Coexistence of Probe Conformations in Lipid Phases—A Polarized Fluorescence Microspectroscopy Study

Abstract

Several well-established fluorescence methods depend on environment-sensitive probes that report about molecular properties of their local environment. For reliable interpretation of experiments, careful characterization of probes’ behavior is required. In this study, bleaching-corrected polarized fluorescence microspectroscopy with nanometer spectral peak position resolution was applied to characterize conformations of two alkyl chain-labeled 7-nitro-2-1,3-benzoxadiazol-4-yl phospholipids in three model membranes, representing the three main lipid phases. The combination of polarized and spectral detection revealed two main probe conformations with their preferential fluorophore dipole orientations roughly parallel and perpendicular to membrane normal. Their peak positions were separated by 2–6 nm because of different local polarities and depended on lipid environment. The relative populations of conformations, estimated by a numerical model, indicated a specific sensitivity of the two probes to molecular packing with cholesterol. The coexistence of probe conformations could be further exploited to investigate membrane organization below microscopy spatial resolution, such as lipid rafts. With the addition of polarized excitation or detection to any environment-sensitive fluorescence imaging technique, the conformational analysis can be directly applied to explore local membrane complexity.

Introduction

Since the original hypothesis about lipid phase organization in cell membranes was proposed (1), lipid rafts, nanodomains, and microdomains have been identified in model and cell membranes and found to modulate numerous biological processes (2), as demonstrated by a variety of fluorescence methods, infrared spectroscopy, electron and nuclear (para)magnetic resonance spectroscopy, x-ray scattering, atomic force microscopy, etc. (3–15). Fluorescence techniques are frequently chosen because of their applicability to live-cell experiments, the ultimate sensitivity of optical methods, and the ability to visualize the sample. Moreover, fluorescence microscopy contrasted by local spectral, relaxation, diffusion, or energy-transfer characteristics provides additional information about nanometer-sized supramolecular arrangements in the membrane with localization resolution within the optical diffraction limit.

As fluorescence methods typically rely on the use of probes, the latter have been extensively developed in parallel with experimental needs. Besides the probes that mark lipid phases by selective partitioning (3-5), several fluorophores have been introduced that change their quantum yield, absorption/emission spectrum, lifetime, or anisotropy with respect to local polarity, hydration, molecular order, or membrane potential (16). Among such dyes, 7-nitro-2-1,3-benzoxadiazol-4-yl (NBD) has attracted a lot of attention (17,18) because of its significant solvatochromic effect (19–21) and comparatively high amenability to chemical modification, enabling versatile applications (22).

It can be easily overlooked, however, that the probes are subjected to complex specific interactions with neighboring lipids and proteins, often resulting in multiple or unexpected conformations, locations, and H-bonding states (18,23–25). Hence, careful systematic studies are needed to characterize the behavior of each probe in different environments to avoid potential misinterpretations of experimental results. For the probes with NBD attached to the alkyl chain (Fig. 1), it has been shown that the tails loop back, bringing the polar fluorophore to the lipid–water interface (21,26–28). In addition, a bimodal distribution of reported local polarities has been observed, supposedly attributed to snorkeling and extended conformations of alkyl chains (24), which could in addition depend on the lipid phase of the environment (25).

Figure 1.

Chemical structures of the two fluorescent probes used: (A) C6-NBD-PC and (B) C12-NBD-PC.

To explore probe conformations, we incorporated two widely used lipid-based NBD probes into giant unilamellar vesicles of three different lipid compositions, representing the three main lipid phases. Differences in local polarity, experienced by NBD, were detected by fluorescence microspectroscopy (FMS) that visualizes local spectral characteristics of emitted fluorescence light throughout the image. Nanometer spectral peak position resolution was achieved by improved bleaching-corrected spectral fitting (29), placing spectral sensitivity of FMS experiments on microscopic objects in line with bulk spectrofluorimetric measurements. Polarized detection, implemented automatically by spectral acquisition through a liquid crystal tunable filter (LCTF), additionally allowed estimation of preferential orientations of fluorophores’ dipole moments relative to the membrane normal, as in fluorescence polarization microscopy (30). The combination of the two concepts (i.e., spectral and polarized detection) enabled recognition and characterization of two main probe conformations coexisting at distances below spatial resolution. Their relative populations and respective peak positions were found to strongly depend on molecular organization of the environment, especially at high concentrations of cholesterol.

