Multiple locations of peptides in the hydrocarbon core of gel-phase membranes revealed by peptide (13)C to lipid (2)H rotational-echo double-resonance solid-state nuclear magnetic resonance.

Membrane locations of peptides and proteins are often critical to their functions. Solid-state rotational-echo double-resonance (REDOR) nuclear magnetic resonance is applied to probe the locations of two peptides via peptide (13)CO to lipid (2)H distance measurements. The peptides are KALP, an α-helical membrane-spanning peptide, and HFP, the β-sheet N-terminal fusion peptide of the HIV gp41 fusion protein that plays an important role in HIV-host cell membrane fusion. Both peptides are shown to have at least two distinct locations within the hydrocarbon core of gel-phase membranes. The multiple locations are attributed to snorkeling of lysine side chains for KALP and to the distribution of antiparallel β-sheet registries for HFP. The relative population of each location is also quantitated. To the best of our knowledge, this is the first clear experimental support of multiple peptide locations within the membrane hydrocarbon core. These data are for gel-phase membranes, but the approach should work for liquid-ordered membranes containing cholesterol and may be applicable to liquid-disordered membranes with appropriate additional analysis to take into account protein and lipid motion. This paper also describes the methodological development of (13)CO-(2)H REDOR using the lyophilized I4 peptide that is α-helical and (13)CO-labeled at A9 and (2)Hα-labeled at A8. The I4 spins are well-approximated as an ensemble of isolated (13)CO-(2)H spin pairs each separated by 5.0 Å with a 37 Hz dipolar coupling. A pulse sequence with rectangular 100 kHz (2)H π pulses results in rapid and extensive buildup of REDOR (ΔS/S0) with a dephasing time (τ). The buildup is well-fit by a simple exponential function with a rate of 24 Hz and an extent close to 1. These parameter values reflect nonradiative transitions between the (2)H spin states during the dephasing period. Each spin pair spends approximately two-thirds of its time in the (13)CO-(2)H (m = ±1) states and approximately one-third of its time in the (13)CO-(2)H (m = 0) state and contributes to the ΔS/S0 buildup during the former but not the latter time segments.

Residue-specific membrane locations are an important feature of membrane-bound peptides and proteins and in some cases are correlated with function.[1,2] These locations have been semiquantitatively probed by relaxation-based approaches, including fluorescence, electron paramagnetic resonance (EPR), and solid-state nuclear magnetic resonance (SSNMR).[3−8] More quantitative locations have been based on SSNMR measurements of peptide-to-lipid internuclear distances. For example, the proximity to the lipid headgroups has been probed via rotational-echo double-resonance (REDOR) determination of peptide backbone 13CO to lipid 31P internuclear distances.[2,9] Locations within the membrane hydrocarbon core have been probed by 13CO–19F REDOR of samples with ∼0.1 mole fraction of lipids with selective 1H → 19F substitutions in their acyl chains.[1,9] A 1H → 19F substitution is a chemical change and has the potential disadvantage of changing acyl chain and/or peptide location within the membrane.[10] A better approach is 13CO–2H REDOR of samples with lipids with 1H → 2H substitutions in their acyl chains. 1H and 2H are chemically equivalent, so there is no perturbation of the acyl chain or peptide locations.[11−15]

It is typical to develop a model with a single membrane location of the peptide based on data from the previously described approaches and the known peptide structure. It is atypical to consider a quantitative model with a distribution of membrane locations (e.g., multiple distinct locations with different populations), in part because the effects of this distribution on the data cannot be readily deconvolved from other effects. For example, even for a single membrane location, there is a broad distribution of peptide 13C to lipid 19F distances because the samples contain only a small fraction of fluorinated lipids.[9] To the best of our knowledge, this study describes the first clear experimental support for a distribution of membrane locations of a peptide within the hydrocarbon core and relies on peptide 13CO to lipid 2H distances of samples with large (≥0.8) mole fractions of deuterated lipids, which minimizes the intrinsic distribution of 13CO–2H distances. The experimental results of this study are compared to recent molecular dynamics simulations of the distribution of membrane locations of a transmembrane helix. This study was conducted with gel-phase membranes lacking cholesterol, but the approach should be readily applicable to liquid-ordered-phase membranes containing cholesterol. The approach may also be applicable to fluid-phase membranes lacking cholesterol with additional analysis to take into account peptide and lipid motion.

The 13CO–2H REDOR experiment and analysis are developed and validated using the 17-residue I4 peptide that is lyophilized and has a regular α-helical structure.[16,17] I4 is synthesized with a specific 13CO label at A9 and 2Hα label at A8. To a good approximation, the sample has isolated 13CO–2H spin pairs that all have the same internuclear separation (r) of 5.0 Å with a corresponding dipolar coupling (d) of 37 Hz.

