Bicelles and Other Membrane Mimics: Comparison ... - ACS Publications

Article pubs.acs.org/JPCB

Bicelles and Other Membrane Mimics: Comparison of Structure, Properties, and Dynamics from MD Simulations Mikkel Vestergaard,† Johan F. Kraft,† Thomas Vosegaard,‡ Lea Thøgersen,§,∥ and Birgit Schiøtt*,† †

Center for Insoluble Protein Structures (inSPIN), Interdisciplinary Nanoscience Center (iNANO), and Department of Chemistry, Aarhus University, Langelandsgade 140, DK-8000 Aarhus C, Denmark ‡ Danish Center for Ultrahigh-Field NMR Spectroscopy and Center for Insoluble Protein Structures (inSPIN), Interdisciplinary Nanoscience Center (iNANO) and Department of Chemistry, Aarhus University, Gustav Wieds Vej 14, DK-8000 Aarhus C, Denmark § Center for Membrane Pumps in Cells and Disease (PUMPKIN), Bioinformatics Research Centre, Aarhus University, C.F. Møllers Alle 8, DK-8000 Aarhus C, Denmark S Supporting Information *

ABSTRACT: The increased interest in studying membrane proteins has led to the development of new membrane mimics such as bicelles and nanodiscs. However, only limited knowledge is available of how these membrane mimics are affected by embedded proteins and how well they mimic a lipid bilayer. Herein, we present molecular dynamics simulations to elucidate structural and dynamic properties of small bicelles and compare them to a large alignable bicelle, a small nanodisc, and a lipid bilayer. Properties such as lipid packing and properties related to embedding both an α-helical peptide and a transmembrane protein are investigated. The small bicelles are found to be very dynamic and mainly assume a prolate shape substantiating that small bicelles cannot be regarded as well-defined disclike structures. However, addition of a peptide results in an increased tendency to form disc-shaped bicelles. The small bicelles and the nanodiscs show increased peptide solvation and difference in peptide orientation compared to embedding in a bilayer. The large bicelle imitated a bilayer well with respect to both curvature and peptide solvation, although peripheral binding of short tailed lipids to the embedded proteins is observed, which could hinder ligand binding or multimer formation.

■

INTRODUCTION Membrane-embedded proteins account for up to 30% of the encoded proteins in the human genome.1 This makes it of utmost importance to be able to study this type of proteins, since a large variety of biological processes are controlled by membrane proteins such as signal transduction and transport of ions and nutrients across cell membranes. These proteins are furthermore the molecular target for more than 50% of the pharmaceutical drugs on the market.2 However, since it is difficult to characterize the individual proteins in detail when found in the native environment of the cell membrane, membrane mimicking models are needed for detailed structural and physicochemical examinations.3 Lipid bilayers consisting of one or a few types of lipids can be used for such studies either in the form of liposomes or mechanically stacked bilayers.3 However, for experimental techniques such as NMR, small-angle X-ray scattering (SAXS), and X-ray crystallography, which are often preferred for structural characterization of membrane proteins, such membrane mimics have various limitations. These include lowresolution data, low concentration effects, and limited applicable solvents.3,4 Another membrane mimic often used in structural studies of membrane proteins is a surfactant micelle, which is a spherical aggregate formed by small © 2015 American Chemical Society

amphipathic molecules [e.g., 1,2-dihexaoyl-sn-glycero-3-phosphocholine (DHPC)] (Figure 1a). However, some proteins have been found to change to a non-native conformation in micelles with subsequent decrease or complete loss of function.5−8 This effect is attributed to the larger surface curvature of the micelles, as implied in Figure 1a.3,9 Furthermore, protein multimers, which are important for the function of many proteins,4 may not form in micelles due to the small size of the micelles compared to the protein oligomers. To overcome such limitations, other membrane mimics have been developed. Two of such mimics have gained special popularity, due to their superior properties over micelles in, for example, NMR and biophysical experiments, namely bicelles10 and nanodiscs11 (see Figure 1). Bicelles are formed by mixing two different lipids: one with long hydrophobic fatty acid chains [e.g., 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC)], which forms bilayers by itself, and one with short hydrophobic fatty acid chains (e.g., DHPC), which usually forms micelles. Such lipid mixtures were initially proposed to form disc-shaped objects with a bilayer part consisting of DMPC and DHPC Received: August 31, 2015 Revised: November 25, 2015 Published: November 26, 2015 15831

DOI: 10.1021/acs.jpcb.5b08463 J. Phys. Chem. B 2015, 119, 15831−15843

Article

The Journal of Physical Chemistry B

Figure 1. Idealized structures of the membrane mimics. (a) A DHPC micelle, (b) a small bicelle formed from twice the amount of DHPC as DMPC (q = 0.5), (c) a nanodisc consisting of a bilayer part comprised of DMPC molecules encircled by two α-helical membrane scaffold proteins, and (d) a large disc-shaped bicelle (q = 3.2). DHPC is shown in red, DMPC in blue, and protein in green. In the top row of the figure, the molecular models are depicted as snapshots from the simulations in their MARTINI20 coarse-grain description, while schematic illustrations of the models are shown below. The scale bar refers to the MARTINI snapshots. (e) Molecular structure as well as an illustration of the coarse grain mapping of DHPC (left) and DMPC (right).

details. This makes it an ideal tool for investigating different properties like hydrophobic mismatch,21,22 curvature of membrane mimics,22,23 and effects of proteins/peptides on segregation of lipids.24,25 A few studies of bicelles and nanodiscs using either MD26−33 or a combination of MD and Monte Carlo simulations32,34−36 have been published; however, the focus of these have mostly been on the morphology of large alignable bicelles and the shape of the nanodisc; none have studied the morphology of the small q = 0.5 bicelles and how well they compare to other membrane mimics. In this account, we use MD simulations to investigate the morphology of small bicelles, applicable for liquid state NMR, for example, with and without peptides embedded, hereby obtaining an improved characterization of their shape and size. The membrane mimicking properties of these small bicelles are then compared to a bilayer, a nanodisc, and a large bicelle. Finally, interesting consequences of embedding a peptide (KALP21) and a protein (rhodopsin) in the membrane mimics are evaluated with respect to the structure and dynamics of both peptides/proteins and lipids. We find that the shape of the small bicelle is highly affected by embedding the KALP21 peptide. Furthermore, different limitations become apparent for the individual membrane mimics, indicating that the best choice of membrane mimic for a given study may differ.

positioned at the rim to stabilize the bilayer disc (Figure 1, panels b and d).12 It has since then been revealed that they adapt a wide variety of topologies depending on temperature, hydration, salt concentration, and the ratio of long-to-shorttailed lipids (referred to as the q-value).13−15 The dimensions of the lipid aggregates are for example known to increase with increasing DMPC:DHPC ratio, although their shape is highly debated.12,13,16 Bicelles are often used in NMR studies because they are easier to handle than membrane bilayers, more stable, and give sharp peaks in liquid-state NMR studies of membrane proteins when mixtures with low q values are used.17 In contrast, bicelles with a higher q value align in a magnetic field, making them useful for solid-state NMR.10 Bicelles do however also have some limitations, the most important being the requirement for a high lipid concentration to overcome the critical micelle concentration of the short-chained lipid.4 This consequently limits their use in certain experimental techniques such as SAXS where a low concentration of the compound studied is essential. The nanodisc mimicking model does not suffer from issues related to the critical micelle concentration.11 Nanodiscs are disc-shaped and constitute a lipid bilayer encapsulated by (usually two) long α-helical proteins, referred to as membrane scaffold proteins (MSPs), stabilizing the rim.18 By varying the scaffold protein, nanodiscs have been prepared with a large range of dimensions, with diameters between 8.6 and 15.2 nm.19 Multiple reviews have recently been published comparing different membrane mimics and their individual limitations.3,4,17 It is however clear from these accounts that the choice of membrane mimic for a given study is mainly guided by the conditions under which they can be used experimentally and some general assumptions about their physicochemical properties. A detailed understanding of how the structure and dynamics of an embedded protein or peptide is affected by the membrane mimic and vice versa is however limited, as is quantitative knowledge of how well they each compare to a lipid bilayer by various physical-chemical properties, such as morphology and curvature. Molecular dynamics (MD) simulations provide an attractive tool for studying the different membrane mimics on an Ångstrøm to nanometer length scale making it possible to examine the molecular interactions and dynamic properties in

■

COMPUTATIONAL METHODS Simulations of a DMPC bilayer along with two sizes of bicelles, consisting of mixtures of DMPC and DHPC (with q values of 0.5 and 3.2, respectively), and a nanodisc were performed to assess how well the three membrane mimics imitate a DMPC lipid bilayer. Additionally, simulations including either an αhelical transmembrane peptide, KALP21, or a large transmembrane protein, a rhodopsin dimer, were performed to investigate how the membrane mimic may affect the properties of the embedded protein or peptide and vice versa. Rhodopsin is a G protein coupled receptor (GPCR) and was chosen as a typical large membrane protein since it has previously been extensively studied experimentally in a large range of membrane mimics such as in micelles,37 bicelles,37 and nanodiscs.38 KALP21 is a well-characterized transmembrane α-helical peptide, with a length that fits the hydrophobic thickness of a DMPC bilayer.39 This peptide was chosen due to its stable secondary structure. All simulation setups are listed in Table 1. 15832


