Solid-State Nuclear Magnetic Resonance Measurements of HIV Fusion Peptide to Lipid Distances Reveal Intimate Contact of β Strand Peptide with Membranes and Close Proximity of the Ala-14 to Gly-16 Region with Lipid Headgroups

Abstract

Human immunodeficiency virus (HIV) infection begins with fusion between viral and host cell membranes and is catalyzed by the HIV gp41 fusion protein. The ~20 N-terminal apolar residues of gp41 are called the HIV fusion peptide (HFP), interact with the host cell membrane, and play a key role in fusion. In this study, the membrane location of peptides which contained the HFP sequence AVGIGALFLGFLGAAGSTMGARS was probed in samples containing either only phospholipids or phospholipids and cholesterol. Four HFPs were examined which each contained ¹³CO labeling at three sequential residues between G5 and G16. The ¹³CO chemical shifts indicated that HFP had predominant β strand conformation over the labeled residues in the samples. The internuclear distances between the HFP ¹³COs and the lipid ³¹Ps were measured using solid-state nuclear magnetic resonance rotational-echo double resonance experiments. The closest ¹³CO-³¹P distances of 5–6 Å were observed for HFP labeled between A14 and G16 and correlated with intimate association of β strand HFP and membranes. These results were confirmed with measurements using HFPs singly ¹³CO labeled at A6 or A14. To our knowledge, these data are the first measurements of distances between HIV fusion peptide nuclei and lipid P and qualitative models of membrane location of oligomeric β strand HFP are presented which are consistent with the experimental data. Observation of intimate contact between β strand HFP and membranes provides rationale for further investigation of the relationship between structure and fusion activity for this conformation.

The infection of enveloped viruses such as human immunodeficiency virus (HIV) begins with fusion between the viral and host cell membranes (1–4). Fusion may be catalyzed by fusion proteins and several models of fusion protein catalysis have been proposed (2, 5–7). For HIV, fusion is catalyzed by a “gp160” glycoprotein complex which is incorporated in the virus membrane and is composed of two non-covalently associated subunits “gp120” and “gp41”. The gp120 subunit lies outside the virus and binds to receptors in the target cell membrane and the gp41 subunit contains a region inside HIV as well as a single-pass transmembrane domain (8, 9). The ~170- residue ectodomain of gp41 lies outside HIV and is subdivided into a more C-terminal “soluble ectodomain” and a ~20-residue N-terminal fusion peptide (HFP) which is apolar and fairly conserved. The HFP is believed to interact with the target cell membrane after gp120 binds to cellular receptors and fusion is greatly disrupted by mutation or deletion of the HFP (10–13).

Peptides with the HFP sequence catalyze vesicle fusion and there are good correlations between the mutation/fusion activity relationships of HFP-induced vesicle fusion and HIV/target cell fusion (14–17). Studies of membrane-associated HFP should therefore provide useful information about some aspects of biological fusion. Both the conformation and membrane location of the HFP have been hypothesized to be significant structural factors for fusion catalysis by the HFP (16, 18). The conformation of the HFP has been investigated in detergent micelles and membranes using a variety of biophysical techniques. For HFP associated with negatively charged sodium dodecyl sulfate micelles, one liquid-state nuclear magnetic resonance (NMR) study showed that there was uninterrupted α helical structure from I4 to M19 while another study showed a helix from I4 to A14 followed by a β turn (19, 20). For HFP associated with neutral dodecylphosphocholine micelles, helical structure was detected from I4 to L12 (21, C. M. Gabrys and D. P. Weliky, unpublished data). There is not yet a consensus for the micelle location of HFP and there are distinct models based on experiment and simulation of either predominant micelle surface location or micelle traversal by HFP (19–23). In one NMR study, residues I4 to A15 were found to be fully shielded from solvent and residues G3 and G16 were at the micelle/solvent interface (20).

The conformation of membrane-associated HFP has been investigated with different lipid components and different peptide:lipid ratios. A greater fraction of HFPs adopted helical structure at low peptide:lipid while non-helical structure became more favored at higher ratios (24). Helical structure was also promoted by negatively charged lipids while a higher fraction of β strand structure was adopted with neutral lipids or with bound Ca²⁺ (16, 24–26). Solid-state NMR provided residue-specific conformational information of HFP associated with membranes whose lipid headgroup and cholesterol composition was comparable to that of host cells of the virus. A β strand conformation was observed for residues A1 to G16 while A21 appeared to be unstructured (27, 28). Formation of β strand oligomers or aggregates was supported by detection of short distances between labeled ¹³COs on one HFP and labeled ¹⁵N on an adjacent HFP (29). Oligomerization/aggregation has also been detected by other biophysical methods (15, 30). There is evidence that at least the lipid mixing step of membrane fusion can occur with the HFP in either helical or β strand conformation although there is some controversy in the literature about this conclusion (14, 16, 18, 31–35).

HFP location in membranes has been primarily probed using a HFP-F8W mutant and by variation of the tryptophan fluorescence of this mutant with changes in environment (36, 37). Key results have included: (1) fluorescence was higher for membrane-associated HFP-F8W than for HFP-F8W in buffered saline solution; (2) greater fluorescence quenching by acrylamide was observed for a soluble tryptophan analog than for membrane-associated HFP-F8W; and (3) similar fluorescence quenching of membrane-associated HFP-F8W was observed in samples containing either 1-palmitoyl-2-stearoyl-phosphocholine brominated at the 6, 7 carbons of the stearoyl chain or the corresponding lipid brominated at the 11, 12 carbons of the chain. The first two results indicated that solvent exposure of the HFP-F8W tryptophan is reduced with membrane association and the third result indicated that the membrane location of the tryptophan indole group is centered near the carbon 9 position of the brominated lipid stearoyl chain; i.e. ~8.5 Å from the bilayer center and ~10 Å from the lipid phosphorus. Infrared (IR) and solid-state NMR spectra of membrane-associated HFP suggested that the HFP-F8W had predominant β strand conformation under the conditions of the fluorescence experiments (16, 27, 36, 37).

In a different set of experiments, electron spin resonance spectra showed that chromium oxalate in the aqueous phase quenched the signal of membrane-associated HFP which was spin-labeled at M19 but did not quench HFP spin-labeled at A1 (30). These data indicated a M19 location close to the aqueous interface of the membrane and an A1 location away from this interface.

