Article Information

• PDF (263 kb)
• Full Text with Large Figures
• Cited by in Scopus (12)

PubMed

• Articles by L. Lins
• Articles by R. Brasseur

Related Articles

• Membrane-Protein Interactions in a Generic Coarse-Grained Mo...

  Membrane-Protein Interactions in a Generic Coarse-Grained Model for Lipid Bilayers Biophysical Journal, Volume 96, Issue 1, 7 January 2009, Pages 101-115 Beate West, Frank L.H. Brown and Friederike Schmid Abstract We study membrane-protein interactions and membrane-mediated protein-protein interactions by Monte Carlo simulations of a generic coarse-grained model for lipid bilayers with cylindrical hydrophobic inclusions. The strength of the hydrophobic force and the hydrophobic thickness of the proteins are systematically varied. The results are compared with analytical predictions of two popular analytical theories: The Landau-de Gennes theory and the elastic theory. The elastic theory provides an excellent description of the fluctuation spectra of pure membranes and successfully reproduces the deformation profiles of membranes around single proteins. However, its prediction for the potential of mean force between proteins is not compatible with the simulation data for large distances. The simulations show that the lipid-mediated interactions are governed by five competing factors: direct interactions; lipid-induced depletion interactions; lipid bridging; lipid packing; and a smooth long-range contribution. The mechanisms leading to hydrophobic mismatch interactions are critically analyzed. Abstract | Full Text | PDF (825 kb)

• Kinetics of local anesthetic inhibition of neuronal sodium c...

  Kinetics of local anesthetic inhibition of neuronal sodium currents. pH and hydrophobicity dependence Biophysical Journal, Volume 58, Issue 1, 1 July 1990, Pages 69-81 D.M. Chernoff and G.R. Strichartz Abstract This study assesses the importance of local anesthetic charge and hydrophobicity in determining the rates of binding to and dissociation from neuronal Na channels. Five amide-linked local anesthetics, paired either by similar pKa or hydrophobicity, were chosen for study: lidocaine, two tertiary amine lidocaine homologs, a neutral lidocaine homolog, and bupivacaine. Voltage-clamped nodes of Ranvier from the sciatic nerve of Bufo marinus were exposed to anesthetic externally, and use-dependent ("phasic") block of Na current was observed. Kinetic analysis of binding (blocking) rates was performed using a three parameter, piecewise-exponential binding model. Changes in extracellular pH (pHo) were used to assess the role of drug protonation in determining the rate of onset of, and recovery from, phasic block. For those drugs with pKa's in the range of pHo tested (6.2–10.4), the forward binding rate during a depolarizing pulse increased at higher pH, consistent with an increase in either intracellular or intramembrane concentration of drug. The rate for unbinding during depolarization was independent of pHo. The dissociation rate between pulses also increased at higher pHo. The pHo dependence of the dissociation rate was not consistent with a model in which the cation is trapped relentlessly within a closed channel. Quantitative estimates of dissociation rates show that the cationic form of lidocaine dissociates at a rate of 0.1 s-1 (at 13 degrees C); for neutral lidocaine, the dissociation rate is 7.0 s-1. Furthermore, the apparent pKa of bound local anesthetic was found to be close to the pKa in aqueous solution, but different than the pKa for "free" local anesthetic accessible to the depolarized channel. Abstract | PDF (1288 kb)

• Simulation Studies of Protein-Induced Bilayer Deformations, ...

