Article

pubs.acs.org/JPCB

Extension of the Slipids Force Field to Polyunsaturated Lipids Inna Ermilova and Alexander P. Lyubartsev*

Department of Materials and Environmental Chemistry, Stockholm University, SE 106 91 Stockholm, Sweden

*S Supporting Information

ABSTRACT: The all-atomic force field Slipids (Stockholm Lipids) for lipid bilayers simulations has been extended to polyunsaturated lipids. Following the strategy adopted in the development of previous versions of the Slipids force field, the parametrization was essentially based on high-level ab initio calculations. Atomic charges and torsion angles related to polyunsaturated lipid tails were parametrized using structures of dienes molecules. The new parameters of the force field were validated in simulations of bilayers composed of seven polyunsaturated lipids. An overall good agreement was found with available experimental data on the areas per lipids, volumetric properties of bilayers, deuterium order parameters, and scattering form factors. Furthermore, simulations of bilayers consisting of highly polyunsaturated lipids and cholesterol molecules have been carried out. The majority of cholesterol molecules were found in a position parallel to bilayer normal with the hydroxyl group directed to the membrane surface, while a small fraction of cholesterol was found in the bilayer center parallel to the membrane plane. Furthermore, a tendency of cholesterol molecules to form chain-like clusters in polyunsaturated bilayers was qualitatively observed.

■ INTRODUCTION docosahexaenoic acid is not only affecting physical-chemical properties of membranes (such as thickness, lipid packing, Lipids are important components of biological systems, such as fl biomembranes.1,2 A lipid bilayer forming a membrane helps to uidity, elasticity, permeability, etc.), but it is activating the control the transport of different molecules into or out of a important signaling protein kinase C (PKC) as well. The 3,4 activity of this protein was higher in a failed human heart cell. Furthermore, the membrane itself is involved in many 38 important processes, such as cell division, cell signaling, protein compared to a healthy one according to Bowling et al. The fact that PUFA lipids affect so wide variety of biological anchoring, cell fusion, which are happening on a microscopic fi level.5,6 Lipids composed of polyunsaturated fatty acids processes is evidently linked to their speci c chemical structure with methylene-interrupted cis double bonds, which causes (PUFAs) are very common in mammalian cells including fi 39 humans.7,8 In particular, they are highly abundant in the eye their speci c functioning in living organisms. Molecular dynamics (MD) simulations can provide valuable retina and in the brain gray and white matter as one of the side fi chains of phosphatidylethanolamine (PE) and phosphatidylser- information for understanding of speci c features of PUFA ine (PS).7 In addition, linoelic acid is a dominant substituent of lipids, and through the years a number of works on atomistic 7 fl simulations of bilayers composed of PUFA lipids have been cardiolipin, which is found in the heart and in the inner lea et 40−48 of mitochondrial membranes. The amount of PUFAs in carried out. Still, the amount of modeling studies on different cells plays a big role in the human’s health. Some PUFA is much less than what is available for fully saturated nutritionists suggest consumption of products which are rich lipids. Partially this can be related to the lack of reliable with ω-3 and ω-6 fatty acids in order to avoid issues with the parametrization of the interactions in the polyunsaturated health9 (the terms ω-3 and ω-6 mean polyunsaturated fatty hydrocarbon chains. Reliability of results obtained by MD fi  simulations depends crucially on the force field (FF) used. In acids which have the rst double bond C C at the third or fi sixth carbon atom from the end of the chain, respectively). It order to get reliable results the force eld should accurately was also discovered that PUFAs are useful for the treatment of describe all interactions in the system of interest. − mental diseases and memory problems,10 19 heart diseases and For lipid bilayers, there is a relatively large variety of force − fi problems with a high blood pressure,20 25 atherosclerosis, elds. Often MD simulations of lipid bilayers are carried out fl 26 27−30 31 using united atom Gromos FF and its variations: Berger dyslipidemia, in ammation, cancer, Parkinson, prob- 49 50 lems with the vision,32 diabetes, and fatty liver disease.33,34 The lipids, 43A1-S3 FF parameter set, or G53A6L parameter 51 fi significant positive effects of the consumption of PUFAs were set. Use of united-atom force elds is less suitable for discovered on children’s performance inlearning.35 At the same modeling of unsaturated lipids because of a weak polarity at the time some harmful effects of the consumption of PUFAs were discovered: pro-hemorrhagic effect, negative effects on immune Received: May 30, 2016 functions, formation of oxidative products by ω-3 polyunsatu- Revised: November 30, 2016 rated fatty acids.36 Stillwell et al.37 have shown that Published: November 30, 2016

© 2016 American Chemical Society 12826 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article double bonds which is not included into united atom FFs ■ METHODS AND MODELS having zero charges of united atom groups. There are also all- Parametrization Strategy. In this work we have focused atomistic FFs with all hydrogens explicitly included, which were on the polyunsaturated tails of lipid molecules having repeated developed for modeling of lipids: several versions of − methylene-separated cis double bonds. Parameters for the lipid CHARMM,52 55 AMBER family (GAFF, lipid11, − − headgroups and saturated parts of the lipid tails are kept as in 56 59 60 61 63 − lipid14), GLYCAM06, Slipids, see also recent the previous versions of the Slipids FF,61 63 in which bond, 64 fi review and references therein. All these force elds contain angular, Urey−Bradley, and a part of Lennard-Jones parameters  parameters for a double C C cis bond, however they were were inherited from the Charmm36 FF.55 derived and validated for an isolated double bond which, for This FF has a standard molecular mechanics form: example, is present in such lipids as DOPC or POPC. However, application of parameters derived for a single double bond to Uff=+++++EEEEEE bonds angles U−‐ B tors L J electrostatic polyunsaturated lipids having several methylene-separated cis (1) double bonds often leads to inaccurate descriptions of the Particularly in this work we reparametrized electrostatic surface area per lipid, deuterium order parameters, and X-ray E and torsional E parts of the FF which have 65 electrostaic tors form factors. To our knowledge, the only atomistic force field strongest effect on the conformational behavior of lipids and which includes parameters specially developed for PUFA chains their interaction with other molecules. The electrostatic fi 65 is a recent modi cation of Charmm36 which was made by interactions are determined by the partial charges qi of atoms: adjustment of the torsion potential between two methylene- qqij separated cis double bonds. Apart from that, there exists a E = ∑ united atom model based on Berger force field66 and a coarse- electrostatic 4πϵ r i,j 0ij (2) grained model based on MARTINI force field67 which include special parameters for PUFA chains. while the torsional potential is presented by a cosine expansion In this work we extend the Slipids force field to up to the 6-th term: polyunsaturated lipids. The original all-atomistic Slipids 6 61−63 FF was derived for saturated and monounsaturated lipids Ek=+−∑ (1 cos( nϕδ )) ff tors nn with a number of di erent headgroups and validated by n=1 (3) comparisons of computed properties of lipid bilayers with available experimental data: average areas per lipid, bilayer The Lennard-Jones and electrostatic interactions are not thickness, X-ray and neutron scattering form factors, NMR computed for atoms pairs connected by 1 or 2 covalent fi bonds. For atom pairs connected by exactly 3 bonds (the so- deuterium order parameters. Slipids force eld demonstrates − also a good description of the temperature dependencies of called 1 4 neighbors), special rules are applied. In the case of Slipids FF, 1−4 electrostatic interactions are scaled by factor experimentally observed bilayer properties including transition fi to the gel phase, and partitioning of various molecular 0.83, while special rules (de ned by the FF atom types) are − − 61 inclusions across the bilayer.68 70 In the extension of Slipids applied for 1 4 Lennard-Jones interactions. to polyunsaturated lipids we adopt the methodology used in the To optimize FF parameters for atom types related to − previous works61 63 based on high-level ab initio computations repeated cis double bonds we used two model molecules: cis- of partial charges with multiconfigurational averaging, and 3,cis-6-nonadiene and cis-3,cis-6-dodecadiene (Figure S1 and S2 subsequent parametrization of the torsion potentials. of the Supporting Information). The parametrization algorithm The developed force field is validated by simulations of involves several steps: several phosphatidylcholine (PC) lipids containing different 1. We run a MD simulation of model molecules in a liquid numbers of double bonds in unsaturated tails: 18:1(n−9), phase and extract from it randomly 50 molecular 18:2(n−9), 18:3(n−3), 20:4(n−6), 20:5(n−3), 22:5(n−6), conformations. 22:6(n−3), and comparisons with available experimental data. 2. For the chosen conformations we compute partial atomic Here and below we use numerical notations for the charges and for each atom type take average over the phospholipid chains in the form N:k(n−j) where N denotes conformations. the total number of carbon atoms in the chains, k is the number 3. We check suitability of the Lennard-Jones parameters by of double bonds, and j is the position of the first double bond calculating thermodynamic properties of the liquid phase counted from the methyl terminus of the chain, cis- of model molecules and optimize them if necessary conformation of double bonds is implied in all cases. Through 4. After that we optimize the torsional potentials by ab the text, notation (n−j) in the descriptions of lipids is often initio computations of energy while rotating the torsional omitted for brevity. An example of 18:0−22:6(n−3) PC angle. (phosphatidylcholine) lipid is shown in Figure 1. Furthermore, The procedure can then be reiterated with MD simulations the behavior of bilayers containing highly unsaturated 20:4− of steps (1)−(4) run with the optimized parameters. From the 20:4 PC and 22:6−22:6 PC lipids in mixture with cholesterol is experience of previous works61,62 it follows that self-consistency studied and discussed in relation to experimental observation of is reached already after the first iteration, so in this work only behavior of cholesterol in PUFA bilayers. one iteration was performed. After optimization of the parameters, validation simulations are run for a number of bilayers and computed properties are compared with available experimental data for the respective systems. Computations of Partial Charges. The initial MD simulations of each of the two dienes were performed using the original Slipids FF61,62 at temperature 30 °C, in which Figure 1. Structure of a 18:0−22:6(n−3) PC lipid. parameters for monounsaturated chains were used to describe

