Lipid acyl chain protrusion induced by the influenza virus hemagglutinin fusion peptide detected by NMR paramagnetic relaxation enhancement

Abstract

The glycoprotein spikes of membrane-enveloped viruses include a subunit that catalyzes fusion (joining) of the viral and target cell membranes. For influenza virus, this is subunit 2 of hemagglutinin which has a ~20-residue N-terminal fusion peptide (Fp) region that binds target membrane. An outstanding question is whether there are associated membrane changes important for fusion. Several computational studies have found increased “protrusion” of lipid acyl chains near Fp, i.e. one or more chain carbons are closer to the aqueous region than the headgroup phosphorus. Protrusion may accelerate initial joining of outer leaflets of the two membranes into a stalk intermediate. In this study, higher protrusion probability in membrane with vs. without Fp is convincingly detected by larger Mn²⁺-associated increases in chain ¹³C NMR transverse relaxation rates (Γ₂’s). Data analysis provides a ratio Γ_2,neighbor/Γ_2,distant for lipids neighboring vs. more distant from the Fp. The calculated ratio depends on the number of Fp-neighboring lipids and the experimentally-derived range of 4 to 24 matches the range of increased protrusion probabilities from different simulations. For samples either with or without Fp, the Γ₂ values are well-fitted by an exponential decay as the ¹³C site moves closer to the chain terminus. The decays correlate with free-energy of protrusion proportional to the number of protruded -CH₂ groups, with free energy per -CH₂ of ~0.25 k_BT. The NMR data support one major fusion role of the Fp to be much greater protrusion of lipid chains, with highest protrusion probability for chain regions closest to the headgroups.

Introduction

Many zoonotic diseases including AIDS, influenza, and COVID are caused by viral pathogens that are membrane-enveloped.^1–5 An initial step in cellular infection is fusion (joining) of the viral and target cell membranes with consequent deposition of the viral capsid in the cytoplasm. Enveloped viruses have glycoprotein spikes whose protein have a receptor-binding subunit (RbSu) followed by a fusion subunit (FsSu), with typical proteolytic cleavage between the two subunits.^1,6–10 The FsSu has a single transmembrane domain and a large N-terminal ectodomain (Ed) outside the virus membrane. Each spike contains a core with a defined number (often 3) of non-covalently-associated Ed’s of FsSu’s, and the same number of RbSu’s that are non-covalently bound with this core. After the virus is in the host, RbSu’s bind to specific receptor molecules on the exterior of target cells, and for some viruses, there is subsequent endocytosis. The RbSu’s move away from the FsSu Ed core, and the core changes to a new structure, typically a thermostable trimer-of-hairpins with T_m > 90 °C.^11–15 There isn’t sequence homology among the RbSu’s of different virus families which can be partly understood because the RbSu’s of different virus families bind different molecules. More surprisingly, there also isn’t sequence homology or sequence-length homology among the FsSu’s of different virus families. As noted above, the final Ed structure is typically a hairpin, but there are substantial length and structural differences between the hairpins of different families.^16–20

There are also large geometric changes of the membranes during fusion, including intermediate structures, but there aren’t yet clear experimental data about the relative timings of changes in membrane vs. FsSu Ed structure. Fig. 1 displays a common model for the membrane changes. These changes are in time-sequence: (a) initial close (nm) apposition of the viral and target membranes; (b) stalk intermediate that connects and is contiguous with the outer leaflets of the two membranes; (c) hemifusion diaphragm with contiguous inner leaflets of the two membranes; (d) pore formation in the diaphragm; and (e) pore expansion with final state of contiguous membranes and viral contents in the cytoplasm.^4,21,22 There are some experimental data that support this model as well as computational studies. The computational consensus estimates for energy barriers of uncatalyzed fusion are ~25 kcal/mole for step a and ~10 kcal/mole between the step a→b, b→c, and c→d states.⁴

Figure 1.

Pictorial representation of a common membrane fusion model that includes (a) initial close apposition of the viral and target membranes; (b) stalk formed from the outer leaflets of the two membranes; (c) hemifusion diaphragm that is contiguous with the inner leaflets of the two membranes; and (d) pore formation. The estimates of the free energy barriers for membrane apposition and for transformation between membrane intermediates are from computational studies of uncatalyzed fusion. The figure doesn’t show the final step of pore expansion that precedes full contents mixing. The different colors of the headgroups are meant to visually enhance the changes in membrane topology during fusion but don’t describe the locations of specific lipids during fusion. During the ~20 s estimated lifetime of a membrane intermediate structure in viral fusion, a lipid molecule could diffuse over ~10¹⁰ Ų leaflet area.

FsSu’s have a N-terminal region (Ntr) that is not part of the final hairpin structure. The Ntr is often folded within the initial spike and then released as the hairpin forms.^6–10 The Ntr length varies among FsSu’s from different viral families, and the range of lengths is typically between 30 and 250 residues. Within a Ntr, there are one or more proposed “fusion peptide” segments that are hypothesized to bind the target membrane during fusion.^23–26 The membrane-bound fusion peptide(s) may reduce the 25 kcal/mole apposition barrier, in conjunction with the more C-terminal hairpin structure of the Ed and viral transmembrane domain. In addition, a membrane may be modified by fusion peptide so that there is also reduction in the 10 kcal/mole barriers between subsequent membrane intermediates.^27–38 A fusion peptide segment has typically been identified by observation of mutations that reduce viral fusion and/or infection without affecting initial spike structure.^{15,24,39–44} In addition, a fusion peptide sequence is typically highly-conserved and should bind membrane.^41,42,45,46

The present study is specifically focused on the fusion peptide (Fp) of the influenza virus FsSu which is subunit 2 of the hemagglutinin protein (Ha2). The influenza RbSu (Ha1) binds sialic acid followed by endocytosis of the virus and then endosome maturation that includes pH reduction to ~5.¹ At low pH, Ha1 separates from the Ha2 Ed and the Ed then changes to the final trimer-of-hairpins structure with accompanying fusion between the viral and endosome membranes. The Ha2 Fp is the ~20 N-terminal and highly-conserved residues of Ha2. The Fp has been identified by: (1) significant attenuation of fusion when specific Fp residues are mutated; (2) very high sequence conservation among different influenza subtypes; and (3) observation of membrane-bound Fp after influenza virus fusion.^{15,23,40,44,46} The Ha2 Fp often adopts helical hairpin structure and is a mixture of: (i) closed structure in which the two antiparallel helices are in van der Waals contact; and (ii) semi-closed structure in which the Phe-9 sidechain is inserted between the two helices.^47,48

The effects of the Ha2 Fp on membrane have been studied by computer simulations by several different groups. These simulations have typically been done using a membrane with Fp peptide without the rest of Ha2. One commonly-observed effect is a higher (~4–20×) probability for chain protrusion by lipids next to vs. further from the Fp (Fig. 2).^49–51 Protrusion is specifically defined as one or more carbons of the lipid chain being at least 1 Å closer to the aqueous phase than the P nucleus of the lipid headgroup. Protrusion is a functionally-interesting motion. As depicted in Fig. 1, an early fusion step is the topological transition from (a) initial apposition of viral and target membranes to (b) stalk that connects the two membranes and is contiguous with the outer leaflets of these membranes. This step requires protrusion by some outer leaflet lipids of both membranes. The hypothesized correlation between increased lipid chain protrusion near Fp and stalk formation is supported by coarse-grained computer simulations of fusion that begin with full-length Ha2 with final trimer-of-hairpins structure and with Fp’s in one membrane and transmembrane domains in the other membrane.⁵² In the absence of Fp, simulations show that at any given time, ~1% of the lipids have a protruded chain.^49–51 For simulations with Fp, ~ 4–20× increased protrusion probability is observed for both chains of a lipid that is Fp-adjacent and for a variety of Fp structures.^49–51 Both interfacial and transmembrane locations of the Fp have been observed as well as a variety of geometries of the protruded lipid relative to the Fp. These include: (1) “straddling” of the protruded chain over the Fp; and (2) hydrogen bonding between the headgroup phosphate oxygen and one of the four N-terminal residues of the Fp with associated headgroup intrusion into the bilayer.^49,50

Figure 2.