This finding suggests that for well-characterized probes, a coexistence of their molecular conformations could be efficiently exploited by polarized FMS (pFMS) to examine lipid domains with sizes below spatial resolution. Moreover, the concept of polarization-dependent measurements and conformation modeling could be directly transferred to any other fluorescence imaging technique that provides some additional molecular features from specific fluorescence light characteristics.

Materials and Methods

Chemicals

Phospholipids 1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC) and 1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC) were purchased from Avanti Polar Lipids (Alabaster, AL), as were the phospholipid-based probes 1-palmitoyl-2-{6-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]hexanoyl}-sn-glycero-3-phosphocholine (C6-NBD-PC) and 1-palmitoyl-2-{12-[(7-nitro-2-1,3-benzoxadiazol-4-yl)amino]dodecanoyl}-sn-glycero-3-phosphocholine (C12-NBD-PC), schematically presented in Fig. 1. Cholesterol (chol), 1,2-diacyl-sn-glycero-3-phospho-[1-rac-glycerol] (PG), and sucrose were obtained from Sigma Aldrich (St. Louis, MO), and glucose was obtained from Kemika (Zagreb, Croatia). Organic solvents chloroform and methanol were purchased from AppliChem GmbH and Merck KGaA (both from Darmstad, Germany), respectively. All chemicals were used without further purification.

Sample Preparation

Giant liposomes were prepared by the gentle hydration method (31) from DPPC, DOPC, and DPPC+chol (3:2 mol/mol), representing gel (S), liquid disordered (Ld), and liquid ordered (Lo) phase, respectively. To each composition 15 mol% of charged PG lipids were added to induce formation of unilamellar vesicles. Lipids were dissolved in 0.5 ml of chloroform–methanol mixture (7:3 vol/vol) at concentration 0.1 mg/ml together with one of the two probes in probe-to-lipid molar ratio 1:250. The probe concentrations were low enough to prevent efficient energy homotransfer (32) or probe aggregation (33) (see Fig. S1 in Supporting Material).

Organic solvents were evaporated by rotary evaporator (Rotavapor R-200, Büchi Labortechnik AG, Postfach, Switzerland) for 2 h at 60°C, followed by additional 3 h at vacuum pump at room temperature to form a dry lipid film on the walls of the glass tube. The lipids were then left to prehydrate in water vapor-saturated atmosphere for 30 min at 60°C. Next, 2 ml of 0.1 M sucrose solution, preheated to 60°C, were gently added to the test tube, which was then left to incubate at 60°C overnight. After having cooled down slowly, the giant unilamellar vesicles suspension was gently transferred to a glass vial and used for measurements the same day.

Spectrofluorimeter Measurements

Reference fluorescence emission spectra (Fig. S2) were measured at Infinite M1000 microplate reader (Tecan, Männendorf, Switzerland) at room temperature. A 96-well black plate was used in the fluorescence intensity top mode. Fluorescence was excited at 450 nm and emission spectra recorded from 480 to 650 nm, both excitation and emission bandwidths being 10 nm. Reference background of 0.1 M sucrose was subtracted from all fluorescence emission spectra of the samples.

pFMS Measurements

About 40 μl of giant unilamellar vesicles suspension, 10× diluted in 0.1 M glucose, were transferred to a pool constructed from silicone lubrication grease (Klüber Lubrication, Munich, Germany) on a standard microscopy slide and covered by a cover slip. Because of the density difference, the vesicles settled at the bottom of the chamber. The samples were examined at room temperature using CFI Plan Apo IR 60×W/NA 1.27 water-immersion objective (Nikon, Tokyo, Japan).

For spectral detection, a narrow-band LCTF (Varispec VIS-10-20 from CRi, Woburn, MA) was placed in front of an EMCCD camera (iXon3 897 from Andor, Belfast, UK), allowing sequential acquisition of images at different wavelengths as reported previously (29). From each acquired λ-stack of images, spectra from every volume-element of the field of view were extracted. After the dark level of the camera was subtracted, the spectra were corrected for transmittance of LCTF, calibrated against a set of reference dyes.

