Influence of Membrane Composition on the Binding and Folding of a Membrane Lytic Peptide from the Non-Enveloped Flock House Virus

Abstract

Using a combination of coarse-grained and atomistic molecular dynamics simulations we have investigated the membrane binding and folding properties of the membrane lytic peptide of Flock House virus (FHV). FHV is an animal virus and an excellent model system for studying cell entry mechanisms in non-enveloped viruses. FHV undergoes a maturation event where the 44 C-terminal amino acids are cleaved from the major capsid protein, forming the membrane lytic (γ) peptides. Under acidic conditions, γ is released from the capsid interior allowing the peptides to bind and disrupt membranes. The first 21 N-terminal residues of γ, termed γ₁, have been resolved in the FHV capsid structure and γ₁ has been the subject of in vitro studies. γ₁ is structurally dynamic as it adopts helical secondary structure inside the capsid and on membranes, but it is disordered in solution. In vitro studies have shown the binding free energies to POPC or POPG membranes are nearly equivalent, but binding to POPC is enthalpically driven, while POPG binding is entropically driven. Through coarse-grained and multiple microsecond all-atom simulations the membrane binding and folding properties of γ₁ are investigated against homogeneous and heterogeneous bilayers to elucidate the dependence of the microenvironment on the structural properties of γ₁. Our studies provide a rationale for the thermodynamic data and suggest binding of γ₁ to POPG bilayers occurs in a disordered state, but γ₁ must adopt a helical conformation when binding POPC bilayers.

1. Introduction

Our current understanding of the mechanisms employed by non-enveloped animal viruses to penetrate the membranes of their host cells is lacking as compared to our knowledge of enveloped virus cell entry mechanisms. Broadly speaking, enveloped animal viruses employ membrane fusion mechanisms mediated via viral surface proteins, called fusion proteins. The fusion proteins of enveloped viruses undergo large conformational changes in response to receptor binding or environmental changes, which allows the fusion proteins to insert into the host membrane. The fusion proteins mediate the merger of the viral and host membrane by reducing the energy barrier to the fused state, in which the virus can shed its membrane and pass through the host membrane.^1,2 On the other hand, non-enveloped viruses lack a membrane and therefore cannot perform membrane fusion to enter cells or exit from membrane bound cellular compartments (e.g. endosomes). The mechanisms employed by these non-enveloped viruses to cross cellular membrane barriers are less well understood. Many non-enveloped viruses contain a specialized capsid protein(s) that mediates membrane penetration. It has been proposed that these “penetration proteins” are inactive precursors sequestered inside the capsid. They become activated under specific cellular conditions, leading to their externalization from the capsid interior and exposure to cellular membranes. Recent studies have indicated this general feature of non-enveloped viruses is spread across many non-enveloped viral families including, Picornaviridae, Polyomaviridae, Reoviridae, Parvoviridae, Adenoviridae, Alphatetraviridea and Nodaviridae.^3,4

Flock House virus (FHV) is one of the simplest and smallest non-enveloped viruses and provides a model system to study non-enveloped animal viruses.⁵ FHV infects insects and is a member of the Nodaviridae family and has a small (4.5 kb) bipartite positive sense RNA genome. The FHV capsid protein contains 407 residues, which assemble into a T=3 icosahedral capsid containing 180 subunits. The initial assembled particle is immature and must undergo an autocatalytic cleavage event to mature and become infective.⁶ The cleavage happens between residue 363 and 364 and is coordinated by residue Asp75⁷ resulting in the major capsid protein β (residues 1–363) and the γ peptide (364–407). The structure of the mature capsid has been determined by X-ray crystallography at 2.7 Å resolution which shows γ is non-covalently associated with the interior capsid surface (Fig. 1A).⁸ FHV enters cells through the endocytic pathway where the virus encounters an acidic environment. The low pH of the endosome is believed to promote externalization of γ peptides from the capsid interior, which can then disrupt the endosomal membrane creating a channel for the release of the virus or viral RNA into the cytoplasm. This mechanism is supported by in vitro liposome dye leakage experiments which show FHV to have maximal lytic activity at pH 6.⁹ It has been established that cleavage defective mutants display almost no infective ability,⁶ and infectivity of non-cleaving mutants can be rescued by coinfection with wild-type virus-like particles (VLPs),¹⁰ establishing the criticality of cleaved γ peptides to viral infection.

Figure 1.

FHV Structures. A) A view from inside the FHV capsid, looking down a five-fold axis. The pentameric capsid subunits are shown in blue and the γ-peptides in yellow (PDB ID: 4FTB). B) The γpeptide is displayed with side chains drawn and colored as follows, pink=hydrophobic, green=uncharged polar, blue=positively charged, red=negatively charged. The residue numbering and residue types are listed near the corresponding residues.

