Integral membrane proteins can have a marked impact on the morphology of the surrounding bilayer. For example, they might alter its curvature or thickness, or foster the enrichment or depletion of specific types of lipid in their vicinity (Lee, 2004; Andersen and Koeppe, 2007; Marsh, 2008; Phillips et al., 2009). These perturbations develop to accommodate the amino-acid composition and specific structural features of the protein surface. For a broad range of systems of interest, however, these features are not static, but vary as the protein carries out its biological activity. Examples include channel proteins and their gating mechanisms, or the conformational cycles resulting in alternating access in active transporters. The protein-lipid interface also changes when membrane proteins form complexes or oligomers. Because most membrane perturbations entail an energetic or entropic cost, it seems reasonable to expect that the bilayer contributes to the thermodynamics and kinetics of these processes, favoring or disfavoring specific structural states. Such interdependence would help explain why changes in lipid bilayer composition, either resulting from natural regulatory mechanisms or induced artificially, can have a determining effect on protein function (Jensen and Mouritsen, 2004; Andersen and Koeppe, 2007; Denning and Beckstein, 2013; Cordero-Morales and Vásquez, 2018; Haselwandter and MacKinnon, 2018).

Here, we seek to gain insights into this interplay for a class of membrane proteins known to undergo a striking structural transformation, namely secondary-active transporters featuring the so-called elevator mechanism. We specifically focus on the Na⁺-aspartate symporter Glt_Ph from Pyrococcus horikoshii, a member and model system of the Excitatory Amino-Acid Transporter (EAAT) family, which includes the human SLC1 neurotransmitter transporters (Gether et al., 2006). Secondary-active transporters like Glt_Ph interconvert between two major conformational states that alternately expose binding sites for ions and substrate to either side of the membrane (Figure 1) (Jardetzky, 1966; Reyes et al., 2009; Boudker and Verdon, 2010; Forrest et al., 2011). In Glt_Ph, this conformational interconversion takes place only when Na⁺ and aspartate occupy the transporter or when the transporter is empty (Yernool et al., 2004; Boudker et al., 2007; Ryan et al., 2009), which defines this protein as a co-transporter. For either occupancy state, the exchange between outward- and inward-facing conformations is spontaneous, that is it does not require an extrinsic electrochemical driving force. However, the relative populations of these conformational states and the net directionality of the transport cycle do depend on the membrane potential and the relative concentrations of ions and substrates on either side of the membrane. This dependence is why downhill translocation of Na⁺ will power uphill transport of aspartate, and vice versa.

Figure 1 with 1 supplement see all

Download asset Open asset

Glt_Ph forms a homotrimer (Yernool et al., 2004); yet, each protomer works independently from its neighbors (Akyuz et al., 2013; Erkens et al., 2013; Akyuz et al., 2015; Ruan et al., 2017). The protomers consist of two distinct units: the scaffold domain, which provides a stable trimerization interface (Groeneveld and Slotboom, 2007; Verdon and Boudker, 2012; Georgieva et al., 2013; Hänelt et al., 2013), and the transport domain, which encapsulates the ion and aspartate binding sites (Yernool et al., 2004; Reyes et al., 2009). Both domains are exposed to the lipid bilayer, but structures of Glt_Ph in outward- and inward-facing states clearly reveal that it is the movement of the latter domain that results in alternating access (Yernool et al., 2004; Reyes et al., 2009). Specifically, if one assumes that the scaffold domains remain stationary relative to the membrane mid-plane, each transport domain would move perpendicularly to that plane by approximately 15 Å, that is about one-third of the total bilayer width. How this drastic structural change impacts the relationship between protein and membrane has, to our knowledge, not been previously evaluated. One might envisage that the transport domains simply traverse the membrane (like an elevator), moving polar sidechains on their surface into the bilayer interior and exposing hydrophobic ones to the solvent (Reyes et al., 2009). Alternatively, the morphology of the lipid bilayer could adapt to the conformational state of the protein, and match the amino-acid make-up of the protein surface throughout the transport cycle. Which of these two possibilities entails the least energetic cost is entirely unclear. The absence of functional cooperativity among protomers would appear to rule out the latter, as a pronounced membrane deformation induced by one protomer could impact the conformational dynamics of its two neighbors. On the other hand, the cumulative cost of dehydration of polar and charged sidechains could be exceedingly large. Experimental studies of the Glt_Ph homolog EAAT2 also appear to show that bilayer-facing regions in the outward-facing state are similarly solvent-accessible throughout the transport cycle (Silverstein et al., 2013).

Molecular dynamics simulations in principle provide a means to examine this interplay in great detail (Cui et al., 2013; Marrink et al., 2019). Current computing power makes it feasible to examine the morphology of simple phospholipid membranes around protein structures, for many cases of interest. Highly complex membranes remain beyond reach, however, if represented in atomic detail, as the relaxation time of an arbitrarily-configured multi-component lipid mixture is slower than typically attainable simulation times, thus imposing a starting-condition bias on any analysis. In such cases, so-called coarse-grained simulations are a viable approach despite their reduced level of detail (Corradi et al., 2018). By contrast, simulation methods that permit a direct quantification of the energetic footprint of a membrane morphological change have been lacking. Computational approaches in this area have typically relied on continuum-mechanics theories of membrane elasticity, which necessarily pre-suppose important parameters defining the bilayer energetics (Argudo et al., 2017). Although such approaches can be predictive and insightful in some cases (Mondal et al., 2011; Bethel and Grabe, 2016), it is unclear whether macroscopic models are generally transferable to the length-scales of individual membrane proteins (Lee et al., 2013; Fiorin et al., 2019). To circumvent this problem, we have recently developed a free-energy simulation method (Fiorin et al., 2019) with which the potential-of-mean-force of a membrane deformation can be probed directly from a simulation system, much in the same way processes such as ligand recognition or ion permeation are commonly characterized. Here, in addition to conventional simulations and structural bioinformatic approaches, we apply this new technology to obtain quantitative insights into the nature of the interplay between Glt_Ph and the surrounding membrane. The implications of the major conformational change required for transport are discussed in the light of available experimental observations.

Our simulations predict that the conformational cycle of Glt_Ph induces a long-ranged remodeling of the surrounding lipid membrane, as a result of the elevator-like motion of the transport domains, each of which displaces its protein-lipid interface by about one-half the width of the bilayer hydrophobic core. Strikingly, this perturbation is abruptly suppressed at the point where the bilayer meets the trimerization domain; therefore, each protomer induces an independent deformation, consistent with existing experimental evidence demonstrating no protomer-protomer cooperativity in this class of transporters (Grewer et al., 2005; Koch and Larsson, 2005; Koch et al., 2007; Leary et al., 2007; Akyuz et al., 2013; Erkens et al., 2013; Akyuz et al., 2015; Ruan et al., 2017). These results provide a working hypothesis and as such require experimental verification; while direct structural data confirming the drastic effects that we predict here are still lacking, such information is not beyond reach. As demonstrated by recent structural studies of Ca²⁺-dependent lipid scramblases, single-particle cryo-EM can reveal the morphological features of detergent micelles and lipid nanodiscs in considerable detail (Falzone et al., 2018; Falzone et al., 2019; Kalienkova et al., 2019); crystallographic studies can also reveal the contours of the lipid bilayer, as observed for stacked membrane crystals of the SERCA Ca²⁺-ATPase pump (Sonntag et al., 2011). Analogous data for alternate conformational states of Glt_Ph, or a close homolog thereof, ought to validate or refute the predicted impact of this transporter on the membrane morphology.

Single-molecule FRET measurements of substrate-loaded and substrate-free Glt_Ph in outside-out proteoliposomes have indicated that the intrinsic free-energy difference between the outward- and inward-facing states of the transporter is approximately zero, that is, both states are approximately equally populated (Akyuz et al., 2013; Erkens et al., 2013; Akyuz et al., 2015). Here, we have shown that the membrane deflection induced by inward-facing Glt_Ph translates into a substantial energetic penalty, which we estimate to be in the order of 6–7 kcal/mol per protomer, or about 20 kcal/mol in total. To balance out this cost, therefore, the inward-facing conformation must somehow recoup this energy, through distinct protein-protein, protein-lipid and/or protein-water interactions. Admittedly, at this point we can only infer the magnitude of free-energy penalty from simulations where the Multi-Map method mimics the impact of the protein on the membrane. While imperfect, it is worth underscoring the technical breakthrough made by this approach: it provides a means to estimate the membrane energetics directly from simulated molecular-dynamics trajectories, without a priori theoretical assumptions. Thus, we believe that further methodological developments and systematic applications of this methodology, in combination with other enhanced-sampling techniques, will result in increasingly precise estimates, and will also enable us to dissect the compensating interactions that must occur in this and other membrane-protein systems. These caveats notwithstanding, we believe our estimate of the free-energy penalty of membrane bending not to be at all unrealistic. For example, based on electrophysiological studies of the gating mechanism of the Shaker voltage-gated K⁺ channel, it has been deduced that at zero membrane potential the conformational free-energy of the open state is about 15 kcal/mol lower than that of the closed state (Chowdhury and Chanda, 2012). Similarly, a value of 16 kcal/mol was deduced for the Na⁺-channel Na_v1.4 (Chowdhury and Chanda, 2012). Regardless of the specifics, this study underscores that the configurational energetics of the lipid bilayer are non-negligible and must therefore be incorporated into the conceptual models and theories used to describe membrane-protein mechanisms – much in the same way one would consider, for example, the dehydration energetics of different ions (also in the order of tens of kcal/mol) when rationalizing the selectivity or conductance rates of a channel protein. It is hoped, therefore, that the calculations presented here will spur further biophysical studies designed to assess the impact of bilayer composition and morphology on the energetics of transporters – in analogy with research other membrane proteins such as receptors (Brown and Chawla, 2017) and channels (Phillips et al., 2009; Andersen, 2013). For example, it would be of interest to dissect how the functional dynamics of a transporter such as Glt_Ph are influenced by the amino-acid make-up of the protein-lipid interface, as well as by lipid bilayer composition. Indeed, functional studies have shown that subtle lipid modifications (methylation) and single-point mutations (Tyr33) have detectable effects on the rates of Glt_Ph transport (McIlwain et al., 2015). It is also intriguing that long-chain polyunsaturated fatty acids (PUFA) modulate the activity of neuronal EAAT transporters (Zerangue et al., 1995; Fairman et al., 1998; Tzingounis et al., 1998; Grintal et al., 2009) and are also known to influence the bending energetics of lipid bilayers (Cordero-Morales and Vásquez, 2018).

