Conformational Distributions and Hydrogen Bonding in Gel and Frozen Lipid Bilayers: A High Frequency Spin-Label ESR Study

Abstract

The ESR parameters of PC spin labels in frozen membranes do not simply represent the membrane polarity or water penetration profile. Instead, they show a distribution between hydrogen-bonded (HB) and non-hydrogen bonded (non-HB) states, which is affected by a number of factors in the membrane composition. Similar to the exclusion of solutes from crystallizing solvents, the pure bulk gel phase excludes nitroxides, forcing acyl chains to take bent conformations. In these conformations the nitroxide is hydrogen-bonded. Furthermore, upon gradual cooling in the supercooled gel PC labels undergo slow lateral aggregation resulting in a broad background signal. However, if the sample is instantly frozen, this background is replaced by the HB component. In membranes with cholesterol the observed HB/ non-HB ratio can best be described by a partition-like equilibrium between nitroxides located in defects of lipid structure within the hydrophobic core and those close to the membrane surface.

Introduction

Lipid spin labels containing nitroxide groups at different positions in the fatty acid chain, such as 1-palmitoyl-2-stearoyl-(n-doxyl)-sn-glycero-3-phosphocholines (n-PC spin labels) are a useful and proven tool in lipid research. Through a vast number of studies, they have provided crucial insights into the structure of model and biological membranes, reported on the membrane fluidity, polarity, phase state and presence of microscopic domains¹, accessibility of different depth positions in the lipid bilayer for oxygen and other polar and non-polar paramagnetic compounds^2,3,4. PC spin labels have also provided valuable information on protein/lipid interactions and boundary lipids^5,6.

It is generally accepted, that, unlike bulky fluorescent labels^7,8, nitroxides are well incorporated into fluid lipid bilayers⁹ and not excluded from them. However, it has been shown by NMR that although the most probable location of the nitroxide group for 5-, 10- and 16- PC spin labels in the fluid POPC membrane corresponds to the fully extended conformation, the distribution is relatively broad and other conformations should also be present¹⁰. Bent conformations were previously found for doxylstearic acids in monomolecular films¹¹, water/hydrocarbon emulsion particles¹² and micellar systems¹³. In fluid membranes the fluidity, polarity and accessibility parameters reported by ESR using PC spin labels and n-doxylstearic acids are, in general, in good agreement with the extended conformation of the spin labeled alkyl chain and change monotonically with an increase in n^3,14, although there are indications that the spin label groups on the stearates are located nearer to the membrane exterior than the analogous positions of the unlabeled phospholipid chains¹⁵. However, in the gel phase, which is characterized by denser chain packing and higher order, the behavior and location of nitroxides may differ. Although most cell membranes in vivo exist in the fluid L_α phase, the gel phase of lipid bilayers has biological interest for specialized membranes such as stratum corneum^16,17 or lens¹⁸.

ESR spectroscopy and High Field ESR, in particular, have been used at low temperatures to determine polarity profiles in membranes of various compositions: Earle et al.¹⁹ at 250GHz, Kurad et al., Subczynski et al.^20,21 at 94 GHz. In these studies the polarity of the local environment was monitored by g- tensor and hyperfine splitting values. However, some observations on the behavior of PC spin labels and features of their ESR spectra are yet to be completely understood. For example, High Field ESR, which is very sensitive to local polarity and proticity²², in DMPC/cholesterol membranes containing n-PC labels at 94 GHz shows the presence of two components with different g_xx values^21,14. Although these components cannot be completely resolved at W-band (94 GHz), the more polar component (lower g_xx) is clearly detectable for n-values higher than 7 on the background of the component with a higher g_xx value. The superior g-factor resolution, which is provided by a 170/240 GHz quasioptical spectrometer available at ACERT was expected to completely resolve and identify these components. Also, while all PC spin labels in DPPC containing gramicidin A were characterized at low temperatures by well resolved High Field rigid limit spectra¹⁹, 7-, 10- and 12-PC in pure DPPC show a singlet-like background signal as the major spectral component. This background signal, however, is much less pronounced in cholesterol-containing membranes, even with low cholesterol content²¹. This difference could be attributed to a specific behavior of PC spin labels in the L′_β phase of pure phospholipids. This densely packed phase (tilted gel phase) could behave as a two-dimensional analog for a bulk fluid phase which upon freezing and crystallization excludes dissolved spin labels into high concentration areas with strong spin-spin interactions. Addition of a third component like GA¹⁹ or cholesterol²¹ for this case can perturb the lipid chain packing and provide more defects to incorporate and spatially separate the nitroxide rings.

The purpose of the current study is to better understand how spin-labeled molecules report on the lipid polymorphic phases in particular the gel phase, to explain previously observed aggregation patterns and their connection to the membrane phase state and composition. This work illustrates superior resolution of HF ESR, its advantages in membrane studies and benefits of simultaneous application of several ESR techniques; it also points at conformational flexibility of PC spin labels, which should be taken into account in all studies where they are used.

MATERIALS AND METHODS

Materials

Spin-labeled phosphatidylcholines, n-PCSL (1-acyl-2-[n-(4,4-dimethyloxazolidine-N-oxyl)stearoyl]-sn-glycero-3-phosphocholine) with n = 5, 7, 10, 12, 14 or 16, were purchased from Avanti Polar Lipids and, if necessary, additionally purified using preparative TLC. Synthetic phosphatidylcholines, DMPC (1,2-dimyristoyl-sn-glycero-3-phosphocholine), DPPC (1,2-palmitoyl-sn-glycero-3-phosphocholine), egg yolk PC were from Avanti Polar Lipids (Alabaster, AL). 3-carboxy-2,2,5,5 tetramethylpyrrolidine - 1-oxyl was from Acros Organics. Cholesterol, organic solvents and other chemicals including TEMPO nitroxides were from Sigma-Aldrich.

Preparation of membrane samples

Spin-labeled phosphatidylcholines were incorporated in bilayer membranes at a relative concentration of 0.5 mol% by drying down the lipid solutions in chloroform/methanol and then suspending the dry lipid in water above the chain melt temperature for ~ 20 min. The suspension was spun down in an Eppendorf tube, and supernatant removed. For High Field ESR measurements about 8 μL of the lipid pellet containing ~ 2 mg of the lipid was placed between a flat quartz coverslip and an etched quartz coverslip with a circular concavity in the middle. The diameter of the concavity to hold the sample is about 10 mm, the diameter of the coverslips is 18 mm. A thin layer of vacuum grease (Dow Corning) was applied on the edge to create a seal between the two coverslips. Such a sample holder was discussed in detail elsewhere^23,24.