Multiple lines of data point toward the γ-peptide being externalized from the FHV capsid and then acting independently (apart from the capsid) to perform membrane disruption.^9,11,12 Therefore, several in vitro studies have investigated the properties of the FHV γ-peptide in solution and in the presence of bilayers to provide mechanistic insights into the membrane disruption process. Many of these studies have focused on the region of γ that was resolved in the capsid crystal structure (Fig. 1), which are the first 21 N-terminal residues. This minimal version of γ is termed γ₁, and it retains membrane lytic activity, though at a weaker level than the full length γ.^13,14 The solution behavior of γ₁ has been examined by circular dichroism (CD) and nuclear magnetic resonance (NMR), and it is found that the structure is largely disordered in aqueous environments.^12,15 In the NMR study, it was shown that increasing the hydrophobicity of the solvent by addition of trifluoroethanol (TFE) increases the ordering of the peptide, leading to a partially α-helical conformation at 50% TFE.¹² Biophysical experimental studies showed γ₁ spontaneously partitions into large unilamellar vesicles (LUVs) made of either 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) or 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoglycerol (POPG) lipids.¹⁵ CD titration experiments demonstrated that γ₁ orders upon addition of POPC or POPG LUVs, increasing in helical content up to 75%. Attenuated total reflectance infrared spectroscopy (ATR-IR)¹³ as well as thermodynamic analysis,¹⁵ support a model in which γ₁ orients laterally, burying about half the peptide surface area into the hydrophobic core of the bilayer, while maintaining partial exposure to the solvent. Native tryptophan fluorescence indicates an insertion depth for γ₁ to be approximately 1 nm from the bilayer center and deeper insertion occurs into POPC than POPG bilayers.¹³ Thermodynamic analysis using isothermal titration calorimetry (ITC) revealed that even though the γ₁ binding free energies to POPC (−7.5 kcal/mol) or POPG (−8.0 kcal/mol) are nearly equivalent, the association with negatively charged POPG is entropy driven (ΔH =10.2 kcal/mol; ΔS = 61.1 cal/mol·K), whereas the association with neutral POPC is enthalpy dominated (ΔH = −4.1 kcal/mol; ΔS = 11.4 cal/mol·K)¹³. The activity of the peptides in membrane lysis has been examined by liposome dye release studies and the activity is much higher against POPC than POPG; a ~10-fold higher peptide:lipid ratio was needed to lyse POPG vesicles at comparable levels to POPC vesicles.¹³ There is minimal pH dependence to the lytic activity of the free peptides, indicating the release mechanism is pH-triggered while the membrane binding/insertion/disruption processes appear pH-independent.^9,12

Through these previous studies a link between the membrane composition and the activity of these peptides has been established, though how this relates to the structural and dynamic features of the peptides remains unclear. There are, however, definite specifications involved that govern the binding propensity of the membrane active lytic peptides of the FHV. These governing factors are both kinetic and thermodynamic which are sensitive to the microenvironment. To probe the structural dynamics and energetics of FHV γ₁, we have employed multi-scale molecular simulations, using a combination of coarse-grained (CG), all-atom (AA) and enhanced sampling molecular dynamics (MD) simulations. Through this approach we are able to examine 1) the structural and energetic properties of γ₁ in solution, 2) the structure and energetics of γ₁ binding to bilayers of varying composition, and 3) the structural evolution and membrane insertion properties of γ₁ interacting with bilayers of varying composition.

2. Methods

We performed multi-scale simulations to characterize the structure and dynamics of γ₁ peptide (21 residues: ASMWERVKSIIKSSLAAASNI) in aqueous solution and in membrane environments. All system configurations and snapshots were visualized using VMD software.¹⁶

2.1. Well-Tempered Metadynamics (WT-MetaD)

To evaluate the conformational dynamics of γ₁ in explicit solvent in neutral pH conditions we applied the well-tempered metadynamics method¹⁷ implemented in NAMD version 2.10.¹⁸ In brief, this method adds time dependent biases along a chosen collective variable(s) (CV) that is used to accelerate the system dynamics in predefined set of CV. The free energy surface (FES) at time t is derived from the histogram of the occupancy of the CV (N(s,t)) in the simulations

$$F(s,t)=-(T+\mathrm{\Delta }T)ln\left(1+\frac{\omega N(s,t)}{\mathrm{\Delta }T}\right)$$ (1)

where, F is the FES, s is the vector of the CV and ΔT is a bias temperature that regulates the extent of the energy barriers that can be overcome. For our simulations we used a Gaussian width of 0.2 Å, a Gaussian height of 0.2 kcal/mol, a bias temperature ΔT = 3000 K, and an energy deposition frequency of 1 ps⁻¹. We have performed WT-MetaD simulations on FHV γ₁ using an end-to-end distance CV, as defined by the difference between the center of geometry of the first two and last two residues backbone atoms. The simulations were run for 190 ns using the CHARMM22 force field¹⁹ with CMAP correction.²⁰ The simulations were performed in the NPT ensemble at 300 K and 1 ATM, with period boundary conditions in X,Y, and Z-directions. A time step of 2 fs was used with a non-bonded cutoff distance of 12 Å, which were smoothly shifted to zero after 10 Å, long range electrostatics were computed with the particle mesh Ewald (PME)²¹ every 4 fs.

2.2. Coarse-Grained (CG) simulations

We performed six different γ₁-bilayer CG simulations using the MARTINI force field^22–24 with the GROMACS 4.6.5 software package.²⁵ Proteins in MARTINI are not capable of spontaneously forming or breaking secondary structure (SS), therefore we performed simulations both when γ₁ was restrained to an α-helical conformation and when γ₁ is in a disordered configuration. The description of the systems has been summarized in Table 1. System I and II consist of 270 POPC CG lipid molecules, while System III and IV comprise of 288 POPG CG lipid molecules. Furthermore, system V and VI consist of a 50:50 mix of POPC:POPG, with 100 CG lipid molecules each of POPC and POPG. The bilayer for each system was generated using the insane.py script.²⁶ The bilayers were first energy-minimized and then equilibrated in the NPT ensemble for 400 ns with polarizable water²⁷ for the solvent. The system was ionized with a proper number of Na⁺ and Cl⁻ ions to make the systems electrically neutral at 0.1 M salt concentration. The simulations were done starting with different initial velocity distributions of γ₁ peptide placed ~ 4 nm from the pre-equilibrated bilayer center, with γ₁ helical axis parallel to the bilayer. The initial FHV γ₁ peptide structure was obtained from the crystal structure, PDB ID: 4FTB.