The aforementioned smFRET studies have also indicated that the balance of outward- and inward-facing states hardly differs when measured in liposomes (containing mostly PE, PG and PC lipids) or in DDM micelles (Akyuz et al., 2015). Taken together with our findings, this observation challenges the notion that detergent micelles have no distinct morphological preference and will comply to the conformation of a protein at little or no energetic cost. To the contrary, when a detergent solution is concentrated above a certain threshold, the micelles that form have an inherently preferred geometry and size (Lipfert et al., 2007; Oliver et al., 2013), that is, there exists a well-defined free-energy minimum of micelle formation. When a micelle assembles around a protein, one might expect that free-energy minimum to naturally shift, that is, the micelle morphology will adapt to the volume and surface of the protein. However, this adaption is not necessarily identical for all conformations of a protein; indeed, one might expect that different free-energy minima will exist for alternative structural states (both in magnitude and morphology), and that this difference in will be greater for some detergents than for others. The expectation that the deformation of a micelle entails little or no energetic cost is even less intuitive if one assumes that the micelle does not reassemble when a protein changes structure. Indeed, the above-mentioned cryo-EM data for nhTMEM16 shows that a highly localized feature of the protein surface causes nearly identical morphological changes in a lipid nanodisc and in a DDM micelle, namely a perturbation that gradually decays along the protein perimeter (Kalienkova et al., 2019). If the DDM micelle was indeed much softer or more compliant than the lipid nanodisc, the resulting curvature would be more localized. Thus, on this matter too we believe our data highlights the need for further work probing how the thermodynamics of micellar or lipid solvation might depend on the conformational state of a protein, or on its oligomerization state.

The dynamics of Glt_Ph at the single-molecule level have also been evaluated using high-speed AFM measurements in protein-dense 2D preparations (Ruan et al., 2017). These elegant measurements clearly confirmed that each protomer exchanges between outward- and inward-facing states stochastically (provided the adequate occupancy state) and independently from each other. Intriguingly, though, in these measurements the population ratio between outward- and inward-facing is shifted significantly against the latter, by about 5-fold (Ruan et al., 2017; Heath and Scheuring, 2019). A possible explanation for the discrepancy between the smFRET and AFM measurements is that the curvature of the outside-out proteoliposomes used in the former experiment favors the inward-facing state, while the flat membranes in the latter do not. However, our finding that the membrane perturbation caused by Glt_Ph projects away from the protein surface for several nanometers provides an alternative explanation: transition to the inward-facing state would be more energetically costly when this deformation must occur in a more confined space. In other words, it is conceivable that the aforementioned balance between large, competing energetic contributions is altered as a result of molecular crowding, in this case favoring the outward-facing conformation. Interestingly, some members of the EAAT family are expressed in the membranes of certain cell types at very high densities (in the order of thousands of transporters per square-micron [Danbolt, 2001]), raising the possibility that crowding effects occur in physiological settings. On a related, more technical note, it is worth noting that molecular simulation studies aiming to evaluate the alternating-access mechanism and protein-lipid interplay for cases such as Glt_Ph might be significantly skewed by finite-size effects if the lipid bilayer patch is not sufficiently large to accommodate the full range of the membrane perturbation created by the protein.

Large-scale structural changes are not exclusive to membrane proteins in the EAAT superfamily. Elevator-like secondary-active transporters are, however, appealing model systems to evaluate the interplay between proteins and the lipid bilayer, given the range of their motions and that these motions occur spontaneously and, in some cases, on time-scales amenable to single-molecule measurements like smFRET or high-speed AFM. As an example of another elevator-like mechanism, an initial analysis of the bacterial sodium-dicarboxylate symporter VcINDY is shown in Figure 8. VcINDY is a dimer, but is similar to Glt_Ph in that each of the constituent protomers features two distinct domains, one facilitating dimerization and the other responsible for substrate translocation (Mancusso et al., 2012; Mulligan et al., 2016). Our simulations show that upon transition from the (predicted) outward-facing state to the (experimentally determined) inward-facing conformation, the transport domains would induce a deformation in the membrane that is comparable to that observed for Glt_Ph. Again, this deformation is suppressed by the scaffold, suggesting that the protomers function independently. Despite the commonalities between Glt_Ph and VcINDY, however, it is important to note that other elevator-like transporters, perhaps even those in the same structural family, might influence the membrane differently, as this interplay is ultimately determined by the amino-acid make-up of the protein surface. For example, it is entirely possible that, in some cases, the outward-facing state perturbs the membrane the most. To document and dissect this variability is no doubt of interest and will be the focus of future studies.

Figure 8

Download asset Open asset

Glt_Ph and VcINDY are stark examples of membrane proteins that cause large-scale morphological changes in the bilayer. Needless to say, numerous channels and transporters also undergo major conformational changes during function, which might impact the membrane to varying degrees. The computational analysis presented here makes it clear that the inherent configurational energetics of the lipid bilayer can be both non-negligible and markedly state-dependent, and must therefore be integrated in our conceptualizations of membrane protein mechanisms.

Input and output files for 1 (out of 3) replica of each simulation system/condition in our study have been uploaded to Zenodo, a public repository free of charge, and is available at the DOI: https://doi.org/10.5281/zenodo.3558957.

1. 1. Zhou Wenchang
   2. Fiorin Giacomo
   3. Anselmi Claudio
   4. Karimi-Varzaneh Hossein Ali
   5. Poblete Horacio
   6. Forrest Lucy Rachel
   7. Faraldo-Gómez José Diego

   (2019) Zenodo

   Simulation files for "Large-scale state-dependent membrane remodeling by a transporter protein".

   https://doi.org/10.5281/zenodo.3558957

1. 1. Gouaux E
   2. Boudker O
   3. Ryan R
   4. Yernool D
   5. Shimamoto K

   (2007) Protein Data Bank

   ID 2NWX. Crystal structure of GltPh in complex with L-aspartate and sodium ions.

   https://www.rcsb.org/structure/2NWX
2. 1. Reyes N
   2. Ginter C
   3. Boudker O

   (2009) Protein Data Bank

   ID 3KBC. Crystal structure of GltPh K55C-A364C mutant crosslinked with divalent mercury.

   https://www.rcsb.org/structure/3KBC
3. 1. Nie R
   2. Stark S
   3. Symersky J
   4. Kaplan RS
   5. Lu M

   (2017) Protein Data Bank

   ID 5ULD. Structure and function of the divalent anion/Na+ symporter from Vibrio cholerae and a humanized variant.

   https://www.rcsb.org/structure/5ULD
4. 1. Mulligan C
   2. Fenollar-Ferrer C
   3. Fitzgerald GA
   4. Vergara-Jaque A
   5. Kaufmann D
   6. Li Y
   7. Forrest LR
   8. Mindell JA

   (2016) Protein Model Data Base

   ID PM0080216. Model of outward-facing VcINDY.

   http://srv00.recas.ba.infn.it/PMDB/zip_model.php?target=5554

1. 1. Akyuz N
   2. Altman RB
   3. Blanchard SC
   4. Boudker O

   (2013) Transport dynamics in a glutamate transporter homologue

   Nature 502:114–118.

   https://doi.org/10.1038/nature12265

   • PubMed
   • Google Scholar
2. 1. Akyuz N
   2. Georgieva ER
   3. Zhou Z
   4. Stolzenberg S
   5. Cuendet MA
   6. Khelashvili G
   7. Altman RB
   8. Terry DS
   9. Freed JH
   10. Weinstein H
   11. Boudker O
   12. Blanchard SC

   (2015) Transport domain unlocking sets the uptake rate of an aspartate transporter

   Nature 518:68–73.

   https://doi.org/10.1038/nature14158

   • PubMed
   • Google Scholar
3. 1. Andersen OS

   (2013) Membrane proteins: through thick and thin

   Nature Chemical Biology 9:667.

   https://doi.org/10.1038/nchembio.1368

   • PubMed
   • Google Scholar
4. 1. Andersen OS
   2. Koeppe RE

   (2007) Bilayer thickness and membrane protein function: an energetic perspective

   Annual Review of Biophysics and Biomolecular Structure 36:107–130.

   https://doi.org/10.1146/annurev.biophys.36.040306.132643

   • PubMed
   • Google Scholar
5. 1. Argudo D
   2. Bethel NP
   3. Marcoline FV
   4. Wolgemuth CW
   5. Grabe M

   (2017) New continuum approaches for determining Protein-Induced membrane deformations

   Biophysical Journal 112:2159–2172.

   https://doi.org/10.1016/j.bpj.2017.03.040

   • PubMed
   • Google Scholar
6. 1. Berendsen HJC
   2. Postma JPM
   3. van Gunsteren WF
   4. DiNola A
   5. Haak JR