Representative picture of lipid acyl chain protrusion near a Fp. A set of 128 POPC and 32 POPG lipids in a pre-assembled bilayer were energy-minimized in a water box using the CHARMM/Membrane Builder/Bilayer Builder/Membrane Only System molecular dynamics program. A membrane cross-section is displayed with lipid acyl chains in light green. A representative protruded chain in magenta is next to a Fp backbone in turquoise and near a Mn²⁺ in purple that is bound to a lipid headgroup. The picture shows a protruded palmitoyl chain and a Fp with semi-closed structure with marked N- and C- termini. There is increased protrusion in simulations for both palmitoyl and oleoyl chains and for lipids next to a variety of Fp structures. In addition, Fp in simulations is observed with both interfacial and transmembrane locations and protruded lipids exhibit a variety of geometries relative to the neighboring Fp.

Despite its potential significance, to our knowledge there hasn’t yet been direct experimental observation of increased lipid protrusion for membrane with vs. without Fp. If chain protrusion is increased with Fp, there will be associated decreases in chain order parameters, and such decreases were observed and quantified in some of the computational simulations with membrane with Fp peptide.^49,51 However, these computational predictions were contradicted by electron paramagnetic resonance (EPR) and fluorescence spectra that showed increases in chain order parameters for samples with Fp.^28,33 Such increases were also observed for membrane with putative fusion peptides from other virus families.^34–36 The sequences are non-homologous with the Ha2 Fp sequence. However, more recently, ²H nuclear magnetic resonance (NMR) spectra of membrane with perdeuterated lipids showed decreases in chain order parameters with vs. without Fp.³⁷ The NMR-derived order parameters were in semi-quantitative agreement with the simulation-derived parameters.^49,51,53 The seeming contradiction between EPR and NMR/simulation may be explained by: (i) EPR detects order for the spin labels and only 0.005 mole fraction of the sample lipids are spin-labeled, whereas (ii) NMR and simulation detect order for all lipids in the sample. The EPR-detected ordering of the spin label with Fp may be due to preferential binding of Fp to the spin label. This hypothesis is supported by the dose response of ordering with respect to Fp mole fraction. The ordering reaches its maximum value when the Fp:lipid mole ratio ≈ 0.002, which is similar to the spin-labeled lipid:total lipid mole ratio.

There is other indirect experimental support for increased lipid protrusion from analysis of the large increases in ²H NMR transverse relaxation rates (R₂’s) of deuterated lipids in samples with vs. without Fp.³⁸ The increases are interpreted to be due to modulation of the ²H NMR frequency as the lipid laterally diffuses in the membrane leaflet. For lipid next to vs. further from Fp, there are larger vs. smaller amplitudes of mean-squared ²H quadrupolar fields that are correlated with the smaller vs. larger chain order parameters. The experimental increases in R₂’s with vs. without Fp are in semi-quantitative agreement with values calculated using experimentally-based estimates of order parameters and the time for a lipid to diffuse past a Fp. Both the decreases in chain ²H order parameters and large increases in ²H R₂’s were also observed for a membrane sample with the putative fusion peptide of the HIV FsSu (gp41).³⁸ The gp41 fusion peptide and the Ha2 Fp have non-homologous sequences and also adopt very different membrane-bound structures.^54–58

Although these earlier experimental data are consistent with increased chain motion for lipid next to vs. further from Fp, they don’t directly evidence increased protrusion (Fig. 2). This knowledge gap motivated the present experimental NMR study in which chain protrusion is probed using comparison of chain ¹³C R₂’s in samples with vs. without the paramagnetic Mn²⁺ species.^59–61 The Mn²⁺ binds to the phosphate oxygens of the lipid headgroup, so a chain ¹³C R₂ is augmented when the ¹³C site is protruded into the headgroup region. The change in protrusion probability with vs. without Fp is probed by the difference in Mn²⁺-associated increase in R₂, i.e. paramagnetic relaxation enhancement (PRE). There is spectral resolution of some of the NMR signals from different −CH₂ sites, so the approach also yields information about how Fp affects protrusion for −CH₂ groups closer to the headgroup vs. the chain terminus.⁶²

Materials and Methods

Lipids and Fp.

Fig. 3 displays the lipids 1-palmitoyl-2-oleoyl-glycero-3-phosphocholine (POPC) and1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-(1’-rac-glycerol) (sodium salt) (POPG) that were purchased from Avanti Polar Lipids (Alabaster, AL, USA). The influenza Ha2 Fp was purchased from GL Biochem (Shanghai, China). Other materials were typically purchased from Sigma-Aldrich (St. Louis, MO). The Fp sequence is GLFGAIAGFIENGWEGMIDGGGKKKKG. The underlined segment is the 20 N-terminal residues of the Ha2 subunit of the hemagglutinin protein (H3 subtype) and the C-terminal segment is non-native residues that greatly increase Fp solubility in aqueous solution so that Fp binding to membrane can be done without organic solvent or detergent additives. The Fp was prepared by Fmoc solid-phase peptide synthesis and purified by reverse-phase HPLC. GL Biochem stated that Fp purity > 95% and this statement is consistent with the electrospray ionization mass spectrum that we acquired (Fig. S1).

Figure 3.

Chemical structures of POPC and POPG lipids with site numbering of the acyl chains with prime (‘) for the palmitoyl chain and no prime for the oleoyl chain.

Membrane samples.

(1) Lipid (~50 μmole) was dissolved in 2 ml chloroform:methanol (9:1 v/v) and the solvent removed by nitrogen gas and then overnight vacuum. (2) The dried lipid was suspended in ~2 mL of 10 mM HEPES/5 mM MES buffer at pH 5.0 and the suspension was subjected to freeze-thaw cycles (~10×). (3) The suspension was subjected to ultra-centrifugation at 150000g at 4 °C for 2 hours. (4) The harvested lipid pellet was lyophilized. (5) A hydrated lipid sample was prepared in a 3.2 mm outer-diameter NMR rotor by adding in sequence a ~10 μL aliquot of water, a portion of the lipid pellet, and then another aliquot of water. The total volume was ~40 μL. The top cap was placed on the rotor followed by overnight incubation at ambient temperature for membrane hydration. Evidence for hydration includes wet appearance and reproducible ¹³C NMR linewidths (~0.3 ppm full-width at half-maximum).

Membrane samples with Mn²⁺.

An aqueous solution was prepared with [MnCl₂] either ~40 mM or ~4 mM and an aliquot added to the lipid suspension after step 2. The suspension was then subjected to freeze-thaw cycles (~5×) to promote homogeneous distribution of Mn²⁺ in the sample.⁶⁰ After step 3 ultracentrifugation, [Mn²⁺]_free was detected in the supernatant using an Agilent/Varian AA240 atomic absorption spectrometer with air-acetylene flame and 279.5 nm wavelength. Instrument calibration was done with MnCl₂ standard solutions in HEPES/MES buffer at pH 5. The Mn²⁺ not in the supernatant was considered bound to the membrane and %Mn²⁺ = (mole bound Mn²⁺)/(mole lipid) × 100.

Membrane samples with Fp.

