CHINESE JOURNAL OF CHEMICAL PHYSICS OCTOBER 30, 2019

ARTICLE

Coarse-Grained Free-Energy Simulations of Conformational StateTransitions in an ABC Exporter

Yun Huang, Hao-chen Xu, Jie-Lou Liao∗

Department of Chemical Physics, University of Science and Technology of China, Hefei 230026, China

(Dated: Received on August 6, 2019; Accepted on August 30, 2019)

ATP-binding cassette exporters transport many substrates out of cellular membranes viaalternating between inward-facing and outward-facing conformations. Despite extensive re-search efforts over the past decades, understanding of the molecular mechanism remainselusive. As these large-scale conformational movements are global and collective, we havepreviously performed extensive coarse-grained molecular dynamics simulations of the poten-tial of mean force along the conformational transition pathway [J. Phys. Chem. B 119, 1295(2015)]. However, the occluded conformational state, in which both the internal and externalgate are closed, was not determined in the calculated free energy profile. In this work, weextend the above methods to the calculation of the free energy profile along the reactioncoordinate, d1−d2, which are the COM distances between the two sides of the internal (d1)and the external gate (d2). The potential of mean force is thus obtained to identify thetransition pathway, along which several outward-facing, inward-facing, and occluded statestructures are predicted in good agreement with structural experiments. Our coarse-grainedmolecular dynamics free-energy simulations demonstrate that the internal gate is closed be-fore the external gate is open during the inward-facing to outward-facing transition and viceversa during the inward-facing to outward-facing transition. Our results capture the unidi-rectional feature of substrate translocation via the exporter, which is functionally importantin biology. This finding is different from the previous result, in which both the internal andexternal gates are open reported in an X-ray experiment [Proc. Natl. Acad. Sci. USA 104,19005 (2007)]. Our study sheds light on the molecular mechanism of the state transitions inan (ATP)-binding cassette exporter.

Key words: ATP-binding cassette exporter, Conformational state transition, Coarse-grained molecular dynamics, Potential of mean force

I. INTRODUCTION

Adenosine 5′-triphosphate (ATP)-binding cassette(ABC) exporters are molecular machines that translo-cate a wide variety of substrates out of biological mem-branes [1–3]. ABC exporters share a common archi-tecture with two nucleotide-binding domains (NBDs)and two transmembrane domains (TMDs) that form apathway for substrate translocation (see FIG. 1). As theNBDs dimerize upon ATP binding, they dissociate afterATP hydrolysis and the release of the hydrolytic prod-ucts. The NBD dimerization/dissociation drives large-scale conformational changes for unidirectional trans-port of substrates out of the membrane by alternatingbetween inward-facing (IF) and outward-facing (OF)conformations in the TMDs.

The TMDs comprise twelve transmembrane helices

∗Author to whom correspondence should be addressed.E-mail: liaojl@ustc.edu.cn

(TMs), each has six TMs (TMi and TMi′, i=1−6, re-spectively in FIG. 1). The TMD connects its associatedNBD with two intracellular coupling helices, ICH1 andICH2, which are oriented roughly parallel to the mem-brane plane. An exporter is characterized by a domain-swapped arrangement, i.e., TM4 and TM5 in one sub-unit reach across and contact the NBD in the other sub-unit [4, 5]. This structural arrangement allows TM4 andTM5 in one subunit to interact tightly with TM2′ in theother subunit (approximately in a parallel orientation),facilitating the formation of a TM4-TM5-TM2′ (TM4′-TM5′-TM2) bundle. The TMDs comprise two gates, aninternal gate facing the cytoplasmic side inward facingand an external gate facing the periplasmic side (out-ward facing), which are arranged roughly perpendicu-lar to each other. In an IF state, the internal gate isopen whereas the external gate is closed, allowing sub-strates to access from the cell interior, and vice versa inthe OF conformation with the extrusion pocket exposedto the periplasm [6]. While both gates are closed, theprotein is located at the occluded (OC) state. Despitesubstantial efforts over the past decades, mechanistic

DOI:10.1063/1674-0068/cjcp1908149 c⃝2019 Chinese Physical Society

Chin. J. Chem. Phys. Yun Huang et al.