NBD probes were excited by nonpolarized light from a Xe-Hg source (Sutter Lambda LS, Novato, CA) through 460/60 broad-band filter (all band-pass filters and dichroics are BrightLine from Semrock, Rochester, NY). Fluorescence was detected through 550/88 emission filter that limited the wavelength scan with LCTF to the range from 515 to 581 nm. Within this interval, λ-stacks of 23 images with 3-nm step and 0.2-s exposure time were acquired.

To reliably correct for NBD photobleaching, which distorts the measured spectra due to sequential wavelength acquisition (29,34) (Fig. 2 A), we introduced stochastic wavelength sampling. The procedure covered the wavelength–time space more uniformly (Fig. 2 B) and hence allowed numerical decoupling of the time decay and the spectral tail lineshape. The resulting “saw-tooth curves” (Fig. 3 A) still carried all the spectral information with a well-defined bleaching fingerprint, which was effectively modeled by numerical spectral analysis.

Figure 2.

Schematic presentation of photobleaching effect on the spectrum for (A) linear and (B) stochastic wavelength sampling.

Figure 3.

Outline of the applied methodology: (A) fitting of bleaching-distorted experimental spectra (open circles; for explanation of the saw-tooth lineshape, see Materials and Methods and Fig. 2B) by single-component model (red solid line) and reproduction of bleaching-corrected (BC) spectra (light-red solid line); (B) creation of spectrally contrasted images according to optimized peak position (λ_MAX); α indicates the angle between the direction of LCTF-transmitted polarization (e_LCTF) and membrane normal at every point at the perimeter; (C) schematic cartoon of the two suggested conformations of the probes with their dipoles wobbling in the cones with preferential orientations parallel (||) and perpendicular (⊥) to the membrane normal (n); (D) simulation of obtained intensity (I₀) and λ_MAX variations around the perimeter (solid circles) by the two-conformation model (gray lines; see legend in the center); (E) bleaching-distorted and -corrected spectra (solid gray and light-gray lines, respectively) generated as a superposition of spectral components (dotted and dashed lines) using the parameters from the two-conformation model.

Spectral Lineshape Model

To push spectral peak position resolution below experimental λ-step and LCTF bandwidth, we applied spectral fitting (Fig. 3 A), as in particle tracking (35). In principle, fluorescence spectra could be simulated by computationally demanding quantum-mechanical models, based on electron transitions and (an)harmonic vibrational potentials (36). However, because too many parameters are required to describe very broad experimental spectra (37), significant correlations often preclude reliable and efficient inverse problem solving. Therefore, a simplified model was applied based on an asymmetrically skewed Gaussian function and related to the standard log-normal curve (37), which can describe three main spectral characteristics: peak position (λ_MAX), full width at half maximum (w), and asymmetry (a) (Eq. S1). The lineshape nicely fits simple fluorescence spectra such as those of NBD probes (Fig. S3).

Photobleaching was modeled by a mono-exponential decay of simulated intensity (38) with bleaching rate b, counting the time from the beginning of the experiment. The parameters λ_MAX, w, a, and b were obtained by Nelder–Mead minimization (39,40) of the standard reduced χ², whereas intensity (I₀) was determined analytically by Eq. S4.

By the developed algorithm, a typical 512 × 512 image, averaged over 5 × 5 pixels, was analyzed in less than 5 s at a standard quad-core desktop computer. The achieved 1-nm peak position resolution is in agreement with the theory (41) adapted to our application (data not shown).

Analysis of pFMS Experiments

To minimize artifacts due to motion of vesicles during acquisition, we automatically aligned images from each λ-stack using algorithms built in Mathematica (Wolfram Research, Champaign, IL). All images were averaged over 5 × 5 pixels to achieve the desired spectral resolution. Because the scanning range within the region of the broad-band emission filter was smaller than NBD-spectrum full width at half maximum, w and a could not be reliably resolved from FMS experiments. Instead, w and a were determined by fitting the spectrofluorimeter data (Fig. S3) and kept fixed at 78 nm and 0.24, respectively, during the optimization of λ_MAX and b. Images were spectrally contrasted with respect to the optimized λ_MAX and I₀ (Fig. 3 B) according to Eq. S6.

Systematic variations of I₀ and λ_MAX around the vesicle perimeter were analyzed with respect to the polar angle α (depicted in Fig. 3 B) in steps of π/10 radian (solid circles in Fig. 3 D). To decrease the noise, we averaged results of up to six similar vesicles of the same lipid composition and labeled by the same probe. Intensities were normalized to their respective maximal values.

