1H NMR Shows Slow Phospholipid Flip-Flop in Gel and Fluid Bilayers.

We measured the transbilayer diffusion of 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) in large unilamellar vesicles, in both the gel (Lβ') and fluid (Lα) phases. The choline resonance of headgroup-protiated DPPC exchanged into the outer leaflet of headgroup-deuterated DPPC-d13 vesicles was monitored using 1H NMR spectroscopy, coupled with the addition of a paramagnetic shift reagent. This allowed us to distinguish between the inner and outer bilayer leaflet of DPPC, to determine the flip-flop rate as a function of temperature. Flip-flop of fluid-phase DPPC exhibited Arrhenius kinetics, from which we determined an activation energy of 122 kJ mol-1. In gel-phase DPPC vesicles, flip-flop was not observed over the course of 250 h. Our findings are in contrast to previous studies of solid-supported bilayers, where the reported DPPC translocation rates are at least several orders of magnitude faster than those in vesicles at corresponding temperatures. We reconcile these differences by proposing a defect-mediated acceleration of lipid translocation in supported bilayers, where long-lived, submicron-sized holes resulting from incomplete surface coverage are the sites of rapid transbilayer movement.

Introduction

The eukaryotic plasma membrane (PM) is characterized by an asymmetric distribution of lipids between the exoplasmic and cytoplasmic bilayer leaflets.[1−3] The physiological fate of cells depends on the strict maintenance of this asymmetry through the interplay of active and passive lipid translocation events.[4] Active mechanisms are thought to rely on the so-called floppases that move newly synthesized lipids from the inner to the outer leaflet and on flippases that restore the asymmetry of passively translocated lipids.[5,6] The selectivity of these enzymes for different classes of lipids regulates compositional asymmetry within the PM.[4] For example, the exoplasmic leaflet of mammalian PM is enriched with sphingolipids and neutral phosphatidylcholines, whereas the cytoplasmic leaflet contains most of the aminophospholipids, including phosphatidylethanolamines and negatively charged phosphatidylserines.[7] Still, the exact mechanisms by which these lipids arrive and remain at their locations in the PM are not fully understood.[8] Reliable values of passive lipid translocation rates are a necessary starting point for a detailed mechanistic understanding of PM asymmetry, but such values are both scarce and scattered, with reported flip-flop half times for PC lipids in the fluid phase ranging from minutes[9] to hours[10] to days or weeks.[11−13] Even less is known about flip-flop kinetics of lipids in the highly ordered gel state.[14−17] The frequent use of extrinsic probe molecules has undoubtedly contributed to the controversy surrounding spontaneous lipid translocation.

Model phospholipid bilayers have long served as surrogates for the systematic and well-controlled investigation of biological membrane phenomena. A variety of sample geometries have been developed, including freely floating liposomes and solid-supported bilayers (SSBs).[18] Different sample geometries have associated advantages and disadvantages and are often chosen out of necessity for a particular experiment. For example, recent studies have utilized SSBs for measuring passive lipid translocation with sum-frequency generation (SFG) vibrational spectroscopy.[14−17] In these studies, SSBs were chosen for the facile preparation of asymmetric bilayers as well as because of the surface-sensitive nature of the measurement technique. Inherent differences between SSBs and vesicles—most notably, the close proximity of the support to the membrane and the presence of edges and defects in SSBs—raise the possibility of differences in lipid translocation behavior.[19] Recent developments in the methodology for preparing and characterizing asymmetric vesicles[20] now allow for a comparison of lipid translocation in bilayers having different sample geometries without the need for bulky labels, which may not faithfully report on the behavior of the host lipid.[14]

Making use of these developments, we report on the passive lipid transbilayer diffusion of the gel (Lβ′) and fluid (Lα) phase 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) in tensionless, probe-free liposomes.[20] Use of asymmetric vesicles having different DPPC isotopes in the inner and outer leaflets enabled measurement of intrinsic flip-flop rates using solution 1H NMR. We find that DPPC flip-flop in the gel phase is too slow to accurately measure, whereas fluid-phase DPPC undergoes flip-flop with a temperature-dependent half time ranging from days to weeks. Vesicles incubated within the main transition at 40 °C exhibited approximately a twofold faster flip-flop rate than fully melted fluid vesicles at 50 °C. At all temperatures we examined, the translocation rate in vesicles is slower by orders of magnitude than that previously measured with SFG on supported DPPC bilayers.[14,16] We propose a plausible defect-mediated mechanism, supported by Monte Carlo simulations, to explain the dramatic acceleration of lipid translocation in supported bilayers compared to vesicles.

