Quantitative Coherent Anti-Stokes Raman Scattering Imaging of Lipid Distribution in Coexisting Domains

Lipids

DOPC, DPPC, d62-DPPC, and cholesterol were purchased from Avanti Polar Lipids (Alabaster, AL) and used without further purification. BODIPY PC was purchased from Molecular Probes (Eugene, OR).

Preparation of supported lipid bilayers

Homogeneous supported bilayers are prepared by the small unilamellar vesicle (SUV) fusion method (54). A chloroform solution of a lipid mixture of desired molar ratio is evaporated under a nitrogen stream and then vacuumed for 3 h. The lipid mixture is rehydrated with a HEPES buffer (50 mM HEPES, 100 mM NaCl, 0.1 mM EDTA, pH = 7.0). The suspension is sonicated in an ultrasonic cleaner (No. G112SP1, Laboratory Supplies, Hicksville, NY) for 10 min and then centrifuged at 13200 rpm for 5 min. The SUV solution is mixed with an equal amount of HEPES buffer, which has a higher concentration of NaCl (250 mM) to increase the vesicle fusion speed. The mixed solution is immediately dropped in a perfusion chamber attached on a coverslip. The coverslips have been cleaned with a 7× detergent (No. IC7667094, ICN, Irvine, CA) and then baked at 600°C for 6 h. The DPPC and d62-DPPC samples are incubated at 60°C for 1 h and then cooled slowly down to room temperature at an average cooling rate of 0.3°C/min. The DOPC sample is directly incubated at room temperature for 1 h. The residual SUVs are washed away using the same buffer.

Preparation of giant unilamellar vesicles and planar bilayer patches

Giant unilamellar vesicles are prepared using the electroformation method (52). The total concentration of the lipid mixtures in a chloroform solution is 0.2 μmol/mL, containing 0.4 molar percent of BODIPY PC. A 10-μL glass syringe is used to add 6 μL of the mixture solution onto each of the two parallel Pt wires. After 1 h vacuum, the wires are inserted into a glass cell (No. 83301-01, Cole-Parmer, Vernon Hills, IL) filled with milli-Q water. The cell is then sealed with wax. Electroformation is carried out by using a 10-Hz sine wave with a V_peak-peak of 3 V for 2 h at 50°C. After that a 4-Hz sine wave at the same voltage is applied to dislodge the GUVs from the Pt wires. Lipid domains are formed after the GUVs are cooled slowly down to the room temperature. To make bilayer patches, the GUV solution is mixed with an equal amount of the HEPES buffer (100 mM NaCl) to assist fusion of GUVs to the glass substrate (53). After at least 1 h incubation at room temperature, the adjacent GUVs are fused and then rupture onto the coverslip to form planar bilayer patches.

CARS imaging

A schematic of our CARS microscope is shown elsewhere (47). Two co-linearly combined 2.5-ps pulse trains at frequencies ω_p and ω_s are generated from two synchronized Ti:sapphire oscillators (Mira 900, Coherent, Santa Clara, CA). A Pockels' cell (No. 350-160, Conoptics, Danbury, CT) is used to lower the repetition rate to 3.8 MHz. CARS images are acquired by raster scanning the two laser beams using a confocal microscope (FV300/IX70, Olympus, Tokyo, Japan). A water immersion objective with 1.2 N.A. is used to focus the laser beams into the sample. The confocal pinhole is not used for CARS imaging, which has a lateral resolution of 0.23 μm and an axial resolution of 0.75 μm (42). The forward and backward CARS signals around 600 nm are collected by an air condenser (N.A. = 0.55) and the water objective, respectively, and are simultaneously detected by two photomultiplier tubes (R3896 and H7422-40, Hamamatsu, Hamamatsu City, Japan) with a quantum efficiency of 0.15 and 0.4, respectively.