For X-band CW saturation measurements the lipid pellets were re-suspended in ~ 40 μl of water and transferred into 50 μl, 0.7 mm i.d. glass capillaries and spun down for 10 min at 10000g using microhematocrit centrifuge. The excess water was removed; sample sizes were trimmed to 5 mm length to avoid inhomogeneities in H₁ and H_m fields²⁵. To remove oxygen from the lipid and ensure anaerobic conditions in final ESR samples, the samples in the capillaries were frozen in liquid nitrogen, briefly evacuated using an oil pump and thawed in the atmosphere of high purity argon. After repeating this procedure three times, the capillaries were flame-sealed under vacuum with the sample part submerged in liquid nitrogen.

X-band ESR spectroscopy

ESR spectra were obtained on a Bruker EMX spectrometer (Bruker, Billerica, MA) at a frequency of 9.55 GHz under standard conditions. The field sweeps were calibrated with a Bruker ER 035 Gaussmeter. The microwave frequency was monitored with a frequency counter. The temperature was stabilized and monitored by a nitrogen gas flow temperature unit (±0.2°C accuracy) or, for 77K measurements, using a finger dewar filled with liquid nitrogen.

For CW saturation experiments sample capillaries were positioned along the symmetry axis of the standard 4 mm quartz EPR sample tube that contained dodecane for thermal stability. The root-mean-square microwave magnetic field 〈H₁²〉^1/2 at the sample was measured as described by^26,25 and corrections were made for the cavity Q value as described in the same reference. The ESR spectra were recorded at a modulation frequency of 100 kHz, the modulation field measured at the sample was 0.32 G p-p.

The dependence on the microwave field for the amplitude of the central component of the EPR spectra were measured and fitted to the equation:

$$I=\frac{I_0·H_1}{{\left(1+P·H_1^2\right)}^{3/2}}$$ (1)

from which the saturation parameter P was obtained. Here I₀ is the I – H₀ slope at low power, where the dependence can be considered linear.

The parameter Δ(1/P) = 1/P − 1/P₀, where P₀ is the saturation parameter in the absence of paramagnetic relaxant, was used as an experimental parameter of relaxation enhancement.

For relaxation enhancement studies by paramagnetic ions present in the water phase of the membrane we used nickel perchlorate (10 mM), which is shown to be the strongest relaxant among nickel salts for PC spin labels in DMPC and EYL membranes²⁷.

High Field ESR

High-field ESR spectra were recorded on a 170/240 GHz (6–9 T) ESR spectrometer, which was designed and built at ACERT at Cornell University²⁸. The data collection routine is based on LabView virtual instrument drivers written at ACERT. A Lakeshore 340 temperature controller was used to stabilize the temperature at 80–82K. It corresponds to the rigid-limit conditions, as evidenced by experiments showing no change in the ESR lineshape between this temperature and 100–110K. MATLAB procedures based on the EasySpin software package²⁹ were used for simulations of High Field ESR spectra at the rigid-limit condition.

Sample Freezing Procedures for High Field ESR

Typically, a low temperature High Field ESR experiment uses a helium or nitrogen cryostat. For our experimental setup, it usually takes ~ 2+ hours to reach the target temperature, with an initial drop from ambient temperature to ~ 260K taking ~ 30 min. During this cooling process, the lipid bilayer, as discussed below, may pass through a number of slow forming metastable phases. Lateral distribution of lipids and spin labels in these phases may differ from the initial gel phases. Alternatively, to prevent any slow lateral redistribution of lipids during this slow cooling procedure, and to trap the initial distribution, a quick freeze-quench was performed by submerging the whole Fabry-Perot resonator assembly into liquid nitrogen. The cooling rate obtained in this procedure was monitored in test experiments by a temperature-dependent resistance Wheatstone bridge with a 15 Ohm thin test tungsten wire placed inside the etched cavity of a standard sample holder for HFHF ESR. The resistance vs. time curves were recorded on the oscilloscope using a Stanford Instruments lock-in amplifier. Analysis of these curves shows an initial slope of ~ 140 K/s; this corresponds to reaching 260K within 0.2s. The same immersion procedure with liquid helium instead of liquid nitrogen gives a substantially slower cooling rate of ~ 40K/s. After freezing in liquid nitrogen the Fabry-Perot resonator assembly was quickly inserted into the bore of the magnet, which was pre-cooled by the nitrogen flow cryostat to ~ 120K. The cryostat was further used to stabilize the temperature at the target temperature.

Before the freeze-quench procedure, samples of both spin-labeled DMPC and DPPC were kept at room temperature, 292–295K, for at least one hour. This temperature is below the main transition for DMPC (297K), which stays in the P_β rippled gel phase at these conditions. On the other hand DPPC at this temperature exists in the L′_β phase, which becomes metastable and slowly transforms to the subgel phase only below 280K³⁰.

RESULTS AND DISCUSSION

9 GHz ESR

Table.1 shows the values of the ¹⁴N hyperfine constant (2A_zz) determined by X-band ESR as the outer splitting of the rigid limit spectra at 77K. As seen from the data, DMPC/Cholesterol (cf. Fig. S1A) and DPPC/Cholesterol (cf. Fig. S1C) membranes, consistent with previous observations³, show abrupt drop in the polarity, as monitored by the 2A_zz value, between n-PC positions 7 and 10. However, for DMPC, in the absence of cholesterol, the 2A_zz value shows little trend throughout the PC spin labels series and remains ~ 69.5G (cf. Fig. 1), which correspond to a relatively polar environment, more polar than ethanol (see below). DPPC in the absence of cholesterol (cf. Fig. S1B), in general, shows a profile similar to a cholesterol-containing membrane, though the spectra, especially for positions 7–12 show signs of a broad singlet-like component (Suppl Fig. 1C).

Fig. 1.

PC spin labels in DMPC at 77K. The apparent value of 2A_zz does not change in the PC series.