Two-Conformation Model

Intensity variations around the vesicle perimeter originated in linear-polarization transmittance of LCTF (e_LCTF in Fig. 3 B) and nonisotropic distribution of fluorophore dipole orientations. The latter was modeled by a reorienting potential (U) that aligned the wobbling dipoles along its preferential orientation relative to the membrane normal, e.g., as in simulations of electron paramagnetic resonance spectra (42,43). The directional distribution was used to assess the LCTF-transmitted light intensity for every membrane orientation by calculating the expected value of the squared projection of the dipole to the polarization direction of LCTF. A finite reorienting potential (U > 0) yielded sinusoidal-like variations of I₀ around the perimeter (dashed and dotted curves in left panel of Fig. 3 D). Because excitation was not polarized, a potential emission depolarization due to homotransfer (32) was not considered.

To explain λ_MAX variations around vesicle perimeters (Fig. 3 D, right panel), we introduced two coexisting populations of the probes with different positions of spectral maxima with their respective dipoles preferentially oriented parallel (||) and orthogonal (⊥) to the membrane normal (n), as illustrated in Fig. 3 C. The two chosen orientations represent the minimal basis set to approximate the observed distribution of orientations. Both peak positions (${\lambda }_{\text{MAX}}^{||}$ and ${\lambda }_{\text{MAX}}^{\perp }$) and corresponding relative portions of the two populations (p^|| and p^⊥) were determined from angular dependences of I₀ and λ_MAX, obtained from mono-component spectral optimization, by calculating superposition of the signal from both populations (solid and dashed/dotted gray lines in Fig. 3 D). Fluorescence resonance energy transfer between || and ⊥, which could affect the overall spectral lineshape, was neglected considering low probe concentrations (Fig. S1).

Reorienting potential strengths, used in the model, were taken from the literature. For both labels in Ld phase, the potentials were calculated from the two order parameters reported for the appropriate positions on the NBD-labeled chains by molecular dynamics (MD) simulations of the very same probe molecules (28). The comparison yielded U values close to 4 k_BT for both labels. In Lo phase, for which we have not found any applicable study with the two probes, the reported 30–50% relative increase of the order parameters for natural lipids (44) was used, which translated into U = 8 k_BT for the probes. The same potential strength was also used for S phase given that at such high reorienting potentials the results were not very sensitive to U (Fig. S5). In principle, fluorescence anisotropy measurements, yielding wobbling cone angles, could be used as well (45).

To check that the resulting parameters of the two-conformation model still faithfully represented the original data, we took the obtained ${\lambda }_{\text{MAX}}^{||,\perp }$ and appropriate relative intensities, calculated from p^||,⊥ at any given α, into the multicomponent spectral model to generate the composite spectra (solid gray line in Fig. 3 E, as superposition of the dashed and dotted spectra from the two probe populations) and compared them with the original mono-component fits and with the experimental spectra (Fig. S7). Other spectral parameters (w, a, and b) were kept the same as with the originally optimized spectra.

See Supporting Material for a complete derivation and detailed explanation of calculation procedures.

Results

pFMS

To characterize the coexisting conformations of the two probes, we labeled vesicles of three lipid compositions with either C6- or C12-NBD-PC probe and investigated them by pFMS. Given the environmental sensitivity of NBD, the corresponding λ_MAX-coded images (Fig. 4, A–F) show significant distinctions in brightness and color patterns, which were analyzed in terms of normalized intensity and peak position variations around the perimeter of the liposomes (Fig. 4, G and H). Absolute intensities, often studied because of NBD quantum yield sensitivity, were not reliable enough to be considered in our case because of nonuniformity of the illumination pattern across the field of view, possible deviations in photobleaching prior to measurements, and potentially different lamellarity of liposomes.

Figure 4.

(A–F) Images of representative vesicles for different lipid-probe mixtures, color coded by λ_MAX; all scale bars correspond to 10 μm and represent a color legend for (G) normalized intensity (I₀) and (H) λ_MAX around the perimeter of the vesicles. Each curve represents an average of up to six similar liposomes of the same lipid composition.