Results

Structure of Asymmetric DPPC Vesicles

We make use of asymmetric large unilamellar vesicles (aLUVs) to measure lipid translocation. To this end, aLUVs of DPPC and its deuterated variants (Figure S1) were prepared by cyclodextrin-mediated lipid exchange as previously described.[20] We exchanged headgroup-protiated/chain-perdeuterated donor lipid (DPPC-dC) into LUVs initially composed of headgroup-deuterated/chain-protiated acceptor lipid (DPPC-dH). The use of differentially deuterated donor and acceptor lipids allowed us to determine the composition of each leaflet after exchange using isotope-sensitive techniques, namely, 1H NMR combined with gas chromatography (GC), as previously described.[20] On average, aLUVs had an outer leaflet donor concentration of 59 mol % (i.e., 59% of outer leaflet acceptor lipid was replaced by donor lipid) and an inner leaflet donor concentration of 30 mol % immediately following exchange. The appearance of inner leaflet donor lipid may result from a small population of otherwise undetectable donor vesicle contamination or from accelerated flip-flop during the exchange step when vesicles are subjected to a relatively high cyclodextrin concentration.[21] However, neither of these cases affects the analysis or interpretation of kinetic data presented below.

Because bilayer defects can promote lipid translocation,[8] vesicle structure was assessed before and after lipid exchange. Vesicle size and polydispersity were measured using dynamic light scattering (DLS), and the bilayer structure was determined from SAXS measurements. SAXS is particularly useful as these measurements probe the internal bilayer structure, which can be compared to that of symmetric LUVs prepared using conventional methods. Figure shows experimental SAXS form factors, , for symmetric and asymmetric DPPC vesicles. Within experimental uncertainty, these form factors are identical. Bilayer structural parameters were obtained by fitting the form factors to a slab model, as described previously[22] and are listed in Table . The area per lipid (AL), headgroup–headgroup spacing (DHH), and hydrocarbon thickness (2DC) are similar for symmetric and asymmetric vesicles and compare well to the literature values. Furthermore, vesicle size and polydispersity measured using DLS were similar before and after the lipid exchange (Figure c) and did not change significantly during sample incubation (Figure d). Together, these results indicate that the asymmetric vesicle preparation did not introduce structural artifacts. A similar conclusion was drawn in a previous study that compared symmetric and asymmetric LUVs composed of 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) and its deuterated analogs and which used the same asymmetric vesicle preparation method.[20]

Figure 1

Structure of symmetric and asymmetric DPPC vesicles. Experimental small-angle X-ray scattering (SAXS) form factors (open circles) and fits (solid colored lines) for DPPC LUVs prepared using standard extrusion (a) and aLUVs (DPPC-dHinner/DPPC-dCouter) prepared by cyclodextrin-mediated exchange (b), both at 55 °C. (c) Vesicle diameter of acceptor LUVs (white bars) and aLUVs (solid bars) immediately after preparation. (d) aLUV diameter before (white bars) and after (solid bars) incubation at different temperatures (incubation times are indicated on the bars).

Table 1

DPPC Structural Parameters Obtained from Refinement of SAXS Dataa,b

	LUV	aLUV	literaturec
AL (Å2)	65.7	65.7	64.05
DHH (Å)	35.2	34.8	36.55
2DC (Å)	27.5	27.5	28.18

Best fit parameters for extruded symmetric and asymmetric DPPC bilayers (LUV and aLUV, respectively) at 55 °C: area per lipid (AL), headgroup–headgroup distance (DHH), and hydrocarbon thickness (2DC).

Parameter uncertainty is estimated to be ±2%.

Data from Kučerka et al.[23]

Thermotropic Behavior of Asymmetric DPPC Vesicles

We used differential scanning calorimetry (DSC) to compare the thermotropic behavior of symmetric and asymmetric DPPC vesicles in D2O. Figure shows exotherms for vesicles composed of DPPC and its isotopic variants. Protiated DPPC multilamellar vesicles (MLVs) (Figure , upper) exhibited a broad Lβ′ → Pβ′ pretransition (TP) centered at 35.5 °C and a relatively narrow Pβ′ → Lα main transition (TM) centered at 42.0 °C, consistent with previous reports.[24] Deuteration of the PC headgroup resulted in only a slight decrease in TP and TM (<1 °C), whereas chain perdeueration resulted in a larger decrease of 3–4 °C, also in agreement with the literature.[25] These results, summarized in Table S1, are consistent with hydrocarbon chain melting as the dominant contribution to the transitions.

Figure 2

Differential scanning calorimetry of DPPC vesicles in D2O. Upper: exotherms of DPPC MLVs composed of different isotopic variants. Lower: exotherms for DPPC LUVs and DPPC-dHin/DPPCdCout aLUVs. Data were collected at a scan rate of 0.5 °C/s.

Extrusion to form symmetric 100 nm LUVs caused a broadening of the phase transitions (Figure , lower), as observed previously.[26] In aLUVs with a DPPC-dH-enriched inner leaflet and a DPPC-dC-enriched outer leaflet, an even broader transition that includes multiple peaks was apparent; indeed, the aLUV exotherm is well-fit by a sum of a single broad Gaussian representing the pretransition and three Gaussians representing the main transition (Figure S2 and Table S2). It was previously found that in homogeneous mixtures of protiated and chain-perdeuterated DPPC, TM shifted linearly as a function of mixture composition and was only slightly broadened.[27] The presence of two main transition peaks is therefore an expected outcome from aLUVs having two weakly coupled leaflets of different isotopic compositions. The additional complexity of the aLUV thermogram may indicate sample heterogeneity, including heterogeneous exchange and/or minor contamination from symmetric donor or acceptor vesicles. Alternatively, it may reflect stronger coupling of the asymmetric leaflets. Importantly, the kinetic analysis presented below is not influenced by the presence of symmetric vesicle contamination.