  Simulation Studies of Protein-Induced Bilayer Deformations, and Lipid-Induced Protein Tilting, on a Mesoscopic Model for Lipid Bilayers with Embedded Proteins Biophysical Journal, Volume 88, Issue 3, 1 March 2005, Pages 1778-1798 Maddalena Venturoli, Berend Smit and Maria Maddalena Sperotto Abstract Biological membranes are complex and highly cooperative structures. To relate biomembrane structure to their biological function it is often necessary to consider simpler systems. Lipid bilayers composed of one or two lipid species, and with embedded proteins, provide a model system for biological membranes. Here we present a mesoscopic model for lipid bilayers with embedded proteins, which we have studied with the help of the dissipative particle dynamics simulation technique. Because hydrophobic matching is believed to be one of the main physical mechanisms regulating lipid-protein interactions in membranes, we considered proteins of different hydrophobic length (as well as different sizes). We studied the cooperative behavior of the lipid-protein system at mesoscopic time- and lengthscales. In particular, we correlated in a systematic way the protein-induced bilayer perturbation, and the lipid-induced protein tilt, with the hydrophobic mismatch (positive and negative) between the protein hydrophobic length and the pure lipid bilayer hydrophobic thickness. The protein-induced bilayer perturbation was quantified in terms of a coherence length, , of the lipid bilayer hydrophobic thickness profile around the protein. The dependence on temperature of , and the protein tilt-angle, were studied above the main-transition temperature of the pure system, i.e., in the fluid phase. We found that depends on mismatch, i.e., the higher the mismatch is, the longer becomes, at least for positive values of mismatch; a dependence on the protein size appears as well. In the case of large model proteins experiencing extreme mismatch conditions, in the region next to the so-called , there appears an (or ) region where the bilayer hydrophobic thickness is locally lower (or higher) than in the unperturbed bilayer, depending on whether the protein hydrophobic length is longer (or shorter) than the pure lipid bilayer hydrophobic thickness. Proteins may tilt when embedded in a too-thin bilayer. Our simulation data suggest that, when the embedded protein has a small size, the main mechanism to compensate for a large hydrophobic mismatch is the tilt, whereas large proteins react to negative mismatch by causing an increase of the hydrophobic thickness of the nearby bilayer. Furthermore, for the case of small, peptidelike proteins, we found the same type of functional dependence of the protein tilt-angle on mismatch, as was recently detected by fluorescence spectroscopy measurements. Abstract | Full Text | PDF (612 kb)

• …more

Copyright © 2006 The Biophysical Society. All rights reserved.
Biophysical Journal, Volume 90, Issue 2, 470-479, 15 January 2006

doi:10.1529/biophysj.105.068213

Membranes

“De Novo” Design of Peptides with Specific Lipid-Binding Properties

L. Lins, B. Charloteaux, C. Heinen, A. Thomas and R. Brasseur^,  

Centre de Biophysique Moléculaire Numérique, Faculté des Sciences Agronomiques de Gembloux, Gembloux, Belgium

Address reprint requests to R. Brasseur, Tel.: 00-32-816-2-2521.

Designed peptides have been used in the last decade to improve our understanding in the structure/function relationship of proteins. This design approach notably helps us to understand the structural determinants of folds, such as those of β-sheets 1. The current methods use statistical analysis of protein-structure databank and empirical information 2,3,4. Rotamer analysis can also be performed to check the compatibility of side-chain packing.

Modified or non-natural amino acids are notably used to study the folding, the conformational stability, and the flexibility of peptides and proteins 5,6,7,8,9. Non-natural amino acids can be more resistant to protease degradation and have a decreased immunogenicity 10.

Model peptides can also be used to explore features responsible for the association of hydrophobic peptides derived from a water-soluble coiled-coil structure in membrane 11. In the same way, designed pore-forming peptides are useful models to enable us to better understand the insertion of proteins in biomembranes 12.

Peptides with antitumoral or antibacterial activity were also designed to investigate structural parameters of their lytic activity and selectivity, notably their membrane specificity 13,14.

In these cases, design of peptides with improved activity and/or selectivity might be helpful for biological and therapeutic uses.