   (1984) Molecular dynamics with coupling to an external bath

   The Journal of Chemical Physics 81:3684–3690.

   https://doi.org/10.1063/1.448118

   • Google Scholar
7. 1. Best RB
   2. Zhu X
   3. Shim J
   4. Lopes PE
   5. Mittal J
   6. Feig M
   7. Mackerell AD

   (2012) Optimization of the additive CHARMM all-atom protein force field targeting improved sampling of the backbone φ, ψ and side-chain χ (1) and χ(2) dihedral angles

   Journal of Chemical Theory and Computation 8:3257–3273.

   https://doi.org/10.1021/ct300400x

   • PubMed
   • Google Scholar
8. 1. Bethel NP
   2. Grabe M

   (2016) Atomistic insight into lipid translocation by a TMEM16 scramblase

   PNAS 113:14049–14054.

   https://doi.org/10.1073/pnas.1607574113

   • PubMed
   • Google Scholar
9. 1. Boudker O
   2. Ryan RM
   3. Yernool D
   4. Shimamoto K
   5. Gouaux E

   (2007) Coupling substrate and ion binding to extracellular gate of a sodium-dependent aspartate transporter

   Nature 445:387–393.

   https://doi.org/10.1038/nature05455

   • PubMed
   • Google Scholar
10. 1. Boudker O
    2. Verdon G

    (2010) Structural perspectives on secondary active transporters

    Trends in Pharmacological Sciences 31:418–426.

    https://doi.org/10.1016/j.tips.2010.06.004

    • PubMed
    • Google Scholar
11. 1. Brannigan G
    2. Brown FL

    (2006) A consistent model for thermal fluctuations and protein-induced deformations in lipid bilayers

    Biophysical Journal 90:1501–1520.

    https://doi.org/10.1529/biophysj.105.075838

    • PubMed
    • Google Scholar
12. Book

    1. Brown MF
    2. Chawla U

    (2017)

    The Biophysics of Cell Membranes, Biological Consequences

    Springer.

    • Google Scholar
13. 1. Canham PB

    (1970) The minimum energy of bending as a possible explanation of the biconcave shape of the human red blood cell

    Journal of Theoretical Biology 26:61–81.

    https://doi.org/10.1016/S0022-5193(70)80032-7

    • PubMed
    • Google Scholar
14. 1. Chowdhury S
    2. Chanda B

    (2012) Estimating the voltage-dependent free energy change of ion channels using the median voltage for activation

    The Journal of General Physiology 139:3–17.

    https://doi.org/10.1085/jgp.201110722

    • PubMed
    • Google Scholar
15. 1. Cordero-Morales JF
    2. Vásquez V

    (2018) How lipids contribute to ion channel function, a fat perspective on direct and indirect interactions

    Current Opinion in Structural Biology 51:92–98.

    https://doi.org/10.1016/j.sbi.2018.03.015

    • Google Scholar
16. 1. Corradi V
    2. Mendez-Villuendas E
    3. Ingólfsson HI
    4. Gu RX
    5. Siuda I
    6. Melo MN
    7. Moussatova A
    8. DeGagné LJ
    9. Sejdiu BI
    10. Singh G
    11. Wassenaar TA
    12. Delgado Magnero K
    13. Marrink SJ
    14. Tieleman DP

    (2018) Lipid-Protein interactions are unique fingerprints for membrane proteins

    ACS Central Science 4:709–717.

    https://doi.org/10.1021/acscentsci.8b00143

    • PubMed
    • Google Scholar
17. 1. Cui Q
    2. Zhang L
    3. Wu Z
    4. Yethiraj A

    (2013) Generation and sensing of membrane curvature: where materials science and biophysics meet

    Current Opinion in Solid State and Materials Science 17:164–174.

    https://doi.org/10.1016/j.cossms.2013.06.002

    • Google Scholar
18. 1. Danbolt NC

    (2001) Glutamate uptake

    Progress in Neurobiology 65:1–105.

    https://doi.org/10.1016/S0301-0082(00)00067-8

    • PubMed
    • Google Scholar
19. 1. Denning EJ
    2. Beckstein O

    (2013) Influence of lipids on protein-mediated transmembrane transport

    Chemistry and Physics of Lipids 169:57–71.

    https://doi.org/10.1016/j.chemphyslip.2013.02.007

    • PubMed
    • Google Scholar
20. 1. Erkens GB
    2. Hänelt I
    3. Goudsmits JM
    4. Slotboom DJ
    5. van Oijen AM

    (2013) Unsynchronised subunit motion in single trimeric sodium-coupled aspartate transporters

    Nature 502:119–123.

    https://doi.org/10.1038/nature12538

    • PubMed
    • Google Scholar
21. 1. Fairman WA
    2. Sonders MS
    3. Murdoch GH
    4. Amara SG

    (1998) Arachidonic acid elicits a substrate-gated proton current associated with the glutamate transporter EAAT4

    Nature Neuroscience 1:105–113.

    https://doi.org/10.1038/355

    • PubMed
    • Google Scholar
22. Preprint

    1. Falzone ME
    2. Rheinberger J
    3. Lee B-C
    4. Peyear T
    5. Sasset L
    6. Raczkowski A
    7. Eng E
    8. Lorenzo AD
    9. Andersen OS
    10. Nimigean CM
    11. Accardi A

    (2018) Cryo-EM structures reveal bilayer remodeling during Ca2+ activation of a TMEM16 scramblase

    bioRxiv.

    https://doi.org/10.1101/420174

    • Google Scholar
23. 1. Falzone ME
    2. Rheinberger J
    3. Lee BC
    4. Peyear T
    5. Sasset L
    6. Raczkowski AM
    7. Eng ET
    8. Di Lorenzo A
    9. Andersen OS
    10. Nimigean CM
    11. Accardi A

    (2019) Structural basis of Ca²⁺-dependent activation and lipid transport by a TMEM16 scramblase

    eLife 8:e43229.

    https://doi.org/10.7554/eLife.43229

    • PubMed
    • Google Scholar
24. 1. Fiorin G
    2. Marinelli F
    3. Faraldo‐Gómez JD

    (2019) Direct derivation of free energies of membrane deformation and other solvent density variations from enhanced sampling molecular dynamics

    Journal of Computational Chemistry 118:.

    https://doi.org/10.1002/jcc.26075

    • Google Scholar
25. 1. Flood E
    2. Boiteux C
    3. Lev B
    4. Vorobyov I
    5. Allen TW

    (2019) Atomistic simulations of membrane ion channel conduction, gating, and modulation

    Chemical Reviews 119:7737–7832.

    https://doi.org/10.1021/acs.chemrev.8b00630

    • PubMed
    • Google Scholar
26. 1. Forrest LR
    2. Krämer R
    3. Ziegler C

    (2011) The structural basis of secondary active transport mechanisms

    Biochimica Et Biophysica Acta (BBA) - Bioenergetics 1807:167–188.

    https://doi.org/10.1016/j.bbabio.2010.10.014

    • Google Scholar
27. 1. Georgieva ER
    2. Borbat PP
    3. Ginter C
    4. Freed JH
    5. Boudker O

    (2013) Conformational ensemble of the sodium-coupled aspartate transporter

    Nature Structural & Molecular Biology 20:215–221.

    https://doi.org/10.1038/nsmb.2494

    • PubMed
    • Google Scholar
28. 1. Gether U
    2. Andersen PH
    3. Larsson OM
    4. Schousboe A

    (2006) Neurotransmitter transporters: molecular function of important drug targets

    Trends in Pharmacological Sciences 27:375–383.

    https://doi.org/10.1016/j.tips.2006.05.003

    • PubMed
    • Google Scholar
29. 1. Goetz R
    2. Gompper G
    3. Lipowsky R

    (1999) Mobility and elasticity of Self-Assembled membranes

    Physical Review Letters 82:221–224.

    https://doi.org/10.1103/PhysRevLett.82.221

    • Google Scholar
30. 1. Grewer C
    2. Balani P
    3. Weidenfeller C
    4. Bartusel T
    5. Tao Z
    6. Rauen T

    (2005) Individual subunits of the glutamate transporter EAAC1 homotrimer function independently of each other

    Biochemistry 44:11913–11923.

    https://doi.org/10.1021/bi050987n

    • PubMed
    • Google Scholar
31. 1. Grintal B
    2. Champeil-Potokar G
    3. Lavialle M
    4. Vancassel S
    5. Breton S
    6. Denis I

    (2009) Inhibition of astroglial glutamate transport by polyunsaturated fatty acids: evidence for a signalling role of docosahexaenoic acid

    Neurochemistry International 54:535–543.

    https://doi.org/10.1016/j.neuint.2009.02.018

    • PubMed
    • Google Scholar
32. 1. Groeneveld M
    2. Slotboom DJ

    (2007) Rigidity of the subunit interfaces of the trimeric glutamate transporter GltT during translocation

    Journal of Molecular Biology 372:565–570.

    https://doi.org/10.1016/j.jmb.2007.06.067

    • PubMed
    • Google Scholar
33. 1. Hänelt I
    2. Wunnicke D
    3. Bordignon E
    4. Steinhoff HJ
    5. Slotboom DJ

    (2013) Conformational heterogeneity of the aspartate transporter glt(Ph)

    Nature Structural & Molecular Biology 20:210–214.

    https://doi.org/10.1038/nsmb.2471

    • PubMed
    • Google Scholar
34. 1. Harpole TJ
    2. Delemotte L

    (2018) Conformational landscapes of membrane proteins delineated by enhanced sampling molecular dynamics simulations

    Biochimica Et Biophysica Acta (BBA) - Biomembranes 1860:909–926.