The large forward-detected CARS (F-CARS) signal is dominated by the nonresonant background from glass and water. The epi-detected CARS (E-CARS) images display high vibrational contrast because the nonresonant CARS background from the surrounding medium is mostly canceled out by the destructive interference among the backward CARS fields (36). Due to the refractive index mismatch between water and the coverslip, the forward CARS field generated from glass is partially reflected and coherently mixed with the backward resonant CARS field from the bilayer (30). It should be noted that the back-reflection amplitude highly depends on the axial position of the foci relative to the water/glass interface. To record the CARS images at the same focal depth, we carefully adjust the height of the objective lens to keep the F-CARS/E-CARS intensity ratio in a bare glass area to be the same. We also realize that both F-CARS and E-CARS intensities change with the detector efficiency at different wavelengths. To perform quantitative analysis, we normalize the E-CARS image intensity by the nonresonant F-CARS intensity from the glass/water interface. This preserves the vibrational contrast but rules out the signal change due to other factors.

The ratio of the pump to Stokes beam power was set as 2:1 to maximize the CARS signal at a certain excitation power. The pulse duration is very repeatable when we tune the laser wavelengths back and forth. We have been able to reproduce the CARS data owing to the high stability of Ti:Sapphire lasers.

CARS image analysis

We have developed a model for calculating the density of a specific vibrational oscillator in the excitation volume from the CARS intensities based on the following assumptions: 1), the resonant CARS signal arises from an isolated Raman band. This assumption is not valid for the symmetric CH₂ stretch vibration that is spectrally overlapped with the antisymmetric CH₂ stretch band and the CH₃ stretch bands. However, it holds for the symmetric CD₂ stretch band of d62-DPPC. 2), The nonresonant background from the scatterer is negligible compared to that from the surrounding medium. We have observed that the contrast for a bilayer patch on a coverslip is negligible when is tuned away from Raman resonance. Therefore, this assumption is valid for either forward- or epi-detected CARS signals from a single bilayer. 3), The resonant CARS signal is much smaller than the nonresonant background (). For a supported bilayer of pure d62-DPPC, the contrast, I_CARS(bilayer) / I_CARS(no bilayer) − 1, is measured to be 0.15 (see Fig. 1 A). Therefore, this assumption holds for the epi-detected resonant CARS signal from CD₂ groups in a single bilayer.

The CARS signal arises from the third-order susceptibility, of a material (55). When one Raman band is involved, can be written as

(1)

where δ represents the detuning, Ω is the vibrational frequency. Γ is the halfwidth at half maximum of an isolated Raman line. A is a constant representing the Raman scattering strength; n is the number of oscillators with the vibrational frequency Ω. represents the nonresonant contribution and is a constant in the absence of electronic resonance (32).

We consider a Raman scatterer (e.g., a lipid bilayer) surrounded by a medium that only contributes to the nonresonant background (). Assuming that the nonresonant background from the scatterer is negligible, the CARS intensity can be written as

(2)

The second item in the right side of Eq. 2 contributes to a constant background. The third term disperses the CARS spectral profile (32) and results in a peak (ω₊) and dip (ω₋) at

(3)

For a weak Raman scatterer (), the difference () between the intensities near the peak position () and the dip position () is proportional to The nonresonant background from surrounding medium, I_bg, is proportional to We have

(4)

The value of is proportional to the number of specific vibrational oscillators (e.g., the CD₂ groups) in the excitation volume. The above model can be applied to both F-CARS and E-CARS. In our experiment, I_bg is measured as the E-CARS intensity from the bare coverslip part of the CARS images. Because the excitation area is much smaller than the domain sizes, is proportional to the density of specific vibrational oscillators in a membrane domain. CARS intensities averaged over a constant area in a certain domain are used in our calculation.