In this study, we describe an original in silico method to design peptides with desired features, notably with specific membrane-interacting properties. We applied this method to the design of tilted peptides. The latter are short protein fragments able to destabilize membranes 15. When tilted peptides are modeled as α-helices, an asymmetric distribution of hydrophobicity is responsible for an oblique orientation at a hydrophobic/hydrophilic interface, such as a membrane 16. Tilted peptides were notably found in viral fusion proteins where they are involved in the first steps of fusion between the host cell and the virus membrane 17. They were also detected in proteins involved in lipid metabolism, in signal sequences, in toxins, etc. 18. Many among the in silico detected peptides were experimentally confirmed. Most were shown to induce liposome fusion in vitro when taken as isolated fragments 19,20,21,22,23,24. Furthermore, when the peptides were mutated to modify the hydrophobicity distribution, fusogenic properties were significantly decreased. When the same mutations were introduced in the protein, its activity was modified 23,25. Neutron diffraction experiments showed that one structural conformer of the tilted peptide from the gp32 fusion protein of simian immunodeficiency virus (SIV) is oriented at an angle of 55° in a model membrane 26. In the same way, the NMR structures of the HA2 fusion peptide from the Influenza virus were shown to be obliquely oriented in model membranes at the active pH 27.

Comparison of tilted peptides demonstrated no sequence homology 28; the only characters they shared were their mean hydrophobicity (following the Eisenberg consensus scale 29) between 0.2 and 0.9, an asymmetric hydrophobicity gradient when helical and an insertion angle of this helix into lipids between 30° and 60°.

In this article, we report a modeling approach for the rational design of hydrophobic peptides entirely made from non-natural amino acids. The goal was to make a tilted peptide, i.e., a short 12-residue fragment with a mean hydrophobicity between 0.6 and 0.7 and a lipid-insertion angle close to 45° due to the asymmetric distribution of hydrophobic residues. One predicted sequence has been synthesized and experimentally tested for its lipid-destabilizing capacities.

Egg phosphatidylcholine (PC), phosphatidylinositol (PI), phosphatidylserine (PS), cholesterol (CHOL), and sphingomyelin (SM) were purchased from Sigma (St. Louis, MO). Egg phosphatidylethanolamine (PE) is from Lipid Products (Redhill, Surrey). Octadecyl rhodamine chloride (R18) is from Molecular Probes (Eugene, OR).

The peptides were obtained from Neosystem (Strasbourg, France). They were 85% pure.

Each molecule structure is oriented with respect to the spatial distribution of its hydrophobicity. The line joining the hydrophilic and hydrophobic centers is set to be perpendicular to the hydrophobicity interface 30. The hydrophobic center () is defined by the equation

(1)

(2)

We inserted peptides into an implicit bilayer integral membrane protein and lipid association (IMPALA) developed by Ducarme et al. 31. It simulates the insertion of any molecule (protein, peptide, or drug) into a bilayer by adding energy restraint functions to the usual energy description of molecules 32,33.

The lipid bilayer is defined by C_(z), which represents an empirical function describing membrane properties. This function is constant in the membrane plane (x- and y-axes) but varies along the bilayer thickness (z-axis) and more specifically, at the lipid/water interface corresponding to the transition between lipid acyl chains (no water=hydrophobic core) and the hydrophilic aqueous environment,

(3)

Two restraints simulate the membrane—one the bilayer hydrophobicity (E_pho), and the other, the lipid perturbation (E_lip).

The hydrophobicity of the membrane is simulated by E_pho,

(4)

The perturbation of the bilayer by insertion of the molecule is simulated by the lipid perturbation restraint (E_lip),

(5)

The environment energy (E_env) applied on the peptide that inserts into the membrane becomes equal to

(6)

A systematic procedure is performed to insert and orient the peptide into the membrane. During this process, the peptide systematically crosses the force field of the membrane from −40 to +40Å with respect to the membrane center by steps of 1Å. For each position along the z axis, 2000 random orientations are tested. Among these 2,000 positions, the minimal energy position is selected. At the end of the systematic analysis, the procedure selects the position and the orientation of minimum energy among all selected minima.

To analyze structural variations of the peptide inserted in the membrane, we have used the angular dynamics procedure previously defined to simulate the protein folding 34.

