Simulations of simple linoleic acid-containing lipid membranes and models for the soybean plasma membranes

Lipids are a major group of biological molecules that play a critical role in storing energy, signaling, and forming membranes of cells.¹ These natural occurring building blocks are incredibly diverse. The types of lipids found in cell plasma membranes (PMs) vary widely amongst different types of organisms and amongst different parts of the same organism. This is because the lipid diversity essentially limits and characterizes what a cell membrane can and cannot do. A membrane’s lipid composition defines its structural and functional properties, such as fluidity, permeability, and enzyme activity.² For instance, there is a great amount of variation between the sterol types across organisms. In yeast, ergosterol is the most common sterol.³ In mammals, this tends to be cholesterol. In plants, campesterol, sitosterol, and stigmasterol are present.⁴ These differences influence properties such as lipid domain formation, chain order, and membrane thickness.⁵ 

Compared to bacteria and fungi, considerably less work has been done on the PM composition of a more advanced organism, such as plants. In the past, studies have been hindered because highly enriched fractions of plasma membranes could not be procured sufficiently for the analysis.⁶ However, the later studies have provided more concrete characterizations through different isolation techniques, resulting in PM fractions that may exceed 80% purity for the soybean,⁶ and a more recent study shows that 95% purity is possible for general plants with the techniques introduced in Ref. 7. Additionally, lipid analysis techniques have continued to improve over time. Mass spectrometry, for instance, has been largely successful in recent years due to its high sensitivity and high specificity.⁸ The compositions of plant chloroplast and mitochondrial membranes have been extensively studied due to the early advancements in purifying plant cell organelles—primarily through the systematic use of sucrose density gradients.⁹ Organelle membranes have been found to largely consist of phospholipids, galactolipids, and sterols⁹ and the proportions of the phospholipids are similar compared to those of mammals, but they do not contain sphingolipids and have high concentrations of polyenoic fatty acids.¹⁰ 

Molecular dynamics (MD) simulations are a new method of studying lipid membranes in addition to experimental studies. Although these are limited by time scales proportional to the number of proteins and lipids, they help provide a better understanding of membranes and their interactions. Besides the extensive study of varied types of simplified single-component homogenous phospholipid bilayer^11,12 by the all-atom CHARMM36 lipid force field (AA-C36FF),¹³ binary to ternary phospholipid/cholesterol membrane models have also been studied. Simple models of the chlamydia membrane with cholesterol¹⁴ and ocular lens membranes with a majority of cholesterol¹⁵ using the AA-C36FF have been investigated. However, real biological membranes are generally composed of more than three types of lipid components. The biologically relevant membrane models have been previously developed for lower microorganisms, such as E. coli and yeast. The E. coli inner membrane model was developed with the compositions presented in Ref. 16. This membrane model contains four types of phosphatidylethanolamine (PE) lipids and two types of phosphatidylglycerol (PG) lipids, and three among the six types of phospholipids have the cyclopropane moiety (cyC17:0).^16,17 E. coli cannot produce sterol, so the membrane model lacks any sterol. Moreover, three yeast organelle membrane models were also developed,³ which represent the endoplasmic reticulum (ER), PM, and trans-Golgi network (TGN). Each membrane model has a unique composition of phospholipids and ergosterol,³ which more accurately describes the membranes compared to the earlier developed yeast membrane model.¹⁸ 

The membranes of higher plants, which are potentially more complex, have not been studied as frequently as bacteria and fungi. However, along with the improved accuracy in complex membrane models, there is increased interest to model the plant membrane to facilitate membrane-related studies. The resorcinolic lipids (with the substituted benzene ring as the hydrophilic head group) are found in high concentrations in the membranes of the bran of cereals (wheat, oats, and rye), based on which the simplified membrane model consisted of 1,2-dimyristoyl-sn-glycero-3-phosphocholine (DMPC) and resorcinolic lipids was developed to study the effect of resorcinolic lipids on the PM.¹⁹ The thylakoid membrane, which is the compartment membrane inside chloroplasts of plants and cyanobacteria, is important for the photosynthesis of plants;²⁰ therefore, a more complex thylakoid membrane model has been developed, which consisted of PG phospholipids and three other non-phosphate glycerol lipids.²¹ Soy membranes, however, have not been extensively studied through computational models.