Materials and Methods

Peptides and Lipids

The 13CO–2H REDOR experiment and analysis were developed using the lyophilized I4_A9C peptide that has predominant α-helical structure (sequence of AEAAAKEAAAKEAAAKAW) and A8 2Hα and A9 13CO labels separated by 5.0 Å (Table 1).[16,17] The membrane-associated peptides were (1) KALP_A11C (sequence of GKKLALALALALALALALALKKA and A11 13CO label) and (2) HFP_G5C and HFP_F8C (sequence of AVGIGALFLGFLGAAGSTMGARSWKKKKKKA and either G5 or F8 13CO labels). KALP is a designed transmembrane peptide with α-helical structure, and HFP contains the N-terminal fusion peptide of the HIV gp41 protein.[1,18] HFP plays a key role in gp41-catalyzed fusion between the HIV and host cell membranes. HFP has oligomeric antiparallel β-sheet structure.[19,20] Membranes contained dipalmitoylphosphatidylcholine (DPPC) lipid deuterated at palmitoyl carbons 2 (D4), 7 and 8 (D8), or 15 and 16 (D10), or dimyristoylphosphatidylcholine (DMPC) perdeuterated at myristoyl carbons 2–14 (D54) (Figure 1). The D4 2H’s are close to the membrane headgroups, and the D8 and D10 2H’s are at the leaflet midpoint and bilayer center, respectively (Figures 3C and 4E,F). The DPPC and DMPC data are directly comparable because of very similar chemical structures. The HFP_G5C samples also included 0.2 mole fraction DTPG (ditetradecylphosphatidylglycerol) lipid, which reflects some negatively charged lipids in host cell membranes.[21]

Table 1

Peptide Sequences

peptide	sequencea
I4_A9C	AEAAAKEAAAKEAAAKAW
KALP_A11C	GKKLALALALALALALALALKKA
HFP_G5C	AVGIGALFLGFLGAAGSTMGARSWKKKKKKA
HFP_F8C	AVGIGALFLGFLGAAGSTMGARSWKKKKKKA

An underlined residue has a backbone 13CO label, and a bold residue has a 2Hα label.

Figure 1

Chemical structures and 2H labeling of the lipids.

Figure 3

(A) τ = 40 ms S0 (black) and S1 (colored) 13CO–2H REDOR spectra of KALP_A11C. (B) Experimental (ΔS/S0)exp (points with uncertainties) and best-fit exponential buildup (solid line) plots of ΔS/S0 vs τ. The D4 data are not fit. The uncertainties are displayed for all (ΔS/S0)exp data, but in many cases, the uncertainties are comparable to or smaller than the displayed symbols. (C) Membrane locations of KALP with major (left) and minor (right) populations. The colored bands are 2H positions; the A11 13CO nuclei are orange dots, and the lysine side chains are brown lines. The samples contain ∼1 μmol of peptide and ∼50 μmol of lipid. For the displayed τ = 40 ms spectra, the numbers of summed S0 or S1 scans are 18165 for D4, 31416 for D8, 25998 for D10, and 26882 for D54.

Figure 4

τ = 40 ms S0 (black) and S1 (colored) 13CO–2H REDOR spectra of membrane-associated (A) HFP_F8C and (B) HFP_G5C. Experimental (ΔS/S0)exp (points with uncertainties) and best-fit exponential buildup (solid line) plots of ΔS/S0 vs τ of (C) HFP_F8C and (D) HFP_G5C. The uncertainties are displayed for all (ΔS/S0)exp data, but in many cases, the uncertainties are comparable to or smaller than the displayed black squares. (E and F) Deep and shallow membrane insertion of HFP. Residues A1 and A14 are close to the membrane surface in concurrence with peptide 13CO–lipid 31P distances of ∼5 Å for these residues.[1] For the sake of clarity, only one β-strand is displayed. The samples contain ∼2 μmol of peptide and ∼50 μmol of lipid. For the displayed τ = 40 ms spectra of the HFP_F8C samples, the numbers of summed S0 or S1 scans are 21799 for D8, 38797 for D10, and 14113 for D54. For the displayed HFP_G5C spectra, the numbers are 34727 for D8 with DTPG and 35602 for D10 with DTPG.

Peptides were made by Fmoc solid-phase peptide synthesis and purified by reversed-phase high-performance liquid chromatography using a C4 column. The purity was >95% as assessed by mass spectrometry. I4 and KALP had N-terminal acetylation, and KALP also had C-terminal amidation. Lipids were purchased from Avanti Polar Lipids (Alabaster, AL) and synthesized using deuterated carboxylic acids purchased from CDN Isotopes (Pointe Claire, QC).

