Tuning lipid mixtures to induce or suppress domain formation across leaflets of unsupported asymmetric bilayers

Symmetric Planar Bilayers are Equivalent to Vesicle Membranes.

Here we produce Montal–Mueller planar lipid bilayers across a hole in a polytetrafluoroethylene sheet (16). The bilayer is surrounded by bulk water on both sides. Fluorescence micrographs in Fig. 1 b–d demonstrate that our planar bilayers have the same progression of phases as membranes in vesicles: first uniform, then with dark domains on a bright background, and then the reverse. Miscibility transition temperatures are equivalent in planar bilayers and in vesicles within ±2–3°C, and domain perimeters in the two leaflets align in registration, giving us confidence that phases observed in planar bilayers are equivalent to those in vesicles. Liquid domains diffuse within both planar bilayers and vesicle membranes. Upon colliding, domains coalesce into one large liquid domain. These critical controls show that any effects of solvent straying into the planar bilayer from the surrounding torus are minimal and that the bilayer adequately captures the intended lipid composition of an equivalent vesicle membrane.

Domains are Either Induced or Suppressed Across Asymmetric Planar Bilayers.

To test whether domains appear in membranes in which one leaflet's composition would phase-separate in a symmetric bilayer and the other's would not, we produced asymmetric bilayers with leaflets of compositions labeled A, B, and C in Fig. 1a. The miscibility transition for a pure A bilayer occurs 2–3°C below room temperature of 24 ± 0.5°C, whereas transitions for pure B and pure C bilayers lie ≈20°C above (15). In other words, whereas A is uniform, B and C lie within the L_d/L_o coexistence region.

An asymmetric A/C membrane, in which leaflet 1 has composition A and leaflet 2 has composition C, does not form domains. In contrast, an asymmetric A/B membrane does form domains (Fig. 2). We conclude that it is necessary, but not sufficient, for one leaflet to have a composition that phase-separates in bilayers (composition B or C) to induce domains in a second leaflet (composition A) whose corresponding bilayer would not phase separate. An A/C membrane remains uniform even though a bilayer of composition A is closer in temperature to its miscibility transition than is composition C.

In the asymmetric A/B bilayer in Fig. 2, we observe three fluorescence intensities from three liquid regions. Through time, the liquid domains diffuse and coalesce. The labeled dark region is constrained to diffuse within a region of intermediate intensity, which in turn diffuses within the bright background [see supporting information (SI) Movie 1].

Our experiments demonstrate that tuning lipid composition in one leaflet produces or suppresses domain formation in both leaflets. This result has significance for asymmetric cell membranes. Plasma membrane inner leaflets are rich in unsaturated phosphatidylcholines (PC), phosphatidylethanolamines (PE), and phosphatidylserines (PS) (1). Vesicles composed of at least one “inner” mixture do not phase-separate (5). We produced an “inner” mixture of 18:1 PC, 18:1 PE, 18:1 PS, and Chol in the molar ratio 29/11/17/33. We find that asymmetric inner/B and inner/C membranes yield the same results as above: Domains form in asymmetric inner/B membranes but are inhibited in inner/C membranes (Fig. 2).

Three Fluorescence Intensities Appear in Each Bilayer Leaflet Separately.

In an asymmetric planar bilayer, domains are induced from one leaflet to the other. In addition, compositions in one leaflet are affected by compositions in the apposed leaflet. To illustrate this concept, the schematic bilayer in Fig. 3b contains six different compositions denoted by six different shades of gray; three are L_o phases, and three are L_d phases. Fig. 3b is consistent with two separate additional experimental observations. In the first experiment, when we construct an asymmetric A/B bilayer in which only the A lipids (Fig. 4a) or only the B lipids (Fig. 4b) contain fluorescent dye, we observe three distinct fluorescence levels. These three levels result from dye partitioning into regions of three different lipid compositions in only one leaflet. Our results are not consistent with a simple model of domains out of registry in the two leaflets, which would never produce three different fluorescence levels in only one leaflet. Our second experiment confirms this result. We produce an asymmetric bilayer in which one leaflet has a composition that is the average of A and B and the other has composition B. Both leaflets contain fluorescent dye, for a total of ≈2.3 nmol on each side of the chamber, of which ≈10 fmol are in each leaflet of the bilayer itself. The B side of the chamber contains a gross excess of 150 μmol sodium dithionite, which quenches all fluorescent dye in the B leaflet. As before, we observe three fluorescence levels from three different lipid compositions in only one leaflet.