Article

The Journal of Physical Chemistry B Table 1. List of Simulations Used in This Study system bilayer simulations build bilayer equilibrated bilayer, KALP21 bilayer self-assembly, rhodopsin bicelle simulations random, q = 0.5 discs, r = 3 nm, q = 0.5 discs, r = 3 nm, q = 0.5, no simulated annealing (SA) discs, r = 5 nm, q = 0.5 random, q = 0.5, KALP21 disc, r = 15 nm, q = 3.2 disc, r = 15 nm, q = 3.2, rhodopsin disc, r = 15 nm, q = 3.2, KALP21 nanodisc simulations discs, MSPs discs, KALP21 + MSPs discs, rhodopsin dimer + MSPs a

simulation name

DMPC

Bilayer BilayerKALP BilayerRho

300 300 300

Bicelle0q5Ran Bicelle0q5r3 Bicelle0q5r3NoSA Bicelle0q5r5 Bicelle0q5KALP Bicelle3q2 Bicelle3q2Rho Bicelle3q2KALP

1000 1044 1044 984 1000 1719 1610 1719

Nanodisc NanodiscKALP NanodiscRho

1200 1120 768

DHPC

repeats × time

Other

1 × 2 μs 1 × 2 μs 5 × 2.1 μs

4 KALP21, 16 Cl− rhodopsin dimer, 4 Na+ 2000 2124 2124 1968 2000 537 505 537

× × × × × × × ×

2a μs 2a μs 2 μs 2a μs 2a μs 2 μs 2.1 μs 2 μs

1 rhodopsin dimer, 4 Na+ 30 KALP21, 120 Cl−

3 1 1 1 3 1 1 1

16 MSP, 80 Na+ 16 MSP, 16 KALP21, 16 Na+ 16 MSP, 8 rhodopsin dimers, 112 Na+

1 × 2 μs 1 × 2 μs 1 × 2.1 μs

40 KALP21, 160 Cl−

Listed times include the simulated annealing protocol time (see Supporting Information for details).

nm13,48 to discs with radii of 4−8 nm.16,39,49,50 The choice of method for constructing the q = 3.2 bicelle was based on the fact that the simulated annealing protocol used for the small bicelles is only appropriate when more than one bicelle is expected to form. Since the system with a single q = 3.2 bicelle consists of more than 330000 beads, it was not possible to simulate systems with multiple bicelles of this size. Nanodiscs. A nanodisc was built by combining the two MSP1-Δ1−11 (MSP1 with the first 11 residues removed) used by Shih et al.31 with 150 DMPC lipids cut as a circle from a bilayer. This resulted in a peptide:lipid ratio of 1:75 in accordance with experimental data.38 The nanodisc was replicated such that the simulation box contained eight nanodiscs. Peptide and Protein Setups. KALP was embedded in q = 0.5 bicelles, a q = 3.2 bicelle, nanodiscs, and a bilayer with a peptide:lipid ratio of approximately 1:75, while the rhodopsin dimer was incorporated into a q = 3.2 bicelle, nanodiscs, and bilayers. Analysis. For comparing the membrane mimics to DHPC micelles, additional analyses of a previously published simulation of DHPC micelles were also performed.46 Most analyses were conducted on 501 snapshots distributed evenly over 1000 ns except for simulations including rhodopsin, where the snapshots were distributed evenly over 1050 ns due to the difference in time step. Increasing the number of snapshots used for the analyses did not affect the results. Only exceptions were the calculation of 31P spectra, and the lipid order parameters of DMPC in the q = 3.2 bicelles where 4001 snapshots distributed evenly over 1000 ns were used. Further details can be found in the Supporting Information.

Bicelle and nanodisc simulations had a lipid concentration of approximately 300 mM, which for systems with 3000 lipids resulted in setups consisting of approximately 440000 particles. General Simulation Setup. All simulations were performed in GROMACS 4.5.440 with the widely used coarse grain force field MARTINI 2.1.20,41 MARTINI has proved to be exceptional for studying lipid systems including self-assembly processes.42−44 In the MARTINI force field, approximately four heavy atoms and the associated hydrogen atoms are modeled as one bead. Such a simplification decreases the number of particles needed from 70 to 6 beads for describing DHPC and from 118 to 10 for DMPC (see Figure 1e). The polarizable water model45 was applied because it has been found essential for proper modeling of DHPC micelles.46 A time step of 25 fs was employed except in simulations with rhodopsin where a time step of 15 fs was applied. The temperature was kept at 308 K by use of Langevin dynamics with a coupling constant of 1.0 ps, while the pressure was controlled by the Berendsen weak coupling algorithm47 using a time constant of 3.0 ps, a compressibility of 3 × 10−5 bar−1, and a target pressure of 1.0 bar. In the bilayer simulations, a semiisotropic pressure coupling was employed with the dimension parallel to the membrane normal uncoupled from the other two while an isotropic pressure coupling was applied in all other simulations. A comprehensive description of the construction of the simulated systems and the analyses can be found in the Supporting Information. A short summery is given below. Bicelles. Two bicelle mimics were modeled with q values of 0.5 and 3.2, respectively. As detailed further in the Supporting Information, a self-assembly strategy using a simulated annealing (SA) protocol was developed for creating the small bicelles (q = 0.5). The large bicelles (q = 3.2) were made by cutting a disc with a radius of 15 nm from a DMPC bilayer and randomly mutating lipids to obtain the correct DMPC/DHPC ratio. This difference in approach was deemed necessary due to the known variation in the observed size of the small bicelles in previous studies and the slow equilibration of the self-assembly of the bicelles. Experimentally, it has been reported that the sizes and shapes of q = 0.5 bicelles vary from small almost spherical aggregates with a hydrodynamics radius of 2.9−4.3

■

RESULTS AND DISCUSSION First, the results from the simulations of the q = 0.5 bicelles are described followed by a comparison to the other membrane mimics. Then, the effects of including KALP21 in the bicelles focusing on the mutual adaptations in both interaction partners are presented. Finally, comparisons of the structural and dynamic properties of the mimics with embedded proteins/ peptides are made and general trends discussed. 15833


Article

The Journal of Physical Chemistry B Pure Membrane Mimics. Bilayer Simulations. Since the polarizable water model has not previously been tested with DMPC, we started out testing this for a simple DMPC-bilayer (simulation named Bilayer). The area per lipid was measured to 60.9 ± 0.1 Å2, which fits nicely with the experimental value of 60.6 ± 0.5 Å2,51 hereby confirming that the polarizable water model is able to reproduce experimental data for this system. Furthermore, the thickness of the bilayer, measured from the zdistance between the mean position of the glycerol beads of the two leaflets, was found to be 25.5 ± 0.2 Å, while the equivalent value for a previously simulated atomistic bilayer52 was 27.3 ± 0.3 Å. Considering that the diameter of a bead is approximately 5 Å, these two values are comparable. Since the polarizable water model in MARTINI is required to model DHPC micelles properly,46 it is likely also important for modeling the more advanced membrane mimics in this study, bicelles and nanodiscs. The polarizable water model was for this reason used in all the simulations. Small Bicelle Simulations. Due to the large discrepancy in the reported topology of q = 0.5 bicelles,13,16,39,49 we decided to perform an extensive self-assembly study of the small bicelles, hereby obtaining a bicelle distribution, rather than manually constructing a single bicelle with a predefined size. However, the diffusion of the bicelles in the simulations is rather slow at 308 K, leading to a low probability for the self-assembled aggregates to meet and fuse. The likelihood for one bicelle breaking into two on the simulated time scale is also low, making it difficult to form multiple small aggregates from larger ones. Together, these limitations result in challenges for obtaining a realistic size distribution for small bicelles within a reasonable simulation time. To overcome this, a simulated annealing approach was developed, hereby increasing both the diffusion speed and the possibility for bicelles to break (see the Supporting Information). The data reported in this section is for the simulations named Bicelle0q5Ran, except where otherwise noted. First, the size and shape of the bicelles were measured as described in the Supporting Information. The aggregation number distribution of the bicelles with q = 0.5 was rather broad (see Figure 2b) with a value of 91 ± 44 lipids. These bicelles had an average size of 5 × 6 × 8 nm (see Figure 2a) which corresponds well with the lowest reported experimental values.13,48 It is seen from Figure 2a that the thickness and the width were rather well-defined while the length of the bicelles varied more. No consensus is found in the literature regarding whether these small aggregates are in fact disc shaped with a pronounced bilayer part or if they are better described as mixed micelles.13,16,39,49,53 Figure 2a shows that the shape of the bicelles in this simulation was 65 ± 6% prolate (cigar shaped), 25 ± 5% oblate (disc shaped), and 9 ± 2% spherical. The oblate aggregates had a diameter of about 6.3 nm and a thickness of 4.7 nm (orange line). The prolate aggregates (blue line) displayed a broader distribution in calculated width, averaging at about 5.4 nm and with a large variation in the calculated length. Examples of the differently shaped q = 0.5 bicelles are shown in Figure 3. Simulations started from built q = 0.5 bicelles with a radius of 3 nm (Bicelle0q5r3) and 5 nm (Bicelle0q5r5) gave similar results (Figures S6 and S7), although not all of the bicelles with a radius of 5 nm (Bicelle0q5r5) fell apart during the equilibrium protocol but were obtained by performing multiple SA cycles (data not shown). These results indicate that a

Figure 2. Distribution of the shape and size of the small bicelles. The distribution in the three dimensions for (a) the q = 0.5 bicelles simulation (Bicelle0q5Ran) and for (c) the simulations of q = 0.5 bicelles with KALP21 peptides (Bicelle0q5KALP). The insets (b and d) show the distribution in lipid aggregation number for Bicelle0q5Ran and Bicelle0q5KALP, respectively. The fraction of the bicelles being prolate, oblate, and spherical are given with the standard error of the mean calculated between three independent runs.