Sample Preparation

The I4 sample was lyophilized peptide without lipid. The other samples contained 50 μmol of lipid and either 1 μmol of KALP or 2 μmol of HFP. Lipid and peptide were dissolved in a 2:3:2 volume ratio mixture of 2,2,2-trifluoroethanol, chloroform, and 1,1,1,3,3,3-hexafluoroisopropanol. Cosolubilization minimized the fraction of kinetically trapped peptide on the membrane surface. Solvent was removed by nitrogen gas and overnight vacuum. The peptide/lipid film was then suspended in 5 mM HEPES/10 mM MES buffer (pH 7.4) and homogenized by 10 freeze/thaw cycles in liquid nitrogen. After centrifugation at 270000g for 4 h, the harvested pellet was lyophilized, packed in a 4 mm diameter magic angle spinning (MAS) rotor, and hydrated with 20 μL of buffer.

Solid-State NMR Spectroscopy and Data Analysis

Data were acquired with a 9.4 T spectrometer and a triple-resonance probe tuned to 1H, 13C, and 2H frequencies. Samples were cooled with gaseous nitrogen at −50 °C, which corresponded to a sample temperature of approximately −30 °C. The magic angle spinning (MAS) frequency is 10.00 kHz, and the pulse sequence is as follows: (1) 1H π/2 pulse, (2) 1H–13C cross-polarization (CP), (3) dephasing period of duration τ, and (4) 13C detection (Figure S1 of the Supporting Information). S0 and S1 acquisitions did not and did have, respectively, 2H π pulses in the middle of each rotor cycle during the dephasing period. Both acquisitions had 13C π pulses at the end of each rotor cycle during the dephasing period (except the last one) as well as 1H decoupling during the dephasing and detection periods. Parameters included (1) a 5.0 μs 1H π/2 pulse, (2) a 1.5 ms CP with 50 kHz 1H and 62–66 kHz ramped 13C fields, (3) 8.3 μs 13C π pulses and 5.1 μs 2H π pulses, and (4) 75 kHz 1H decoupling with two-pulse phase modulation. The recycle delay was 1 s for τ values of 2, 8, and 16 ms, 1.5 s for τ values of 24 and 32 ms, and 2 s for τ values of 40 and 48 ms. Spectra were processed with 100 Hz Gaussian line broadening and baseline correction and externally referenced to the methylene peak of adamantane at 40.5 ppm.[22]

The experimental (ΔS/S0)exp = (S0 – S1)/S0 values at each τ were calculated using S0 and S113CO intensities determined with a 3.0 ppm integration window. The uncertainty of the S0 or S1 intensities (σ) is determined as the standard deviation of the experimental intensities from spectral regions that do not contain signal. The 3 ppm integration window in each region matches that used for S0 and S1 integration. The uncertainty in ΔS/S0 is calculated from propagation of error as σ[(1/S0)2 + (S1/S02)2]1/2. For the I4 peptide, similar dephasing was observed with 6, 8, and 10 kHz MAS frequencies; e.g., for the 24 ms dephasing time, the respective (ΔS/S0)exp values were 0.38(3), 0.40(1), and 0.38(1), respectively.

Comparison was made between the (ΔS/S0)exp of the I4 sample and the (ΔS/S0)sim calculated for isolated 13CO–2H spin pairs using the SIMPSON simulation program.[23] Inputs to the program include the rf fields of the pulses, the MAS frequency, and the 2H quadrupolar anisotropy. The program was also used to calculate the contribution of natural abundance (na) 13CO nuclei to the (ΔS/S0)exp of the I4 sample and then determine the (ΔS/S0)lab associated with the labeled (lab) A9 13CO nuclei. The r ≡ 13COna–2HαA8 separation was determined for each na 13CO site using a model of regular α-helical structure. r was used to determine the corresponding 13CO–2H dipolar coupling d. The (ΔS/S0)na versus τ was calculated using the SIMPSON program with input d. The (ΔS/S0)lab was then calculated usingwhere (ΔS/S0)na = ∑[(ΔS/S0)na]/∑ and flab and fna are the total fractional populations of lab and na 13CO nuclei, respectively. The typical (ΔS/S0)lab ≈ 1.08(ΔS/S0)exp for the I4 peptide. Table S2 of the Supporting Information presents all (ΔS/S0)exp and (ΔS/S0)lab values versus τ.