Laser scanning fluorescence imaging

To correlate the CARS images with the membrane phases, we have labeled the bilayers with BODIPY PC that preferentially partitions into the more disordered phase (56,57). The peak emission wavelength of BODIPY PC at 510 nm is spectrally separated from the CARS wavelengths around 600 nm. Laser-scanning fluorescence imaging is carried out using the same microscope. A 488-nm Ar⁺ laser is used to excite BODIPY PC. For each bilayer sample, a fluorescence image is acquired before the CARS images. The fluorescent probe is quickly bleached by the picosecond excitation pulses and does not contribute to any contrast in the CARS images. Fluorescence images of GUVs and bilayer patches at different temperatures are acquired using a homebuilt temperature-controlled chamber.

CARS spectra of single lipid bilayers

Because of the interference between the vibrationally resonant CARS signal and the nonresonant background, the spectral profile of a CARS band differs from the corresponding Raman band (32). To identify the optimal wavelength for CARS imaging, we have recorded the CARS contrast spectra of a single DOPC bilayer and a single d62-DPPC bilayer. The CARS spectra shown in Fig. 1 were obtained from a series of epi-detected CARS images at different Raman shifts. Each image covers a bilayer part and a bare coverslip part. The CARS contrast was calculated as the signal difference between the bilayer and the bare coverslip part divided by the signal from the bare coverslip. Because of the interference between the resonant signal from the bilayer and the back-reflected nonresonant background, the peak positions for the CARS bands of symmetric CH₂ and CD₂ stretch vibration appear at 2840 and 2080 cm⁻¹, red-shifted from the center vibration frequencies at 2850 and 2100 cm⁻¹, respectively. Meanwhile, a dip is observed around 3000 and 2125 cm⁻¹ for the CH₂ and CD₂ vibration, respectively. Accordingly, we observed a positive contrast at the peak frequency and a negative contrast at the dip frequency in the CARS images shown beside the spectra. It should be noted that the spectral profiles in Fig. 1, A and B, are quite different. The CH₂ stretch vibration spectrum in the 2700–3100-cm⁻¹ region is complicated by several spectrally overlapped vibrational bands (48,49), whereas the CD₂ vibrational band exhibits a much simpler profile in the 2000–2200-cm⁻¹ region for the following reasons. The 2700–3100-cm⁻¹ region contains the Raman bands of symmetric and antisymmetric CH₂ and CH₃ stretch vibration (23). The Raman cross section for the antisymmetric CH₂ vibration is comparable to that for the symmetric CH₂ vibration. The CH₃ groups that are mainly located in the choline region have higher stretch vibration frequencies. The antisymmetric CH₂ vibration and the CH₃ vibration cancel the dip of the symmetric CH₂ stretch band and create a new dip around 3000 cm⁻¹. On the other hand, the antisymmetric CD₂ stretch band around 2200 cm⁻¹ is much weaker than the symmetric CD₂ stretch band around 2100 cm⁻¹. Additionally, the polar head of d62-DPPC is not deuterated. Only the stretch vibrational bands of the two CD₃ groups in the deuterated acyl chains could interfere with the 2100-cm⁻¹ band. Therefore, the CD₂ symmetric stretch vibration band can be treated as an isolated Raman band that exhibits a symmetrically dispersed CARS spectral profile (Fig. 1 A).

CARS intensity versus d62-DPPC molar fraction

Supported bilayers composed of different molar fractions of DPPC and d62-DPPC were used to test the sensitivity and the image analysis model. Fig. 2 A shows the CARS images of a bilayer composed of 10% d62-DPPC and 90% DPPC taken at room temperature. The nonresonant background is removed by subtracting the image at the dip frequency from the image at the peak frequency. In the difference image (Fig. 2 A, right), we observed a clear vibrational contrast from the 10% d62-DPPC in the bilayer whereas only noise in the bare coverslip part. Such sensitivity permits the detection of specific lipid molecules in different domains.

Fig. 2 B shows the difference CARS images between 2080 and 2125 cm⁻¹ for bilayers at different d62-DPPC molar percentages. The gel phase bilayers are homogeneous judging from the CARS contrast. We have calculated the values of for the CD₂ groups from the CARS intensity difference between 2080 and 2125 cm⁻¹. The validity of our model (Eq. 4) can be seen from the linear relationship between the measured value and the molar fraction of d62-DPPC in the sample (Fig. 2 C). With 0% d62-DPPC, we can hardly see any contrast (Fig. 2 B), indicating that the nonresonant background is completely removed in the difference images.