Intensity variations around vesicle perimeter are generally observed when fluorophores are either excited by polarized light, as exploited in fluorescence polarization microscopy (30) and fluorophore photoselection (3), or their emission is detected through a polarization analyzer, in our case implemented by Lyot filter-based LCTF. In addition, a nonisotropic distribution of dipoles relative to the membrane normal is required because totally random polarizations would otherwise result in uniform intensity attenuation around the perimeter. Considering that LCTF transmitted only the light that was polarized in the horizontal plane (depicted as e_LCTF in Fig. 3 B) and suppressed the vertical polarization, the maximal fluorescence intensity at the vesicles’ poles for both probes in DOPC and DPPC (Fig. 4) revealed that the dipoles were preferentially oriented in the membrane plane, that is, perpendicular to the membrane normal (⊥). On the contrary, polarized detection of both probes in cholesterol-rich membranes showed the highest intensity at the equator, meaning that their dipoles were oriented roughly parallel to the membrane normal (||). It is noteworthy that C6-NBD-PC in DPPC and C12-NBD-PC in DPPC+chol (Fig. 4, B and F, respectively) showed only weak intensity variations around the perimeter, which could be mistakenly attributed to nearly isotropic motion or a tilted preferential dipole orientation. However, together with significant variations in λ_MAX, revealed by spectral detection, they could only be consistently explained by a coexistence of probe conformations with different dipole orientations, as discussed later.

The peak position analysis (Fig. 4 H) revealed that, on average, λ_MAX blue-shifted for approximately 4.5 and 3 nm when comparing either of the probes in Ld and S, or in S and Lo phase, respectively, which agreed with the data measured at spectrofluorimeter on the whole liposome suspension (Fig. S2). In addition, a 1- to 2-nm difference between the two probes in the same environment was consistently observed (Fig. 4 H). And most important, both probes exhibited significant λ_MAX variations around the perimeter, most notably in Lo phase (Fig. 4 H). The differences in λ_MAX within the same membrane, observed at different polarizations relative to the membrane normal (α), were ascribed to coexisting conformations of the probe molecules. The latter caused fluorophores to experience different local polarities or were subjected to different H-bonding, and exhibited different preferential dipole orientations (see above). Consequently, their relative intensities, polarization filtered due to LCTF, varied with respect to the orientation of membrane normal, resulting in the observed λ_MAX variations of their spectral superposition. Note that both polarized and spectral detection of pFMS were needed to unambiguously characterize probe conformation coexistence.

Two-Conformation Model

To evaluate the hypothesis, we built a numerical model consisting of two sets of dipoles with different λ_MAX, wobbling in reorienting potentials with orthogonal preferential orientations (i.e., || and ⊥ with respect to membrane normal). Relative portions of the two populations and their peak position wavelengths were determined from the pFMS data for each sample, as outlined in Materials and Methods and described in detail in Supporting Material. The results nicely reproduced both intensity and λ_MAX variations around the perimeter of the vesicles (Fig. 5). The obtained parameters, graphically presented in Fig. 6 and listed in Table S1, confirmed that high concentrations of cholesterol had the strongest effect on the probing molecules: The portion of the conformation || was the greatest, as was the reported polarity difference between the two populations, reflected in the difference between ${\lambda }_{\text{MAX}}^{||,\perp }$. The effect was especially pronounced for C6-NBD-PC, which seemed to adopt the conformation || about twice as readily as its C12 analog (Table S1).

Figure 5.

Simulations of I₀ and λ_MAX angular dependences by the two-conformation model (solid lines; the contributions of || and ⊥ components are shown with dashed and dotted lines, respectively), plotted against the data points from mono-component spectral optimization (solid circles) for all six samples (A–F).

Figure 6.

Graphical presentation of the results from the two-conformation model, yielding characteristic peak positions of the two populations of the probe (${\lambda }_{\text{MAX}}^{||}$ and ${\lambda }_{\text{MAX}}^{\perp }$, presented as vertical positions of the bubbles) and their respective portions (p^|| and p^⊥, proportional to bubble area). Numerical values and their uncertainties are listed in Table S1.

To check the fidelity of the two-conformation model, we reconstructed several spectra from the obtained model parameters (${\lambda }_{\text{MAX}}^{||,\perp }$ and p^||,⊥), which were found to completely reproduce the mono-component fits and to nicely represent the experimental data (Fig. S7). Note that the two probe conformations were required to consistently explain systematic λ_MAX variations around vesicle perimeter, revealed by pFMS, and not to fit individual spectra, which were themselves well described by a single spectral component.