Figure 3. Example of spherical-, oblate-, and prolate-shaped bicelles and equivalent model shapes.

monodisperse distribution is most efficiently obtained by starting from a random distribution of lipids. The width in the distribution of the aggregation number for the bicelles signifies that the individual bicelles will have different q values and contain a varying number of lipids. The relation between these values for each aggregate is displayed in Figure 4a, where the values are plotted for all bicelles over the trajectory. Each point refers to a given bicelle at a given snapshot and describes its DMPC/DHPC ratio (the q value) and the total number of lipids it contains. Two clear trends are observed from this graph: first of all, the points form curved lines, and second, the q value increases with the aggregation number. The points for the bicelles fall on separate curves, and each curve represents the variation in lipid composition of a specific bicelle over the time of the trajectory. Since DHPC is soluble in water, the DHPC molecules have a tendency to dissociate from one bicelle and fuse with another. The varying numbers of DHPC molecules in each of the bicelles consequently result in a change in both the aggregation number, and the q value of the individual bicelles. This shows 15834


Article


the smaller ones (see Figure 4a), the diffusion coefficient is likely underestimated from such an approach, resulting in overestimating their sizes. Other studies have been performed at low concentrations, once again resulting in a higher effective q value due to the large critical micelle concentration of DHPC which brings DHPC into solution, ultimately resulting in larger bicelles.16 Possible domain formation was analyzed by measuring the fraction of the DMPC lipid neighbors for both types of lipids in the bicelles to assess if the lipids are randomly mixed or not. The measured values for the q = 0.5 bicelles are plotted in Figure 5a with the amount of DMPC neighbors shown as red

Figure 4. (a) The effective q-value (DMPC:DHPC) in each of the q = 0.5 bicelles in Bicelle0q5Ran as a function of the number of lipids in the bicelle (aggregation number). (b) as in (a) but for the simulations including KALP21 (Bicelle0q5KALP) and colored according to the number of peptides in the bicelle.

Figure 5. Average fraction of the neighbors that are DMPC are plotted for DHPC (red), DMPC (blue), KALP21 (green), and the same number if the lipids were randomly mixed in the bicelles (black). (a) The results are plotted for the q = 0.5 bicelle simulations (All), the same setup where only bicelles with an aggregation number above 90 is counted (Min 90), the simulation without use of the simulated annealing protocol named Bicelle0q5r3NoSA, and the q = 0.5 bicelle simulations with KALP21. (b) The same for the q = 3,2 bicelle without (no KALP21) and with (with KALP21) KALP21 included. The error bars specify the standard error of the mean calculated between three independent runs (when available).

that we not only had a large variation in the size of the bicelles but also had a dynamic change in the individual bicelles resulting in large variation in their individual q values over time. It is furthermore apparent from Figure 4a, that the larger the bicelle (large aggregation number), the higher a q value that bicelle will have. The largest bicelles therefore have a q value higher than the overall DMPC/DHPC ratio of the simulation setup implies. It is thus evident that the DMPC/DHPC ratio of the individual bicelles is not a static well-defined value. Rather, the analysis reveals that great variations both over time and between the individual bicelles are found. Bicelles which correspond to the largest experimentally reported disc-shaped aggregates for q = 0.5, possessing a diameter above 10 nm at a temperature of 308 K, as determined by NMR,39 were not observed during the simulations. The simulation started from discs with a diameter of 10 nm (Bicelle0q5r5) revealed that such large aggregates rapidly changed into elongated assemblies (Figure S8). The simulations hereby propose that the high concentration of short-tailed lipids in the q = 0.5 bicelle mixtures cannot support large disc-shaped aggregates but rather promotes smaller assemblies or elongated morphologies. Similar findings, based on atomistic bicelle simulations, have been reported by Beaugrand et al. for bicelles with q = 0.25.53 One can speculate that such large discs, which have been proposed experimentally, could originate from an assumption of disc-shaped bicelles in the model fitting,16,39,50 since the hydrodynamic radius of ∼3− 4 nm obtained by dynamic light scattering fits the simulated bicelles.13,16,48 Furthermore, the existence of a broad size distribution, similar to what was observed from the simulations herein, may not have been accounted for in the experimental model fitting. The largest bicelles with a hydrodynamic radius of ∼5 nm was proposed based on the diffusion coefficient of DMPC measured in NMR experiments.39,50 However, since the DMPC content is elevated in the larger bicelles compared to

points for DHPC and blue points for DMPC. To have an indication of what a random mixture would result in, the total number of DMPC contacts was divided with the total number of lipid−lipid contacts (black points). A point above the “randomly mixed”-point (black) indicates that the molecules on average are positioned in DMPC-enriched domains, while a value below the “randomly-mixed”-point indicates a DMPCdepleted environment. Figure 5 thus demonstrates a partial segregation of the lipids, as DMPC was most often located in a DMPC-enriched environment while DHPC was found in a DMPC depleted region (“all” in Figure5a). This may be related to DMPC mainly participating in the larger aggregates while DHPC forms the primary component of the smaller aggregates, as described above. Therefore, the neighbor analysis was refined now using only the bicelles with an aggregation number of at least 90 (“Min 90” in Figure 5a) to explore if the segregation was similarly observed within the larger aggregates. While all the values moved toward a larger fraction of DMPC neighbors, segregation was still observed. The q = 0.5 bicelles started from a disc with a radius of 3 nm and simulated without use of the SA protocol (Bicelle0q5r3NoSA), although holding approximately the same number of lipids as the average value observed in the simulations with SA (Bicelle0q5Ran), showed a DMPC neighbor fraction of 37.1 (DMPC), 34.4 (DHPC), and 36.0 (random) (“No SA” in Figure 5), signifying that less segregation was present in the highly monodisperse bicelles. From these observations, two conclusions can be drawn: (i) a broad size distribution leads to more segregation and (ii) the segregation will then be partly due to the formation of small 15835


Article


variation in a sample with multiple bicelles will presumably be larger. The segregation analysis of the q = 3.2 bicelle clearly shows that DMPC remains mainly in a DMPC-enriched environment while DHPC was located in a DMPC-depleted region, indicating a segregation of the lipids (“No KALP21” in Figure 5b). This is also apparent from visual inspection (see Figure 1d), where it can be seen that DMPC is positioned mainly in a bilayer part, while DHPC lipids move to the rim. Furthermore, unlike what was observed for q = 0.5 bicelles, the DMPC cluster analysis (Figure S9b) reveals two large DMPC clusters, one containing ∼820 DMPCs and one with ∼1640 DMPCs, corresponding to a situation where the two leaflets of the bicelle were each assigned their own cluster or when they both were cluster together, respectively. A well-defined bilayer is thus present. It is, however, also clear that the segregation is not complete as some of the DHPC molecules are mixed into the bilayer part, as proposed by the mixed bicelle model.58 Approximately 12 ± 1% of the lipids in the bilayer part of the q = 3.2 bicelle were found to be DHPC, which fits reasonably well with the NMR measurements estimating this to be around 8% at a temperature of 308 K.56,58 As for the q = 0.5 bicelles, a rapid exchange of the DHPC molecules was also observed in the q = 3.2 bicelle. During 1 μs of simulation, each of the 537 DHPC molecules on average moved from the bicelle into solution or vice versa once. The observed difference between the q = 0.5 and q = 3.2 bicelles might be due to the difference in free lipid concentration which was found to be 8 and 4 mM for the q = 0.5 and q = 3.2 bicelles, respectively, which is close to the concentrations of 5−7 mM reported from experiments at both q values.13,53,55,59 Nanodisc Simulations. Simulations of nanodiscs, of a size that is applicable under similar conditions as q = 0.5 bicelles, were undertaken to elucidate the differences and similarities between nanodiscs and bicelles. A simulation setup including eight nanodiscs each consisting of two membrane scaffold proteins (MSPs) and 150 lipids was conducted (Nanodisc). The nanodiscs were found to be much less dynamic than the q = 0.5 bicelles due to the rigidity imposed by the MSPs, and the overall topology of the nanodiscs thus remained discshaped. The nanodiscs were mostly flat, although a slight bending of the bilayer part was observed. Comparison of the Membrane Mimics. In this comparison of the membrane mimics, we will focus on the properties important when assuming transferability to a regular lipid bilayer, such as surface curvature and solvent exposure. To get an impression of the curvature, the solvent accessible surface area (SASA) per lipid was computed and plotted in Figure 6. A large SASA of a given lipid is consistent with a large curvature since a larger fraction of each of the lipids then becomes accessible to water molecules. As shown in Figure 6, the DHPC micelles46 possess a large SASA (311 Å2/lipid) in accordance with the large curvature of a micelle, and the DHPC lipids show decreased solvation in both the q = 0.5 (277 ± 4 Å2/lipid) and q = 3.2 (233 Å2/lipid) bicelles, due to the presence of less curved surface areas formed by DMPC and the DHPC lipids partly mixing with the DMPC lipids. The SASA of the DMPC molecules also decreases as the q value increases; from 230 ± 4 Å2/lipid for q = 0.5 to 181 Å2/lipid for q = 3.2 coming closer to the value of the DMPC bilayer (177 Å2/lipid). This once more indicates that the q = 0.5 bicelles are more micelle-like than would be expected by the ideal bicelle model,12 where the DMPC lipids are expected to form a regular bilayer. It can be

