Subnanometer Structure of an Asymmetric Model Membrane: Interleaflet Coupling Influences Domain Properties

Cell membranes possess a complex three-dimensional architecture, including nonrandom lipid lateral organization within the plane of a bilayer leaflet, and compositional asymmetry between the two leaflets. As a result, delineating the membrane structure–function relationship has been a highly challenging task. Even in simplified model systems, the interactions between bilayer leaflets are poorly understood, due in part to the difficulty of preparing asymmetric model membranes that are free from the effects of residual organic solvent or osmotic stress. To address these problems, we have modified a technique for preparing asymmetric large unilamellar vesicles (aLUVs) via cyclodextrin-mediated lipid exchange in order to produce tensionless, solvent-free aLUVs suitable for a range of biophysical studies. Leaflet composition and structure were characterized using isotopic labeling strategies, which allowed us to avoid the use of bulky labels. NMR and gas chromatography provided precise quantification of the extent of lipid exchange and bilayer asymmetry, while small-angle neutron scattering (SANS) was used to resolve bilayer structural features with subnanometer resolution. Isotopically asymmetric POPC vesicles were found to have the same bilayer thickness and area per lipid as symmetric POPC vesicles, demonstrating that the modified exchange protocol preserves native bilayer structure. Partial exchange of DPPC into the outer leaflet of POPC vesicles produced chemically asymmetric vesicles with a gel/fluid phase-separated outer leaflet and a uniform, POPC-rich inner leaflet. SANS was able to separately resolve the thicknesses and areas per lipid of coexisting domains, revealing reduced lipid packing density of the outer leaflet DPPC-rich phase compared to typical gel phases. Our finding that a disordered inner leaflet can partially fluidize ordered outer leaflet domains indicates some degree of interleaflet coupling, and invites speculation on a role for bilayer asymmetry in modulating membrane lateral organization.

Leaflet compositions for an asymmetric LUV sample prepared from POPC-dHC acceptor and POPC donor, determined from GC, 1 H-NMR, and SANS modeling. Data columns as described in Table S2 legend.  Table S4 Leaflet compositions for an asymmetric LUV sample prepared from POPC-dH acceptor and DPPC-dC donor, determined from GC, 1 H-NMR, and SANS modeling. Data columns as described in Table S2 Table S8 LUV size and polydispersity determined by dynamic light scattering before and after centrifugal filtration. Uncertainty is estimated to be ± 5 nm. All measurements were performed at room temperature.
Step 2, aqueous methyl-beta cyclodextrin (mβCD, 35 mM) is added to the MLV pellet in an 8:1 mβCD:donor ratio and incubated for 2 h at room temperature while stirring.
Step 3, a suspension of acceptor large unilamellar vesicles (LUVs) is added to the donor/mβCD sample to achieve a desired donor:acceptor molar ratio and an mβCD concentration of ~ 29 mM, then incubated for 1 h at room temperature while stirring. Acceptor LUVs can be prepared in low osmolarity buffer (e.g., 10−30 mM NaCl) to balance residual solutes remaining after the asymmetric sample preparation. Depending on the mβCD concentration, a small fraction of the heavy donor vesicles may be dissolved and reformed as light small unilamellar vesicles (SUVs).
Step 4, the mixture is diluted 8-fold with H 2 O, then centrifuged for 30 min at 20K × g (pellet is discarded).
Step 5, supernatant from Step 4 is first concentrated to 0.5−1 mL with a 100K MWCO centrifugal filter. The remaining mβCD is then removed by repeated dilution (with H 2 O or D 2 O, depending on experimental needs) and concentration steps, with the filtrate discarded between steps, to achieve a desired dilution factor.
Step 6, asymmetric sucrose-free LUVs in water are recovered from the retentate following the final wash.