A solution was prepared with [Fp] ≈ 1 mM in HEPES/MES buffer at pH 5.0. The solution was added dropwise to the lipid suspension after step 2 so that the Fp:lipid ≈ 1:30 molar ratio. The Fp/lipid suspension was subjected to freeze/thaw cycles (~5×) and then gently agitated overnight. If Mn²⁺ was to be included in the sample, the MnCl₂ solution was added before the freeze/thaw cycles. After step 3 ultracentrifugation, [Fp]_free was measured in the supernatant using A₂₈₀ and if MnCl₂ had been added, [Mn²⁺]_free was measured in the supernatant using flame atomic absorption spectroscopy.

NMR spectroscopy.

Data were acquired on a NMR spectrometer with 9.4 T magnet, Bruker Neo console, and Bruker Efree magic angle spinning (MAS) probe designed for lower dielectric heating of aqueous samples and for a rotor with 3.2 mm outer diameter. NMR data were acquired at 298 K with 8.0 kHz MAS frequency, ¹³C transmitter at 100.0 ppm, and ¹H transmitter at 3.5 ppm. Fig. 4 displays the pulse sequence with ¹H→ ¹³C cross polarization (CP) followed by dephasing with Hahn echo, and then ¹³C acquisition, with ¹H decoupling during dephasing and acquisition.^63,64 CP parameters were varied to yield highest aliphatic intensity and typically included a 2.5 μs ¹H π/2 pulse followed by 1.4 ms contact time with 42 kHz ¹C radiofrequency (rf) field and ¹H rf field with a linear ramp between 44 and 60 kHz. The Hahn echo was τ/2- ¹³C π pulse-τ/2. Data were collected for a range of τ typically between ~2 and ~40 ms and the 10.0 μs ¹³C π pulse was rotor-synchronized with the start of dephasing. There was 50 kHz rf field of the ¹H SPINAL-64 decoupling, acquisition with 25 μs dwell time and 1400 complex points, 1 s recycle delay, and sum of 8 or 16 K scans. The phases of ¹H π/2, ¹³C CP, ¹³C π, and receiver were cycled: (y, x, x, x); (-y, x, -x, -x); (y, -x, -x, -x); (-y, -x, x, x); (y, y, y, y); (-y, y, -y, -y); (y, -y, -y, -y); (-y, -y, y, y). The 8-step phase cycle includes 180° alternation of the ¹H π/2 phase and correlated quadrature alternation of the ¹³C CP, ¹³C π, and receiver phases. Spectra were referenced using the methylene peak of adamantane at 40.5 ppm peak and the terminal lipid chain ¹³CH₃ peak at 13.9 ppm.

Figure 4.

CP-MAS/Hahn-echo sequence displayed as rf field vs. time. Typical parameters included ¹H transmitter at 3.5 ppm, ¹³C transmitter at 100.0 ppm, 8.0 kHz MAS frequency, 2.5 μs ¹H π/2 pulse, 1.4 ms CP contact time, an array of τ between ~2 and ~40 ms with τ/2 an integral number of rotor periods, 10.0 μs ¹³C π pulse, SPINAL-64 ¹H decoupling during the Hahn-echo and acquisition periods, ¹³C acquisition with 25 μs dwell time and 1400 complex points, 1 s recycle delay, and sum of 8 or 16 K acquisitions using the phase cycle described in the text. Typical ¹H rf fields are: CP, linear ramp between 44 and 60 kHz; and decoupling, 50 kHz. The typical ¹³C CP rf field is 42 kHz.

The ¹H recycle delay of 1 s was chosen based on earlier studies that showed the ¹H longitudinal relaxation rate is ~1.6 s⁻¹ for lipid -CH₂ groups in the liquid-crystalline phase.^65,66 The 1 s recycle delay is close to the ~0.8 s delay for optimal (signal-to-noise)/time. Longer ¹H recycle delay results in larger initial ¹H magnetization for ¹H-¹³C cross-polarization, but because the same delay is used throughout a data array, all the ¹³C intensities from the array are scaled by the same factor, and there isn’t an effect on the ¹³C R₂ transverse relaxation rate.

The NMR data were typically processed with 8K zero-filling, 20 Hz exponential line broadening, Fourier transform, zero- and first-order phase correction, and baseline correction. Spectral peaks were integrated over specific chemical shift ranges and the peak integrals S vs. τ were fitted to S = A × exp(−R₂ × τ) with A and R₂ as fitting parameters.

Results

NMR sample preparation.

The goal of this study is to probe whether or not lipid acyl chains have greater probability of protrusion into the aqueous phase when Fp is membrane-bound, as has been observed in some simulations. There are several considerations for sample preparation including lipid composition of the membrane and optimal mole% values of membrane-bound Fp and Mn²⁺, where %Fp or %Mn²⁺ = (mole Fp or Mn²⁺)/(mole lipid) × 100. One consideration for %Fp is simulation data showing that protrusion probability is increased by a factor of 4–20× for lipids next to a Fp, whereas changes are much smaller for more distant lipids.^49–51 During NMR data collection, a lipid molecule experiences rapid lateral diffusion in the liquid-crystalline phase and will spend time both next to and further from a Fp.^67–69 Larger %Fp is anticipated to result in greater fraction of time next to Fp but there will also be undesired effects if %Fp is too large. These include oligomeric β sheet rather than monomer helical hairpin Fp structure, with the latter being the likely Fp structure in full-length Ha2.^15,70 For the 3% Fp of our samples, earlier studies evidence that the Fp adopts monomer helical hairpin structure and the membrane retains the liquid-crystalline bilayer phase.^{29,37,38,48,71,72} This retention was supported by comparison between lipid samples without vs. with Fp. The two sample types exhibited similar lineshapes, linewidths, and relaxation rates for both their ³¹P and ²H static NMR spectra.

We first prepared POPC samples because the earlier simulations had typically used POPC. For sample preparation with an aliquot corresponding to 5% Mn²⁺, the [Mn²⁺]_free ≈ 0 in the supernatant after centrifugation which correlates with ~5% bound Mn²⁺, i.e. complete binding to lipid. By contrast, most Mn²⁺ did not bind to POPC with 3% bound Fp. The low binding may be due to electrostatic repulsion between Mn²⁺ and Fp, where the calculated Fp charge is +1.6. We switched to POPC:POPG (4:1), referred to as “PC:PG”, with the choice of POPG based on its −1 charge and on otherwise similar properties to POPC.⁷³ For sample preparation with an aliquot corresponding to 5% Mn²⁺, the [Mn²⁺]_free correlated with ~5% bound Mn²⁺ for PC:PG without Fp and ~4% bound Mn²⁺ for PC:PG with 3% Fp. For PC:PG either without or with Mn²⁺, there was complete binding of Fp to lipid, based on A₂₈₀ ≈ 0 in the supernatant after centrifugation. This result is consistent with Fp_bound:Fp_free ≈ 10⁴ calculated by K_bind × [lipid] where K_bind ≈ 10⁶ M⁻¹ is the previously-determined binding constant and [lipid] ≈ 10⁻² M in our sample preparation.²⁵

¹³C CP NMR spectra without Hahn echo are displayed for the lipid samples in Fig. 5 either (a) without or (c) with Fp. Assignments are shown for the lipid acyl chain peaks use the Fig. 3 carbon numbering with prime (‘) for the palmitoyl chain and no prime for the oleoyl chain.⁶² There typically isn’t resolution of individual peaks for POPC vs. POPG. There are resolved peaks for sites with distinctive bonding environments, but with superposition of signals from at least two sites. The 9,10 peak exhibits partial resolution of the 9 and 10 signals at higher and lower shift, respectively. The * peak is a superposition of signals from the 4–7, 12–15, and 4’−13’ sites. For the same lipid site peak, there are negligible spectral shift or width differences for samples without vs. with Fp. There aren’t peaks that could correspond to Fp signals, with possible explanations being the: (1) 3% Fp concentration; and (2) broader Fp linewidths because of less motional averaging for Fp vs. lipid.