The compositions of soybean PMs vary depending on the species, stage of development, and the part of the plant.^2,6 The two parts of the plant that were examined in this study were the hypocotyl and the root. The hypocotyl is the stem of the germinating seedling, found between the seed leaves and the embryonic root. The root membrane under study was that of the seedling soybean embryonic root. The compositions of these membranes were weighted and averaged from the past experimental studies^4,6,22 and modeled using lipid bilayers consisting of 100 lipids on the top and bottom leaflets. Complex hypocotyl and root membranes are studied along with the basic 1-stearoyl-2-linoleoyl-sn-glycero-3-phosphocholine (SLPC, 18:0/18:2) and 1,2-dilinoleoyl-sn-glycero-3-phosphocholine (DLiPC, di-18:2) systems. Please note that L has been used to represent the lauroyl (12:0) lipid tail, e.g., 1,2-dilauroyl-sn-glycero-3-phosphocholine (DLPC, di-12:0), and also linoleoyl (18:2), e.g., 1-palmitoyl-2-linoleoyl-sn-glycero-3-phosphocholine (PLPC, 16:0/18:2). Therefore, we cannot use L to represent linolenoyl (18:3). To distinguish the lipid tail types, we name lipids based on both tails instead each individual tail, i.e., DLi (di-18:2) and LL (18:2/18:3). The purpose of this paper is to model and characterize the properties of the PMs of two parts of the soybean plant through MD simulations for use in future studies.

Since soybean plant membranes contain lipids with ω-6 18:2 (cis Δ9, 12) linoleoyl chains,^6,22,23 we first built pure lipid membranes to compare to the available experimental data.^24,25 Single-component membranes of SLPC and DLiPC (Figure 1(a)) were generated using the CHARMM graphical user interface (CHARMM-GUI) Membrane Builder.^18,26–28 The bilayers consisted of 40 lipids per leaflet, with a hydration number of 35 water molecules per lipid. Nanoscale Molecular Dynamics (NAMD)²⁹ was used to perform triplicate simulations of each bilayer at a temperature of 303.15 K. The bilayer systems were first equilibrated using the standard Membrane Builder six-step equilibration process.^18,26–28 Production runs were then conducted for 100-124 ns as equilibration for these membranes was relatively fast and all simulations used three independent replicas.

In the primary portion of this study, two soybean PM bilayer systems consisting of 100 lipids per leaflet were simulated to model the hypocotyl and root of the soybean seedling. Experimentally, the composition of the phospholipids (each head group and tail type) in the mature root tissue (1.5-4 cm behind the meristematic zone) of the germinated soybean seedlings was measured.⁶ The molar ratio of the phospholipid to the free sterols and the composition of the free sterols (sitosterol, stigmasterol, and other minor sterol components) in the root plasma membrane of the germinated soybean seedlings were studied in Ref. 4. The full composition of the hypocotyl plasma membrane of soybean seedlings was measured in Ref. 22. The bilayers were built simplified by taking the most common phospholipids and sterols and then proportionally adjusting their percentages. Essential lipids with less than 4% composition were not considered in the building of these model membranes. The CHARMM-GUI Membrane Builder^18,26–28 does not contain all the lipids needed to make these membranes, so a model membrane (Tables I and II) was made by mutating the originally built lipids to the desired lipids (Figure S1 of the supplementary material). Plants produce various sterols but sitosterol and stigmasterol (see Figure 1(a) for the plant sterol structure) are the most common (90% of sterols)² and are used in our models. For both membrane models, the sterol composition was set to be at 29% of all lipids. The hypocotyl PM has lipids that contain mainly 18:2 chains with a small amount of 18:3 chains (Table I). Phosphatidylcholine (PC), PE, and phosphoinositol (PI) are the main lipid head groups with decreasing amounts, respectively. In total, the hypocotyl PM is composed of 9 different lipids. The root PM contains more 18:3 chains compared to the hypocotyl PM. This membrane also includes phosphatidic acid (PA) and phosphatidylserine (PS) lipids and together a total of 10 different lipids were used to model the root PM (Table II). Triplicate runs were equilibrated and ran at 298.15 K and hydrated with 30 water molecules per lipid. As with the single-component membranes, the standard Membrane Builder six-step equilibration process^18,26–28 was used with these modified membranes and were followed by a production simulation for 150 ns (see Figure 1(b) for a snapshot at the end of one simulation).