The dependence of relaxation enhancement Δ(1/P) by 10 mM of Ni(ClO₄)₂ on the spin-labeling position for the gel and liquid crystal phase of DMPC are given in Fig. 2. In the liquid crystal phase (T=39°C) the interaction with the paramagnetic relaxant decreases with increasing n, consistent with an increase in the average immersion depth of the spin label moiety in the membrane. On the contrary, the Δ(1/P) profile in the gel phase (P_β, T=19°C) is almost flat, (with a minor spike at position 10).

Fig. 2.

Relaxation enhancement from 10 mM of Ni(ClO₄)₂ introduced into the water phase of lipid DMPC dispersions as a function of n at the P_β (19°C) and L_α (39°C). An almost flat accessibility profile in the gel might indicate similar average membrane immersion depth for the nitroxide moieties of all studied PC labels.

High field/ High frequency ESR

High frequency ESR and PC spin labels were successfully used to study polarity profiles in phospholipid membranes^19,21,14. For example, in a detailed ESR study at 250GHz 5,7,10,12,14 and 16 PC were studied in DPPC and DPPC/gramicidin systems¹⁹. It was found that in pure DPPC most spins are strongly aggregated and the spectrum consists mostly (especially for 7–12 PCs) of a singlet-like signal. Although the g-factor values for the resolved rigid-limit component clearly show increasing immersion of the nitroxide into the hydrophobic core of the membrane with increasing n, it is difficult to obtain any information on the location of aggregates which manifest themselves in the broad component. Several W-band (94GHz) studies by Marsh and coworkers^21,31,14 on DMPC/Cholesterol membranes utilized all n-PC spin labels in the 4–16 range (except 15-PC). They allowed, based on the g_xx values, for obtaining detailed polarity profiles in these systems.

Resolution of different hydrogen-bond states by 240GHz HFHF ESR in bulk organic solvents

It has been well established that the g-tensor and hyperfine components of nitroxide radicals are very sensitive to the local environment, to polarity and proticity in particular. In general, changing the local environment of the nitroxide moiety from water (polar) to hydrocarbon (non-polar) causes an increase in the g-tensor components and a concomitant decrease in the values of the components of the hyperfine tensor. This effect is most pronounced for the tensor components g_xx and A_zz. For example, for the PDT (4-Oxo-,2,2,6,6-tetramethylpiperidine-d₁₆-1-oxyl) probe in the polar glycerol/water solvent vs. toluene, g_xx changed by ~0.001³², while A_zz decreases. In the 1980’s, Lebedev and co-workers extensively studied the effects of solvent polarity and demonstrated a high degree of correlation between the values of A_zz and g_xx³³. However, the separation of the polarity and proticity effects could complicate the analysis of the ESR spectra. Proticity refers to the propensity to donate hydrogen bonds; whereas aprotic refers to solvents which cannot donate a hydrogen bond. At relatively low frequencies, up to 95GHz, the separation of hydrogen-bonded vs. non hydrogen-bonded states of the nitroxide often relied upon different g versus A plots, discovered for these two states³⁴. However, as shown in a recent study using TEMPO, if the correlations are indeed different for TEMPO in protic and aprotic solvents, the difference is rather small³⁵.

On the other hand, superior g-factor resolution of very high-field ESR allows for observation of two resolved spectral components corresponding to two (Smirnova et al.³⁶ at 130GHz) or possibly more (Bordignon et al.³⁷ at 95, 275 and 360GHz ) states of hydrogen-bonding. Two hydrogen bonding states coexisting in frozen deuterated alcohols were previously demonstrated for perdeuterated TEMPONE by X-band ESR³⁸.

Tables 2 and 3 show the values of A_zz and a_iso hyperfine splitting and the g_xx component of the g-tensor for several nitroxide radicals.

Table 2.

Values of isotropic hyperfine splitting constant a_iso and 2A_zz determined at X-band at 295 and 77K respectively for several solvents. These A_zz values were also used to obtain the best fits for the corresponding 240 GHz rigid limit spectra.

Solvent, dielectric constant at room temperature68	Oxo-TEMPO	TEMPO	4-Hydroxy TEMPO	3-carboxy-2,2,5,5 tetramethylpyrrolidine-1-oxyl
Isopentane, 1.8	14.27/**	15.29/68.2	15.15/**	13.92/**
MCH, 2.02	14.315/**	15.32/68.2	15.20/**	13.96/**
Toluene, 2.4	14.49/67.6	15.53/69.3	15.43/69.1	14.22/66.7
DBPh, 6.4	14.61/67.6	15.64/69.4	15.55/69.2	14.37/66.9
Ethanol, 24.3	15.06/70.0	16.21/72.9	16.05/71.0	15.02/70.1
TFE, 26.14	15.63/73.5	17.00/79.7	16.73/77.3	15.77/74.4
Water/Glycerol, 80.4	15.97/73.8	17.19/76.5	16.96/75.5	16.12/73.2

^**2A_zz value at 77K cannot be reliably determined due to the presence of a singlet-like background.

Table 3.

g_xx component of the g-tensor determined by 240GHz ESR at 80–85K in several glass forming solvents. A common value of 2.00233 was assigned as g_zz for all spin labels and g_xx value was accurately determined relative to this g_zz value from the corresponding spectral splitting^47,46. If two components are present in the spectrum, the component with higher fraction is marked bold. “Sh” denotes the presence of a high/low field component, which manifests itself not as a distinct peak but as a shoulder on the main component.