Figure 5.

¹³C NMR spectra of samples containing (a) Lipid, (b) Lipid + Mn²⁺, (c) Lipid + Fp, and (d) Lipid + Fp + Mn²⁺, with 0.5% Mn²⁺ and 3% Fp that are calculated as (mole Mn²⁺ or Fp)/(mole lipid) × 100. There is POPC:POPG (4:1) lipid composition. The data were acquired after CP without the Hahn echo. The vertical scales of the four spectra have been adjusted so that the “*” peaks at 30 ppm have the same height. Assignments are displayed for the lipid acyl chain peaks use the Fig. 2 carbon numbering with prime (‘) for the palmitoyl chain and no prime for the oleoyl chain. There isn’t resolution of individual peaks for POPC vs. POPG. There are resolved peaks for sites with distinctive bonding environments, but with superposition of signals from at least two sites. Peaks are assigned for: (1) the 2,2’ and 3,3’ sites that are one and two bonds from the carbonyl groups; (2) the 9,10 C=C and 8,11 C=C adjacent sites of the oleoyl chain; (3) the 18,16’ -CH₃ sites; and (4) the 17,15’ and 16,14’ sites that are one and two bonds from the -CH₃ groups. The * peak is a superposition of signals from the 4–7, 12–15, and 4’−13’ sites. The carbonyl peak and headgroup peak regions are also noted.

We first measured the effect of %Mn²⁺ on ¹³C R₂, with the goal of finding the optimal %Mn²⁺ for detection of lipid protrusion using the R₂ difference between samples with vs. without Mn²⁺, i.e. Γ₂ = R_2,Mn – R_2,NoMn. The Fp effect on protrusion is assessed by comparison of Γ_2,Fp vs. Γ_2,noFp. The Mn²⁺ is likely close to the negatively-charged lipid phosphate oxygens in a lipid headgroup. The Γ₂ ∝ 〈r⁻⁶ 〉 where r is the ¹³C-Mn²⁺ distance and 〈…〉 is the average over the ~10 ms NMR measurement time.⁵⁹ The 〈r⁻⁶ 〉 will depend on %Mn²⁺ as well as lipid ¹³C site (Figs. 2, 3). We hypothesized that increased lipid protrusion with Fp could be mostly clearly observed if Γ_2,noFp were comparable or smaller than R_2,noMn,noFp. This is based on the idea that Fp-induced increases in 〈r⁻⁶ 〉 would be more readily detectable if Γ_2,noFp isn’t already approaching its maximal value. We first tried 5% Mn²⁺ in lipid without Fp and observed complete loss of the 2,2’ and 3,3’ signals as well as apparent Γ₂/R_2,noMn ratios of ~10 for 8,11 and 9,10, ~6 for *, and ~5 for 16,14’. We then investigated %Mn²⁺ in the 0.5 – 1.25% range and observed that even at 0.5%, Γ₂ > 0 for many peaks and Γ₂/R₂ < 1 for all peaks (Tables 1, S1, and S2). We then prepared samples with 0.5% Mn²⁺ for more complete analysis. There was no detectable [Mn²⁺]_free for the lipid without Fp sample and small [Mn²⁺]_free for the lipid with Fp sample. The %Mn²⁺ _bound were 0.50 and 0.48%, respectively.

Table 1.

Site-specific ¹³C transverse relaxation rates of acyl chains of POPC:POPG (4:1) membrane and % Mn²⁺ dependence (fitting uncertainties in parentheses)^a

Mn2+	13C R2 (s−1)				13C Γ2 (s−1)
%	2,2′	3,3′	*	16,14′	2,2′	3,3′	*	16,14′
0	28.1(1.2)	20.3(1.0)	15.9(0.2)	13.9(0.9)
0.5	50.2(1.8)	34.3(2.2)	20.5(0.9)	15.4(1.0)	22.1(2.5)	14.0(2.4)	4.6(0.9)	1.4(1.3)
0.75	81.9(6.3)	59.1(3.4)	26.4(0.9)	13.7(1.9)	53.8(6.5)	38.8(3.5)	10.5(0.9)	−0.2(2.1)
1.00	92.1(3.4)	60.8(4.6)	28.4(1.5)	22.0(1.4)	64.0(3.8)	40.5(4.7)	12.5(1.5)	8.1(1.6)
1.25	93.9(7.0)	67.8(5.8)	29.2(1.4)	20.7(1.9)	65.8(7.2)	47.6(5.9)	13.3(1.4)	6.8(2.1)

^aEach ¹³C transverse relaxation rate (R₂) was determined from best-fitting the integrated NMR peak intensity S vs. delay time τ using S = A × exp(−R₂ × τ) where A and R₂ are fitting parameters. The fitting uncertainty of R₂ is given in parentheses. The Γ₂ values are the differences between the best-fit R₂ values of samples with vs. without Mn²⁺. The * peak is the superposition of the 4–7, 12–15, and 4’−13’ signals. The typical ppm integration ranges for peaks are: 2,2’, 33.00–37.00; 3,3’, 24.00–26.30; *, 28.30–31.50; 16,14’, 31.50–33.00. The % Mn²⁺ = (mole bound Mn²⁺)/(mole lipid) × 100.

¹³C NMR spectra and relaxation.

Fig. 5 displays the ¹³C CP NMR spectra without Hahn echo of the (b) lipid + Mn²⁺ and (d) lipid + Fp + Mn²⁺ samples with comparison to the (a, c) samples without Mn²⁺. The vertical scales of the four spectra have been adjusted so that the * peaks at 30 ppm have the same height. For the same lipid site peak, there is negligible spectral shift without vs. with Mn²⁺. For both lipid and lipid + Fp, there is attenuation of the lipid 2,2’ and 3,3’ peak intensities with vs. without Mn²⁺, whereas there is not obvious attenuation for the 16,14’, 17,15’, and 18,16’ peaks. In addition, the lipid headgroup and CO signals are highly attenuated by Mn²⁺.

Fig. 6 displays 2,2’ spectral signal intensities vs. Δτ (increment in dephasing time) of the four samples. The spectra were acquired with the CP-Hahn echo sequence (Fig. 4). The vertical scales of the spectra of each sample were adjusted so that the Δτ = 0 spectral peaks of all samples have the same height. The signal intensities of the (a) lipid and (c) lipid + Fp samples exhibit semi-quantitatively similar attenuation of signal intensities with Δτ. There is greater attenuation for the (c) lipid + Mn²⁺ sample and even greater attenuation for the (d) lipid + Fp + Mn²⁺ sample. Fig. 6 provides qualitative spectral evidence which supports the hypothesis that bound Fp induces higher probability of lipid protrusion. Although the shortest τ value was 1.25 ms for panel (a) vs. 2.00 ms for (b-d), the NMR intensity S(τ) for all data was close to the S(0) intensity prior to relaxation-associated decay, i.e. the S(τ)/S(0) ratio in (a-d) is ~ 0.96, 0.90, 0.95, and 0.88, respectively, as calculated using the R₂ rates described in the next paragraph.

Figure 6.

The 2,2’ spectral signals vs. Δτ (increment in dephasing time) of samples containing (a) Lipid, (b) Lipid + Mn²⁺, (c) Lipid + Fp, and (d) Lipid + Fp + Mn²⁺, with 0.5% Mn²⁺ and 3% Fp that are based on (mole Mn²⁺ or Fp)/(mole lipid) × 100. Spectra were acquired with the CP-Hahn echo sequence. The vertical scales of the spectra of each sample were adjusted so that the Δτ = 0 spectral peaks have the same height. For these top spectra, τ = 1.25 ms for (a) and 2.00 ms for (b-d).