    https://doi.org/10.1016/j.bbamem.2017.10.033

    • Google Scholar
35. 1. Haselwandter CA
    2. MacKinnon R

    (2018) Piezo's membrane footprint and its contribution to mechanosensitivity

    eLife 7:e41968.

    https://doi.org/10.7554/eLife.41968

    • PubMed
    • Google Scholar
36. 1. Heath GR
    2. Scheuring S

    (2019) Advances in high-speed atomic force microscopy (HS-AFM) reveal dynamics of transmembrane channels and transporters

    Current Opinion in Structural Biology 57:93–102.

    https://doi.org/10.1016/j.sbi.2019.02.008

    • PubMed
    • Google Scholar
37. 1. Helfrich W

    (1973) Elastic properties of lipid bilayers: theory and possible experiments

    Zeitschrift Für Naturforschung C 28:693–703.

    https://doi.org/10.1515/znc-1973-11-1209

    • Google Scholar
38. 1. Hess B
    2. Kutzner C
    3. van der Spoel D
    4. Lindahl E

    (2008) GROMACS 4: Algorithms for Highly Efficient, Load-Balanced, and Scalable Molecular Simulation

    Journal of Chemical Theory and Computation 4:435–447.

    https://doi.org/10.1021/ct700301q

    • Google Scholar
39. 1. Jardetzky O

    (1966) Simple allosteric model for membrane pumps

    Nature 211:969–970.

    https://doi.org/10.1038/211969a0

    • PubMed
    • Google Scholar
40. 1. Javanainen M
    2. Martinez-Seara H
    3. Vattulainen I

    (2017) Excessive aggregation of membrane proteins in the martini model

    PLOS ONE 12:e0187936.

    https://doi.org/10.1371/journal.pone.0187936

    • PubMed
    • Google Scholar
41. 1. Jensen Morten Ø.
    2. Mouritsen OG

    (2004) Lipids do influence protein function—the hydrophobic matching hypothesis revisited

    Biochimica Et Biophysica Acta (BBA) - Biomembranes 1666:205–226.

    https://doi.org/10.1016/j.bbamem.2004.06.009

    • Google Scholar
42. 1. Jorgensen WL
    2. Chandrasekhar J
    3. Madura JD
    4. Impey RW
    5. Klein ML

    (1983) Comparison of simple potential functions for simulating liquid water

    The Journal of Chemical Physics 79:926–935.

    https://doi.org/10.1063/1.445869

    • Google Scholar
43. 1. Kalienkova V
    2. Clerico Mosina V
    3. Bryner L
    4. Oostergetel GT
    5. Dutzler R
    6. Paulino C

    (2019) Stepwise activation mechanism of the scramblase nhTMEM16 revealed by cryo-EM

    eLife 8:e44364.

    https://doi.org/10.7554/eLife.44364

    • PubMed
    • Google Scholar
44. 1. Khelashvili G
    2. Kollmitzer B
    3. Heftberger P
    4. Pabst G
    5. Harries D

    (2013) Calculating the bending Modulus for multicomponent lipid membranes in different thermodynamic phases

    Journal of Chemical Theory and Computation 9:3866–3871.

    https://doi.org/10.1021/ct400492e

    • PubMed
    • Google Scholar
45. 1. Klauda JB
    2. Venable RM
    3. Freites JA
    4. O'Connor JW
    5. Tobias DJ
    6. Mondragon-Ramirez C
    7. Vorobyov I
    8. MacKerell AD
    9. Pastor RW

    (2010) Update of the CHARMM all-atom additive force field for lipids: validation on six lipid types

    The Journal of Physical Chemistry B 114:7830–7843.

    https://doi.org/10.1021/jp101759q

    • PubMed
    • Google Scholar
46. 1. Koch HP
    2. Brown RL
    3. Larsson HP

    (2007) The glutamate-activated anion conductance in excitatory amino acid transporters is gated independently by the individual subunits

    Journal of Neuroscience 27:2943–2947.

    https://doi.org/10.1523/JNEUROSCI.0118-07.2007

    • PubMed
    • Google Scholar
47. 1. Koch HP
    2. Larsson HP

    (2005) Small-scale molecular motions accomplish glutamate uptake in human glutamate transporters

    Journal of Neuroscience 25:1730–1736.

    https://doi.org/10.1523/JNEUROSCI.4138-04.2005

    • PubMed
    • Google Scholar
48. 1. Kumar S
    2. Rosenberg JM
    3. Bouzida D
    4. Swendsen RH
    5. Kollman PA

    (1992) THE weighted histogram analysis method for free-energy calculations on biomolecules. I. the method

    Journal of Computational Chemistry 13:1011–1021.

    https://doi.org/10.1002/jcc.540130812

    • Google Scholar
49. 1. Leary GP
    2. Stone EF
    3. Holley DC
    4. Kavanaugh MP

    (2007) The glutamate and chloride permeation pathways are colocalized in individual neuronal glutamate transporter subunits

    Journal of Neuroscience 27:2938–2942.

    https://doi.org/10.1523/JNEUROSCI.4851-06.2007

    • PubMed
    • Google Scholar
50. 1. Lee AG

    (2004) How lipids affect the activities of integral membrane proteins

    Biochimica Et Biophysica Acta (BBA) - Biomembranes 1666:62–87.

    https://doi.org/10.1016/j.bbamem.2004.05.012

    • Google Scholar
51. 1. Lee KI
    2. Pastor RW
    3. Andersen OS
    4. Im W

    (2013) Assessing smectic liquid-crystal continuum models for elastic bilayer deformations

    Chemistry and Physics of Lipids 169:19–26.

    https://doi.org/10.1016/j.chemphyslip.2013.01.005

    • PubMed
    • Google Scholar
52. 1. Lipfert J
    2. Columbus L
    3. Chu VB
    4. Lesley SA
    5. Doniach S

    (2007) Size and shape of detergent micelles determined by small-angle X-ray scattering

    The Journal of Physical Chemistry B 111:12427–12438.

    https://doi.org/10.1021/jp073016l

    • PubMed
    • Google Scholar
53. 1. MacCallum JL
    2. Bennett WF
    3. Tieleman DP

    (2007) Partitioning of amino acid side chains into lipid bilayers: results from computer simulations and comparison to experiment

    The Journal of General Physiology 129:371–377.

    https://doi.org/10.1085/jgp.200709745

    • PubMed
    • Google Scholar
54. 1. Mancusso R
    2. Gregorio GG
    3. Liu Q
    4. Wang DN

    (2012) Structure and mechanism of a bacterial sodium-dependent dicarboxylate transporter

    Nature 491:622–626.

    https://doi.org/10.1038/nature11542

    • PubMed
    • Google Scholar
55. 1. Marks DS
    2. Colwell LJ
    3. Sheridan R
    4. Hopf TA
    5. Pagnani A
    6. Zecchina R
    7. Sander C

    (2011) Protein 3D structure computed from evolutionary sequence variation

    PLOS ONE 6:e28766.

    https://doi.org/10.1371/journal.pone.0028766

    • PubMed
    • Google Scholar
56. 1. Marrink SJ
    2. de Vries AH
    3. Mark AE

    (2004) Coarse grained model for semiquantitative lipid simulations

    The Journal of Physical Chemistry B 108:750–760.

    https://doi.org/10.1021/jp036508g

    • Google Scholar
57. 1. Marrink SJ
    2. Risselada HJ
    3. Yefimov S
    4. Tieleman DP
    5. de Vries AH

    (2007) The MARTINI force field: coarse grained model for biomolecular simulations

    The Journal of Physical Chemistry B 111:7812–7824.

    https://doi.org/10.1021/jp071097f

    • PubMed
    • Google Scholar
58. 1. Marrink SJ
    2. Corradi V
    3. Souza PCT
    4. Ingólfsson HI
    5. Tieleman DP
    6. Sansom MSP

    (2019) Computational modeling of realistic cell membranes

    Chemical Reviews 119:6184–6226.

    https://doi.org/10.1021/acs.chemrev.8b00460

    • PubMed
    • Google Scholar
59. 1. Marsh D

    (2008) Protein modulation of lipids, and vice-versa, in membranes

    Biochimica Et Biophysica Acta (BBA) - Biomembranes 1778:1545–1575.

    https://doi.org/10.1016/j.bbamem.2008.01.015

    • Google Scholar
60. 1. McIlwain BC
    2. Vandenberg RJ
    3. Ryan RM

    (2015) Transport rates of a glutamate transporter homologue are influenced by the lipid bilayer

    Journal of Biological Chemistry 290:9780–9788.

    https://doi.org/10.1074/jbc.M114.630590

    • PubMed
    • Google Scholar
61. 1. Mondal S
    2. Khelashvili G
    3. Shan J
    4. Andersen OS
    5. Weinstein H

    (2011) Quantitative modeling of membrane deformations by multihelical membrane proteins: application to G-protein coupled receptors

    Biophysical Journal 101:2092–2101.

    https://doi.org/10.1016/j.bpj.2011.09.037

    • PubMed
    • Google Scholar
62. 1. Monticelli L
    2. Kandasamy SK
    3. Periole X
    4. Larson RG
    5. Tieleman DP
    6. Marrink SJ

    (2008) The MARTINI Coarse-Grained force field: extension to proteins

    Journal of Chemical Theory and Computation 4:819–834.

    https://doi.org/10.1021/ct700324x

    • PubMed
    • Google Scholar
63. 1. Moon CP
    2. Fleming KG

    (2011) Side-chain hydrophobicity scale derived from transmembrane protein folding into lipid bilayers

    PNAS 108:10174–10177.