Solvent, dielectric constant	Oxo-TEMPO	TEMPO	4-Hydroxy TEMPO	3-carboxy-2,2,5,5 tetramethylpyrrolidine-1-oxyl
Toluene	2.009438	****	*****	****
DBPH	2.009445	2.010103	2.010129	2.009230
Ethanol	2.009485/2.008838	sh/2.009411	2.010040/2.009456	2.009250/2.008533
TFE	2.008793/sh	2.008669	2.009460/2.009089	2.008487/sh
Water/Glycerol	2.008784/~2.00850	2.008805	2.009460/2.008951	2.008532/sh

As one sees from Fig. 3 and Fig. 3S supplement, whereas at X-band one sees a continuous increase in the ¹⁴N hyperfine splitting with change of the local environment from non-polar/aprotic to polar/protic, the 240 GHz ESR shows three distinct values of the g_xx parameter. Although the presence/ratio of these components strongly depends on the polarity-proticity of the solvent, there is little variation in the g_xx measured for each such component. For all four spin labels studied three distinct components could be detected: (1) “non-polar”, as in toluene, DBPh or the minor component in alcohols, (2) “polar”, the major component for ethanol and major or minor component in TFE and water/glycerol, depending on the nitroxide used, and (3) “very polar” component observable in TFE and water/glycerol. Although components 2 and 3 cannot be separated at 240GHz as two distinct peaks, their presence is quite obvious (compare Fig. 3 and Fig. 3S for TEMPOL and TEMPO). We assign these components to different hydrogen-bonding states of nitroxide radicals and speculate that state 2 corresponds to a single hydrogen bond, while state 3 is double-bonded. Existence of multiple hydrogen bonding to a nitroxide has been predicted theoretically^39,40,41 and later suggested as an explanation for complex ESR lineshapes observed in spin labeled proteins³⁷. Interestingly, the g_xx value of the non-hydrogen bonded component for all four spin labels studied shows little dependence on the polarity of the frozen glass-forming solvent (Table 3). This contrasts with the conspicuous dependence of the hyperfine splitting for non-protic solvents at room temperature on the dielectric constant ε(cf. Table 2). For protic solvents at 77K the resolution at X-band is insufficient for resolving hydrogen-bonded and non-hydrogen bonded spectra. However, the hyperfine splitting, a_iso at room temperature for aprotic solvents, show a clear dependence on ε Also, theoretical predictions for the g-factor⁴², as well as some room temperature measurements for g_iso point to a higher g-factor for lower ε^34,35.

Fig 3.

240 GHz spectra of 4-oxo-TEMPO (TEMPONE) in a series of glass-forming solvents at 80–85K. Black dotted line is a two-component rigid simulation of the ethanol spectrum. g values 2.009450, 2.008830 and 2.008500 are noted “1”, “2” and “3” respectively.

It is well known that at room temperature the lifetime of a hydrogen bond is very short and measured on the picosecond timescale⁴³. Existence of two distinct spectral states in alcohols at liquid nitrogen temperatures shows that at these conditions the lifetime of the hydrogen bond is long on the time scale of ESR at this frequency and should be longer than ~100 ns.

However, if the ethanol solution containing TEMPONE warms up above ~ 115K, some signs of exchange between the components appear: they broaden and start to move towards each other. Eventually, at ~ 140K, the low field component of the doublet disappears (Fig. 4). High field ESR can reliably attribute this effect to exchange between the hydrogen-bonded and non-bonded states of the nitroxide because of shortening of the hydrogen bond lifetime rather than to motional effects in the ethanol glass (melting point 159K). Indeed, as seen in Fig. 4, the broadening affects only the X-region of the spectrum and leaves the triplet in the Z-region intact. To attribute this spectral change to molecular motion one would have to assume extremely high-axial Z-rotation, i.e. motion about the axis perpendicular to the nitroxide ring. This is highly unlikely for a relatively spherical molecule like TEMPONE in an isotropic glassy medium⁴⁴.

Fig. 4.

Disappearance of the two-component feature in the X-region of TEMPONE/ethanol spectrum above 120K. High field ESR can reliably attribute this effect to the shortening of the hydrogen bond lifetime.

At this point we would like to call attention to the superior orientational resolution of High Field ESR^45,46: this conclusion could not be drawn from ESR at low frequencies.

PC spin labels in DMPC and DPPC in the presence of cholesterol

The polarity profiles previously obtained by HFHF ESR¹⁴ appeared to be almost identical to the polarity profiles previously obtained by rigid limit X-band ESR from the hyperfine splitting³ values. However, a close inspection of the spectra of ²¹ shows two components, which correspond to hydrogen-bonded and non-hydrogen-bonded states of the nitroxide radical states 2 and 1 (as marked in Fig. 3). These two components are discernible not only for the area of abrupt “polarity change” (PC 5–10), but also for PC14-16. With the higher resolution of 240GHz ESR we expected to better separate and identify these components. Fig. 5A,B show the spectra for DMPC and DPPC in the presence of cholesterol. The ESR spectra in the two different lipids are very similar, nearly identical. There are two components discernible for n≥7, which are completely resolved for n≥10, with g_xx= 2.009435 and g_xx= 2.008820 (with g_zz taken as 2.00233, see ⁴⁷). The two g values are nearly the same for all n. For 5-PC only the polar component is present. While at 94GHz and even at 170GHz these components could not be completely separated, see Kurad et al.²¹ and our 170GHz data (not shown), 240GHz appears as a good frequency to observe them as separate discernible peaks.

Fig. 5.

Rigid-limit 240 GHz ESR spectra of n-PC spin labels in DMPC and in DPPC containing 30% mol cholesterol. T=80–82K. The samples are gradually frozen in the flow of gaseous nitrogen. Red lines show simulations of these spectra using two components with the g_xx = 2.009435±0.000005 and 2.008820±0.000005 for all spectra. g_yy and g_zz are taken 2.006100 and 2.00233 for both components.

It has previously been assumed that polarity profile reported by spin labels for fluid and frozen membranes containing cholesterol in general follows the membrane water penetration profile^48,40,49. However, the exact features of the ESR spectrum may be defined by interplay of this water penetration and the flexibility of spin-labeled lipid chains, see below. While the short acyl tether for the nitroxide of 5PC is always located in the area with higher water content and shows only hydrogen-bonded component, longer acyl tethers can reach a less polar area with lower water content. Yet, even for 14 and16PC there is a substantial fraction of hydrogen-bonded component. Moreover, the fraction of the hydrogen-bonded component does not decrease monotonically with an increase in the n number. As seen in Fig. 5A, B and Table 4, this fraction experiences a dramatic drop between 7 and 10PC and then gradually increases further for positions 12,14 and 16. Based on these observations and what is currently known about flexibility of spin-labeled lipids¹⁰, we believe that the hydrogen-bonded component mainly corresponds to bent conformations of the spin-labeled lipid chain, rather than penetration of water molecules to the middle of the membrane and forming a hydrogen bond with the nitroxide moiety of the PC spin label in a fully extended conformation.

Table 4.