The area fraction of the three fluorescence levels can be tuned. Decreasing the fraction of DPPC in the A leaflet decreases the fraction of the darkest fluorescence level of the A leaflet until only two phases remain (Fig. 4, from a to c to d).

Differences between Monolayers and Bilayers.

We know that the domains we observe in bilayers are not simply captured from the monolayers used to assemble the planar bilayer. At 27 ± 0.5°C for a composition between points A and C in Fig. 1a, each monolayer separately exhibits domains occupying ≈50% of the monolayer area (data not shown). These domains disappear when we join the two monolayers into a bilayer. When we cool the bilayer to 24 ± 0.5°C, new domains suddenly appear. The new domains occupy ≈10% of the bilayer area, whether a uniform bilayer at 27°C was cooled to 24°C or whether the bilayer was formed at 24°C. The result is independent of path. In summary, planar bilayers are not simply two apposed monolayers. Our conclusion is supported by the result that large-scale liquid domains are observed in monolayers containing binary mixtures of lipids and Chol but not in bilayers (4, 11).

A Qualitative Model Captures Essential Phase Behavior.

We can qualitatively understand our results with a simplified free-energy model. Consider a lipid monolayer containing only two components (1 and 2) with a positive interaction energy. An equimolar mixture of 1 and 2 lies at B′ in Fig. 3a. Mixture B′ substantially lowers its free energy by separating into two distinct phases with compositions at the L_d and L_o minima. A mixture with composition C′ also separates but produces less L_o phase and lowers its free energy less. Compositions outside the two minima remain in uniform phases (L_d or L_o). For example, an attempt to separate composition A′ into two phases by placing 10% of the mixture's lipids at the L_o minimum fails because the remaining 90% of the mixture's lipids would be at point A″, which has high free energy, resulting in a net unfavorable increase in the system's free energy.

If monolayer leaflets of A′ and B′ couple such that both must phase-separate or both must be uniform, then the free-energy gain of separating leaflet B′ offsets the cost of separating A′ and both separate as in the schematic in Fig. 3a. Unlike coupled A′ and B′ leaflets, coupled A′ and C′ leaflets cannot both separate. The free-energy gain of separating leaflet C′ does not offset the cost of separating A′.

The correspondence of points A′, B′, and C′ in Fig. 3a to our experimental compositions A, B, and C in Fig. 1a is compelling. However, the description above is overly simplified on three important counts. First, if we assume that A, B, and C lie along a single tie-line [which is roughly true (15)], then the x axis in Fig. 3a should be replaced by separate order parameters (φ₁, φ₂) describing each leaflet's composition.

Second, coupling the leaflets via a composition-independent coupling constant b between the Landau order parameters φ₁ and φ₂ modifies the free energy, g, according to

This free energy is similar, although not identical, to one introduced recently (17) and is described in greater detail in SI Text. Setting dimensionless parameters a = 40, b = 5, T₁ = T₂ = 350, and temperature T = 335 results in Fig. 3d. Points along a 45° line in Fig. 3d correspond to symmetric bilayers for which each leaflet has a free energy as in Fig. 3a. All other points indicate asymmetric bilayers. Four local minima exist: two for symmetric bilayers and two for strongly asymmetric bilayers. Increased coupling makes minima for asymmetric bilayers more shallow and brings them closer to the symmetric line. Conversely, symmetric minima become deeper and move away from the origin.

Third, the phase-separated bilayer contains six distinct lipid compositions, represented by six different gray levels in Fig. 3b, of which three are L_o phases and three are L_d phases. Equivalently, one can divide the membrane into three separate asymmetric bilayer-spanning phases as in Fig. 3c. Fig. 3d represents a Helmholtz free-energy landscape. Compositions of the three bilayer phases are points on a plane that is tangent to the bottom of the landscape, here at points L_o^bilayer (or L_o^b), L_d^bilayer (or L_d^b), and mixed. Each phase has a distinct line tension and, in equilibrium, is constrained to diffuse within one of the other phases, much like cells segregate during tissue development due to cell surface tensions (18). An initial composition at the white circle in Fig. 3d separates into all three distinct phases, whereas the composition at the white square cannot separate at all. A small set of compositions between the circle and square would separate into only two phases. Three separate liquid phases fall within the maximum number allowed by the Gibbs phase rule, particularly when one considers that, in the absence of phospholipid translocation, the chemical potential of phospholipids need not be equal across the leaflets. As a result, our system and, indeed, any asymmetric bilayer with leaflets formed from Chol and two phospholipids contains the equivalent of five lipid components, not counting the dye.