Table 1.

Simulation System Details. The Initial γ₁ Configuration column describes which CG simulation end state was used to initiate the AA simulations.

System	Lipid Composition	CG Spontaneous Binding Simulations		AA bilayer bound folding simulations
		SS	γ1 binds	Initial γ1 Configuration	Simulation Time	Final% Helicity
I	POPC	on	yes	System I	1 μs	78
II	POPC	off	no	n/a	n/a	n/a
III	POPG	on	yes	System III	1 μs	73
IV	POPG	off	yes	System IV	1 μs	0
V	POPC/POPG (50/50)	on	yes	System V	1 μs	72
VI	POPC/POPG (50/50)	off	yes	System VI	1 μs	16
VII	POPC	n/a	n/a	System VI	1 μs	63
VIII	POPG	n/a	n/a	System VI	1 μs	23

After the initial setup, each system was energy minimized using the steepest descent algorithm. All production runs were simulated in the NPT ensemble using Berendsen coupling scheme²⁸ with the temperature maintained at 323 K and pressure kept at 1.0 bar with semi-isotropic coupling. The time constants for the pressure and temperature couplings were 3.0 and 1.0 ps, respectively, and the compressibility value was 3×10⁻⁵ bar⁻¹. Electrostatics were computed by the PME method using a relative dielectric constant of ε_r = 2.5, as is the suggested protocol for MARTINI simulations using polarizable water. The simulation was performed utilizing periodic boundary conditions (PBC), with a time step of 20 fs. The nonbonded Lennard-Jones (LJ) interactions were cut off at 1.2 nm. These simulations were run for 1.5 μs. GROMACS analysis tool g_dist was used to calculate the difference between the center of mass (COM) distance between the γ₁ peptide and the bilayer.

To generate systems VII and VIII, which involved converting a mixed bilayer to a pure bilayer, the following protocol was used. In the CG representation both POPC and POPG contain 13 beads, with only one different bead type differentiating between the choline and glycerol moiety in the head group region of POPC and POPG, respectively. Specifically, the bead NC3 (type Q0; charge +1) of POPC was changed to bead GL0 (type P4; charge zero) of POPG and vice-versa. Therefore, the bound state configuration of system VI (mix, SSoff) was converted to either pure POPC (System VII) or pure POPG (System VIII), without altering the peptide configuration. Both systems, VII and VIII were re-neutralized with a proper number of Na⁺ and Cl⁻ ions to make the systems electrically neutral at 0.1 M salt concentration. Each system was energy minimized using the steepest descent algorithm before being converted to an all-atom representation.

Membrane binding free energies were calculated in the MARTINI model for Systems I–VI. With the exception of System II (POPC, random coil γ₁), we were able to observe spontaneous bilayer association when γ₁ is initially placed in the bulk solvent. We took the final configuration from the 1.5 μs binding simulation and perform steered MD (SMD) to generate intermediate conformations for umbrella sampling. The SMD runs were performed by applying a harmonic biasing force to the COM Z-coordinate distance between the peptide and the bilayer. A force constant of 1000 kJ mol⁻¹ nm⁻² was used and the center of the bias was moved at a rate of 0.001 nm/ps. Positional harmonic restraints with a force constant of 1000 kJ mol⁻¹ nm⁻² were applied on the lipid atoms during the SMD and intermediate states were extracted in 0.1 nm spacing ranging up to 3 nm displacement from the bound state. Positional restraints were released and umbrella sampling was performed on the COM Z-coordinate distance between the bilayer and the peptide for 300 ns with an umbrella potential of 2000 kJ mol⁻¹ nm⁻² for each window. The data from the last 100 ns of each umbrella window was analyzed using the weighted histogram analysis method (WHAM)²⁹ implemented in the GROMACS g_wham tool, to calculate the potential of mean force (PMF) of peptide-bilayer binding. The PMF for System II was generated in the same manner as all other systems, except that the initial configuration for the SMD was the same as System I, but with the SS restraints turned off (random coil). All reported CG times are the actual simulation times without any scaling factor applied; a common practice is to estimate the “true” time by multiplying the simulation time by a factor of 4.

2.3. All-Atom (AA) simulations

Bound configurations from the CG simulations were transformed to AA descriptions for systems I–VIII to allow for more detailed structural interrogation of the binding mode and folding characteristics of γ₁. This was achieved by “back-mapping” to atomic coordinates, utilizing the geometrical approach by Teileman³⁰ to convert from MARTINI to an AA description for simulations in the CHARMM36 forcefield.^31,32

For FHV γ₁, 21 CG beads were mapped to 328 atoms. The atomistic γ₁ had a charge of +2 with a protonated N-terminus and a negatively charged C-terminus. For the POPC lipid, 134 atoms were mapped from the 13 CG beads, and for POPG lipid 127 atoms were mapped from the 13 CG beads. The three bead CG polarizable water was mapped to the TIP3P water model. The ions Na⁺ and Cl⁻ were also back-mapped into the AA configuration. Each back-mapped system was minimized using the steepest descent algorithm. Following the minimization, the systems were equilibrated in a constant volume, constant temperature (NVT) ensemble for 10 ns. Post equilibration, the production simulations were carried out for 1 μs in the NPT ensemble. Parrinello-Rahman semi-isotropic pressure coupling was used to maintain the pressure at 1.0 bar and the temperature was maintained at 303 K using the velocity (v)-rescale method. The time constants for the pressure and temperature couplings were 1.0 ps and the compressibility value was 4.5 × 10⁻⁵ bar⁻¹. Non-bonded short range interactions were cutoff at 1.2 nm with the LJ interaction being force-switched starting at 1.0 nm. Long-range electrostatics were calculated by PME, bonds involving hydrogen atoms lengths were constrained with the LINCS constraint algorithm. The simulations were performed using PBC in X,Y and Z-directions with a time step of 2 fs.