The integrated intensities vs. τ of each acyl chain -CH₂ signal were fitted to single exponential decays, i.e. S(τ) = A × exp(−R₂ × τ) where A and R₂ are fitting parameters. Fig. 7 displays plots and fittings of S(τ)/A vs. τ for some of the signals and Table 2 gives the best-fit R₂’s for all fittings as well as the Γ₂’s, the R₂ changes for samples with vs. without Mn²⁺. Table 2 provides the uncertainties in the R₂ and Γ₂ values in parentheses. Fitting is done both for the full integration range of the * peak and for smaller integration ranges of partially-resolved peaks with predominant contributions from a more limited number of sites (Fig. S2). The uncertainties of the experimental peak intensities were calculated as the standard deviations of the integrated intensities in noise regions of the spectra. For two-site peaks, these experimental uncertainties were typically between 10⁻³ and 10⁻², i.e. smaller than the dimensions of points in the plots in Figure 7. Fig. S3 displays the data and fittings on a logarithmic scale. Fig. 8 displays a bar plot of the Γ₂’s of samples without vs. with Fp, with data from each of the resolved peaks that are due to two ¹³C sites in the acyl chains (Fig. 5). Table 3 lists the experimental linewidths (Δν’s) of the peaks in the four samples as well as differences Δ[Δν] with vs. without Mn²⁺. Table 4 provides the inhomogeneous contributions to the linewidths of the resolved peaks, calculated using Δν_inhom = Δν_exp – R₂/π. The Δν_inhom values are shown for resolved peaks of individual samples, along with average values for the four samples and associated standard deviations.

Figure 7.

Integrated peak intensities vs. dephasing time (τ) and best-fit single exponential decays for the Lipid, Lipid + Mn²⁺, Lipid + Fp, and Lipid + Fp + Mn²⁺ samples. Data and fittings are displayed for the (a) 2,2’; (b) 3,3’; (c) 9,10; and (d) * peaks. The * peak is a superposition of signals from the 4–7, 12–15, and 4’−13’ sites. The peak intensities vs. τ were fitted to A × exp(−R₂ × τ) with A and R₂ as fitting parameters. The displayed intensities have been divided by A so that the best-fit intensity = 1 for all peaks when τ = 0. The best-fit R₂’s and their uncertainties are presented in Table 2. The uncertainties in the peak intensities were calculated as the RMSD’s of ten different integrals in noise regions of the spectra. For the two-site peaks, these uncertainties were typically between 10⁻³ and 10⁻² and less than the dimension of the points in the plots.

Table 2.

Site-specific ¹³C transverse relaxation rates of acyl chains of POPC:POPG (4:1) membrane and Mn²⁺ and Fp dependences (uncertainties in parentheses)^a

13C	R2 (s−1)				Γ2 (s−1)
	w/o Fp		3% Fp		w/o Fp	3% Fp
	w/o Mn2+	0.5% Mn2+	w/o Mn2+	0.5% Mn2+
2,2′	28.8(1.4)	51.5(2.1)	28.1(1.2)	63.5(3.2)	22.8(2.6)	35.4(3.4)
3,3′	20.0(1.5)	35.2(2.4)	21.3(2.0)	51.1(4.0)	15.2(2.8)	29.8(4.1)
8,11	9.1(0.7)	12.6(0.9)	13.6(1.6)	19.8(1.7)	3.5(1.2)	6.2(2.3)
9,10	8.4(0.6)	11.9(0.8)	8.7(1.0)	13.0(1.2)	3.5(1.0)	4.3(1.2)
16,14′	15.5(0.7)	15.0(0.6)	13.8(0.8)	15.4(0.7)	−0.5(0.9)	1.5(1.0)
17,15′	5.9(0.7)	7.3(0.4)	8.6(0.8)	9.7(1.0)	1.4(0.8)	1.1(1.2)
*{4–7,12–15,4′−13′}	15.8(0.3)	20.5(0.9)	17.8(0.7)	24.1(1.1)	4.7(0.9)	6.4(1.3)
*1{ 6′−9′}	24.8(0.2)	27.2(1.5)	25.4(0.5)	30.3(0.3)	2.5(1.6)	4.9(0.6)
*2{ 7,10′,11′}	15.1(0.2)	16.4(0.9)	16.0(0.7)	23.2(1.0)	1.3(0.9)	7.2(1.2)
*3{4–6,12–15,4′,5′,12′,13′}	11.7(0.3)	18.8(0.9)	15.8(1.2)	23.1(1.8)	7.1(1.0)	7.4(2.2)

^aEach ¹³C transverse relaxation rate (R₂) was determined from best-fitting the integrated NMR peak intensity S vs. delay time τ using S(τ) = A × exp(−R₂ × τ) where A and R₂ are fitting parameters. The fitting uncertainty of R₂ is given in parentheses. The Γ₂ values are the differences between the best-fit R₂ values of samples with vs. without Mn²⁺. Typical ppm integration ranges for peaks are: 2,2’, 33.00–37.00; 3,3’, 24.00–26.30; 8,11, 26.50–28.20; 9,10, 128.00–131.00; 16,14’, 31.50–33.00; 17,15’, 21.50–23.60; *, 28.30–31.50; *1, 30.24–31.50; *2, 30.04–30.24; *3, 28.30–30.04 (Fig. S2). The ¹³C sites that make the largest contributions to the *1, *2, and *3 integration ranges are listed between the brackets. The % Mn²⁺ = (mole bound Mn²⁺)/(mole lipid) × 100. The % Fp is calculated using the same type of expression.

Figure 8.

Bar plot of the Γ₂’s, i.e. the differences between the R₂’s for samples with vs. without Mn²⁺. The Γ₂’s are displayed for several peaks. Each peak is due to signals from two ¹³C sites in the acyl chains. The Γ₂’s are shown for samples without vs. with Fp. The Γ₂’s and their uncertainties are presented numerically in Table 2.

Table 3.

Site-specific ¹³C FWHM NMR linewidths of acyl chains of POPC:POPG (4:1) membrane and Mn²⁺ and Fp dependences^a

13C	Δν (Hz)				Δ[Δν] (Hz)
	w/o Fp		3% Fp		w/o Fp	3% Fp
	w/o Mn2+	0.5% Mn2+	w/o Mn2+	0.5% Mn2+
2,2′	33.1	43.2	34.8	49.4	10.1	14.6
3,3′	37.3	43.1	38.7	47.8	5.8	9.1
8,11	39.7	42.8	38.8	44.1	3.1	5.3
9,10	33.0	39.4	33.7	36.4	6.4	2.7
16,14′	30.0	34.7	30.0	31.1	4.7	1.1
17,15′	27.5	32.9	28.1	28.9	5.4	0.8
*{4–7,12–15,4′−13′}	110.7	112.4	111.7	116.0	1.7	4.3

^aThese are experimental full-width at half-maximum linewidth of the ¹³C signal of the CP NMR spectrum. The Δ[Δν] is the difference in linewidths between samples with vs. without Mn²⁺.

Table 4.

Inhomogeneous contributions to site-specific ¹³C FWHM NMR linewidths of acyl chains of POPC:POPG (4:1) membrane^a

13C	Δνinhom (Hz)				Average
	w/o Fp		3% Fp
	w/o Mn2+	0.5% Mn2+	w/o Mn2+	0.5% Mn2+
2,2′	23.9	26.8	25.9	29.2	26.4(1.9)
3,3′	30.9	31.9	31.9	31.5	31.6(0.4)
8,11	36.8	38.8	34.5	37.8	37.0(1.6)
9,10	30.3	35.6	30.9	32.6	31.6(2.0)
16,14′	25.1	29.9	25.6	26.2	27.0(1.9)
17,15′	25.6	30.6	25.4	25.8	26.8(2.2)

^aThe Δν_inhom = Δν_exp – (R₂/π) i.e. the difference between the experimental full-width at half-maximum linewidth and the experimental relaxation contribution to the linewidth. The average is for the four samples with the standard deviation in parentheses.