12827 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

Figure 2. Derived partial charges for unsaturated lipid tails (charges of hydrogens are given in parentheses): (a) = −0.27 (0.05), (b) = −0.081 (0.016), (c) = 0.047 (−0.007), (d) = −0.28 (0.14), (f) = 0.28 (0.00), (g) = 0.26 (0.00), (h) = −0.06 (0.03), (i) = 0.12 (0.01), (j) = 0.00 (0.00). the polyunsaturated part of the molecules. After 1 ns (−0.2/0.11)61 which were derived using the same method- equilibration, 50 molecular conformations were randomly ology. Stronger polarity of PUFA chains, which follows from chosen from the simulated trajectory. For each selected our quantum-chemical calculations, may result in a weaker molecular conformation, density functional theory (DFT) hydrophobicity of the interior of PUFA bilayers and affect single point calculations were carried out with the purpose to partitioning of various molecules in the bilayers. extract partial atom charges. In these calculations we used the After optimization of the partial atom charges, MD − B3LYP exchange-correlation functional71 73 with cc-pVTZ simulations of several dienes were carried out in the NPT basis set.74 As in the previous works we assumed that the ensemble at 303 K for 10 ns and the computed density was molecules were placed in a polarizable continuum with compared to the experimental ones.79 New recomputed partial IEFPCM75,76 as a solvent model with ϵ =2.04which charges (shown in Figure 2) and torsion potentials from ref 62 corresponds to the typical dielectric permittivity of a hydro- were used in these simulations. The density computations for phobic media.61 For calculations of partial charges we used the several isomers of nonadiene such as 1,8-nonadiene (0.74 g/ restrained electrostatic potential (RESP) approach77 as cm3 at 25 °C), 2-butyl-1,4-pentadiene (0.747 g/cm3 at 25 °C, implemented in the R.E.D. software,78 in which the difference 2-(1,1-dimethylethyl)-1,3-pentadiene (0.751 g/cm3 at 25 °C), between the electrostatic potential determined from the 2,4-dimethyl-1,5-heptadiene (0.746 g/cm3 at 25 °C) showed electron density and the classical point charges was minimized, agreement with experimental data79 within 1% in all cases. For while maintaining the total sum of squares of atomic charges cis-3,cis-6-dodecadiene we compared simulated density (0.777 possibly low. The data on computed atomic charges for all g/cm3 at 30 °C) with experimental density of isomers cis-5,cis- molecular structures and their conformational averaging are 7-dodecadiene (0.777 g/cm3)[http://www.guidechem.com/ presented in the Supporting Information, Tables S1−S3. dictionary/en/6108-62-9.html, accessed October 18, 2016], The computed average values of the atomic charges were and cis-5,cis-7-dodecadiene (0.767 g/cm3)[http://www. further averaged over symmetrical atoms and were used as a guidechem.com/dictionary/en/5876-87-9.html, accessed Octo- guide, taking also into account partial charges of the previous ber 18, 2016]. We however did not find data on evaporation version of the Slipid force field,62 in the final assignment of the enthalpy for these molecules. Given good results for the partial charges for typical PUFA chains as shown in Figure 2.It density, we did not proceed with an update of Lennard-Jones is remarkable that charges for atoms at double bonds turned parameters and kept them as in the original Slipids FF. out to be larger for PUFA chains (−0.28/0.14) compared to Parametrization of Torsion Potentials. The final step in charges for monounsaturated lipids in the Slipids force field the FF parametrization was derivation of parameters for the

12828 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article α − ° − ° torsional potentials by high-quality ab initio computations. adjacent angle 2 deviates from 110 120 to 80 90 (Figure  − − fi Dihedrals of our interest were: CH CH CH2 CH (FF types S4). These results show that xing of other degrees of freedom CEL1 = CEL1-CTL2-CEL1 denoted in this work as α) and while rotating the torsional angle of interest does not bring  − − CH CH CH2 CH2 (FF types CEL1 = CEL1-CTL2-CTL2, major deviations from the optimized transition path. denoted as β, see Figure 3). The torsion potential was fitted by the form of eq 3 to the energies obtained by

Etorsional =

()()EEEECCSD(TT )/BBS−− MD CCSD( )/BBS − MD min (5)

Figure 3. Parametrized dihedrals α (CEL1 = CEL1-CTL2-CEL1) and where EMD is the value of the potential energy obtained by a β (CEL1 = CEL1-CTL2-CTL2) and associated force field atom types. MD software with zero parameters for the considered torsion angle, ECCSD(T)/BBS is the energy computed according to eq 4, − For accurate computations of the relative quantum- and (ECCSD(T)/BBS EMD)min is the minimal value of the ff mechanical energies between different conformations, the di erence between results from quantum chemical computa- hybrid method for interaction energies (HM-IE)80 was used tions and MD providing zero value of the torsional potential which is described by the following formula: energy in the minimum. The optimized parameters for two types of dihedrals are fi EEECCSD(TTT )/BBS=+ CCSD( )/SBS CCSD( )/BBS given in Table 1. Figure 4 shows tting of the torsion potential

−≈+−EEEECCSD(TT )/SBS CCSD( )/SBS MP2/BBS MP2/SBS Table 1. Parameters for Dihedrals CEL1 = CEL1-CTL2- (4) CEL1 (α) and CEL1 = CEL1-CTL2-CTL2 (β) where notations CCSD(T), MP2 refer to the ab initio method nα δα, degrees kα. kJ/mol nβ δβ, degrees kβ, kJ/mol used, and SBS and BBS mean “small” and “big” basis set, respectively. The gain of using eq 4 is that precision of this 1 0.00 4.331 1 0.00 3.967 81 2 0.00 5.161 2 0.00 3.89 method is as high as coupled clusters CCSD(T) theory with a − − big basis set while the computations involve either a smaller 3 0.00 0.415 3 0.00 0.138 basis set or less expensive method (MP2). Here we used cc- 4 0.00 3.135 4 0.00 1.892 pVDZ basis set for SBS and cc-pVQZ74 for BBS. 5 0.00 1.59 5 0.00 0.421 Previously, Klauda et al.65 considered the two-dimensional 6 0.00 0.137 6 0.00 0.421 energy surface for two neighboring dihedrals between double bonds, computed by high-level ab initio computations for 2,5- computed with the optimized FF parameters to the points heptadiene. They found that a direct fit to this quantum- obtained from ab initio computations resulting from eq 5.In mechanical energy surface resulted in too high chain order, order to make an accurate approximation for the dihedrals especially for SDPC bilayer, and in the final expression for the potential energy, each function was presented by 6 terms in eq torsion term they increased the well breadth and reduced the 3. The phase δ was set equal to 0° for all terms (note that in transitional barrier between the minima, to get better order to satisfy the symmetry condition, the phase can be either agreement with NMR order parameters. In our work, we 0° or 180°, but these two values can be interchanged by the ffi have followed the approach adopted in derivation of torsion simultaneous change of sign of the coe cient kn). Furthermore, parameters for Slipids FF for saturated61 and monounsatu- during the fitting we excluded the range of torsion angles which rated62 chains, using one-dimensional rotation around torsion are close to cis-conformations between −50° and 50° for angle of interest. First, the geometries of model dienes (cis-3,cis- dihedral α and between −60° and 60° for β, which showed 6-nonadiene and cis-3,cis-6-dodecadiene) were optimized using quantum energies exceeding values of the energy in the the second order Møller−Plesset perturbation theory (MP2)82 minimum by more than 20 kJ/mol. The reason for this with correlation-consistent cc-pVDZ basis set,74 and then exclusion was that such torsional angles, together with cis- molecular conformations were rotated around the torsion angle conformation of the double bond, will always lead to too close of interest with the step 10° keeping other degrees of freedom contacts (or even overlapping) of some atoms, and too high frozen. values of the potential energy. Fitting of the energy surface in In order to ensure that fixing of the other degrees of freedom the high-energy region may deteriorate fitting in the most while varying torsion angle does not bring major artifacts, we populated low energy area, which is much more important for have carried out additional series of computations in which, for the correct reproduction of the conformational behavior. each fixed value of the torsion angle of interest, we performed We have computed distributions of dihedrals α and β for the geometry optimization over all other degrees of freedom of the model molecules by MD simulations at T = 303 K, the results molecule. Results for the quantum energies computed for the are displayed in Figure 5. Two probability maxima are seen at fixed and optimized conformations obtained at different values values of dihedrals at around −120° and 120° for both types of α of torsional angle 1 are shown in Figure S3 of the Supporting dihedrals. Similar results for the distribution of dihedrals in Information. The transition barrier over the trans-conformation polyunsaturated chains were achieved in works of Saiz and (180°) shows only a marginal decrease by 0.5 kJ/mol when Klein83 and Klauda et al.65 Figure 5 shows that dihedrals which α optimization of molecular structure at each value of 1 angle were excluded from the optimization of the potential for was made. Somewhat larger difference in energy between the torsional angles (around cis-conformation) never appear during fixed and optimized conformations (within 2 kJ/mol) is seen in simulations. The region of conformations around 180° is α − ° α the range of 1 angle 110 140 , and just in this range of 1 the weakly populated showing that these dihedrals can change

12829 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

Figure 4. Fitting of the potentials for dihedrals α (left) and β (right).

α α β Figure 5. Distribution of dihedrals: 1 is dihedral CEL1 = CEL1-CTL2-CEL1, 2 is dihedral CEL1-CTL2-CEL1 = CEL1, 1 is dihedral CEL1 = β CEL1-CTL2-CTL2, 2 is dihedral CTL2-CTL2-CEL1 = CEL1, as shown in the Supporting Information, Figures S1 and S2. conformations from “right” to “left” and vise versa on the time Table 2. Simulated Systems Details scale of the simulations. time, Simulation Details. All molecular dynamics simulations lipid, PC name N N N T (K) μs were done with Gromacs-4.6.784 software using the leapfrog lip w chol 18:0−18:1 SOPC 128 5120 0 303 0.4 integrator with the time step 2 fs. Covalent bonds were (n−9) 85 constrained by the LINCS algorithm with exception for water 16:0−18:2 PiLPC 128 3840 0 303, 293, 0.4 for which bonds were constrained by the SETTLE method.86 (n−9) 303, 313 Model molecules simulations were carried out in a cubic 16:0−22:6 PDPC 128 3840 0 303 0.4 − periodic box containing 340 molecules in the isotropic NPT (n 3) 18:0−18:3 128 3840 0 303 0.4 ensemble at 1 bar pressure. Bilayer simulations were carried out (n−3) in the NPT ensemble with semianizotropic Berendsen − 87 18:0 20:4 SAPC 128 3840 0 303 0.4 barostat keeping the constant pressure at 1 bar separately (n−6) in Z-direction and in the XY-plane. The barostat coupling 18:0−20:5 128 3840 0 303 0.4 constant was set to 10.0 ps and the isothermal compressibility (n−3) 18:0−22:5 128 3840 0 303 0.4 was equal to 0.000045 1/bar. The temperature was kept − constant using V-Rescale thermostat88 with coupling constant (n 6) 89 18:0−22:6 SDPC 128 3840 0 303 0.4 0.5 ps. The particle mesh Ewald scheme was used for (n−3) treatment of the long-range electrostatic interactions with 20:4(n−6)- DAPC 128 3840 0 293 0.4 Fourier spacing of 0.12 nm. The cutoff for Lennard-Jones 20:4(n−6) interactions was set to 1.4 nm and the isotropic long-range 22:6(n−3)- DDPC 128 3840 0 293 0.4 corrections were applied to energy and pressure. 22:6(n−3) The lipid bilayer simulation were carried out with 128 lipids 20:4(n−6)- DAPC− 200 10000 100 293 2.4 20:4(n−6) CHOL (64 lipids per a leaflet) and 30 TIP3P water molecules90 per 22:6(n−3)- DDPC− 200 10000 100 293 2.4 one lipid except simulation of the 18:0−18:1 PC bilayer which 22:6(n−3) CHOL was carried out with 40 water molecules per lipid (the latter simulation repeated the one from previous publication62 but with the updated torsion potential for the β angle). Full account bilayer system was simulated during 400 ns of which initial 120 of the simulated bilayer systems is given in Table 2. Each pure ns were considered as equilibration. Analyses of the simulated