We have also taken the CARS images of the same samples at 2840 and 3000 cm⁻¹ and performed the same calculation for the CH₂ stretch band. As shown in Fig. 2 D, the obtained value is still linear with respect to the molar fraction of DPPC although several spectrally overlapped Raman bands are located in the CH₂ vibration region. The ratio of at CH₂ vibration for 100% DPPC to that for 100% d62-DPPC (0% DPPC) is 17.8/2.1 = 8.5. This value agrees with the 8:1 ratio between the 32 CH₂ groups in DPPC and the four CH₂ groups in the polar head of d62-DPPC.

Comparison of phase segregation in d62-DPPC-containing and DPPC-containing bilayers

Before we study the lipid domains with CARS microscopy, we have evaluated the effect of chain deuteration on the bilayer phase segregation. The chain melting temperature (T_m) of DPPC is 42°C. The T_m of d62-DPPC is lowered to 37.75°C due to the chain deuteration (58,59). Fig. 3, A and B, display the fluorescence images of the GUVs composed of DOPC/DPPC (1:1) and DOPC/d62-DPPC (1:1) taken at different temperatures. The DOPC/d62-DPPC GUVs are segregated into coexisting liquid and gel phases at room temperature as the DOPC/DPPC GUVs, with the liquid phase labeled by BODIPY PC. The miscibility temperature for the DOPC/DPPC GUVs is 35°C, consistent with previous measurement (60). The miscibility temperature for the DOPC/d62-DPPC GUVs is lowered by 4°C because of the deuteration.

When the GUVs rupture into bilayer patches, the lipid domains become nearly immobile in the bilayer patches as shown in previous studies (53). Only two BODIPY PC fluorescence intensities were observed, indicating that the patches were single bilayers and the domains symmetrically occupy both leaflets (11). To evaluate the effect of the glass substrate on the miscibility temperature of bilayers, we have monitored a single DOPC/DPPC (1:1) patch in a temperature-controlled chamber. As shown in Fig. 3 C, the domains become smaller with increasing sample temperature and eventually disappear at 37°. The higher miscibility temperature for supported bilayer patches is likely due to the stabilizing effect of the substrate (15). In summary, deuteration of hydrocarbon chains lowers the miscibility temperature between DPPC and DOPC by 4°C, but the domain pattern is the same in the GUVs containing deuterated and nondeuterated lipids.

Two-component bilayers

After a systematic characterization of the method, we studied the DOPC/d62-DPPC bilayer patches prepared by rupturing GUVs on a chambered coverslip. Fig. 4, A–D, shows the E-CARS images of the DOPC/d62-DPPC (1:1) bilayer at the peak and dip positions of CH₂ and CD₂ vibration, respectively. Similar to the bilayers of DOPC and d62-DPPC shown in Fig. 1, we observed an inversion of the CARS contrast when was tuned from the peak to the dip position for both CH₂ and CD₂ stretch bands. The bright CD₂ vibration contrast shown in the difference CARS image (Fig. 4 F) demonstrates that d62-DPPC is highly concentrated in the gel phase that is not labeled by BODIPY PC (Fig. 4 E). Noting that the CARS signal is affected by molecular orientation (44), we monitored the CARS signal while rotating the excitation polarization direction in the x-y plane. The CARS signal is invariant, indicating that the ordered phase is likely in the P_β′ phase but not in the tilted L_β′ phase. The P_β′ phase has been observed as a rippled phase in AFM studies (61). The molecular axis in the P_β′ phase is vertical to the substrate, thus the CARS intensities reflect the density of specific molecules in the excitation volumes. It is interesting that the bilayer patch displays little contrast at the CH₂ stretch vibration frequency (Fig. 4 A). The resonant CH₂ signal from the liquid phase is mainly contributed by the abundant DOPC molecules. The resonant CH₂ signal from the gel phase could be contributed by the d62-DPPC molecules that have CH₂ groups in the head region and also by the DOPC molecules that are incorporated into the gel phase. A quantitative analysis is presented below.