Models for HFP location in membranes have also been developed from simulations of a single HFP molecule in membranes and have shown either partial insertion or traversal of the membrane. The HFP always adopted predominant α helical conformation and in one simulation was generally near the membrane surface with the F8 backbone and sidechain nuclei respectively 4 Å and 6 Å deeper than the phosphorus longitude (38). For a different simulation, HFP traversed the membrane and the backbone and sidechain F8 nuclei were at the bilayer center, i.e. ~19 Å from the phosphorus longitude (39).

This paper includes solid-state NMR measurements of distances between ¹³C labeled carbonyl (¹³CO) nuclei in HFP and lipid ³¹P nuclei. These studies provide information about the location of specific HFP residues relative to the phosphorus headgroups and are complementary to other solid-state NMR methods to probe membrane location of peptides and proteins (40–50). The ¹³CO-³¹P distance approach has previously been used to probe the locations of antimicrobial peptides, antibiotics, and sterols in membranes (51–53). Measurements were made both on HFP associated with membranes containing only phospholipids and on HFP associated with membranes which contained both phospholipids and cholesterol. The potential significance of cholesterol-containing membranes is suggested both by the cholesterol:phospholipid mol ratios of ~0.5 and 0.8 for HIV host cell and HIV membranes, respectively, and by the observation that β strand conformation of HFP is promoted by membrane cholesterol (27, 35, 54–57).

The ¹³CO-³¹P distances (r) were probed with the rotational-echo double resonance (REDOR) technique which is a solid-state NMR method to measure magnitudes of dipolar couplings (d) between spin ½ heteronuclei such as ¹³C and ³¹P (58). For a ¹³CO-³¹P spin pair, r = 23.05/d^1/3 where r and d are in units of Å and Hz, respectively. The upper limit of REDOR detection of r is ~10 Å (d ≥ 10 Hz) (35).

MATERIALS AND METHODS

Materials

Resins and 9-fluorenylmethoxycarbonyl (FMOC) amino acids were obtained from Peptides International (Louisville, KY). ¹³CO-isotopically labeled amino acids were obtained from Cambridge Isotope Laboratories (Andover, MA). The lipids 1,2-di-Otetradecyl-sn-glycero-3-phosphocholine (DTPC), 1,2-di-O-tetradecyl-sn-glycero-3-[phospho-rac-(1-glycerol)] (DTPG), and [1-¹³C]-1,2-dipalmitoyl-sn-glycero-3-phosphocholine (DPPC-¹³C) were obtained from Avanti Polar Lipids (Alabaster, AL). The 5 mM HEPES buffer at pH 7 contained 0.01% (w/v) NaN₃ preservative.

Peptides

All peptides contained the sequence AVGIGALFLGFLGAAGSTMGARS which is the 23 N-terminal residues of HIV-1 gp41, LAV_1a strain. A set of peptides were synthesized to probe peptide/lipid headgroup distances between G5 and G16. “HFP1-⁸FLG” had the sequence AVGIGALFLGFLGAAGSTMGARS-NH₂ and was ¹³CO labeled at F8, L9, and G10. “HFP2-⁵GAL”, “HFP2-¹¹FLG”, and “HFP2-¹⁴AAG” had the sequence AVGIGALFLGFLGAAGSTMGARSKKK-NH₂ and were ¹³CO labeled at G5, A6, L7; F11, L12, G13; or A14, A15, G16, respectively. The three non-native lysines increased aqueous solubility and resulted in monomeric peptide in the buffer solution prior to membrane binding (59). “HFP3-⁸FLG” had sequence AVGIGALFLGFLGAAGSTMGARSKKKA_β and was ¹³CO labeled at F8, L9, and G10 and “HFP4”, “HFP4-⁶A”, and “HFP4-¹⁴A” had sequence AVGIGALFLGFLGAAGSTMGARSWKKKKKKA_β and were unlabeled or ¹³CO labeled at A6 or A14, respectively. The β-alanine resin used for the HFP3 and HFP4 syntheses had a low substitution which helped to increase yield. All peptides were synthesized using an ABI 431A peptide synthesizer (Foster City, CA) and FMOC chemistry. Peptides were cleaved from the resin for 2–3 hours using either a mixture of trifluoroacetic acid (TFA):water:phenol:thioanisole:ethanedithiol:water in a 33:2:2:2:1 volume ratio or a mixture of TFA:thioanisole:ethanedithiol:anisole in a 90:5:3:2 volume ratio. TFA was removed from the cleavage filtrate with nitrogen gas and peptides were precipitated with cold t-butyl methyl ether. Peptides were purified by reversed-phased high performance liquid chromatography using a semi-preparative C₁₈ column and a water-acetonitrile gradient containing 0.1% TFA. Mass spectroscopy was used for peptide identification.

NMR sample preparation

Samples were made with the ether-linked lipids DTPC and DTPG because these are commercially available lipids which do not contain carbonyl groups. The liquid-crystalline to gel phase transition temperatures of DTPC and DTPG are ~28 C and are close to the ~23 C phase transition temperatures of the corollary ester-linked lipids 1,2-dimyristoyl-sn-glycero-3-phosphocholine and 1,2-dimyristoyl-sn-glycero-3-[phospho-rac-(1-glycerol)] (60). If NMR samples had been made with the more typical ester-linked lipids, there would be large natural abundance lipid ¹³CO signals which would overlap with the peptide ¹³CO signals (35). For such samples, analysis of the peptide ¹³CO-lipid ³¹P distances would then be complicated by the close proximity of the lipid ¹³CO to lipid ³¹P. All samples contained a 4:1 DTPC:DTPG mol ratio which reflected the approximate ratio of neutral:negatively charged headgroups in membranes of host cells of HIV (54, 57). For each triply ¹³CO-labeled HFP, a sample was made with DTPC:DTPG = 4:1 ≡ “PC:PG” and a sample was made with DTPC:DTPG:cholesterol = 8:2:5 ≡ “PC:PG:CHOL”. The latter total lipid:cholesterol mol ratio reflected the ratio observed in membranes of host cells of HIV and the former PC:PG composition without cholesterol was similar to membrane compositions used in previous structural studies of fusion peptides (54, 57, 61–63). The samples containing singly ¹³CO-labeled HFP were prepared with PC:PG:CHOL.