All MD simulations run using the CHARMM36 (C36) lipid force field¹³ with updates for polyunsaturated chains,³⁰ cholesterol,³¹ and PI lipids.³² The TIP3P water model^33,34 was used, and the counter potassium ions (K⁺) were added to maintain an electroneutral system. A 2-fs timestep was used and coordinates were saved every 2 ps. The constant number of molecules, pressure, and temperature (NPT) ensemble was used. The simulations were performed with 1 bar at the temperatures specified in Table III . The Langevin dynamics was used to maintain a constant temperature, while the Nosé-Hoover Langevin-piston algorithm^35,36 was used for constant pressure. Long-range electrostatics was obtained by using the particle-mesh Ewald (PME) method³⁷ with an interpolation order of 6 and a direct space tolerance of 10⁻⁶. The van der Waals interaction was switched off from 10 to 12 Å by the force-based switching function.³⁸ 

All analyses were performed on the last 60 ns of the simulation as all systems reached equilibrium at or before this time. Data were analyzed for the surface area per lipid (SA/lip), deuterium order parameters (S_CD), electron density profiles (EDPs), tilt angle distributions of sterols, area compressibility modulus, nuclear magnetic resonance (NMR) T₁ relaxation time, and lipid hydrogen bonding and clustering. For single-component bilayers, the SA/lip was obtained based on $X ⋅ Y$ divided by the number of lipids per leaflet. For multiple-component soybean membranes, the X and Y coordinates of the representative atoms of each lipid (O3 for sterols and C2, C21, and C31 for glycerol phospholipids) of the primary cell and the surrounding 8 images are imported into Qhull,³⁹ which constructs the convex polygons for each atom (with all the interior angles less than 180°). The total area of the polygons is equal to the box size, from which the averaged estimated SA/lip for each lipid component can be obtained.

Most analyses were performed using in-house scripts and the CHARMM^40,41 program. The EDP was obtained by recentering the bilayer so that the lipids are at the center of the box, calculating the densities for each atom, based on which the electronic density can be obtained. S_CD was obtained from the formula $S C D = | ⟨ 3 2 cos 2 θ − 1 2 ⟩ | ,$ where θ is the instantaneous angle between the vector of the C–H bond and the bilayer normal. The tilt angle distribution of sterols was obtained by calculating the angles for each residue of the C3 to C17 vector (Figure 1(a)) to that of the bilayer normal. The area compressibility modulus, K_A, was calculated using

where k_B is the Boltzmann constant, $⟨ 𝐴 ⟩$ is the average surface area, T is the temperature, and $σ ⟨ A ⟩ 2$ is the variance of the area.

The average nuclear magnetic resonance (NMR) spin-lattice relaxation times were also calculated for each carbon in the hydrophobic tail of the DLiPC membranes, and they are compared to experiment.²⁵ Assuming pure dipolar relaxation between the heavy-atom nucleus (¹³C) and its N attached protons,⁴² 

where $ℏ$ is Planck’s constant divided by 2π and r_C–H is the effective C–H bond length 1.117 Å.⁴³ The gyromagnetic ratios and Larmor frequencies (radian/s) for hydrogen and carbon are γ_H, γ_C and ω_H, ω_C, respectively. The Larmor frequencies are obtained by ω_C = γ_CH and ω_H = γ_HH, where H is the field strength. μ₀ is the vacuum permeability. The spectral density of the second rank reorientational correlation function, J(ω), is

where P₂ is the second order Legendre polynomial and $μ ^ ( t )$ is the unit vector along the C–H bond direction at time t. The spectral densities were obtained from numerical integration and the values reported below are for ω_C = 90.80 MHz.