12830 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article trajectories were done with MDynaMix software package91 and electron density profiles.97,100 We collected simulated and TRAJMAN in-house software (a software for analyzing available experimental data for 18:0−18:1 PC, 16:0−22:6 PC, trajectories, written by Jon Kapla and available at https:// 18:0−22:5 PC, and 18:0−22:6 PC lipids in Table 4. All github.com/kaplajon/trajman). Statistical uncertainties were volumetric data for the simulated bilayers were found in calculated from the variance of the subaverages computed agreement with experimental results within computational and from 10 consecutive fragments of the production part of the experimental uncertainty, in cases when such data were trajectories. available. All quantum chemical computations were carried out using Deuterium Order Parameters. A profile of deuterium Gaussian09 package92 order parameters measurable in NMR experiments is a property widely used in validation of force fields for lipid bilayer ■ RESULTS AND DISCUSSION simulation. The order parameter for a specific CH (or CD) Volumetric Parameters of the Bilayers. Average area per bond is determined by lipid is a commonly used fast test to evaluate quality of the fi ff 1 2 force eld. The problem of such validation is that di erent SCD =⟨3cosθ −⟩ 1 experimental methods use different model assumptions which 2 (6) may lead for a large spread of experimental average areas for the where θ is the angle between direction of the bond and the same lipid. For example for 18:0−18:1 PC experimental values bilayer normal. of the average area per lipid were found in different works in We have computed order parameters for saturated and 64 the range between 61.4 and 71 Å2. Neutron and X-ray unsaturated lipid tails of the studied bilayers using eq 6. Figure scattering data are considered as the most reliable sources of 7 shows comparison of the results for 18:0−22:6(n−3) PC, experimental determination of the average areas, however they 18:0−22:5(n−6) PC, and 16:0−22:6(n−3) PC obtained in our are not available for many lipid species. Still, due to its simulations with experimental data from several sour- simplicity a comparison of the simulated and experimental ces.40,101,103 For bilayers consisting of 18:0−18:3(n−9) PC, average lipid area becomes a necessary element of the analysis, 18:0−20:4(n−6) PC, and 18:0−20:5(n−3) PC simulated order aiming also to determine whether equilibration of the bilayer parameters are given in the Supporting Information, Figures was reached or not. S8−S10, together with experimental data for sn-1 chains from The average areas per lipids were computed for all bilayers Holte et al.104 Statistical uncertainty of the simulated order simulated in this work and they are given in Table 3 in parameters is always within 0.01. In all cases simulations show overall good agreement with experimental order parameters. In Table 3. Average Areas per Lipids (T = 303K) all simulated bilayers, order parameters for saturated sn-1 chains are about 0.2 in the beginning of the chain and gradually computed, lipid, PC Å2 experimental area/lipid, Å2 decrease toward the chain end. One can note slightly higher − 18:0−18:1 66.7 ± 0.5 67 (low angle X-ray scattering)93 order parameters in the beginning of sn-1 chain for 18:0 22:5 ± 94 PC than for other lipids both in simulation and experimental 65.5 1.3 (small angle neutron scattering) ff − ± fi 95 data. The di erence between order parameters for the sn-1 16:0 18:2 67.2 0.5 66 (Langmuir lm balance) − − 18:0−18:3 68.9 ± 0.6 66.6 (NMR)96 chain of the 18:0 22:5 PC and 18:0 22:6 PC bilayers, 18:0−20:4 69.6 ± 0.6 70.6 (NMR)96 displayed in Figure S11 of the Supporting Information, − ± 96 demonstrates remarkable similarity between computed and 18:0 20:5 70. 0.6 69.1 (NMR) 40 − ± ± 40 experimental data of Eldho et al., with reproduction of the 18:0 22:5 67.1 0.6 68.7 0.4 (X-ray) ff 16:0−22:6 68.8 ± 0.6 70. (Langmuir film balance)95 maximum of the di erence in the middle of the chain. − ± ± 40 For unsaturated tails the order parameters are generally 18:0 22:6 68.6 0.7 68.2 0.4 (X-ray) − − 71.6 (NMR)96 lower. For most carbons (except C4 and C5 in 18:0 22.5(n 6) PC), the difference between simulations and experimental results is within the same interval (about 0.02) as the difference comparison with experimental results. Evolution of the area between two experiments in case of 18:0−22:6(n−3) bilayer. during 400 ns of the simulated trajectory can be seen in Figure Simulated order parameters for C2 carbon of sn-2 chains show 6 and in Figures S5−S7 of the Supporting Information, where splitting for the two hydrogens, more noticeable for the most experimental lipid areas are given as horizontal lines. For some unsaturated PUFA (16:0−22:6, 18:0−22:6, 18:0−20:4, and lipids several sources of experimental data can be found and we 18:0−20:5). Splitting of deuterium order parameters at C2 cite such data in Table 3. A more complete collection of carbon was also observed in some experiments for other experimental and simulated (by other force fields) results on lipids.105 For 18:0−22:5(n−6) PC, simulations overestimate the average areas for unsaturated lipids can be found in the the order parameter for C4 and C5 carbons of the unsaturated recent review.64 chain by 0.05−0.07, but also reproduce the experimentally For all considered bilayers, deviations between experimental observed minimum of the order parameter at around 15-th and simulated lipid areas are within 2 Å2 (except an older NMR carbon and its increase toward the end of the chain. result for 18:0−22:6 PC bilayer where deviation is 3 Å2). In For 16:0−18:2(n−9) PC bilayer we have carried out some cases deviation is somewhat larger than simulated and simulations in the range of temperatures from 283 to 313 K experimental uncertainty given in specific studies, but the and compared computed order parameters with results of − differences are still within a typical experimental spread of data. Baenziger et all.106 108 Figure 8 illustrates that simulations Thus, we can conclude that the force field reproduces well the reproduce well the shape of the order parameter profile of the experimentally estimated average areas per lipid. unsaturated tail with the fall in the area of double bonds and Distances between headgroups (DHH), Luzatti (DB), and subsequent increase in the lower part of the tails, with the hydrophibic (DC) thicknesses can be determined from the exception of carbons 9 and 10, for which simulations do not

12831 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

Figure 6. Average areas per lipid, Å2. Red lines (comp) show evolution of the area during 400 ns production part of the simulations; straight green lines are experimental values from different works (for the references, see Table 3).

D D D Table 4. Distance between Headgroups HH, Hydrophobic C, and Luzzati B Thicknesses

lipid, PC DHHcomp,Å DHH exp,Å 2DCcomp,Å 2DC exp,Å DBcomp,Å DB exp,Å 18:0−18:1 38.5 ± 0.8 39.293 28.50 29.997 41.00 40.0097 16:0−22:6 37.5 ± 0.8 38.198,99 29.00 2899 44.00 no data 18:0−22:5 38.1 ± 0.8 37.9 ± 140 31.00 30.5 ± 140 46.00 48.5 ± 140 18:0−22:6 38.1 ± 1 37.9 ± 140 33.00 30.5 ± 140 46.00 48.5 ± 140 show the experimentally observed inversion, although they where ρ(z) is the electron density along z-coordinate follow the right trend (increase for C9, decrease for C10 with (computed in simulations taking into account partial charges ρ increasing temperature). It is also worth noting that for this of atoms), w is the electron density of bulk water and D is size lipid the order parameters of the unsaturated tail are rather of the simulation box in Z-direction, which corresponds to the insensitive to the temperature in the considered range which is D-spacing (the distance between parallelly oriented bilayers in different from the temperature behavior in saturated lipids in the sample) of bilayers in scattering experiments. − which order parameters noticeably decrease with increase of the The scattering factor for 18:0 18:1 PC lipid bilayer, temperature.61 computed using eq 7 with integration step 0.2 Å, is displayed All results of Figures 7 and 8 illustrate the well-known feature in Figure 9 together with experimental form factor from a paper ̌ 97 fi of order parameters profiles in unsaturated chains: the drop of by Kucerka et al. This result (red line in the gure) essentially 48,83,109−112 repeats the simulation for SOPC lipid with the previous version SCD at positions of the double bonds. It was shown 62 previously that the low order parameters in the area of double of Slipids FF (blue line), and taken together with results for bonds are mainly related to the specific orientations of the CH the average lipid area, volumetric properties, and order parameters for this lipid shows that update of the torsion bonds at the angle to the bilayer normal close to the magic potential for dihedral next to a double bond (β in Figure 3) angle of 54°,45,48 and such behavior was also observed in our does not affect the computed properties for monounsaturated simulations. lipids. X-ray Form Factors. X-ray scattering form factors are For polyunsaturated lipids, we found in the literature considered as the most direct test of the quality of the FF for scattering data only for 18:0−22:6 PC bilayer in the work by lipid bilayer simulation since experimental form factors are not 40 113 Eldho et el. We plot these data, together with our results subjected to model-based interpretations. In simulations, X- obtained by the Fourier transform of the electron density, in ray scattering factor can be straightforwardly computed by the 40 114 Figure 10. The experimental results combine data from three Fourier transform of the electron density: different experiments which were performed at three different

D/2 values of D-spacing. Experimental points are placed close to the simulated ones that give additional arguments that our model S(qzqzdz )=−∫ (ρρ ( )w )cos( ) −D/2 (7) provides realistic description of PUFA lipids.

12832 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

Figure 7. Deuterium order parameters for 18:0−22:5(n−6), 18:0−22:6(n−3), and 16:0−22:6 PCs. Experimental values (marked as “exp”) are taken from the following works: 18:0−22:5 PC (top),40 18:0−22:6 PC (middle), sn-1 chain,40 sn-2 chain, “exp1”,40 “exp2”;101 16:0−22:6 PC (bottom), sn- 1 chain, “exp1”102 (323 K; a plateau in the beginning of tail corresponds to unresolved values), “exp2”103 sn-2 chain,98 “comp” are order parameters obtained in the present simulations.