Each sample preparation began with dissolution in chloroform of 20 µmol total PC:PG or 30 µmol total PC:PG:CHOL. The chloroform was removed under a stream of nitrogen followed by overnight vacuum pumping. The lipid film was suspended in 2 mL buffer and homogenized with ten freeze-thaw cycles. Large unilamellar vesicles were formed by extrusion through a 100 nm diameter polycarbonate filter (Avestin, Ottawa, ON). HFP (0.8 µmol by weight) was dissolved in 2 mL buffer and the HFP and vesicle solutions were then gently vortexed together. The mixture was refrigerated overnight and ultracentrifuged at ~150000g for five hours. The membrane pellet with associated bound HFP was transferred to a 4 mm diameter magic angle spinning (MAS) NMR rotor. The majority of the HFP binds to membranes under these conditions and the membranes remain bilayers for HFP:lipid ~ 0.04 (27, 56, 59, 64).

A sample was also prepared for calibration of the NMR experiments and contained HFP4 (0.8 µmol) and DPPC-¹³C (20 µmol). The ¹³CO signal of this sample was dominated by DPPC-¹³C.

Solid-state NMR

Experiments were done on a 9.4 T solid-state NMR spectrometer (Varian Infinity Plus, Palo Alto, CA) equipped with a triple resonance MAS probe. The detection channel was tuned to ¹³C at 100.8 MHz, the decoupling channel was tuned to ¹H at 400.8 MHz, and the third channel was tuned to ³¹P at 162.2 MHz. ¹³C shifts were externally referenced to the methylene resonance of adamantane at 40.5 ppm, ³¹P shifts were referenced to 85% H₃PO₄ at 0 ppm, and the ¹³C and ³¹P transmitter chemical shifts were 156 and −16 ppm, respectively. The ¹³C referencing allowed direct comparison with ¹³C shift databases derived from liquid-state NMR assignments of proteins (65, 66). These databases are appropriate for solid-state NMR data as evidenced by similar ¹³C shifts observed for the same protein in either aqueous solution or the microcrystalline state (67–69). experiments were done at −50 C to enhance ¹³C signal and to prevent motional averaging of the ¹³C-³¹P dipolar coupling which was the parameter used to assess HFP location in membranes. The ¹³C shifts and presumably the HFP conformation were comparable at −50 C and ambient temperature (70). At −50 C, the lipids were likely in the gel phase for the PC:PG samples and in the liquid-ordered phase for the PC:PG:CHOL samples (60, 71). Analyses of slow-spinning spectra yielded a ¹³CO chemical shift anisotropy range of ~90 to 240 ppm and a ³¹P chemical shift anisotropy range of ~ −75 to 100 ppm (72). The REDOR experiment contained in sequence: (1) a 50 kHz ¹H π/2 pulse; (2) 1 ms cross-polarization with 52 kHz ¹H field and 58–69 kHz ramped ¹³C field; (3) a dephasing period of duration τ which contained ~50 kHz ¹³C π and in some cases ~60 kHz ³¹P π pulses with XY-8 phase cycling on each channel; and (4) ¹³C detection with a four scan phase cycle (29, 35, 73–75). Two-pulse phase modulation ¹H decoupling of ~100 kHz was applied during the dephasing and detection periods, the recycle delay was 1 s, and the MAS frequency was 8000 ± 2 Hz (76).

For each sample and each τ, two spectra were acquired. The dephasing period during the “S₁” acquisition contained a ¹³C π pulse at the end of each rotor cycle except for the last cycle and a ³¹P π pulse in the middle of each cycle. The ³¹P pulses were absent during the “S₀” acquisition. MAS averaged the ¹³C-³¹P dipolar coupling to zero over each rotor cycle of the dephasing period of the S₀ acquisition, while incorporation of two π pulses per rotor cycle during the S₁ acquisition resulted in a non-zero average value of the dipolar coupling and concomitant reduction in signals of ¹³C nuclei close to ³¹P. Determination of d was based on the difference in ¹³C signal intensity of the two spectra.

The ¹H and ¹³C rf fields were initially calibrated with adamantane and the ¹³C cross-polarization field was then adjusted to give the maximum ¹³CO signal of the sample containing unlabeled HFP4 and DPPC-¹³C. The ³¹P π pulse length was set by minimization of the S₁ signal in this sample for τ = 8 ms and the ¹H TPPM pulse length was set to give the maximum S₀ signal.

REDOR data analysis

All spectra were processed with Gaussian line broadening and with baseline correction. For each sample and each value of τ, spectra were integrated over a defined chemical shift range and the integrals of the S₀ and S₁ spectra were denoted as “S₀” and “S₁” and were used to calculate a normalized experimental dephasing parameter (ΔS/S₀)^exp = 1 − (S₁/S₀). Uncertainties in (ΔS/S₀)^exp were calculated:

$${\sigma }^{\text{exp}}=\frac{\sqrt{({S_0}^2\times {{\sigma }_{S_1}}^2)+({S_0}^2\times {{\sigma }_{S_0}}^2)}}{{S_0}^2}$$ (1)

where σS₀ and σS₁ were the experimental root-mean-squared-deviations of integrated intensities in regions of the spectra without signal (77). relative to the other labeled HFP samples, the HFP2-¹⁴AAG and HFP4-¹⁴A samples had significantly larger values of (ΔS/S₀)^exp and data from these samples were used to determine an approximate distance between the ³¹Ps and the labeled ¹³COs. The distance determination was done with (ΔS/S₀)^lab calculated to remove the natural abundance (na) contribution from (ΔS/S₀)^exp:

$${\left(\frac{\Delta S}{S_0}\right)}^{\mathit{\text{lab}}}=\left[\left(1+\frac{S_0^{na}}{S_0^{\mathit{\text{lab}}}}\right)\times {\left(\frac{\mathrm{\Delta }S}{S_0}\right)}^{\mathit{\text{exp}}}\right]-\left[\left(\frac{S_0^{na}}{S_0^{\mathit{\text{lab}}}}\right)\times {\left(\frac{\mathrm{\Delta }S}{S_0}\right)}^{na}\right]$$ (2)

The values of (S₀ ^na/S₀^lab) were 0.084 and 0.32 for the HFP2-¹⁴AAG and HFP4-¹⁴A samples, respectively. The values of (ΔS/S₀)^na were calculated as the average of (ΔS/S₀)^exp for the HFP2-⁵GAL, HFP3-⁸FLG, and HFP2-¹¹FLG samples and for τ = 2, 8, 16, and 24 ms were 0.000, 0.029, 0.094, and 0.134 for the PC:PG samples and 0.026, 0.028, 0.083, and 0.092 for the PC:PG:CHOL samples. The σ^lab were calculated with σ^lab = (1 + S₀^na/S₀^lab) × σ^exp which neglected the contribution from the far-right term in Eq. 2. Discussion of this approximation and derivation of Eq. 2 are given in the Supporting Information.