2.4. Trajectory analysis

The trajectories were analyzed using multiple analysis tools from the GROMACS, MDTraj³³ and CHARMM³⁴ packages. Total helicity was computed every 1 ns using the COOR SECS function in CHARMM with a cutoff distance of 2.6 Å. Orientation (tilt) angle of γ₁ was measured by calculating the arccosine of the dot product between the bilayer normal (0, 0, 1) and the normalized vector along the helical axis of γ₁. The helicity per residue was computed using the DSSP³⁵ tool of MDTraj. Distance calculations between COM of bilayer phosphate group and peptide (in Z-direction) were done using the g_traj tool in GROMACS. Also, the g_mindist tool from GROMACS was utilized to calculate the number of contacts γ₁ makes with lipid molecules within a 5 Å radius. The ion release entropy calculation was based upon constructing a normalized probability distribution for Na⁺ ions in the range of 0 to 2.8 nm from the bilayer. This was done by calculating the minimum distance (R) for each Na⁺ ion to it’s nearest lipid phosphate in the upper leaflet and we assumed that p(R)≈ p(Z) for the histogramming.

3. Results

3.1 γ₁ is partially disordered in solution

While the γ₁ peptide is known to be well ordered inside the viral capsid, it undergoes an order-disorder transition, when the peptide exits the capsid and is free in solution. However, structural insight into the disordered state or insight into how energetically unfavorable it is to regain structure is lacking. We have conducted AA WT-MetaD simulations¹⁷ in explicit solvent to calculate the free energy profile of structural transitions of the γ₁ peptide in an aqueous environment. The simulation was conducted for 190 ns and the FES displays good convergence (Fig. S1). We observe that the energy landscape along the end-to-end distance variable, shows a compact structure to be the free energy minimum (Fig. 2). Based upon our definition, a full helix would have an end-to-end distance of ~29 Å, which is not a stable region in the landscape. We have performed structural clustering at several points along the free energy profile and have found the average helicity in the different clusters ranges from 29% to 43%. The cluster centroids, shown in Fig. 2, show that helicity is generally maintained in the N-terminal region, while the C-terminal region adopts a random coil configuration. The results are in reasonable agreement with CD measurements, which have predicted ~20% α-helical and 40% random coil for γ₁ in buffer.³⁶

Figure 2.

Free energy profile of γ₁ from WT-MetaD. Six cluster centroid structures are shown, the clusters were constructed based upon the end-to-end CV. The following CV ranges were used in constructing the clusters: 16–18 Å, 24–26 Å, 30–32 Å, 35–37 Å, 38–40 Å, 42–44 Å.

3.2. γ₁ binding to membranes is dependent on peptide structure and lipid composition

The spontaneous binding of a small peptide from solution to the membrane surface can be a slow process due to dependencies on peptide folding and lipid diffusion rates.³⁷ Therefore the peptide membrane binding process may not be well suited for investigation by AA simulations. The MARTINI CG model^22,24 offers an alternative model in which microsecond timescales can be reached with reasonable computational expense. The model has been shown to be effective at predicting membrane partitioning free energies of short peptides,³⁸ but the model is not designed to capture secondary-structure (SS) changes in a protein. Therefore, as a way to circumvent this limitation in MARTINI, we performed membrane binding simulations both when no secondary structure restraints were imposed on γ₁ (referred to as SSoff), and when γ₁ was restrained to an α-helical conformation (referred to as SSon). We examined the spontaneous membrane binding of γ₁ as function of SS restaints and as a function of lipid composition of the bilayer. The six CG binding simulations are described in Table 1, and include bilayers composed purely of POPC (zwitterionic), POPG (negatively charged) or a heterogeneous bilayer containing an equal number of POPC and POPG lipids, referred to as a mixed (mix) bilayer.

The center of mass (COM) separation between γ₁ and the bilayer was monitored (Fig. 3) in simulations in which γ₁ was initiated in the solvent, ~4 nm away from the bilayer. We find that γ₁ only transiently binds to a POPC bilayer when it is in a disordered state (System II). However, upon applying the SS restraints (System I) a helical γ₁ binds to the neutral POPC bilayer after ~500 ns, as shown in Fig. 3. We next examined the spontaneous binding of γ₁ to charged bilayers as described by Systems III–VI in Table 1. We observed that by introducing negatively charged POPG lipids, γ₁ would bind bilayers in both helical and random coil configurations, whereas binding to neutral POPC was dependent on formation of helical structure. Fig. 3 illustrates that γ₁ binds pure POPG or a mixed POPC/POPG bilayer much faster (under 200 ns) than it binds a POPC bilayer. This feature can be attributed to the electrostatic attraction between the positively (+2) charged γ₁ and the presence of net negative charge in the bilayer. It may also be noted from Fig. 3 that the helical γ₁ peptide inserts deeper into the POPC bilayer than the disordered peptides insert into the charged bilayers.