For the membrane samples, semiquantitative understanding of the relative contributions of (ΔS/S0)lab and (ΔS/S0)na to (ΔS/S0)exp is obtained with a model for which half the na sites have r values similar to rlab and half are more distant such that their (ΔS/S0)na ≈ 0. For this model, the average (ΔS/S0)na ≈ (ΔS/S0)lab/2, and for flab ≈ 0.8 and fna ≈ 0.2, the consequent (ΔS/S0)lab ≈ 1.1(ΔS/S0)exp. Because the differences between the (ΔS/S0)exp and (ΔS/S0)lab values are small and somewhat uncertain, data analysis for the membrane samples is based on (ΔS/S0)exp.

Results

Representative S0 (black) and S1 (colored) spectra with a τ of 40 ms are displayed in the insets of Figures 2–4. For S0 and S1, there was respective complete and incomplete averaging of 13CO–2H dipolar coupling, with an S113CO spectral intensity consequently smaller than that of S0. The I4_A9C and KALP_A11C13CO peak shifts are 179 ppm and consistent with α-helical structure.[22,24] The HFP_F8C and HFP_G5C shifts are 174 and 171 ppm, respectively, and consistent with β-sheet structure.[19] The 13CO S0 intensities have ∼0.8 fractional contribution from the labeled (lab) nuclei; e.g., for KALP, the lab signal = 0.99 and the natural abundance (na) signal = 0.24 (0.011 × 22). The experimental buildup (ΔS/S0)exp = (S0 – S1)/S0 versus τ is calculated from 13CO S0 and S1 intensities. The experimental (ΔS/S0)exp values typically differ by at most 10% from the (ΔS/S0)lab values of the lab 13CO nuclei as explained in Materials and Methods.

Figure 2

13CO–2H REDOR spectra with a 40 ms dephasing time for the I4_A9C peptide as well as ΔS/S0 vs τ. Black squares are the (ΔS/S0)lab values of the labeled A9 13CO nuclei and are numerically very close to the experimental (ΔS/S0)exp. For all values of τ, the difference between the (ΔS/S0)lab and (ΔS/S0)exp values is ≤0.06 (Table S2 of the Supporting Information). The uncertainties are displayed for all (ΔS/S0)lab data, but in many cases, the uncertainties are comparable to or smaller than the displayed black squares. Blue triangles are (ΔS/S0)sim calculated using the SIMPSON program without 2H relaxation. The (ΔS/S0)sim values are best-fit to the (ΔS/S0)lab with a 22 Hz 13CO–2H dipolar coupling. The red line is the best-fit exponential buildup. The peptide is synthesized with a specific 13CO label at A9 and a 2Hα label at A8. To a good approximation, the sample has isolated 13CO–2H spin pairs that all have the same internuclear separation (r) of 5.0 Å with a corresponding dipolar coupling (d) of 37 Hz. The sample contains ∼15 μmol of lyophilized peptide. The number of summed S0 or S1 scans is 500 for τ values of 2–48 ms and 2000 for τ values of 56–80 ms.

The membrane samples were likely in the thermodynamic equilibrium state. This assertion was evidenced by ΔS/S0 values of replicate samples that agreed to within ±0.03. In addition, similar ΔS/S0 values were observed in a sample prepared using a different means of incorporation of peptide into the membrane, in particular: (1) mixing HFP_G5C with unilamellar lipid vesicles (D10+DTPG) in an aqueous solution, (2) centrifugation to form a pellet containing vesicles and bound HFP with unbound HFP in an aqueous solution, and (3) lyophilization of the pellet, packing in the rotor, and hydration with buffer. The ΔS/S0 of this sample agreed within ±0.05 with the ΔS/S0 of the comparable HFP_G5C sample prepared by organic cosolubilization.

Analysis of the REDOR Data of the I4_A9C Sample

This is an excellent model system because it contains isolated 13CO–2H spin pairs that have r = 5.0 Å and d = 37 Hz (Figure 2). The REDOR experiment was conducted with an ∼100 kHz 2H rf field, and the (ΔS/S0)lab buildup is rapid and extensive. This is generally consistent with buildups calculated using the SIMPSON quantum mechanics simulation program with 2H fields of ≥100 kHz (Figure S3 of the Supporting Information). For ≤40 kHz fields, the calculated buildups are attenuated because of the ∼65 kHz distribution of 2H resonant offsets associated with a rigid C–2H bond. Attenuated buildups have been observed in previous experiments performed with lower 2H rf fields.[15]