Simulations of the experimental data were based on a single ¹³CO-³¹P spin pair model:

$${\left(\frac{\mathrm{\Delta }S}{S_0}\right)}^{\mathit{\text{sim}}}=1-{\left[J_0(\sqrt{2}\lambda )\right]}^2+\left\{2\times {\displaystyle \sum _{k=1}^5\frac{{[J_k(\sqrt{2}\lambda )]}^2}{16k^2-1}}\right\}$$ (3)

with λ = d × τ and J_k as the k^th order Bessel function of the first kind (78). The samples contained multiple ¹³CO-³¹P distances and couplings and these are approximated as a single r and a single d in Eq. 3 (29, 51, 79).

For the HFP4/DPPC-¹³C sample, χ²(d) were calculated for an array of values of d:

$${\chi }^2(d)={\displaystyle \sum _{i=1}^T\frac{{\left[{\left(\frac{\mathrm{\Delta }S}{S_0}\right)}_i^{\mathit{\text{exp}}}-{\left(\frac{\mathrm{\Delta }S}{S_0}(d)\right)}_i^{\mathit{\text{sim}}}\right]}^2}{{({\sigma }_i^{\mathit{\text{exp}}})}^2}}$$ (4)

where T was the number of experimental τ values. The best-fit d corresponded to minimum χ²(d).

At larger τ, the (ΔS/S₀)^lab for the HFP2-¹⁴AAG and HFP4-¹⁴A samples reached plateau values which were significantly smaller than 1 while the (ΔS/S₀)^sim had plateau values of ~1. This inconsistency was resolved using a model of two populations of membrane-associated HFPs. The “f ” fraction represented ¹³COs close to the lipid ³¹Ps with corresponding d ≠ 0 while the “1−f” fraction represented ¹³COs far from the lipid ³¹Ps with corresponding d = 0. Fitting was done with an array of values of d and f:

$${\chi }^2(d)={\displaystyle \sum _{i=1}^T\frac{{\left[{\left(\frac{\mathrm{\Delta }S}{S_0}\right)}_i^{\mathit{\text{lab}}}-\left\{f\times {\left(\frac{\mathrm{\Delta }S}{S_0}(d)\right)}_i^{\mathit{\text{sim}}}\right\}\right]}^2}{{({\sigma }_i^{\mathit{\text{lab}}})}^2}}$$ (5)

The uncertainty of d was calculated with the χ² = χ²_min + 1 criterion (77).

RESULTS

Overall strategy

Our long-term goal is a detailed structure of the membrane location of the HFP in helical and β strand conformations. Prior to beginning this study, there was relatively little information about the membrane location of HFP, particularly for the β strand conformation. It was likely that a large number of ¹³CO sites had ¹³CO-³¹P distances beyond the REDOR detection limit. In addition, HFP ¹³C linewidths are fairly broad which leads to overlap of ¹³CO resonances from different residues and the need for specific ¹³CO labeling. In an effort to reduce the numbers of specifically labeled peptides needed to develop a membrane location model, samples were first made with four peptides each of which had ¹³CO labels at three sequential residues between G5 and G16. The G5–G16 region was therefore rapidly scanned for ¹³CO-³¹P proximity. Although the (ΔS/S₀)^exp data for each of the samples had contributions from three distinct ¹³CO sites, the individual (ΔS/S₀) would only be appreciably greater than zero for ¹³CO-³¹P distances ≤ 8 Å. The regions of HFP close to ³¹P would be defined from the REDOR data on the triply labeled samples and these regions would then provide a basis for choosing sites for single ¹³CO labeled peptides. ¹³CO-³¹P distances are more straightforwardly derived from (ΔS/S₀) of singly labeled HFPs and REDOR data for two such HFPs are presented to refine the basic HFP membrane location model developed from the triply labeled HFP data.

REDOR calibration experiments

As an initial control experiment, ¹³CO-³¹P REDOR spectra were obtained for HFP2-¹¹FLG lyophilized from water and resulted in (ΔS/S₀)^exp ~ 0 for values of τ between 1 and 19 ms, Fig. 1a–c. Non-zero values of (ΔS/S₀)^exp for ¹³COs in membrane-associated HFP samples can therefore be definitively ascribed to ¹³CO-³¹P proximity. As displayed in Fig. 1d–g, REDOR spectra were also obtained for the HFP4/DPPC-¹³C sample. Because HFP4 is unlabeled and DPPC-¹³C is labeled, the signals were primarily due to the DPPC ¹³COs and have non-zero (ΔS/S₀)^exp because of the proximity of the headgroup ³¹P. Fig. 2 displays (ΔS/S₀)^exp and best-fit (ΔS/S₀)^sim for this sample and yielded d = 68 Hz and r = 5.6 Å. The best-fit NMR value of r is comparable to the 5–6 Å values of r observed in the crystal structures of the related lipids, 1,2-dimyristoyl-sn-glycero-3-phosphocholine and 1,2-dipalmitoyl-sn-glycero-[phospho-rac-(1-glycerol)] (which had both been dehydrated) and in molecular dynamics simulations of gel-phase DPPC (80–82). The differences between (ΔS/S₀)^exp and (ΔS/S₀)^sim are likely due to: (1) contributions to (ΔS/S₀)^exp from intra- and intermolecular ³¹P with comparable values of r which contrasts with the single ¹³CO-³¹P spin pair model used to calculate (ΔS/S₀)^sim; (2) two structurally distinct ¹³COs in each headgroup with different intra- and intermolecular r values; and (3) structural disorder within the headgroups (79). Overall, the DPPC-¹³C fitting yielded good agreement between the NMR r value and the expected range of r values in the lipid.