Differences Between Planar Bilayers and Supported Bilayers.

Bilayers on solid supports have previously been used to study interleaflet interactions (10, 19), but the proximity of the substrate prevents domains from coming into registration (and reaching equilibrium) when leaflets are deposited from monolayers. Inserting a ≥58-Å polymer cushion between a supported bilayer and its substrate brings domains into registration but still produces results distinct from our unsupported bilayers. For example, domains in supported bilayers are induced from the leaflet near the substrate to the far leaflet but never in reverse, and a uniform leaflet does not suppress domains in a phase-separated leaflet (12). We surmount experimental challenges of supported bilayers by producing asymmetric, unsupported membranes.

Domain Induction and Membrane Proteins.

Induction of domains across leaflets of cell membranes is applicable to problems in immunology, where there is evidence that stimuli in a cell membrane's outer leaflet are colocalized with inner leaflet responses (20–23). Our results compliment opinions that protein–protein interactions are important in signaling but relieve constraints on those proteins. Moreover, we can speculate on ways that membrane proteins could influence domain formation. If the lipid composition in each leaflet is not optimal for inducing domains, it could be modified by the transfer of lipids between membrane leaflets. In turn, lipid transfer could be altered through the presence or conformational change of peptides (24) or translocase proteins (25, 26), as has been observed for alamethicin and PS lipids (27). In Fig. 3a, transfer would transform an asymmetric membrane with uniform leaflets initially at the L_d and L_o minima into a symmetric membrane with domains and overall composition B′. Alternately, increased activity of a lipid translocase may increase the leaflet asymmetry and transform a membrane with liquid domains into a uniform membrane. By actively regulating asymmetry and coupling between membrane leaflets, cells could regulate proteins whose functions depend on the lipid environment.

Preparation and Observation of Planar Membranes.

Here lipids (Avanti Polar Lipids) and <0.2% Texas red dipalmitoylphosphatidylethanolamine (DPPE) are self-assembled at 24 ± 0.5°C in asymmetric Montal–Mueller planar bilayers pretreated with hexadecane in pentane (16). Asymmetric bilayers have been produced previously (e.g., refs. 16 and 28). Pentane was allowed to evaporate before planar bilayers were constructed. Hexadecane was chosen because it is saturated and therefore less likely to photooxidize and because very little hexadecane incorporates between leaflets of planar bilayers (29). We find equivalent miscibility transition temperatures and domain appearance in symmetric bilayers of vesicles without the use of solvent and in planar bilayers with the use of hexadecane. We find that substituting squalene for hexadecane does not change the appearance or behavior of domains or the specific capacitance of bilayers in either symmetric or asymmetric bilayers.

Successful planar bilayers had either no visible hexadecane lenses or small visible hexadecane lenses, and specific capacitances of 0.8–0.9 μF/cm², consistent with solvent-free hydrocarbon thicknesses of 2.0–2.3 nm (30), assuming a hydrocarbon dielectric constant of 2.2 (31). Variation in bilayer thickness results from solvent trapped upon formation. Only the thinnest bilayers with the least solvent, yielding the highest capacitances, were used. Capacitances remained typically within ≈0.5% and always <3% of their initial values. Unless noted, all experiments were conducted at 24.0 ± 0.5°C.

Planar membranes are in metastable equilibrium (32). Moreover, area fractions slowly vary over an hour. In principle, compositions of the planar bilayer, the annulus, and the original stock solution need not be equal. However, our controls show that, on average, symmetric planar bilayers have the same phase behavior as vesicles from the same stock solution, implying that compositions of at least two of the three are similar. We know that there is greater variation in planar bilayer compositions (≤5 mol%) than in vesicles (≈1 mol%) because we see greater variation between trials in area fraction of bright to dark phases. Effects of composition variations are most pronounced near phase boundaries. For example, experiments using composition C were reproduced in ≈80% of trials, as opposed to 100% for compositions farther from the boundary. All experiments were conducted at least three times.