Figure 3.

The COM separation between γ₁ and the bilayer. The peptide was modeled in an unstructured state for the binding against PC (red), PG(green) and the mixed bilayer (pink), as well as in a helical state against a PC bilayer (blue). Stable associations were observed for all systems except for the unstructured peptide against PC (red). For clarity purposes the data was smoothed by performing 6 ns running averages.

We further characterized the γ₁ membrane binding properties by performing umbrella sampling simulations to obtain the binding free energy for Systems I–VI listed in Table 1. Since disordered γ₁ (SSoff) never forms a stable complex with a POPC bilayer, the bound state of System I was used as the starting configuration for the umbrella sampling of System II. For all other systems the initial configuration and SS restraints were consistent with the spontaneous binding simulations. Fig. 4 presents the 1D free energy profile (potential of mean force, PMF) for the six systems that were umbrella sampled. We observe a ΔG of binding of −1.5 kcal/mol, −11.0 kcal/mol, −7.5 kcal/mol with POPC, POPG and the mixed bilayers, respectively, when SS restraints are not imposed (SSoff). However, the bilayer-γ₁ binding with a helical γ₁ (SSon) show considerably stronger binding for all systems. With SSon we observe a ΔG of binding of −15.0 kcal/mol, −22.5 kcal/mol, −18.5 kcal/mol with POPC, POPG and the mixed bilayer, respectively.

Figure 4.

PMF of γ₁ binding to membranes from umbrella sampling the COM displacement from the bound configuration. For all membrane compositions, the umbrella sampling was performed both with γ₁ in a helical configuration (SSon) and in an unstructured state (SSoff). When the peptide was in a helical state, the binding free energy increased, though the binding to PC as a helix (red solid) and PG as a coil (blue dashed) show compared binding free energies.

Experimentally the γ₁ adsorption free energies have been measured against POPC and POPG bilayers by ITC and nearly equivalent values are observed (−7.5 and −8.0 kcal/mol, respectively).¹⁵ While the magnitudes of our PMFs are larger, we observed similar magnitude ΔGs (−11.0 vs −15.0 kcal/mol) when comparing ordered γ₁ binding to POPC and disordered γ₁ binding to POPG. This observation appears reasonable in consideration that the calorimetry studies also reported that binding to POPC membranes is enthalpically driven while binding to POPG membranes is entropically driven. Our studies provide a rationale for the thermodynamic measurements, that γ₁ can bind to POPG in a disordered state but must adopt a helical conformation when binding POPC. Energetically, association of γ₁ to POPG offers a lower conformational entropic loss thus providing a means to compensate for the decreased enthalpic contribution to binding.

Solvent effects can also have a significant contribution to the protein binding thermodynamics,³⁹ and we investigate an aspect of this by computing the entropic gain from cation desorption from the bilayer surface when γ₁ binds to the bilayer. We have computed the sodium ion probability distributions (p(Z)) in the direction normal to the peptide binding bilayer (upper) leaflet, for POPC and POPG bilayers when γ₁ is bound and when the peptide is in solution. From the probability distributions the entropy can be computed as

$$S=-k_{B}N\sum _{i}p(Z_i)\text{ln}p(Z_i)$$ (2)

where k_B is Boltzmann’s constant, and N is the number of ions in the system.⁴⁰ The change in ion translational entropy upon peptide binding can be computed as ΔS = S_bound-S_unbound, and the contribution to the free energy of binding as −TΔS. The changes to p(Z) upon binding γ₁ are slight for both the POPC and POPG systems, however due to the larger number of cations in the POPG simulations to neutralize the system, we find that the entropic gain due to ion release is approximately 10-fold larger in the POPG system than the POPC system. With a histogram bin width of 0.5 Å, we compute −TΔS_{POPG =} −6.5 kcal/mol and −TΔS_POPC = −0.7 kcal/mol, indicating that the ion release entropy gain is a significant contribution in the binding of γ₁ to POPG bilayers. Regarding the magnitudes of the computed binding ΔGs it is possible the CG model overestimates these quantities. However a more recent study measuring surface pressure changes in a Langmuir film balance, calculated a K_d = 2.8 nM for γ₁ binding to DPPC/DPPS (4:1) monolayers.⁴¹ This K_d would correspond to a ΔG = −14 kcal/mol, according to eq (3)

$$\mathrm{\Delta }G=-kT\text{ln}\left(\frac{55.5}{K_d}\right)$$ (3)

which is more consistent with our estimates of the binding free energy magnitudes.

3.3. Influence of lipid composition on the membrane bound folding properties of γ₁

To understand the effect of lipid composition on the structural aspects of γ₁ binding, we move to AA simulations to study the structural evolution of γ₁ when interacting with bilayers. We first analyze Systems I and IV, which displayed similar binding free energies despite significant differences in the secondary structure content of the peptide. We converted CG structures from the end of the 1.5 μs spontaneous binding simulations to AA configurations and then performed 1μs AA simulations. The initial and final snapshots of these AA simulations are shown in Fig. 5. Our final configuration for POPC (System I) and POPG (System IV) shows that γ₁ remains helical on POPC and embeds into the hydrophobic region of the bilayer, while γ₁ on POPG remains disordered with 0% helicity on the bilayer periphery. From these detailed AA simulations, the difference in conformational state of γ₁ on POPC (ordered/low entropy) as compared to POPG (disordered/high entropy) again corroborates with the thermodynamic decomposition of γ₁ binding to POPC and POPG bilayers.

