Introduction
Cell membranes are complex assemblies of lipids and proteins that separate the cell interior from the outside environment. Despite the simple bilayer nature of the membrane, lipidomics studies have identified a tremendous variety of lipid species involved in its constitution. Typical plasma membranes (PM) contain hundreds of different lipids.(1-3) The lateral mixing of these components is presumably highly non-uniform.(4, 5) In fact, plasma membranes are close to a miscibility critical point, (6) and large-scale phase separation can occur near physiological conditions.(7, 8) This lateral segregation potential or “patchiness” of the membrane has important implications for many cellular processes, e.g., protein trafficking and aggregation, membrane fusion, and signal transduction. Changes in expression levels of individual lipid species have been implicated in many diseases including: cancers, diabetes, Alzheimer’s disease, HIV entry, and atherosclerosis.(9, 10) Despite recent progress,(11, 12) in vivo characterization of the structural heterogeneity remains very challenging due to the high spatiotemporal resolution required to study fluctuating nanoscale assemblies of lipids and proteins in living cells.
To understand the driving forces governing the structural and dynamical organization of cellular membranes, computer simulation has become an indispensible tool. “Computational microscopy” can be considered a complement to traditional microscopy methods.(13, 14) Notably, simulation studies of membranes are becoming increasingly sophisticated.(15) Nevertheless, membrane complexity is still limited to bilayers consisting of, at most, four or five lipid components.(16-19)
Here we developed a dynamic model of a realistic plasma membrane at near-atomic detail, containing more than 60 different lipids. As PM lipid composition varies significantly between different organisms and cell types and depends on the stage of the cell cycle as well as environmental factors,(1, 2) the composition represents an average idealized mammalian plasma membrane, with phosphatidylcholines (PC), sphingomyelin (SM), and gangliosides (GM) predominantly in the outer leaflet and phosphatidylethanolamine (PE), phosphatidylserine (PS), and other charged lipids in the inner leaflet (Figure 1 and Tables S1–S2). Lipid tails range between fully saturated and polyunsaturated, with a larger fraction of unsaturated tails assigned to the inner leaflet (Figure 1 and Tables S1–S2). Eukaryotic plasma membranes also contain around 20–50% sterols;(1, 20) we included 30 mol % cholesterol in our PM model. To make the simulations computationally feasible, we used a coarse-grained (CG) model in which small groups of atoms (3–4 heavy atoms) are united into chemical entities.(21) The combined headgroups and lipid acyl chains represented in our PM model resulted in 63 different lipid species that, to our knowledge, is by an order of magnitude the most complex bilayer composition that has been simulated to date.
figure
Figure 1. Plasma membrane lipid and cholesterol distribution. (a) Pie charts showing the distribution of the main lipid headgroups and the levels of tail unsaturation in the inner and outer leaflets. (b) A snapshot of the plasma membrane after 40 μs of simulation viewed on the outer leaflet. Underneath are two zoomed in cross sections viewed from the outer and inner PM side, respectively. Cholesterols are colored yellow, lipid headgroups are colored by type (PC, blue; SM, gray; PE, cyan; GM, red; PIPs, magenta; PI, pink; PS, green; PA, white; CE, ice blue; DG, brown; LPC, orange), and tails by number of unsaturated bonds (0, white; 1, light gray; 2, dark gray; 3–6, black). (c) The PMs density profile across the membrane, averaged from 38 to 40 μs, shows the asymmetric cholesterol distribution; 54% in the outer vs 46% in the inner leaflet. (d) The Z-position, with respect to the bilayer center, of the polar head of a cholesterol molecule, demonstrating the fast cholesterol flip-flop rate between the leaflets, see Methods in Supporting Information for analysis and rates.
Determining the lateral distribution of lipids and its influence on membrane protein function is one of the key challenges in cell biology and is crucial for understanding eukaryotic life. Our study provides a detailed view of lipid organization in the PM at near-atomic resolution, across length scales from individual lipids to 70 nm and time scales ranging from nano- to microseconds. Our results provide insight on the much debated trans-leaflet distribution of cholesterol, the extent of nanoscale heterogeneity in the membrane, and preferential clustering of lipid types.
