phase separation in lipid...

Phase Separation in Lipid Membranes

Frederick A. Heberle and Gerald W. Feigenson

Department of Molecular Biology and Genetics, Field of Biophysics, Cornell University,Ithaca, New York 14853

Correspondence: [email protected]

Cell membranes show complex behavior, in part because of the large number of differentcomponents that interact with each other in different ways. One aspect of this complex be-havior is lateral organization of components on a range of spatial scales. We found that lipid-only mixtures can model the range of size scales, from approximately 2 nm up to microns.Furthermore, the size of compositional heterogeneities can be controlled entirely by lipidcomposition for mixtures such as 1,2-distearoyl-sn-glycero-3-phosphocholine (DSPC)/1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC)/1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-choline (POPC)/cholesterol or sphingomyelin (SM)/DOPC/POPC/cholesterol. In oneregion of special interest, because of its connection to cell membrane rafts, nanometer-scaledomains of liquid-disordered phase and liquid-ordered phase coexist over a wide range ofcompositions.

SCALES OF SPATIAL ORGANIZATION OFCELL MEMBRANES

Cell membranes have daunting complexity,including but not limited to the scales of

spatial organization that emerge from a com-plex system. Complex behavior arises in partfrom the large number of different kinds of lip-ids and proteins, the different ways in whichthese membrane components interact with eachother and with the rest of the cell, and thedynamic processes that locally and dramaticallychange significant fractions of the membrane.To appreciate the nature of this behavior, vari-ous key aspects of the complexity must be exam-ined. Although a complete picture has not yetemerged, the available data are compelling that

compositional heterogeneity, rather than ran-dom mixing of membrane lipids and proteins,describes cell membranes (Lingwood andSimons 2010). Here, we focus on one specificsubset of complex membrane behavior: lateralheterogeneity based on lipid–lipid interactionsin multicomponent bilayer mixtures. The impe-tus for examining such mixtures is the apparentconnection between the coexistence of liquid-disordered (Ld) and liquid-ordered (Lo) phasesobserved in some simple, lipid-only bilayers,and the properties of lipid rafts observed insome cellular membranes. One possible startingpoint for experimental study is to find the sim-plest system that shows a wide range of organi-zation comparable to what is found in cells, thatis, from nanometers to microns. This turns out

Editor: Kai Simons

Additional Perspectives on The Biology of Lipids available at www.cshperspectives.org

Copyright # 2011 Cold Spring Harbor Laboratory Press; all rights reserved; doi: 10.1101/cshperspect.a004630

Cite this article as Cold Spring Harb Perspect Biol 2011;3:a004630

1

to be a four-component lipid bilayer mixture atequilibrium, containing cholesterol.

Membranes of living cells show spatial orga-nization on all relevant size scales, ranging frommolecular lengths to the dimensions of the cellitself. The plasma membrane is the site of con-trolled crossing of molecules and ions betweenthe cell and the environment, and is the aniso-tropic medium in which proteins find theirpartners and catalyze chemical reactions. Thesize scale for such molecular interactions is inthe range of a few nanometers. Some largerbilayer environments are known to have special-ized purposes, for example, membrane regionson the order of 50–100 nm that characterizecaveolae and other sites of membrane fusionand endocytosis. A larger, micron-size scaledescribes the clustering of proteins and lipidsin the membrane site of an immune synapse(Manz and Groves 2010). At a scale of tens ofmicrons, epithelial cell plasma membranes areorganized into the very different morphologicaland compositional regions of apical and baso-lateral domains (Simons and van Meer 1988).

Small size scales of organization in a cellmembrane are difficult to characterize experi-mentally. Several techniques have proven usefulfor detecting heterogeneities on the scale ofsuch small collections of molecules, althoughnot providing quantitative size measurement.Methods that are sensitive to the immediateenvironment of a molecule are probing a veryshort spatial scale, �1 nm. Fluorescence tech-niques including anisotropy (Kinosita et al.1984), lifetime (Haluska et al. 2008), quantumyield (Zhao et al. 2007a), and quenching byspin-labels (London and Feigenson 1981), aswell as the ESR spectrum itself (Schneider andFreed 1989), are such highly locally sensitivemethods. Although first-order phase separationcreates different environments on both smalland large spatial scales, simple nonideal lipidmixing also creates different local environmentsat the �1 nm scale. Experimental sensitivity tosuch small heterogeneities is needed, along withquantitative measure of spatial scale and (ifpossible) lifetime of any such domains. Super-resolution imaging is highly promising in thisregard. In a recent study, STED microscopy of

fluorescently labeled lipids revealed transienttrapping of sphingomyelin for an average timeof �15 msec within regions of diameter ,20nm (Sahl et al. 2010). Of course, these STED-detected domains themselves must last longerthan 15 msec. This highly significant findingrules out simple nonideal mixing, which wouldresult in molecular clusters that dissipate on atimescale of tens of microseconds (Abney andScalettar 1995). In addition to techniques em-ploying a single probe species, FRET betweendonor and acceptor fluorophores is in principlecapable of providing a measure of domain sizein the range 1 – 10 x R0, which is in the rangeof 2–80 nm for available membrane fluorescentprobes (Feigenson and Buboltz 2001). Finally,we note that any technique involving additionof an extrinsic probe carries the caveat that theprobe molecule might perturb the native spatialorganization of the membrane.