Ratio of hydrogen bonded, HB (g_xx= 2.00882) and non-hydrogen bonded components (g_xx=2.009435) determined from rigid-limit simulations of n-PC spin labels in DMPC and DPPC containing 30% mol cholesterol.

PC label	Ratio HB/non-HB in DMPC	Ratio HB/non-HB in DPPC
5	~6	> 8
7	1.8	2.3
10	0.32	0.36
12	0.45	0.62
14	0.55	0.8
16		0.8

Indeed, theoretical estimates⁵⁰ show that the concentration of water in the middle of the DPPC membrane is ~ 1 mM. It should be even lower for membranes containing cholesterol, since cholesterol is known to substantially decrease water permeability across lipid membranes^51,52. For the equilibrium constant between hydrogen bonded/unbonded forms of 16-sasl in toluene - trifluoroethanol mixtures Marsh⁴⁸ gives a value ~ 1M⁻¹. This value is determined from the isotropic hyperfine splitting at room temperature. Our estimates for this constant based on measurement of 2A_zz values for TEMPONE and TEMPO in ethanol – DBPH mixtures by 9GHz ESR quickly frozen at 77K give values of the same order, 0.5M⁻¹ and 0.3M⁻¹ respectively (see Supplementary Fig. 4S). As seen from Fig. 3, a two-component rigid–limit simulation for TEMPONE in pure ethanol (~ 17M hydroxyl concentration) gives their ratio of 1:3.6 and, hence, a rough estimate of 0.2M⁻¹ for the hydrogen bonding constant. A similar procedure for the 3-carboxy-2,2,5,5 tetramethylpyrrolidine spin-label (Supplementary Fig. 3C), which is structurally more similar to the nitroxide moiety on PC labels and shows very similar magnetic parameters, gives the component ratio 1:6 and hence a hydrogen bonding constant of ~ 0.35 M⁻¹. A 1:1 ratio of the two spectral components for a nitroxide located in the middle of the bilayer would thus yield a water concentration in the membrane core of ~0.5M, about three orders of magnitude higher than expected from theoretical estimates. In our test experiments (not shown) we did not see any appearance of a hydrogen bonded component for nitroxides dissolved in frozen nearly saturated (~16 mM) solutions of water in toluene. And finally, a strong point in favor of considering bent conformations of n-PC spin labels is the non-monotonic dependence of the hydrogen-bonded fraction on n (Table 4). In a fluid membrane the membrane depth distribution of nitroxide moieties for each spin-labeled lipid correlates to its n-value¹⁰. Higher n show deeper average immersion although the distribution is broad and even high n numbers show a substantial fraction of conformations with the nitroxide touching the carbonyl area. In the much denser packed gel phase the situation can be different and the spin labels may prefer defects in the lipid structure^53,54. One of the areas with such defects is just above the cholesterol rings and this should correspond to a hydrophobic local environment. It can be reached by the nitroxide of 10PC, but not 7PC. This would explain the jump in the fraction of non-polar component between 7 and 10PC, while further decrease in this fraction could be attributed to U-shaped conformations for higher values of n. These conformations put the nitroxide moiety back to the surface region with high water content, while the hydrocarbon chain mostly remains located in the hydrophobic part of the membrane.

DMPC and DPPC without cholesterol

It has been previously observed that in the absence of cholesterol, PC spin labels in DPPC¹⁹ and DMPC²¹ at rigid limit conditions have a strong broad background signal. This signal points at aggregation of spin labels and is most pronounced for the7–12 PC positions. Since no significant signs of such aggregation are seen in the fluid phase of the membrane or in the gel phase in the presence of cholesterol, this aggregation can be specifically attributed to the gel or crystalline (sub-gel) state of pure-lipid membranes. Also, since diffusion in the gel phase is slow, one can expect hysteresis effects and effects of the sample-treatment procedure, see^55,56. Fig. 6 shows spectra of PC spin labels in DMPC membranes without cholesterol. These spectra were recorded after the sample was slowly cooled from 295 to 85K within ~ 2 hours. As in the presence of cholesterol, two components with different g_xx are clearly discernible. However, another broad unresolved component, which is present for all spin labels but most pronounced for 7–12PC, follows a previously observed¹⁹ pattern for DPPC at 250GHz. Our 240 GHz data for DPPC obtained under the same conditions as those for DMPC are very similar to the 250 GHz results by Earle et al.¹⁹ and shown in Fig. 7. A non-polar (non -hydrogen bonded) component similar to the non-polar component in membranes containing cholesterol can be identified by its characteristic g_xx value. However, there is an important and obvious difference between DMPC and DPPC at first inspection. In DPPC there is only one resolved component which shows the g_xx value corresponding to the non-hydrogen bonded state of the nitroxide, while in DMPC two components are clearly discernible in the spectrum. This observation is in good accord with the X-band results, which indicate a less polar environment reported by high values of n in DPPC compared to DMPC (cf. Table 1, Figs. 1A, 1C).

Fig. 6.

n-PC spin labels in DMPC with no cholesterol at 80–82K. The samples are gradually frozen in the flow of gaseous nitrogen. The vertical lines indicate g values of 2.00882 and 2.009435. (The additional superposition signal for 7PC originates from an occasional manganese impurity in the sample holder).

Fig. 7.

n-PC spin labels in DPPC with no cholesterol at 80–82K. The samples are gradually frozen in the flow of gaseous nitrogen.

Table 1.

Hyperfine splitting parameter 2A_zz in DMPC and DPPC membranes with and without cholesterol determined by X-band ESR at 77K. To record the ESR spectra the samples in 1.2 mm ID capillaries after long exposure at 19°C were quickly submerged into liquid nitrogen.

	DMPC	DMPC/Chol	DPPC	DPPC/Chol
5	68.9	69.7	69.2	70.1
7	69.3	70.1	69.9 (broadening)	69.1
10	69.6	66.6	69.5 (more broadening)	66.5
12	68.4	66.3	67.2	66.4
14	69.3	66.4	66.3	66.8
16	69.3	66.2	66.7	66.5

It is very unlikely that small structural differences between the gel phase bilayers of DMPC and DPPC^57,58 cause dramatically different water penetration into these membranes. Moreover, it has been recently shown that water penetration into the membrane depends rather on the surface area of the lipid (which is nearly identical for DMPC and DPPC) than on the membrane thickness⁵⁹. Thus, it makes the explanation of the difference between DMPC and DPPC through dramatically different water concentration very unlikely. Even more intriguing, the average environment of the nitroxide in DPPC looks, at the first glance, less polar in the absence of cholesterol, since no component with a clear peak at smaller g_xx can be detected, cf. Fig 8.