Figure 1.

¹³CO-³¹P REDOR spectra of (a–c) lyophilized HFP2-¹¹FLG and (d–g) the HFP4:DPPC-¹³C sample. The left and right spectra in each pair of ¹³C-detected spectra are S₀ and S₁ spectra, respectively. The dotted lines are drawn for visual comparison of S₀ and S₁ peak intensities. Each spectrum was processed with 200 Hz Gaussian line broadening and baseline correction. The τ values and numbers of S₀ or S₁ scans in each pair of spectra were: a, 1 ms, 120; b, 11 ms, 126; c, 19 ms, 138; d, 1 ms, 8; e, 7 ms, 56; f, 13 ms, 104; g, 19 ms, 152.

Figure 2.

(ΔS/S₀)^exp (vertical lines with error bars) and best-fit (ΔS/S₀)^sim (diamonds) vs dephasing time (τ) for the HFP4/DPPC-1-¹³C sample. Lines are drawn between points with adjacent values of τ. Each (ΔS/S₀)^exp valuewas obtained from a 1 ppm integration region centered at 173 ppm. The total (S₀ + S₁) numbers of scans for τ = 1, 3, 5, 7, 9, 11, 13, 15, 17 and 19 ms were 16, 48, 80, 112, 144, 176, 208, 240, 272 and 304, respectively. The displayed best-fit (ΔS/S₀)^sim values corresponded to d = 68 Hz and r = 5.6 Å.

For the HFP4/DPPC-¹³C sample, experiments were also done with a “one-channel” version of the REDOR sequence for which the S₁ acquisitions contained a single ¹³C π pulse at the center of the dephasing period and ³¹P π pulses in the middle and end of each rotor cycle except for the center and end of the dephasing period. The S₀ acquisition did not have ³¹P π pulses. relative to the “two-channel” version of REDOR described in Materials and Methods, one-channel REDOR has a reduced number of ¹³C π pulses which could result in reduced ¹³C-¹³C dipolar coupling and larger overall signals (83, 84). In fact, the experimental S₀ intensities were comparable for the two versions of REDOR while (ΔS/S₀)^exp for one-channel REDOR was ~2/3 that of two-channel REDOR (29). All subsequent experiments were done with two-channel REDOR.

Triply labeled HFP

Local peptide conformation was examined by analysis of the ¹³CO chemical shift distributions in S₀ spectra of HFPs obtained with τ = 2 ms, cf. Fig. 3. The data supported the following models: (1) the major fraction of peptides in PC:PG and PC:PG:CHOL adopted a β strand conformation from G5 to G16; and (2) there is a minor fraction of peptides in PC:PG with helical conformation. The detailed experimental support for the models is based on the known correlation between larger ¹³CO chemical shifts and local helical conformation and smaller ¹³CO chemical shifts and local β strand conformation. For example, average database values in ppm units of ¹³CO chemical shifts of helix (strand) conformations are: Gly, 175.5 (172.6); Ala, 179.4 (176.1); Leu, 178.5 (175.7); and Phe, 177.1 (174.2) (66). For the HFP2-⁵GAL, HFP3-⁸FLG, HFP2-¹¹FLG, and HFP2-¹⁴AAG samples, the peak chemical shifts were ~175, 174, 175, and 176 ppm, respectively, and correlated with β strand conformation for the Ala, Leu, and Phe residues. For the HFP3-⁸FLG and HFP2-¹⁴AAG samples associated with PC:PG, there were shoulders at ~178 and 179 ppm, respectively, which correlated with helical conformation of Ala, Leu, and Phe residues. These results were consistent with previous studies of the conformation of membrane-associated HFP with peptide:lipid ~ 0.04 and with previous observations of greater preference for β strand conformation in cholesterol-containing membranes (27, 28, 35, 55, 56, 59, 70).

Figure 3.

S₀ spectra for membrane-associated HFP with peptide:lipid ~0.04. The dotted lines are at 175 ppm. All spectra were obtained with τ = 2 ms and were processed with 200 Hz Gaussian line broadening and baseline correction. The membrane composition for samples a–d was PC:PG and the membrane composition for samples e–h was PC:PG:CHOL. The peptides were: a, e, HFP2-⁵GAL; b, f, HFP3-⁸FLG; c, g, HFP2-¹¹FLG; and d, h HFP2-¹⁴AAG. The numbers of scans summed to obtain spectra a-h were 4823, 3867, 4823, 8500, 3259, 1001, 4320 and 6992, respectively.

Fig. 4 displays the τ = 16 and 24 ms REDOR spectra of triply-labeled membrane-associated HFP samples and Fig. 5a,b displays comparative plots of (ΔS/S₀)^exp for the different samples. The data demonstrated that samples containing HFP2-¹⁴AAG have qualitatively larger (ΔS/S₀)^exp than do samples containing HFP labeled at other residues. Using the conformational results from Fig. 3, it appears: (1) a significant fraction of β strand HFP are in close contact with membranes; and (2) the ¹⁴AAG (A14 to G16) region is closer to the lipid ³¹P than is the ⁵GALFLGFLG (G5 to G13) region. Fig. 5 c,d displays plots of (ΔS/S₀)^lab and best-fit (ΔS/S₀)^sim for HFP2-¹⁴AAG in PC:PG and PC:PG:CHOL. The best-fit r was ~5.2 Å in both membrane compositions and the best-fit f in PC:PG and PC:PG:CHOL were 0.45 and 0.32, respectively. It was not possible to fit the HFP2-¹⁴AAG data well without inclusion of the f parameter. Although the (ΔS/S₀)^lab had contributions from three ¹³CO sites which would each have a distinct r, the number of data points and signal-to-noise dictated fitting to a single r value. The best-fit r should therefore be considered as both approximate and as likely representing the population of ¹³CO sites with greatest d and smallest r. Fitting was not done for data from the other samples because of the small (ΔS/S₀)^exp and because the (ΔS/S₀)^exp do not always reach asymptotic values at large τ.

Figure 4.