MODELING SPATIAL ORGANIZATION INCELL MEMBRANES

The size scales, time scales, and patterns of het-erogeneity found in membranes of live cells canbe treated in terms of general physical principles(Seul and Andelman 1995; Imperio and Reatto2006; Elson et al. 2010; Fan and Sammalkorpi2010). One way to organize such general con-siderations is to classify the driving forces asequilibrium or nonequilibrium processes. Forseparated phases at equilibrium, competing in-teractions can give rise to patterns of spatialorganization with variations in size, lifetime,and morphology. For example, when line tensiondominates any other interactions that dependon domain (perimeter) size, coexisting liquidphases round up into large circular domains con-taining vast numbers of molecules to minimizetheir perimeter. This simple morphology can bemodulated when interactions of sufficient mag-nitude competewith line tension, so that patternsappear—a maze of stripes, tiny round domains,curved lines, and branched lines (Seul and Andel-man 1995). Also at equilibrium, and dependingon proximity to a critical point, fluctuating do-mains showing a range of size and timescalesappear (Honerkamp-Smith et al. 2008).

F.A. Heberle and G.W. Feigenson

2 Cite this article as Cold Spring Harb Perspect Biol 2011;3:a004630

Processes occurring far from equilibriumcan strongly influence the nature of coexistingdomains. An especially simple case is the failureof phase-separated Ld þ Lo macroscopic do-mains to merge into an equilibrium largedomain because of a high-energy barrier frommembrane curvature at the domain boundaries(Semrau et al. 2009). Nanodomains can also bestabilized by entropy: If line tension is suffi-ciently low, the entropic penalty for mergingmany small domains into a single large domaincan produce a “quasi-equilibrium” state of do-mains smaller than 50 nm (Frolov et al. 2006).A more complex and interesting process, alsofar from equilibrium and connected to otherevents in the cell, is the removal and deliveryof patches of membrane having a compositiondistinct from that of the larger membrane.This dynamic process can generate a wide rangeof domain sizes and morphologies (Fan andSammalkorpi 2010).

LIPIDS FOUND IN CELL MEMBRANES

Considering only the lipids (a mixture chosenby evolution in part to spontaneously formthe bilayer phases that are the fundamentalcell membrane structure), a first step for studyof heterogeneity is to find the lipid compositionof cell membranes (van Meer et al. 2008). Acomplete analysis would involve assaying thelipids of the given cell membrane, both qualita-tively and quantitatively, and finding the partic-ular composition of each leaflet. The latterhas been a vexing problem for decades. Wholemembrane assays are more feasible, and large-scale efforts from “lipidomics” initiatives(Wenk 2005) have been successful in purifyingparticular cell membranes and measuring theiroverall composition, albeit without being ableto characterize separate leaflet compositions sofar. Different types of lipids have been foundto predominate in different organisms, and indifferent membranes within a given cell. Evenwith so much new information, the connectionbetween events of cell biology and particularlipid compositions—a kind of “lipid roleassignment”—can be made in but a few cases.These exceptions include the phosphoinositide

binding of particular proteins to the plasmamembrane inner leaflet (McLaughlin and Mur-ray 2005) and polyunsaturated acyl chains oflipids in the retinal rod, which are neededfor full activity of rhodopsin (Polozova and Lit-man 2000). More information is known aboutplasma membranes than about any other cellmembranes, making the plasma membrane astarting point for thinking about lipid roles inmammalian cell membranes.

In the favorable case that a particular mem-brane can be obtained in high purity, the overalllipid composition can be found. In comparisonwith the total lipids of a cell, two categories of lip-ids stand out for their abundance in the plasmamembrane compared with othercell membranes:high melting temperature lipids, in particularsphingomyelin and gangliosides, and cholesterol(van Meer et al. 2008). These types of lipids arealways present in the plasma membrane alongwith much lower melting temperature lipids,which in the outer leaflet are mostly phosphati-dylcholines (Gennis 1989). After so many yearsof effort, it is perhaps surprising that the preciselipid composition has not been separately deter-mined for outer and inner leaflets for any plasmamembrane. Because the best understood case isthat of the mature human erythrocyte, we willtake these results as an approximate model forother mammalian plasma membranes: the outerleaflet has all of the gangliosides, the great major-ity of the sphingomyelin (SM), and much of thephosphatidylcholine (PC) and cholesterol; theinner leaflet has most of the membrane’s phos-phatidylethanolamine (PE), possibly all of itsphosphatidylserine (PS) and phosphatidylinosi-tol (PI) and some PC, and much cholesterol(Gennis 1989). It is important to rememberthat these numbers reflect an average over theentire membrane. Even if we know the composi-tion of a leaflet, this does not imply that exactlythis composition is found at all locations overthe entire membrane surface, (e.g., apical andbasolateral regions of an epithelial cell). In addi-tion, manyother lipid components, including theplasmalogen versions of glycerophospholipids,are also present. The phase behaviorof these mol-ecules in mixtures with other membrane compo-nents is largely unexplored.


Cite this article as Cold Spring Harb Perspect Biol 2011;3:a004630 3

A more complete story of plasma mem-brane behavior should describe both the sepa-rate behavior of each leaflet, and the couplingbetween the leaflets (Collins and Keller 2008;Kiessling et al. 2009). The understanding ofthis “asymmetry” of the bilayer membrane isstill at an early stage. Almost the entirety ofwhat we know about the phase behavior ofthe lipid mixture in each leaflet of the plasmamembrane comes from study of mixtures thatseparately model each leaflet.

CAN NEGLECTING MEMBRANEPROTEINS BE A STARTING POINT FORSIMPLE MODELS?