Reorientational Dynamics. Experimentally, dynamical where N is the number of protons bound to the considered ω ω properties of lipid bilayers can be probed by NMR relaxation carbon atom, C and H are resonance frequencies of C and H 13 − experiments. Particularly, C spin lattice relaxation time T1 nuclei, γ and γ are gyromagnetic ratios, respectively, and r can provide information on the re-orientational dynamics of C H CH is the length of the CH bond. CH-bonds, which happens on pico- and nanosecond time scale ω and which is also accessible by molecular dynamics In computer simulations, the spectral density J( ) can be 40 calculated from the re-orienational time correlation function simulations. T1 time is related to the spectral density function 40 ⟨ μ⃗ μ⃗ ⟩ μ⃗ J(ω): C(t)= P2( (0) (t)) of the CH unit vector (t)(P2 being the second Legendre polynomial): ⎛ ⎞2 11 h 1 =×−+++⎜γγ ⎟ ((JJJωω ) 3( ω ) 6( ωω )) ∞ NT 10⎝ CH 2π 3 ⎠ CH C CH 1 rCH JCttdt(ωω )= ∫ ( )cos( ) (8) 0 (9)

12833 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

Figure 8. Deuterium order parameters for 16:0−18:2(n−9) PC, sn-2 tail, for different temperatures. “comp” are values determined from the − simulations, ”exp” are values from works by Baenziger et al.106 108

Figure 9. Scattering form factors for 18:0−18:1 PC bilayer. comp (1): Figure 10. Computed (red line) and experimental X-ray scattering computed with the FF of this work; comp (2): computed with the form factors for 18:0−22:6 PC bilayer. The experimental points are previous version of Slipids FF;62 “exp”: experimental X-ray form taken from article40 for different D-spacings. D-spacing in the factor.97 simulations is 66.5 Å.

In the limit of motional narrowing (suggesting that τω ≪ 1, We have computed time correlation functions C(t) for all where τ is the motional correlation time and ω is the resonance carbons of unsaturated chain in the 18:0−22:5(n−6) PC and − − frequency), T1 can be computed directly from the decay 18:0 22:6(n 3) PC bilayers, by averaging over all lipid constant of the C(t)59,115 and become independent of the molecules and over all possible start times in the production magnetic field: part of the trajectory. As an example, time correlation functions − ∞ ⃗ ⃗ for C2 and C21 of 18:0 22:6 PC are displayed in Figure 11. 1 10− 2 40 =·(1.855 10 s )∫ ( ⟨PtCdt2 (μμ (0) ( )) ⟩−∞ ( )) They show similar behavior to C(t) obtained by Eldho et al. NT1 0 in the same time interval, with much slower decay of the C2 re- (10) orientation correlations on a nanosecond time scale compared where C(∞) is the limit of the time correlation function at to fast (of order tens of picoseconds) decay of C(t) for C21 infinite time (equal to the square of the order parameter). carbon.

12834 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

workers have performed a number of 31P NMR experiments on polyunsaturated phospholipids and found that at higher temperatures the presence of cholesterol could induce the destabilization of a bilayer while its absence could help the PC molecules to stay organized in the lamellar phase. Later in 1999 Brzustovicz et al.120 carried out a solid state 2H NMR investigation on polyunsaturated membranes containing 20:4−20:4 PC and cholesterol and discovered a low solubility of cholesterol in the membrane. Furthermore, in works of Wassall et al.121,122 and Shaikh123 it was demonstrated that cholesterol has the lowest solubility in membranes consisting of 22:6−22:6 PC. It was also suggested that PUFAs were pushing away cholesterol molecules, and in a mixed membrane docosahexaenoic acid-rich domains were found. In addition, experiments using neutron scattering techniques124,125 have Figure 11. Second order Legendre polynomial of the reorientation shown that in membranes consisting of 20:4−20:4 PC lipids time correlation functions of CH bond at carbons C2 and C21 in the sn- cholesterol is frequently oriented in a way that its hydroxyl − 2 chain of 18:0 22:6 PC bilayer. group resides in the center of membrane: alongside with upright position with hydroxyl group directed to the membrane The computed time correlation functions were used to surface (position 1 in Figure 13), positions with cholesterol 13 lying in the membrane midplane (2), and with hydroxyl group determine C relaxation time T1 using eq 10 corresponding to the motional narrowing regime, and eqs 8 and 9 for an NMR directed to the membrane center (3) were suggested. On the ω simulation side, phase separation of cholesterol from PUFA spectrometer operating at H = 500 MHz. Results for T1 for all − − lipids and cholesterol coordination in the membrane center was carbons of the unsaturated chain of 18:0 22:5 PC and 18:0 126 22:6 PC are shown in Figure 12 together with the experimental demonstrated within coarse-grained Martini FF. data from work by Eldho et al.40 The simulations qualitatively We have carried out simulations of two membranes − − reproduce the experimental behavior showing substantial consisting each of 22:6 22:6 PC or 20:4 20:4 PC lipids and − increase of T1 toward the end of the chains, but T1 times cholesterol, as well as simulations of pure 22:6 22:6 PC or computed from the simulations are systematically, by a factor 20:4−20:4 PC bilayers. In case of cholesterol mixtures, 3−5 lower than the experimental ones except the carbons near simulation boxes included 200 molecules of the selected lipid, the chain ends, which may be an indication of underestimation 100 molecules of cholesterol and 10000 molecules of water. 63 of the rotational dynamics of lipid tails. Cholesterol was described by the Slipids FF and water model Simulation of Cholesterol in Polyunsaturated Bi- was TIP3P. The temperature was set to 293 K and pressure to layers. PUFA bilayers are known to show remarkable 1 bar. The simulation started from random positioning of properties in interaction with cholesterol, an important cholesterol in the bilayer and they lasted for 2.4 μs for the component of animal biomembranes. Residing in the mixtures and 400 ns for the pure systems. membrane interior and interacting with lipid tails, cholesterol Addition of cholesterol to the bilayers was found to have only decreases membrane fluidity which is known as the a small effect on the total area of the bilayer. The average area condensation effect.116,117 Cholesterol is believed to be per PC lipid increased from 73.7 Å2 to 77.1 Å2 for 20:4−20:4 responsible for formation of lipid rafts, which are rigid PC and from 71.8 Å2 to 72.9 Å2 for 22:6−22:6 PC upon cholesterol-rich nanodomains in biological membranes.118 addition of cholesterol (the uncertainty was within 0.8 Å2 in all Back in 1983, Dekker et al.119 had been studying the effects these cases). We have investigated the distribution of the of cholesterol in lipid bilayers consisting of PUFAs. He and co- location and orientation of cholesterol molecules in the

13 Figure 12. C T1 relaxation time computed in approximation of motional narrowing (Comp, m.n.) and using eqs 8 and 9 for an NMR spectrometer ω − − operating at H = 500 MHz (Comp, 500 MHz), as well as experimental T1 from ref 40 for 18:0 22:5 PC and 18:0 22:6 PC bilayers.

12835 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

Figure 13. (A): Possible positioning of cholesterol molecules inside bilayer. (B): Distributions of mass densities of cholesterol molecules across bilayer. (C): Definition of angle θ between cholesterol and bilayer normal. (D) Distributions of cosine of the angle between the cholesterol and normal to bilayer vector going out of membrane.

Figure 14. Electron density profiles 20:4−20:4 PC-cholesterol and 22:6−22:6 PC-cholesterol mixed bilayers (A), for bilayers consisting of 20:4− 20:4 PC and 22:6−22:6 PC (B), and the difference profiles (C). bilayers. The distributions were calculated from the last the cholesterol distribution in our work can be related to larger microsecond of the simulations. Figure 13B shows cholesterol cholesterol concentration (33% in our work compared to 10% mass density distribution along normal direction to the in works124,125). A drop in the difference electron density at membrane. It is clear that cholesterol is highly present in the 15−20 Å from the membrane center (Figure 14) can be related hydrophobic part of the membrane including membrane center, to a rearrangement of the lipid head groups due to high and in the case of 22:6−22:6 PC bilayer the distribution is cholesterol content in the tail region. more concentrated near the membrane central plane. Figure 14 The distribution of cosine of angle between the cholesterol shows electron density profiles of the mixed lipid-cholesterol and membrane normal going out of bilayer is shown in Figure systems, pure lipid systems, and the difference. The difference 13D. Each cholesterol molecule was prescribed to one of the profile has a wide maximum within 10 Å from the bilayer monolayers according to the position of cholesterol center of center. These results can be compared with the experimental mass relative to the membrane midplane, and direction “out of neutron scattering difference profiles for 20:4−20:4 PC bilayer bilayer” was defined relative to this monolayer. The cholesterol from the works by Harroun et al.124,125 which have more vector was defined as going from the carbon in the end of narrow maximum near the bilayer center. A wider maximum of cholesterol tail to the oxygen of the hydroxyl group as depicted