The number of intra-lipid (⁠$N HB intra$⁠) and inter-lipid (⁠$N HB inter$⁠) hydrogen bonds was calculated with the definition that the distance between proton and the acceptor pairs is less than 2.4 Å and the angle of donor-proton-acceptor is greater than 150°. The lateral clustering of lipids was analyzed using the same method in Ref. 11, which uses Python scikit-learn⁴⁴ with the density-based spatial clustering of applications with noise (DBSCAN) algorithm.⁴⁵ A maximum distance of 5.5 Å between the center of mass of head group atoms (phosphate and the above atoms) and a minimum cluster size of three lipids per cluster were applied.¹¹ 

The C36 lipid force field has been parameterized for lipids with monounsaturated and polyunsaturated (>4 double bonds) chains,^13,30 but has not been tested on linoleoyl lipids with a polyunsaturated sn-1 chain. Therefore, MD simulations of SLPC and DLiPC were performed to compare with the experimental data and provide a baseline to compare with our soy membrane models. The equilibration of SA/lip was extremely fast (less than 20 ns) as shown in Figure S2 of the supplementary material due to the polyunsaturated fluid membranes. The average SA/lip for DLiPC was 4 Ų higher than SLPC with a single chain dual-unsaturated (Figure 1(a)). K_A is slightly higher for SLPC compared to DLiPC (Table III) suggesting that the presence of the 18:0 chain slightly rigidifies the membrane.

Although direct S_CD NMR measurements for DLiPC and SLPC are lacking, there are S_CD data on the isolinoleoyl (18:2^Δ6,9) with its double bonds shifted up from linoleoyl (18:2^Δ9,12).²⁴ Considering this shifted placement of the double bonds, the order parameters are in excellent agreement between the MD simulations and experiment (Figure 2). The amount of disorder near the two double bonds agrees markedly well between the MD simulations and experiment. S_CDs for the sn-2 chain for DLiPC are nearly identical to the sn-1 chain except for carbon-2 (Figures 2(b) and S3b), and the expected trend for saturated tails is observed for the sn-1 chain of SLPC (Figure S3a of the supplementary material).

There are published data on the ¹³C NMR spin-lattice relaxation times for various carbons for DLiPC,²⁵ which can provide details on isomerization time scales of bonds in the acyl chain. Since the carbon assignment is not straightforward and estimated with NMR, Figure 3 shows the ranked order of the NMR T₁ values from low to high. The agreement between experiment and MD is excellent. Table S1a of the supplementary material contains the raw data for the NMR T₁s and also compares the assigned values for T₁ from an NMR experiment in comparison with the directly calculated T₁s from MD simulations (Table S1b). Previous experimental values for C14/C15 were suggested to be 0.59 and 0.51, respectively, but they are close to T₁s we obtained from our MD simulations for the C4-6/C7-8 position (Table S1b). Although MD simulations show a non-monotonic increase in T₁ from C2 to C18 it is less severe than that suggested previously by experiment.²⁵ 

The density profiles show slight changes between the SLPC and DLiPC bilayers (Figure 4). The main effect is in the thickness of these bilayers (Table IV) which is inversely related to the SA/lip in Table III. The presence of two double bonds for each chain influences all measures of the bilayer thickness (head group, water, and hydrocarbon densities). DLiPC bilayers have more density of unsaturated carbons at the center of the bilayer and the maximal peaks shifted toward the center as compared to SLPC. However, this has a minimal effect on the overall density in this hydrocarbon region (Figure 4(a)). The largest effect on thickness is the decrease in the overall bilayer thickness D_B for DLiPC compared to SLPC (3.4 Å) suggesting water can penetrate significantly deeper into the DLiPC bilayer (Table IV). This agrees with the results that the smallest positive z value of non-zero water electron densities in the SLPC bilayer is 7.1 Å (with EDP 0.000 03 Å⁻³) and that in the DLiPC bilayer is 6.3 Å (with EDP 0.000 03 Å⁻³).

Based on C36 MD simulations with SLPC and DLiPC, the bilayer properties resulted from the lipid force field for polyunsaturated lipid tails³⁰ are in excellent agreement with the limited experimental data for di-unsaturated lipid tails. Therefore, our next step is to develop model membranes for the soy membranes of the hypocotyl and root to structurally quantify how these differ within the soybean plant and compare to the simple single-component membranes with two double bonds per chain.