Using Eq. 4 and the CARS images shown in Fig. 4, we have calculated the values of for the CD₂ and CH₂ groups in the gel and liquid domains. The results are illustrated in Fig. 5. For the DOPC/d62-DPPC (1:1) bilayer, the value of CD₂ is 8.40 and 0.96 for the gel and liquid phase, respectively. The value of CD₂ for pure d62-DPPC bilayers is measured to be 11.41.

In our temperature-control experiment of the pure d62-DPPC bilayer patch, we have observed that the value decreases by two times when the temperature is increased from 23 to 46°C. This implies that d62-DPPC in the gel and liquid phase has different configuration and gives different CARS intensities. To evaluate the molar density of d62-DPPC and DOPC in each domain, we make two assumptions. First, d62-DPPC in the gel (or L_d) domains has the same configuration as in the pure d62-DPPC bilayer in the gel (or L_d) phases, respectively. The same assumption is made for DOPC. Second, d62-DPPC and DOPC have the same area per molecules in either the gel or the liquid domain. This assumption is supported by the x-ray measurements of pure lipid bilayers (62). Based on the above assumptions, the molar density of d62-DPPC in the gel phase is calculated to be 8.40/11.41 = 0.74. The molar density of DOPC in the gel phase is 1.0 − 0.74 = 0.26.

Analysis for the CH₂ group is complicated because both DOPC and d62-DPPC contribute to the CH₂ signal. The value for CH₂ in a pure DOPC bilayer is 9.73. We measured the d62-DPPC bilayer at 46°C and obtained the value to be 1.02. The value for the L_d phase of the DOPC/d62-DPPC bilayer is measured to be 8.03. Assume the molar fraction of DOPC is y, we have the following equation,

The molar density of DOPC and d62-DPPC in the L_d domain is found to be 0.80 and 0.20, respectively.

Three-component bilayers

In GUVs composed of DOPC, cholesterol, and DPPC, L_o domains have been observed in fluorescence microscopy (12,63). Using CARS microscopy, we have studied the DOPC/d62-DPPC (1:1) bilayers incorporated with 10, 20, 30, and 40% cholesterol. At 10% cholesterol level, we did not observe large-scale domains in either GUVs or bilayer patches (see Fig. 6 A). For the 20% cholesterol sample, round L_o domains were observed in the percolating L_d domain in the GUV fluorescence images. The domains became static and exhibited irregular shapes after we ruptured the GUVs onto a coverslip (see Fig. 6 B), but the individual lipids in the bilayer patch were shown to be mobile in a fluorescence recovery after photobleaching study by Kaizuka et al. (53). For convenience, we call the observed domains in the bilayer patches as coexisting L_o and L_d domains. For the 30% cholesterol samples, the L_o phase becomes the percolating domain in some GUVs. In the bilayer patches containing 30% cholesterol, the L_o phase appears as the percolating domain (Fig. 6 C).