Proteins and lipids comprise approximatelyequal fractions of the membrane, an obviousproblem for the relevance of lipid-only mixturesto model acell membrane. Yet, a closer inspectionof the protein mass distribution reveals that themajority (perhaps 80%–85%) is located withinthe aqueous milieu either outside or inside thecell. The remaining �15%–20% of the proteinmass is located within the lipid bilayer (Sheetz1993). This consideration leads to an estimatethat the membrane bilayer, which we can defineas the matter bounded by the lipid headgroupsof each leaflet, is 10%–15% protein. Indeed,the very nature of the membrane as a bilayerphase is entirely a property of the lipids, not theproteins. Therefore, lipid-only models are likelyto inform about some aspects of the nature ofcell membranes.

INFORMATION IN LIPID COMPOSITIONALPHASE DIAGRAMS

The history of composition-dependent phasediagrams of lipid bilayer mixtures dates fromthe 1960s (Phillips et al. 1970). Early experimentsshowed that single-component bilayers exist in agel (Lb) phase at low temperatures, characterizedby a high trans/gauche conformer ratio in thehydrocarbon chains, as well as long-range two-dimensional positional order. Following heating,these bilayers will ultimately pass through a first-order melting transition to a disordered fluidphase (Ld), accompanied by a decrease in chain

segmental order and the loss of long-range cor-relations in lipid position. Subsequent work re-vealed transitions to a third lamellar phase (theliquid-ordered, or Lo phase), found only in thepresence of substantial amounts of cholesterol,and possessing extraordinary properties: theabsence of long-range positional order as in afluid, but with a high degree of chain segmentalorder as in a gel (Wu and McConnell 1975; Vistand Davis 1990). Early work also revealed thatreliable data could be obtained to establish simplecorrelations, for example, that more structurallydissimilar lipids in a mixture give rise to widerregions of immiscibility (Tenchov and Koynova1985; Huang and Li 1999).

The apparent connection between themembrane raft hypothesis and coexistence ofdistinct fluid phases (Ld þ Lo) in model sys-tems spurred the study of biomimetic bilayermixtures in the previous decade, with thegoal of solving complete phase diagrams. Theminimal systems that maintain biological rele-vance yet are amenable to relatively completestudy contain both high-melting and low-melting temperature (TM) lipids and choles-terol (Veatch and Keller 2003). A rich phasebehavior is found in these systems, includinga large region of coexisting ordered- anddisordered-liquid phases at biologically rele-vant cholesterol concentrations (Marsh 2009).Furthermore, the size distribution of phasedomains in the Ld þ Lo region depends onthe low-TM lipid: Strongly perturbing lipidslike DOPC or diphytanoyl-PC generate mi-cron-sized domains (Veatch et al. 2006; Zhaoet al. 2007a), whereas biologically abundantlow-TM lipids, such as POPC or 1-stearoyl-2-olyeoyl-sn-glycero-3-phosphocholine (SOPC),yield nanometer-sized domains (de Almeidaet al. 2003; Pokorny et al. 2006). In the follow-ing discussion, we consider the interactionsthat give rise to these phenomena.

Lipid Clustering from Nonideal Mixing inSingle-Phase Regions

We begin by addressing cell membrane orga-nization in the small size scale range of 5–200 nm. A 200 nm bilayer patch contains about



60,000 lipids per leaflet. A domain of this size iscertainly large enough to be considered a phase,or even to have coexisting phases within thedomain. But what is the nature of a 5 nm patchcontaining only approximately 40 molecules?

The thermodynamics of mixing, togetherwith Monte Carlo computer simulations, pro-vides a powerful, systematic basis for examina-tion of lipid distributions based on lipid–lipidinteraction energies. A simple visual inspectionof Monte Carlo snapshots at equilibrium, suchas those shown in Figures 1 and 2, is instructive.Cell membranes are far from random mixturesof lipids and proteins, but we start by consider-ing exactly this case: an ideal mixture of, forexample, phosphatidylcholines, defined by anexcess mixing energy DEm ¼ 0 (Guggenheim1952). Figure 1A shows that, for an idealequimolar binary mixture with only nearest-neighbor interactions (e.g., no long-range orhigher order interactions), groupings of threeto five like lipids in each leaflet are commonly

observed (Huang and Feigenson 1993). First-order phase separation occurs when DEm ex-ceeds 0.55 kT (Guggenheim 1952), and resultsin large phase domains seen in Figure 1D,E.Intermediate values ofDEm generate small clus-ters. For example, with DEm ¼ 0.4 kT, the sizesof the numerous clusters are in the range 10 to20 lipids (see Fig. 1B), a “domain size” of�3 nm (Huang and Feigenson 1993). At DEm¼ 0.5 kT, numerous clusters of approximately100 lipids form, as shown in Figure 1C,F. Suchsmall clusters from nonideal mixing have ashort lifetime on the order of microseconds,disappearing and forming again elsewhere aftermany nearest-neighbor exchanges, each �1027

s (Abney and Scalettar 1995).Further insight into lipid mixing comes

from considering the more complex behaviorof multicomponent bilayer mixtures. Figure 2shows snapshots of mixing behavior in whichthe interaction energies are held constant, butthe concentration of a third component is

Figure 1. Lattice snapshots for Monte Carlo simulations of binary mixtures with different pairwise interactionenergy DEm. (A) A random mixture (DEm ¼ 0) is characterized by groupings of 3 to 5 lipids. (B) An unfavor-able interaction ofDEm ¼ 0.4 kTresults in clusters of 10 to 20 lipids, or a domain size of �3 nm. (C) AtDEm ¼0.5 kT the system is close to phase separation. An enlargement of the snapshot (F) reveals large clusterscomposed of hundreds of lipids. (D) At DEm ¼ 0.55 kT the clusters coalesce, indicating a phase transition.(E) DEm ¼ 0.6 kT. The phase-separated mixture is characterized by a single domain of each phase.



varied. For comparison with Figure 1, MonteCarlo simulation snapshots shown in Figure 2of a three-component mixture are arranged inan order that leads up to phase separation inFigure 2E. In these simulations, increasing cho-lesterol concentration leads to increased mixingof the components. At a cholesterol concentra-tion of 35 mol % the mixture shows highly non-random mixing, with clusters of approximately20 lipids (Fig. 2A). Smaller cholesterol concen-trations lead to increasing cluster sizes. Forexample, Figure 2C,F show clusters of severalhundreds of lipids at 25 mol % cholesterol, cor-responding to a domain size of approximately10 nm.