DHPC rich bicelles and larger DMPC-enriched bicelles, and partly due to segregation within the individual aggregates. These results are consistent with previous results found by NMR showing that DMPC and DHPC are not fully mixed in the q = 0.5 bicelle mixtures.50 To investigate this further, the DMPC cluster size distribution was calculated and plotted in Figure S9a. This analysis shows no single domain size, not even in the highly monodisperse bicelle distribution in Bicelle0q5r3NoSA. Instead, the clusters vary in size with the smallest clusters being most abundant, indicating a large degree of lipid mixing with local DMPC cluster formation. While the q = 0.5 bicelles were found to be mainly prolate (Figure 2), their structure was not static and some bicelles were able to change between a prolate and an oblate shape within 80 ns of simulation, within which time the lipids in that specific bicelle also became totally mixed (Figure S5). Not only did the lipids in the individual bicelles mix rapidly, a large exchange of the DHPC molecules between the aggregates and the solvent was also observed. On average, each DHPC molecule changed from being in an aggregate to being in solution or vice versa three times during 1 μs of simulation. A concentrationdependent variation in the 31P chemical shift value of DHPC in bicelles has previously been found and assigned to a fast exchange of DHPC between the bulk solution and the bicelles.16,53 This agrees with the fast exchange observed in our simulations. All these results add to the conclusion that a bicelle has a very dynamic structure, and the q = 0.5 bicelles should not be regarded as disc-shaped lipid aggregates but rather as dynamic assemblies which are mostly prolate in shape. Our simulation data thus support the findings that the lipids in q = 0.5 bicelles are not largely segregated since mostly local DMPC clusters are formed,13,53 and bilayer parts are small and only transiently present. This is also consistent with the findings that the hydrodynamic radius of the q = 0.5 bicelles only show little temperature dependence,48 indicating that the increased size expected from an increased lipid mixing is limited. Large Bicelles. It is widely debated whether the bicelles at q = 3.2 are disc shaped with a diameter of 20−40 nm54,55 or if they form ribbons/cylinders13,15 or perforated bilayers.56,57 The morphology has been found to depend on several factors such as lipid concentration, temperature, and the presence of ions and buffer.13−15,53 We recently showed that a disc-shaped bicelle can describe the two peaks observed in 31P NMR measurements,52 and we thus here assume the bicelle to be disc-shaped. A bicelle disc with a radius of 15 nm was therefore constructed by cutting a circle out of a bilayer and randomly mutating DMPC into DHPC until a DHPC:DMPC ratio of 3.2 was reached (Bicelle3q2). The q = 3.2 bicelle kept the disc-shaped topology throughout the simulation (Figure S10), which indicates that bicelle discs can be a stable morphology. However, other morphologies cannot be ruled out since a change to either perforated bilayers or ribbons cannot be expected for the simulated system. To address this question, further investigations are needed, which are currently outside the scope of simulations but will be possible as computational power increases. Similarly to the q = 0.5 bicelles, the bicelle in the q = 3.2 system was observed to change aggregation number during the simulation, varying between 2213 and 2234 lipids, which resulted in the DMPC:DHPC ratios fluctuating between 3.34 and 3.48. However, only one large bicelle has been simulated, and the 15836


Article


distribution in the aggregation number is shown in Figure 2d. It is realized that KALP21 only had a minor effect on the aggregation number distribution of the bicelles as the number of lipids in the bicelles was 88 ± 52, close to the value of 91 ± 44 observed for the q = 0.5 bicelles without peptides. A small decrease in the aggregate number distribution is observed near ∼150, likely due to the limited number of bicelles simulated. Although only a small change in the average aggregation number was observed, a significant alteration in the shape distribution could be detected. An increase in the fraction of oblate (from 25 ± 5 to 35 ± 5%) and spherical (from 9 ± 2 to 12 ± 3%) aggregates was observed compared to the pure bicelles, whereas the amount of prolate (from 65 ± 6 to 54 ± 3%) aggregates decreased (see Figure 2). Overall, the occurrence of oblate bicelles in the mixture increased with about 10% when adding the peptide, indicating that hydrophobic transmembrane peptides may be able to induce a more disclike shape of the small bicelles. This effect is most likely due to a preference of such transmembrane hydrophobic peptides to interact with long-tailed lipids such as DMPC due to a lower hydrophobic mismatch compared to the mismatch imposed by DHPC. Support for this hypothesis was established when inspecting the degree of segregation, as the average fraction of DMPC neighbors (“KALP21” in Figure 5a) indicated that KALP21 was positioned in a DMPC-enriched domain with about 42.8 ± 1.5% DMPC neighbors. This was significantly higher than if the lipids were randomly distributed (36.3 ± 0.1%). The increase in the fraction of spherical bicelles (up from 9% to 12%) indicate that the short-tailed lipids (DHPC) were forced to form smaller aggregates due to the need for binding of the long-tailed lipids (DMPC) to the peptides in the large aggregates. This change in the morphology of the bicelles in the ensemble is in accordance with the studies published by Bodor et al., where KALP21 was found to change the hydrodynamic radius of the bicelles while a surface-active peptide only affected the mobility of the lipids in the bicelles.65 Examples of an oblate and a prolate bicelle with KALP peptides embedded are shown in Figure 7.

Figure 6. Average solvent accessible surface area (SASA) per lipid in each of the membrane mimics. The error bars for q = 0.5 bicelles show the standard error of the mean between three independent runs.

speculated that this could induce non-native proteinconformations when embedding proteins into q = 0.5 bicelles as, for example, observed for BtuB.60 However, as the lipids showed much lower SASA in bicelles than in micelles, the related lower curvature in the mimics may potentially provide an explanation of why some proteins (e.g., rhodopsin) are more active in small bicelles than in micelles.61,62 When comparing to the SASA of a DMPC bilayer, it can be concluded that the q = 3.2 bicelles had a curvature almost equivalent to the bilayer, which makes it much more suitable as a membrane mimic than micelles and the q = 0.5 bicelles in this regard. The SASA of the nanodiscs was found to be surprisingly high (211 Å2/lipid) with a value only slightly lower than the q = 0.5 bicelles. This could be due to tilting of the lipid head groups over the MSPs. However, only a small decrease in SASA from 211 Å2/lipid to 206 Å2/lipid was observed when calculating the SASA for the lipids that was not in contact with the MSP. It shows that, although the lipids at the rim have a higher solvation, an increased SASA is present for all the lipids in a nanodisc as compared to the bilayer, indicating that the increased curvature induced by the MSP propagates throughout the entire nanodisc. It may also be that more lipids are needed for the nanodiscs to be fully loaded since some discrepancies in the number of DMPC lipids per nanodisc exist in the literature: fully loaded nanodiscs have been suggested to contain between ∼150 and 190 DMPC lipids.38,63,64 Nanodiscs with low lipid:MSP ratios have furthermore been found to show increased lipid solvation, similar to what is observed here.64 MD simulations at a higher DMPC:MSP ratio of 160:2 showed that the lipid tails of DMPC attain a higher-order parameter, which is consistent with NMR measurement,33 thus indicating that the nanodisc can in fact contain more than 150 DMPC lipids. Membrane Mimics with KALP21 Embedded. Simulations were undertaken to study the effect the bicelles may have on embedded peptides compared to other membrane mimics. Furthermore, the effect the peptides have on the bicelles size and shape was also analyzed. For the q = 0.5 bicelle simulation, 40 KALP21 peptides were added to the mixture of lipids before simulated annealing and self-assembly. For the q = 3.2 bicelles, 30 KALP21 peptides were embedded into the bicelle. This gave a peptide:lipid ratio of approximately 1:75 in both bicelle mixtures as described in the Supporting Information. Q = 0.5 Bicelles with KALP21 Embedded. The size and shape distributions from the simulations of the q = 0.5 bicelles including KALP21 are depicted in Figure 2c, while the

Figure 7. Snapshot of an oblate and a prolate bicelle with KALP21 in a transmembrane conformation. KALP is green, DHPC’s phosphate group is red, DMPC’s phosphate group is blue, and the hydrophobic core of the bicelle is shown in transparent gray.