Interestingly, both LUV exotherms clearly showed the presence of a ripple phase. The ripple phase has previously been observed experimentally in both unilamellar vesicles[28−31] and in symmetric single SSBs.[32,33] However, the cooperativity of lipid phase transitions is much higher in multibilayer stacks due to the confinement of out-of-plane fluctuations[30] and is a likely reason for the failure to detect the Lβ′ → Pβ′ transition in some reports of unilamellar systems.[34,35] On the basis of the DSC data in Figure , it is clear that gel-phase flip-flop measurements should be made below 30 °C to avoid the ripple phase, especially when deuterated lipids are used.

Translocation of DPPC in Gel- and Fluid-Phase Vesicles

Asymmetric lipid distributions in aLUVs were quantified from solution 1H NMR spectra measured in the presence of extravesicular paramagnetic lanthanide ions Pr3+.[36] aLUVs were prepared and incubated at a desired temperature in the absence of Pr3+, and at regular time intervals, a sample aliquot was removed to quantify the asymmetry, with Pr3+ added immediately before the NMR measurement. Externally added Pr3+ (∼70 μM) does not permeate into the vesicle lumen during the ∼15 min NMR measurement,[20,36] interacting only with outer leaflet lipid headgroups and inducing a downfield shift of the DPPC choline resonance (we note that the deuterated choline of DPPC-dH does not contribute to the 1H NMR signal). The observed choline signal is therefore a superposition of shifted and unshifted resonances with relative areas that are proportional to the amount of DPPC in the outer and inner leaflets, respectively (Figure a).[36]

Figure 3

DPPC flip-flop kinetics. (a) Asymmetric lipid distribution determined by 1H NMR. Shown are NMR data (black line) and fitted components (filled peaks) from DPPC-dHin/DPPC-dCout aLUVs in the presence of ∼70 μM Pr3+ shift reagent, measured immediately following aLUV preparation (upper panel) and after 24 h incubation at 65 °C (lower panel). Spectra were modeled as the sum of outer leaflet (red) and inner leaflet (green) choline resonances and residual aqueous cyclodextrin (gray). (b,c) Time decay of normalized bilayer asymmetry at various temperatures in the gel (blue symbols), gel + fluid (green symbols), and fluid (orange symbols) phases and fits to eq (solid lines). ΔC is defined as the difference in inner and outer leaflet NMR peak areas normalized to their initial (time zero) difference, see eq . Inset to (c) shows the Arrhenius behavior of fluid-phase translocation (solid gray line, with dashed lines indicating the 95% confidence interval). (d) DSC exotherm for DPPC-dHin/DPPCdCout aLUVs. Roman numerals denote different bilayer phase states and are color-coded to the data in (b) and (c): I, gel (blue); II, gel + fluid coexistence (green); and III, fluid (orange). (e) Translocation rate constants (kf) for gel phase (blue, SSB values reproduced from Liu and Conboy),[14] gel + fluid coexistence (green), and fluid-phase (orange) DPPC. Horizontal lines represent the minimum kf for SSB fluid-phase DPPC and the maximum gel-phase kf for LUVs.

Freshly prepared aLUVs whose outer leaflets were initially enriched in headgroup-protiated DPPC (i.e., DPPC-dHin/DPPCdCout) exhibited an unequal area ratio that gradually approached unity as lipids equilibrated between the two bilayer leaflets (Figure a, lower panel). The normalized temperature- and time-dependent transbilayer lipid distributions obtained from NMR are shown in Figure b,c. We determined the flip-flop rate constant kf at different temperatures directly from changes in the asymmetric lipid distribution as a function of time. Fitted decay curves for fluid-phase DPPC are shown in Figure c. In the temperature range of 50–65 °C, we observed flip-flop rates ranging from 1.3 × 10–6 to 8.3 × 10–6 s–1, corresponding to flip-flop half-times (t1/2) on the order of days to weeks (Table ).

Table 2

Summary of DPPC Translocation Kinetics

temperature (°C)	phase	kf (×10–6 s–1)	t1/2 (h)
20	Lβ′	<0.085a	>4555a
34
37
40	Pβ’ + Lα	2.5 ± 0.2	75 ± 6
50	Lα	1.3 ± 0.08	147 ± 9
55		1.8 ± 0.09	105 ± 5
60		5.1 ± 0.5	38 ± 3
65		8.3 ± 0.4	23 ± 1

No flip-flop was observed in gel-phase vesicles.

We further investigated DPPC flip-flop in the gel phase (22, 34, and 37 °C) and within the main phase transition at 40 °C (Figure d). We did not observe appreciable flip-flop of gel-phase DPPC for incubation times up to 250 h. However, when LUVs were incubated within the main transition at 40 °C, the flip-flop rate was a factor of 2 greater than that measured for fully melted fluid DPPC at 50 °C and a factor of 5 greater than that expected from an extrapolation of higher-temperature data. Flip-flop rate constants (kf) and half-times (t1/2) for DPPC at various temperatures are summarized in Table .