In the simulations, the total energy (E_tot) is the sum of the intramolecular energy of the peptide (E_intra) and of the energy due to the membrane environment (E_env). E_tot is distributed at each step of the calculation on the peptide torsion k-axis as E(k). This total energy is equal to

(7)

The intramolecular energy E_intra is composed by E_vdw, the Van der Waals energy that represents the interaction of the electronic cloud of the atoms, and is described by the Levitt’s equation 35

(8)

E_elec, the electrostatic energy is given by the classical Coulomb’s law that takes into account the point-charge of the atoms (q_i and q_j), the dielectric constant, and the distance (d_ij) between atoms i and j,

(9)

E_{pho_in} is the intramolecular hydrophobic energy,

(10)

The energy of each axis ((∑E(k)) is supplemented by its own torsion energy (E(k)_tor),

(11)

The total energy associated with each torsion axis (E(k)) is therefore represented by the sum of 1), the torsion energy of the k axis (E(k)_tor); 2), the intramolecular interaction energies between atoms i and j, divided by the number of axes between these atoms; and 3), the energy in the membrane for the atoms i and j divided by the number of atoms of the system minus 1 (N−1) and by the number of axes between atoms i and j,

(12)

The energy E(k) allows us to calculate an angular dynamics that gives rise to an acceleration of torsion axes.

During the dynamics, the length of atomic bonds and the value of valence angles are kept constant; only the torsion angles are modified. The angular acceleration of each torsion axis is determined following Newton’s equation,

(13)

(14)

Knowing the angular acceleration of the torsion axes, the t+1 position of each torsion axis (ξ(k)^theo) can be determined following the equation derived by Verlet 36,

(15)

At each step, E(k)_t is the energy associated to an angle k at t time and is given by Eq. (12).

During the angular dynamics, the rotation velocity v_r(k), expressed as a function of Δt is calculated according to

(16)

The rotation velocity v_r(k) used in our calculations is limited to 0.075°/step. It is an arbitrary empirical value that seems to be sufficient for a rapid molecule movement and slow enough to prevent molecule burst. During all the calculation, the movement quantity (Q) was kept constant,

(17)

As the velocity changes at time t+1 the theoretical movement quantity becomes Q_t+1,

(18)

(19)

(20)

(21)

A t=0, the position and orientation of the peptide in the membrane are those determined from the IMPALA systematic analysis (see above). From there, an angular dynamics is performed on the peptide taking into account the mean force field of the membrane. Random rotations of 1° max and random translations of 0.1Å max of the peptide are allowed.

Calculations are performed on an Intel Pentium 4, 3.80-GHz CPU, with 4.00 Gb of RAM.

Pex2dstats files 37 are generated during the simulations and used for the analysis of each peptide. The mean gain ratio of angular versus molecular dynamics was calibrated to 1/100–1/1000 the number of calculation steps when only side chains or side chains plus backbone movements were calculated.

Large unilamellar vesicles (LUV) were prepared by the extrusion technique 38 using an extruder (Lipex Biomembranes, Vancouver, Canada). In brief, dry lipid films which are mixtures in weight of 30% phosphatidylcholine (PC), 30% phosphatidylethanolamine (PE), 2.5% phosphatidylinositol (PI), 10% phosphatidylserine (PS), 5% sphingomyelin (SM), and 22.5% cholesterol were hydrated for 1h at 37°C. The resulting suspension was submitted to five successive cycles of freezing and thawing and thereafter extruded 10 times through two stacked polycarbonate filters (pore size 0.08μm) under a nitrogen pressure of 20 bars.

The final concentration of liposomes was determined by phosphorus analysis 39.

Mixing of liposome membranes was followed by measuring the fluorescence increase of R18, a lipid soluble probe, occurring after the fusion of labeled and unlabeled liposomes, as described 40. Labeled liposomes were obtained by incorporating R18 in the dry lipid film at a concentration 6.3% of the total lipid weight. Labeled and unlabeled liposomes were mixed at a weight ratio 1:4 and a final concentration of 50μM in 10mM Tris, 150mMNaCl, 0.01% EDTA, and 1mMNaN₃, pH 8. Fluorescence was recorded at room temperature (λ_exc, 560nm; λ_em, 590nm) on an LS-50B Perkin-Elmer fluorimeter (Boston, MA).