The four samples for Figs. 5–8 were prepared around the same time and as similarly as possible other than the prescribed absence vs. presence of Mn²⁺ and/or Fp. Replicate datasets were acquired and analyzed and Table S3 provides the best-fit R₂’s for the replicate data. The differences in best-fit R₂’s between replicates are typically comparable to the uncertainties for the differences. Table S4 displays best-fit R₂’s for replicate samples and the R₂’s are also similar between samples.

Discussion

Hydration and location of Mn²⁺ in samples.

The lipids in the samples were likely close to full hydration. Water was added to the rotor both before and after the lipid is added, i.e. the water was initially both underneath and on top of the lipid. The rotor was then sealed with the top cap followed by >12 hours prior toacquisition of NMR spectra, and this incubation time was intended to promote water permeation and homogenous hydration of the lipid. Such hydration is evidenced by the: (1) wet appearance of the sample, including after the NMR experiments were completed; and (2) ¹³C symmetric lineshapes with narrow (~0.3 ppm FWHM) linewidths. We also anticipate little evaporation of water because the rotor was sealed, the MAS frequency was moderate (8 kHz), and the ¹H rf circuit of the NMR probe was designed to minimize dielectric heating of the water. For a typical sample, ~20 μL water was added to the rotor which has ~40 μL total volume rotor, so there is ~20 μL lipid in the sample. POPC and water have similar densities, so the lipid:water mass ratio was similar to the volume ratio. By this approach, the lipid:water mass ratio in our samples was comparable to the ~3:2 ratio for fully-hydrated lipid, which was calculated using 28 water molecules-per-lipid molecule.⁷⁴

The Mn²⁺ likely binds the lipids rather than Fp. As noted in the Results section, membrane with only POPC lipid (with zwitterionic headgroup) quantitatively binds Mn²⁺ whereas POPC membrane with bound Fp doesn’t bind Mn²⁺. This was the reason that the NMR samples were prepared with 20 mole% POPG lipid with anionic headgroup, with resulting Mn²⁺ binding when peptide was also bound. In addition, the experimentally-determined pK_a’s of the sidechains of the two Glu and one Asp sidechains are >5. For the NMR samples at pH 5.0, the Glu and Asp sidechains have only partial negative charge, whereas the lipid headgroups have full negative charge.⁷⁵

NMR relaxation data support increased probability of lipid chain protrusion with Fp.

There are overall larger Γ₂’s for lipid with vs. without Fp, and the magnitude is most pronounced for the 2,2’ and 3,3’ signals, with reductions in R₂ and Γ₂ values for -CH₂ sites closer to the chain terminus (Figs. 6–8, Table 2). The Γ₂ trend is consistent with the results of the initial molecular dynamics simulation showing increased protrusion probability with Fp and specifically with the result summarized on p. 3 of this article “…our simulations predict the [Fp] effect on tail protrusion to be most profound in the upper region of the acyl chain …”⁴⁹ (Fig. 2).

Earlier NMR relaxation data were used to estimate a correlation time of ~10⁻⁸ s for the lipid chain motion that could lead to protrusion in the liquid-crystalline membrane.^76,77 This time is much smaller than the characteristic (1/R₂) ≈ 10⁻¹ s for ¹³C transverse relaxation so the R₂ is most reasonably considered as a weighted average of the chain’s high-probability unprotruded state and low-probability protruded state. We hypothesize that a site’s Γ₂ is proportional to the probability of chain protrusion (Ω_prot) into the headgroup region.⁶¹ This hypothesis is based on: (1) the r⁻⁻⁶ dependence of Γ₂ where r is the ¹³C-Mn²⁺ distance; and on (2) smaller r and therefore much larger Γ₂ when the chain is protruded. As described in the Results section, only 0.005 fraction of the lipid headgroups in our samples have bound Mn²⁺. The Mn²⁺ are likely also exchanging rapidly between headgroups during ¹³C transverse relaxation. Based on arguments similar to those above for protrusion, we additionally hypothesize that a site’s Γ₂ is also proportional to the probability that a Mn²⁺ is bound to a headgroup close to the protruded chain (Ω_Mn).

We estimate r for the protruded state with nearby Mn²⁺ by combining our two hypotheses with the known expression for Γ₂ with a nuclear spin I with spin ½ and a nearby paramagnetic species with spin quantum number L:

$$\text{r}\approx \{(1/15)\times {\left({\mu }_0/4\pi \right)}^2\times {\gamma }_1{}^2\times {\text{g}}_{\text{e}}{}^2\times {\mu }_{\text{B}}{}^2\times \text{L}(\text{L}+1)\times \left(4{\text{J}}_0+3{\text{J}}_{\omega \text{l}}\right)\times {\Omega }_{\text{prot}}\times {\Omega }_{\text{Mn}})]/{\Gamma }_2{}^{1/6}$$ (Eq. 1)

whereμ₀ is the permeability of free space, γ_I is the nuclear spin gyromagnetic ratio, g_e is the electron spin g factor, μ_B is the Bohr magneton, J_ω is the spectral density at angular frequency ω, and ω_I is the angular NMR Larmor frequency of the nucleus.⁵⁹ Using the known values ofμ₀, γ_I for ¹³C, g_e, andμ_B, and L = 5/2 for Mn²⁺,

$$\text{r}\approx \{\left[\left(9.106\times {10}^{-45}{\text{m}}^6-{\text{s}}^{-2}\right)\times \left(4{\text{J}}_0+3{\text{J}}_{\omega \text{l}}\right)\times {\Omega }_{\text{prot}}\times {\Omega }_{\text{Mn}}\right)]/{\Gamma }_2{}^{1/6}$$ (Eq. 2)

There are estimates below for the other terms but we note that because of the 6^th root dependence, r is fairly insensitive to moderate changes in these estimates, e.g. a 10× increase in the expression in braces correlates with a 1.1× increase in r. The J_ω is calculated using:

$${\text{J}}_{\omega }={\tau }_{\text{c}}/\left[1+\left({\tau }_{\text{c}}{}^2\times {\omega }^2\right)\right]$$ (Eq. 3)

where τ_c is the correlation time.⁶¹ The J₀ = τ_c which is much larger than J_{ω I}, based on τ_c ≈ 10⁻⁸ s, the experimentally-based estimate of the correlation time for chain motion associated with protrusion and on ω_I ≈ 6 × 10⁸ s⁻¹.^76,77 We estimate Ω_Mn ≈ 10⁻² based on a protruded ¹³C being near ~2 lipid headgroups and the experimental Mn²⁺ headgroup occupancy ≈ 0.005. For lipid without Fp, Ω_prot ≈ 0.01 in simulations and the Γ₂ ≈ 10 s⁻¹ from our experimental data (Table 2). The resulting calculated r ≈ 4 Å is plausible for the ¹³C nuclei of a chain that protrudes into the headgroup region of the membrane.