    https://doi.org/10.1073/pnas.1103979108

    • PubMed
    • Google Scholar
64. 1. Mulligan C
    2. Fenollar-Ferrer C
    3. Fitzgerald GA
    4. Vergara-Jaque A
    5. Kaufmann D
    6. Li Y
    7. Forrest LR
    8. Mindell JA

    (2016) The bacterial dicarboxylate transporter VcINDY uses a two-domain elevator-type mechanism

    Nature Structural & Molecular Biology 23:256–263.

    https://doi.org/10.1038/nsmb.3166

    • PubMed
    • Google Scholar
65. 1. Nie R
    2. Stark S
    3. Symersky J
    4. Kaplan RS
    5. Lu M

    (2017) Structure and function of the divalent anion/Na⁺ symporter from Vibrio cholerae and a humanized variant

    Nature Communications 8:15009.

    https://doi.org/10.1038/ncomms15009

    • PubMed
    • Google Scholar
66. 1. Oliver RC
    2. Lipfert J
    3. Fox DA
    4. Lo RH
    5. Doniach S
    6. Columbus L

    (2013) Dependence of micelle size and shape on detergent alkyl chain length and head group

    PLOS ONE 8:e62488.

    https://doi.org/10.1371/journal.pone.0062488

    • PubMed
    • Google Scholar
67. 1. Perez A
    2. Morrone JA
    3. Simmerling C
    4. Dill KA

    (2016) Advances in free-energy-based simulations of protein folding and ligand binding

    Current Opinion in Structural Biology 36:25–31.

    https://doi.org/10.1016/j.sbi.2015.12.002

    • PubMed
    • Google Scholar
68. 1. Phillips JC
    2. Braun R
    3. Wang W
    4. Gumbart J
    5. Tajkhorshid E
    6. Villa E
    7. Chipot C
    8. Skeel RD
    9. Kalé L
    10. Schulten K

    (2005) Scalable molecular dynamics with NAMD

    Journal of Computational Chemistry 26:1781–1802.

    https://doi.org/10.1002/jcc.20289

    • PubMed
    • Google Scholar
69. 1. Phillips R
    2. Ursell T
    3. Wiggins P
    4. Sens P

    (2009) Emerging roles for lipids in shaping membrane-protein function

    Nature 459:379–385.

    https://doi.org/10.1038/nature08147

    • PubMed
    • Google Scholar
70. 1. Reyes N
    2. Ginter C
    3. Boudker O

    (2009) Transport mechanism of a bacterial homologue of glutamate transporters

    Nature 462:880–885.

    https://doi.org/10.1038/nature08616

    • PubMed
    • Google Scholar
71. 1. Ruan Y
    2. Miyagi A
    3. Wang X
    4. Chami M
    5. Boudker O
    6. Scheuring S

    (2017) Direct visualization of glutamate transporter elevator mechanism by high-speed AFM

    PNAS 114:1584–1588.

    https://doi.org/10.1073/pnas.1616413114

    • PubMed
    • Google Scholar
72. 1. Ryan RM
    2. Compton EL
    3. Mindell JA

    (2009) Functional characterization of a ^na+-dependent aspartate transporter from Pyrococcus horikoshii

    The Journal of Biological Chemistry 284:17540–17548.

    https://doi.org/10.1074/jbc.M109.005926

    • PubMed
    • Google Scholar
73. 1. Silverstein N
    2. Crisman TJ
    3. Forrest LR
    4. Kanner BI

    (2013) Cysteine scanning mutagenesis of transmembrane Helix 3 of a brain glutamate transporter reveals two conformationally sensitive positions

    Journal of Biological Chemistry 288:964–973.

    https://doi.org/10.1074/jbc.M112.403576

    • PubMed
    • Google Scholar
74. 1. Sonntag Y
    2. Musgaard M
    3. Olesen C
    4. Schiøtt B
    5. Møller JV
    6. Nissen P
    7. Thøgersen L

    (2011) Mutual adaptation of a membrane protein and its lipid bilayer during conformational changes

    Nature Communications 2:304.

    https://doi.org/10.1038/ncomms1307

    • PubMed
    • Google Scholar
75. 1. Tzingounis AV
    2. Lin CL
    3. Rothstein JD
    4. Kavanaugh MP

    (1998) Arachidonic acid activates a proton current in the rat glutamate transporter EAAT4

    Journal of Biological Chemistry 273:17315–17317.

    https://doi.org/10.1074/jbc.273.28.17315

    • PubMed
    • Google Scholar
76. 1. Verdon G
    2. Boudker O

    (2012) Crystal structure of an asymmetric trimer of a bacterial glutamate transporter homolog

    Nature Structural & Molecular Biology 19:355–357.

    https://doi.org/10.1038/nsmb.2233

    • PubMed
    • Google Scholar
77. 1. Wassenaar TA
    2. Pluhackova K
    3. Böckmann RA
    4. Marrink SJ
    5. Tieleman DP

    (2014) Going backward: a flexible geometric approach to reverse transformation from coarse grained to atomistic models

    Journal of Chemical Theory and Computation 10:676–690.

    https://doi.org/10.1021/ct400617g

    • PubMed
    • Google Scholar
78. 1. Watson MC
    2. Brandt EG
    3. Welch PM
    4. Brown FL

    (2012) Determining biomembrane bending rigidities from simulations of modest size

    Physical Review Letters 109:028102.

    https://doi.org/10.1103/PhysRevLett.109.028102

    • PubMed
    • Google Scholar
79. 1. Yernool D
    2. Boudker O
    3. Jin Y
    4. Gouaux E

    (2004) Structure of a glutamate transporter homologue from Pyrococcus horikoshii

    Nature 431:811–818.

    https://doi.org/10.1038/nature03018

    • PubMed
    • Google Scholar
80. 1. Zerangue N
    2. Arriza JL
    3. Amara SG
    4. Kavanaugh MP

    (1995) Differential modulation of human glutamate transporter subtypes by arachidonic acid

    Journal of Biological Chemistry 270:6433–6435.

    https://doi.org/10.1074/jbc.270.12.6433

    • PubMed
    • Google Scholar

Acceptance summary:

Conventional molecular dynamics simulations and a new enhanced sampling method, presented in another paper (in J Comp Chem), were used to examine the extent of lipid perturbation due to conformational changes of the Na+-aspartate symporter GltPh. This transporter, which follows the 'elevator mechanism', undergoes particularly large conformational changes upon substrate transport; the transition between the outward-facing and inward-facing conformations involves a 15Ang motion of the transport domain with respect to the stationary domain of the protein. Their simulations suggested that this transition radically perturbs the lipid bilayer, which is expected. However, very surprisingly, it leads to a large free energy penalty of about 20 kcal/mol, which of course needs to be balanced by internal components of the free energy. The manuscript addresses a fundamental and very interesting question in membrane biophysics.

Decision letter after peer review:

Thank you for submitting your article "Large-scale state-dependent membrane remodeling by a transporter protein" for consideration by eLife. Your article has been reviewed by three peer reviewers, including Nir Ben-Tal as the Reviewing Editor and Reviewer #1, and the evaluation has been overseen by Richard Aldrich as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Simon Scheuring (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

A new enhanced sampling simulation method, presented in another paper (in review with J Comp Chem), is used to examine the extent of lipid perturbation due to conformational changes of the Na+-aspartate symporter GltPh. This transporter, that typifies the 'elevator mechanism', undergoes particularly large conformational changes upon substrate transport; the transition between the outward-facing and inward-facing conformations involves a 15 Ang motion of the transport domain with respect to the stationary domain of the protein. The simulations suggest that this transition radically perturbs the lipid bilayer, which is expected. However, very surprisingly, it leads to a large free energy penalty of about 20 kcal/mol (which of course needs to be balanced by internal components of the free energy).

Opinion:

The manuscript addresses a fundamental and very interesting question in membrane biophysics. In this respect it is very suitable for publication in eLife. The problem is that it is unclear how trustworthy this large estimate is. The following suggestions may increase the credibility of the manuscript.

Essential revisions:

1) The in-plane distributions of the membrane deformations predicted by the simulations are inaccessible for experimental verifications. Moreover, the character of these deformations (types of strains in the case of 3D description or bending/stretching/chain tilting for a 2D description of the membrane) is not determined.

2) "In a recent breakthrough, we have developed and validated a promising free-energy simulation strategy to address this problem, which we refer to as Multi-Map (Fiorin, 2019)". The Fiorin paper is not in review here so we will not comment about it in detail. However, the draft that is included does not consolidate this bold statement. To our understanding it shows that the new formalism converges to the Helfrich-Canham theory and agrees with measurements at long distances (100Ang and more), as it should. However, we did not see any comparison regarding short distances which are relevant here. This statement should be removed.

3) The authors dismiss the Helfrich-Canham theory and its extensions, arguing that they do not hold for short distances. And yet, it would be insightful for the reader to know what values this approximate model and its extensions give for GltPh. At the very least it would indicate how high the 20 kcal/mol value is compared to existing theory.

4) The derived energies of the membrane deformations, which could be indeed important for understanding the effects of the membrane lipid composition on the protein function, are not supported by any independent estimations and remain, therefore, completely dependent on the parameters of the computational model (forcefield parameters), and other details of the simulations. The authors should admit to it or provide experimental measurements.

5) No analysis is provided either of the effects of membrane tension, which may dramatically change the protein-mediated deformations and their distribution, or of the presence in the membrane of specific lipids such as cholesterol, PIPs, DAGs, etc, whose redistribution to the protein vicinity may moderate the elastic energy. This decreases the general impact of the results.