Figure S3 | GC determines the total composition of lipid mixtures.
A, A lipid sample is subjected to an acid-or base-catalyzed transesterification, converting individual chains to volatile FAMEs suitable for GC analysis (inset cartoon). A binary equimolar mixture of POPC and POPC-dC yields a 2:1:1 ratio of methyl oleate, methyl palmitate, and methyl palmitate-d31, which is reflected in the relative peak areas from the total ion chromatogram. B, Changing the mixture composition alters the relative areas of the methyl palmitate and methyl palmitate-d31 peaks. The composition of an unknown sample can therefore be obtained from its peak area fraction. C, Detection inefficiencies result in a nonlinear dependence of peak area fraction vs. mixture composition. Precise quantitation of an unknown sample requires comparison to a standard curve obtained from mixtures of known composition.      Figure S10 | SANS data and fits for symmetric POPC vesicles. Experimental SANS data (circles) for different contrast symmetric POPC bilayers. Data were modeled individually (solid lines) and jointly (dashed lines) with a symmetric four shell scattering length density profile as described in the Supporting Information text. Predicted scattered intensity is shown as solid lines, with best fit parameters given in Table S5. Data are vertically offset by powers of 10 for clarity.

Figure S11 | Osmotic imbalance generates membrane tension and thins POPS vesicles.
Experimental SANS data for symmetric 100 nm diameter POPS LUVs, with different core/solvent conditions. Osmotically stressed POPS vesicles containing a 25% d-sucrose core (red) in an external D 2 O solvent show a decreased bilayer thickness (−1.8 Å) and an increased area per lipid (+4 Å 2 ) compared to stress-free POPS vesicles prepared in D 2 O (blue), consistent with vesicle swelling and lateral bilayer expansion. Data are vertically offset for clarity.

Figure S12 | 1 H-NMR detection mβCD.
Titration of mβCD into D 2 O (left) reveals the solute's detection limit with proton NMR. The characteristic CD peaks occur at 3.29 ppm and 3.46 ppm.
Comparison of corresponding resonances in an asymmetric LUV sample provides a lower limit for residual mβCD concentration of 1:130 CD:lipid molar ratio. Titration of mβCD into an LUV/D 2 O dispersion (right) reveals the solute's quantity relative to the lipid (CD:lipid). The choline peak (3.17 ppm) was used to determine CD:lipid ratios. Spectra are vertically offset for clarity. Inset, proton NMR spectra of an asymmetric LUV sample reveals a low quantity of CD relative to lipid (~7:100 molar ratio). An octet at 3.6 ppm is attributed to trace glycerol contamination originating in the centrifugal filters, despite extensive pre-washing as described in the Supporting Information Section S2.
Figure S13 | SANS is sensitive to the presence of multilamellar vesicles (MLVs). Shown are SANS data for a 10 mM POPC-dC vesicle suspension in D 2 O, before (MLVs, gray circles) and after (LUVs, black triangles) extrusion to produce 50 nm diameter LUVs. For MLVs, density correlations between the stacked bilayers give rise to Bragg scattering peaks at a length scale corresponding to integer multiples of the lamellar repeat distance (e.g., the first Bragg order at q ~ 0.1 Å -1 , corresponding to a lamellar repeat distance of ~ 63 Å). Following extrusion to produce unilamellar vesicles, Bragg peaks are no longer observed, and vesicles exhibit the typical form factor for a dilute spherical shell particle. The inset shows a weighted sum of the black and gray curves as indicated in the inset legend, demonstrating the sensitivity of SANS to MLV contamination.

Figure S14 | 1 H-NMR is sensitive to the presence of small unilamellar vesicles (SUVs).
Shown are choline resonances from 14.6 mM POPC LUVs (red) and 5.3 mM POPC SUVs (black), as well as the sum of the LUV and SUV spectra (blue curve, offset for clarity). The shoulder at 3.16 ppm demonstrates that ~ 27 mol % SUV can be easily identified in the presence of LUVs. Inset, small unilamellar vesicles (SUVs) give rise to a split choline resonance even in the absence of extravesicular shift reagent. This is due to both lipid number density and packing differences between the inner (green area) and outer (red area) leaflet. SUVs were obtained from an acceptor-free prep with omitted washing steps.  The pellet was discarded, and the supernatant (containing asymmetric LUVs, mβCD, and residual sucrose) was concentrated to ~ 1 mL using a prewashed 100K MWCO centrifugal filter device at 5,000 × g. Soluble contaminants (i.e., mβCD and sucrose) were removed by successive dilution/concentration cycles, whereby the sample was diluted with D 2 O to the filter device's capacity (~ 11 mL) and then centrifuged at 5,000 × g to obtain a final retentate volume of 0.5−1 mL. The time for centrifugal filtration varied depending on the phase state of the lipids and the quantity of asymmetric vesicles being washed, and ranged from 30-60 min per wash. Typically, four such cycles reduced the mβCD concentration by a factor of > 10, and exchanged > 99% of H 2 O with D 2 O. The asymmetric vesicle preparation is summarized in Fig. S2. The final yield was estimated as not less than half of the initial acceptor amount.