A similar relation between the q value and the aggregation number was found for the bicelles with KALP21 (see Figure 4b). The q value is highest for the bicelles with largest aggregation number, and the points fall on lines demonstrating the dynamic behavior of DHPC. It is furthermore evident that the larger bicelles contain the most bound KALP21, as would be expected. 15837


Article


examined the orientation of the headgroup by 14N NMR measurements and found that the angle of the P−N vector with the bilayer plane to increase when KALP23 was added to a DMPC bilayer.69 This trend was also observed in the bicelle simulations where the average angle between the PO4-NC3 vector of the DMPC lipids and the bilayer plane increased from 24.75° to 25.76° upon addition of KALP21. Doux et al. suggested it to be due to electrostatic interaction since neutral peptides did not have this effect on the headgroup orientation. The lipids close to the KALP21 showed an average angle of 25.84° in the simulation, which is close to the average of all the DMPC lipids in the bicelle. It is thus not obvious if the ordering is due to local lipid ordering near KALP21 or an increased rigidity of the bicelle affecting all lipids in the bicelle. However, Figure 5b indicates that KALP21 selectively interact with the bilayer part consisting of DMPC, and effects on the bicelle are therefore likely similar to what would be obtained in a bilayer environment. Comparing Membrane Mimics with Embedded KALP21. Due to the small aggregates observed in the q = 0.5 bicelle simulations, an obvious expectation would be that any peptide embedded in these small bicelles would be more water accessible than when inserted in bilayers. Therefore, we calculated the average SASA of the individual KALP21 when present in the different lipid systems and found that the water accessibilities of KALP21 in either q = 3.2 bicelles (18.2 ± 0.6%) or DMPC bilayers (17.8 ± 0.9%) are alike while the peptides show similarly increased solvation in q = 0.5 bicelles (23.9 ± 5.3%) and nanodiscs (24.6 ± 4.8%). The standard deviation stated for each value describe the variation between the individual peptides, and the variation in the solvation of the peptides in the q = 0.5 bicelles was thus larger than that for the bilayers and the large q = 3.2 bicelles. This large variation indicates that the wide size distribution of the q = 0.5 bicelles leads to a large difference in the solvent accessibility of the individual KALP21 peptides. Whereas some peptides in the small bicelles from the q = 0.5 simulations show increased solvation due to being located in small bicelles without a proper bilayer present, others are similarly solvated to peptides embedded in a bilayer. Interestingly, the same large deviation in solvation of the peptides is observed in nanodiscs, which was found to be due to peptides in contact with the MSPs being much more solvent exposed than the peptides located in the interior of the nanodiscs. Therefore, the solvation of the peptides in both q = 0.5 bicelles and nanodiscs can vary a lot dependent on the specific location of a given peptide. The increased solvation of KALP21 indicates that the hydrophobic part of the peptide, which is usually embedded deeply in the membrane, is exposed to the solvent and the lipid head groups. This could result in un- or misfolding of the peptides. Thus, to check this hypothesis, the number of contacts between the glycerol linker beads of a given lipid type and each bead of the peptide was counted and normalized relatively to the maximum number of contacts counted. This value was subtracted from the equivalent normalized value for the contacts of the glycerol beads of DMPC in a pure DMPC bilayer and illustrated in Figure 9. A red color therefore describes parts of the peptide which show increased interaction with the glycerol linker of each type of lipid compared to the fraction of DMPC glycerol linker contacts observed when the peptide is embedded in a DMPC bilayer. Similarly, a blue color indicates areas with decreased interaction. It is apparent that DMPC in the q = 3.2 bicelle (Figure 9a) interacts with similar

Q = 3.2 Bicelles and Nanodiscs with KALP21. Neither the q = 3.2 bicelle nor the nanodiscs changed shape by addition of KALP21 (Figures S10b and S11b). In our recent work, we demonstrated that it is possible to simulate the 31P NMR spectra of oriented lipid bilayers from AA simulations, and we were able to reproduce the essential features of the 31P NMR spectra from CG simulations as well.52 Here, we investigate the differences in the simulated 31P NMR spectra of large bicelles with and without membrane-bound peptides for comparison

Figure 8. Simulated 31P spectra for oriented q = 3.2 bicelles with and without KALP21. The peaks at approximately −8 ppm and −14 ppm originate from DHPC and DMPC, respectively. The dashed lines indicate the peak positions in the pure bicelles.

with previous experiments. Figure 8 displays the time-averaged 31 P spectra of the q = 3.2 bicelles with and without the peptide KALP21. Both spectra reveal two peaks at approximately −8 and −14 ppm in agreement with typical experimental findings,66 although the experimentally observed peaks often are shifted toward zero because of the nonperfect alignment of the bicelle resulting in an order parameter of ∼0.8 for bicelles,12 which we do not take into account in the present simulations. The addition of KALP21 shifts the two peaks toward more negative 31P frequencies, which we interpret as an increased order of the lipids in the presence of KALP21. In the following, we focus on the DMPC peak at −14 ppm as the position of the DHPC peak is affected by various other properties such as the free DHPC concentration. A similar increase in the chemical shift has been observed when the positively charged toroidalpore-forming antimicrobial peptide novicidin is added to mixed DMPC/DMPG bilayers. 67 We are not aware of any experimental 31P NMR studies of aligned DMPC/DHPC bicelles with KALP21. However, the present molecular assembly shows remarkable similarities with the KALP23/ DMPC system investigated by 2H solid-state NMR.68 In this study, de Planque et al. observed that addition of the peptide slightly increased the 2H quadrupolar splitting, which was interpreted in the same way as when the 31P resonances move to larger negative frequencies in the bicelle system. Although the systems are slightly different, they seem relevant to compare since the CG peptide is shorter (21 instead of 23 residues) and the CG DMPC bilayer is 1.8 Å thinner than what would be expected for an atomistic DMPC bilayer as shown above, implying that the systems are similar. For comparison, the order parameters of the bonds between the CG beads in DMPC were calculated from the simulations (Figure S12 of the Supporting Information) revealing only a very small increase in the order parameter for DMPC lipids close to KALP21. Doux et al. have 15838


Article


KALP21 in nanodiscs (Figure S13b), it is apparent that the peptides which are in contact with the membrane scaffold protein have a higher tilt angle than those which are not, which indicate that the perturbing effect of the nanodisc may depend on the studied peptides, which may have a tendency to interact with the MSPs. For the q = 3.2 bicelles, no correlation is found between the peptides’ distance from the center of the bicelle and the tilt angle (see Figure S13c), indicating that the bending of the large bicelle could affect the apparent tilt angle. To investigate the peptide−peptide interactions, the degree of oligomerization of the KALP21 peptides was measured. Multimer formation was observed in all membrane mimics (Figure S14) and an increase in the number of formed multimers was observed in both q = 3.2 and q = 0.5 bicelles compared to a bilayer while a decrease was observed in the nanodiscs. The increased fraction of multimers in the bicelles correlates with the increased peptide:DMPC ratio going from bilayers over q = 3.2 to q = 0.5 bicelles, since all simulations were done at the same peptide:(DHPC+DMPC) ratio. The multimer formation thus seems to increase as the amount of DMPC decreases. This is consistent with the increased dimerization observed for the related peptide WALP23 in vesicles at increased peptide concentrations.72 Membrane Mimics with Rhodopsin Embedded. As GPCRs are of great interest due to their pronounced importance in cell regulation, signaling, and as drug targets,4,73 we chose to use a rhodopsin dimer as a test case for how well the different lipid bilayer mimics accommodate a large multimeric transmembrane protein. Since it is not known if dimers form in q = 0.5 bicelles and no indication of rhodopsin dimer formation in small bicelles have been found,37 these were not considered in this part of the study. Pictures of the rhodopsin dimer embedded in a q = 3.2 bicelle and a nanodisc are shown in Figure 10a. Comparing Membrane Mimics with Rhodopsin Embedded. The interaction between the lipid glycerol linkers of DMPC and the rhodopsin dimer is very similar to that observed for KALP21 in both q = 3.2 bicelles and nanodiscs (Figure 10b); DMPC shows a similar interaction pattern in q = 3.2 bicelles as in a bilayer, whereas a decreased interaction between the glycerol bead and the protein is observed in the nanodiscs in areas where the rhodopsin dimer is in contact with the MSPs, which results in increased water interactions in these areas (Figure S15). However, in q = 3.2 bicelles, the DHPC molecules bind at areas not normally occupied by lipids, specifically on top of the dimer interface [red color for “Large Bicelle (DHPC)” in Figure 10b], which indicates that the DHPC present in the mixture interact with the peripheral regions of the dimer not usually in contact with lipids, due to the dynamic behavior of DHPC. This may lead to problems in multimer formation, as the DHPC lipids by binding in the interface regions holds the potential for competing with the other rhodopsin monomers. Since the function of some proteins has been found to be affected by the type and dynamics of the lipid environment,74 the residential time between the annular lipids and the rhodopsin dimer was calculated (see Figure S16), and it is clear that the lipids in the nanodiscs remain bound to the rhodopsin dimer for a lot longer time than observed in any of the other membrane mimics. A few percent of the interactions were even observed to prevail for the entire 1 μs simulation. Analyzing the residential time to each of the 7 transmembrane helices (see Figure S17) leads to the conclusion that lipids

Figure 9. Interaction between the glycerol linker of each lipid type and KALP21 in (a) q = 3.2 bicelles, (b) q = 0.5 bicelles, and (c) nanodiscs compared to the number of contacts between the glycerol beads of DMPC in a DMPC bilayer and KALP21.