The CARS images of the DOPC/d62-DPPC(1:1)/cholesterol bilayer patches allow us to quantify the role of cholesterol in lipid organization in the bilayer patches. Quantitative analysis (see Fig. 5) was carried out in a similar way. At 10% cholesterol level, the value for CD₂ is measured to be 2.29 for the whole bilayer. For the 20% cholesterol sample, the positive contrast at the CD₂ stretch vibration frequency of 2080 cm⁻¹ (Fig. 6 B) demonstrates that d62-DPPC preferentially partitions into the L_o domain that is not labeled by BODIPY PC. The smallest domain we can identify is 0.25 μm in width, corresponding to the lateral resolution of our setup. With 20% cholesterol incorporated into the bilayer, the value for CD₂ drops from 8.40 in the gel phase to 7.07 in the L_o phase, corresponding to a decrease of the d62-DPPC molar density from 0.74 to 0.62. Meanwhile, we find an increase of the value from 0.96 to 1.32 in the L_d phase. Our observation suggests that cholesterol “drives” some d62-DPPC molecules from the more-ordered phase to the less-ordered phase. The observed change in d62-DPPC partition is accompanied by the change of the values for CH₂. In the DOPC/DPPC (1:1) bilayer, the CH₂ concentration in the gel phase is lower than that in the L_d phase. With 20% cholesterol, we observed a slightly higher CH₂ concentration in the L_o phase than in the L_d phase (Fig. 5 A), indicating a larger partition of DOPC into the L_o than the gel phase.

When the cholesterol level is increased to 30%, we observed a uniform CARS intensity in the whole bilayer (see Fig. 6 C). Because the BODIPY PC labeled L_d domain is ∼1 μm in size, we should be able to resolve them in the CARS image. Thus, the uniform CARS intensity indicates that the concentration difference of d62-DPPC between the L_o and L_d phase is below 10%, our detection sensitivity. With 30 and 40% cholesterol incorporated into the bilayer, the value of CD₂ for the whole bilayer is measured to be 3.28 and 2.88, respectively. The molar density is calculated to be 0.29 and 0.25, close to the molar fraction of d62-DPPC (0.35 and 0.30). This supports that the d62-DPPC/DOPC bilayers are highly compact in the presence of high-level cholesterol.

Evaluation of CARS microscopy for studies of membrane organization

The major advantage of CARS is its chemical selectivity without fluorophore labeling. In our experiments, we have used the CH₂ and CD₂ stretch Raman bands because coherent addition of CARS radiations from the large number of CH₂ or CD₂ groups builds up a large signal, allowing us to probe a specific lipid with a molar concentration as low as 10% in a single bilayer. With the improvement of CARS imaging sensitivity on the basis of new generations of lasers and detectors, one will be able to use the weaker Raman bands from the polar head region to distinguish different lipid molecules without any labeling. In comparison with secondary ion mass spectrometry that probes dehydrated samples under high vacuum (21), CARS microscopy is superior in that it is noninvasive and can probe membrane samples under physiological conditions. Recently, CARS microscopy has been applied to image myelin sheath in live spinal tissues with three-dimensional spatial resolution and a penetration depth >250 μm (47).

Several strategies have been combined to push the sensitivity limit in the current work. First, we use two synchronized picosecond lasers to generate the CARS signal. Because the spectral profile of a picoseond pulse matches with the line width of most Raman bands, all the energy can be used to drive a single molecular vibration, providing much higher sensitivity than excitations with femtosecond pulses (35). Second, instead of using amplified laser pulses at low repetition rates, we use synchronized Ti:sapphire oscillators and pulse picking devices to generate two pulse trains with pulse energy of a few nanojoules and repetition rate of a few megahertz. This excitation scheme not only avoids nonlinear photodamage, but also permits high-speed imaging. Third, we make use of epi-detection that avoids the forward-going nonresonant background arising from the solvent and the glass substrate. Fourth, by measuring the intensity difference at the peak and the dip positions, we have not only removed the back-reflected nonresonant background from glass/water interface, but also utilized the dispersive feature of a CARS band to further increase the resonant signal. This is superior over other methods such as polarization CARS imaging wherein the resonant signal can be significantly reduced by polarization sensitive detection (37). These careful designs allowed us to generate a clear vibration contrast from d62-DPPC with a 10% molar concentration in a single bilayer (see Fig. 2). We should note that with picosecond excitation, two images need to be acquired to derive quantitative concentration information. This can be avoided by multiplex CARS microscopy that uses a narrowband pump beam and a broadband Stokes beam (40,41,64), providing that the CARS spectral imaging sensitivity will be significantly improved with new generations of electron multiplying CCDs. In addition to molecular mapping, the polarization sensitivity (44) and the spectral characterization capability (40,41,65) can be applied to study the ordering degree and conformation of lipid molecules without any labeling. Finally, the lateral resolution of far-field CARS microscopy is ∼0.25 μm. Thus, we could not probe the nanoscale domains that have been observed by AFM (15). This problem could be solved by using near-field CARS microscopy (66).