The 8-hydroxypyrene-1,3,6-trisulfonic acid (HPTS)/p-xylylenebis[pyridinium] bromide (DPX) assay of Ellens et al. 41 was used to monitor vesicle leakage. The assay is based on the quenching of HPTS by DPX. HPTS and DPX are both encapsulated in the aqueous phase of the same liposomes. Leakage of vesicles was followed by measuring the dequenching of HPTS released into the medium. Fluorescence was recorded at room temperature (λ_exc, 360nm; λ_em, 520nm) on a LS-50B Perkin-Elmer fluorimeter.

The mixing of liposome contents was monitored using the core-mixing assay of Kendall and McDonald 42. Liposomes (LUV) were prepared as described above in 10mM Tris-HCl buffer, 150mMNaCl, and 1mMNaN₃, pH 8.0 and containing calcein at 0.8mM and CoCl₂ at 1.0mM or EDTA at 20mM. Untrapped solutes were removed by one elution on a Sephadex G-75 column (Sigma-Aldrich) with 10mM Tris-HCl, 150mMNaCl, and 1mMNaN₃ buffer, pH 8.0. In a standard experiment, calcein, Co²⁺- and EDTA-containing vesicles were mixed at 1:1 molar ratio in a 10mM Tris-HCl buffer, pH 8.0 (150mMNaCl, 1mMNaN₃). When peptides were added, the calcein fluorescence was monitored at room temperature (λ_exc, 490nm; λ_em, 520nm) as a function of time on a LS-50B Perkin-Elmer fluorimeter. Co²⁺ (0.4mM in chelation with citrate at 1:1mol/mol) was present in the medium to avoid fluorescence due to leakage of vesicle contents. The maximum fluorescence was determined in presence of Triton X-100, 0.5% (10mM EDTA).

Attenuated total reflection (ATR) infrared spectroscopy was used to determine the secondary structure of the peptides alone and bound to lipids.

Spectra were recorded at room temperature on a Bruker Equinox 55 (Bruker AXS, Karlsruhe, Germany) equipped with a liquid nitrogen-cooled Mercury-Cadmium-Telluride detector at a resolution of 2cm⁻¹, by averaging 512 scans. Free peptide samples (20μg peptide) dissolved in TFE and the lipid-bound peptides (see preparation below) were spread out on a germanium ATR plate (50×20×2 mm—Aldrich Chimica, Milan, Italy—with an aperture of 45° yielding 25 internal reflections) and slowly dried under a stream of N₂. Reference spectra of a Germanium plate were automatically recorded after purge for 15min with dry air and subtracted to the recently run sample spectra. The plate was sealed in an universal sample holder and hydrated by flushing the holder with N₂ saturated with D₂O for 3h at room temperature.

A dried mixed film made of 20μg peptide and 100μg lipids (PC/PE 2:1w/w) was hydrated with 100μl of a 10mM Tris, 150mMNaCl, pH 8 buffer.

Phospholipid concentration was determined as mentioned above.

Vibrational bands, especially the amide I band (1600–1700cm⁻¹), are sensitive to the secondary structures of proteins. The CO vibration is representative of 80% of the amide I band. This band accounts for all secondary structures which have different vibration values. The combination of resolution-enhancement methods with curve-fitting procedures allow us to assign quantitatively different secondary structures such as α-helix, β-sheets, and unordered structures. Each band was assigned according to the frequency of its maximum. The areas of all bands assigned to a given secondary structure were then summed and divided by the sum of all areas. This gives the relative ratio of each secondary structure. The bands are assigned as follows: α-helix, 1662–1645cm⁻¹; β-sheets, 1689–1682cm⁻¹ and 1637–1613cm⁻¹; random, 1644.5–1637cm⁻¹; and β-turns, 1682–1662cm⁻¹. It should be noted that the proteins spread on the plate are deuterated to avoid an overlap of α-helix and random-coil structures, as previously described 43,44.