There are also systematic differences between (ΔS/S0)lab and the best-fit SIMPSON-calculated (ΔS/S0)sim of the I4_A9C sample, including respective exponential versus sigmoidal shapes and maximal values of ∼1 versus ∼(2/3). In addition, the best-fit d of 22 Hz is smaller than the expected d of 37 Hz. The differences are not due to deviation from regular α-helical structure. An earlier 13CO–15N REDOR study of the I4 peptide with A9 13CO and A13 15N labeling yielded a buildup of (ΔS/S0)exp versus τ that was very well-fit by SIMPSON to a 45 Hz coupling corresponding to the expected 13COA9–15NA13 distance of 4.1 Å in an α-helix.[17]

The differences for 13CO–2H REDOR between (ΔS/S0)lab and (ΔS/S0)sim are explained by effects of nonradiative transitions between the m = ±1 states and the m = 0 states of individual 2H nuclei during the dephasing period that are not considered in the SIMPSON calculations. These transitions are evidenced by measured 2H T1 relaxation times in our samples in the 50–100 ms range (Table S1 of the Supporting Information). As detailed below, there is an effect on (ΔS/S0)lab because these nonradiative transitions have a much greater effect on the S1 signal than on the S0 signal.

At any specific time during the dephasing period, the I4 sample has a two-thirds fractional population of 13CO–2H (m = ±1) spin pairs that experience dipolar coupling and a one-third fractional population of 13CO–2H (m = 0) pairs that do not experience dipolar coupling. In the absence of 2H relaxation, a spin pair is either 13CO–2H (m = ±1) or 13CO–2H (m = 0) for the entire dephasing period. During each rotor cycle of the S0 acquisition, the dipolar evolution of a 13CO–2H (m = ±1) pair is completely refocused by MAS so the pair makes a full contribution to the 13CO S0 signal. Dipolar evolution is not refocused during a rotor cycle of the S1 acquisition, and the pair makes an attenuated contribution to the S1 signal. A 13CO–2H (m = 0) pair does not experience coupling, so there is not evolution for either the S0 or S1 acquisition and there are complete and equal contributions to the S0 and S1 signals. In the absence of relaxation, the ΔS/S0 buildup is therefore only due to 13CO–2H (m = ±1) pairs, so the consequent (ΔS/S0)max is expected to be ∼(2/3). This is observed for the SIMPSON calculation that does not include relaxation (Figure 2).

In the presence of 2H relaxation with T1 comparable to the dephasing period, there may be one or a few instantaneous and stochastic changes in a 2H m state. Stochastic means the transitions of individual 2H’s are uncorrelated with each other but there is a defined overall transition rate. For the S0 acquisition, there is dipolar evolution for a 13CO–2H pair during the one or few S0 rotor cycles in which the 2H changes from m = ±1 to m = 0 or vice versa. However, the net S0 signal attenuation is negligible because there is no evolution during the other (typically hundreds of) rotor cycles of the dephasing period.

Relaxation has a larger impact on the S1 signal. A spin pair is 13CO–2H (m = ±1) during some time segments of the dephasing period and 13CO–2H (m = 0) during other segments. There is dipolar evolution during the 13CO–2H (m = ±1) segments with a corresponding attenuated contribution to the S1 signal and no evolution during the 13CO–2H (m = 0) segments and corresponding full contribution to the S1 signal. As noted above, the m = ±1 ↔ m = 0 transitions are stochastic, so the time segments are uncorrelated among the 2H nuclei. Qualitatively, each pair spends approximately two-thirds of the dephasing period as 13CO–2H (m = ±1) and approximately one-third as 13CO–2H (m = 0). It is therefore expected that the buildup rate of ΔS/S0 will be ∼(2/3) the dipolar coupling for 13CO–2H (m = ±1) pairs. This matches the fitting of the I4 data with SIMPSON calculations. The best-fit d = 22 Hz is ∼(2/3) of the known d = 37 Hz of the 13CO–2H (m = ±1) pairs of this sample.

Relaxation also affects the value of (ΔS/S0)max. Most pairs are 13CO–2H (m = ±1) for some segments of the dephasing period and therefore contribute to the ΔS/S0 buildup. The ΔS/S0 for large τ values is therefore expected to be >(2/3), which is consistent with the ΔS/S0 of Figure 2.