One notable trend of Fig. 8 and Table 2 is the attenuation of Γ₂ as the ¹³C site moves closer to the chain terminus. This trend holds for samples both without and with Fp. We have hypothesized that Γ₂ ∝ Ω_prot and further hypothesize that protrusion of a specific -CH₂ group also means protrusion of the -CH₂ groups that are closer to the lipid glycerol group. This second hypothesis is based on the location of the glycerol group close to the phosphate group and the chemical bonding of the acyl chain (Fig. 3). For the six ¹³C NMR signals assigned to two -CH₂ sites with sites numbered x and y, respectively (Fig. 5), the average number of protruded -CH₂ groups (n) is calculated:

$$\text{n}=[(\text{x}+\text{y})/2]-1$$ (Eq. 4)

and protrusion of each -CH₂ group is hypothesized to require free energy ΔG_prot so that Γ₂ depends on n as:

$${\Gamma }_2(\text{n})={\Gamma }_2(0)\times \text{exp}\left[-\left(\text{n}\times \Delta {\text{G}}_{\text{prot}}\right)/{\text{k}}_{\text{B}}\text{T}\right]$$ (Eq. 5)

where Γ₂ (0) is the parameter for the rate when no -CH₂ are protruded. Fig. 9 displays fitting with Eq. 5 of the experimental Γ₂ (n) data vs. n, with separate fittings of data without and with Fp. The best-fit ΔG_prot/ k_BT are 0.249 ± 0.018 without Fp and 0.266 ± 0.016 with Fp. The Eq. 5 model is supported both by the quantitative similarity of the two values and the semi-quantitative agreement with the 0.25–0.50 k_BT range for ΔG_prot/ k_BT in one of the simulations.⁴⁹ The best-fit Γ₂ (0) are 27.9 ± 1.8 s⁻¹ without Fp and 47.3 ± 2.6 s⁻¹ with Fp.

Figure 9.

Plots and exponential decay fittings of Γ₂ vs. average number (n) of protruded -CH₂. For each peak corresponding to signals from two -CH₂ sites that are numbered x and y (Figs. 3 and 5), this average number is calculated as [(x+y)/2] – 1, e.g. 2 for the 3,3’ peak and 14 for the 16,14’ peak. Data are displayed for samples without and with Fp. Separate fittings are done for each sample type using Γ₂ (n) = Γ₂ (0) × exp(−n × κ) with Γ₂ (0) and κ as fitting parameters and κ = ΔG_prot/k_BT. Best-fit values with uncertainties in parentheses are: (1) without Fp, Γ₂ (0) = 27.9(1.8) s⁻¹ and κ = 0.249(18); and (2) with Fp, Γ₂ (0) = 47.3(2.6) s⁻¹ and κ = 0.266(16).

The molecular dynamics simulations from different groups show that the large enhancement of Ω_prot is primarily for lipids next to the Fp whereas the Ω_prot of more distant lipids is similar to lipids without Fp. For samples with Fp, the ~10⁻⁸ s time for lateral diffusion of a lipid molecule between Fp-neighboring and more distant locations is much more rapid than the 1/R₂ relaxation time, so the Γ_2,Fp will be a weighted average of the larger Fp-adjacent and smaller more distant values, Γ_2,neighbor and Γ_2,distant, respectively. For “q” lipid molecules neighboring a Fp and the ~3 mole% Fp of a sample:

$${\Gamma }_{2,\text{Fp}}=\left(0.03\times \text{q}\times {\Gamma }_{2,\text{neighbor}}\right)+[1-(0.03\times \text{q})]\times {\Gamma }_{2,\text{distant}}$$ (Eq. 6)

which is algebraically rewritten as:

$${\Gamma }_{2,\text{neighbor}}/{\Gamma }_{2,\text{distant}}=\left[{\Gamma }_{2,\text{Fp}}/{\Gamma }_{2,\text{distant}}-1+(0.03\times \text{q})\right]/(0.03\times \text{q})$$ (Eq. 7)

Using the best-fit Γ_2,Fp(0) and Γ_2,noFp(0) values and the approximation Γ_2,distant(0) ≈ Γ_2,noFp(0):

$${\Gamma }_{2,\text{neighbor}}/{\Gamma }_{2,\text{distant}}=\left[{\Gamma }_{2,\text{Fp}}(0)/{\Gamma }_{2,\text{noFp}}(0)-1+(0.03\times \text{q})\right]/(0.03\times \text{q})$$ (Eq. 8)

The Γ_2,Fp(0)/Γ_2,noFp(0) = 1.70 ± 0.14. The value of q will depend on the Fp surface area that contacts lipids and this area will vary with location and orientation of the Fp in the membrane. There is also the possibility that q is reduced because of increased protrusion probability is mostly for a subset of neighboring lipids with spatiallyspecific Fp interactions. A reasonable possible range of q values is 1 to 8, with q = 8 based on interfacial Fp location and ~4× greater cross-sectional area for Fp vs. lipid.^47,48 From Eq. 8, Γ_2,neighbor/Γ_2,distant = 24.3 ± 4.8 for q =1 and 3.92 ± 0.60 for q = 8. This matches the range of Γ_2,neighbor/Γ_2,distant ratios that were observed in different simulations.^49–51 The inverse correlation between the experimentally-derived Γ2,neighbor/Γ2,distant ratio and q is consistent with the larger simulation-derived Γ2,neighbor/Γ2,distant ratios for transmembrane vs. membrane surface location of the Fp, and the likely larger Fp lipid-contacting area and q value for the surface location.^49,51 In addition, large Γ_2,neighbor/Γ_2,distant is observed in a different simulation in which the effective q ≈ 1 because of the strong correlation between protrusion and a hydrogen bond between one the four N-terminal residues of the Fp and a lipid phosphate oxygen, with consequent headgroup intrusion into the membrane interior.⁵⁰

Besides the aforementioned simulation observations that increased chain protrusion may be associated with Fp/phosphate hydrogen bonding and headgroup intrusion, protrusion may also be augmented by solvation of lipid chains by hydrophobic Fp sidechains at the membrane surface. This could be part of the basis for much greater Ha2-induced intervesicle lipid mixing at endosomal pH 5.0 vs. physiologic pH 7.4.¹⁵ This effect is observed with POPC vesicles for which there isn’t bulk Ha2/vesicle electrostatic energy. At both pH’s, Fp is a mixture of closed and semi-closed helical hairpin structures, and semi-closed has significantly greater hydrophobic surface area.⁴⁸ There is higher semi-closed population at pH 5 vs. 7 and therefore larger Fp hydrophobic surface area.

Linewidth and * peak analysis also support protrusion.

For the 2,2’ and 3,3’ signals, there are larger Mn²⁺-associated increases in linewidth, Δ[Δν], for samples with vs. without Fp (Table 3). The difference is less apparent for two-site signals from ¹³C nuclei closer to the chain termini, and these observations correlate with the trend of Γ_2,Fp – Γ_2,NoFp (Fig. 8 and Table 2). For the two-site signals, Table 4 provides the inhomogenous contributions to linewidths, which are calculated by Δν_inhom = Δν − (R₂/π). Table 4 also provides for each signal the RMSD and average value calculated from the data of the four samples. The typical RMSD is ±2 Hz and the average values are in the 26–37 Hz range. For our spectra, the individual contributions to the signal from each site are typically unresolved, except for 9,10 which has partially-resolved C9 and C10 contributions with respective higher and lower shifts. For earlier ¹³C NMR spectra of POPC with somewhat narrower linewidths than our spectra, there aren’t resolved shift differences (Δδ) between sites for the 2,2’, 16,14’, and 17,15’ signals, and the Δδ are ~0.09, 0.13, and 0.36 ppm for the 3,3’, 8,11, and 9,10 signals, respectively.⁶² For the present study, the 2,2’, 16,14’, and 17,15’ signals have Δν_inhom ≈ 27 Hz which is likely due to 20 Hz exponential line broadening and shimming. There are larger Δν_inhom of ~32 and 37 Hz for the 3,3’ and 8,11 signals, respectively, and the ~5 and 10 Hz increases over the Δν_inhom ≈ 27 Hz baseline value correlate semi-quantitatively with the Δδ values of ~9 and ~13 Hz, respectively. For the 9,10 signal, the Δν_inhom ≈ 32 Hz is smaller than would be expected from the Δδ ≈ 36 Hz. This anomaly is likely a consequence of the larger C9 vs. C10 contribution to the 9,10 signal. There is greater ¹H-¹³C cross-polarization for C9 vs. C10 because of the smaller vs. larger site mobility that was previously described by differences in site order parameters.⁶²

