been established, as exemplified by the crystal structure of the homodimeric multidrug exporter Sav1866. Currently, it is unclear how sequential ATP hydrolysis arises in a symmetric homodimeric transporter, since it implies at least transient asymmetry at the NBSs. We show by molecular dynamics simulation that the initially symmetric structure of Sav1866 readily undergoes asymmetric transitions at its NBSs in a pre-hydrolytic nucleotide configuration. MgATP-binding residues and a network of charged residues at the dimer interface are shown to form a sequence of putative molecular switches that allow ATP hydrolysis only at one NBS. We extend our findings to eukaryotic ABC exporters which often consist of two non-identical half-transporters, frequently with degeneracy substitutions at one of their two NBSs. Interestingly, many residues involved in asymmetric conformational switching in Sav1866 are substituted in degenerate eukaryotic NBS. This finding strengthens recent suggestions that the interplay of a consensus and a degenerate NBS in eukaroytic ABC proteins pre-determines the sequence of hydrolysis at the two NBSs.
ABC transporters are a large family of membrane proteins present in all organisms. Typically, they utilize ATP hydrolysis, the most prominent biological energy source, to translocate substrates into cells (e.g., bacterial nutritient uptake) or out of cells (e.g., multidrug exporters that contribute to antimicrobial resistance in bacteria and resistance to chemotherapeutic drugs in cancer). Also clinically relevant non-transport roles have been identified among ABC proteins. ABC transporters bind two molecules of ATP but do not hydrolyze them simultaneously. Therefore, an ABC transporter that consists of two symmetric halves must temporarily adopt asymmetric conformations at the two ATP-binding sites. Such transient conformational changes are difficult to address biochemically, but may be amenable to study by simulation methods, leading to future experiments. We employ molecular dynamics simulations to study how asymmetric switching might occur in the homodimeric bacterial ABC multidrug exporter Sav1866. The simulations suggest a mechanism of conformational switching that encompasses the ATP-binding sites and their interface towards the substrate-binding site. We extend our findings to show how asymmetric residue substitutions may render the switching process non-stochastic in mammalian Sav1866-like ABC exporters. This contributes to ongoing discussions about the role of two dissimilar ATP-binding sites in clinically relevant ABC proteins.
ATP-binding cassette (ABC) transporters are a large family of membrane proteins that use MgATP hydrolysis to drive the import or export of solutes to or from the cytoplasm. They undertake a number of physiological roles, for example bacterial nutrient uptake, bacterial drug resistance, tumor drug resistance, and peptide secretion
Most ABC transporters consist of two transmembrane domains (TMDs) that provide a pathway across the membrane for the transported substrate, and two nucleotide-binding domains (NBDs) which form two nucleotide-binding sites (NBSs) at their dimer interface (see
A: Sav1866 in a DPPC lipid bilayer. Each TMD contacts the NBDs by two cytosolic loops (CL1 and CL2, labeled in regular font for the green monomer and underlined italic font for the orange monomer). CL2 carries the coupling helix that is buried inside the NBD of the opposite monomer. B: Overview of the MgATP-bound NBD dimer, as seen from the TMDs. Functional motifs are indicated. ATP is colored green, Mg2+ brown. C: Close-up view of the MgATP binding site. Backbones of functional motifs are colored as in B. The signature sequence (orange) and the D-loop (lime) are provided by the other NBD. Selected residues are labeled.
ABC transporters are thought to hydrolyze the two MgATP molecules sequentially, as opposed to simultaneously, possibly involving a mechanism of alternating catalytic sites
In contrast to homodimeric bacterial ABC proteins, many eukaryotic ABC transporters consist of two asymmetric halves, often resulting in one consensus and one degenerate NBS. The degenerate NBS typically displays markedly reduced ATPase activity compared to the consensus NBS
In this work we use molecular dynamics (MD) simulations to explore the transient asymmetry that a homodimeric ABC exporter may adopt when in a pre-hydrolytic, MgATP-bound configuration. We simulate the bacterial multidrug exporter Sav1866, which is thought to represent the core architecture of ABC exporters
A number of other ABC transporters (e.g. BtuCD
Our multiple MD simulations show that an initially symmetrical MgATP-bound state of Sav1866 exhibits rapid (initial steps on ∼10 ns timescale) and stochastic switching into asymmetric NBD conformations. The NBDs seemingly adopt hydrolytically favorable conformations in only one NBS. We further show how the switching at the NBS is reflected in a network of charged residues at the TMD-NBD interface. We extend our observations of stochastic switching in symmetric NBDs to suggest how degeneracy in eukaryotic ABC exporters may lead to preferential switching. This provides a rationale for understanding residue substitutions in ABC proteins, such as CFTR, MRP1, Tap1/2 and SUR.