FIG. 1 Stereoviews of two conformations of an ABC ex-porter (a) inward-facing conformation, in which the inter-nal gate is open whereas the external gate is closed, (b)outward-facing conformation, in which the internal gate isclosed whereas the external gate is open.

understanding of ABC exporter gating movements atthe molecular level is not fully understood [7, 8].

A current understanding of NBD-coupled conforma-tional transitions in the TMDs was based on the full-length crystal structures of the bacterial ABC exporter,MsbA, including one OF conformation (PDB code:3B60; termed 3B60 structure hereafter) and two IFconformations (PDB codes: 3B5X and 3B5W, termedIF-closed (IF-c) and IF-open (IF-o), whose NBDs werefound significantly twisted [9]. Thereby, it was proposedthat the NBD twisting motion drives the IF↔OF tran-sition of MbA. However, recent structural studies ofMsbA [10, 11] have challenged this mechanistic view.Although remarkable progress in the structural biologyfor ABC exporters has been made in the past decade,the nature of the transition pathway in these proteinsremains largely unknown [8].

In principle, detailed elucidation of the molecularmechanism of ABC exporter action requires the eval-uation of free energy profile along a certain “reac-tion coordinate” (i.e., potential mean force, PMF)[6, 8, 12]. These free energy landscapes typically con-tain distinct conformational states separated by highbarriers that prevent an efficient sampling of config-uration space by standard all-atom MD simulations.To address this issue, one way is to apply a biased ornonequilibrium method to enhance sampling of confor-mations with little possibility of access to standard all-atom MD simulations. Moradi and Tajkhorshid usedthe nonequilibrium-driven all-atom MD to study theconformational transition of MsbA with the aforemen-tioned crystal structures, particularly the IF-c one asthe references [12]. Their work offers an interestingmechanistic picture for the highly cooperative move-ments of the transporter as a rigid body. However, the

FIG. 2 Views of the external gate from the periplasmic en-trance. (a) The X-ray IF-c structure of MsbA (PDB code:3B5X) [9], the Cα−Cα distance between the two gatekeeperresidues at the external gate is 18.6 A, showing that the ex-ternal gate is open (see discussion in the main text). (b)A typical IF structure, for example, the C. elegan IF struc-ture (PDB code: 4F4C) [13], the Cα−Cα distance betweenthe gatekeeper residues at the external gate is 5.5 A. (c)The supposition of 3B5X (red) and 4F4C (green). (d) Thesupposition of the above 3B5X (red) and the IF-o structureof MsAb (blue) (PDB code: 3B5W) [9], where the Cα−Cα

distance between the gatekeeper residues is ∼6.5 A.

obtained free energy profile cannot identify the OF andOC states, demonstrating that their calculations wereunable to capture the OF↔IF transition. Most likely,the reaction coordinate, α, which was used in their work[12], is defined as the angle between two sides of the in-ternal gate, but it is too coarse-grained for describingthe gating movements of the protein, as the completeclosure of the (internal) gate is often controlled by a fewgatekeeper residues [6]. Furthermore, careful examina-tion shows that the IFc structure (PDB code: 3B5X) [9]used in their calculation [12] represents a state, in whichthe internal and external gate are both open (see FIG.2(a)). This is contradictory against the unidirectional-ity (i.e., from intracellular to extracellular) of transportsubstrates by an exporter.