parts of KALP21 as when the peptide is inserted in a DMPC bilayer, since no strong colors are found. As expected, the interaction with DHPC glycerol linkers is slightly decreased near the terminals of the peptide, due to the shorter lipid tails of DHPC. However, in the q = 0.5 bicelles (Figure 9b), a clear increase in interaction is observed between the DHPC glycerol bead and the middle of the peptide. This shows that the hydrophobic part of the peptides, due to the size and shape of the q = 0.5 bicelles, is in contact with the headgroup of DHPC lipids which potentially could perturb the peptide. Because the headgroup contacts are typically related to increased solvation, an increased solvation would also be expected in the q = 0.5 bicelles compared to in the q = 3.2 bicelles, which could be tested by comparing the solvent accessibility of the transmembrane part of the peptide in the membrane mimics by paramagnetic relaxation enhancement (PRE) experiments. Previous experiments, showing that bicelles with q < 0.96 decrease the stability of embedded proteins, also indicate that q = 0.5 bicelles can destabilize the embedded proteins.61 Increasing q further does not necessarily increase the stability any more, since, for example, the surfactant concentration and thus the total lipid concentration may affect the unfolding free energy.70 In nanodiscs (Figure 9c), a decrease in interactions is observed between KALP21 and DMPC due to the binding of the peptide to the MSP hereby limiting the number of lipid contacts it can make. Local membrane thinning and thickening have been found important for the function of membrane proteins.71 A property of similar great interest is therefore the tilt angle of a peptide that inserts into a membrane mimic since differences in tilt angles may be an indication of a mismatch between the length of the peptide and the membrane environment. The tilt angle of KALP21 was measured (Figure S13a), and in all the membrane mimics, the peptides show a slight perturbation in the angle distribution toward higher values compared to a pure lipid bilayer. Furthermore, in the q = 0.5 bicelles the peptides display tilt-angles over the full range of possible values, due to the fact that not all bicelles have a bilayer part into which the peptides can insert. From inspection of the tilt angle of 15839


Article


membrane mimic can change upon addition of proteins or peptides as observed in this study where the small bicelles become more disclike with KALP21 included. The assumptions and conclusions made from studies without proteins may therefore not apply when peptides or proteins are added. This has recently also been found to be the case in a nanodisc where an ABC transporter was found to replace a much larger number of lipids than what would be expected from its size, and the remaining lipid belt was substantially thinned.76 The results obtained for one specific peptide may furthermore not be transferable to other peptides, as embedding a transmembrane peptide affects the bicelles in a different way than observed for a surface-bound peptide.65 Moreover, the q = 0.5 bicelles have a wide size distribution. This accounts for part of the observed lipid segregation, as the smallest bicelles in the distribution also holds the lowest q values. The prolate shape of the bicelles and the low segregation of the lipids imply that the lipids are more mixed than suggested by the idealized bicelle model.12 Large bilayers are thus not present in these small bicelles as they mostly contain smaller DMPC domains. Compared to other membrane mimics, the increased solvent accessible surface area indicates that the curvature of the small q = 0.5 bicelles is significantly higher than that for bilayers and large q = 3.2 bicelles, although only slightly higher than found for nanodiscs with a DMPC:MSP ratio of 150:2 and lower than found for DHPC micelles. Small bicelles are from this perspective therefore a fine tool for solution state experiments, but one should be aware of its limitations, such as increased solvation of the embedded peptide, compared to the larger membrane mimics. The KALP21 peptide showed increased solvation in both q = 0.5 bicelles and nanodiscs compared to when embedded in a DMPC bilayer. For avoiding this effect, one would need to go to larger membrane mimics such as q = 3.2 bicelles. The large bicelles were, as expected, found to be much better mimics of a lipid bilayer than smaller membrane mimics; however, the use of these large membrane mimics are limited to techniques where a large mobility of the membrane mimic is not essential, such as in solid state NMR. The presence of the DHPC molecules in the large bicelles does however result in peripheral binding of DHPC to a region of the rhodopsin dimer normally not occupied by lipids. This may then also perturb either multimer formation or the dynamics of the protein. This is not the case with the nanodiscs as no surfactants are present in the nanodisc solution. As the exact morphology of the large bicelles is still debated, further studies are underway, focusing on elucidating some of the different topologies of lipid mixtures and their effect on protein structure and function by combining MD with SAXS and NMR experiments.

Figure 10. Rhodopsin embedded in large bicelle and nanodiscs. (a) Pictures of a rhodopsin dimer embedded in a q = 3.2 bicelle and in a nanodisc. The rhodopsin molecules are visualized by their transmembrane helices colored according to the helix number: gray, yellow, orange, pink, cyan, purple, and green for helix one to seven, respectively. DMPC molecules are colored blue, DHPCs are red, while the backbones of the MSPs are visualized in green. (b) The lipid glycerol linker interaction with the rhodopsin dimer in a q = 3.2 bicelle and nanodiscs compared to the contacts with the DMPC glycerol bead in simulations where the dimer is embedded in a DMPC bilayer. Red areas have increased interaction compared to a bilayer while blue areas show decreased interaction.

bound near the MSP-Rhodopsin interface (near TM3, TM4, and TM5) were much less dynamic when inserted in the nanodiscs compared to the other membrane mimics. Furthermore, the DMPC lipids were found to remain bound near the dimer interface (TM1, TM2, and TM7) for longer time intervals, which shows that the dynamics of the lipids around the protein may vary significantly dependent on the environment of the individual lipids. Crystal structures of GPCRs with membrane components in so-called nonannular binding sites75 have previously been published with cholesterol and lipid tails tightly bound between helices, for example, between TM6 and TM7 of rhodopsin. However, we see no indication of such sites, maybe due to the coarse description of the CG model. However, the reduced dynamics near the dimer interface indicates that the binding time is related to the volume of the lipid binding site. Therefore, it seems likely that specificity of the nonannular sites may be due to an exact match between the size and shape of the bound lipid and the binding site. This decreases the dynamics enough for retaining the lipid during the crystallization process.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jpcb.5b08463. A thorough description of the development of the simulated annealing protocol, the construction of the simulated system, and the performed analyses. Additional figures describing DMPC cluster size in the bicelles, the size and shape of the q = 0.5 bicelles simulations build with a radius of 3 and 5 nm, the shape and size distribution observed for the nanodiscs and large bicelles

■

CONCLUSION Membrane mimics are essential for studying membraneassociated proteins and peptides in a membrane-like environment, but as illustrated in this study, the choice of membrane mimic can affect the obtained results in a vast number of ways. First of all, one needs to be aware that the size and shape of the 15840