From the temperature dependence of kf (Figure c, inset), we found that the kinetics for fluid-phase DPPC flip-flop follow Arrhenius’ law. The activation energy barrier for flip-flop was determined by fitting to the Arrhenius equation, kf = A exp(−Ea/RT), yielding Ea = 122 ± 14 kJ/mol. Additional thermodynamic parameters were determined using transition state theory.[13] These values are summarized in Table S3.

Simulations of Defect-Mediated Translocation

As speculated previously[19] and confirmed in this work, liposomes exhibit remarkably different flip-flop behavior than supported bilayers. One contributing factor might be submicron topological defects that are known to exist in SSBs from atomic force microscopy (AFM) imaging studies.[33,37−46] To investigate the influence of defects on flip-flop, we performed random-walk (Monte Carlo) simulations of lateral lipid diffusion in the presence of 100 nm-diameter holes, a typical size observed in AFM images.[46]Figure a shows a schematic illustration of such a defect (not to scale) in a supported bilayer. We assumed that translocation effectively occurs by unhindered lateral diffusion of a lipid through the pore formed by the lipid headgroups. Flip-flop was therefore incorporated into the simulations by enforcing a translocation event whenever a proposed random step carried a lipid into a hole (full details of the simulations are found in the Experimenal Procedures).

Figure 4

Monte Carlo simulations of defect-mediated flip-flop. (a) Schematic illustration of an SSB, showing a defect site in cross-section. Transbilayer movement is assumed to occur via unhindered lateral diffusion through the pore formed by lipid headgroups. (b) MC simulation snapshot of a particle trajectory that includes segments in the top (black) and bottom (gray) leaflets. The inset shows an expanded view in the vicinity of a circular defect, revealing multiple translocation events. (c) Top–down view of simulation snapshots (defects shown as black circles; scale bar, 1 μm) for different bilayer surface coverages: 99.0% (gray), 99.5% (blue) and 99.9% (yellow). (d) Simulated asymmetry decay curves (open symbols) and fits (solid lines) corresponding to the surface coverages in (c), for a lateral lipid diffusion coefficient of 10–3 μm2 s–1. (e) Lipid translocation rate kf vs lateral diffusion coefficient DT calculated from decay curves corresponding to the surface coverages in (c).

Figure b shows a typical simulated lipid trajectory, characterized by relatively long time periods during which a lipid meandered through the space between holes, before finally approaching a hole edge. This was followed by a series of alternating flip and flop events as the lipid diffused in the vicinity of the hole edge; for the particular trajectory shown in Figure b, 47 separate translocation events occurred. Eventually, the lipid wandered away from the hole to repeat the cycle. The outcome of this sequence of events is effectively a randomization of the lipid’s transbilayer location.

Simulations were performed at three values of bilayer surface coverage, 99.0, 99.5, and 99.9% (Figure c), using diffusion coefficients typical of gel-phase DPPC (10–4 to 10–2 μm2 s–1).[32] We observed asymmetry decay curves (Figure d, open symbols) similar to those measured using SFG for gel-phase DPPC in supported bilayers.[14] These curves were well fit by a double exponential decay (Figure d, solid lines), and the range of obtained values for the slower component kf (Figure e) was consistent with reported rates for gel-phase translocation in SSBs.[14] As expected, either increasing the diffusion coefficient or decreasing the surface coverage (i.e., increasing the defect density) increased the translocation rate (Figure e). Furthermore, a fluid-phase diffusion coefficient of 1 μm2 s–1 (equivalent to compressing the time axis of Figure d by a factor of 103) resulted in complete equilibration of the two leaflets within seconds, even for 99.9% surface coverage. This observation is consistent with reports that SSBs prepared by Langmuir–Blodgett/Langmuir–Schaefer (LB/LS) deposition do not support asymmetry at fluid-phase temperatures.[14] We conclude that defect-mediated translocation is a plausible mechanism to reconcile the observed differences between vesicles and SSBs.

Discussion

We measured flip-flop kinetics in vesicles using 1H NMR by following the transbilayer movement of a headgroup-protiated lipid in a headgroup-deuterated matrix. DPPC in the fluid phase showed the expected increase in kf (decrease in t1/2) with increasing temperature (Figure c,e and Table ). Still, DPPC flip-flop is slow, even at fluid-phase temperatures; for example, a half time of nearly 6 days was found at 50 °C. Slow translocation is nevertheless consistent with many previous reports for PC lipids in vesicles,[13,20,47−49] including the inability of a SANS study to detect intrinsic POPC flip-flop at 37 °C over a period of 2 days.[21] A significant advantage of our method is the ability to measure the intrinsic flip-flop rate of the host lipid (here, DPPC) as opposed to that of an extrinsic probe molecule. For example, Homan and Pownall found a fourfold increase in flip-flop half time of a prenyl phospholipid in a host POPC vesicle when the probe chain length was varied from 8 to 12 carbons.[13]

Translocation in Vesicles is Dramatically Slower than that in SSBs

Our observation of slow DPPC translocation in vesicles stands in contrast to previously published gel-phase DPPC flip-flop rates in SSBs measured using SFG,[14,17] a method that also measures intrinsic lipid translocation and is therefore directly comparable to the present work. Indeed, our data suggest that accurate flip-flop measurements may not even be possible for gel-phase lipids in vesicles because of the extremely slow kinetics of flip-flop when compared with the vesicle lifetime.