Alternatively, as interdomain conformational changesin ABC exporters are global and collective movements[6, 14], a coarse-grained (CG) approach is especially use-ful to reduce the system size and to remove fastest de-grees of freedom [15, 16]. In our previous study, we car-ried out CG-MD free-energy simulations of the OF↔IFtransition in response to the NBD dissociation in a mul-tidrug transporter P-glycoprotein for probing the struc-tural determinants [6]. Although the OF↔IF transitioncan be captured in our calculations, the OC state wasnot yet identified in the free energy profile (see FIG. 2in Ref.[6]) and detailed understanding of the large-scaleconformational changes remains to be completed. In

DOI:10.1063/1674-0068/cjcp1908149 c⃝2019 Chinese Physical Society

Chin. J. Chem. Phys. Conformational State Transitions in an ABC Exporter

this work, we aim to address these issues using the bac-terial ABC exporter, MsbA, as a prototype in the pres-ence of the lipid bilayer and explicit water molecules.The CG-MD umbrella sampling method is employedfor the PMF calculations. Our PMF calculations haveidentified the pathway for the state transitions betweenOF, OC, and IF, and the predicted structures are con-sistent with structural experiments.

II. COMPUTATIONAL METHODS

All simulations in this work were performed with theGROMACS 5.0.5 package [17]. The full-length cryo-EM structure of MsbA, which adopts an IF confor-mation, was taken from the Protein Data Bank (PDBcode: 5TV4) [10] serving as the starting structure forour simulations. MsbA acts as a homodimeric exporterto transport lipopolysaccharide (LPS) responsible forinducing a potent inflammatory response during bacte-rial infection to the outer membrane of gram-negativebacteria. We used the Martini method [15, 16, 18],which has been applied successfully to a large varietyof proteins, lipid membranes, DNAs and RNAs [19–26]to model MsbA together with an elastic network modelfor the stabilization of the backbone conformation of theCG protein [27]. The resulting MsbA CG model was in-serted into a pre-equilibrated POPC lipid bilayer mod-eled with the Martini force field [15, 18]. The MsbA-lipid system was solvated in polarizable Martini water[28, 29] in a rectangular box of 18 nm×14 nm×14 nm(FIG. 3). The final system comprises totally 130898MARTINI particles, including one MsbA protein (2546particles), 990 POPCs, and 116472 polarizable waterCG particles (FIG. 3). In the Martini model, the vander Waals interactions are described using a Lennard-Jones potential function, whereas charged CG particlesinteract via a Coulomb energy function with the dielec-tric constant, εr=15. The cutoff of 1.2 nm was usedfor the Lennard-Jones potential, which was smoothlyshifted to zero between 0.9 and 1.2 nm. The long-rangeelectrostatic interactions were treated using the parti-cle mesh Ewald (PME) method with a real space cut-off of 1.2 nm and a 0.12 nm Fourier grid spacing [30].The simulations performed with the PW model havea van der Waal cut-off of 1.2 nm and a dielectric con-stant, εr=2.5. In all cases, the Verlet scheme for theneighbor list update and the PME method for the elec-trostatic interactions, with a grid spacing of 0.12 nm,are used. The CG-MD simulations were executed witha time step of 20 fs under a periodic boundary condi-tion. The above MsbA-lipid system was weakly cou-pled to the Berendsen thermostat (τT=2 ps) and theParrinello-Rahman barostat (τp=12 ps) fixed at 310 Kand 1 bar in all simulations unless otherwise indicated.Following a 20000-step energy minimization, the abovecoarse-grained protein in the membrane and aqueousenvironment was equilibrated.

FIG. 3 The CG system containing the MsbA exporter(spheres colored red), DOPC lipid bilayer (spheres coloredcyan, blue and orange), and water molecules (spheres col-ored dark blue), a total of 130898 MARTINI particles areused to compute the PMF described in the main text.

The CG-MD simulation and umbrella sampling (US)with the weighted histogram analysis method (WHAM)[31] were then employed to calculate the PMF along thereaction coordinate.