Considering these two simulation studies oftwo- and three-component mixtures, what dowe learn about the spatial distribution of lipids?First, even well-mixed or random mixturesshow numerous clusters of 3 to 5 similar lipids.

Second, lipid clusters grow as the mixture ischanged to approach a boundary of phase sep-aration, but only in a limited way: Cluster sizesapproach a maximal range that is similar for themixtures with different numbers of compo-nents. However, a short distance past the phaseboundary in composition space (Fig. 2) or inenergy (Fig. 1) results in enormous, abruptchange in size of domains.

Structured One-Phase Mixtures

Before exploring further the nature of lipidphase separations, another possibility shouldbe discussed: A particular spatial organizationin a lipid mixture might be described as a struc-tured one-phase region instead of phase co-existence. In this state of matter, moleculesare organized on the nm and tens of nm scale,but chemical potentials can vary with mixture

Figure 2. Lattice snapshots for Monte Carlo simulations of ternary mixtures. (E) An equimolar binarya/bmix-ture with an unfavorable pairwise interaction energyDEab ¼ 0.8kT is characterized by coexistence ofa-rich andb-rich phases. (A–D) Although maintaining the ratio of components a and b, a third component, g (whichcould be cholesterol), is added that interacts favorably with both a and b (DEag ¼20.8 kT, DEbg ¼21.2 kT).(D) At xg ¼ 0.20, large clusters of b within the a-rich phase (and vice versa) are evident. (C) Long-rangestructure is broken up at xg ¼ 0.25. An enlargement of the snapshot (F) shows clusters of hundreds of lipids,and the uneven distribution of component g (gray) between a-rich (white) and b-rich (black) clusters. Furtheraddition of component g reduces the size of clusters. (B) Snapshot for xg ¼ 0.30 and (A) xg ¼ 0.35.



composition (unlike a two-phase tieline, inwhich chemical potentials are constant even asthe bulk composition varies). Examples of suchstructured one-phase mixtures are micelles,formed as a high-order association reaction ofamphiphiles (Tanford 1980), and microemul-sions, which can form when surface tension islow enough compared with kT that large struc-ture breaks up (Gelbart et al. 1999). Such struc-tured single phases provide distinctly differentpossibilities for the local environment of indi-vidual molecules. As an example, DPH fluores-cence increases greatly when detergent formsmicelles, revealing a new environment in whichDPH quantum yield is increased (London andFeigenson 1978).

An important task when modeling cellmembranes is to faithfully describe the spatialorganization of the system of interest. At leastinitially, this does not necessitate distinguishingbetween structured one-phase matter or two-phase nanodomains of Ld þ Lo. Characteriz-ing the spatial organization would includedetermining the size, shape, lifetime, and con-nectivity of the heterogeneities, whatever thenumber of true phases. But the distinctionbetween one-phase and two-phase is of interest.First, along particular directions (namely thethermodynamic tielines of a two-phase system),chemical potential of each component is con-stant. In this case, we can in principle measurethe very informative partitioning of a varietyof membrane-bound lipids and proteins be-tween coexisting Ld and Lo phases, a potentialstep in establishing the rules that describe pro-tein partitioning in cell membranes. Moreover,systematic measurements of Kp might providean experimental test to establish whether nano-scopic domains are one-phase or two-phase.After all, in a one-phase mixture chemicalpotential would vary continuously withoutshowing fixed values along particular lines. Ifthe nanodomain region of a ternary mixturewere a structured one-phase compositional re-gion, then we would expect to find a first-ordertransition to a two-phase region somewhere incomposition space. Sharp transitions can bebiologically useful, for example, in amplifica-tion of stimuli (Stadtman and Chock 1977).

Coexisting Phases: The Special Behaviorof Cholesterol in Bilayers

Phases can separate when unfavorable interac-tions are sufficiently large, resulting in richphase behavior for some lipid mixtures. In ageneral sense, the two-chain lipids that formthe bilayer phase are so very structurally differ-ent from cholesterol that we expect wide regionsof immiscibility. But cholesterol, despite itshigh solubility in phospholipids—about 67mol %, or two cholesterol molecules for eachphospholipid molecule at saturation for manyPCs—ties up its phospholipid neighbors toaccommodate what is effectively strong choles-terol–cholesterol repulsion (Huang 2002). Atcholesterol concentrations greater than that ofcertain regular arrays, cholesterol chemicalpotential spikes (Ali et al. 2007). At these specialconcentrations, a different cholesterol-richphase separates from the cholesterol-saturatedmatrix. This behavior was first described explic-itly for the formation of cholesterol monohy-drate from a cholesterol-saturated Lo bilayer(Huang et al. 1999), but applies to other sep-arations of cholesterol-rich phases as well,including formation of Lo from a cholesterol-saturated Lb phase.