¹³CO-³¹P REDOR spectra of membrane-associated HFP with peptide:lipid ~0.04. Each letter corresponds to a single sample which contained (a–d) PC:PG or (e–h) PC:PG:CHOL and (a, e) HFP2-⁵GAL, (b, f) HFP3-⁸FLG, (c, g) HFP2-¹¹FLG, or (d, h) HFP2-¹⁴AAG. For each letter/sample, S₀ (left), S₁ (right), τ= 16 ms (top), and τ= 24 ms (bottom) spectra are displayed. The dotted lines are drawn for visual comparison of S₀ and S₁ peak intensities. Each spectrum was processed with 300 Hz Gaussian line broadening and baseline correction. The numbers of S₀ or S₁ scans summed to obtain the top and bottom spectra were respectively: a, 30000, 56000; b, 27509, 29463; c, 20000, 40000; d, 44129, 48296; e, 8448, 52384; f, 5488, 21664; g, 28032, 52384; and h, 22576, 50240.

Figure 5.

(ΔS/S₀) vs dephasing time for membrane-associated HFP in (a,c) PC:PG or (b,d) PC:PG:CHOL. For panels a and b, the points correspond to (ΔS/S₀)^exp and the symbol legend is: squares, HFP2-⁵GAL; circles, HFP3-⁸FLG; up triangles, HFP2-¹¹FLG; and diamonds, HFP2-¹⁴AAG. The vertical dimensions of each symbol approximately correspond to the ±1 σ uncertainty limits. Lines are drawn between (ΔS/S₀)^exp values with adjacent values of τ. Each (ΔS/S₀)^exp value was determined by integration of 10 ppm regions of the S₀ and S₁ spectra. Panels c and d respectively correspond to the HFP2-¹⁴AAG/PC:PG and the HFP2-¹⁴AAG/PC:PG:CHOL samples and the points correspond to (ΔS/S₀)^lab (vertical lines with error bars) and best-fit (ΔS/S₀)^sim (down triangles). Lines are drawn between points with adjacent τ values. For plot c, the best-fit d = 91 ± 8 Hz with corresponding r = 5.12 ± 0.16 Å, f = 0.45 ± 0.02, and χ²_min = 5.0. For plot d, the best-fit d = 85 ± 6 Hz with corresponding r = 5.24 ± 0.13 Å, f = 0.32 ± 0.02, and χ²_min = 3.8.

Spectra were also obtained for samples made with HFP1-⁸FLG, the peptide which did not contain C-terminal lysines. For τ = 2, 8, 16, and 24 ms, (ΔS/S₀)^exp = −0.02, 0.06, 0.11, and 0.08 for the HFP1-⁸FLG/PC:PG sample and 0.01, 0.03, 0.01, and −0.01 for the HFP1-⁸FLG/PC:PG:CHOL sample. These values correlated with the (ΔS/S₀)^exp of the respective HFP3-⁸FLG/PC:PG and HFP3-⁸FLG/PC:PG:CHOL samples (circles in Fig. 5) and suggested that the additional C-terminal lysines of HFP2 and HFP3 do not greatly affect the REDOR results.

Singly labeled HFP

The triply labeled HFP results motivated experiments on singly labeled HFP4-⁶A and HFP4-¹⁴A associated with PC:PG:CHOL. As discussed in the Overall strategy subsection, analysis of singly labeled HFP data should result in more quantitative assessment of r. In addition, we were interested in studying HFP in a single conformation and the Fig. 3 data suggested that this was best achieved with PC:PG:CHOL and the resulting predominant β strand conformation. Finally, the Fig. 5 data suggested that (ΔS/S₀)^lab would be about zero for a HFP4-⁶A/PC:PG:CHOL sample and could be nonzero for a HFP4-¹⁴A/PC:PG:CHOL sample. REDOR data from these two sites could therefore further test the qualitative membrane location model derived from the triply labeled HFP results.

Fig. 6a,b displays the respective S₀ spectra at τ = 8 ms for HFP4-⁶A and HFP4-¹⁴A associated with PC:PG:CHOL. Single peaks were observed with peak shifts of ~175 ppm which correlated with β strand conformation at these residues. The spectrum of HFP4-¹⁴A is similar to a difference spectrum representing the Ala-14 ¹³CO signal for HFP (with no lysines) associated with an ester-linked lipid and cholesterol composition close to that of host cells of HIV (27). Fig. 6c,d displays the τ = 16 and 24 ms REDOR spectra of the singly labeled HFP4 samples and Fig. 7 shows (ΔS/S₀)^exp plots and data fitting for the HFP4-¹⁴A data. At large τ, (ΔS/S₀)^exp ≈ 0 for the HFP4-⁶A sample and (ΔS/S₀)^exp were significantly greater than zero for the HFP4-¹⁴A sample. Fitting of the HFP4-¹⁴A data with a single ¹³CO-³¹P spin pair model yielded best-fit r and f of 5.1 Å and 0.29, respectively. Thus, there was general consistency between the REDOR data of the HFP4-⁶A/PC:PG:CHOL and the HFP2-⁵GAL/PC:PG:CHOL samples and the REDOR data and fitting of the HFP4-¹⁴A/PC:PG:CHOL and the HFP2-¹⁴AAG/PC:PG:CHOL samples.

Figure 6.

¹³CO-³¹P REDOR spectra of (a,c) HFP4-⁶A/PC:PG:CHOL and (b,d) HFP4-¹⁴A/PC:PG:CHOL samples with peptide:lipid ~0.04. Panels a and b are S₀ spectra obtained with τ = 8 ms and were processed with 200 Hz Guassian line broadening and baseline correction. Panels c and d display S₀ (left), S₁ (right), τ= 16 ms (top), and τ= 24 ms (bottom) spectra. The dotted lines are drawn for visual comparison of S₀ and S₁ peak intensities. Each spectrum was processed with 300 Hz Gaussian line broadening and baseline correction. The numbers of S₀ or S₁ scans summed were: a, 2304; b, 3680; c, 5504 (top), 28288 (bottom); and d, 5120 (top), 28736 (bottom).

Figure 7.

(ΔS/S₀) vs dephasing time for HFP4-⁶A/PC:PG:CHOL and HFP-¹⁴A/PC:PG:CHOL samples. In panel a, (ΔS/S₀)^exp points with error bars are displayed with legend: HFP4-⁶A, squares; and HFP4-¹⁴A, diamonds. Each (ΔS/S₀)^exp value was determined from integrals of the entire S₀ and S₁ peaks. Panel b represents analysis of the HFP4-¹⁴A data with legend: (ΔS/S₀)^lab, vertical lines with error bars; and best-fit (ΔS/S₀)^sim, down triangles. Lines are drawn between points with adjacent τ values. The best-fit parameters were d = 93 ± 10 Hz with corresponding r = 5.08 ± 0.19 Å, f = 0.29 ± 0.02, and χ²_min = 0.1.