12836 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article in Figure 13C, thus cosine value 1 corresponds to the cholesterol orientation in “upright” position with the hydroxyl group oriented toward the bilayer surface, cosine value −1 corresponds to cholesterol aligned to the bilayer normal with headgroup directed to the membrane interior and cosine value 0 corresponds to the cholesterol oriented parallel to the membrane plane. Additionally, probability distribution of the cholesterol hydroxyl group along the membrane normal is presented in the Supporting Information (Figure S12). The latter figure shows that besides the main maxima at 15−17 Å from the bilayer center, there is a local distribution maximum exactly in the bilayer center which is more pronounced for 22:6−22:6 PC bilayer. Analysis of the simulation results shows that most of cholesterol molecules are present in the “upright” position (1) (Figure 13A), however there is a non-negligible (above 1%) fraction of molecules with the headgroups directed toward membrane interior (case 3) and with cholesterol oriented parallel to the membrane plane (case 2). Both cases (2) and (3) correspond to location of cholesterol with hydroxyl group close to the bilayer midplane. This result can be related to findings of Harroun et al.124 that cholesterol hydroxyl group is highly present in the center of polyunsaturated lipid membrane, however in our simulations only a small fraction of cholesterol have such coordination. Note that simulations of cholesterol containing 20:4−20:4 PC and 22:6−22:6 PC bilayers by C36p FF65 did not show any local free energy minima at all for cholesterol in the middle of bilayer. Furthermore, in works by Brzustowicz et al. the cholesterol average tilt angle was determined as 25° for 20:4−20:4 PC,127 and 24° for 22:6−22:6 PC128 (both results for T = 293 K), which exclude prevailing orientation of cholesterol parallel to the membrane plane. The cholesterol tilt angle, determined in ± our work from the angle distribution in Figure 13D as 19.5 Figure 15. (A): Two-dimensional radial distribution functions in x,y− 1° for 20:4−20:4 PC and 16.5 ± 1° for 22:6−22:6 PC, is directions for cholesterol molecules in a monolayer of lipids bilayers somewhat lower than the tilt found experimentally, but shows consisting of 20:4−20:4 PCs and 22:6−22:6 PCs. (B,C): Views from correct trend (larger tilt for 20:4−20:4 PC than for 22:6−22:6 the top of boxes on cholesterol molecules in the last frame of PC). simulations for bilayers consisting of 20:4−20:4 PC and 22:6−22:6 PC We have also computed two-dimensional RDF between lipids respectively, lipids of the bottom monolayer are hightlighted in cholesterol centers of mass within each monolayer. The RDFs, red. collected during the last 500 ns of the simulations, are shown in Figure 15A. The snapshots of the last frames of simulations for not consider the final structures as fully equilibrated. Never- extracted cholesterol molecules from both bilayers are displayed theless a tendency to association of cholesterol molecules can in Figures 15(B,C). A qualitative analysis of the snapshots and be noted already after a few hundred nanoseconds of simulation RDFs shows that cholesterol molecules tend to form “clusters” when most of cholesterol molecules form “pairs” which then mostly in the form of chains in two-dimensions, so that several slowly evolve to systems of chain clusters shown in Figure 15. cholesterol molecules are connected to each other. Corre- The cluster structure is more expressed for 22:6−22:6 PCs spondingly, two-dimensional RDFs have clearly distinguishable cholesterol bilayers which correlates with the experimental maxima coming from the first, second, and so on neighbors in result that cholesterol has lowest solubility in this type of such chains. Previously, two-dimensional RDFs between bilayer. cholesterol molecules with similar structure were observed for cholesterol in DMPC lipid bilayer,129 but in the case of ■ CONCLUSION polyunsaturated bilayer of the present work the RDF structural In this work we have derived a new set of force field parameters features are stronger, and they propagate on longer distances. describing polyunsaturated lipid tails having multiple methyl- This means that in the case of polyunsaturated lipids ene-separated cis double bonds, which became an extension of cholesterol clustering is stronger, which is in agreement with the existing Slipids FF. We have validated the new set of the general tendency of lower affinity of cholesterol to parameters by extensive simulations of a number of bilayers unsaturated lipids. composed of PUFA lipids and comparison with experimental Despite the long simulation time (2.4 μs), for a binary system data on the average area per lipid, bilayer thickness, order it may be not enough to reach the thermodynamic equilibrium. parameters and scattering factors, which showed a fair The lateral diffusion of lipids and cholesterol in the mixed agreement with the experimental structural properties of the − systems was in the range (4−6)·10 8 cm2/s which corresponds bilayer. The experimental temperature dependence of the order to the average traveling distance about 70 Å (less than the box parameters for 16:0−18:2 PC bilayer, and a tiny difference in size) for the whole time of simulations. By this reason we do the order parameters of saturated tails of 18:0−22:5 and 18:0−

12837 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

22:6 bilayers were qualitatively reproduced. NMR T1 relaxation validation of the FF. Still, the current version of the FF for times of CH bonds were computed and found to be shorter PUFA lipids is consistent with the experimental data that are than the experimental ones which may be an indication of available at the moment. Our confidence in the reliability of the underestimating the rotational dynamics of lipids. new parameter set is strengthened by the fact that the Furthermore, the new set of FF parameters for polyunsatu- parameters were derived exclusively on the basis of high-level rated tails of lipids has led to appearance of interesting effects in ab initio computations. interaction of cholesterol with PUFA bilayers, which, on one hand, have not been observed in previous simulations with ■ ASSOCIATED CONTENT atomistic force fields, and on the other hand are in line with *S Supporting Information experimental observations. Our simulations of mixtures of lipids The Supporting Information is available free of charge on the having symmetric PUFA chains with cholesterol have shown ACS Publications website at DOI: 10.1021/acs.jpcb.6b05422. certain degree of clustering of cholesterol molecules in the Archive containing the force field, molecular topology bilayers, which can be related to known from experiments low files, input files with MD parameters, and equilibrated solubility of cholesterol in bilayers with high content of fi 121−123 con gurations of bilayers in Gromacs format (ZIP) PUFA. We have also demonstrated that cholesterol can Details on model molecules and torsion angles; quantum have several types of positioning and orientation in PUFA energy of cis-3,cis-6-nonadiene as a function of the “ ” bilayers. Apart from the upright position with the headgroup torsional angle; time evolution of the average area per directed to the bilayer surface, we have found, in simulations of lipids for some bilayers; order parameters for some − − 20:4 20:4 PC and 22:6 22:6 PC bilayers, a small (an order of bilayers; partial charges computed for individual percent) population of lipids with the headgroup oriented to structures (PDF) the bilayer center, as well as cholesterol molecules lying in the fi bilayer center parallel to the bilayer surface. Such con gurations ■ AUTHOR INFORMATION were previously suggested in the experimental studies of PUFA Corresponding Author bilayers with cholesterol. * The optimization of the FF parameters in this work was done E-mail: [email protected]; Telephone: +46- with respect to partial atomic charges and torsional potentials. 8161193. Parametrization of dihedrals was important since they directly ORCID affect flexibility of hydrocarbon chains and molecular Alexander P. Lyubartsev: 0000-0002-9390-5719 conformation, which have consequences for the structural Notes properties of the lipid bilayers. By using ab initio computations The authors declare no competing financial interest. we have developed parameters that provide realistic con- formations of polyunsaturated lipid tails which were confirmed ■ ACKNOWLEDGMENTS by experimental data on the geometric characteristic of the The work has been supported by the Swedish Research Council bilayers and order parameters. Similar re-parametrization of the (Vetenskapsrådet), grant 621-2013-4260.) The simulations torsion parameters for PUFA chains was done in the recent were performed on resources provided by the Swedish National 65 update of the Charmm36 FF. Our work is different from that Infrastructure for Computing (SNIC) at PDC (Stockholm), work in that we also re-parametrized partial atom charges. The NSC (Linköping), and HPC2N (Umeå). We thank Joakim new set of partial charges (obtained from DFT computations Jambeck̈ and Arnold Maliniak for useful discussions and Jon carried out for ensemble of conformations) assigns more polar Kapla for the software suite TRAJMAN. partial charges for carbons and hydrogens at double bonds compared with those for monounsaturated bonds, leading to ■ REFERENCES higher polarity of such double bonds, and weaker hydro- (1) Singer, S. J.; Nicolson, G. L. The Fluid Mosaic Model of the phobicity of the membrane interior. Indirectly, this can be Structure of Cell Membranes. Science 1972, 175, 720−731. supported by the increase of water permeability observed (2) Cullis, P. R.; de Kruijff, B. Lipid Polymorphism and the experimentally in PUFA bilayers upon the increase of the Functional Roles of Lipids in Biological Membranes. Biochim. Biophys. number of double bonds.130 Furthermore, stronger partial Acta, Rev. Biomembr. 1979, 559, 399−420. charges increase local dipole−dipole interactions between (3) Stein, W. D. The Movement of Molecules Across Cell Membranes; PUFA chains, which, according to the regular solution theory, Academic Press: NY, 1967. (4) Singer, S. J. The Molecular Organization of Membranes. Annu. favors demixing of PUFA chains and nonpolar compounds. The Rev. Biochem. 1974, 43, 805−833. clustering of cholesterol molecules (having very small partial (5) Mileykovskaya, E.; Dowhan, W. Role of Membrane Lipids in charges except the hydroxyl group), as well as alternative Bacterial Division Site Delection. Curr. Opin. Microbiol. 2005, 8, 135− positioning of cholesterol hydroxyl group in PUFA membranes 142. obtained in this study is likely related to the increased partial (6) Chen, Y. A.; Scheller, R. H. SNARE-Mediated Membrane Fusion. atom charges at double bonds. Similar effect was observed in a Nat. Rev. Mol. Cell Biol. 2001, 2,98−106. recent paper on parametrization of the FF for perfluor- (7) Marsh, D. Handbook of Lipid Bilayers, Second ed.; CRC Press, alkanes131 where increase of partial charges at CF groups has 2013. led to (experimentally observed) phase separation of (8) Jacquot, A.; Francius, G.; Razafitianamaharavo, A.; Dehghani, F.; perfluoralkanes and hydrocarbons. Tamayol, A.; Linder, M.; Arab-Tehrany, E. Morphological and Physical Analysis of Natural Phospholipids-Based Biomembranes. PLoS One Finally we remark on somewhat lower availability of the 2014, 9, e107435. experimental data for PUFA bilayers in comparison to bilayers (9) Lunn, J.; Theobald, H. E. The Health Effects of Dietary with fully saturated or monounsaturated lipids. New exper- Unsaturated Fatty Acids. Nutrition Bull. 2006, 31, 178−224. imental studies of PUFA bilayers, and especially data from X- (10) Schlögelhofer, M.; Amminger, G. P.; Schaefer, M. R.; Fusar-Poli, ray and neutron scattering would be of great help in further P.; Smesny, S.; McGorry, P.; Berger, G.; Mossaheb, N. Polyunsatu-