Polar hydrogens were added to the ADP-bound crystal structure of Sav1866 (pdb id 2HYD) using GROMACS
The protein was positioned manually in a preformed DPPC bilayer containing 512 lipid molecules. A shell of lipid molecules around the protein was removed, and the remaining 376 lipids were relaxed around the protein by a 0.5 ns simulation with position restraints on all non-hydrogen protein and MgATP atoms (harmonic restraints, force constant 1000 kJ mol−1 nm−2). In a similar fashion, we inserted the Sav1866 structure into a POPC bilayer. A snapshot of the bilayer-inserted Sav1866 structure is shown in
MD simulations were performed using the GROMACS 3.3.1 software
We first performed six repeat simulations of Sav1866 with MgATP nucleotide in a DPPC lipid bilayer, each with different starting velocities and a duration of 40 ns. In order to probe the effects of the lipid environment in our simulations, we also performed a 30 ns simulation of Sav1866 in a POPC bilayer. In addition, we performed two further control simulations with bound ADP and DPPC lipid, both to a duration of 30 ns.
Root-mean-square deviations (rmsd) of Cα atoms in our six simulations with DPPC lipid and one simulation with POPC lipid are within typical values observed for protein simulations: 2.5–4 Å for the whole protein, and 2–2.5 Å for individual NBD subunits (see
Root-mean-square deviations of C-alpha atoms in six repeat simulations with DPPC bilayer (40 ns duration), and one repeat with POPC bilayer (30 ns duration).
simulation | whole protein | NBD A+B | NBD A | NBD B | TMD A+B | TMD A | TMD B |
DPPC 1 | 3.6 | 2.5 | 2.2 | 2.5 | 3.9 | 3.2 | 4.3 |
DPPC 2 | 2.9 | 2.3 | 2.1 | 2.2 | 2.6 | 2.3 | 2.5 |
DPPC 3 | 4.1 | 2.7 | 2.4 | 2.6 | 4.1 | 3.8 | 4.4 |
DPPC 4 | 3.9 | 2.1 | 2.0 | 1.9 | 4.4 | 4.3 | 4.5 |
DPPC 5 | 3.5 | 2.4 | 2.1 | 2.4 | 3.4 | 3.4 | 3.0 |
DPPC 6 | 3.4 | 2.6 | 2.4 | 2.3 | 3.1 | 2.6 | 3.2 |
POPC | 2.6 | 2.1 | 2.0 | 1.9 | 2.1 | 2.2 | 2.1 |
The domain-specific rmsd values for the NBDs in the 30 ns simulation performed with POPC lipid are within the range observed in the 40 ns simulations with DPPC lipid, but those of the TMDs are lower in the POPC simulation (
To identify possible binding modes for MgATP at the two NBSs we determined atomic contacts between MgATP and the most important nucleotide-binding residues. In addition to inspecting various contacts as a function of time for each simulation (e.g.
B–E: Inter-residue contact numbers (4 Å cut-off) of selected residues at the MgATP binding sites visualized in scatter plots with the two NBS on the two axes. Higher numbers of contact are indicative of stronger interactions. Scatter plots were generated from simulation snapshots in 50 ps intervals (see SI
From this analysis, MgATP appears to form several distinct binding modes (
Distinct levels of contact are also seen between MgATP and the Walker-B Glu (E503;
Our simulations also feature contacts of the switch motif H534 to MgATP (
As noted above, the off-diagonal elements in the scatter plots of contact numbers (
The ADP-bound simulations show hardly any interactions of the studied residues with the nucleotide (see SI
The conserved Q-loop motif has been implicated in distinguishing between ATP and ADP at the NBS
In our simulations, inter-NBD D423-K483 interactions were frequent (
A–D: Scatter plots of contacts formed by the Q-loop D423. Contacts observed in the protonated Sav1866 crystal structure are indicted by green dots. Scatter plots were generated as described for
In the Sav1866 crystal structures, the D423 residues of each of the two Q-loops interact with the R474 residues of the so-called x-loops of the opposite NBDs (
Interactions formed by the Q-loop D423 in the ADP-bound simulations also show remarkable drift away from their contact numbers in the starting crystal structure, as shown in a scatter plot similar to that in
Sav1866 contains a short sequence motif (the x-loop) at the NBD-TMD interface that is specific to ABC exporters
The asymmetric GERG motif interactions with the NBDs arise from a stacking of the R474 pair on the Sav1866 symmetry axis. As indicated by the trajectory of the projection of R474 onto the lipid bilayer normal from one simulation (
Residue pairs along the Sav1866 symmetry axis, at the NBD-TMD interface, in a protonated form of the crystal structure. The CL2 loops of the TMDs are shown as green and orange ribbons. The NBDs are shown as transparent C-alpha traces. Exemplary plots of z-coordinates (along the bilayer normal) of the x-loop R474 and H204 in one of the repeat simulations indicate asymmetric stacking. See SI
A: Initial symmetric arrangement before simulation. B: Asymmetric arrangement observed in simulation, involving stacking of the two x-loop R474 and interactions of Q-loop D423 with the TMDs. Labels referring to the two monomers are distinguished by regular black font and bold-italic gray font.