Fig. 8.

Comparison of 14 PC spin label in DMPC and DPPC with/without cholesterol. All samples are slowly frozen in the flow of gaseous nitrogen. While in the presence of cholesterol the spectra can be described as a superposition of two rigid-limit components (cf. Fig. 5A,B), the most salient feature in the absence of cholesterol is a broad singlet-like background signal.

However, the most salient feature of all spectra, both DMPC and DPPC, in the absence of cholesterol, is a broad singlet-like background signal which accounts for most of the spins. The location of spins responsible for this background is unclear. In Fig. 8 one can see an apparent broadening of the hydrogen-bonded component in DMPC and complete absence of this component in DPPC, concomitant with an increase in the background. A likely explanation for these observations is that the hydrogen-bonded and non hydrogen- bonded forms of PC spin labels aggregate differently. If the hydrogen bonded form is more prone to aggregation than the non- bonded one, it likely becomes broader due to more interactions between spins. In the superposition, spectral peaks from this broad component are less intense and eventually, with an increase in broadening, not discernible at all. To test this hypothesis we cooled our samples very rapidly in an attempt to trap the hydrogen-bonded component before aggregation could occur.

Slow cooling vs. quick submerging into liquid nitrogen

Quantitative study of aggregation in the gel or subgel phases is difficult because of the slow diffusion rates and various hysteresis effects. For example, even though the gel phase of DPPC does not favor formation of gramicidin channels, it takes an hour to several days for these channels to dissociate⁵⁶. There are also indications pointing to a relatively slow time scale for aggregation of PC spin labels in the gel-phase. Fig. 9 shows saturation curves for 0.5 mol. % of 7PC in DMPC. At 19°C, in the P_β phase, this system show good saturation, with a P parameter of ~ 19. Saturation measurements performed immediately after cooling to 2°C give a similarly high P value. However, after longer exposure at this temperature the P value starts to decrease and after several hours drops below 2. This increase in relaxation is very likely due to aggregation, and this aggregation appears to be a relatively slow process at 2°C. Also, this increase in relaxation is reversible; return to 19°C reverses the aggregation and the P value. Although a temperature of 2°C should correspond to the subgel phase, the exact phase state of the lipid at this condition is not obvious due to likely supercooling, see below. It usually takes days at this temperature to form the L_C phase, which then can be characterized by X-Ray diffraction.

Fig. 9.

Change in relaxation curve of 7PC in DMPC under exposure at 2°C. This increase in relaxation occurs gradually within several hours and it can be completely reversed by a brief exposure at 19°C.

Previously, similar observations of exclusion of 5PC spin labels from DPPC after exposure at 0°C by using Saturation Transfer ESR⁶⁰ were attributed to formation of a sub-gel phase.

We expected that quick freezing (within ~ 0.2s) of the sample by instantly submerging into liquid nitrogen should substantially eliminate these slow lipid rearrangement effects compared to the standard slow cooling procedure using a flow cryostat. Indeed, spectra recorded after this simple quick-freeze procedure were different from the spectra obtained by gradual cooling in a flow of gaseous nitrogen. As seen in Fig. 10A the spectrum of 14PC in DPPC does not have a broad singlet-like component and can be well simulated by a simple superposition of two rigid limit spectra with g_xx of 2.00943 and g_xx= 2.00882. Interestingly, most 14PC spins are present in the form of the hydrogen bonded component, which greater than the non-bonded one by a 2.2:1 ratio.

Fig. 10.

14PC in DPPC (A) and DMPC (B): gradual cooling vs. quick freeze by submerging into liquid nitrogen. The spectra are recorded at 80–82K. In the case of quick freezing the broad unresolved component disappears and the spectrum can be well described as a superposition of HB/non-HB forms of the nitroxide. The HB/non-HB ratio is 2.2:1 for DPPC and 3.1:1 for DMPC.

As seen from Fig 11, quick freezing helps to dramatically decrease the amount of the broad background signal and yields well resolved rigid-limit spectra even when a standard cooling procedure results in nearly unresolvable, broad-feature spectra.

Fig. 11.

Obtaining better resolved spectra by quick freezing in liquid nitrogen.

PC spin labels in gel/subgel phases. A foreign body looking for a harbor

One of the possible artifacts of using molecular probes as reporters in biological and model membranes is their exclusion from the membrane interior to the membrane surface. Such exclusion of certain molecules or structural units is a natural feature of membrane biological function. For example, exclusion of tryptophans and their affinity to the membrane interface is an important factor in lipid-protein interactions and stabilization of certain conformations of transmembrane protein/peptides⁶¹. Exclusion from the bilayer and formation of some U-shaped conformations have been known for a long time for a number of fluorescent labels^11,8,9.

Spin labels are usually considered as more adequate molecular probes for lipid bilayers, as evidenced by studies in fluid membranes (see Introduction). Indeed, in the L_α phase of model membranes or natural biological membranes, ESR parameters of polarity, accessibility for paramagnetic molecules and molecular dynamics parameters^3,4,1 reported by PC spin labels change monotonically with an increasing value of n. NMR also shows that, despite a relatively broad distribution of conformations, the average depth position of n-PC nitroxides increases with n.

However, our current study shows that in the gel phase the situation is very likely different. Several ESR parameters (accessibility for Ni²⁺ ions absorbed on the membrane surface, hyperfine tensor and g-tensor components) unambiguously indicate that in the gel phase a substantial fraction of nitroxide is located in the membrane region with high water content.