Single-letter and three-letter codes were assigned to the 14 non-natural amino acids that we used throughout the study (Table 1). These amino acids, which are commercially available, are called modified or non-natural because they have the same backbone as the 20 residues found in eukaryotic proteins but have different side chains. None of the selected amino acids contains a halogen or a potentially toxic atom.

The structures of the 14 selected amino acids were modeled using HyperChem 5.0 (Hypercube, Gainesville, FL), starting from the backbone of alanine and optimized by the Polak-Ribiere algorithm using an AMBER force field with a gradient δ inferior to 0.1kcal/(Å mol).

The peptide length is fixed to 12 residues. All positions can be taken by any of the 14 non-natural residues, allowing 14¹² possible sequences. Due to time constraints, we reduced the number of sequences to 10,000,000, randomly generated. Three independent runs of 10,000,000 peptides have been performed. The results from a single run were selected for presentation in this article; the two others gave similar results.

One parameter of selection is the mean peptide hydrophobicity. This value is obtained from individual residue hydrophobicity, which is unknown for the non-natural residues.

For current amino acids, we use the hydrophobicity scale of Eisenberg 29. For non-natural amino acids, we developed a method for estimating an Eisenberg-like hydrophobicity value. Hydrophobicity values are usually obtained from experimental measurements of the transfer energy (E_tr) of molecules from water to an organic phase such as octanol. Molecular transfer energy can also be calculated from atomic E_tr (Table 1) and the solvent-accessible surface of atoms in the molecule three-dimensional structures as previously described 45. Both Eisenberg’s amino-acid hydrophobicity values and Brasseur’s calculated E_tr values are linearly correlated with a regression coefficient of 0.75 (Fig. 1). A 75% correlation was here estimated as sufficient since hydrophobicity was used as a filter for selection. Hence from the calculated three-dimensional structures and the atomic E_tr scales, the E_tr of non-natural amino acids were calculated. The Eisenberg hydrophobicity values were then extrapolated using the linear correlation (Fig. 1). Table 1 provides all Brasseur-calculated E_tr and Eisenberg’s consensus hydrophobicity values for the 20 natural amino acids and for the 14 non-natural amino acids. Hydrophobicity values of non-natural amino acids are often a bit lower or higher than those of natural amino acids. Amino acids of similar structures, such as (cyclo)hexalanine and (cyclo)hexetamine, or phenylethanine and methylphenylalanine, have very close values. For residues with increasing numbers of carbons in the lateral chain, i.e., ethanine, norvaline, allylglycine, terleucine, and norleucine, the increase in hydrophobicity values goes with the increase in the number of carbons, as expected for organic compounds 46. Two non-natural amino acids have a negative hydrophobicity: orthinine and citrulline; both are charged. Their hydrophobicity values are close to those of charged and polar natural amino acids, aspartic acid, asparagine, glutamic acid, and glutamine.

Display large version of this figure

Figure 1

Linear correlation between Eisenberg’s mean hydrophobicity values of natural amino acids and Brasseur’s calculated E_tr of those molecules (open circles). The linear regression coefficient (R²) is indicated; amino acids are named by their one-letter code. The calculated E_tr values of the 14 non-natural amino acids are also plotted (solid squares).

The peptide selection process is based on several steps corresponding to specific criteria (Fig. 2). The first criterion is the mean hydrophobicity of the peptide. The value is calculated using the scale described above (Table 1). Sequences are kept when the mean peptide hydrophobicity is between 0.6 and 0.7kcal/residue. This range was previously shown to correspond to tilted peptides interacting with lipids 28.

Display large version of this figure

Figure 2

Schematic representation of the different selection steps in the de novo generation of tilted peptides.

Sequences fulfilling this criterion were then three-dimensionally constructed as an α-helix. The distribution of hydrophobicity with respect to the helix axis was calculated by taking into account the geographic centers of hydrophobic and hydrophilic atoms. The angle between the helix axis and the molecule hydrophobicity interface of each helix structure was calculated; only those with angles between 30° and 60° were selected.