The cytosolic ends of the TMDs, which form the interface to the x-loops, contain a series of residues along the dimer symmetry axis: Q208, H204, Q200 (see
The simulations described above reveal a pronounced asymmetry between the dynamic behavior of two initially identical Sav1866 monomers, such that several key interactions are found to exist only at one monomer at a time. These interactions include a salt bridge (D423-K483) that has also been reported to form in only one monomer in crystal structures of the homologous MgATP-bound NBD homodimer HlyB
Our simulations support this idea of a structural coupling of the inter-NBD Q-loop D423-K483 charge pair. This pairing may be dependent on the association of the Q-loop Q422 with MgATP. Furthermore, when D423 is unbound from K483, it can flip towards the TMDs and interact with the coupling helix on CL2 of the opposite monomer (
Another region in which we observe asymmetric switching is the GERG motif of the x-loop. Our simulations support its proposed role in linking the CLs together
Interestingly, Dawson and Locher
Our observation of asymmetric switching in Sav1866 is in line with the asymmetric function of its NBDs. As a comparison to this finding, we sought for a protein that requires perfect symmetry for proper function. Selectivity filters of potassium channels are likely to be dependent on their tetrameric symmetry to ensure K+ conduction and selectivity. Using a simulation trajectory generated in our laboratory for another study, we examined an inter-residue interaction at a crucial position near the selectivity filter: that between (protonated) E71 and D80. The E71-D80 interaction has been shown to be involved in slow inactivation of KcsA, underlining its functional importance
In the structure determination of ADP-bound Sav1866, two-fold non-crystallographic symmetry was applied to large parts of the dimer, although some pairs of residues were refined in non-symmetric conformations
While most asymmetric residue pairs are located on the outside of the Sav1866 dimer and presumably contribute to crystal packing, a cluster of initially asymmetric residues is found at the cytosolic end of the TMDs, at the dimer interface along the symmetry axis (see
It should also be noted that our simulations were performed in the absence of a transported substrate. The binding of substrate is known to promote ATPase activity in ABC transporters including Sav1866
Our simulations were set up by changing the nucleotides in the pseudo-symmetric post-hydrolytic Sav1866 structure into a model of the pre-hydrolytic nucleotide configuration. This allowed us to study possible first steps of a putative pre-hydrolytic switching. We observe asymmetric structural rearrangements relative to the ADP-bound crystal structure, suggesting that the substitution of MgATP into the binding sites initiated a switching process. However, since our simulations are relatively short (and do not allow for formation or breaking of covalent bonds), we cannot expect to see beyond the initial steps of this process.
We observe several regions of asymmetric switching. Firstly, the MgATP-sensing Q-loop Q422 forms a high-contact MgATP binding mode typically only at one NBS. Secondly, the neighboring Q-loop residue D423 charge pairs with a basic residue (K483) near the signature sequence only at one NBS at a time, as has been reported in the ABC transporters HlyB and Tap1/2
Conceivably, the asymmetric interactions in the Sav1866 NBDs effectively constitute a series of molecular switches (
Ultimately, the behaviour of Sav1866 appears stochastic at the level of these molecular switches. A picture emerges in which, at the atomic level, ABC transporters do not function as molecular motors with perfectly defined steps. Rather, coincidental stochastic steps underpin the larger-scale motions required for function. Therefore, our sequence of simulation snapshots (SI
Many interactions in our system showed considerable variability in contact numbers and possible binding modes between individual runs within the ensemble of seven simulations. It seems likely that substitution of ADP for MgATP at both NBSs left the initial system in an energetically activated (i.e. non-equilibrium) state, thus enhancing the extent of conformational space accessible in relatively short (40 ns) simulations. Thus, our approach of repeated MD simulations helps to extend the range of conformations accessed. However, our simulations provide far from exhaustive coverage of the conformational space. This is apparent in the lack of diagonal symmetry in many of the scatter plots of contact numbers. Asymmetric coverage of conformational space could also be indicative of an inherent bias in the starting structure, such as the asymmetric starting orientations of residues along the symmetry axis (see above).