12838 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article rated Fatty Acids in Emerging Psychosis: a Safer Alternative? Early Cause Mortality in Patients with Acute Myocardial Infarction. Circ. J. Interv. Psychiatry 2014, 8, 199−208. 2009, 73, 2250−2257. (11) Zugno, A. I.; Chipindo, H. L.; Volpato, A. M.; Budni, J.; (26) Dessi, M.; Noce, A.; Bertucci, P.; Manca di Villahermosa, S. M.; Steckert, A. V.; De Oliveira, M. B.; Heylmann, A. S.; Da Rosa Silveira, Zenobi, R.; Castagnola, V.; Addessi, E.; Di Daniele, N. Atherosclerosis, F.; Mastella, G. A.; Maravai, S. G.; et al. Omega-3 Prevents Behavior Dyslipidemia, and Inflammation: The Significant Role of Polyunsatu- Response and Brain Oxidative Damage in the Ketamine Model of rated Fatty Acids. ISRN Inflammation 2013, 2013, 191823. Schizophrenia. Neuroscience 2014, 259, 223−231. (27) Vasudevan, A.; Yu, Y.; Banerjee, S.; Woods, J.; Farhana, L.; (12) Widenhorn-Muller, K.; Schwanda, S.; Scholz, E.; Spitzer, M.; Rajendra, S. G.; Patel, A.; Dyson, G.; Levi, E.; Maddipati, K. R.; et al. Bode, H. Effect of Supplementation with Long-chain ω −3 Omega-3 Fatty Acid Is a Potential Preventive Agent for Recurrent Polyunsaturated Fatty Acids on Behavior and Cognition in Children Colon Cancer. Cancer Prev. Res. 2014, 7, 1138−1148. with Attention Deficit/Hyperactivity Disorder (ADHD): A Random- (28) Wang, W.; Zhu, J.; Lyu, F.; Panigrahy, D.; Ferrara, K. W.; ized Placebo-Controlled Intervention Trial. Prostaglandins, Leukotrienes Hammock, B.; Zhang, G. ω-3 Polyunsaturated Fatty Acids-Derived Essent. Fatty Acids 2014, 91,49−60. Lipid Metabilites on Angiogenesis, Inflammation and Cancer. (13) Hawkey, E.; Nigg, J. T. Omega-3 Fatty Acid and ADHD: Blood Prostaglandins Other Lipid Mediators 2014, 113-115,13−20. Level Analysis and Meta-Analytic Extension of Supplementation (29) Bassett, J. K.; Severi, G.; Hodge, A. M.; MacInnis, R. J.; Gibson, Trials. Clin. Psychol. Rev. 2014, 34, 496−505. R. A.; Hopper, J. L.; English, D. R.; Giles, G. G. Plasma Phospholipid (14) Grosso, G.; Galvano, F.; Marventano, S.; Malaguarnera, M.; Fatty Acids, Dietary Fatty Acids and Prostate Cancer Risk. Int. J. Bucolo, C.; Caraci, F.; Drago, F. Omega-3 Fatty Acids and Depression: Cancer 2013, 133, 1882−1892. Scientific Evidence and Biological Mechanisms. Oxid. Med. Cell. (30) Siddiqui, R. A.; Harvey, K.; Stillwell, W. Anticancer Properties of Longevity 2014, 2014, 313570. Oxidation Products of Docosahexaenoic acid. Chem. Phys. Lipids 2008, (15) El-Ansary, A. K.; Bacha, A. G. B.; Al-Ayadhi, L. Y. Impaired 153,47−56. Plasma Phospholipids and Relative Amounts of Essential Polyunsatu- (31) Bousquet, M.; Calon, F.; Cicchetti, F. Impact of Omega-3 Fatty rated Fatty Acids in Autistic Patients from Saudi Arabia. Lipids Health Acids in Parkinson’s Disease. Ageing Res. Rev. 2011, 10, 453−463. Dis. 2011, 10, 63. (32) Schnebelen, C.; Viau, S.; Gregoire,́ S.; Joffre, C.; Creuzot- (16) Rondanelli, M.; Giacosa, A.; Opizzi, A.; Pelucchi, C.; La Vecchia, Garcher, C. P.; Bron, A. M.; Bretillon, L.; Acar, N. Nutrition for the C.; Montorfano, G.; Negroni, M.; Berra, B.; Politi, P.; Rizzo, A. M. Eye: Different Susceptibility of the Retina and the Lacrimal Gland to Long Chain Omega-3 Polyunsaturated Fatty Acids Supplementation in Dietary Omega-6 and Omega-3 Polyunsaturated Fatty Acid Incorpo- the Treatment of Elderly Depression: Effects on Depressive ration. Ophthalmic Res. 2009, 41, 216−224. Symptoms, on Phospholipids Fatty Acids Profile and on Health- (33) Lou, D. J.; Zhu, Q. Q.; Si, X. W.; Guan, L. L.; You, Q. Y.; Yu, Z. Related Quality of Life. J. Nutr., Health Aging 2011, 15,37−44. M.; Zhang, A. Z.; Li, D. Serum Phospholipid Omega-3 Polyunsatu- (17) Suominen-Taipale, A. L.; Partonen, T.; Turunen, A. W.; rated Fatty Acids and Insulin Resistance in Type 2 Diabetes Mellitus Mannistö ̈, S.; Jula, A.; Verkasalo, P. K. Fish Consumption and Omega- and Non-Alcoholic Fatty Liver Disease. J. Diabetes Complicat. 2014, 28, 3 Polyunsaturated Fatty Acids in Relation to Depressive Episodes: A 711−714. Cross-Sectional Analysis. PLoS One 2010, 5, e10530. (34) Rossmeisl, M.; Medrikova, D.; van Schothorst, E. M.; Pavlisova, (18) Vakhapova, V.; Cohen, T.; Richter, Y.; Herzog, Y.; Korczyn, A. J.; Kuda, O.; Hensler, M.; Bardova, K.; Flachs, P.; Stankova, B.; Vecka, D. Phosphatidylserine Containing ω-3 Fatty Acids May Improve M.; et al. Omega-3 Phospholipids from Fish Suppress Hepatic Memory Abilities in Non-Demented Elderly with Memory Com- Steatosis by Integrated Inhibition of Biosynthetic Pathways in Dietary plaints: A Double-Blind Placebo-Controlled Trial. Dementia Geriatr. Obese Mice. Biochim. Biophys. Acta, Mol. Cell Biol. Lipids 2014, 1841, Cognit. Disord. 2010, 29, 467−474. 267−278. (19) Clayton, E. H.; Hanstock, T. L.; Hirneth, S. J.; Kable, C. J.; Garg, (35) Stonehouse, W. Does Consumption of LC Omega-3 PUFA M. L.; Hazell, P. L. Long-Chain Omega-3 Polyunsaturated Fatty Acids Enhance Cognitive Performance in Healthy School-Aged Children and in the Blood of Children and Adolescents with Juvenile Bipolar throughout Adulthood? Evidence from Clinical Trials. Nutrients 2014, Disorder. Lipids 2008, 43, 1031−1038. 6, 2730−2758. (20) Rudkowska, I.; Julien, P.; Couture, P.; Lemieux, S.; Tchernof, A.; (36) Serini, S.; Fasano, E.; Piccioni, E.; Cittadini, A. R. M.; Calviello, Barbier, O.; Vohl, M.-C. Cardiometabolic Risk Factors are Influenced G. Dietary n-3 Polyunsaturated Fatty Acids and the Paradox of Their by Stearoyl-CoA Desaturase (SCD) − 1 Gene Polymorphisms and n-3 Health Benefits and Potential Harmful Effects. Chem. Res. Toxicol. Polyunsaturated Fatty Acid Supplementation. Mol. Nutr. Food Res. 2011, 24, 2093−2105. 2014, 58, 1079−1086. (37) Stillwell, W.; Shaikh, S. R.; Zerouga, M.; Siddiqui, R.; Wassall, S. (21) Macartney, M. J.; Hingley, L.; Brown, M. A.; Peoples, G. E.; R. Docosahexaenoic Acid Affects Cell Signalling by Altering Lipid McLennan, P. L. Intrinsic Heart Rate Recovery after Dynamic Exercise Raffs. Reprod., Nutr., Dev. 2005, 45, 559−579. Is Improved with an Increased Omega-3 Index in Healthy Males. Br. J. (38) Bowling, N.; Walsh, R. A.; Song, G.; Estridge, T.; Sandusky, G. Nutr. 2014, 112, 1984−1992. E.; Fouts, R. L.; Mintze, K.; Pickard, T.; Roden, R.; Bristow, M. R.; (22) Wilusz, M.; Wasilewski, G.; Przybyłowski, P.; Janik, L.; Bugajska, et al. Increased Protein Kinase C Activity and Expression of Ca2+- J.; Sztefko, K. Effect of Immunosuppressive Therapy on the Serum Sensitive Isoforms in the Failing Human Heart. Circulation 1999, 99, Fatty Acids of Phospholipids Fraction in Patients After Heart 384−391. Transplantation. Transplant. Proc. 2014, 46, 2825−2829. (39) Lyubartsev, A. P.; Rabinovich, A. L. Recent Development in (23) Masson, S.; Marchioli, R.; Mozaffarian, D.; Bernasconi, R.; Computer Simulations of Lipid Bilayers. Soft Matter 2011, 7,25−39. Milani, V.; Dragani, L.; Tacconi, M.; Marfisi, R. M.; Borgese, L.; (40) Eldho, N. V.; Feller, S. E.; Tristram-Nagle, S.; Polozov, I. V.; Cirrincione, V.; et al. Plasma n-3 Polyunsaturated Fatty Acids in Gawrisch, K. Polyunsaturated Docosahexaenoic vs Docosapentaenoic Chronic Heart Failure in the GISSI-Heart Failure Trial: Relation with Acid Differences in Lipid Matrix Properties from the Loss of One Fish Intake, Circulating Biomarkers, and Mortality. Am. Heart J. 2013, Double Bond. J. Am. Chem. Soc. 2003, 125, 6409−6421. 165, 208−215. (41) Pitman, M. C.; Suits, F.; MacKerell, A. D.; Feller, S. E. (24) Liu, J. C.; Conklin, S. M.; Manuck, S. B.; Yao, J. K.; Muldoon, Molecular-Level Organization of Saturated and Polyunsaturated Fatty M. F. Long-Chain Omega-3 Fatty Acids and Blood Pressure. Am. J. Acids in a Phosphatidylcholine Bilayer Containing Cholesterol. Hypertens. 2011, 24, 1121−1126. Biochemistry 2004, 43, 15318−15328. (25) Lee, S.-H.; Shin, M.-J.; Kim, J.-S.; Ko, Y.-G.; Kang, S.-M.; Choi, (42) Hyvönen,M.T.;Kovanen,P.T.MolecularDynamics D.; Jang, Y.; Chung, N.; Shim, W.-H.; Cho, S.-Y.; et al. Blood Simulations of Unsaturated Lipid Bilayers: Effects of Varying the Eicosapentaenoic Acid and Docosahexaenoic Acid as Predictors of All- Number of Double Bonds. Eur. Biophys. J. 2005, 34, 294−305.