In the third step, a systematic IMPALA screening investigated the helical peptide insertion into a membrane. The optimal orientation of each helix in membrane was calculated and peptides presenting an orientation between 35° and 55° were kept. At that step a complementary criterion was introduced, which is that the position of the helix mass center had to be between 8 and 13Å with respect to the bilayer center, i.e., near the phospholipid (PL) acyl-chain/headgroup interface. This restraint was set to choose peptides able to disturb the PL acyl-chain organization, the first step thought to induce fusion processes 16. The peptides matching the different criteria were then optimized at the interface by an angular dynamics, to analyze their conformational stability. The whole strategy is illustrated in Fig. 2.

From 10,000,000 sequences tested, a quantity of 9,109,270 were rejected by taking into account the mean sequence hydrophobicity value. Among the remaining number, 822,337 were rejected by taking into account the peptide interface orientation, and another 67,196 were rejected after the IMPALA screening, taking into account the angle and the depth of insertion. At that step, 1197 sequences were retained. To further decrease the number of candidates, one criterion was added: the peptide had to include one hydrophilic residue, i.e., citrulline or ornithine. Only 12 peptides fitted that last criterion. The angular dynamics demonstrated that four of them were conformationally stable. They correspond to sequences number 48,482; 51,208; 88,682; and 97,848 (Table 2).

The energy minima profiles of peptide 97,484 insertion in the model membrane IMPALA (Fig. 3) demonstrates that both the pre- and post-angular dynamics structures maximally insert in a similar way—with their center of mass between 5 and 13Å from the membrane center (i.e., at the level of phospholipid headgroup/acyl-chain interface)—and have a low energy barrier to cross the membrane. This suggests that the latter structures might have access to a wide range of membrane levels. If the angular dynamics has had little consequence on the 97,484 backbone structure supporting our conclusion of helical stability, it modified the side-chain orientation and the final structure was even more penalized in water (IMPALA energy values at z over ±20Å) than the pre-optimized structure. This is illustrated in Fig. 3. It is lso seen while comparing the best configurations in membrane for the pre- and post-angular dynamics structures (Fig. 5 as compared to Figure 6f) and also accounts for the increase of the IMPALA energy when the peptide is sliding toward the membrane surface after 8000 steps of dynamics (Figure 6ac). After angular dynamics, side chains of naphtylalanine and phenylethaline are bent toward the phospholipid polar head interface and the citrulline polar side chain is partly protruding out of the membrane (Figure 6ef).

Display large version of this figure

Figure 3

Best position of peptide 97,848 in the IMPALA membrane before the angular dynamics. Midplane, bilayer center (z=0); first upper (beneath) plane; lipid acyl chain/polar headgroups interface at 13.5Å from the center; and second upper (beneath) plane, lipid/water interface (z=18Å).

Display large version of this figure

Figure 4

Time course of lipid mixing of liposomes (LUV) induced by peptide 97,848. The peptide is added at different concentrations (from 30μM to 600μM corresponding to peptide to lipid molar ratio from 1:50 to 2:5) to a mixture (1:4 w/w ratio) of R18-labeled and unlabeled LUVs. Increase of the R18 relative fluorescence due to probe dilution is followed at room temperature. The SIV tilted peptide at 150μM is used as positive control, the addition of TFE (1.6% final concentration) as negative control. TFE (blank), ●; and SIV peptide at 150μM (positive control), ♢. Addition of peptide 97,848 at 30μM (P/L molar ratio, 1:50), ○; 60μM (P/L ratio 1:25), ▴; 150μM (P/L ratio 1:10), ×; and 300μM (P/L ratio 1:5), ▵.

Display large version of this figure

Figure 5

Profile of the minimal IMPALA energy values (in kcal/mol) versus the mass center position (Å) of the 97,848 peptide after a systematic screening of peptide membrane traverse. Peptide 97,848 before the angular dynamics (solid line); and peptide 97,848 after the angular dynamics (dotted line).