One-, two-, and three-phase regions arefound in the case of ternary mixtures with cho-lesterol that contain, for example, the phospho-lipid pairs SM þ DOPC (Veatch and Keller2005), or 1,2-dipalmitoyl-sn-glycero-3-phospho-choline (DPPC) þ DOPC (Veatch and Keller2003), or DSPC þ DOPC (Zhao et al. 2007a).The interaction of DOPC with the high-meltinglipid is so unfavorable that over an extensiveregion of composition space, phase separationof Ld and Lo occurs. Phase domains are macro-scopic, and a single round domain can extendfor many microns. This observation also impliesline tension between Ld and Lo domains that issufficiently large to drive their coalescence.

Figure 3 shows the phase behavior over allpossible compositions at 238C for the particularmixture DSPC/DOPC/cholesterol, but we notethat this can be considered a general phase dia-gram for mixtures of cholesterol with a high-TM

and low-TM lipid. As cholesterol concentration



increases in the Ld phase (arrow 1 in Fig. 3),apparently cholesterol continuously dissolveswithout phase separation right up to its maxi-mum solubility of 67 mol %. The same is nottrue of the gel phase (arrow 3 in Fig. 3). In theLb phase, cholesterol reaches a maximum solu-bility of about 16 mol % (for SM, DPPC, orDSPC in mixtures with DOPC, POPC, orSOPC). Cholesterol chemical potential risessteeply at this special concentration, resultingin the formation of a Lo phase having the highercholesterol concentration of �27 mol %. Theconcentration of 16 mol % corresponds toeach cholesterol molecule in the gel being sur-rounded by about six phospholipids—everyphospholipid in the gel “solvating” but onecholesterol in the cholesterol-saturated solidLb phase. Cholesterol at concentration .16mol % must be accommodated in a differenttype of lattice than that of the Lb phase, namelythat of the Lo phase. Apparently, lateral androtational positions have larger ranges available

for optimal cholesterol solvation when notdictated by the demands of the highly-orderedLb lattice. Perhaps the details of cholesterolmolecular shape make anisotropic demandson its now approximately four or fewer phos-pholipid neighbors in Lo, rather than the sixphospholipid neighbors of the cholesterol-saturated Lb phase. Packing adjustments inLo, unconstrained by the Lb lattice, seem toenable each cholesterol to be shielded fromwater by fewer lipid headgroups.

As cholesterol concentration increases,eventually reaching its maximum solubility inthe Lb phase, the cholesterol also responds tothe presence of any coexisting Ld phase. Forexample, Figure 3 arrow 2 corresponds to theaddition of cholesterol to a two-phase mix-ture of Ld (DSPC-saturated DOPC) and Lb(DOPC-saturated DSPC). In this interesting sit-uation, the cholesterol has modest preferencefor Lb over Ld, but dissolves into both phases.In Ld and Lb, cholesterol chemical potential

Cholesterol

0.67

0.40

0.27

0.16

High–TMLow–TM

LdLd + Lβ

Ld + Lo

Ld + Lo + Lβ

1 2 3

Lo

LoLβ

Lβ

+

Figure 3. Illustrative phase diagram for a ternary lipid mixture containing low- and high-melting temperaturelipids and cholesterol. Tielines are shown in phase-coexistence regions, and the Ld þ Lo critical point is markedwith a star. The effect of cholesterol addition to Ld (arrow 1), Lb (arrow 3), or phase-separated Ld þ Lb mix-tures (arrow 2) is discussed in the text.



does not rise very steeply as its concentrationincreases (until near saturation in Lb), and infact is lower in Lb than in Ld, as evidenced byits higher concentration in the Lb phase thatis in equilibrium with Ld phase. However, asteep jump in cholesterol chemical potentialoccurs in the Lb phase for concentration greaterthan �0.16 mol %. Once the Lo phase separatesout from Lb, more added cholesterol entersboth Ld and Lo, with an initial preference ofabout 2.5 for Lo over Ld, a preference that getssmaller as cholesterol concentration increasesin the Ld phase. In the DOPC-rich Ld phasethat is in equilibrium with a DSPC-rich orSM-rich Lo phase, at a cholesterol concentra-tion of about 40 mol %, the phases becomeindistinguishable; this is the critical point. Inthe POPC-rich or SOPC-rich Ld phase that isin equilibrium with a DSPC-rich Lo phase,the lower cholesterol concentration of about30 mol % makes the Ld and Lo phases indistin-guishable (Heberle et al. 2010).

However, in the POPC- or SOPC-contain-ing mixtures, in the Ld þ Lo region far fromthis critical point, another interesting behavioris manifest: the phase domains are smallerthan the wavelength of light, about 200–300 nm, and so are not resolved in ordinary-resolution optical microscopy. However, FRET(Silvius 2003), fluorescence anisotropy (deAlmeida et al. 2003), ESR spin label orderparameter and rotational diffusion (Heberleet al. 2010), and d-Lysin-induced dye efflux(Pokorny et al. 2006) are all consistent withcoexistence of Ld and Lo phases.

If nanodomains of Ld þ Lo can be treatedas two-phase coexistence, some questions arise:

1. Given that minimization of domain perime-ter is driven by lowering the free energy fromline tension between Ld and Lo domains,what is an opposing interaction that wouldbecome unfavorable as domain size grows?