Discussion

NBD-PC Snorkeling

Several fluorescence methods (21,26,27), as well as NMR (26) and MD simulations (28) revealed that in Ld phase the polar NBD group attached to one of the acyl chains preferentially locates in the region of glycerol backbones rather than in the membrane core (schematically presented in Fig. 3 C). Consequently, the reported spectrum peak position (541 nm) for C12-NBD-PC in egg yolk PC (21) corresponds to an environment with dielectric constant (ε) around 38 (Fig. S8 A), characteristic for the lipid–water interface (46). A model for ε variations in a PC membrane (21,46,47) (Fig. S8 B) translates this ε value into an average NBD group position of z = 1.98 ± 0.05 nm, measured from the membrane core (Fig. S8 C), which agrees with the value obtained from nonradiative rate constants (1.98 nm) (21). Similar relative positions reported from other membrane systems (26–28) prove that λ_MAX of NBD-PCs reliably characterizes molecular conformations.

Our FMS experiments on DOPC vesicles showed average λ_MAX of 537 and 538.5 nm for C6- and C12-NBD-PC, respectively (Fig. 4 H), confirming the snorkeling effect in the Ld environment. An additional 1- to 2-nm shift between the two probes revealed that the fluorophore attached to the longer tail experienced a slightly more polar environment, indicating its higher vertical location in the membrane. A similar spectral shift, although uncommented, has already been reported (27). According to the ε(z) model, 1.5 nm higher λ_MAX for C12-NBD-PC translates into a 0.27 ± 0.03-nm upward drift, again in a qualitative agreement with the study exploiting NMR and fluorescence quenching by a water-soluble agent (26) as well as with the MD simulations (28). On the contrary, parallax position determination through fluorophore quenching by spin-labeled PCs did not see any difference between the positions of the two analogs (27), probably due to very broad vertical distributions of nitroxide moieties, spanning the entire membrane leaflet (48).

Conformation Coexistence

The same MD study revealed another, much more significant difference between the two probes: The dipole of C12-NBD-PC lies roughly in the membrane plane, whereas in the case of C6-NBD-PC, it spends a considerable amount of time roughly parallel to the membrane normal because of jumps between conformations (28), directly corroborating our two-conformation model. As a consequence of different orientations of NBD rings, the fluorophore of C12-NBD-PC, on average lying flat in the membrane plane, forms H-bonds with water molecules more readily than C6-NBD-PC (28), which elucidates the above-mentioned difference in water-soluble quencher accessibility (26) and explains the difference in λ_MAX between the two spectral components of distinct dipole orientations.

The two conformations experiencing slightly different polarities might also underlie the observed bicomponent fluorescence lifetimes of the two probes in DOPC (27), with their corresponding relative portions similar as obtained herein from pFMS. In addition, the probe with the longer average lifetime, i.e., C6-NBD-PC (27), should bleach faster according to the inverse relationship between fluorescence lifetime and bleaching constant (38), which was indeed observed for our DOPC as well as DPPC vesicles (data not shown). Even if the absolute values of our bleaching rates and relative portions of the two probe conformations might have been influenced by some above-mentioned inherent experimental limitations and model simplifications, e.g., reorientation potential implementation and neglected variations in quantum yield (19,21), it is nevertheless clear that C6-NBD-PC exhibits a richer repertoire of motional patterns, reflected in higher portions of the conformation || (cf. Table S1 and Fig. 6).

Lipid Phase Sensitivity

To our knowledge, the behavior of both NBD-PC probes in the other two main membrane phases (i.e., Lo and S) has only rarely been studied systematically (24,25). Our measurements show that the observed values of λ_MAX—and thus the apparent local polarity—decreased in the order Ld > S > Lo for both labels and for either probe conformation (Fig. 6), which is in direct agreement with Laurdan measurements (49) as well as with attenuated total reflectance Fourier transform infrared spectroscopy data on interactions between carbonyl groups of lipids and water molecules (50,51), both methods probing local polarity approximately at the same depth in the membrane as the two snorkeling NBD-PCs.