The SA/lip as a function of time (Figure S4 of the supplementary material) demonstrates that the replicas reached equilibrium for this metric after 50 ns. The average SA/lip of the root membrane is 0.8 Ų lower than the hypocotyl (Table III). This is the result of having slightly more sterols and the presence of PA lipids with small head group in the root membrane. This is evident from the component SA/lip (Table V) with PLPA being several Ų lower than the similar PLPC or PLPE lipids. The component SA/lip for the same lipid shows statistically identical values between these two model membranes, further demonstrating that sterols and PA lipids are influencing the change in the overall SA/lip. K_A of soybean membranes nearly doubled compared to the single-lipid membranes (Table III) due to the presence of the sterols.

S_CD provides a way to compare with the available NMR data (as above) and also to provide information on the acyl chain order in the lipid bilayer. The order parameters for most of the lipids were approximately the same between the hypocotyl and root membranes. However, for lipid LLPE (Figure 5), the order parameters of some carbons of the hypocotyl model were slightly higher than the equivalent values in the root LLPE. The opposite was true for the PLPC lipids on the sn-1 chain (Figure S5 of the supplementary material). Although the shape of the order parameters was more consistent between the hypocotyl and root, S_CD of PLPC in the root membrane tends to be slightly higher for most of the carbons.

Comparing two sterols in each membrane model, the average tilt angle for sitosterol is greater than stigmasterol in both hypocotyl and root membranes, and the difference is more evident in the root membrane (Table VI). The distribution of the tilt angles shows limited variation for each sterol between different soy membranes (Figure 6). The tilt angles of sitosterol are very similar in hypocotyl and root membranes. However, as shown in Figure 6, the probability is slightly larger for stigmasterol having tilts greater than 17.5° in hypocotyl (0.45) than the root (0.41) membrane model. This indicates that stigmasterol is more upright in the root membrane compared to the hypocotyl model, which agrees with slightly lower SA/lip of sterols in the root membranes (Table V). The difference of tilt angle distribution of stigmasterol in two membrane models may be the consequence of the hydrogen bonding or clustering effects, and the analysis results of each will be presented Sections III B 5 and III B 6, respectively, to understand the possible reason.

The general position and shape of the EDP are nearly identical between the hypocotyl and root membrane models (Figure 7(a)). D_HH and 2D_C are statistically identical between the two soy membrane models (Table IV). The main difference is that compared to the hypocotyl membrane, the overall bilayer thickness (D_B, which measures water penetration) of the root membrane model is 0.8 Å smaller. The replacement of the PC/PE lipid with PA allows for more water penetration into the bilayer for the root membranes, which is consistent with the result that single-component PA bilayers have lower D_B than the PC/PE bilayer for the same tail type.¹¹ 

Secondary peaks at roughly ±10 Å in Figure 7(a) are the result of increased density due to the sterols (Figure 7(c)). This increased density is the maximum in the ring density for the sterols (Figures 7(d) and S6b). The EDP for sitosterol has the same shape in the two soy membranes, whereas stigmasterol shows some variation between these membranes. The hydroxyl and ring carbons for stigmasterol show a slight positive shift from the center of the bilayer (Figure S6b of the supplementary material). This results in additional peaks of the tail density that are shifted from the center of the bilayer for the root membrane. The more upright orientation of stigmasterol suggests that this ordering influences the position of this sterol for the root membrane. Moreover, we would like to mention that even though the EDP (Figure 7) indicates the symmetry of the soybean plasma membranes, the plasma membranes of the eukaryotic cells are actually asymmetrical due to the spontaneous lipid flip or by a lipid transporter (“flippase”).⁴⁶ For instance, PS and phosphatidylethanolamine (PE) are found more enriched in the cytosolic leaflet of the plasma membranes.¹ However, the lipid membrane asymmetry is hard to capture due to the limit of the current experimental techniques.

The lateral ordering of lipids is promoted through either hydrogen bonding or electrostatic interactions. Considering the equivalence of ensemble and time average, the result that the sterols form 0.24 inter-lipid hydrogen bonds per lipid suggests that on average sterols form hydrogen bonds with the surrounding lipids during 24% of the time (Table S2 of the supplementary material). Although the PC-PC lipid cannot form hydrogen bonds, these can act as acceptors to other lipids. For the hypocotyl and root membranes, there are 0.40 and 0.33 hydrogen bonds per PC lipid (Figure 8(a) and Table S2). As expected the PE, PI, and PS lipids have the highest occurrence of hydrogen bonding and PA has intermediate values between these and the PC lipids, which are similar in trends to what has been observed in single-component membranes.¹¹ Although several lipids can form intra-lipid hydrogen bonds, only the PI lipids have the significant formation of this class of hydrogen bonding at roughly 0.25 intra-lipid hydrogen bonds per lipid (Table S2).