The result of such competing interac-tions would be modulated phase behavior.Depending on the relative magnitude ofthe competing terms, a variety of domainmorphologies and size scales can be ob-served, including round domains, straight

lines, branched lines, and maze patterns(Seul and Andelman 1995). Possible candi-dates for such an interaction that becomesunfavorable as domain size grows can beidentified: (1) Lipid intrinsic curvature hasbeen proposed to favor domains of a partic-ular size (Brewster and Safran 2010); (2) Thespecial demand of each individual choles-terol molecule to be “solvated” by neighbor-ing lipids is not symmetrical around eachcholesterol molecule, and each “lipid solva-tion shell” could interact unfavorably withothers to oppose domain growth.

2. To what extent are the chemical and physicalproperties of any nanodomains in POPC-and SOPC-containing mixtures similar to,or different from, the macroscopic phase do-mains of Ld þ Lo in the related mixturesSM/DOPC/chol or DSPC/DOPC/chol, forexample?

Simons and colleagues have emphasizedthat the partition of “raft-associated” pro-teins in cell membranes differs from thepartition observed in simple lipid mixturemodels of cell membranes (Kaiser et al.2009). But perhaps protein partition behav-ior for coexisting macroscopic domains ismisleading. This hypothesis can be testedby comparing the partitioning of even simplefluorescent dyes in the nanoscopic and mac-roscopic mixtures. Of great interest would beto find the Kp of various lipids, peptides, andproteins. Furthermore, the physical proper-ties of order parameter and diffusion coeffi-cient differences between nanoscopic andmacroscopic domains might inform as tothe fundamental domain characteristics.For example, the order parameter S0 of theacyl chain spin label 16-PC responds in asimilar way to the phase separation of eithermacroscopic or nanoscopic domains, yet thenanoscopic Ld-phase domains show greaterorder than macroscopic Ld-phase domains(Fig. 4).

3. Are interesting morphologies observed whenLd and Lo phases coexist?

Given two-phase coexistence, Seul andAndelman (1995) and others have described



modulation of spatial organization thatshows a number of possible morphologies,including the possibility of progressionthrough different morphologies. For mix-tures that show a range of size scales, for ex-ample the four-component mixture DSPC/DOPC/POPC/chol, the extremes of fluidphase morphology are known: micron-sized, round domains at 0 mol % POPC,and domains much smaller than 300 nm(and likely in the range of 2–5 nm) at0 mol % DOPC. However, in the intermedi-ate regime, different morphologies are fre-quently observed (Fig. 5). The systematicexamination of such behavior in lipid-onlymodels might inform a search for similarmorphologies in cell membranes.

4. Are the domains observed with light mi-croscopy strongly sensitive to intense illu-mination?

For any imaging techniques in whichthe question is a transition from “close-to-phase-separated” to actually phase-sepa-rated, we expect sensitivity to light-inducedmacroscopic domains. This artifact can beminimized by use of low dye concentration,multiphoton excitation, and sensitive cam-eras (Morales-Penningston et al. 2010).

Value of a Phase Diagram for aFour-Component Bilayer Mixture

Because DSPC/DOPC/chol has macroscopicLd þ Lo phase separation, whereas DSPC/POPC/chol has nanometer-scale Ld þ Lo sep-aration, a minimal system with domain sizecontrolled by composition would be the four-component mixtures DSPC/DOPC/POPC/chol. Finding the boundaries of phase regionsin four-component mixtures will be data-intensive because a three-dimensional compo-sition space must be examined. In addition tophase boundaries, domain size scales must bedetermined; GUV imaging and FRETare usefulfor obtaining such information. Domain life-time distributions are necessary for under-standing their role in the enzymatic activity ofmembrane proteins. Are there distinct proteinbehaviors in response to distinct lipid phasebehaviors, or does a short domain lifetimeeffectively present the protein with an averagedenvironment?

Figure 5. Confocal microscopy of giant unilamellarvesicles reveals interesting domain morphology ina four-component mixture that is intermediate insize between the smallest nanoscopic domains ofDSPC/POPC/chol and the macroscopic domainsof DSPC/DOPC/chol. Vesicle composition DSPC/DOPC/POPC/chol ¼ 45/4.5/25.5/25. Scale bar10 microns.

0.25

0.15S0

0.05

0 0.25 0.5χDSPC

0.75

Figure 4. ESR reveals similarity of macroscopic andnanoscopic phase properties. Compositional trajec-tories run in the approximate direction of Ld þ Lotielines (see Fig. 3) and differ only in the identityof the low-TM lipid. Composition-dependent orderparameters obtained from ESR spectral simulationsin DSPC/DOPC/chol (triangles) and DSPC/POPC/chol (circles).



CONCLUSIONS

It is clear that cell membranes possess domainswith a range of sizes. Compositional heteroge-neities also occur in lipid-only bilayer mixtures,with a range of sizes and perhaps other proper-ties. Model mixtures in which domain size iscontrolled by composition from nanometer-scale to microns would provide a valuableexperimental system for examining this aspectof membrane rafts. So far, no such systematicstudy has been published, though we haverecently identified a four-component mixturethat meets this criterion. Also of interest forthe behavior of cell membranes, even singlephase nonideal lipid mixtures can be inducedto increase the size of their heterogeneities.Here, a case that plagues the use of fluorescencemicroscopy is the introduction of artifactualvisible domains by intense illumination (Zhaoet al. 2007b). This might be caused by free rad-ical initiated polymerization that starts from thereactive excited singlet state (Ayuyan and Cohen2006). More interesting examples come fromaggregating ganglioside by binding choleratoxin B subunit, resulting in large domains(Hammond et al. 2005). Such controlled in-crease of domain size by protein binding orcross-linking could be used by cells in a con-trolled fashion to effect cellular functions (Ling-wood et al. 2008). Furthermore, perhaps the cellhas mechanisms for the complementary role ofcontrolled decrease of domain size, for example,via protein “obstacles” that interfere with phaseseparation.