The comparison to the results of the spin-label studies that, in contrast, report increased local polarity in the outer region of cholesterol-rich Lo membranes (Lo > Ld down to 7th–9th C-atom of alkyl chains) and decreased polarity in the membrane core (Ld > Lo) (52,53) is slightly more perplexing. Similar to NBD probes, certain types of spin labels also exhibit the snorkeling effect (54,55) or they induce notable packing perturbations to the neighboring lipids (55,56). However, the fluorescence and spin probe molecules might respond differently to local interactions and consequently undertake diverse conformations because of differences in polarity between NBD and oxazolidine groups or differences in their positioning at the end and in the middle on the alkyl chains, affecting hydrophobic anchoring of the reporting groups. Therefore, considerable care with data interpretation is required.

From the observed λ_MAX values, characteristic for environments with ε well above 10 (Fig. S8 A), we can only suggest that in both probe conformations NBD probably locates on average somewhere close to the headgroup region also in S and Lo phases, as ε of 1–2 and accordingly lower peak positions would be anticipated in the membrane core (46), similar to those shown by NBD-labeled cholesterol (25). Within this interpretation, the λ_MAX difference between || and ⊥ would be primarily due to the change in dipole orientation, as discussed before. However, our experiments do not exclude the possibility that NBD-PC molecules of the lower polarity-experiencing population (||) extend their chains deep into the membrane, as has been suggested before (24,25). In this case, the observed ${\lambda }_{\text{MAX}}^{||}$ require an additional red-shifting mechanism, e.g., a few water molecules accompanying the fluorophore in the hydrophobic region (25).

The portions of the alternative || conformation (Fig. 6) show that the effect of the surrounding lipids on probe conformations increased in the order Ld < S < Lo, opposite to the average area per lipid (57,58). Increased lateral density of surrounding molecules obviously forced NBD rings into the upright conformation more often, which required less space and therefore allowed tighter packing, and was most notably induced by planar, rigid structures of cholesterol. This is more pronounced for C6-NBD-PC, as expected from the reduced flexibility of its shorter chain.

Spectral Background due to Motional Averaging

Detection of probe conformations, discussed herein, requires the existence of molecular states with preferential dipole orientations that are not averaged out because of molecular rotational diffusion within the excited state lifetime. For the two probes studied, it has indeed been shown by fluorescence anisotropy decay measurements and supported by MD simulations (28) that in the most dynamic Ld phase, the description of molecular motion involved a substantial component with rotational correlation times on the order of NBD fluorescence lifetime (27). In addition, several studies in Ld phase have noted small but significant orientational restrictions, indicating a wobbling-in-cone type of motion (27,28) and thus supporting the choice of our model. Its application to Lo and S phases was therefore even more justified because rotational correlation times are at least an order of magnitude longer than in Ld membranes (59), whereas the prolongation of fluorescence lifetimes because of lower polarity is expected to be less than 20% (19).

In Ld phase, however, a small component with rotational correlation times below NBD fluorescence lifetime (27) would imply an averaged spectral background superimposed onto the two resolvable components. In this case, the latter would require more separated ${\lambda }_{\text{MAX}}^{||,\perp }$ values to reproduce the observed peak position variations. Nevertheless, because the observed difference in λ_MAX between the two conformations in Lo phase was just slightly smaller than in S (Table S1 and Fig. 6), for which such a fast isotropic movement is not expected, this background contribution should not affect our characterization of probe conformations significantly.

Applicability of pFMS to Explore Membrane Heterogeneities

Even if exact molecular processes are not yet completely understood, the ability of NBD-based labels to probe their local environment was demonstrated in a new way. From this point of view, snorkeling of the alkyl chain-bound fluorophore or tilting of the dipole should not necessarily be considered as misbehavior or complication. Instead, they can be regarded as an additional degree of sensitivity, offering ample opportunities to be exploited through minor spectral shifts, as applied here, changes in fluorescence lifetimes, correlation decay rates, or any other feature of fluorescence light that carries information about local molecular environment. We show that imaging the differences of such a variable detected at various polarizations with respect to membrane normal can reveal probe conformation coexistence within the resolution-limited volume element of the sample. A careful characterization of such heterogeneities could bring new insights into local (supra)molecular organization of biomembranes, e.g., lipid rafts and nanodomains.

For future FMS experiments, spectral fitting with peak position resolution down to 1 nm offers an additional opportunity to exploit numerous probes with desired biochemical characteristics that have been so far considered as not sensitive enough. The bleaching correction algorithm together with stochastic sampling extend the applicability of experimental systems with sequential wavelength acquisition to photosensitive dyes, which has not been reliable before.