Figure 5.

AA membrane bound simulations. Simulations were initiated from CG bound structures to POPC (A) and POPG (C) bilayers. Simulations were run for 1 μs on each bilayer and end states are shown in (B) and (D). We observe that γ₁ in the simulation on POPC remains helical and embeds deeper in the membrane, while the simulation on POPG shows that γ₁ remains unstructured and bound to the membrane surface without partitioning into the hydrophobic core.

The AA simulations are supported by the strong agreement with experimental measurements of the configuration of γ₁ on POPC bilayers. We have computed the helicity, TRP4 (reside 367 by capsid numbering) insertion depth, and helical orientation relative to the bilayer normal during the final 100 ns of γ₁-POPC simulation (System I) and find all these quantities to be well correlated with experimental measurements.¹³ The helicity in our simulations is calculated to be 78 +/− 4%, which matches the 75% helicity as determined from CD. The TRP4 distance to the bilayer mid-plane is measured to be approximately 1 nm from the center of a POPC bilayer and our simulations show TRP4 inserts to a distance of 1.2 +/− 0.2 nm from the bilayer center. The angle the helical axis makes with the bilayer normal was measured to between 66° – 78°, while in the simulation the peptide samples a wider range of angles between 65° – 90°, with an average value of 81° +/− 6° This excellent agreement with the structural and thermodynamic experimental measurements gives us confidence in predicting and analyzing the binding, folding and insertion properties of FHV γ₁ in systems which have not been experimentally examined.

The simulations described above support a conformational dependence of γ₁ as a function of bilayer composition. However, even during 1 μs simulations, we cannot discount that kinetic barriers may exist, to either nucleate folding or unfolding, which are not overcome on the timescale of the simulations. Therefore, we further explored the membrane composition sensitive folding propensities, by starting simulations on POPC, POPG and the mixed bilayer that are initiated in a consistent, partially folded configuration. To do this we back-mapped the CG configuration of γ₁ bound to the mixed bilayer (Fig. 6A; System VI) and transformed the bilayer to either a pure POPC (System VII) or pure POPG bilayer (System VIII). This approach was adopted to try to eliminate any bias introduced by differing initial peptide configurations in the folding behavior of γ₁.

Figure 6.

Membrane folding of γ₁ on different composition bilayers. A) The initial configuration for all simulations is derived from the CG simulation of γ₁ binding to a mixed bilayer with SSoff. The final configuration after 1 μs of simulation is shown when the configuration from (A) is run on mixed POPC/POPG bilayer, pure POPG bilayer and pure POPC bilayer in (B), (C) and (D), respectively. The progression of helicity is shown in (E), and the per-residue helicity averaged over the final 100 ns is shown in (F). The residue insertions with respect to the lipid phosphate groups is shown in (G), which shows insertion depth and folding (F) are strongly correlated. In (A)–(D) the POPG lipids are colored greenm the POPC lipids are gray and the peptide is blue.

The results from these simulations initiated from a consistent peptide conformation are presented in Fig. 6. Consistent with the AA simulations of System I and System IV, the simulation on POPC gains helical content reaching 63% helicity during the final 100 ns (Fig. 6D,E). Whereas γ₁ on POPG maintains a relatively low helical content of only 23% helicity over the final 100 ns (Fig. 6C,E). These “lipid-substitution” simulations, re-affirm our understanding that the lipid composition affects the bound conformational state of γ₁ as the peptide folds or remains helical on a POPC bilayer (System I and VII). On the other hand, when bound to an anionic POPG bilayer, γ₁ remains in a low helicity disordered state (System IV and VIII).

However, the behavior of γ₁ on the mixed POPC/POPG bilayer cannot be as simply interpreted. There does not appear to be a linear relationship between the charge content in the membrane and the folding capacity of the peptide, as the mixed bilayer does not display an intermediate degree of helicity between POPC and POPG. After 1 μs of equilibrium simulation, as shown in Fig. 6B γ₁ appears as a random coil structure with a low helical content of 16%. To comprehend the structural dynamics of γ₁ in this system we focus on the micro-environment around γ₁ during the simulation. The local environment around the initial conformation of γ₁ on the mixed bilayer (Fig. 6A) is dominated by POPG contacts. During the first 100 ns of the 1 μs simulation the average ratio of PG to PC contacts is 4.6:1. The contacts evolve to a more equalized ratio, through the simulation, but PG contacts still dominate, the average ratio during the final 100 ns is 1.4:1::PG:PC. The ratio of PC:PG contacts during the 1 μs simulation are presented in Fig. S2. One possibility why a greater degree of folding is not seen in the mixed system is that slow diffusion of the lipids prevents formation of a POPC dominated microdomain to nucleate γ₁ folding and insertion.