DISCUSSION

Insertion models

The position of the HFP in the membrane has been postulated to be a significant structural factor in its fusion activity and to our knowledge, this study is the first example of direct distance measurements between the HFP and the lipid headgroups. Values of r ~5–6 Å were detected between the ¹³COs of residues from A14 to G16 and the lipid ³¹Ps. These r values support intimate association of the HFP and membranes containing either only phospholipids or phospholipids and cholesterol. The average r for ⁵GALFLGFLG ¹³COs was likely greater than 8 Å (d ≤ 25 Hz) as evidenced by the significantly smaller (ΔS/S₀)^exp, cf. Fig. 5, Fig 7. Thus, relative to the ⁵GALFLGFLG residues, the ¹⁴AAG residues are much closer to the lipid ³¹P.

The G5 to G16 ¹³CO chemical shift distributions of this study were consistent with a major population of HFP with β strand conformation for these residues. This result correlated with previous studies which supported the following structural features: (1) β strand HFP was fully extended between A1 and G16; (2) β strand HFP formed hydrogen bonded oligomers or aggregates; and (3) a significant fraction of the oligomers have an antiparallel arrangement with adjacent strand crossing between F8 and L9 (25, 27–29, 35, 37, 85, 86). Some of these studies also supported conformational disorder at A21 (27, 28). Although there are some data supporting a population of parallel strand arrangement, “partial membrane insertion (PMI)” and “full membrane insertion (FMI)” models are only presented for the antiparallel arrangement, cf. Fig. 8 (29, 35, 87). There have been high-resolution structures for the ~130-residue “soluble ectodomain” region of gp41 which begins about ten residues C-terminal of the HFP and ends about twenty residues N-terminal of the gp41 transmembrane domain (88–92). These structures showed trimeric gp41 with the residues closest to the HFPs in a parallel in-register coiled-coil. Antiparallel HFP strand arrangement in the context of gp41 would then require at least two gp41 trimers. Strands from trimer A (A₁, A₂, A₃) would be parallel to one another and strands from trimer B (B₁, B₂, B₃) would be parallel to one another and an antiparallel interleaved strand arrangement could be formed as A₁B₃A₂B₂A₃B₁. There is solid-state NMR evidence for the antiparallel arrangement of membrane-associated HFPs which were cross-linked at their C-termini (35).

Figure 8.

(a) Partial membrane insertion (PMI) and (b, c) full membrane insertion (FMI) models for antiparallel β strand HFP. The red arrows represent the A1 to G16 residues in strand conformation and the black lines represent the S17 to S23 residues in random coil conformations. For clarity, black lines are not displayed in c. Lipids are represented in blue and grey and cholesterol is not displayed. Three antiparallel strands are displayed in a, b and twelve strands are displayed in c but the actual number of strands in the oligomer/aggregate is not known. The curvature and angle of the strands with respect to the bilayer normal are not known but the models consider that A1-G16 has ~55 Å length and that the transbilayer distance is ~48 Å (100). The experiments do not provide information about the membrane locations of residues S17 to S23. relative to FMI model (b), the FMI β barrel variant (c) could have reduced energy because all of the residues in the membrane interior have backbone hydrogen bonds.

For antiparallel strands between A1 and G16, the ¹⁴AAG residues in both the PMI and FMI models are at the ends of the hydrogen-bonded oligomer and are closer to the lipid headgroups than residues ⁵GALFLGFLG. The F8 and L9 residues are at the center of the hydrogen-bonded oligomer and are most deeply membrane-inserted in all models. This result is consistent with the smallest (ΔS/S₀)exp values observed for the HFP1-⁸FLG and HFP3-⁸FLG samples and with the large number of apolar sidechains in the central ⁷LFLGFL (L7 to L12) region. relative to the HFP1-⁸FLG and HFP3-⁸FLG samples, the models also predict smaller r and larger (ΔS/S₀)^exp for the HFP2-⁵GAL and HFP2-¹¹FLG samples which generally correlates with the experimental data, cf. Fig. 5a–b. The models suggest small r and significant (ΔS/S₀)expression for HFPs labeled at the N-terminal residues and future studies could examine samples labeled in this manner.

The PMI model in Fig. 8a would likely perturb the membrane and has some similarity with: (1) the PMI of extended conformation internal fusion peptides postulated from structures of dengue, Semliki forest, herpes, and vesicular stomatitis viral fusion proteins; (2) the PMI of helical influenza fusion peptide determined from electron spin resonance experiments; and (3) a PMI model based on the HFP-F8W fluorescence measurements (37, 61, 62, 93–96). However, the locations of lipids in the perturbed leaflet in the PMI model are not clear. For the FMI model of Fig. 8b, the positions of the lipids are more clear but there are non-hydrogen bonded CO and NH groups at the sheet edges with large Born energies. These energies would be reduced for a FMI β barrel structure, Fig. 8c. There is correlation between the FMI model and the deep insertion of the Trp sidechain suggested from fluorescence studies of the HFP-F8W mutant (36, 37). In the context of gp41, individual HFP trimers would be on the same side of the membrane in the PMI model but would be on different sides of the membrane in the FMI model. It is not clear how this FMI trimer topology would relate to the positions of the viral membrane-anchored gp41 trimers and the host cell membranes. The free energy difference between the A1 to G16 FMI state and a non-inserted state is ~3.9 kJ/mol as calculated from the sum of individual residue free energy values derived from transmembrane helices (97). The calculated difference for the I4 to G13 sequence is −2.3 kJ/mol and leads to the general conclusion that the free energy calculations do not strongly distinguish between the PMI and FMI models. Future studies could discriminate between the PMI and FMI models using REDOR distance measurements between peptide nuclei and lipid acyl chain nuclei (51).