In NBD crystal structures with Mg-nucleotides, interactions of the Walker-B aspartate with Mg2+ are water-mediated. This suggests that our protocol of solvating the MgATP may have been insufficient (see
Finally, we wish to note that our method of visualizing atomic interactions in scatter plots provides a compact and visually intuitive picture of the distinct interaction patterns and the arising asymmetry accessed by our repeat simulations. Similar methods of visualization may be of help in future MD simulation studies, especially as it looks to become more common to extend phase space coverage by repeat simulations.
Although our simulations have been performed on a homodimeric bacterial ABC transporter, the results have possible implications for its (more complex) eukaryotic counterparts. Eukaryotic ABC transporters are often encoded with the two half-transporters in one protein, thus allowing for sequence and (potential) functional differences between the two halves. Indeed, eukaryotic ABC proteins often contain one “consensus” NBS and one “degenerate” NBS
Sequence alignments of conserved NBD motifs in Sav1866 and selected eukaryotic ABC proteins. Asterisks indicate identical residues. Colons/dots indicate closely/less similar residues. Degeneracy substitutions in functional motifs are highlighted by surrounding boxes.
It has been suggested before that the combination of one consensus and one degenerate NBS may lead to
Time-resolved graphs of interactions at the MgATP binding sites. Selected atomic contacts at the MgATP binding sites. Data is plotted separately for the two binding sites in the seven simulations.
(2.04 MB TIF)
Time-resolved graphs of interactions at the NBD-TMD interface. Selected atomic contacts of Q-loop residues Q422-D423 plotted separately for the two NBDs in the seven simulations.
(1.35 MB TIF)
Interactions of the x-loop GERG motif. Scatter plots of contacts formed by residues E473-R474 of the x-loop GERG motif and the TMDs. Contacts observed in the protonated Sav1866 crystal structure are indicated by green dots. Scatter plots were generated as described for
(0.31 MB TIF)
Asymmetric stacking of H204 and R474. Coordinates along the symmetry axis of Sav1866 (z-coordinates) of selected residues in the seven repeat simulations.
(0.30 MB TIF)
Example rearrangement of residue pairs along the Sav1866 symmetry axis.
(1.88 MB TIF)
Schematic overview of studied interactions. One-nanosecond windows are represented by dots. R474 is defined to be closer to the TMDs if its z-position is different by >2 Å to the other R474 z-position. Q-loop Q422-MgATP interactions are colored by three different contact levels: <5 contacts blue, >5 and <14 contacts green, >14 contacts red. Q-loop D423 interactions to x-loop R474 and to K483 are colored by two levels: <1 contact blue, >1 contact red. Q-loop D423 interactions to TMD CL2 are colored: <8 contacts blue, >8 contacts red. All contact values are determined as the mean value over the respective 1 ns window.
(0.25 MB TIF)
Interactions at the ADP binding site in the ADP-bound control simulation. Figure was created in the same fashion as
(0.13 MB TIF)
Interactions at the NBD-TMD interface in the ADP-bound control simulation. Figure was created in the same fashion as
(0.17 MB TIF)
Functionally important interactions in a simulation of the KcsA K+ channel. Distance between the delta-C of protonated E71 and the gamma-C of D80, and E71-D80 contact numbers, derived from a KcsA simulation produced in our laboratory (P. W. Fowler and M. S. P. Sansom, unpublished) and plotted separately for the four monomers. The E71-D80 interaction is important for KcsA slow inactivation
(0.36 MB TIF)
Snapshots from the simulations. Snapshots show the protein, MgATP, and the 40 closest water molecules to MgATP. The PDB file contains six snapshots as separate models. All are from one simulation, at times 0, 5, 10, 20, 30, and 40 ns.
(4.79 MB TXT)
Additional details of method used for generation of scatter plots (for
(0.11 MB DOC)