6) Evidently, the authors are aware of the exceedingly high energy penalty due to lipid perturbation and try to explain it in Discussion. However, the arguments raised are not compelling. Thus, the tune of the manuscript should reflect the fact that all we have here is a computation that may or may not be correct.

7) Introduction, third paragraph: The question is nicely set between (1) "moving polar sidechains on their surface into the bilayer interior and exposing hydrophobic ones to the solvent" and (2) "the morphology of the lipid bilayer could adapt to the conformational state of the protein, and match the amino-acid make-up of the protein surface". The authors obviously come to the conclusion that the latter (2) is the case. However, at a considerable (20kcal/mol) cost. Given that the structure of all conformations are known, the authors should provide an estimate of the energetic cost for the rejected hypothesis (1).

The simulations:

8) The boundary conditions imposed on the membrane fragment are not specified. At the same time, in the absence of membrane tension, which seems to be the case here, the effects of the boundary conditions might be substantial.

9) The method is based on combining coarse-grained and all-atom simulations, which can be tricky. In particular, for the all-atom simulations, the authors take the course-grained system they simulated and convert it to all-atom, while also (naturally) changing force-fields. Then they let the system relax and report that the trends observed are consistent with the course-grain simulations. It is theoretically possible that transitioning the system from the course-grained to the all-atom parameters is not that trivial, and maybe the fact that they start from an already equilibrated state introduces some bias. Could it be examined somehow?

10) It is unfortunate that DPPC (transition temperature 41C) was chosen for many simulations. Why? Also, simulations using POPC (transition temperature -2C) are shown. The fact that the results are quasi identical seems rather worrying than comforting. Why these choices? Why not using a mixture, it is not unthinkable that in a mixture the protein would recruit specific lipids on these interfaces between transporter and scaffold domains. Please comment. Or, better, examine in simulations.

Presentation:

11) Introduction: "These perturbations develop to accommodate the amino-acid composition and specific structural features of the protein surface.": This deserves a reference to the Piezo channel, the only protein that displays clear structural features that force the membrane into bending, and for which a theory and experiments about how membrane bending (and flattening / the physics of the membrane) is exploited to gate the channel was presented.

12) The general reader might not know what the second-rank order parameter of the lipid alkyl chains is.

13) Figure 2, Figure 3 and associated, in the context of "Transport domains bend the membrane, while scaffold domains anchor it": While we understand the measurements, the results and the importance of all these results, the presentation is unclear with respect to the setting of the 0-level. From the captions "2(A) The deflection is quantified by calculating the mean value of the Z coordinate of the bilayer across the X-Y plane." and "(A) Deflection of the membrane mid-plane relative to a flat surface". For example, the image in 3A), all the membrane has negative deformation. Wouldn't that correspond to creation of a net elastic and potential energy? Shouldn't the bilayer as a whole still go towards flat and as a result one has negative deformation next to the transporter domains and positive deformation next to the scaffold domain. This does not change anything to the results in terms of local deformation, right? Is the bilayer held in place at the periphery of the simulation box?

14) Further on this: Why is the scaffold domain considered the 'anchor'? Why does the membrane next to the transporter domains (that are in the inward (down) orientation) move down, rather than the scaffold domain moving upwards? Overall – after sufficiently long relaxation of the all-inward structure – shouldn't the overall level of the membrane remain 0 and the membrane next to the transporter domains shift downwards, but the membrane next to the scaffold domains shift upwards? Especially, in light of Figure 3C, which seems to show that there are many more key amino acids in the transporter than in the scaffold domain, one would expect that the transport domain dominates the relative motion. It is unclear how in panels like Figure 2A left, 2A right, Figure 3A, all membrane deformation values can be positive or negative? Is this the reason why such huge energy penalties result from the analysis?

15) Discussion section: The authors find a large energetic penalty for the inward facing state due to membrane deformation. They note that the smFRET studies displayed almost zero energy difference between the states. In contrast the HS-AFM study revealed that the inward facing state indeed was the high energy state (like here). The authors mention that the HS-AFM study was performed in rather densely packed membrane where no such long range lipid relaxations are possible as in the study here, yet it is a flat membrane. In this context, it is also noticeable that the smFRET studies are performed either in detergent or on transporters in small vesicles that are tethered in outside-out configuration only – which goes against the bowl-shaped structure of GltPh – of which the authors see the preference for membrane bending in Figure 1 (side view), which might favor the adoption of the inward-facing state in the smFRET experiments.

Opinion:

The manuscript addresses a fundamental and very interesting question in membrane biophysics. In this respect it is very suitable for publication in eLife. The problem is that it is unclear how trustworthy this large estimate is. The following suggestions may increase the credibility of the manuscript.

We appreciate the editor’s remarks in regard to the broad significance of the subject of our study, and are thankful to all reviewers for their suggestions and constructive criticisms. We trust that the new data and clarifications provided in the revised manuscript will address the reviewers’ concerns.

For the record, however, we believe it is important to clarify that our analysis of the extent of the membrane perturbations induced by Glt_Ph did not involve the use of a new enhanced sampling simulation method, as the Summary states. This component of the study, which has been substantially modified and expanded in the revision, entailed standard simulation methods, in different conditions and using both coarse-grained and all-atom forcefields – the latter being particularly challenging due to the very large size of the system. The new enhanced-sampling methodology mentioned in the Summary, now published in J Comp Chem, was used only to estimate the energetic cost of this perturbation. This second component has also been expanded in the revision.

Essential revisions:

1) The in-plane distributions of the membrane deformations predicted by the simulations are inaccessible for experimental verifications. Moreover, the character of these deformations (types of strains in the case of 3D description or bending/stretching/chain tilting for a 2D description of the membrane) is not determined.

We understand that by in-plane distributions the reviewer refers to the two-dimensional representations of the deflection (depression/elevation) of the membrane mid-plane. We believe this kind of representation is important as it quantifies what is observed in the simulations, which in turn permits comparison of different conditions and conformational states. In order to facilitate a direct comparison with experiment, however, we also provided three-dimensional distributions (Figure 2, Figure 5, Video 1 and Video 2, Figure 8) of the lipid density around the protein – for both Glt_Ph and VcINDY. We believe that in time it will be possible to directly contrast this data with three-dimensional density maps derived from single-particle cryo-EM imaging of these transporters reconstituted in lipid nanodiscs – as has been possible for other membrane proteins.

With regard to the character of the deformation induced by all-inward Glt_Ph, additional figures have been provided to underscore that the impact of the protein is primarily a change in the curvature of both membrane leaflets, as originally stated. Specifically, Figure 4—figure supplement 1 shows that the thickness of the membrane is, relatively speaking, influenced only minimally, both in terms of magnitude and range. Figure 4—figure supplement 2 reiterates the observation made in the original version that a significant change in alkyl-chain tilt (as quantified by the second-rank order parameter) is observed only in the inner leaflet, and only at the interface between the membrane and the scaffold. By contrast, the alkyl-chain tilt along the periphery of the transport domain, where the membrane is most significantly depressed, is essentially bulk-like. This data correlates with what can be inferred from the graphical 3D representations of lipid density in Figure 2, Figure 5, and Video 1and Video 2.

2) "In a recent breakthrough, we have developed and validated a promising free-energy simulation strategy to address this problem, which we refer to as Multi-Map (Fiorin, 2019)". The Fiorin paper is not in review here so we will not comment about it in detail. However, the draft that is included does not consolidate this bold statement. To our understanding it shows that the new formalism converges to the Helfrich-Canham theory and agrees with measurements at long distances (100Ang and more), as it should. However, we did not see any comparison regarding short distances which are relevant here. This statement should be removed.

We have no objection to rephrasing this statement to make it strictly factual: “In a recent development, we have reported a novel free-energy simulation strategy to address this problem, which we refer to as Multi-Map (Fiorin, 2019).”

We must note, however, that in this context the term “validation” refers to the fact we established the correctness of the enhanced-sampling algorithm, i.e., that it reproduces the expected thermodynamic ensemble, the target membrane morphology, and the mean forces along the biased collective variable from which the free energy is derived.

One way in which the proposed method was validated (as defined above) was by comparing it with unbiased-sampling simulations, for a problem that is fully quantifiable through both approaches (the free-energy of hydration of a hydrophobic cavity). In addition, the draft of Fiorin et al., enclosed with the original submission of the current article did include a comparison of calculated and measured bending moduli for small bilayers of different lipid types, in the presence and absence of cholesterol. Because experiments and computations probe very different length scales, what was evaluated is the relative change in this bending modulus upon addition of cholesterol. The results of this analysis were positive for the lipid types for which the experimental data is uncontroversial (POPC and DMPC). Thus, at least in regard to the effect of cholesterol on membrane rigidity of small lipid bilayers, the computational Multi-Map method performs no worse than existing experimental techniques.

3) The authors dismiss the Helfrich-Canham theory and its extensions, arguing that they do not hold for short distances. And yet, it would be insightful for the reader to know what values this approximate model and its extensions give for GltPh. At the very least it would indicate how high the 20 kcal/mol value is compared to existing theory.

We have not argued, here or elsewhere, that extensions to the Helfrich-Canham theory do not hold for small wave-length deformations, nor have we dismissed them. To the contrary, in Fiorin et al., we discuss how these extensions address the known limitations of the Helfrich-Canham model (described by us and others before us), in some cases quite elegantly. We have however argued that an alternative, direct PMF approach is advantageous as it does not entail important ad hoc assumptions (e.g. a bending modulus, a homogenous membrane, etc.) and avoids ill-defined methodological choices (as discussed below). In any case, it is beyond our specific expertise to carry out a systematic comparison of the multiple variations and extensions of the Helfrich-Canham functional that have been formulated over the years. As per eLife’s editorial policy, we believe this request is outside the scope of this revision.