12839 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

(43) Carrillo-Tripp, M.; Feller, S. E. Evidence for a Mechanism by (63) Jambeck,̈ J. P. M.; Lyubartsev, A. P. Another Piece of the which ω-3 Polyunsaturated Lipids May Affect Membrane Protein Membrane Puzzle: Extending Slipids Further. J. Chem. Theory Comput. Function. Biochemistry 2005, 44, 10164−10169. 2013, 9, 774−784. (44) Ollila, S.; Hyvönen, M. T.; Vattulainen, I. Polyunsaturation in (64) Lyubartsev, A. P.; Rabinovich, A. L. Force Field Development Lipid Membranes: Dynamic Properties and Lateral Pressure Profiles. J. for Lipid Membrane Simulations. Biochim. Biophys. Acta, Biomembr. Phys. Chem. B 2007, 111, 3139−3150. 2016, 1858, 2483−2497. (45) Rabinovich, A. L.; Kornilov, V. V.; Balabaev, N. K.; Leermakers, (65) Klauda, J. B.; Monje, V.; Kim, T.; Im, W. Improving the F. A. M.; Filippov, A. V. Properties of Unsaturated Phospholipid CHARMM Force Field for Polyunsaturated Fatty Acid Chains. J. Phys. Bilayers: Effect of Cholesterol. Biochemistry (Moscow), Ser. A: Membr. Chem. B 2012, 116, 9424−9431. Cell Biol. 2007, 1, 343−357. (66) Bachar, M.; Brunelle, P.; Tieleman, D. P.; Rauk, A. Molecular (46) Feller, S. E. Acyl Chain Conformations in Phospholipid Bilayers: Dynamics Simulation of a Polyunsaturated Lipid Bilayer Susceptible to a Comparative Study of Docosahexaenoic Acid and Saturated Fatty Lipid Peroxidation. J. Phys. Chem. B 2004, 108, 7170−7179. Acids. Chem. Phys. Lipids 2008, 153,76−80. (67) Marrink, S. J.; de Vries, A. H.; Harroun, T. A.; Katsaras, J.; (47) Bennett, W. F. D.; MacCallum, J. L.; Hinner, M. J.; Marrink, S. Wassall, S. R. Cholesterol Shows Preference for the Interior of − J.; Tieleman, D. P. Molecular View of Cholesterol Flip-Flop and Polyunsaturated Lipid Membranes. J. Am. Chem. Soc. 2008, 130,10 Chemical Potential in Different Membrane Environments. J. Am. 11. ̈ Chem. Soc. 2009, 131, 12714−12720. (68) Jambeck, J. P. M.; Lyubartsev, A. P. Implicit Inclusion of Atomic (48) Rabinovich, A. L.; Lyubartsev, A. P. Bond Orientation Polarization in Modeling of Partitioning between Water and Lipid − Properties in Lipid Molecules of Membranes: Molecular Dynamics Bilayers. Phys. Chem. Chem. Phys. 2013, 15, 4677 4686. ̈ Simulations. J. Phys.: Conf. Ser. 2014, 510, 012022. (69) Jambeck, J. P. M.; Lyubartsev, A. P. Exploring the Free Energy (49) Berger, O.; Edholm, O.; Jahnig,̈ F. Molecular Dynamics Landscape of Solutes Embedded in Lipid Bilayers. J. Phys. Chem. Lett. − Simulations of a Fluid Bilayer of Dipalmitoylphosphatidylcholine at 2013, 4, 1781 1787. ́ ́ Full Hydration, Constant Pressure and Constant Temperature. (70) Paloncyova, M.; Fabre, G.; DeVane, R. H.; Trouillas, P.; Berka, − K.; Otyepka, M. Benchmarking of Force Fields for Molecule- Biophys. J. 1997, 72, 2002 2013. − (50) Chiu, S. W.; Pandit, S. A.; Scott, H. L.; Jakobsson, E. An Membrane Interactions. J. Chem. Theory Comput. 2014, 10, 4143 Improved United Atom Force Field for Simulations of Mixed Lipid 4151. − (71) Becke, A. D. Density Functional Thermochemistry. III. The Bilayers. J. Phys. Chem. B 2009, 113, 2748 2763. − (51) Poger, D.; van Gunsteren, W. F.; Mark, A. E. A New Force Field Role of Exact Exchange. J. Chem. Phys. 1993, 98, 5648 5652. (72) Lee, C.; Yang, W.; Parr, R. G. Development of the Colle-Salvetti for Simulating Phosphatidylcholine Bilayers. J. Comput. Chem. 2010, Correlation Energy Formula into a Functional of the Electron Density. 31, 1117−1125. 1988 − (52) Feller, S. E.; Yin, D.; Pastor, R. W.; MacKerell, A. D., Jr. Phys. Rev. B: Condens. Matter Mater. Phys. , 37, 785 789. (73) Stephens, P. J.; Devlin, F. J.; Chabalowski, C. F.; Frisch, M. J. Molecular Dynamics Simulation of Unsaturated Lipid Bilayers at Low Ab-initio Calculation of Vibrational Absorption and Circular Dichroism Hydration: Parametrization and Comparison with Diffraction Studies. Spectra using Density Functional Force Fields. J. Phys. Chem. 1994, 98, Biophys. J. 1997, 73, 2269−2279. 11623−11627. (53) Feller, S. E.; MacKerell, A. D., Jr. An Improved Empirical (74) Kendall, R. A.; Dunning, T. H., Jr.; Harrison, R. J. Electron Potential Energy Function for Molecular Simulations of Phospholi- Affinities of the First-Row Atoms Revisited. Systematic Basis Sets and 2000 − pids. J. Phys. Chem. B , 104, 7510 7515. Wave Functions. J. Chem. Phys. 1992, 96, 6796−6806. (54) Klauda, J. B.; Brooks, B. R.; MacKerell, A. D.; Venable, R. M.; (75) Tomasi, J.; Mennucci, B.; Cances,́ E. The IEF Version of the Pastor, R. W. An ab initio Study on the Torsional Surface of Alkanes PCM Solvation Method: an Overview of a New Method Addressed to and its Effect on Molecular Simulations of Alkanes and a DPPC Study Molecular Solutes at the QM ab initio Level. J. Mol. Struct.: Bilayer. J. Phys. Chem. B 2005, 109, 5300−5311. − ’ THEOCHEM 1999, 464, 211 226. (55) Klauda, J. B.; Venable, R. M.; Freites, J. A.; O Connor, J. W.; (76) Pomelli, C. S.; Tomasi, J.; Barone, V. An Improved Iterative Tobias, D. J.; Mondragon-Ramirez, C.; Vorobyov, I.; MacKerell, A. D.; Solution to Solve the Electrostatic Problem in the Polarizable Pastor, R. W. Update of the CHARMM All-Atom Additive Force Field Continuum Model. Theor. Chem. Acc. 2001, 105, 446−451. for Lipids: Validation on Six Lipid Types. J. Phys. Chem. B 2010, 114, (77) Cornell, W. D.; Cieplak, P.; Bayly, C. I.; Kollman, P. A. − 7830 7843. Applicattion of RESP Charges to Calculate Conformational Energies, ́ ̈ (56) Siu, S. W. I.; Vacha, R.; Jungwirth, P.; Bockmann, R. A. Hydrogen Bond Energies, and Free Energies of Solvation. J. Am. Chem. Biomolecular Simulations of Membranes: Physical Properties from Soc. 1993, 115, 9620−9631. Different Force Fields. J. Chem. Phys. 2008, 128, 125103. (78) Dupradeau, F. Y.; Pigache, A.; Zaffran, T.; Savineau, C.; Lelong, (57) Wang, J.; Wolf, R. M.; Caldwell, J. W.; Kollman, P. A.; Case, D. R.; Grivel, N.; Lelong, D.; Rosanski, W.; Cieplak, P. The R.E.D. Tools: A. Development and Testing of a General Amber Force Field. J. Advances in RESP and ESP Charge Derivation and Force Field Library − Comput. Chem. 2004, 25, 1157 1174. Building. Phys. Chem. Chem. Phys. 2010, 12, 7821−7839. (58) Skjevik, A. A.; Madej, B. D.; Walker, R. C.; Teigen, K. LIPID11: (79) Wilhoit, R. C.; Marsh, K. N.; Hong, X.; Gadalla, N.; Frenkel, M. A Modular Framework for Lipid Simulations Using Amber. J. Phys. Landolt-Bornstein:̈ Numerical Data and Functional Relationships in Chem. B 2012, 116, 11124−11136. Science and Technology. Group IV: Physical Chemistry; Springer: Berlin (59) Dickson, C. J.; Madej, B. D.; Skjevik, A. A.; Betz, R. M.; Teigen, Hidelberg, 1999; Vol. 8C; pp 237−261. K.; Gould, I. R.; Walker, R. C. Lipid14: The Amber Lipid Force Field. (80) Klauda, J. B.; Garrison, S. L.; Jiang, J.; Arora, G.; Sandler, S. I. J. Chem. Theory Comput. 2014, 10, 865−879. HM-IE: Quantum Chemical Hybrid Methods for Calculating (60) Tessier, M. B.; DeMarco, M. L.; Yongye, A. B.; Woods, R. J. Interaction Energies. J. Phys. Chem. A 2004, 108, 107−112. Extension of the GLYCAM06 Biomolecular Force Field to Lipids, (81) Raghavachari, K.; Trucks, G. W.; Pople, J. A.; Head-Gordon, M. Lipid Bilayers and Glycolipids. Mol. Simul. 2008, 34, 349−363. A Fifth-Order Perturbation Comparison of Electron Correlation (61) Jambeck,̈ J. P. M.; Lyubartsev, A. P. Derivation and Systematic Theories. Chem. Phys. Lett. 1989, 157, 479−483. Validation of a Refined All-Atom Force Field for Phosphatidylcholine (82) Møller, C.; Plesset, M. S. Note on an Approximation Treatment Lipids. J. Phys. Chem. B 2012, 116, 3164−3179. for Many-Electron Systems. Phys. Rev. 1934, 46, 618−622. (62) Jambeck,̈ J. P. M.; Lyubartsev, A. P. An Extension and Further (83) Saiz, L.; Klein, M. L. Structural Properties of a Highly Validation of an All-Atomistic Force Field for Biological Membranes. J. Polyunsaturated Lipid Bilayer from Molecular Dynamics Simulations. Chem. Theory Comput. 2012, 8, 2938−2948. Biophys. J. 2001, 81, 204−216.