There are similarities between these PMI and FMI models of oligomeric β strand HFP and PMI and FMI models which have been developed for a single HFP in a helical conformation (38, 39, 98). Much of the experimental data for helical HFP insertion has been based on detergent rather than membrane samples and there has been support for both micelle surface location and micelle traversal by HFP (19–23). Our results on oligomeric β strand HFP were consistent with the previous observations that the A15 and G16 residues of monomeric helical HFP were close to the water-micelle interface and that the F8 to G10 residues were furthest from this interface. Thus, there may be common features shared by the micelle and membrane locations of helical and β strand HFP.

Origin of f and effects of cholesterol

The HFP2-¹⁴AAG and HFP4-¹⁴A data could only be fit well with addition of the f parameter which approximately reflected the fractional population of peptides whose labeled residues were close to the ³¹P. Analysis of ¹³CO-³¹P REDOR data of a membrane-associated antimicrobial peptide also required an f parameter and the best-fit f and r values were similar to our results (51).

The membranes of host cells of HIV have cholesterol:lipid ~ 0.45 and the membranes of HIV have cholesterol:lipid ~ 0.8 (54, 57). These data suggest that it is interesting to probe the effect of cholesterol on HFP location in the membrane. Similar values of (ΔS/S₀)^exp were obtained for the ⁵GALFLGFLG residues in PC:PG and PC:PG:CHOL samples and the best-fit r for the ¹⁴AAG residues were comparable in both membrane compositions. These results suggest: (1) the inserted HFP population has similar location in membranes with or without cholesterol; and (2) the membrane location of the A14 to G16 residues may be similar in both helical and β strand conformations because there appeared to be some helical conformation in PC:PG and negligible helical conformation in PC:PG:CHOL. There was a difference in best-fit f for HFP2-¹⁴AAG in PC:PG and PC:PG:CHOL with values of 0.45 and 0.32, respectively. There are several potential explanations for this variation. First, the HFP/PC:PG samples likely had a small population of helical conformation which was absent in the HFP/PC:PG:CHOL samples and this helical population could have contributed to the larger f in PC:PG. Second, the presumably gel phase PC:PG and the liquid-ordered phase PC:PG:CHOL had lateral molecular densities of ~0.213 Å⁻² and ~0.256 Å⁻² as calculated from gel-phase PC and PG areas of 47 Ų, liquid-ordered PC and PG areas of 40 Ų, and cholesterol area of 37 Ų (60, 71, 99, 100). The denser packing in PC:PG:CHOL could have shifted an inserted HFP:surface HFP equilibrium to surface HFP and led to reduced f. Finally, relative to the PC:PG sample, there was likely a reduced number of phospholipids close to the ¹⁴AAG residues in the PC:PG:CHOL sample because of statistical substitution of nearby phospholipids with cholesterol. This lipid dilution may have also reduced f in the PC:PG:CHOL sample.

In summary, ¹³CO-³¹P distance measurements have demonstrated close proximity of the ¹⁴AAG residues to the lipid ³¹P for a significant fraction of membrane-associated HFP. This proximity was observed both for membranes with and without cholesterol. The chemical shifts of this study as well as results from previous studies correlated with a predominant population of HFP with β strand conformation. Models of partial and full insertion of β strands are proposed which are consistent with the experimental data. Although there have been numerous previous observations of β strand HFP, the role of this conformation in fusion has been controversial (14, 16, 18, 31, 45). The present study demonstrates that β strand HFP is in intimate contact with membranes and merits serious consideration as a fusogenic conformation.

Interesting future work could include studies of cross-linked HFPs which are thought to mimic the HFP oligomeric topology in the gp41 protein and which have increased fusion rates relative to the non-cross-linked HFPs of the present study (34). There may be a distinct membrane location of the cross-linked HFPs which correlates with their fast fusion rate. It is also known that the cross-linked HFP trimer will form helical or β strand conformation in membranes without or with cholesterol, respectively, so that studies of the trimer in different membrane compositions can provide information about the conformational dependence of peptide location in membranes (35).

Supplementary Material

File0011. SUPPORTING INFORMATION AVAILABLE.

Eq. 2 and the σ^lab expression are derived and discussed. This material is available free of charge via the Internet at http://pubs.acs.org.

ABBREVIATIONS

¹⁴AAG

residues A14, A15, and G16

¹³CO

¹³C labeled carbonyl

d

magnitude of ¹³C-³¹P dipolar coupling

DPPC-¹³C

1,2-dipalmitoyl-sn-glycero-3-phosphocholine with ¹³C labeling at the both carbonyl sites

DTPC

1,2-di-O-tetradecyl-sn-glycero-3-phosphocholine

DTPG

1,2-di-O-tetradecyl-sn-glycero-3-[phospho-rac-(1-glycerol)]

f

fraction of ¹³COs close to ³¹Ps

FMI

full membrane insertion

⁵GALFLGFLG

residues G5, A6, L7, F8, L9, G10, F11, L12, and G13

HEPES

N-(2-hydroxy-ethyl)piperazine-N’-2-ethanesulfonic acid

HFP

HIV fusion peptide

HFP1

AVGIGALFLGFLGAAGSTMGARS-NH₂

HFP1-⁸FLG

HFP1 ¹³CO labeled at F8, L9, and G10

HFP2

AVGIGALFLGFLGAAGSTMGARSKKK-NH₂

HFP2-⁵GAL

HFP2 ¹³CO labeled at G5, A6, and L7

HFP2-¹¹FLG

HFP2 ¹³CO labeled at F11, L12, and G13

HFP2-¹⁴AAG

HFP2 ¹³CO labeled at A14, A15, and G16

HFP3

AVGIGALFLGFLGAAGSTMGARSKKKA_β

HFP3-⁸FLG

HFP3 ¹³CO labeled at F8, L9, and G10

HFP4

AVGIGALFLGFLGAAGSTMGARSWKKKKKKA_β

HFP4-⁶A

HFP4 ¹³CO labeled at A6

HFP4-¹⁴A

HFP4 ¹³CO labeled at A14

HIV

human immunodeficiency virus

IR

infrared

⁷LFLGFL

residues L7, F8, L9, G10, F11, and L12

MAS

magic angle spinning

NMR

nuclear magnetic resonance

PC:PG

4:1 DTPC:DTPG

PC:PG:CHOL

8:2:5 DTPC:DTPG:cholesterol

PMI

partial membrane insertion

r

¹³C-³¹P internuclear distance

REDOR

rotational-echo double resonance

τ

duration of dephasing period

TFA

trifluoroacetic acid