The peptide insertion depth does appear to be an important quantity as it correlates strongly with formation of secondary structure in γ₁. In Systems VI–VIII simulations, helix formation is observed in the central region (residue 9–13) of peptide, but only in system VII (pure POPC) does helicity develop in the C-terminal region (Fig. 6F). We also observe that γ₁ inserts into the hydrophobic region to a much greater extent, especially in the C-terminal region, when comparing POPC against POPG for the simulations converted from the mixed bilayer system (Fig. 6G). The average insertion depth of γ₁ on POPC is 6.9 Å below the headgroups, while the helical residues (6–19) insert further to an average depth of 8.6 Å. Similar trends are seen on POPG where the overall average residue insertion depth is 1.3 Å and 3.6 Å for the helical residues (9–14). Likewise, on the mixed bilayer γ₁ has an average insertion depth of 0.1 Å above the headgroups while the helical residue (9–13) are inserted to an average depth of 2.5 Å. The decreased insertion depth into POPG bilayers has been noted in the experimental literature as well.¹⁵

3.4 To fold or not to fold?

Functionally we expect folding to be beneficial to the membrane disrupting activity of the peptide to create transmembrane structures and hence an important component to viral infection. The observed folding characteristics in this study may help in understanding why γ₁ elicits more robust membrane disruption of POPC vs. POPG vesicles in-vitro.¹³ However, from an energetic perspective, our CG simulations indicate that the folded state is always thermodynamically more favorable, regardless of membrane composition (Fig. 4). The nature of the MARTINI model and its inability to correctly model the protein dynamics, as well as the noted enthalpy/entropy imbalance⁴² could be causes for this discrepancy between the CG and AA simulation. Another possibility is inadequate sampling in the AA simulations due to large energy barriers in the folding pathway of γ_1. To address these possibilities we conducted additional simulations (System III and V) to explore if helical conformations were stable on the POPG or mixed bilayers in the AA model. If the presences of anionic lipids in the bilayer were highly destabilizing to the helical conformation, we would expect to see rapid unfolding of γ₁ when initiated in a helical conformation. However, we do not observe unfolding in these simulations, and γ₁ maintains high helicity content through the simulation.

A plot showing the insertion depth for each residue with respect to the COM of the top leaflet phosphates is shown in Fig. 7. For each system, we have compared the initial 100 ns to the final 100 ns of the 1 μs simulations, to judge the extend of insertion stability. We observe that the insertion depth is stable in POPC (Fig. 7A) and POPG (Fig. 7B), while the mixed system shows the N-terminus inserting deeper while the C-terminus moves toward the surface (Fig. 7C). The insertion profile for POPC shows the deepest insertion and is consistent with the insertion profile of System VII (Fig. 6G). The average insertion depth per residue during the final 100 ns residue is 6.0 Å, 3.3 Å, and 4.9 Å, for the POPC, POPG and mix bilayer systems, respectively. The increased insertion depth into the POPC and mixed bilayers compared to the POPG system may indicate a lower energy barrier to transition to a transmembrane configuration.

Figure 7.

Residual membrane insertion depth for systems with γ₁ initiated in a helical conformation. The depths are averaged over the first (red) and last (blue) 100 ns. Simulations were performed on A) POPC (System I), B) POPG (System III) and C) the mixed bilayer (System V).

4. Discussion

The results from these simulation studies provide qualitative explanations of the relatively high ΔG values when helical γ₁ binds to the membrane, (Fig. 4) and underscores the time-scale limitation of AA MD (ns to μs) simulations to access processes like membrane insertion and folding of a peptide. An illustrative example is System IV (Fig. 6, S2), which shows that towards the end of 1 μs simulation, the micro-environment around γ₁ has a comparable amount of POPC and POPG molecules. However, due to the limitation of the time-scales for folding, membrane insertion and/or lipid diffusion the peptide does not encounter a more concentrated POPC environment to undergo an entropic loss and gain helicity. Integrating our findings with previous studies we suggest an energetic pathway as shown in Fig. 8, that once a membrane-γ₁ complex is formed a “folding” barrier must be overcome to gain helicity and insert into the membrane. This barrier appears to be a function of membrane charge (and other possible factors), and may be non-existent for neutral membranes where the binding and folding process can be coupled, which is schematically represented in Fig. 8. It is possible that the barrier is too high for the POPG system, and the peptide remains trapped in the disordered state, resulting in the observed low lytic activity.¹³

Figure 8.

Proposed energy landscape for γ₁ to transition from an unstructured, solvated state to a helical, membrane bound state. We propose the pathway is modified based upon the lipid composition, and specifically the increase in anionic lipids may introduce/increase an energy barrier to reach the folded state.

Investigations into the molecular dynamics of membrane active peptides from non-enveloped viruses are quite limited as compared with anti-microbial peptides (AMP). However, a recent study has also examined the FHV γ-peptide using MD and experimental methods.⁴³ In that study γ₁, the full-length γ-peptide and a construct peptide in which the 8 C-terminal residues of γ were fused to γ₁ (termed Δ385–399), were investigated to characterize their interactions with membranes. The Δ385–399 construct contained three Phe residues in the C-terminal region, which had been previously implicated to be critical for membrane disruption,¹⁴ yet Δ385–399 was less effective in disrupting membranes than γ₁ or even a non-cleaving maturation defective mutant. Based upon 100 ns simulations of the peptides in the presence of a POPC bilayer, it was proposed that the full C-terminal region is critical in stabilizing the helicity in the N-terminal region. This proposal was based upon their simulations that showed i) γ₁ displayed low helicity when bound to the bilayer and ii) a homology model of the full length γ peptide, which formed a helical hairpin structure, displayed a stabilized N-terminal helix on the bilayer. Both our simulations and previous experimental results indicate that γ₁ should retain a high degree of helicity on a POPC bilayer and therefore the proposal that the C-terminal region acts as a stabilizer of the N-terminal helix may be disputable. Furthermore, the initial helical hairpin predicted structure may deviate from the authentic structure and the initial condition could significantly bias the simulation, especially on the 100 ns timescale.