The SIMPSON program is based on coherent quantum mechanics, and it is not straightforward to incorporate stochastic changes of the 2H m state into this program. Analysis of (ΔS/S0)exp versus τ is instead based on excellent fitting to the simple exponential buildup function A(1 – e–γτ), where γ and A are fitting parameters (Figures 2–4 and Table 2). The buildup rate γ is correlated to the 13CO–2H dipolar coupling, and the buildup extent A is correlated to the fraction of lab 13CO nuclei with this coupling. The corresponding value of 1 – A is correlated to the fraction of lab 13CO nuclei not coupled to 2H. These correlations are most quantitatively understood from the fitting results for the I4 sample that contains isolated 13CO–2H spin pairs with r = 5.0 Å and corresponding d = 37 Hz. The best-fit γ = 24 Hz ≈ (2/3)d is consistent with an individual 2H spending approximately two-thirds of the dephasing period in the m = ±1 states and approximately one-third in the m = 0 state. For this model of 2H relaxation, all 13CO–2H pairs contribute to the ΔS/S0 buildup, and the expected (ΔS/S0)max ≈ 1. However, the best-fit A = 0.87 implies that 0.13 fraction of the pairs remain as 13CO–2H (m = 0) for the longest measured dephasing period. This fraction matches the expected value for τ ≈ T1, as calculated using (1/3)(1 – e–τ/).

Table 2

REDOR Fitted Parametersa

sample	A	γ (Hz)	d (Hz)	r (Å)
I4_A9C	0.87(5)	24(2)	37	5.0
KALP_A11C in D8	0.15(2)	47(10)	72(15)	4.0(3)
KALP_A11C in D10	0.48(4)	34(5)	52(8)	4.5(2)
KALP_A11C in D54	0.96(1)	85(4)	131(6)	3.3(1)
HFP_F8C in D8	0.21(1)	71(10)	109(15)	3.5(2)
HFP_F8C in D10	0.82(20)	16(5)	25(8)	5.7(6)
HFP_F8C in D54	0.99(1)	122(1)	188(2)	2.9(1)
HFP_G5C in D8 with DTPG	0.45(5)	27(5)	42(8)	4.8(3)
HFP_G5C in D10 with DTPG	0.85(3)	37(3)	57(5)	4.3(1)

The fitting function is A(1 – e–γτ). The γ = 0.65d is determined from the known 13CO–2H dipolar coupling d = 37 Hz for the I4 sample. This correlates to γ = (2/3)d expected when a 2H spin spends approximately equal times during the dephasing period in the m = ±1 and m = 0 states because of m = ±1 ↔ m = 0 nonradiative transitions. The calculated 13CO–2H distance r = (4642/d)1/3 is quantitative for a sample like I4 that contains isolated 13CO–2H spin pairs with a single separation. For the KALP and HFP samples, a 13CO may be coupled to multiple 2H’s, although the buildup rate γ is probably dominated by coupling to the closest 2H. The small r values of the perdeuterated D54 samples are closer than the expected 13CO–2H van der Waals separation and may reflect couplings to multiple nearby 2H’s. The fitted A is correlated to the fraction of molecules with the fitted γ and corollary-calculated d and r. The term 1 – A is correlated to the fraction of molecules with d ≈ 0. The AD8 + AD10 > 1 of HFP_G5C may mean that some G5 13CO nuclei contact both D8 and D10 2H nuclei.

We do not have a rigorous argument to explain the observation of exponential buildup of ΔS/S0 versus τ. However, stochastic processes often lead to exponential time dependence, e.g., NMR longitudinal and transverse relaxation.

KALP and HFP

Membrane locations of peptides are assessed by comparative analysis of the (ΔS/S0)exp with lipids with perdeuterated (D54) or selectively deuterated (D4, D8, or D10) acyl chains (Figures 3 and 4). In the absence of 2H relaxation, sigmoidal buildups are expected. However, like I4, all membrane peptide data are well-fit by single-exponential but not sigmoidal buildups (Table 2). This is consistent with a 2H relaxation effect that is additionally evidenced by measured T1’s typically in the range of 50–100 ms (Table S1 of the Supporting Information). Each fitted γ is interpreted with a single 13CO–2H spin pair model with calculated d = γ/0.65 and r (Å) = [4642/d (Hz)]1/3. The (ΔS/S0)exp values reflect coupling of peptide 13CO to multiple lipid 2H’s. However, because of the r–3 dependence of coupling, the buildup rates for the D4, D8, and D10 samples are likely dominated by the closest pair. This is supported by r values of ≈4–5 Å that match the peptide 13CO–lipid 2H van der Waals separation. The perdeuterated D54 samples likely have at least two close 2H’s, which results in buildup that is faster than that for a single 2H. This is consistent with a calculated r of ≈3 Å, which is smaller than the van der Waals separation.

For either KALP or HFP, the D54 (ΔS/S0)exp ≈ 1 at longer values of τ results in best-fit A ≈ 1. This supports the locations of A11 and F8 within the membrane hydrocarbon core for all KALP and HFP molecules. Rapid and quantitative (ΔS/S0)exp buildup also supports monomer rather than oligomer KALP because 13CO’s in the oligomer interior are distant from lipid 2H’s and would have slower buildup. Previous work showed oligomeric antiparallel β-sheet HFP. The rapid and quantitative HFP buildup supports small (∼10 molecule) oligomers in which all HFP molecules contact lipid acyl chains.