There are some statistical differences in hydrogen bonding between the hypocotyl and root membranes (Figure 8(a) and Table S2 of the supplementary material). The common PE lipids (PLPE and LLPE) show a reduction in inter-lipid hydrogen bonds for the hypocotyl membrane. There is a reduction in the number of PE lipids in the root membrane that appears to slightly enhance the hydrogen bonding with PE lipids. The number of inter-lipid hydrogen bonds per lipid of sterols does not depend on the membrane type as the glycerol phospholipids. Other common lipids and sterols show no statistical difference in hydrogen bonding between soy membranes. Moreover, comparing sitosterol and stigmasterol in the hypocotyl and root membranes, the lowest number of hydrogen bonds per lipid of stigmasterol in the root membrane (0.230 ± 0.0002) may explain the lowest tilt angle among the four (Tables VI and S2). This suggests that hydrogen bonding is one of the causes of the tilt angle difference of stigmasterol for the two membrane models.

For the root membrane, PL and LL lipids have the same number of lipids with the same head group types (PA, PC, and PE); therefore, it is used for the further lipid type dependence comparison. The LL (18:2/18:3) lipids have both tails unsaturated while PL (16:0/18:2) lipids have only one tail unsaturated (Figure 1 and Figure S1 of the supplementary material). As shown in Table VII, the overall number of hydrogen bonds per lipid between PL-PL and LL-LL are higher than PL-LL. Moreover, the sterols prefer to form hydrogen bonds with PL than LL lipids. Among the two types of sterols, the more saturated sitosterol shows a stronger preference with PL lipids. These results suggest that the lipids with the same or similar unsaturation level prefer to form hydrogen bonds.

Due to the difference of the interaction strength among lipids, the lipid clusters can be formed in the membrane. The number of clusters in each frame (Ncf) and the average number of lipid per cluster (Nlc) fluctuate around stable values (Figures S7a and S7b of the supplementary material), and the overall cluster composition distribution is similar along the simulation time (Figure S7c). These results suggest the metastability of lipid clustering. For the soybean membrane, the van der Waals interaction between the hydrophobic tails is the main cause, the hydrogen bonding and the electrostatic interactions also contribute to clustering formation. The interaction between the sterols and glycerol phospholipids is stronger than the interactions between glycerol phospholipids. As shown in Figure 8(b), the existence of sterols within the soybean membrane induces the formation of large clusters. With the presence of sterols, more phospholipids form clusters due to the strong interactions between sterols and phospholipids. The fraction of sterols forming a cluster (Y_C) is higher than glycerol phospholipids for both hypocotyl and root membranes (Table S3). Sitosterol is less common in clusters than stigmasterol in the hypocotyl membrane, while it forms slightly more in the root membrane. However, the difference of the fraction Y_C is within the standard error (Table S3). Based on the result that sitosterols form more hydrogen bonds with phospholipids and having higher title angles, we expect that sitosterols interact stronger with phospholipids than stigmasterols. Therefore, the situation that sitosterol having lower tendency to form cluster in the hypocotyl membrane is very unlikely. The average fraction of lipid in the cluster (Xc) of PL lipids (0.286) higher than LL (0.269) in the root membrane suggests that PL has a higher tendency to form a cluster, and this is also the case for the hypocotyl membrane (Table S3). Stigmasterol in the root membrane has lower Yc than in the hypocotyl membrane, meaning that slightly less stigmasterol in the root membrane form the clusters, which also may explain the low tilting angle (Table VI).

To study the effect of tail type on lipid clustering in the root membrane, as shown in Figure 8(c), the unsaturation level difference in the tail shows no effect in small clusters (N_PL ≤ 2), in which the probability of (2, 2) cluster composition (nomenclature: (N_PL, N_LL)), is higher than (1, 2) and (2, 1) implying that PL and LL lipids interactions have no preference. However, for the large clusters (N_PL ≥ 3), the equal lipid (diagonal) composition demonstrates the lowest probability, suggesting that PL and LL lipids prefer to form clustering with the same lipid tails due to stronger van der Waals interactions, which agrees with the above hydrogen bonding results. The cluster compositions of sitosterol with PL and LL lipids are slightly different while there is much less difference for stigmasterol (Figure S8 of the supplementary material). The clusters with the number of PL or LL lipids greater than 3 is present in sitosterol, but absent in stigmasterol, implying that sitosterol interacts with glycerol phospholipids stronger than stigmasterol.