An alternative explanation through water penetration into the membrane interior is unsuccessful because (1) it requires several orders of magnitude higher water content in the middle of membrane than any known theoretical prediction, (2) it cannot explain non-gradual change in the fraction of hydrogen-bonded nitroxide along the PC – series. On the other hand, assuming a high statistical weight of bent conformations with the nitroxide moiety close to the membrane surface can better explain these observations. Pure lipids DMPC and DPPC form the L′_β “tilted gel” phase with densely packed and relatively well ordered hydrocarbon chains. The ability of nitroxides groups on spin labeled stearic acids to be excluded from hydrocarbon environments and form U-shaped conformations has been previously reported^13,12. In principle, formation of the L′_β phase can be compared with freezing of a bulk three-dimensional solvent. In this case solutes are excluded from crystallizing solvent and form some regions with high local concentration. If the solute is a nitroxide radical, freezing of its solution in non glass-forming media is well known to yield a broad signal similar to one observed in DMPC or DPPC without cholesterol (Figs 6, 7).

In the case of a lipid in the gel phase the complete exclusion of spin labels into a separate phase with a high spin concentration appears to take two steps. The first one, exclusion of nitroxides from the hydrophobic area of the membrane, occurs once the lipid forms an L_β′ or even a P_β phase (cf. Fig 1A, Fig 2 for 19°C, the P_β phase of DMPC, also the quick-freeze spectra in Fig. 10). The second stage, which is a formation of separate phase by these bent-conformation molecules, can be attributed to lateral aggregation in the supercooled gel, possibly at the gel/subgel phase transition.

Although formation of subgel phases was initially observed after storing a multilamellar suspension of DPPC at ~0°C for several days³⁰, there are reports of quicker gel/subgel transformations in DPPC and DMPC. For example, in one protocol⁶² the subgel (SGII) phase in DPPC forms at about 7°C upon cooling at 2°C/min. These conditions are very similar to cooling conditions observed in our “slow cooling” experiment – and, in general, the cooling conditions usually existing in standard helium/nitrogen cryostats. For DMPC, formation of the subgel phase was reported after incubation at temperatures of −5°C or lower for 2h⁶³. But, the kinetics of subgel formation is complex⁶⁴ and in some cases the final subgel phase contains spectroscopic features characteristic for the L_β′ phase⁶⁵. It could be possible that aggregation of spin labels does not require a complete transformation into the subgel phase, in the sense of giving a clean X-ray pattern, and can happen sooner during cooling.

Again, we would like to stress that exclusion of nitroxides from the hydrophobic membrane core occurs in all gel phases and is a separate effect from the succeeding lateral aggregation in the supercooled gel phase.

Our quick freeze vs. slow cooling experiments can also help answer an important question about the nature of the two components in the 240GHz spectrum. Does the component with a larger g_xx correspond to the location of the nitroxide in the hydrophobic part of the membrane? Or both components arise from equilibrium of the hydrogen-bonded/non-hydrogen bonded forms in the same location, like TEMPONE in ethanol (cf. Fig. 3A)? The different aggregation behavior of spins contributing into these components points to their different location. That is, the major hydrogen-bonded component comes from nitroxides excluded from the hydrophobic part into the area with high water content, while the minor non-polar component may arise from spins somehow trapped in the defects of the hydrophobic membrane core.

Frozen DMPC/DPPC membranes with cholesterol: partition-like depth distribution of spin labels

As seen in Fig. 5 and Table 4, in frozen membranes of both DPPC and DMPC in the presence of 30% cholesterol all PC spin labels show well-resolved rigid limit spectra with no broad features. There is little change in the ESR lineshape for quick freeze vs. slow cooling in this case. Starting from 7PC, (i.e. n ≥ 7) all spectra consist of two components. Fractions of these components approximately determined by spectral simulations are given in Table 4. Clearly, there is an abrupt jump in the fraction of non-hydrogen-bonded compound from 7 to 10PC which is followed by a gradual decrease in this fraction with further increase in n. This observation is difficult to explain using simple models. Indeed, if we assume that the spin-labeled sn-chain takes on mostly the fully extended conformation, we would indeed observe a similar jump after the nitroxide moiety reaches the hydrophobic core of the membrane, but with further increase in the n number the hydrogen-bonded fraction would decrease and quickly disappear.

Another limiting case suggests that the acyl-chain hydrocarbon tether connecting the nitroxide to the lipid head group can take all possible coil conformations of the chain. It will put the possible position of the nitroxide of an n-PC spin label anywhere within a half sphere with a radius of the all-stretched conformation for the nitroxide tethers. This half sphere would rest on the membrane surface (cf. Fig. 12). The ratio of hydrogen bonded/unbonded components would be opposite to the volume ratio of the spherical cap, which is cut off by the border of the hydrophobic core of the membrane, and the rest of the half sphere. If the cap height is half of the sphere radius, the ratio will be 2.2, similar to what we see for 7PC (Table 4). Using this observation to set (in a rather arbitrary fashion) the cutoff of the hydrophobic core at “3.5” PC, half of the acyl tether for 7PC in the fully extended conformation, and further applying the formula for the spherical cap we get following predictions for the ratio of components.

Fig. 12.

A model of random distribution of conformations of the spin label tethers for 7PC. It puts the spin label in some random position within the half-sphere. The spherical cap is the intersection of the area available for the nitroxide and the hydrophobic core of the membrane.

Although this model is qualitatively better when compared to Table 4, it does not reproduce the observed abrupt drop between positions 7 and 10. Choosing different cutoff values or slightly modifying the model by using a cone instead of a whole half-sphere does not substantially improve it. Also, no model assuming some random distribution of nitroxide depth position would reproduce the decrease in the hydrophobic fraction with further increase in n beyond 10.

To better explain the observed effects one could assume some set of preferential depth positions in the membrane to be occupied by the nitroxide ring. These positions may be areas of defects in the membrane structure, which can more easily accommodate a structure-disturbing nitroxide moiety than areas with compact alignment of hydrocarbon chains. One of the areas of such defects could start above the end of the rigid fused-ring system of cholesterol⁶⁶, which is about the level reachable by n-PC labels starting from n=9. Once the tether length is sufficient, nitroxides start populating these favorable locations causing the change in the component ratio. However, the preference of the nitroxide ring for the location in the hydrophobic part of a DMPC/Cholesterol or DPPC/Cholesterol membrane does not completely overwhelm its affinity to some sites close to the membrane surface. It gives a partition-like distribution between the two sites which is observable in the spectrum of all n-PC labels with n>7 as two components.