Article


■

(12) Vold, R. R.; Prosser, R. S. Magnetically Oriented Phospholipid Bilayered Micelles for Structural Studies of Polypeptides. Does the Ideal Bicelle Exist? J. Magn. Reson., Ser. B 1996, 113, 267−271. (13) van Dam, L.; Karlsson, G.; Edwards, K. Direct Observation and Characterization of DMPC/DHPC Aggregates Under Conditions Relevant for Biological Solution NMR. Biochim. Biophys. Acta, Biomembr. 2004, 1664, 241−256. (14) Nieh, M.; Glinka, C. J.; Krueger, S.; Prosser, R. S.; Katsaras, J. SANS Study on the Effect of Lanthanide Ions and Charged Lipids on the Morphology of Phospholipid Mixtures. Biophys. J. 2002, 82, 2487− 2498. (15) Harroun, T. A.; Koslowsky, M.; Nieh, M.; de Lannoy, C.; Raghunathan, V. A.; Katsaras, J. Comprehensive Examination of Mesophases Formed by DMPC and DHPC Mixtures. Langmuir 2005, 21, 5356−5361. (16) Glover, K. J.; Whiles, J. A.; Wu, G.; Yu, N.; Deems, R.; Struppe, J. O.; Stark, R. E.; Komives, E. A.; Vold, R. R. Structural Evaluation of Phospholipid Bicelles for Solution-State Studies of MembraneAssociated Biomolecules. Biophys. J. 2001, 81, 2163−2171. (17) Dürr, U. H. N.; Gildenberg, M.; Ramamoorthy, A. The Magic of Bicelles Lights Up Membrane Protein Structure. Chem. Rev. 2012, 112, 6054−6074. (18) Jonas, A.; Kézdy, K. E.; Wald, J. H. Defined Apolipoprotein A-I Conformations in Reconstituted High Density Lipoprotein Discs. J. Biol. Chem. 1989, 264, 4818−4824. (19) Ritchie, T. K.; Grinkova, Y. V.; Bayburt, T. H.; Denisov, I. G.; Zolnerciks, J. K.; Atkins, W. M.; Sligar, S. G. Reconstitution of Membrane Proteins in Phospholipid Bilayer Nanodiscs. Methods Enzymol. 2009, 464, 211−231. (20) Marrink, S. J.; Risselada, H. J.; Yefimov, S.; Tieleman, D. P.; de Vries, A. H. The MARTINI Force Field: Coarse Grained Model for Biomolecular Simulations. J. Phys. Chem. B 2007, 111, 7812−7824. (21) Klingelhoefer, J. W.; Carpenter, T.; Sansom, M. S. P. Peptide Nanopores and Lipid Bilayers: Interactions by Coarse-Grained Molecular-Dynamics Simulations. Biophys. J. 2009, 96, 3519−3528. (22) Parton, D.; Klingelhoefer, J.; Sansom, M. P. Aggregation of Model Membrane Proteins, Modulated by Hydrophobic Mismatch, Membrane Curvature, and Protein Class. Biophys. J. 2011, 101, 691− 699. (23) Fuhrmans, M.; Marrink, S. J. Molecular View of the Role of Fusion Peptides in Promoting Positive Membrane Curvature. J. Am. Chem. Soc. 2012, 134, 1543−1552. (24) Domański, J.; Marrink, S. J.; Schäfer, L. V. Transmembrane Helices can Induce Domain Formation in Crowded Model Membranes. Biochim. Biophys. Acta, Biomembr. 2012, 1818, 984−994. (25) Polyansky, A. A.; Ramaswamy, R.; Volynsky, P. E.; Sbalzarini, I. F.; Marrink, S. J.; Efremov, R. G. Antimicrobial Peptides Induce Growth of Phosphatidylglycerol Domains in a Model Bacterial Membrane. J. Phys. Chem. Lett. 2010, 1, 3108−3111. (26) de Joannis, J.; Jiang, F. Y.; Kindt, J. T. Coarse-Grained Model Simulations of Mixed-Lipid Systems: Composition and Line Tension of a Stabilized Bilayer Edge. Langmuir 2006, 22, 998−1005. (27) Vácha, R.; Frenkel, D. Stability of Bicelles: A Simulation Study. Langmuir 2014, 30, 4229−4235. (28) Shih, A. Y.; Freddolino, P. L.; Arkhipov, A.; Schulten, K. Assembly of Lipoprotein Particles Revealed by Coarse-Grained Molecular Dynamics Simulations. J. Struct. Biol. 2007, 157, 579−592. (29) Shih, A. Y.; Arkhipov, A.; Freddolino, P. L.; Schulten, K. Coarse Grained Protein-Lipid Model with Application to Lipoprotein Particles. J. Phys. Chem. B 2006, 110, 3674−3684. (30) Shih, A. Y.; Arkhipov, A.; Freddolino, P. L.; Sligar, S. G.; Schulten, K. Assembly of Lipids and Proteins into Lipoprotein Particles. J. Phys. Chem. B 2007, 111, 11095−11104. (31) Shih, A. Y.; Denisov, I. G.; Phillips, J. C.; Sligar, S. G.; Schulten, K. Molecular Dynamics Simulations of Discoidal Bilayers Assembled from Truncated Human Lipoproteins. Biophys. J. 2005, 88, 548−556. (32) Shih, A. Y.; Sligar, S. G.; Schulten, K. Maturation of HighDensity Lipoproteins. J. R. Soc., Interface 2009, 6, 863−871.

as well as the order parameters of DMPC in q = 3.2 bicelles. Figure of the flexibility of the q = 0.5 bicelles, figures of the angle distribution and oligomerization of KALP21, and figures describing the solvation and lipid− protein interaction time of rhodopsin in different membrane mimics (PDF)

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. Tel: +45 8715 5975. Fax: +45 8619 6199. Present Address ∥

L.T.: QIAGEN-Aarhus, Silkeborgvej 2, DK-8000 Aarhus C, Denmark. Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS We thank Xavier Periole and Siewert-Jan Marrink for providing the structure of the rhodopsin dimer. Financial support has been obtained from the Danish Council for Independent Research | Technology and Production Sciences (FTP), the Danish National Research Foundation (DNRF59), and the Danish Centre for Scientific Computing, which are all acknowledged.

■

REFERENCES

(1) Wallin, E.; Von Heijne, G. Genome-Wide Analysis of Integral Membrane Proteins from Eubacterial, Archaean, and Eukaryotic Organisms. Protein Sci. 1998, 7, 1029−1038. (2) Drews, J. Drug Discovery: A Historical Perspective. Science 2000, 287, 1960−1964. (3) Warschawski, D. E.; Arnold, A. A.; Beaugrand, M.; Gravel, A.; Chartrand, É; Marcotte, I. Choosing Membrane Mimetics for NMR Structural Studies of Transmembrane Proteins. Biochim. Biophys. Acta, Biomembr. 2011, 1808, 1957−1974. (4) Serebryany, E.; Zhu, G. A.; Yan, E. C. Y. Artificial Membrane-Like Environments for in Vitro Studies of Purified G-Protein Coupled Receptors. Biochim. Biophys. Acta, Biomembr. 2012, 1818, 225−233. (5) Cross, T. A.; Sharma, M.; Yi, M.; Zhou, H. Influence of Solubilizing Environments on Membrane Protein Structures. Trends Biochem. Sci. 2011, 36, 117−125. (6) Georgieva, E. R.; Ramlall, T. F.; Borbat, P. P.; Freed, J. H.; Eliezer, D. Membrane-Bound α-Synuclein Forms an Extended Helix: Long-Distance Pulsed ESR Measurements using Vesicles, Bicelles, and Rodlike Micelles. J. Am. Chem. Soc. 2008, 130, 12856−12857. (7) Poget, S. F.; Girvin, M. E. Solution NMR of Membrane Proteins in Bilayer Mimics: Small is Beautiful, but Sometimes Bigger is Better. Biochim. Biophys. Acta, Biomembr. 2007, 1768, 3098−3106. (8) Jastrzebska, B.; Maeda, T.; Zhu, L.; Fotiadis, D.; Filipek, S.; Engel, A.; Stenkamp, R. E.; Palczewski, K. Functional Characterization of Rhodopsin Monomers and Dimers in Detergents. J. Biol. Chem. 2004, 279, 54663−54675. (9) Chou, J. J.; Kaufman, J. D.; Stahl, S. J.; Wingfield, P. T.; Bax, A. Micelle-Induced Curvature in a Water-Insoluble HIV-1 Env Peptide Revealed by NMR Dipolar Coupling Measurement in Stretched Polyacrylamide Gel. J. Am. Chem. Soc. 2002, 124, 2450−2451. (10) Sanders, C. R.; Schwonek, J. P. Characterization of Magnetically Orientable Bilayers in Mixtures of Dihexanoylphosphatidylcholine and Dimyristoylphosphatidylcholine by Solid-State NMR. Biochemistry 1992, 31, 8898−8905. (11) Bayburt, T. H.; Grinkova, Y. V.; Sligar, S. G. Self-Assembly of Discoidal Phospholipid Bilayer Nanoparticles with Membrane Scaffold Proteins. Nano Lett. 2002, 2, 853−856. 15841


Article

The Journal of Physical Chemistry B (33) Debnath, A.; Schäfer, L. V. Structure and Dynamics of Phospholipid Nanodiscs from all-Atom and Coarse-Grained Simulations. J. Phys. Chem. B 2015, 119, 6991−7002. (34) Jiang, Y.; Wang, H.; Kindt, J. T. Atomistic Simulations of Bicelle Mixtures. Biophys. J. 2010, 98, 2895−2903. (35) Wang, H.; de Joannis, J.; Jiang, Y.; Gaulding, J. C.; Albrecht, B.; Yin, F.; Khanna, K.; Kindt, J. T. Bilayer Edge and Curvature Effects on Partitioning of Lipids by Tail Length: Atomistic Simulations. Biophys. J. 2008, 95, 2647−2657. (36) Jiang, Y.; Kindt, J. T. Simulations of Edge Behavior in a MixedLipid Bilayer: Fluctuation Analysis. J. Chem. Phys. 2007, 126, 045105. (37) Gautier, A.; Mott, H. R.; Bostock, M. J.; Kirkpatrick, J. P.; Nietlispach, D. Structure Determination of the Seven-Helix Transmembrane Receptor Sensory Rhodopsin II by Solution NMR Spectroscopy. Nat. Struct. Mol. Biol. 2010, 17, 768−774. (38) Bayburt, T. H.; Grinkova, Y. V.; Sligar, S. G. Assembly of Single Bacteriorhodopsin Trimers in Bilayer Nanodiscs. Arch. Biochem. Biophys. 2006, 450, 215−222. (39) Lind, J.; Nordin, J.; Mäler, L. Lipid Dynamics in Fast-Tumbling Bicelles with Varying Bilayer Thickness: Effect of Model Transmembrane Peptides. Biochim. Biophys. Acta, Biomembr. 2008, 1778, 2526−2534. (40) Hess, B.; Kutzner, C.; van der Spoel, D.; Lindahl, E. GROMACS 4: Algorithms for Highly Efficient, Load-Balanced, and Scalable Molecular Simulation. J. Chem. Theory Comput. 2008, 4, 435−447. (41) Monticelli, L.; Kandasamy, S. K.; Periole, X.; Larson, R. G.; Tieleman, D. P.; Marrink, S. The MARTINI Coarse-Grained Force Field: Extension to Proteins. J. Chem. Theory Comput. 2008, 4, 819− 834. (42) Velinova, M.; Sengupta, D.; Tadjer, A. V.; Marrink, S. Sphere-toRod Transitions of Nonionic Surfactant Micelles in Aqueous Solution Modeled by Molecular Dynamics Simulations. Langmuir 2011, 27, 14071−14077. (43) Periole, X.; Huber, T.; Marrink, S.; Sakmar, T. P. G ProteinCoupled Receptors Self-Assemble in Dynamics Simulations of Model Bilayers. J. Am. Chem. Soc. 2007, 129, 10126−10132. (44) Periole, X.; Knepp, A. M.; Sakmar, T. P.; Marrink, S. J.; Huber, T. Structural Determinants of the Supramolecular Organization of G Protein-Coupled Receptors in Bilayers. J. Am. Chem. Soc. 2012, 134, 10959−10965. (45) Yesylevskyy, S. O.; Schäfer, L. V.; Sengupta, D.; Marrink, S. J. Polarizable Water Model for the Coarse-Grained MARTINI Force Field. PLoS Comput. Biol. 2010, 6, e1000810. (46) Kraft, J. F.; Vestergaard, M.; Schiøtt, B.; Thøgersen, L. Modeling the Self-Assembly and Stability of DHPC Micelles using Atomic Resolution and Coarse Grained MD Simulations. J. Chem. Theory Comput. 2012, 8, 1556−1569. (47) Berendsen, H. J. C.; Postma, J. P.; van Gunsteren, W. F.; DiNola, A.; Haak, J. R. Molecular Dynamics with Coupling to an External Bath. J. Chem. Phys. 1984, 81, 3684−3690. (48) Ye, W.; Lind, J.; Eriksson, J.; Mäler, L. Characterization of the Morphology of Fast-Tumbling Bicelles with Varying Composition. Langmuir 2014, 30, 5488−5496. (49) Luchette, P. A.; Vetman, T. N.; Prosser, R. S.; Hancock, R. E. W.; Nieh, M.; Glinka, C. J.; Krueger, S.; Katsaras, J. Morphology of Fast-Tumbling Bicelles: A Small Angle Neutron Scattering and NMR Study. Biochim. Biophys. Acta, Biomembr. 2001, 1513, 83−94. (50) Andersson, A.; Mäler, L. Size and Shape of Fast-Tumbling Bicelles as Determined by Translational Diffusion. Langmuir 2006, 22, 2447−2449. (51) Kučerka, N.; Liu, Y.; Chu, N.; Petrache, H. I.; Tristram-Nagle, S.; Nagle, J. F. Structure of Fully Hydrated Fluid Phase DMPC and DLPC Lipid Bilayers using X-Ray Scattering from Oriented Multilamellar Arrays and from Unilamellar Vesicles. Biophys. J. 2005, 88, 2626−2637. (52) Hansen, S. K.; Vestergaard, M.; Thøgersen, L.; Schiøtt, B.; Nielsen, N. C.; Vosegaard, T. Lipid Dynamics Studied by Calculation of 31P Solid-State NMR Spectra using Ensembles from Molecular Dynamics Simulations. J. Phys. Chem. B 2014, 118, 5119−5129.