Using our own simulation data, however, it is relatively straightforward to illustrate why an alternative to the Helfrich-Canham approach is advantageous. In the newly provided Figure 7—figure supplement 1A, we compare free-energy curves calculated with the direct PMF method (Figure 7) with 3 alternative results deduced from the Helfrich-Canham equation, for the same set of molecular configurations (i.e. for each of the umbrella-sampling simulations carried out to calculate the PMF). The two inputs for the calculation of the Helfrich-Canham energy are the assumed bending modulus (here we use k_c = 18 kcal/mol, standard for POPC) and the membrane curvature distribution across the bilayer plane, c(x,y). We evaluate this curvature distribution from analysis of the average mid-plane in from each umbrella-sampling simulation (18 μs per window) – see Materials and methods section. The Helfrich-Canham energy is then obtained by integrating the energy density [0.5 k_cc² (x,y)] over the area of the membrane. The three calculations in Figure 7—figure supplement 1A use the same value of the bending modulus, but differ in the level of resolution used to probe c(x,y). Linear regressions of the data in each case show that the Helfrich energy can be much larger, similar or somewhat smaller depending on the resolution with which the average local curvature of the membrane is evaluated. Needless to say, a smaller or larger value of the bending modulus might improve this correlation for a given resolution, but make it worse for another. Nevertheless, it seems clear that there is no reasonable choice of either k_c or the curvature-map resolution that would lead to Helfrich energies that are, say, 2-fold smaller than those we calculate with the Multi-Map method. The opposite is however true, i.e. Helfrich energies that are significantly larger than those deduced with a direct PMF method can be easily envisaged. In sum, going back to reviewers’ point, the 20 kcal/mol value reported in the manuscript is no larger than what would be reasonably inferred from non-microscopic theories of membrane energetics.

As illustrated in Figure 7—figure supplement 1B, a rigorous evaluation of the local membrane curvature c(x,y) is itself challenging and prone to large statistical errors despite extensive simulation time (18 μs per point, split in 1-μs fragments). Thus, when comparing RMS-curvatures across the bilayer obtained at different resolutions, the expected correlation is degraded at moderate to low curvature values. These ambiguities involved in extracting membrane curvature from molecular configurations further highlight the value of direct PMF calculations, of the kind now made possible by the Multi-Map method.

4) The derived energies of the membrane deformations, which could be indeed important for understanding the effects of the membrane lipid composition on the protein function, are not supported by any independent estimations and remain, therefore, completely dependent on the parameters of the computational model (forcefield parameters), and other details of the simulations. The authors should admit to it or provide experimental measurements.

The reviewer is right, evidently, that the results from any computer simulation or theoretical analysis are hypothetical until proven by experiment. This point has been clarified in the Discussion section.

5) No analysis is provided either of the effects of membrane tension, which may dramatically change the protein-mediated deformations and their distribution, or of the presence in the membrane of specific lipids such as cholesterol, PIPs, DAGs, etc, whose redistribution to the protein vicinity may moderate the elastic energy. This decreases the general impact of the results.

We appreciate the opportunity to discuss in detail the effect of membrane tension and membrane composition, as this is a recurrent question. As shown in the newly provided Figure 3, tensions of increasing magnitude do not dramatically change the nature of the deformations induced by all-inward Glt_Ph. At the structural level, the effect of increased membrane tension is to gradually reduce the magnitude of the depression induced by the transport domains. Nevertheless, this perturbation is discernable even when the applied membrane tension is as high as 10 mN/m. It is important to note, however, that the energetic cost of the membrane deformation induced by Glt_Ph increases under membrane tension, as we show in the revised version of Figure 7B. Thus, while the amplitude of the perturbation caused by the transporter would be somewhat diminished if the membrane was under tension, the resulting energetic cost might be comparable to (or greater than) a condition where the membrane is not under tension.

With regard to the effect of lipid composition, we now compare 4 different conditions, namely POPC, POPE and 2:1 POPE-POPG at 298 K and DPPC at 323 K. Albeit simplified, we would argue these model membranes are not entirely unlike the conditions that have been probed in published liposome reconstitutions. As shown in Figure 4, the nature of the deformations induced by all-inward Glt_Ph is not dramatically changed by different lipid compositions either.

The notion that the functional regulation of membrane proteins by specific lipids (PIP2, cholesterol, etc) might be in part due to changes in the energetics of membrane plasticity is indeed an intriguing possibility, which we are currently pursuing for other systems where systematic experimental analyses have been conducted. We are unaware of such data for Glt_Ph and thus we believe that addressing this question is beyond the scope of this study.

6) Evidently, the authors are aware of the exceedingly high energy penalty due to lipid perturbation and try to explain it in Discussion. However, the arguments raised are not compelling. Thus, the tune of the manuscript should reflect the fact that all we have here is a computation that may or may not be correct.

See our response to points 4 and point 7.

7) Introduction, third paragraph: The question is nicely set between (1) "moving polar sidechains on their surface into the bilayer interior and exposing hydrophobic ones to the solvent" and (2) "the morphology of the lipid bilayer could adapt to the conformational state of the protein, and match the amino-acid make-up of the protein surface". The authors obviously come to the conclusion that the latter (2) is the case. However, at a considerable (20kcal/mol) cost. Given that the structure of all conformations are known, the authors should provide an estimate of the energetic cost for the rejected hypothesis (1).

We thank the reviewers for this suggestion, as this analysis enables us to quantify our claim that the cost of membrane bending is considerable smaller than the cost of the alternative (rejected) hypothesis.

This analysis is summarized in the newly provided Figure 6, for an equilibrated simulation snapshot of all-inward Glt_Ph, and in Figure 6—figure supplement 1 for the crystal structure. The question that we pose is: what would be the hypothetical free-energy gain/cost of preserving a flat membrane when the transporter is in the all-inward state? To estimate this cost, we considered the two membrane configurations described in the figure – one deformed, as in our simulation, and a version thereof that has the same thickness but flat. (By ‘membrane’, we refer to the solvent-excluded region of the bilayer.) We then calculated the solvent-accessible surface area (SASA) for all protein residues in either case. The difference, i.e. ΔSASA, was then normalized by the SASA of each residue-type when fully solvent-exposed in a Gly-X-Gly peptide. The resulting ‘dehydration factor’ was then multiplied by the transfer free-energy of that residue type into the membrane, according to two independent hydrophobicity scales: one based on experimental measurements of the stability of WT and mutagenized OMPLA, published by Karen Fleming; and another published by Peter Tieleman, based on MD simulations.

Taken together, these results underscore how strikingly unfavorable it would be for the membrane to remain flat when the transport domains adopt the inward facing state. Small energy penalties from a large number of residues add up to very large totals – specifically about 30 kcal/mol for the Fleming scale and about 50 kcal/mol for the Tieleman scale. As originally stated, these penalties result from either exposing polar sidechains to the bilayer interior (R105, K230), or from exposing hydrophobic sidechains to the solvent; indeed, more than a dozen hydrophobic residues disfavor the flat membrane by 1 k_BT or more, each. As one might expect, the polar residues that contribute the most are on the extracellular side of the transport domains, while the hydrophobic groups are on the intracellular side. A few hydrophobic residues on the extracellular side favor the flat membrane, but their total contribution is comparatively much smaller than the energetic penalty that opposes it.

It should be noted that, by definition, this ‘transfer free-energy’ examines the energetics of hydration/dehydration, neglecting the stabilizing effect of the hydrogen-bonding interactions between polar sidechains and the lipid headgroups described in Figure 5; many of these lipid contacts would be unfeasible if the membrane was flat, and would be replaced by water, penalizing this hypothetical state further.

In summary, the adaptation of the membrane to the conformational state of the transporter, while clearly costly, is nevertheless much more favorable than the alternative.

The simulations:

8) The boundary conditions imposed on the membrane fragment are not specified. At the same time, in the absence of membrane tension, which seems to be the case here, the effects of the boundary conditions might be substantial.

The boundary conditions are now specified in the Materials and methods section, i.e., semi-isotropic pressure coupling and periodic boundary conditions. As mentioned, simulations are now presented with and without applied membrane tension. See Figure 3 and our response to point 5.

9) The method is based on combining coarse-grained and all-atom simulations, which can be tricky. In particular, for the all-atom simulations, the authors take the course-grained system they simulated and convert it to all-atom, while also (naturally) changing force-fields. Then they let the system relax and report that the trends observed are consistent with the course-grain simulations. It is theoretically possible that transitioning the system from the course-grained to the all-atom parameters is not that trivial, and maybe the fact that they start from an already equilibrated state introduces some bias. Could it be examined somehow?

This possibility was examined in the original submission, specifically in Figure 5—figure supplement 1AB (Formerly Figure 3—figure supplement 1). In this calculation, a perturbation of magnitude comparable to that induced by Glt_Ph is created in an all-atom lipid bilayer and allowed to relax over time. The perturbation dissipates well within 100 ns, which is shorter than the length of the Glt_Ph all-atom simulations.

In the revised version of the article, we further examine this question by quantifying the degree to which lipid molecules exchange between the first solvation shells around the protein and the rest of the membrane in the course of our 3 all-atom simulations of all-inward Glt_Ph. As shown in the newly-provided Figure 5—figure supplement 1C, we find that, by the end of the simulation, about 30% of the lipids that are initially in these first solvation shells are replaced by other lipid molecules from further away from the protein – with the overall lipid number in each shell approximately constant. These exchanges would permit a change in membrane morphology, if the such change was favorable. The fact that the membrane remains deformed, despite the results from these two control calculations, suggests that this deformation is the preferred state of the molecular system, and not an artifact of the coarse-grained forcefield, preserved at the all-atom level.