The D4, D8, and D10 data provide more detailed location information. One potential concern is that variation among buildup extents, i.e., A values, is primarily due to differences in 2H quadrupolar anisotropies (characterized by the Pake doublet splitting ΔνQ) with smaller A values correlated to greater ΔνQ values via the 2H pulse resonance offset. However, this concern is not supported by several lines of evidence. The samples are cooled with N2 gas at −50 °C, and the 2H’s have rigid ΔνQ ≈ 120 kHz with D4 and D8 and a superposition of −CD2 ΔνQ ≈ 120 kHz and −CD3 ΔνQ ≈ 30 kHz with D10 (Figure S2 of the Supporting Information). I4 and D54 samples have dominant ΔνQ ≈ 120 kHz, yet their buildups and A values are larger than those of D10 samples with smaller average ΔνQ values. In addition, other peptide buildups show AD8 > AD10, AD4 > AD8, or A(-CD,-CD2) > A(-CD3) (Figures S5–S7 of the Supporting Information). SIMPSON simulations incorporating the experimental 100 kHz 2H π pulses show superimposable buildups for a ΔνQ of either 120 or 0 kHz (Figure S3A of the Supporting Information).

Both KALP and HFP show significant buildups in D10 and D8 but not D4 membranes (Figures 3 and 4 and Figure S4 of the Supporting Information). This is consistent with D54 data and supports an α-helical KALP monomer with transmembrane topology and a β-sheet HFP with insertion into the membrane hydrocarbon core. Typical fittings show AD10 ≠ AD8 with both A values being <1 and also r values of ≈4–5 Å, which correspond to peptide 13CO–lipid 2H van der Waals contacts (Table 2). These parameters are inconsistent with a single membrane location for which AD10 ≈ AD8 ≈ 1 and rD10 ≠ rD8 would be expected. For example, KALP centered in the membrane would be evidenced by rD10 ≈ 4 Å and rD8 ≥ 8 Å. The actual KALP parameters could be understood only using two distinct membrane locations with major and minor populations (Figure 3C and Figure S8 of the Supporting Information). These populations have A11 contact with the D10 and D8 2H nuclei, respectively, and the A11 13CO nuclei are 2–3 and 7–8 Å from the membrane center, respectively. One reason for multiple KALP locations may be a hydrophobic KALP length of ∼26 Å (L4–L20), which is shorter than the hydrophobic length of the DPPC membrane (∼31 Å).[25,26] The KALP hydrophobic length could be increased by “snorkeling” of lysine side chains to the headgroup region, and the different KALP locations may be correlated with different snorkeling geometries.[27,28] The two experimentally determined locations in Figure 3C differ by an ∼5 Å translation along the membrane normal, which is comparable to the ∼8 Å range in molecular dynamics simulations for transmembrane helices.[8,29] KALP likely has additional membrane locations because the sum AD8 + AD10 < 1.

For either HFP_F8C or HFP_G5C, there are very different values of AD10 and AD8, and for HFP_G5C, rD8 ≈ rD10 ≈ 4.5 Å. As with KALP, these trends support two distinct membrane locations of HFP. The larger AD10 values are attributed to a major HFP population with deep membrane insertion and HFP contact with D10 2H nuclei, and the smaller AD8 values are attributed to a minor population with shallow membrane insertion and HFP contact with D8 2H nuclei (Figure 4E,F). The major:minor population ratio is ∼7:3 as calculated from the τ = 48 ms (ΔS/S0)D10-to-(ΔS/S0)D8 ratio for either the HFP_F8C or HFP_G5C samples. The 7:3 ratio is also supported by the ΔS/S0 values at large τ values for HFP_G5C samples prepared with membranes with different deuterated cholesterols (Figure S7 of the Supporting Information). The AD10/AD8 ratios are ∼4 for HFP_F8C and ∼2 for HFP_G5C. However, there are ∼30% uncertainties in the best-fit A and γ values for the HFP_F8C (D10) sample, which are much larger than for other samples. The reason for these larger uncertainties is not well-understood, but it might be related to the use of neutral membrane in this sample rather than a membrane with a 0.2 fraction of negatively charged lipid as was used for the HFP_G5C samples. Relative to a neutral membrane, there is higher and more reproducible binding of the positively charged peptide to the negatively charged membrane, probably because of the contribution of the attractive electrostatic energy.