III. RESULTS AND DISCUSSION

The calculations of the free energy profile often re-quire the definition of a “reaction coordinate” designedto monitor the conformational changes in the protein.A frequently used reaction coordinate is the root-mean-squared deviation (RMSD) of an evolving structure rel-ative to its target. However, the quality of the targetstructure is a determinant factor for the reliability ofthe results. As mentioned above, α, defined as the an-gle between the roll axes of two sides of the internalgate, was used as the reaction coordinate to investi-gate the conformational changes in the MsbA protein[12]. This variable is useful to measure the extent ofthe opening/closure of the internal gate as a rigid-body.However, this angle coordinate seems unable to deter-mine whether the gate is closed completely because itusually is a couple of gatekeeper residues that lock thegate [6] as mentioned above. The OF state was thusnot determined in the resulting PMF and the OF↔IFtransition was unable to be described in their study (seeFIG. 4 in Ref.[12]). Intuitively, the COM distance, d1,was employed as a reaction coordinate to probe the freeenergy landscape in our previous study [6]. The thus-obtained PMF captures the OF↔IF transition and thepredicted OF and IF structures are in good agreementwith structural experiments [6]. Unfortunately, the OCstate was not well identified and the OF↔OC transi-tion cannot be described in the calculations (see FIG. 2in Ref.[6]). Careful examination shows that during theOF↔OC transition, the COM distance, d2, which de-scribes the opening/closure extent of the external gate(FIG. 1), is decreased by ∼5.0 A while d1 is increased

DOI:10.1063/1674-0068/cjcp1908149 c⃝2019 Chinese Physical Society

Chin. J. Chem. Phys. Yun Huang et al.

only by ∼2.0 A (2.8−3.0 nm). Therefore, to address theabove issue, we use d1−d2 as the reaction coordinate toevaluate the free energy profile.

In the following discussion, the CG-MD simulationand umbrella sampling with the weighted histogramanalysis method (WHAM) were then employed to calcu-late the potential of mean force (PMF) along the reac-tion coordinate [6]. To this end, we applied weak pullingforces only to the NBDs and a small portion of theTMDs at the cytoplasmic end of the internal gate. Con-sequently, the COM distances, d1 and d2, were gradu-ally changed via the successive pulling and equilibra-tion steps. Structural snapshots were then taken outfrom the CG-MD trajectory to generate the configura-tions for the umbrella sampling simulations. This COMpulling method has been widely applied in various com-plex biological systems [6, 31–33]. In this study, wecomputed PMF along the reaction coordinate, d1−d2,generating 141 windows. For each window, 60 ns CG-MD umbrella sampling simulations were executed. Atotal of effective simulation time [32] was ∼33.8 µs inour calculations.

The resulting PMF along the reaction coordinate,d1−d2, is presented in FIG. 4. The PMF has severalminima, whose atomistic structures can be obtainedusing the back mapping method [34]. These mini-mum points a−d represent the OF, OC, IFc and IF-ostate, respectively. Except for the OC state, all otherstructures predicted from our calculations are similarto those from our previous study [6] and are in goodagreement with structural experiments. Alignment ofthe calculated OC structure (i.e., FIG. 4(b)) to the5TTP structure [10], which adopts an OC conforma-tion, leads to a root-mean-squared deviation (RMSD)of 2.8 A (3.0 A RMSD with the 4S0F structure [35]),in good agreement with experiments. In addition, thefree energies of the b, c, and d states relative to that ofa are 2.3 kcal/mol, −2.4 kcal/mol, and −5.0 kcal/mol,respectively.

Our PMF simulations demonstrate that the OF andIF state transition occurs via an OC state. This is cru-cial for the unidirectionality of the transport function ofan ABC exporter. If the internal and external gate bothopen, it might cause reverse transport of substratesthrough the membrane. Intriguingly, the IFc backboneconformation (i.e., (c) in FIG. 4) is similar to that ofthe heterodimeric ABC exporter TM287/TM288 (PDBcode: 3QF4) (RMSD of 3.3 A for C atoms) [36] ratherthan that of the MsbA IFc (PDB code: 3B5X) (Cα

RMSD of 10.5 A) [9]. Careful examination of the 3B5Xstructure shows that although it was acclaimed to be atan IF state, its external gate is actually open (see FIG.2), demonstrating that this structure corresponds to astate in which the internal and external gate are bothopen.

FIG. 4 PMF presented as a function of d1−d2. Here, d1 andd2 are defined as the COM distances between the two sidesof the internal gate (each containing the associated NBD)and the external gate, respectively.