10) It is unfortunate that DPPC (transition temperature 41C) was chosen for many simulations. Why? Also, simulations using POPC (transition temperature -2C) are shown. The fact that the results are quasi identical seems rather worrying than comforting. Why these choices? Why not using a mixture, it is not unthinkable that in a mixture the protein would recruit specific lipids on these interfaces between transporter and scaffold domains. Please comment. Or, better, examine in simulations.

Although we did not observe a phase transition in any of our original simulations, we recognize that the use of DPPC near the (coarse-grained) transition temperature is a concern. To address this concern, we have repeated all the coarse-grained simulations using different conditions. Specifically, we repeated the triplicated simulations of all-outward, one-inward, two-inward, and all-inward Glt_Ph using POPC at 298 K, and show the results in Figure 2. We also carried out triplicated simulations of all-inward Glt_Ph in POPE and in a 2:1 POPE:POPG mixture at 298 K, as well as in DPPC at 323 K, i.e., well above the transition temperature. The results of these simulations are shown in Figure 4. The triplicated simulations of VcINDY, in all-outward and all-inward conformations were also repeated using POPC at 298 K, and the results shown in Figure 8. Taken together, these new simulation data convey that the effects we describe are not critically dependent on variations in the lipid composition of the membrane. Please see also our response to point 5.

Presentation:

11) Introduction: "These perturbations develop to accommodate the amino-acid composition and specific structural features of the protein surface.": This deserves a reference to the Piezo channel, the only protein that displays clear structural features that force the membrane into bending, and for which a theory and experiments about how membrane bending (and flattening / the physics of the membrane) is exploited to gate the channel was presented.

A reference has been added.

12) The general reader might not know what the second-rank order parameter of the lipid alkyl chains is.

An explanation has been added (see caption of Figure 4—figure supplement 2).

13) Figure 2, Figure 3 and associated, in the context of "Transport domains bend the membrane, while scaffold domains anchor it": While we understand the measurements, the results and the importance of all these results, the presentation is unclear with respect to the setting of the 0-level. From the captions "2(A) The deflection is quantified by calculating the mean value of the Z coordinate of the bilayer across the X-Y plane." and "(A) Deflection of the membrane mid-plane relative to a flat surface". For example, the image in 3a), all the membrane has negative deformation. Wouldn't that correspond to creation of a net elastic and potential energy? Shouldn't the bilayer as a whole still go towards flat and as a result one has negative deformation next to the transporter domains and positive deformation next to the scaffold domain. This does not change anything to the results in terms of local deformation, right? Is the bilayer held in place at the periphery of the simulation box?

We have clarified this point in the revised version.

The Z-coordinate of the membrane mid-plane is obtained from an average over all snapshots collected in a given simulation. The protein starts perfectly centered in the membrane, but in the course of a simulation it tumbles and rotates freely (and the membrane, which is initially flat, adapts accordingly). To analyze the simulation data, therefore, all snapshots must be first transformed so that the center-of-mass and orientation of the Glt_Ph trimer are consistent across the data set. To do so, all atomic coordinates (protein, membrane, solvent) are roto-translated equally, in a manner that results in an optimal (least-squares) fitting of the protein backbone. Note that this transformation does not alter in any way the relative coordinates of the atoms in the molecular system. Through this procedure, the membrane shape is defined with good precision in the vicinity of the protein. Conversely, at the edges of the simulated membrane patch, this procedure leads to large “apparent” fluctuations and an ill-defined average. Nevertheless, with extensive sampling it becomes possible to define a 0-level with confidence, i.e. a broad region far from the protein where the membrane is flat on average, and where the “apparent” fluctuations induced by the analysis process are small. In our case, that 0-level is about ~200 Å from the protein center.

The original version of Figure 5A (formerly 3A) was a close-up view of the membrane close to the protein – hence this 0-level wasn’t clear. To avoid confusion, in the new version we zoom out and present the data exactly as for the coarse-grained simulations in Figure 2. As mentioned, the 0-level becomes clear at about 200 Å from the protein center.

The membrane is not held in place, anywhere along the periphery of the system, or elsewhere. The perturbation caused by the transporter does however decay naturally away from the protein surface and so the membrane does indeed become flat, as it should. That is the region where the 0-level is set.

14) Further on this: Why is the scaffold domain considered the 'anchor'? Why does the membrane next to the transporter domains (that are in the inward (down) orientation) move down, rather than the scaffold domain moving upwards? Overall – after sufficiently long relaxation of the all-inward structure – shouldn't the overall level of the membrane remain 0 and the membrane next to the transporter domains shift downwards, but the membrane next to the scaffold domains shift upwards? Especially, in light of Figure 3C, which seems to show that there are many more key amino acids in the transporter than in the scaffold domain, one would expect that the transport domain dominates the relative motion. It is unclear how in panels like Figure 2A left, 2A right, Figure 3A, all membrane deformation values can be positive or negative? Is this the reason why such huge energy penalties result from the analysis?

As explained above, the 0-level can be defined reliably at about 200 Å from the center of the protein, which is about 150 Å from its surface. Relative to this 0-level, the change in the membrane mid-plane is much smaller at the interface with the scaffold domain than at the interface with the transport domains (see 1D profiles in Figure 2D, formerly Figure 3C). The relative change in this mid-plane when comparing all-outward to all-inward is also much larger for the transport domains. Therefore, it seems logical to think of the scaffold domain as the ‘anchor’. As noted, however, the scaffold isn’t entirely stationary. Comparing the all-outward with the all-inward state, we detect a small net displacement inwards and the transport domains move inwards too. However, we certainly do not observe an upwards displacement, relative to the 0-level. [Of course, one could arbitrarily redefine the 0-level so that the membrane mid-plane appears unperturbed at the transport domain, but that 0-level would not be consistent with the lipid bilayer further from the protein.]

Ultimately, the geometry of the membrane and the position of the protein therein is determined by a complex free-energy function that, as we argue, includes the conformational preference of the membrane itself and not only a protein-centric perspective. The geometry described by the reviewer for all-inward Glt_Ph seems entirely plausible a priori (i.e., the membrane is elevated near the scaffold domain and approximately balances out the opposite perturbation caused the transport domains), but in our view is not more intuitive than other alternatives. Arguably, the purpose of molecular simulations such as those presented in our article is precisely to investigate complex, multi-component problems such as this – notwithstanding the caveat in point 4.

15) Discussion section: The authors find a large energetic penalty for the inward facing state due to membrane deformation. They note that the smFRET studies displayed almost zero energy difference between the states. In contrast the HS-AFM study revealed that the inward facing state indeed was the high energy state (like here). The authors mention that the HS-AFM study was performed in rather densely packed membrane where no such long range lipid relaxations are possible as in the study here, yet it is a flat membrane. In this context, it is also noticeable that the smFRET studies are performed either in detergent or on transporters in small vesicles that are tethered in outside-out configuration only – which goes against the bowl-shaped structure of GltPh – of which the authors see the preference for membrane bending in Figure 1 (side view), which might favor the adoption of the inward-facing state in the smFRET experiments.

We thank the reviewers for pointing out this interesting possibility, which has been noted in the Discussion section.

Author details

1. Wenchang Zhou

   Theoretical Molecular Biophysics Laboratory, National Heart, Lung and Blood Institute, National Institutes of Health, Bethesda, United States

   Contribution

   Formal analysis, Investigation, Visualization

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-0397-1032

2. Giacomo Fiorin

   Theoretical Molecular Biophysics Laboratory, National Heart, Lung and Blood Institute, National Institutes of Health, Bethesda, United States

   Contribution

   Formal analysis, Investigation, Visualization

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-8793-8645

3. Claudio Anselmi

   Theoretical Molecular Biophysics Laboratory, National Heart, Lung and Blood Institute, National Institutes of Health, Bethesda, United States

   Present address

   Children’s National Hospital, Washington, United States

   Contribution

   Formal analysis, Investigation, Visualization

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-3017-5085

4. Hossein Ali Karimi-Varzaneh

   Computational Structural Biology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, United States

   Present address

   Continental Reifen Deutschland GmbH, Hannover, Germany

   Contribution

   Formal analysis, Investigation, Visualization

   Competing interests

   No competing interests declared

5. Horacio Poblete

   1. Theoretical Molecular Biophysics Laboratory, National Heart, Lung and Blood Institute, National Institutes of Health, Bethesda, United States
   2. Computational Structural Biology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, United States

   Present address

   Centro de Bioinformática y Simulación Molecular, University of Talca, Talca, Chile

   Contribution

   Formal analysis, Investigation, Visualization

   Competing interests

   No competing interests declared

6. Lucy R Forrest

   Computational Structural Biology Section, National Institute of Neurological Disorders and Stroke, National Institutes of Health, Bethesda, United States

   Contribution

   Conceptualization, Supervision, Investigation, Visualization

   For correspondence

   lucy.forrest@nih.gov

   Competing interests

   Reviewing editor, eLife

   "This ORCID iD identifies the author of this article:" 0000-0003-1855-7985

7. José D Faraldo-Gómez

   Theoretical Molecular Biophysics Laboratory, National Heart, Lung and Blood Institute, National Institutes of Health, Bethesda, United States

   Contribution

   Conceptualization, Supervision, Investigation, Visualization

   For correspondence

   jose.faraldo@nih.gov

   Competing interests

   Senior editor, eLife

   "This ORCID iD identifies the author of this article:" 0000-0001-7224-7676

• 3,043

  Page views

• 430

  Downloads

• 24

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.