Implications of the role of cholesterol in bilayer organization

It is known that cholesterol broadens the gel to liquid phase transition of saturated lipids (67) and creates the L_o phase in the bilayer (68). It is generally believed that cholesterol associates strongly with saturated lipid and poorly with the unsaturated lipids (69). However, cholesterol may have complicated interactions with phospholipids as reviewed in a number of articles (69–73). The information about d62-DPPC distribution obtained from our CARS images allows a better understanding of the role of cholesterol in a three-component bilayer, as discussed below.

In the DOPC/d62-DPPC (1:1) bilayer patches, we have shown that the L_d domain contains ∼80% DOPC and 20% d62-DPPC. This result is supported by our observation that no domains are formed in the bilayer patches when the molar percentage of d62-DPPC is below 25%. The gel domain contains ∼74% d62-DPPC and 26% DOPC, indicating that the phase segregation is induced by hydrophobic interactions that aggregate saturated lipids in the ordered gel phase.

With 10% cholesterol added to the 1:1 DOPC/d62-DPPC mixture, the domains disappear as reported before (63). As shown in Fig. 6 A, d62-DPPC is uniformly distributed in the whole bilayer on the scale of our 250-nm spatial resolution. A similar observation has been reported by Feigenson and Buboltz in the DPPC/DLPC/cholesterol GUVs (74). They suggested that a three-component bilayer at a certain level of cholesterol could form a fluid ordered phase. The fluid ordered phase provides PC molecules with a larger capacity to shield cholesterol from bulk water than the L_d phase. The value of CD₂ for the 10% cholesterol bilayer is measured as 2.29, much smaller than 8.40 (gel phase in DOPC/d62-DPPC bilayer) but higher than 0.96 (L_d phase in DOPC/d62-DPPC bilayer). This supports that the bilayer is in a fluid ordered phase.

Interestingly, the addition of 20% cholesterol causes the formation of percolating L_d and circular L_o domains in GUVs. Our CARS data show that d62-DPPC is enriched in the L_o domain. Meanwhile, cholesterol increases the partition of d62-DPPC into the DOPC-enriched phase, with a decrease of the partition ratio between the ordered and disordered domains from 8.7 at 0% cholesterol level to 5.4 at 20% cholesterol level (see Fig. 5 B). Our observation can be interpreted as an interplay of different types of lipid-lipid interactions. On one hand, cholesterol tends to have a uniform distribution due to long-range repulsive forces (71). On the other hand, hydrophobic interactions associate cholesterol with d62-DPPC in the form of molecular complex (73).

When the level of cholesterol is increased to 30 and 40%, d62-DPPC becomes homogeneously distributed in the whole bilayer. The homogenization effect of high-level cholesterol has been qualitatively observed in fluorescence imaging studies. The whole bilayer becomes homogeneous when the cholesterol molar ratio exceeds a certain value (12,63,74), while depleting cholesterol leads to phase separation in the plasma membrane (75).

Keller and co-workers have measured the fraction of d62-DPPC in the coexisting domains using NMR spectroscopy (17). For multilamellar vesicles composed of DOPC/d62-DPPC (1:1) and 30% cholesterol at 25°C, the fraction of d62-DPPC in the L_o and L_d phases is 53 and 18%, respectively (17). This is quite different from our measurements of single bilayers shown in Fig. 5. The difference could be due to the interbilayer interactions. The domain sizes of their sample were ∼80 nm. As the authors have suggested, those nanoscale organizations could be molecular complexes instead of separate phases (17).