IV. CONCLUSION

Quantification of the free energy landscape is of im-portance to elucidate the molecular mechanism of thestate transitions in an ABC exporter [8]. In a previousstudy as mentioned above, the nonequilibrium-drivenMD method was applied to calculate the PMF for theMsbA exporter, but no OF and OC states were deter-mined in their calculations [12]. Our previous studyevaluated the PMF for a C. elegans exporter, however,we were unable to identify the OC state [6]. In thispresent work, we used the CG-MD and umbrella sam-pling approaches with the weighted histogram methodto evaluate the free energy profile treating the differ-ence of the COM distances, d1−d2 as the reaction co-ordinate. We thus obtained the PMF that identified areliable pathway, along which the transitions from theOF to the IF state via the OC conformation were cap-tured. The calculated OF, OC, and IF structures are ingood agreement with X-ray experiments. To the best ofour knowledge, this is the first time that the PMF is cal-culated to capture all these conformational transitionsfor an ABC exporter.

Detailed analysis of the structural changes along thetransition pathway (FIG. 4) demonstrates that duringthe transition from the OF to IF state driven by theNBD dissociation, the external gate is closed before theinternal gate is opened, and vice verse in the transitionfrom the IF to OF state. This feature is essential forthe unidirectionality of the substrate transport via anABC exporter. The determination of the OC state inthe PMF is important for understanding the mechanismof the OF↔IF transition. It should be pointed out thatno state with both gates open was found in our PMFcalculations, indicating that such conformations wouldbe located at high free-energy levels.

DOI:10.1063/1674-0068/cjcp1908149 c⃝2019 Chinese Physical Society

Chin. J. Chem. Phys. Conformational State Transitions in an ABC Exporter

V. ACKNOWLEDGMENTS

This work was supported by the National Natu-ral Science Foundation of China (No.21073170 andNo.21273209). We gratefully acknowledge the Super-computing Center at University of Science and Tech-nology of China for computational resources.

[1] K. Hollenstein, R. J. Dawson, and K. P. Locher, Curr.Opin. Struct. Biol. 17, 412 (2007).

[2] A. L. Davidson and P. C. Maloney, Trends Microbiol.15, 448 (2007).

[3] D. C. Rees, E. Johnson, and O. Lewinson, Nat. Rev.Mol. Cell Biol. 10, 218 (2009).

[4] R. J. Dawson and K. P. Locher, Nature 443, 180 (2006).[5] R. J. Dawson and K. P. Locher, FEBS Lett. 581, 935

(2007).[6] Z. Wang and J. L. Liao, J. Phys. Chem. B 119, 1295

(2015).[7] K. P. Locher, Nat. Struct. Mol. Biol. 23, 487 (2016).[8] D. Szollosi, D. Rose-Sperling, U. A. Hellmich, and T.

Stockner, BBA Biomembranes 1860, 818 (2018).[9] A. Ward, C. L. Reyes, J. Yu, C. B. Roth, and G. Chang,

Proc. Natl. Acad. Sci. USA 104, 19005 (2007).[10] W. Mi, Y. Li, S. H. Yoon, R. K. Ernst, T. Walz, and

M. Liao, Nature 159, 233 (2017).[11] H. Ho, A. Miu, M. K. Alexander, N. K. Garcia, A.

Oh, I. Zilberleyb, M. Reichelt, C. D. Austin, C. Tam,S. Shriver, H. Hu, S. S. Labadie, J. Liang, L. Wang,Jian Wang, Y. Lu, H. E. Purkey, J. Quinn, Y. Franke,K. Clark, M. H. Beresini, M. -W. Tan, B. D. Sellers,T. M., Michael, F. T. Koehler, A. T. Wecksler, J. R.Kiefer, V. Verma, Y. Xu, M. Nishiyama, J. Payandeh,and C. M. Koth, Nature 557, 196 (2018).

[12] M. Moradi and E. Tajkhorshid, Proc. Natl. Acad. Sci.USA 110, 18916 (2013).