Methods
Bilayer Composition
All major lipid headgroups known to reside in mammalian PMs were included, and their distribution and molar ratios were set to fit an idealized PM.(1, 2, 22-24) The charged species PS, phosphatidic acid (PA), phosphatidylinositol (PI), and the PI-phosphate, -bisphosphate, and -trisphosphate (PIPs) were placed in the inner leaflet and the glycolipids (GM) in the outer leaflet. The zwitterionic lipids PC, PE, and SM are present in both leaflets, with PC and SM primarily in the outer leaflet (70%) and PE in the inner leaflet (80%). A few of the more prominent minor species were also included: ceramide (CER), diacylglycerol (DAG), lysophosphatidylcholine (LPC), with all the LPC in the inner leaflet and CER, and DAG primarily in the outer leaflet (60–65%). A total of 63 different lipid species were created by combining the lipid headgroup with various fatty acid tails, dependent on their relative prevalence.(22, 24-27) Table S2 lists all the lipids used and their relative abundance in the inner/outer PM leaflets. The lipid tails include fatty acids from palmitoyl to lignoceroyl (16–24 carbons), and the level of saturation was varied from fully saturated to unsaturated palmitoleic acid and oleic acid tails all the way to polyunsaturated arachidonic acid and docosahexaenoic acid tails. The cholesterol concentration was 30 mol % in the main PM model and 40 mol % in an additional control model.
Force Fields
The Martini biomolecular CG model(21, 28, 29) was used throughout this study. Details on the lipid topologies used can be found in the Supporting Information, Force Fields section, and all the lipid parameters are available on the Martini portal, http://cgmartini.nl/. The Martini force field has been tested and successfully used in a large variety of lipid membrane simulations and has been shown to reproduce experimental data and all-atom simulation data across a broad range of lipid membrane properties (recently reviewed in ref 30). In particular, lipid phase diagrams are well reproduced with Martini, including liquid-ordered/liquid-disordered (Lo/Ld) phase coexistence in ternary mixtures of dipalmitoyl-PC/dilinoleyl-PC/cholesterol(31, 32) and dioleoyl-PC/SM/cholesterol (Arnarez et al., unpublished) as well as formation of GM1 nanodomains.(18) Phase segregation in mixtures of dipalmitoyl-PC/dioleoyl-PC/cholesterol is only observed at reduced temperature (15 CG waters per lipid, corresponding to 60 real waters per lipid) plus counterions and 150 mM NaCl. After initial energy minimization and equilibrium runs all subsequent simulations were performed with a 20 fs time step, a temperature of 310 K set using the Bussi et al.(35) velocity rescaling thermostat, and a semi-isotropic pressure of 1 bar maintained with a Parrinello–Rahman barostat.(36)
The main PM simulation has ∼20,000 lipids, ∼300,000 CG water beads, ∼6000 Na+ and, ∼3200 Cl–, totaling over half a million particles in a box of 71 × 71 × 11 nm and was simulated for 40 μs. Table S2 lists the quantity of each lipid type present and their respective ratios in the two leaflets. Figure S1 shows equilibration of the area per lipid and the total potential energy within the first 5 μs of the simulation. To check reproducibility of our results, a series of smaller simulations were used (replicas with the same asymmetric lipid composition, with 40 mol % cholesterol, with improved treatment of water and electrostatics, and with the inner and outer leaflet mixtures setup as symmetrical bilayers, see Supporting Information, Methods section). All demonstrated similar overall characteristics as the larger PM simulation, see Figure S4 and Supporting Information, Methods section. In all simulations large-scale bilayer undulations were limited using weak Z-direction position restraint on some of the lipids in the outer leaflet, mimicking the effect of a cytoskeleton network underpinning real plasma membranes. We verified that these constraints did not influence our results, see Figure S7 and Supporting Information, Methods section. To attain the correct cholesterol ratio between the two leaflets, based on cholesterols chemical potential in the two leaflets, the PM simulation cholesterol distribution was iteratively varied until the initial cholesterol ratio remained stable throughout, see and Supporting Information, Methods section. The last 200 ns or 2 μs of each simulation were used for analysis, unless otherwise specified; see Supporting Information, Methods section for details on all analysis methods.
Results and Discussion
Outer Leaflet More Ordered and Moderately Enriched in Cholesterol
Given the i