12840 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

(84) Hess, B.; Kutzner, C.; van der Spoel, D.; Lindahl, E. GROMACS Change of Molecular Shape for Phosphatidylcholines Containing a 4: Algorithms for Highly Efficient, Load-Balanced, and Scalable Polyunsaturated Acyl Chain. Biophys. J. 1995, 68, 2396−2403. Molecular Simulation. J. Chem. Theory Comput. 2008, 4, 435−447. (105) Aussenac, F.; Laguerre, M.; Schmitter, J.; Dufourc, E. Detailed (85) Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. LINCS: A Structure and Dynamics of Bicelle Phospholipids Using Selectively Linear Constraint Solver for Molecular Simulations. J. Comput. Chem. Deuterated and Perdeuterated Labels. 2H NMR and Molecular 1997, 18, 1463−1472. Mechanics Study. Langmuir 2003, 19, 10468−10479. (86) Miyamoto, S.; Kollman, P. A. SETTLE: An Analytical Version of (106) Baenziger, J. E.; Smith, I. C. P.; Hill, R. J. Biosynthesis of a the SHAKE and RATTLE Algorithm for Rigid Water Models. J. Specially Deuteriated Fatty Acid 18:2 for H NMR Membrane Studies. Comput. Chem. 1992, 13, 952−962. Biochemistry 1987, 26, 8405−8410. (87) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; (107) Baenziger, J. E.; Smith, I. C. P.; Hill, R. J.; Jarrell, H. C. Direct DiNola, A.; Haak, J. R. Molecular Dynamics with Coupling to an Measurements of Deuterium-Deuterium Dipolar Coupling and External Bath. J. Chem. Phys. 1984, 81.368410.1063/1.448118 Analysis of the Ordering of a Specifically Deuteriated Diunsaturated (88) Bussi, G.; Donadio, D.; Parrinello, M. Canonical Sampling Lipid. J. Am. Chem. Soc. 1988, 110, 8229−8231. through Velocity Rescaling. J. Chem. Phys. 2007, 126, 014101. (108) Baenziger, J. E.; Jarrell, H. C.; Hill, R. J.; Smith, I. C. P. Average (89) Darden, T.; York, D.; Pedersen, L. Particle Mesh Ewald: An N · Structural and Motional Properties of a Diunsaturated Acyl Chain in a log(N) Method for Ewald Sums in Large Systems. J. Chem. Phys. 1993, Lipid Bilayer: Effects of Two Cis-Unsaturated Double Bonds. − 98, 10089−10092. Biochemistry 1991, 30, 894 903. (90) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. (109) Saiz, L.; Klein, M. L. Influence of Highly Polyunsaturated Lipid W.; Klein, M. L. Comparison of Simple Potential Functions for Acyl Chains of Biomembranes on the NMR Order Parameters. J. Am. − Simulating Liquid Water. J. Chem. Phys. 1983, 79, 926−935. Chem. Soc. 2001, 123, 7381 7387. (91) Lyubartsev, A. P.; Laaksonen, A. A General and Scalable Parallel (110) Saiz, L.; Klein, M. L. Electrostatic Interactions in a Neutral Software Package for Arbitrary Mixtures of Molecules. Comput. Phys. Model Phospholipid Bilayer by Molecular Dynamics Simulations. J. − Commun. 2000, 128, 565−589. Chem. Phys. 2002, 116, 3052 3057. (92) Frisch, M. J.; Trucks, G. W.; Schlegel, H. B.; Scuseria, G. E.; (111) Petrache, H. I.; Dodd, S. W.; Brown, M. F. Area per Lipid and Robb, M. A.; Cheeseman, J. R.; Scalmani, G.; Barone, V.; Mennucci, Acyl Length Distributions in Fluid Phosphatidylcholines Determined 2 − B.; Petersson, G. A. et al. Gaussian 09, Revision E.01.; Gaussian Inc.: by H NMR Spectroscopy. Biophys. J. 2000, 79, 3172 3192. (112) Gawrisch, K.; Eldho, N. V.; Holte, L. L. The Structure of DHA Wallingford CT, 2009. − (93) Pan, J.; Tristram-Nagle, S.; Nagle, J. F. Effect of Cholesterol on in Phospholipid Membranes. Lipids 2003, 38, 445 452. Structural and Mechanical Properties of Membranes Depends on Lipid (113) Nagle, J. F.; Zhang, R.; Tristram-Nagle, S.; Sun, W.; Petrache, Chain Saturation. Phys. Rev. E 2009, 80, 021931. H. I.; Suter, R. M. X-ray Structure Determination of Fully Hydrated Lα ̌ Phase Dipalmitoylphosphatidylcholine Bilayers. Biophys. J. 1996, 70, (94) Kucerka, N.; Nieh, M.-P.; Katsaras, J. Fluid Phase Lipid Areas − and Bilayer Thicknesses of Commonly Used Phosphatidylcholines as a 1419 1431. (114) Benz, R. W.; Castro-Roman,́ F.; Tobias, D. J.; White, S. H. Function of Temperature. Biochim. Biophys. Acta, Biomembr. 2011, Experimental Validation of Molecular Dynamics Simulations of Lipid 1808, 2761−2771. Bilayers: A New Approach. Biophys. J. 2005, 88, 805−817. (95) Smaby, J. M.; Momsen, M. M.; Brockman, H. L.; Brown, R. E. (115) Lipari, G.; Szabo, A. Effect of Librational Motion on Phosphatidylcholine Acyl Unsaturation Modulates the Decrease in Fluorescence Depolarization and Nuclear Magnetic Resonance Interfacial Elasticity Induced by Cholesterol. Biophys. J. 1997, 73, − Relaxation in Macromolecules and Membranes. Biophys. J. 1980, 30, 1492 1505. 489−506. (96) Separovic, F.; Gawrisch, K. Effect of Unsaturation on the Chain (116) Barrett, M. A.; Zheng, S.; Toppozini, L. A.; Alsop, R. J.; Dies, Order of Phosphatidylcholines in a Dioleoylphosphatidylethanolamine H.; Wang, A.; Jago, N.; Moore, M.; Rheinstadter,̈ M. C. Solubility of Matrix. Biophys. J. 1996, 71, 274−282. ̌ Cholesterol in Lipid Membranes and the Formation of Immiscible (97) Kucerka, N.; Gallova, J.; Uhrikova, D.; Balgavy, P.; Bulacu, M.; Cholesterol Plaques at High Cholesterol Concentrations. Soft Matter Marrink, S.-J.; Katsaras, J. Areas of Monosaturated Diacylphosphati- − − 2013, 9, 9342 9351. dylcholines. Biophys. J. 2009, 97, 1926 1932. (117) Alsop, R. J.; Barrett, M. A.; Zheng, S.; Dies, H.; Rheinstadter,̈ (98) Petrache, H. I.; Salmon, A.; Brown, M. F. Structural Properties M. C. Acetylsalicylic Acid (ASA) Increases the Solubility of of Docosahexaenoyl Phospholipid Bilayers Investigated by Solid-State − Cholesterol when Incorporated in Lipid Membranes. Soft Matter 2H NMR Spectroscopy. J. Am. Chem. Soc. 2001, 123, 12611 12622. 2014, 10, 4275−4286. ̈ (99) Koenig, B. W.; Strey, H. H.; Gawrisch, K. Membrane Lateral (118) Niemela,̈ P. S.; Ollila, S.; Hyvönen, M. T.; Karttunen, M.; Compressibility Determined by NMR and X-Ray Diffraction: Effect of Vattulainen, I. Assessing the Nature of Lipid Raft Membranes. PLoS − Acyc Chain Polyunsaturation. Biophys. J. 1997, 73, 1954 1966. Comput. Biol. 2007, 3, 304−312. ̌ (100) Kucerka, N.; Nagle, J. F.; Sachs, J. N.; Feller, S. E.; Pencer, J.; (119) Dekker, C. J.; van Kessel, W. S. M. G.; Klomp, J. P. G.; Pieters, Jackson, A.; Katsaras, J. Lipid Bilayer Structure Determined by the J.; de Kruijff, B. Synthesis and Polymorphic Phase Behaviour of Simultaneous Analysis of Neutron and X-Ray Scattering Data. Biophys. Polyunsaturated Phosphatidylcholines and Phosphatidylethanol- − J. 2008, 95, 2356 2367. amines. Chem. Phys. Lipids 1983, 33,93−106. (101) Mihailescu, M.; Soubias, O.; Worcester, D.; White, S. H.; (120) Brzustowicz, M. R.; Stillwell, W.; Wassall, S. R. Molecular Gawrisch, K. Structure and Dynamics of Cholesterol-Containing Organization of Cholesterol in Polyunsaturated Phospholipid Polyunsaturated Lipid Membranes Studied by Neutron Diffraction and Membranes: a Solid State 2H NMR Investigation. FEBS Lett. 1999, NMR. J. Membr. Biol. 2011, 239,63−71. 451, 197−202. (102) Salmon, A.; Dodd, S. W.; Williams, G. D.; Beach, J. M.; Brown, (121) Wassall, S. R.; Brzustowicz, M. R.; Shaikh, S. R.; Cherezov, V.; M. F. Configurational Statistics of Acyl Chains in Polyunsaturated Caffrey, M.; Stillwell, W. Order from Disorder, Coralling Cholesterol Lipid Bilayers from Deuterium NMR. J. Am. Chem. Soc. 1987, 109, with Chaotic Lipids. The Role of Polyunsaturated Lipids in Membrane 2600−2609. Raft Formation. Chem. Phys. Lipids 2004, 132,79−88. (103) Huber, T.; Rajamoorthi, K.; Kurze, V.; Beyer, K.; Brown, M. (122) Wassall, S. R.; Stillwell, W. Polyunsaturated Fatty Acid - Structure of Docosahexaenoic Acid Containing Phospholipid Bilayers Cholesterol Interactions: Domain Formation in Membranes. Biochim. Studied by 2H NMR and Molecular Dynamics Simulations. J. Am. Biophys. Acta, Biomembr. 2009, 1788,24−32. Chem. Soc. 2002, 124, 298−309. (123) Shaikh, S. R.; Cherezov, V.; Caffrey, M.; Soni, S. P.; LoCascio, (104) Holte, L. L.; Peter, S. A.; Sinnwell, T. M.; Gawrisch, K. 2H D.; Stillwell, W.; Wassall, S. R. Molecular Organization of Cholesterol Nuclear Magnetic Resonance Order Parameter Profiles Suggest a in Unsaturated Phosphatidylethanolamines: X-Ray Diffraction and

12841 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842 The Journal of Physical Chemistry B Article

Solid State 2 H NMR Reveal Differences with Phosphatidylcholines. J. Am. Chem. Soc. 2006, 128, 5375−5383. (124) Harroun, T. A.; Katsaras, J.; Wassall, S. R. Cholesterol Hydroxyl Group Is Found to Reside in the Center of a Polyunsaturated Lipid Membrane. Biochemistry 2006, 45, 1227−1233. (125) Harroun, T. A.; Katsaras, J.; Wassall, S. R. Cholesterol Is Found To Reside in the Center of a Polyunsaturated Lipid Membrane. Biochemistry 2008, 47, 7090−7096. (126) Kucerka,̌ N.; Marquardt, D.; Harroun, T. A.; Nieh, M.-P.; Wassall, S. R.; de Jong, D. H.; Schafer,̈ L. V.; Marrink, S. J.; Katsaras, J. Cholesterol in Bilayers with PUFA Chains: Doping with DMPC or POPC Results in Sterol Reorientation and Membrane-Domain Formation. Biochemistry 2010, 49, 7485−7493. (127) Brzustowicz, M. R.; Cherezov, V.; Caffrey, M.; Stillwell, W.; Wassall, S. R. Molecular Organization of Cholesterol in Polyunsatu- rated Membranes: Microdomain Formation. Biophys. J. 2002, 82, 285−298. (128) Brzustowicz, M. R.; Cherezov, V.; Zerouga, M.; Caffrey, M.; Stillwell, W.; Wassall, S. R. Controlling Membrane Cholesterol Content. A Role for Polyunsaturated (Docosahexaenoate) Phospho- lipids. Biochemistry 2002, 41, 12509−12519. (129) O’Connor, J. W.; Klauda, J. B. Lipid Membranes with a Majority of Cholesterol: Applications to the Ocular Lens and Aquaporin 0. J. Phys. Chem. B 2011, 115, 6455−6464. (130) Olbrich, K.; Rawicz, W.; Needham, D.; Evans, E. Water Permeability and Mechanical Strength of Polyunsaturated Lipid Bilayers. Biophys. J. 2000, 79, 321−327. (131) Nikitin, A.; Milchevskiy, Y.; Lyubartsev, A. P. AMBER-ii: New Combining Rules and Force Field for Perfluoroalkanes. J. Phys. Chem. B 2015, 119, 14563−14573.

12842 DOI: 10.1021/acs.jpcb.6b05422 J. Phys. Chem. B 2016, 120, 12826−12842