[13] M. S. Jin, M. L. Oldham, Q. Zhang, and J. Chen, Na-ture 490, 566 (2012).

[14] J. L. Liao and D. N. Beratan, Biophys. J. 87, 1369(2004).

[15] S. J. Marrink, H. J. Risselada, S. Yefimov, D. P. Tiele-man, and A. H. Vries, J. Phys. Chem. B 111, 7812(2007).

[16] L. Monticelli, S. K. Kandasamy, X. Periole, R. G. Lar-son, D. P. Tieleman, and S. J. Marrink, J. Chem. The-

ory Comput. 4, 819 (2008).[17] M. J. Abraham, M. Murtola, R. Schulz, S. Pall, J. C.

Smith, B. Hess, and E. Lindahl, Software X 1-2, 19(2015).

[18] T. A. Wassenaar, H. I. Ingolfsson, R. A. Boeckmann,D. P. Tieleman, and S. J. Marrink, J. Chem. TheoryComput. 11, 2144 (2015).

[19] C. A. Lopez, Z. Sovova, F. J. Van Eerden, A. H. DeVries, and S. J. Marrink, J. Chem. Theory Comput. 9,1694 (2013).

[20] H. I. Ingolfsson, C. A. Lopez, J. J. Uusitalo, D. H. DeJong, S. M. Gopal, X. Periole, and S. J. Marrink, WiresComput. Mol. Sci. 4, 225 (2014).

[21] J. J. Uusitalo, H. I. Ingolfsson, P. Akhshi, D. P. Tiele-man, and S. J. Marrink, J. Chem. Theory Comput. 11,3932 (2015).

[22] J. J. Uusitalo, H. I. Ingolfsson, S. J. Marrink, and I.Faustino, Biophys. J. 113, 246 (2017).

[23] H. I. Ingolfsson, T. S. Carpenter, H. Bhatia, P. T. Bre-mer, S. J. Marrink, and F. C. Lightstone, Biophys. J.113, 2271 (2017).

[24] F. Grunewald, G. Rossi, A. H. De Vries, S. J. Marrink,and L. Monticelli, J. Phys. Chem. B 122, 7436 (2018).

[25] M. Xue, L. Cheng, I. Faustino, W. Guo, and S. J. Mar-rink, Biophys. J. 115, 494 (2018).

[26] S. J. Marrink, V. Corradi, P. C. T. Souza, H. I.Ingolfsson, D. P. Tieleman, and M. S. P. Sansom, Chem.Rev. 119, DOI:10.1021/acs.chemrev.8b00460 (2019).

[27] X. Periole, M. Cavalli, S. J. Marrink, and M. A. Ceruso,J. Chem. Theory Comput. 5, 2531 (2009).

[28] S. J. Marrink, PLoS Comput. Biol. 6, E1000810 (2010).[29] J. Michalowsky, L. V. Schaefer, C. Holm, and J. Smi-

atek, J. Chem. Phys. 146, 054501 (2017).[30] T. Darden, D. York, and L. Pedersen, J. Chem. Phys.

98, 10089 (1993).[31] J. K. Lemkul and D. R. Bevan, J. Phys. Chem. B 114,

1652 (2010).[32] X. Periole, A. M. Knepp, T. P. Sakmar, S. J. Marrink,

and T. Huber, J. Am. Chem. Soc. 134, 10959 (2012).[33] J. Domanski, G. Hedger, R. B. Best, P. J. Stansfeld,

and M. S. P. Sansom, 121, 3364 (2017).[34] T. A. Wassenaar, K. Pluhackova, R. A. Bockman, S. J.

Marrink, and D. P. Tieleman, J. Chem. Theory Com-put. 10, 676 (2014).

[35] D. Y. Lin, S. Huang, and J. Chen, Nature 523, 425(2015).

[36] M. Hohl, C. Briand, M. G. Grutte, and M. A. Seeger,Nat. Struct. Mol. Biol. 19, 395 (2012).

DOI:10.1063/1674-0068/cjcp1908149 c⃝2019 Chinese Physical Society