A gradual decrease in the hydrogen-bonded fraction between PC10 and PC16 can be also explained using this partition model. For an n-PC spin label, the same carbon atom of the nitroxide ring that is connected to the acyl tether also has a hydrocarbon tail with a length of 18-n carbons attached. Bringing the nitroxide label of 10PC to the membrane surface will require, on the average, placing more hydrophobic CH₂ groups to the polar area than for 16PC. This will be associated with some energy penalty preventing the nitroxide of 10PC from leaving the hydrophobic core and affecting the partition. Note also a slight drop in the accessibility of the 10PC position for Ni²⁺ ions for cholesterol-free DMPC membranes, which may have the same origin.

What about biological membranes?

The gel phase rarely exists in biological membranes and the subgel phase is unknown for them. However, polarity measurements in frozen natural membranes using PC spin labels can also be affected by the affinity of spin labels to some structural defects, especially in the presence of proteins. For example, all PC labels in mixtures of DPPC with gramicidin A show about the same g_xx value, indicative of the same polarity¹⁹, most likely due to an interaction with the peptide’s α-helical structure. Also, natural membranes usually have complex lipid head group composition and variety of lengths and unsaturation in acyl chains. This can affect the free volume available for spin labels both in the hydrophobic core and in the membrane interface. Fig. 13 shows 240 GHz spectra for 14PC in egg-yolk lecithin with and without cholesterol.

Fig 13.

14 PC in EYL with and without cholesterol at 82K. The samples are gradually frozen in the flow of gaseous nitrogen. The corresponding one and two-component (the ratio of components with g_xx=2.009410 and 2.008820 is 0.5) rigid limit simulations are shown in red.

One can see that, similar to DMPC and DPPC membranes, the fraction of the hydrogen-bonded component drops with addition of cholesterol (cf. Figs 5 and the quick-freeze spectra in Fig. 10 ). However, compared to DMPC and DPPC membranes this fraction in EYL is lower. Actually, in EYL/Cholesterol only a non-polar component can be observed in the spectrum. It can be explained by the presence of a double bond in the acyl chain of EYL. This unsaturated bond has a kink, which disrupts the packing of lipids. It creates extra free space in the hydrophobic core of the membrane⁶⁷, and thus may shift the partition of nitroxides in favor of the membrane interior. Also, a variety of headgroups may improve packing in the polar head area, additionally forcing nitroxides into the hydrophobic core.

CONCLUSIONS

Location of the reporter nitroxide moiety in frozen membranes with and without cholesterol was studied using ESR at 9 and 240GHz.

1. Test experiments on four different nitroxide radicals in solvents of different polarity showed that 240GHz ESR can clearly detect three components having three distinct g_xx values: (1) “non-polar”, (2) “polar”, (3) “very polar”. In some solvents these component coexist, for example (1) and (2) in ethanol, (2) and (3) in water/glycerol and TFE, and can be separated as distinct peaks in the X-region of the spectrum. These three components should be attributed to different hydrogen-bonding states of the nitroxide.

2. In frozen lipid membranes with or without cholesterol, n-PC spin labels with n> 7 usually show two-component 240 GHz ESR spectra. These components correspond to hydrogen-bonding states (1) and (2). We attribute the hydrogen-bonded component for n> 7 to bent conformations of the spin labeled molecule, with the nitroxide close to the membrane surface. In membranes without cholesterol, ESR spectra obtained by relatively slow sample cooling in the flow of gaseous nitrogen show also a broad background signal, most pronounced for PC labeling positions 7–12.

3. Additional evidence for the presence of such bent conformations and their prevalence in the gel-phase DMPC membranes without cholesterol was obtained at 9.5 GHz ESR from A_zz values determined at 77K and by relaxation enhancement for PC spin labels from Ni²⁺ ions absorbed at the membrane surface at 19°C (P_β phase).

4. If instead of slow cooling a sample of cholesterol-free membrane of DMPC or DPPC is quickly frozen by being instantly submerged into liquid nitrogen, the broad background signal is replaced by the hydrogen-bonded component.

5. We believe that the appearance of bent sn-2 chain conformations due to exclusion of nitroxides from pure lipids occurs at the formation of all gel phases, including the P_β rippled phase. In a sense, this process is similar to exclusion of solutes from a crystallizing three-dimensional solvent. However, further slow formation of a separate phase of spin-labeled lipid which manifests itself in the broad background ESR signal occurs in the supercooled gel. It could be likely attributed to the effects of a gel/subgel phase transition.

6. In membranes containing cholesterol the observed ratio of the hydrogen-bonded components can be best described by a partition-like equilibrium between nitroxides located in defects of lipid structure in the hydrophobic core of the membrane and close to the membrane surface.

7. The main conclusion of this work is that the ESR parameters of PC spin labels in frozen model membranes and, quite likely, in majority of biological membranes do not follow the membrane polarity profile, water penetration profile or any other continuously changing property of membrane environment. Instead, they reflect a distribution between hydrogen-bonded and non-hydrogen-bonded forms of the nitroxide. This distribution is affected by a number of factors in the membrane composition, chain packing in the lipid phase and folding properties of the sn-2 spin labeled chain itself for each specific PC label.

Table 5.

Estimates for the ratio of hydrogen bonded and non-hydrogen bonded components based on the simple model shown in Fig. 12. The location of the nitroxide moiety is considered in the hydrophobic core of the membrane if the distance to the membrane surface is more than half of the acyl tether for 7PC in the fully extended conformation (“3.5 PC”).

n-PC	5	7	10	12	14	16
HB/non-HB	7.2	2.2	1.01	0.74	0.58	0.48

LIST OF ABBREVIATIONS

HF ESR,HFHF ESR

High Field ESR, High Field/High Frequency ESR

DMPC

1,2-dimyristoyl-sn-glycero-3-phosphocholine

DPPC

1,2-dipalmitoyl-sn-glycero-3-phosphocholine

EYL

egg yolk lecithin

n-PC spin labels

1-palmitoyl-2-stearoyl-(n-doxyl)-sn-glycero-3-phosphocholines

TFE

2,2,2-trifluoroethanol

MCH

methylcyclohexane

TEMPO

2,2,6,6-tetramethylpiperidine-N-oxyl

TEMPOL

4-Hydroxy-TEMPO, 4-hydroxy-2,2,6,6-tetramethylpiperidine-N-oxyl

TEMPONE

Oxo-TEMPO, 4-oxy-2,2,6,6-tetramethylpiperidine-N-oxyl