S3. Gas chromatography (GC).
Phospholipids were converted to fatty acid methyl esters (FAMEs) via acid catalyzed methanolysis. Briefly, 5−10 μL of an aqueous vesicle suspension (containing 20−100 μg total lipid) was dispensed into a 13 × 100 mm screw top glass culture tube, followed by addition of 1 mL methanolic HCl (1 M) prepared with concentrated HCl and methanol. 5 The sample was vortexed, sealed under Ar, and incubated at 85 °C for 1 h. After cooling to room temperature, 1 mL H 2 O was added and the sample was vortexed. FAMEs were extracted with 1 mL hexane and vigorous vortexing, followed by low-speed centrifugation (500 × g) for 10 min. Finally, 800 µL of the upper (hexane) phase were transferred to an autosampler vial and brought to 1 mL with hexane, for injection into the GC column.
GC analysis was performed on an Agilent 5890A gas chromatograph (Santa Clara, CA) with a 5975C mass-sensitive detector operating in 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. A 1 μL aliquot of FAME dissolved in hexane was injected in splitless mode using an Agilent where denotes the i th chain peak area and the denominator is a sum over all mixture components , ∈ . For this relationship to be strictly valid, chain peak area fractions must vary linearly with mixture composition. In practice, we found a slight deviation from linearity which necessitated the use of standard curves (Fig. S3). Briefly, standard two-component FAME samples covering a range of compositions and containing 20−100 μg total lipid, were prepared by dispensing lipid stock solutions into glass culture tubes. Chloroform was removed by an N 2 stream and gentle heating, followed by derivatization and extraction of FAMEs as described above. Peak area fractions were plotted vs. component mole fraction and fitted to a four-parameter function: using Mathematica 10.0 (Wolfram Research, Champaign, IL) to obtain a standard curve, from which the composition of unknown samples was determined. proportional to the number of molecules having protiated headgroups in the corresponding leaflet. We define the outer leaflet peak fraction:

S5. Proton nuclear magnetic resonance spectroscopy ( 1 H-NMR). 1 H-NMR
where the superscript denotes the inner (in) or outer (out) leaflet. If all bilayer components possess protiated headgroups, directly yields the mole fraction of all bilayer lipids found in the outer leaflet, Χ : where and denote number of molecules in the whole bilayer and in the outer leaflet, respectively, where ( ) is the projected scattering length density (SLD) in the direction normal to the bilayer plane, is the scattering length density of the solvent (water), and the integral is evaluated over the full bilayer thickness. 10 We simplify the asymmetric bilayer's SLD profile by considering four slabs of independent and thickness , in addition to the SLD of the vesicle core and external solvent. This model is represented graphically in Fig. S9 and mathematically as: where superscripts in and out refer to the inner and outer leaflets, respectively. Provided the core SLD matches that of the solvent ( = ), the asymmetric form factor has an analytical solution: where is the headgroup volume in Å 3 , is the coherent neutron scattering length in fm, is the average number of bound waters per headgroup, is the component mole fraction, the subscript i indexes the bilayer's lipid components, and the superscript j indexes the two leaflets. The analogous expressions for the hydrocarbon slabs are: The slab thicknesses and scattering length densities follow directly: where is the slab thickness in Å, and is the average area per lipid in Å 2 . With lipid volumes and scattering lengths constrained by independent measurements (Table S1) where and are obtained with GC and NMR experiments, respectively, and Χ (cf. Eq. 4) is given by the inner and outer leaflet areas per lipid: Compositional parameters for isotopically asymmetric POPC bilayers obtained from GC, NMR, and SANS experiments are given in Tables S2-S3. Structural parameters obtained from SANS analysis of isotopically asymmetric POPC bilayers are given in Table S6, with data and best-fit curves shown in Fig.   3a. For comparison, structural parameters obtained from a joint refinement of symmetric POPC variants are found in Table S5, with data and best-fit curves shown in Fig. S10.
For k coexisting bilayer phases, assuming negligible interference, the observed intensity is given by a weighted sum of asymmetric form factors for each phase: where is the phase area fraction given. For a binary mixture of two coexisting phases (e.g., phases 1 and 2), four bilayer compartments must be considered, namely the outer and inner leaflets of phases 1 and 2. Using superscripts i and o to refer to the inner and outer leaflets, respectively, the unconstrained model has eight structural parameters ( 1 , 2 , 1 , 2 , 1 , 2 , 1 , 2 ) and four compositional parameters and the inner leaflet phase 2 mole fraction of component A is given by: leaving 11 free parameters, an unacceptably large number. Therefore, additional assumptions were made when modeling DPPC/POPC aLUVs in order to reduce the number of free parameters to a reasonable level: (1) the number of headgroup waters for both leaflets was fixed to 7; and (2) given the small amount of DPPC found in the inner leaflet by NMR, we assumed that the inner leaflet was a uniform fluid phase and that the DPPC-poor phase was symmetric, allowing us to jointly vary three parameters (i.e., 1 = 2 = 1 ). The final model had four adjustable parameters as indicated in Table S7, which lists all structural and compositional parameters obtained from the fit (data and best-fit curves are shown in Fig.   3b). Compositional parameters obtained from GC, NMR, and SANS analysis are given in Table S4, where Χ is given by: Χ = 1 / 1 + (1 − 1 )/ 2 1 (1/ 1 + 1/ 1 ) + (1 − 1 )(1/ 2 + 1/ 2 ) . (24) S9. Assessment of sample contamination. We define a contaminant as any impurity whose presence can affect (or hinder the determination of) the asymmetric LUV bilayer structure. In this context, the most problematic contaminants are mβCD and residual donor or mixed donor/acceptor vesicles that resist sedimentation. These are typically lighter donor multilamellar vesicles (MLVs), or small unilamellar vesicles (SUVs, diameter < 30 nm) formed during the exchange process. As a lipid carrier molecule, mβCD facilitates outer leaflet exchange and may perturb the bilayer structure, while residual vesicles can bias the determination of the asymmetric leaflet compositions.
We assessed mβCD contamination by establishing the 1 H-NMR detection limits of specific mβCD resonances in D 2 O in the absence and presence of LUVs (Fig. S12), from which we estimate a lower detection limit of 1:130 CD:lipid molar ratio. Asymmetric LUVs typically contained < 1:10 CD:lipid after three wash steps (Fig. S12 inset).
The presence of residual vesicles can be assessed with a variety of techniques. Vesicles containing multiple lamellae (including contaminating donor MLVs) exhibit a series of Bragg peaks in the SANS intensity at = 2 / , where D is the lamellar repeat distance and n is an integer. As illustrated in Fig. S13, mixtures of LUVs and MLVs show characteristic excess scattering near the first Bragg order (n = 1, ~ 0.1 Å -1 ). The absence of Bragg scattering in asymmetric samples is confirmed by a good fit between data and model (which assumes unilamellar vesicles), as well as simple visual inspection of the data.
A less obvious form of residual vesicle contamination is the presence of SUVs with diameter < 30 nm, which can be generated upon lengthy exposure of lipid vesicles to mβCD. GC measurements of donor-only control samples revealed 1−2% of the total donor mass in the recovered sample, which was subsequently identified as SUVs with 1 H-NMR lineshape analysis (Fig. S14). Briefly, the SUV choline resonance is characterized by the appearance of two peaks even in the absence of shift reagent, attributable to packing differences in the inner and outer leaflets of highly curved vesicles. 7 The SUV choline resonance width is ~ 0.25 ppm-considerably narrower than the LUV width-with inner and outer resonances separated by 0.019 ppm (7.6 Hz). However, despite their considerable line width differences, the similarity of the spin-lattice relaxation time (T1) values for LUV and SUVs (Fig. S15) precludes the ability to isolate the SUV component by varying the delay time. NMR measurements of asymmetric samples did not indicate the presence of SUVs (Fig. 2a, Figs. S4-S6), although low contamination levels may fall below the detection threshold (Fig. S14). If complete removal of SUVs is necessary, sucrose density gradients can be employed prior to mβCD removal.