The SA/lip of DLiPC in the hypocotyl membrane is 62.3 ± 0.4 Ų (Table V), which is much lower than 70.7 ± 0.2 Ų (Table III) in the pure membrane. The 2D_C considering DLiPC only in the hypocotyl membrane is 32.7 ± 0.2 Å, which is larger than 26.9 ± 0.03 Å (Table IV) in the pure membrane. These results suggest the tighter packing of the hypocotyl mixture membrane than the single-component pure DLiPC membrane. As shown in Figure 9(a), the EDP of the hydrophilic head group region, interface, and tail region of DLiPC in the hypocotyl membrane shift outward which agrees with the elongation of DLiPC in the mixture due to tighter packing. As shown in Figures 9(b) and 9(c), S_CD of a single component membrane such as the DLiPC bilayer seems to resemble that of the hypocotyl membrane. Moreover, S_CD is categorically higher for the hypocotyl soybean membrane lipids than for the single-component membrane lipids. This suggests that the presence of multiple lipids, including sterols, increases the chain order in the soy membrane systems.

The soybean hypocotyl and root membrane models that we present in this paper consist of seven or eight types of linoleoyl phospholipids and two types of sterols (sitosterol and stigmasterol). The compositions of these two membrane bilayers were based on the collective results of multiple existing experimental studies. We believe this is the first study on a complex representation of the plant membrane that contains sterols and provides insight into how membrane properties differ from other membranes in biology. The physical properties of the soy membrane models vary from the previous work on yeast and E. coli cytoplasmic models. The SA/lipid of the soy membrane models lie between the values for the TGN (60.7 Ų) and PM (47 Ų) of yeast³ and is lower than the sterol-free E. coli membrane.^16,17 K_A for the soy membrane models is comparable to the PM of yeast and twice as rigid as the E. coli cytoplasmic membrane. Sitosterol has a similar tilt to that of ergosterol in the yeast PM.³ Although D_B is similar between the soy membrane models and the yeast PM membranes, D_HH and 2D_C are ∼2 Å smaller for the soy membrane models. The different head group and acyl-chain composition are likely the cause for this difference. It is clear that changes in the lipid composition can have an effect to lipid properties but sterol composition appears to have the largest effect to gross structural properties in bilayers and influences lateral packing even in membranes with polyunsaturated fatty acid (PUFA)-containing lipids (soy membrane models). The soy membrane models contain sterols, which results inthe lower SA/lip (Table III) than the thylakoid lipid membrane (66.0 ± 0.3 Ų) of the plant.²¹ 

The results of hydrogen bonding and lipid clustering analyses indicate that the lipids of similar tail saturation interact stronger than the lipids with different tail saturations. The lower tilt angle of the stigmasterol in the root membrane (Table VI) suggests a slightly weaker interaction between stigmasterols and phospholipids in the root membrane, which is consistent with the slightly lower number of hydrogen bonds (Table S2 of the supplementary material) and clusters formation fraction (Table S3b). Along with these complex membrane systems, a DLiPC membrane was generated to compare with the experimental data on DLiPC membranes. Compared to the pure DLiPC membrane, DLiPC in the hypocotyl soybean membrane has a higher SA/lip and a lower S_CD. These results indicate tighter packing of the soybean membrane than the pure PUFA DLiPC membrane. The great agreement of the relaxation time between pure linoleoyl phospholipid membrane models and NMR indicated that the linoleoyl phospholipid is accurate, from which we conclude that the soybean membranes are realistic and they can facilitate the future study on the membranes of soybean and other plants. Moreover, having accurate models for membranes with linoleoyl allow for further studies on membrane-associated proteins in plants.

See supplementary material for more details of many additional membrane properties.

This research is supported by NSF Grant No. MCB-1149187. The high performance computational resources used are Deepthought and Deepthought2, which are maintained by the Division of Information Technology at the University of Maryland.