(53) Beaugrand, M.; Arnold, A. A.; Hénin, J.; Warschawski, D. E.; Williamson, P. T. F.; Marcotte, I. Lipid Concentration and Molar Ratio Boundaries for the use of Isotropic Bicelles. Langmuir 2014, 30, 6162− 6170. (54) Arnold, A.; Labrot, T.; Oda, R.; Dufourc, E. J. Cation Modulation of Bicelle Size and Magnetic Alignment as Revealed by Solid-State NMR and Electron Microscopy. Biophys. J. 2002, 83, 2667−2680. (55) Ottiger, M.; Bax, A. Characterization of Magnetically Oriented Phospholipid Micelles for Measurement of Dipolar Couplings in Macromolecules. J. Biomol. NMR 1998, 12, 361−372. (56) van Dam, L.; Karlsson, G.; Edwards, K. Morphology of Magnetically Aligning DMPC/DHPC Aggregates - Perforated Sheets, Not Disks. Langmuir 2006, 22, 3280−3285. (57) Gaemers, S.; Bax, A. Morphology of Three Lyotropic Liquid Crystalline Biological NMR Media Studied by Translational Diffusion Anisotropy. J. Am. Chem. Soc. 2001, 123, 12343−12352. (58) Triba, M. N.; Warschawski, D. E.; Devaux, P. F. Reinvestigation by Phosphorus NMR of Lipid Distribution in Bicelles. Biophys. J. 2005, 88, 1887−1901. (59) Wu, H.; Su, K.; Guan, X.; Sublette, M. E.; Stark, R. E. Assessing the Size, Stability, and Utility of Isotropically Tumbling Bicelle Systems for Structural Biology. Biochim. Biophys. Acta, Biomembr. 2010, 1798, 482−488. (60) Fanucci, G. E.; Lee, J. Y.; Cafiso, D. S. Membrane Mimetic Environments Alter the Conformation of the Outer Membrane Protein BtuB. J. Am. Chem. Soc. 2003, 125, 13932−13933. (61) McKibbin, C.; Farmer, N. A.; Jeans, C.; Reeves, P. J.; Khorana, H. G.; Wallace, B. A.; Edwards, P. C.; Villa, C.; Booth, P. J. Opsin Stability and Folding: Modulation by Phospholipid Bicelles. J. Mol. Biol. 2007, 374, 1319−1332. (62) Sanders, C. R.; Landis, G. C. Reconstitution of Membrane Proteins into Lipid-Rich Bilayered Mixed Micelles for NMR Studies. Biochemistry 1995, 34, 4030−4040. (63) Nakano, M.; Fukuda, M.; Kudo, T.; Miyazaki, M.; Wada, Y.; Matsuzaki, N.; Endo, H.; Handa, T. Static and Dynamic Properties of Phospholipid Bilayer Nanodiscs. J. Am. Chem. Soc. 2009, 131, 8308− 8312. (64) Miyazaki, M.; Nakano, M.; Fukuda, M.; Handa, T. Smaller Discoidal High-Density Lipoprotein Particles Form Saddle Surfaces, but Not Planar Bilayers. Biochemistry 2009, 48, 7756−7763. (65) Bodor, A.; Kövér, K. E.; Mäler, L. Membrane Interactions in Small Fast-Tumbling Bicelles as Studied by 31P NMR. Biochim. Biophys. Acta, Biomembr. 2015, 1848, 760−766. (66) Triba, M. N.; Zoonens, M.; Popot, J.; Devaux, P. F.; Warschawski, D. E. Reconstitution and Alignment by a Magnetic Field of a β-Barrel Membrane Protein in Bicelles. Eur. Biophys. J. 2006, 35, 268−275. (67) Bertelsen, K.; Dorosz, J.; Hansen, S. K.; Nielsen, N. C.; Vosegaard, T. Mechanisms of Peptide-Induced Pore Formation in Lipid Bilayers Investigated by Oriented 31P Solid-State NMR Spectroscopy. PLoS One 2012, 7, e47745. (68) de Planque, M. R.; Kruijtzer, J. A.; Liskamp, R. M.; Marsh, D.; Greathouse, D. V.; Koeppe, R. E., II; de Kruijff, B.; Killian, J. A. Different Membrane Anchoring Positions of Tryptophan and Lysine in Synthetic Transmembrane α-Helical Peptides. J. Biol. Chem. 1999, 274, 20839−20846. (69) Doux, J. P. F.; Hall, B. A.; Killian, J. A. How Lipid Headgroups Sense the Membrane Environment: An Application of 14N NMR. Biophys. J. 2012, 103, 1245−1253. (70) Chang, Y.; Bowie, J. U. Measuring Membrane Protein Stability Under Native Conditions. Proc. Natl. Acad. Sci. U. S. A. 2014, 111, 219−224. (71) Andersen, O. S.; Koeppe, R. E. Bilayer Thickness and Membrane Protein Function: An Energetic Perspective. Annu. Rev. Biophys. Biomol. Struct. 2007, 36, 107−130. (72) Sparr, E.; Ash, W. L.; Nazarov, P. V.; Rijkers, D. T. S.; Hemminga, M. A.; Tieleman, D. P.; Killian, J. A. Self-Association of Transmembrane α-Helices in Model Membranes: Importance of Helix 15842


Article

The Journal of Physical Chemistry B Orientation and Role of Hydrophobic Mismatch. J. Biol. Chem. 2005, 280, 39324−39331. (73) Fotiadis, D.; Liang, Y.; Filipek, S.; Saperstein, D. A.; Engel, A.; Palczewski, K. The G Protein-Coupled Receptor Rhodopsin in the Native Membrane. FEBS Lett. 2004, 564, 281−288. (74) Grouleff, J.; Irudayam, S. J.; Skeby, K. K.; Schiøtt, B. The Influence of Cholesterol on Membrane Protein Structure, Function, and Dynamics Studied by Molecular Dynamics Simulations. Biochim. Biophys. Acta, Biomembr. 2015, 1848, 1783−1795. (75) Lee, A. G. Lipid-Protein Interactions. Biochem. Soc. Trans. 2011, 39, 761−766. (76) Eggensperger, S.; Fisette, O.; Parcej, D.; Schafer, L. V.; Tampe, R. An Annular Lipid Belt is Essential for Allosteric Coupling and Viral Inhibition of the Antigen Translocation Complex TAP (Transporter Associated with Antigen Processing). J. Biol. Chem. 2014, 289, 33098− 33108.

15843


Bicelles and Other Membrane Mimics: Comparison ... - ACS Publications

Recommend Documents