That flip-flop rates measured in vesicles and SSBs are different, is perhaps unsurprising.[19] Membrane properties including lateral diffusion[50,51] and phase behavior[52,53] are known to exhibit a complex dependence on the properties of the support as well as the bilayer deposition technique.[54,55] The sheer variety of SSB systems—including different substrates, cushions, tethers, and deposition techniques—precludes a simple explanation of their flip-flop behavior. For example, Tamm and co-workers have successfully retained asymmetry for hours in fluid-phase SSBs prepared using the Langmuir–Blodgett/vesicle fusion (LB/VF) technique.[54,55] We therefore restrict the following discussion to supported bilayers prepared using LB/LS deposition onto uncushioned quartz substrates, for which extensive flip-flop data exist.[14,17] Below, we discuss three possible explanations for the differences in flip-flop rates in this system compared with vesicles: these include the phase state of the sample, the presence of bilayer curvature in vesicles, and the presence of topological bilayer defects such as edges and pores in SSBs.

Translocation is Accelerated in Phase Coexistence Regions

We found that DPPC flip-flop in vesicles is accelerated when the sample is incubated within the main phase transition, an effect that was previously reported for a fluorescent nitrobenzoxadiazole-labeled lipid (C6-NBD-PC) in host DPPC vesicles.[9] Bilayer permeability reaches a maximum in the vicinity of the main transition,[56] where enhanced area fluctuations lead to a greater probability of spontaneous pore formation.[57] The interface of gel/fluid domain boundaries may be particularly leaky.[58] One possible explanation for accelerated flip-flop in SSBs is therefore the presence of a broadened gel + fluid coexistence region (33–41.5 °C) in supported DPPC bilayers, as shown by Wu and co-workers using SFG and AFM.[46] Other SFG studies have also found a similarly broad main phase transition for DPPC SSBs.[59] Although published values for DPPC flip-flop rates measured in the temperature range 27.7–36.6 °C were all ascribed to the gel phase,[14] it is possible that at least some of these rates were actually measured within the gel + fluid coexistence region and would therefore be expected to have enhanced translocation. Nevertheless, our observations in vesicles suggest that this would result in at most a factor of five increase in the translocation rate, which cannot fully account for the orders-of-magnitude discrepancy between vesicles and SSBs.

Influence of Vesicle Curvature

Although we did not directly address the influence of curvature on flip-flop rate in the present study, two lines of evidence suggest that bilayer curvature effects should be minimal for the 100 nm-diameter DPPC vesicles used in our NMR measurements. First, for a wide variety of neutral PC lipids, X-ray scattering data obtained from flat oriented bilayer stacks and extruded LUVs (50 nm pore size) are in excellent agreement over a large q range, suggesting a negligible influence of curvature on the bilayer structure.[23,60,61] Moreover, these data were well-fit by a symmetric bilayer form factor, implying an identical structure for the inner and outer leaflets. (A caveat is that vesicle curvature may have a more pronounced effect on the structure of bilayers composed of charged lipids, as observed by Brustowicz and Brunger for SOPS vesicles[62] and Kučerka and co-workers for DOPS vesicles,[61] although this finding is not relevant to our experiments on zwitterionic PC lipids.) Second, it was previously reported that lipid asymmetry is stable for days in a variety of systems having vastly different vesicle curvature, namely, SUVs of <30 nm diameter,[47] LUVs of ∼50–150 nm diameter,[63,64] and GUVs of >20 micron diameter.[49,65] This finding suggests that flip-flop rate is not strongly dependent upon curvature, and more importantly that the slow flip-flop reported here for 100 nm-diameter DPPC LUVs is not an artifact of vesicle curvature.

Topological Defects in Solid-Supported Bilayers

The topology of a sealed vesicle is fundamentally different from that of a supported bilayer. In the former, the outer and inner leaflets are distinct surfaces, whereas in the latter the two leaflets are continuous, meeting at the macroscopic edges of the substrate and at any defect sites, as shown schematically in Figure a. In an SSB, lipids can therefore move from the proximal to the distal leaflet (and vice versa) effectively by lateral diffusion through a permanent or defect edge. Although the permanent substrate edge is an unavoidable consequence of the sample geometry, it is likely not a significant contributor to translocation for centimeter-sized substrates, as most lipids would need to diffuse long distances to reach the nearest edge. On the other hand, AFM images of SSB generally show numerous patches of bare substrates appearing as holes with submicron dimensions (tens to hundreds of nanometers in diameter); these are especially ubiquitous in gel-phase bilayers.[33,37−46] Moreover, for typical values of bilayer surface coverage, the average distance between holes is on the order of microns or less.[46] Although these defects are often too small to be seen with conventional fluorescence microscopy, they can be detected with other fluorescence techniques. For example, in LB/LS supported bilayers containing fluorescent NBD-labeled lipids in both leaflets, external addition of the reducing agent dithionite—which does not permeate intact bilayers[66]—nevertheless quenched the fluorescence on both sides of the supported bilayer, implying the existence of numerous small defects in SSB that could not otherwise be visualized with fluorescence microscopy.[54] These defects are apparently long-lived and are most likely due to incomplete coverage and lipid contraction and desorption during the bilayer deposition process.[40]