For HFP_G5C, the relationship AD8 + AD10 > 1 may mean that some G5 13CO nuclei contact both D8 and D10 2H nuclei. There is negligible HFP localized to the membrane surface, as evidenced by ΔS/S0 ≈ 0 for D4 samples (Figure S4 of the Supporting Information). The multiple membrane locations of HFP are attributed to the distribution of antiparallel β-sheet registries.[20] Specifically, the membrane insertion depth of a HFP registry likely depends on the lengths of its contiguous hydrophobic regions, and these lengths vary among registries. Deep and shallow insertions may also have a distribution of membrane locations of HFP. The predominant deep insertion of HFP could significantly perturb the membrane bilayer and lower the activation energy of membrane fusion. This is consistent with the observed strong positive correlation between membrane insertion depth and fusion rate for several HFP constructs.[1] Models in panels E and F of Figure 4 show insertion of the antiparallel β-sheet into a single leaflet rather than membrane traversal analogous to that of bacterial porins. The major deep insertion into a single leaflet is more consistent with the observed close contact of multiple HFP residues with the D10 2H’s near the center of the membrane. Insertion into a single leaflet is also likely for the intermolecular β-sheet formed from multiple gp41 proteins during HIV–cell fusion.[30] This topology allows the other parts of all the ectodomains (∼160 residues per gp41) to be on the same face of the membrane.

Discussion

13CO–2H REDOR SSNMR reveals multiple locations within the hydrocarbon core of a gel-phase membrane for both the monomeric α-helical KALP peptide and the oligomeric β-sheet HFP peptide. The KALP locations are attributed to hydrophobic mismatch and consequent snorkeling of lysine side chains. The HFP locations are attributed to the distribution of antiparallel β-sheet registries. We consider whether multiple locations in gel-phase membranes that are derived from our data reflect multiple locations for peptides and proteins in cell membranes. One general difference between model and cell membranes is the higher ∼1:1 (w/w) protein:lipid ratio of a typical cell membrane. The cell membrane also contains a mixture of distinct lipids and sterols specific to the cell type. HFP should be considered in the context of the membrane composition of host cells of HIV with a ∼1:2 cholesterol:lipid mole ratio and ∼50% PC and ∼10% negatively charged lipid.[21] This composition is likely liquid-ordered phase, and in such membranes with significant cholesterol, the HFP HIV gp41 has β-sheet structure with a distribution of antiparallel registries.[20,31] There may be multiple locations for HFP in the cell membrane because of this distribution. KALP is a designed sequence, and there may be multiple membrane locations in higher-temperature single-component fluid-phase membranes because of hydrophobic mismatch. We note a recent study proposing two locations for a peptide in a fluid-phase membrane based on fluorescence and molecular dynamics simulations.[8]

We also consider future application of the protein 13C membrane 2H REDOR approach to probe the membrane locations of proteins. Experiments in liquid-ordered membranes containing cholesterol should be straightforward and could include detection of contacts between specific protein residues and specific cholesterol regions (Figure S7 of the Supporting Information). Experiments in fluid-phase membranes lacking cholesterol may be challenging because of motion by both the protein and the lipid. Such motion may result in reduced 13CO–2H dipolar coupling and a corresponding ΔS/S0 buildup rate via averaging of ⟨3 cos2 θ – 1⟩, where θ is the angle between the 13CO–2H internuclear vector and the external magnetic field. Even with motion, it should still be possible to distinguish the relative proximity of a residue to the D4, D8, and D10 deuterated regions, and a model transmembrane peptide like KALP could be used to validate the approach. 13CO–2H distances could be semiquantitatively derived using ⟨3 cos2 θ – 1⟩ values that are estimated from lipid order parameters and/or molecular dynamics simulations. The dipolar coupling and corresponding ΔS/S0 buildup typically have an only minor dependence on averaging of ⟨r–3⟩. One practical change in the pulse sequence for obtaining higher signals for samples with motion may be use of an initial 13C π/2 pulse rather than 1H → 13C cross-polarization.

Although our experiments were conducted with selectively labeled protein to allow unambiguous resonance assignment of one-dimensional spectra, the approach also can be applied to proteins with extensive 13C and 15N labeling with the caveat that the 13C and 15N line widths are narrow enough to allow resonance assignment via multidimensional SSNMR.[32] These experiments were also conducted with deuterated DPPC lipid because of the commercial availability of palmitic acid with different deuteration patterns. Future experiments could be conducted with more common biological lipids that contain a palmitoyl and an oleoyl chain. Experiments can also be conducted with deuterated cholesterol (Figure S7 of the Supporting Information).