While the literature on non-enveloped membrane active viral peptides is small, there has been a tremendous amount of work investigating the mechanism of AMP interactions with membranes We will briefly expand upon some studies on AMP systems which have a degree of similarity to FHV γ₁. The size of FHV γ₁ (21 amino acids) and it’s amphipathic character is similar to several AMPs such as alamethicin, piscidin, magainin and melittin, which have 20, 22, 23 and 26 amino acids, respectively. These AMPs also exhibit a similar structural behavior to γ₁ as they are disordered in the solution phase but gain helicity upon interactions with a membrane.^44–47 γ₁ is a cationic peptide with a net charge of +2 at neutral pH, which is consistent with most AMPs,⁴⁸ including melittin (+6), piscidin (+3) and magainin (+3). The thermodynamics of membrane binding of of the aforementioned AMPs is also similar to FHV γ₁ (≈−7.5 kcal/mol), for example, the free energy of melittin binding to a 80/20 POPC:POPG bilayer is −8.7 kcal/mol⁴⁹ and −6.2 kcal/mol for alamethicin binding to a POPC bilayer.⁵⁰ Numerous mechanisms of action have been proposed for AMPs and for many systems the mode of membrane disruption remains unclear. Recent long time scale (multi μs) have been conducted on pore structures for melittin, piscidin 1, magainin (MAG2 and MG-H2) and PGLa, and the results vary between the systems. The peptides generally show dynamic pores, with the peptides being able to transition between a transmembrane (T) and surface bound state (S). For melittin, magainin, and PGLa a toroidal pore structure is supported, in which lipid head groups are pulled into the bilayer interior and (some) peptides are stable in the T-state.^51,52 However, for piscidin 1 a pore structure was not stable as the peptides preferred S-state orientations.⁵³ Transient membrane defects were observed for piscidin 1, which were characterized by the deep insertion of the C-terminal region of the peptide.⁵³ Based upon the size and charge similarity of γ₁ to piscidin 1, and that we also observe deep C-terminal insertion (Fig. 6G), we would predict these peptides to share a similar mechanism of action, but further studies will be required to test this prediction.

Enveloped virus membrane fusion is a heavily studied field and these systems have key differences but also share some commonalities with non-enveloped virus membrane disruption. ⁵⁴ Influenza virus is best characterized system and several molecular dynamics studies have investigated the fusion peptide-membrane interactions to understand the structural and energetic aspects of this interaction.^55–59 The role of enveloped virus fusion proteins is not to directly disrupt the cellular membrane, as is the case for non-enveloped viruses, but to lower the energy barrier associated with fusing the viral and cellular membranes. Enveloped viruses can have multiple surface glycoproteins, one of which is the fusion protein. There are three classes of fusion proteins: class I consist of mainly α-helical, class II are largely β-structures and class III contain both α-helical and β-structures. Fusion proteins require a priming step which involves a proteolytic cleavage event, to make them fusion competent, this is analogous to the maturation process in FHV which cleaves the γ-peptide from the capsid protein. Enveloped fusion is often triggered by exposure to low pH. In many cases the low-pH environment is required but not sufficient for fusion, while in other cases such as influenza and dengue virus, low-pH alone is sufficient to trigger fusion, similar to FHV. Fusion proteins have either an N-terminal fusion peptide or an internal fusion loop, generally on the tips of an extended β-sheet. The fusion peptides are conformationally diverse and dynamics, as the undergo conformational changes when they transition to the membrane pre-fusion state. A key difference between FHV and enveloped viruses is that there is a well defined oligomeric state in the post-fusion conformation, as all known cases exist in a trimer of hairpins, whereas the oligomeric state of FHV is unknown and may not be well-defined. Presenting the fusion pepetides/loops to the cellular membrane and the folding-back process to fuse the two membranes both involve large conformational changes to the fusion proteins, which is not analogous to the FHV mechanism. Overall enveloped virus fusion machinery is larger and the mechanisms are more complex than FHV membrane disruption.

In this work we have extended our understanding of FHV γ₁ membrane interactions through a multi-scale computational approach, which combined CG MARTINI, AA metadynamics and 8 μs of equilibrium AA simulations. We were able to observe that γ₁ membrane binding is consistent with several experimental measurement and provides a justification of experimentally known thermodynamic energy decomposition of γ₁ binding to POPC and POPG bilayers. Binding of γ₁ to a negatively charged POPG membrane is entropically driven and strongly influenced by electrostatic effects, including the release of counter ions form the membrane surface. However, γ₁ adopts a helical conformation while undergoing an entropic loss when binding to neutral POPC bilayers, which is an exothermically driven binding process. Our simulations also revealed that the structural dynamics of γ₁ are governed not only by the bulk lipid composition, but the micro-environment in the vicinity of γ₁ also plays a crucial role. Assimilating the CG and AA simulations conducted in this study we have proposed an energetic pathway of γ₁ binding. γ₁ will be rapidly attracted to negatively charged membranes and will adsorb in a disordered state, where the peptide has to overcome an energy barrier in order to fold and insert into the membrane, and the size of this barrier may limit the membrane activity of the peptide. Overall this study provides insights into the conformational dynamics of FHV γ₁ and suggests the virus may be evolved for efficient binding and folding against the endocytic membranes the peptides would encounter during the virus infection pathway. Further investigations utilizing enhanced sampling methods and a wider range of lipid compositions are warranted to further tease apart these details.

Highlights.

• Thermodynamic driving force of γ₁ membrane binding is dissected

• Membrane composition affects peptide folding rates

• Increased insertion depth in neutral membranes may relate to increased lysis