Translocation is Accelerated by Defects

MC simulations revealed dramatically accelerated translocation in bilayers containing even a low density of defects (Figure c–e). The translocation rate exhibited a linear dependence on the lateral diffusion coefficient (DT), increasing by 2 orders of magnitude with a 100-fold increase in DT, which may help explain the observed strong dependence of flip-flop rate on temperature in gel-phase SSBs.[14] Using supported bilayers, Tamm and McConnell found that the lateral diffusion coefficient of gel-phase DPPC increased from 10–4 μm2 s–1 at room temperature to 10–2 μm2 s–1 at the pretransition, with an associated activation energy of 400 kJ mol–1 (compared with ∼40 kJ mol–1 for fluid-phase diffusion).[32] The authors attributed this unusually large activation energy to a defect-mediated diffusion process in the gel phase. Given the strong temperature dependence of DT in the gel phase, any process that is limited by translational diffusion would also be expected to exhibit a strong temperature dependence. A second factor that influences translocation kinetics in our simulations is defect density, but this parameter has not been systematically explored experimentally. Typical reported values of surface coverage are >95%, but these may also depend on temperature.[67]

Our simulations, while highly simplified, nevertheless indicate that a better understanding of lateral diffusion is needed to explain translocation in supported bilayers. There are three potentially distinct environments in an SSB—the distal leaflet, the proximal leaflet, and the defect edge—and therefore at least three diffusion rates to consider. For example, there is evidence to suggest that lateral diffusion may be slower in the proximal leaflet compared to the distal leaflet.[50] Much less is known about the nature of the bilayer in the vicinity of a defect. Molecular dynamics (MD) simulations suggest that lipids located within a defect are more disordered and that lipid movement through the edge may actually be faster than lateral diffusion in the bulk.[68] Experimental studies of mechanically stressed vesicles also indicate that translocation is accelerated by the presence of defects.[69] The implication of these findings is that in a bilayer with a sufficient density of long-lived defects, the translocation rate will be controlled by the time that it takes for a lipid to diffuse to the nearest defect and not by the actual translocation event at the defect edge itself. Indeed, an MD simulation of pore-mediated flip-flop found that after spontaneous formation of a small water channel in the bilayer, the majority of ensuing flip-flop events were for lipids that were initially remote from the pore but that diffused to the pore site with time.[70] Still, other experimental studies have found increased order and reduced mobility of fluorescent lipid probes located near the edges of nanopatterned supported bilayers,[71] suggesting the existence of an energy barrier for a lipid to enter or leave the edge region that might slow translocation. More experimental and simulation work is needed to understand the nature of the defect edge environment and how it contributes to flip-flop rates in supported bilayers.

Influence of Pr3+ on Translocation Measurements

In the current study, samples were incubated in the absence of Pr3+, which was introduced immediately before NMR measurements. We demonstrated in previous work that Pr3+ does not cross the bilayer on the time scale of our NMR measurements, even upon temperature cycling through TM.[20] Previous studies have shown an increase in TM with increasing Pr3+ concentration.[72−74] We also observe a concentration-dependent increase in TM of DPPC due to Pr3+ (Figure S3). However, for conditions within the NMR sample tube (∼70 μM Pr3+, 75:1 lipid/Pr3+), there is only a minor increase in TM, such that aLUVs were fully melted at the measurement temperature of 50 °C. Furthermore, after the addition of Pr3+ to the sample, successively recorded NMR spectra were identical (Figure S4), indicating that Pr3+ does not measureably accelerate flip-flop during the data collection interval. We therefore conclude that Pr3+ reports faithfully on flip-flop rates and does not itself influence the transbilayer distribution of lipids.

Summary and Conclusions

We used NMR to determine passive lipid flip-flop rates in asymmetric vesicles composed of DPPC and its deuterated variants. This novel experimental approach allowed us to avoid complications associated with bulky fluorescence or spin labels. We were unable to observe flip-flop in gel-phase vesicles over the course of weeks. Even at elevated temperatures, fluid-phase DPPC exhibited relatively slow flip-flop, with half times ranging from ∼1 week at 50 °C to ∼1 day at 65 °C. From the observed Arrhenius behavior, we determined the activation free energy, enthalpy, and entropy of DPPC translocation in vesicles.

We found significant differences in flip-flop rates of lipids in vesicles compared with the literature data from supported bilayers. Specifically, DPPC flip-flop in SSBs prepared by LB/LS[14] deposition is at least several orders-of-magnitude faster than that in LUVs at the same temperature. The ubiquity of submicron holes in SSBs seen with AFM imaging suggests that these differences are most likely due to the different nature of bilayer defects in these systems. This conclusion is supported by MC simulations, which show rapid equilibration of the proximal and distal leaflets (seconds to hours) in the presence of holes, even for gel-phase diffusion and what would be considered excellent surface coverages. We are not the first to urge caution when equating the behaviors of model systems having different sample geometries,[75] let alone when extrapolating these results to biological membranes,[76,77] which do not fit neatly into simple categories.