ACKNOWLEDGMENTS

We gratefully acknowledge the help of J. Huangin teaching us details of the Monte Carlo simu-lation, and J. Wu for the GUV image. Supportwas received from research awards from theNational Institutes of Health R01 GM077198and the National Science Foundation MCB0842839 (G.W.F.). F.A.H. was supported inpart by National Institutes of Health researchaward 1-T32-GM08267.

REFERENCES

Abney JR, Scalettar BA. 1995. Fluctuations and membraneheterogeneity. Biophys Chem 57: 27–36.

Ali MR, Cheng KH, Huang J. 2007. Assess the nature ofcholesterol-lipid interactions through the chemicalpotential of cholesterol in phosphatidylcholine bilayers.Proc Natl Acad Sci 104: 5372–5377.

Ayuyan AG, Cohen FS. 2006. Lipid peroxides promote largerafts: Effects of excitation of probes in fluorescencemicroscopy and electrochemical reactions during vesicleformation. Biophys J 91: 2172–2183.

Brewster R, Safran SA. 2010. Line active hybrid lipids deter-mine domain size in phase separation of saturated andunsaturated lipids. Biophys J 98: L21–L23.

Collins MD, Keller SL. 2008. Tuning lipid mixtures to induceor suppress domain formation across leaflets of unsup-ported asymmetric bilayers. Proc Natl Acad Sci 105:124–128.

de Almeida RF, Fedorov A, Prieto M. 2003. Sphingomyelin/phosphatidylcholine/cholesterol phase diagram: Boun-daries and composition of lipid rafts. Biophys J 85:2406–2416.

Elson EL, Fried E, Dolbow JE, Genin GM. 2010. Phase sep-aration in biological membranes: integration of theoryand experiment. Annu Rev Biophys 39: 207–226.

Fan J. Sammalkorpi M. 2010. Influence of nonequilibriumlipid transport, membrane compartmentalization, andmembrane proteins on the lateral organization of theplasma membrane. Physical Review E 81: 011908.

Feigenson GW, Buboltz JT. 2001. Ternary phase diagram ofdipalmitoyl-PC/dilauroyl-PC/cholesterol: Nanoscopicdomain formation driven by cholesterol. Biophys J 80:2775–2788.

Frolov VA, Chizmadzhev YA, Cohen FS, Zimmerberg J.2006. "Entropic traps" in the kinetics of phase separationin multicomponent membranes stabilize nanodomains.Biophys J 91: 189–205.

Gelbart WM, Sear RP, Heath JR, Chaney S. 1999. Array for-mation in nano-colloids: Theory and experiment in 2D.Faraday Discuss Chem Soc 112: 299–307.

Gennis RB. 1989. Biomembranes: Molecular structure andfunction. Springer-Verlag, New York.

Guggenheim EA. 1952. Mixtures; the theory of the equilib-rium properties of some simple classes of mixtures, solutionsand alloys. Clarendon, Oxford, United Kingdom.

Haluska CK, Schroder AP, Didier P, Heissler D, Duportail G,Mely Y, Marques CM. 2008. Combining fluorescence life-time and polarization microscopy to discriminate phaseseparated domains in giant unilamellar vesicles. Biophys J95: 5737–5747.

Hammond AT, Heberle FA, Baumgart T, Holowka D, BairdB, Feigenson GW. 2005. Crosslinking a lipid raft compo-nent triggers liquid ordered-liquid disordered phase sep-aration in model plasma membranes. Proc Natl Acad Sci102: 6320–6325.

Heberle FA, Wu J, Goh SL, Petruzielo RS, Feigenson GW.2010. Comparison of three ternary lipid bilayer mixtures:FRET and ESR reveal nanodomains. Biophys J doi:101016/jbpj201009064.



Honerkamp-Smith AR, Cicuta P, Collins MD, Veatch SL,den Nijs M, Schick M, Keller SL. 2008. Line tensions,correlation lengths, and critical exponents in lipidmembranes near critical points. Biophys J 95: 236–246.

Huang J. 2002. Exploration of molecular interactions incholesterol superlattices: Effect of multibody interac-tions. Biophys J 83: 1014–1025.

Huang J, Feigenson GW. 1993. Monte Carlo simulation oflipid mixtures: finding phase separation. Biophys J 65:1788–1794.

Huang C, Li S. 1999. Calorimetric and molecular mechanicsstudies of the thermotropic phase behavior of membranephospholipids. Biochim Biophys Acta 1422: 273–307.

Huang J, Buboltz JT, Feigenson GW. 1999. Maximum solu-bility of cholesterol in phosphatidylcholine and phospha-tidylethanolamine bilayers. Biochim Biophys Acta 1417:89–100.

Imperio A, Reatto L. 2006. Microphase separation in two-dimensional systems with competing interactions. JChem Phys 124: 164712.

Kaiser HJ, Lingwood D, Levental I, Sampaio JL, KalvodovaL, Rajendran L, Simons K. 2009. Order of lipid phasesin model and plasma membranes. Proc Natl Acad Sci106: 16645–16650.

Kiessling V, Wan C, Tamm LK. 2009. Domain coupling inasymmetric lipid bilayers. Biochim Biophys Acta 1788:64–71.

Kinosita K Jr, Kawato S, Ikegami A. 1984. Dynamic structureof biological and model membranes: Analysis by opticalanisotropy decay measurement. Adv Biophys 17: 147–203.