The * peak is a superposition of signals from eighteen ¹³C sites in the middle of the two chains, 4–7, 12–15, and 4’−13’. Table 2 lists the best-fit R₂ and Γ₂ values both for full integration of * intensities and for integration ranges denoted *1, *2, and *3 for which a subset of the ¹³C sites make the largest contributions, respectively 6’−9’; 7,10’,11’; and 4–6,12–15,4’,5’,12’,13’.⁶² The Γ₂ values from the * fittings are similar to those of 8,11 and 9,10, and are intermediate between the larger 2,2’ and 3,3’ Γ₂ values and smaller 16,14’ and 17,15’ values. The *1 and *2 integrations are dominated by signals from ¹³C sites closer to CO whereas the *3 integration has large contributions from ¹³C sites closer to the chain terminus. These different locations correlate with the Γ_2,Fp > Γ_2,NoFp for *1 and *2 and for 2,2’; 3,3’; 8,11; and 9,10 fittings, whereas Γ_2,Fp ≈ Γ_2,NoFp for *3; 16,14’; and 17,15’ fittings. The Mn²⁺-associated increases of the linewidths of the * peaks are similar to those of resolved ¹³C sites in the middle and terminal regions of the chains, and the increase for * is a little larger with vs. without Fp (Table 3). The * linewidth has substantial inhomogeneous contribution from the superposition of unresolved signals from many ¹³C sites.

Comparison between PRE and other experimental approaches to probe chain motions relevant to fusion.

The hypothesis of increased protrusion induced by Fp was initially proposed based on molecular dynamics simulations.^49–51 The typical Ω_prot,NoFp ≈ 0.01 and the Ω_prot,Fp was larger but still < 0.2, i.e. protrusion was always a low-probability state. It is therefore anticipated that the observables from some experimental approaches have large contributions from the high-probability unprotruded chains. The Mn²⁺ paramagnetic relaxation enhancement (PRE) approach of the present study has the advantage that the Γ₂ observable is dominated by the small population of protruded chains, based on: (1) Mn²⁺ is predominantly bound to the lipid headgroup phosphate; (2) the 〈r⁻⁶ 〉 dependence of Γ₂; and (3) the larger τ_c for protrusion vs. other chain motions.⁷⁷ Some other approaches wouldn’t have this advantage. As one example, protrusion would also affect lipid ¹³C-³¹P dipolar coupling that can be measured by NMR.²⁶ The coupling is proportional to 〈⁻³〉 where ρ is the internuclear distance. For the 2,2’ and 3,3’ sites, the 〈⁻³〉 for membrane without vs. with Fp are estimated to be ~2.01 × 10⁻³ vs. ~2.26 × 10⁻³ Å⁻³, i.e. only ~12% increase, as calculated from ρ_C-P,unprot ≈ 8 Å, ρ_C-P,unprot ≈ 5 Å, Ω_prot,NoFp ≈ 0.01, and Ω_prot,Fp ≈ 0.05. Another consideration is that investigation of protrusion by simulation has been done in fluid rather than gel membrane phases, in part because fluid phases are similar to those of membranes in viral fusion. Lipids in fluid phases experience rapid lateral diffusion and also other large-amplitude motions.^68,77 These motions are usually advantageous for the PRE approach because they result in smaller R₂’s that can generally be measured more accurately than larger R₂’s. The motions also reduce dipolar couplings; however, the NMR spectra often have lower signal-to-noise when measuring smaller vs. larger couplings, so smaller couplings are less accurately-determined.⁷⁸

Splay is the term used to describe the large-amplitude movement by the terminal region of the lipid chain into the headgroup region. Splay may be relevant to fusion and has been detected using the ¹H-¹H NOESY NMR cross-relaxation rate between the terminal methyl and the headgroup nuclei.⁶⁵ The NOESY rates have been positively-correlated with the extents of fusion between vesicles with transmembrane peptides.⁷⁹ However, these vesicle fusion rates are ~3 × 10⁻⁴ s⁻¹ which are ~1000× smaller than rates of vesicle fusion induced by Fp’s.⁴⁸ These results suggest that splay is a less important motion for fusion than the protrusion of chain -CH₂ groups closest to the glycerol linkage, i.e. the motion probed in the present study. This conclusion is supported by the statement on p. 5 of Ref. 49, “…our simulations predict the effect on tail protrusion to be most profound in the upper region of the acyl chain, so a difference in tail exposure might be suboptimal probe for fusion peptide activity.”⁴⁹

Conclusions

The present study presents convincing experimental data that support a large increase in lipid acyl chain protrusion caused by the membrane-bound Fp domain of the influenza virus Ha2 protein. Increased protrusion had previously been observed in computational simulations and may play an important role in fusion between the viral and the endosome membranes. In particular, protrusion may accelerate the transition from the initial separate apposed membranes to the stalk intermediate that connects and is contiguous with the outer leaflets of the two bodies. For the present study, protrusion was detected by larger Mn²⁺-associated increases in transverse relaxation rates of lipid chain ¹³C nuclei for samples with vs. without Fp. Analysis of the Γ_2,Fp vs. Γ_2,NoFp rate increases resulted in a calculated ratio Γ_2,neighbor/Γ_2,distant in the range of 4–24 where the ratio is for lipids neighboring vs. more distance from Fp. The ratio values within this range are inversely-correlated with the number of neighboring lipids. The experimental range is similar to the range in simulations for increased protrusion probability of a lipid neighboring vs. more distant from the Fp. For samples either with or without Fp, the Γ₂ values are well-fitted by an exponential decay as the ¹³C site moves closer to the chain terminus. The decay correlates with a positive free-energy of protrusion that is proportional to the number of protruded -CH₂ groups. The experimentally-determined free energy per -CH₂ is ~0.25 k_BT which matches the value in one of the simulations. Overall, the NMR data support one major fusion role of the Fp to be much greater chain protrusion with highest probability for chain regions closest to the headgroups.

Highlights.

• NMR PRE shows 4–24 times higher chain protrusion probability for lipid next to Fp

• Fp is 20-residue influenza virus hemagglutinin fusion peptide

• Protrusion increase from Fp is greatest for chain -CH₂ closest to lipid headgroup

• Higher experimental protrusion probability matches MD simulations

• More protrusion should increase rate of stalk formation in membrane fusion

Abbreviations

CP

cross-polarization

Δδ

shift difference between sites of a two-site peak

Δν

experimental FWHM linewidth

Δν_inhom

inhomogeneous contribution to linewidth

Ed

ectodomain

EPR

electron paramagnetic resonance

FHa2

Ha2 ectodomain

Fp

fusion peptide

FsSu

spike fusion subunit protein

FWHM

full-width at half-maximum

Γ₂

difference between R₂ for samples with vs without Mn²⁺

gp41

HIV FsSu protein

Ha2

hemagglutinin protein subunit 2 (influenza virus FsSu)

HIV

human immunodeficiency virus

k_B

Boltzmann’s constant

MAS

magic angle spinning

n

number of protruded -CH₂ groups

NMR

nuclear magnetic resonance

Ntr

N-terminal region of FsSu

PC:PG

POPC:POPG (4:1)

Ω_Mn

probability that a protruded chain is near a Mn²⁺

Ω_prot

probability of lipid chain protrusion

POPC

1-palmitoyl-2-oleoyl-phosphatidylcholine

POPG

1-palmitoyl-2-oleoyl-phosphatidylglycerol

PRE

paramagnetic relaxation enhancement

q

number of lipid molecules next to Fp

r

¹³C-Mn²⁺ distance

R₂

NMR transverse relaxation rate

RbSu

spike receptor-binding subunit protein

rf

radiofrequency

S

integrated NMR peak intensity

τ

dephasing time