Despite these caveats, we do not wish to imply that SSBs have no value for translocation studies. On the contrary, SSBs may provide a more realistic model for cellular membranes that are often supported by a cytoskeleton. SSBs are also potentially a powerful platform for systematic investigation of defect-mediated translocation. In addition to facile access to translocation rates through SFG measurements, the use of SSBs enables direct visualization of even submicron defects with AFM, allowing for a thorough characterization of defect size and lifetime distributions. Such information will likely be crucial for modeling translocation. Moreover, it may be possible to control defect characteristics by using micro- or nanopatterned substrates. There has long been speculation that all lipid translocation is fundamentally a defect-mediated process,[78] irrespective of the sample geometry, owing to the prohibitively large energetic cost of desolvating the polar headgroup.[79] In recent years, MD simulations have lent support to this idea.[80,81] Because supported bilayers potentially offer greater control over defect characteristics, SSB-based translocation studies may ultimately shed new light on this long-standing problem in membrane biophysics.

Experimental Procedures

Materials

1,2-Dipalmitoyl-sn-glycero-3-phosphocholine (16:0/16:0 PC, DPPC), 1,2-dipalmitoyl-d62-sn-glycero-3-phosphocholine [16:0(d31)/16:0(d31) PC, DPPC-dC], 1,2-dipalmitoyl-sn-glycero-3-phosphocholine-1,1,2,2-d4-N,N,N-trimethyl-d9 [16:0/16:0 PC(d13), DPPC-dH], and 1,2-dipalmitoyl-sn-glycero-3-phospho-(1′-rac-glycerol) (sodium salt) [16:0/16:0 PG, DPPG] were purchased from Avanti Polar Lipids (Alabaster, AL) and used as received. Lipid stock solutions were prepared by dissolving dry lipid powder in HPLC-grade chloroform. Methyl-β-cyclodextrin (mβCD) was purchased from Acros Organics (Thermo Fisher Scientific, Waltham, MA) and prepared as a 35 mM stock solution in H2O. Praseodymium(III) nitrate hexahydrate Pr(NO3)3·6H2O (Pr3+) was purchased from Alfa Aesar (Ward Hill, MA) and prepared as a 100 mM stock solution in D2O. Centrifugal filter devices (Amicon Ultra-15, 100 000 Da molecular weight cutoff) were purchased from EMD Millipore (Billerica, MA) and washed seven times with H2O before use to remove trace glycerol. Ultrapure H2O was obtained from a High-Q purification system (Wilmette, IL), and 99.9% D2O was purchased from Cambridge Isotopes (Andover, MA).

Preparation of aLUVs

We prepared aLUVs from DPPC and its deuterated headgroup (DPPC-dH) and acyl chain (DPPC-dC) variants using cyclodextrin-mediated lipid exchange, described in detail elsewhere.[20,82] Briefly, extruded 100 nm-diameter acceptor LUVs (10–12 mg/mL) provided lipids for the aLUV inner leaflet, whereas donor MLVs provided different lipids for the outer leaflet. An excess of donor lipid (threefold over acceptor) was used. Donor MLVs were prepared in a 20% (w/w) sucrose solution to increase their density and facilitate their separation from aLUVs following the exchange step, in which donor and acceptor vesicles were gently stirred for 1 h at room temperature in the presence of 30 mM mβCD. The mixture was then diluted eightfold with H2O and centrifuged at 20 000 × g for 30 min to pellet donor MLVs. The supernatant containing aLUVs was concentrated using a prewashed 100 kDa molecular weight cutoff centrifugal filtration device at 5000 × g. Three subsequent cycles of dilution with D2O and concentration in the centrifugal filter allowed the efficient removal of residual sucrose and mβCD as well as exchange of H2O with D2O for NMR measurements.

Characterization of aLUVs

Samples for SAXS were concentrated to 20 mg/mL and measured with a Rigaku BioSAXS-2000 home source system with a Pilatus 100K detector and a HF007 copper rotating anode (Rigaku Americas, The Woodlands, TX). SAXS data were collected at a fixed sample-to-detector distance using a silver behenate calibration standard. Samples for 90° DLS were diluted to ∼60 μM and measured with a Brookhaven BI-200SM system (Brookhaven Instruments, Holtsville, NY). Samples for DSC were measured at ∼5 mg/mL using a Nano DSC (TA Instruments, New Castle, DE). Sample composition was determined with GC using procedures outlined by Heberle et al.[20] Analysis was performed on an Agilent 5890A gas chromatograph (Santa Clara, CA) with a 5975C mass-sensitive detector operated in the electron-impact mode. An HP-5MS capillary column (30 m × 0.25 mm, 0.25 μm film thickness) was used with a helium carrier at 1 mL/min and an inlet temperature of 270 °C.