For experiments involving fluorescence quenching, planar bilayers were constructed containing 1–1.5 mol% nitrobenzoxadiazole (NBD)-DPPE. The quencher, sodium dithionite, was purchased at 85% purity from Sigma and mixed in a 1 M stock solution buffered to pH 10.5 in 10 mM 3-cyclohexlamino-1-propanesulfonic acid. The ratio of sodium dithionite to NBD-DPPE in contact with the quencher was ≈50,000:1. Sodium dithionite quenches all available NBD under these conditions. We performed a separate experiment by adding 100 μmol of sodium dithionite to an aqueous suspension of 0.22 μmol of NBD-DPPE in a cuvette (≈500:1) and observed that NBD fluorescence quickly dropped to background levels.

Lipid Oxidation, Translocation, and Interactions with Solid Supports.

Oxidation of lipids is not a concern in our experiments. We find that planar bilayer phase behavior is insensitive to light exposure for >30 min. Moreover, electroformation was not used, the phospholipids had no oxidizable double bonds, and light exposure was an order of magnitude lower than used to investigate vesicles. DiPhyPC was chosen as the low chain-melting temperature lipid because it is saturated. Phase diagrams are similar if DiPhyPC, a lipid with branched acyl chains, is replaced by another low chain-melting temperature lipid, such as dioleoylphosphatidylcholine, with unbranched acyl chains (15, 33).

The translocation of lipids, including our fluorescent probe, between membrane leaflets either by “flip-flop” across the planar bilayer or by transfer through the hexadecane annulus also is not a concern in our experiments. The formation of domains cannot depend on lipid translocation because domain formation occurs quickly whereas translocation proceeds slowly. Domains appeared immediately after construction of both symmetric and asymmetric bilayers or never appeared at all. We captured images of planar bilayers as soon as we confirmed a stable capacitance, usually within 1 min. In contrast, no significant translocation of fluorescently labeled lipid occurs over at least 5 min. We established this result by producing a planar bilayer containing ≈1.5% NBD-DPPE fluorescently labeled lipid, for a total of ≈2.3 nmol of dye on the quenched side of the membrane, of which ≈10 fmol are in the bilayer itself. We then added a gross excess of sodium dithionite (50 μmol), which irreversibly quenches the fluorescence of NBD, to the aqueous compartment on one side of the membrane. Membrane fluorescence was quenched over the course of 1 min, and the new, diminished fluorescence level remained essentially constant for the duration of the experiment, an additional 5 min (see SI Fig. 5).

A slow rate of translocation of fluorescently labeled lipid across planar bilayers is consistent with our knowledge that the transbilayer exchange rate of unlabeled phospholipids by all mechanisms is slow. We find that changes in the appearance of asymmetric bilayers containing domains are imperceptible over at least 15 min and that >1 hour is required for leaflet compositions to become equal. As a specific example, we constructed an asymmetric bilayer in which one leaflet was 100% DiPhyPC, which would have been uniform in a vesicle bilayer. The other leaflet was 33.3%:33.3%:33.3% DiPhyPC/DPPC/Chol, which would have formed domains in a vesicle bilayer. If lipids completely translocated across the leaflets, the composition of the symmetric membrane would have been 66% DiPhyPC, 16% DPPC, and 16% Chol. In the symmetric membrane of a vesicle, this composition would have formed a few dark domains on a bright background, consistent with the lipid composition lying at one of the endpoints of a tie-line in a region of coexisting liquid phases. We formed the asymmetric bilayer. We observed no phase separation for >2 h. Finally, small dark domains appeared, as predicted. Thus, significant lipid translocation took >2 h to complete in this system, whether by flip-flop or via the hexadecane torus. Flip-flop rates have been measured to be several hours for phospholipids in single-component vesicles (34) and seconds for Chol (35).

Important differences exist between planar bilayers and supported bilayers, even when tethers mitigate interactions between the bilayer and the substrate. Domains on supported bilayers are typically captured from a deposited monolayer (10), and supported bilayers often reflect the phase behavior of the two lipid monolayers from which they are made, rather than of a bilayer (11). Lipids in monolayers are usually oxidized to exhibit domains above the surface pressure of 32 mN/m before deposition as a supported bilayer (11). Moreover, Chol compositions in the two liquid phases differ widely in monolayers but are comparable in bilayers, so that tie-lines run oppositely in the two systems (11). Domains successfully induced from one supported leaflet to another can produce counterintuitive dye distributions (7, 10). Finally, the bilayer's proximity to the substrate often means that domain diffusion is too slow to measure on experimental time scales (10). In contrast, planar bilayers produce domains in asymmetric, unsupported, unoxidized model membranes.