Lingwood D, Simons K. 2010. Lipid rafts as a membrane-organizing principle. Science 327: 46–50.

Lingwood D, Ries J, Schwille P, Simons K. 2008. Plasmamembranes are poised for activation of raft phase coales-cence at physiological temperature. Proc Natl Acad Sci105: 10005–10010.

London E, Feigenson GW. 1978. Fluorescence quenching ofCa2þ-ATPase in bilayer vesicles by a spin-labeled phos-pholipid. FEBS Lett 96: 51–54.

London E, Feigenson GW. 1981. Fluorescence quenching inmodel membranes. 1. Characterization of quenchingcaused by a spin-labeled phospholipid. Biochemistry 20:1932–1938.

Manz BN, Groves JT. 2010. Spatial organization and signaltransduction at intercellular junctions. Nat Rev Mol CellBiol 11: 342–352.

Marsh D. 2009. Cholesterol-induced fluid membranedomains: A compendium of lipid-raft ternary phase dia-grams. Biochim Biophys Acta 1788: 2114–2123.

McLaughlin S, Murray D. 2005. Plasma membrane phos-phoinositide organization by protein electrostatics.Nature 438: 605–611.

Morales-Penningston NF, Wu J, Farkas ER, Goh SL, Konya-khina TM, Zheng JY, Webb WW, Feigenson GW. 2010.GUV preparation and imaging: Minimizing artifacts.Biochim Biophys Acta 1798: 1324–1332.

Phillips MC, Ladbrooke BD, Chapman D. 1970. Molecularinteractions in mixed lecithin systems. Biochim BiophysActa 196: 35–44.

Pokorny A, Yandek LE, Elegbede AI, Hinderliter A, AlmeidaPF. 2006. Temperature and composition dependence ofthe interaction of delta-lysin with ternary mixtures ofsphingomyelin/cholesterol/POPC. Biophys J 91: 2184–2197.

Polozova A, Litman BJ. 2000. Cholesterol dependentrecruitment of di22:6-PC by a G protein-coupled recep-tor into lateral domains. Biophys J 79: 2632–2643.

Sahl SJ, Leutenegger M, Hilbert M, Hell SW, Eggeling C.2010. Fast molecular tracking maps nanoscale dynamicsof plasma membrane lipids. Proc Natl Acad Sci 107:6829–6834.

Schneider DJ, Freed JH. 1989. Calculating slow motionalmagnetic resonance spectra. A user’s guide. In Spin label-ing: Theory and application (ed Reuben J). Plenum,New York.

Semrau S, Idema T, Schmidt T, Storm C. 2009. Membrane-mediated interactions measured using membranedomains. Biophys J 96: 4906–4915.

Seul M, Andelman D. 1995. Domain shapes and patterns:The phenomenology of modulated phases. Science 267:476–483.

Sheetz MP. 1993. Glycoprotein motility and dynamicdomains in fluid plasma membranes. Annu Rev BiophysBiomol Struct 22: 417–431.

Silvius JR. 2003. Fluorescence energy transfer reveals micro-domain formation at physiological temperatures in lipidmixtures modeling the outer leaflet of the plasma mem-brane. Biophys J 85: 1034–1045.

Simons K, van Meer G. 1988. Lipid sorting in epithelial cells.Biochemistry 27: 6197–6202.

Stadtman ER, Chock PB. 1977. Superiority of intercon-vertible enzyme cascades in metabolic regulation: analy-sis of monocyclic systems. Proc Natl Acad Sci 74:2761–2765.

Tanford C. 1980. The hydrophobic effect: The formationof micelles and biological membranes, 2nd ed. Wiley,New York.

Tenchov BG, Koynova RD. 1985. The effect of nonideallateral mixing on the transmembrane lipid asymmetry.Biochim Biophys Acta 815: 380–391.

van Meer G, Voelker DR, Feigenson GW. 2008. Membranelipids: Where they are and how they behave. Nat RevMol Cell Biol 9: 112–124.

Veatch SL, Keller SL. 2003. Separation of liquid phases ingiant vesicles of ternary mixtures of phospholipids andcholesterol. Biophys J 85: 3074–3083.

Veatch SL, Keller SL. 2005. Miscibility phase diagrams ofgiant vesicles containing sphingomyelin. Phys Rev Lett94: 148101.

Veatch SL, Gawrisch K, Keller SL. 2006. Closed-loop misci-bility gap and quantitative tie-lines in ternary mem-branes containing diphytanoyl PC. Biophys J 90:4428–4436.

Vist MR, Davis JH. 1990. Phase equilibria of cholesterol/dipalmitoylphosphatidylcholine mixtures: 2H nuclear



magnetic resonance and differential scanning calorime-try. Biochemistry 29: 451–464.

Wenk MR. 2005. The emerging field of lipidomics. Nat RevDrug Discov 4: 594–610.

Wu SH, McConnell HM. 1975. Phase separations in phos-pholipid membranes. Biochemistry 14: 847–854.

Zhao J, Wu J, Heberle FA, Mills TT, Klawitter P, Huang G,Costanza G, Feigenson GW. 2007a. Phase studies of

model biomembranes: Complex behavior of DSPC/DOPC/cholesterol. Biochim Biophys Acta 1768:2764–2776.

Zhao J, Wu J, Shao H, Kong F, Jain N, Hunt G, Feigenson G.2007b. Phase studies of model biomembranes: Macro-scopic coexistence of Lalpha þ Lbeta, with light-inducedcoexistence of LalphaþLo Phases. Biochim Biophys Acta1768: 2777–2786.



phase separation in lipid...

Documents