Translocation Measurements with 1H NMR

Lipid translocation rates were determined from proton nuclear magnetic resonance (1H NMR) spectroscopic measurements. Isotopically asymmetric aLUVs were prepared, and an aliquot was immediately measured using the shift reagent technique described below to establish a time zero (t = 0) data point. Vesicles were incubated at 22, 34, 37, 40, 50, 55, 60, and 65 °C, with aliquots removed at regular time intervals for measurement. Separate sample preparations were used to populate different temperature data to minimize preparation-specific bias.

1H NMR spectra were collected on an Avance III 400 MHz spectrometer using the Bruker TopSpin acquisition software and analyzed with TopSpin 3.2. Lipid suspensions in D2O were brought to a total volume of 0.6 mL (for a total lipid concentration of ∼5 mM) and loaded into 5 mm NMR tubes. A standard 1H pulse sequence with a 30° flip angle and 2 s delay time was employed to collect 1664 transients at 50 °C. Data were processed with a line-broadening parameter of 2 Hz. The distribution of protiated choline between inner and outer vesicle leaflets was determined by the addition of the shift reagent Pr3+. Briefly, immediately before measurement, 2 μL of 20 mM Pr3+/D2O solution was dispensed directly into the NMR tube, which was then capped and inverted a minimum of three times to mix the contents. Several Pr3+ additions were made, with spectra obtained between between titrations, and the sample was discarded after measurement. The choline resonance was modeled using one or two Lorentzian peaks in the absence or presence of Pr3+, respectively. Spectra obtained from 2–3 successive Pr3+ additions were modeled separately to determine the inner and outer leaflet area fractions.

Kinetic Model for Lipid Translocation

In general, an analysis of vesicle exchange kinetics must consider both intervesicle and intravesicle (flip-flop) contributions to the decay of signals. For example, Nakano and co-workers were able to determine both inter- and intravesicle exchange rates in a time-resolved SANS experiment by mixing two populations of initially symmetric LUVs (one protiated and one deuterated).[21] However, owing to our initial conditions (i.e., a single population of identical aLUVs), vesicle–vesicle exchange does not contribute to the observed kinetics. This allowed us to determine the flip-flop rate constant kf directly from changes in the asymmetric lipid distribution as a function of time.

Asymmetry decay curves were modeled by solving a first-order, homogeneous system of equations that accounts for both inter- and intravesicle lipid transport (a complete derivation is found in the Supporting Information). Briefly, the inner and outer leaflet peak areas obtained from 1H NMR are proportional to the inner and outer leaflet concentrations of choline-protiated lipid Cin and Cout, given byTaking the difference in peak areas and normalizing to their initial difference gives the decay curveThe translocation half time is found by setting eq equal to 0.5 and solving for t, yieldingMeasurements at multiple temperatures allowed the determination of the activation energy Ea, using the Arrhenius equationwhere R is the universal gas constant and A is the pre-exponential factor. Other thermodynamic quantities including the enthalpy and entropy of formation of the activated state (ΔΗ‡ and ΔS‡, respectively) and the free energy of activation ΔG‡ were calculated using transition state theory following Homan and Pownall.[13]

Monte Carlo Simulations of Defect-Mediated Translocation

Simulations of 2D lipid diffusion in the presence of defects were performed using custom code written in Mathematica 11.0 (Wolfram Research, Champaign, IL). Particle trajectories were generated by first choosing a random initial particle position (x0, y0) within a periodic box of side length l (representing a square patch of bilayer) and assigning the particle to the top or bottom leaflet. During each uniform time step τ, the particle was advanced a uniform distance δ in a random direction within its leaflet, specified by an angle θ drawn from a uniform probability distribution θ ∼ U(0, 2π). Particle trajectories generated in this way satisfy the probability distribution for a 2D random walkwhere the diffusion coefficient D = δ2/4τ.

To simulate defect-mediated translocation, nonoverlapping circular holes of radius R were randomly placed within the simulation box. The number of holes NH in a given box was drawn from a Poisson distribution NH ∼ Poisson(λ), where λ is the mean number of holes, given byand σ is the fractional bilayer surface coverage (e.g., 0.99). Because holes represent bare substrate, particles were not allowed to occupy a position within a hole. The following rule was therefore applied to each proposed distance step: if the step carried the particle into a hole, the particle maintained its previous (x, y) position but was reassigned to the opposite leaflet, a movement that corresponds to translocation via lateral diffusion through the hole edge. To minimize the unphysical possibility of a particle “jumping” over a hole in a single step, the distance step was chosen such that δ < 0.1R.

To relate an ensemble of particle trajectories to asymmetry decay curves, each particle was initially and arbitrarily assigned a random position within the top leaflet. The fractional asymmetry parameter A was then calculated aswhere NP was the total number of simulated particles, and NP,top and NP,bot were the number of particles in the top and bottom leaflets, respectively. For the asymmetry decay curves presented in Figure d, a total of 104 particle trajectories were simulated in 10 boxes of side length l = 50 μm, each having a different random configuration of holes. The resulting decay curves were averaged and fit to a double exponential decay,with the slower decay component reported as the flip-flop rate constant kf in Figure e.