Display large version of this figure

Figure 6

Angular dynamics course of peptide 97,848 in IMPALA membrane. The total number of steps is 10,000. (a) Evolution of IMPALA restraint energy (kcal/mol). (b) Evolution of peptide intramolecular energy. (c) Evolution of the peptide mass center position. (d) Evolution of the helix axis insertion angle (°). (e) Membrane position of peptide 97,848 residues in the most stable configuration of the angular dynamics simulation. The dotted line (interface) indicates where the water concentration is equal to 0.5. (f) The stick view of the most stable conformation of the angular dynamics simulation. This view is made in conditions similar to those shown in Fig. 5.

Those results suggest that the peptide 97,848 is stable in the membrane, and can move from the lipid polar head to the acyl-chain domain in that membrane. This mode of insertion generates conditions where lipid perturbation may occur and be responsible for a membrane fusion process. Peptide 97,848 was synthesized and experimentally tested for fusogenic activity and conformational analysis.

Lipid fusion (lipid and core mixing) and leakage assays were carried out. The SIV fusogenic peptide was used as a positive control, since its lipid-destabilizing properties are well known 20. The induction of vesicular lipid mixing by the different peptides is tested with PC/PE/SM/PI/PS/Chol LUVs. The R18-labeled and R18-free liposomes were mixed and the time-course increase of fluorescence intensity due to the dequenching of the probe was measured to follow-up lipid fusion. Fig. 4 clearly shows that the peptide 97,848 induces lipid mixing. The process is dose-dependent. The fusogenic activity of the peptide is further assessed by core mixing experiments (Table 3). Leakage assays confirmed the lipid-destabilizing properties of the peptide, as shown in Table 3. The process was also dependent on the peptide concentration. In three independent experiments, the 97,848 peptide appeared slightly but constantly more potent than the SIV peptide.

The conformation of the peptide alone and in the presence of lipids was analyzed by attenuated total reflection FTIR spectroscopy (Table 4). The peptide is mainly helical in both conditions, in agreement with our calculations. It should be noted that the purity of the peptide is 85%.

Automatic design of peptides is little described in the literature. Some computer programs, such as LUDI,were set to design protein ligands, from organic compounds to peptidomimetics 3. It was shown on inhibitors of HIV-1 protease that this algorithm essentially helps in the initial steps of the design procedure 47. Other procedures, such as DESIGNER, use a structure template, a rotamer library, and an empirical force field; this was notably applied to MHC class I binding peptides 2. Very recently, Decaffmeyer et al. 48 developed an in silico approach, called Pep Design, to design complementary peptides. The assay was used to design a partner to the C-terminal fusion domain of the Aβ peptide involved in the Alzheimer’s disease. The peptide was made from natural amino acids. The leading idea was derived from the observation that the lipid-binding domain (and more specifically the C-terminal helix) of apolipoprotein E (apoE) is able to decrease the destabilizing properties of the Aβ peptide 49,50. The “de novo” peptide, derived from the apoE C-terminal helix is able to interact with the C-terminal fusogenic domain of Aβ peptide and to decrease its lipid-destabilizing properties with a higher efficiency than the apoE wild-type peptide.

In this study, we aimed to design peptides made of non-natural amino acids and eliciting peculiar lipid-interacting properties mimicking those of lipid-destabilizing tilted peptides. The in silico selection was based on several criteria common to tilted peptides 16,28. Tilted peptides are able to interact with and insert obliquely within a lipid bilayer. This behavior is notably linked to their mean hydrophobicity value (the most efficient peptides being between 0.6 and 0.7kcal/residue according to the hydrophobicity scale of Eisenberg). Tilted peptides also present an asymmetric distribution of hydrophobic residues when helical.

A de novo peptide (12 residues long) exclusively made of non-natural amino acids was developed. Fourteen commercially available non-natural amino acids were selected. They have a normal backbone configuration but modified lateral chains such as aliphatic chains with variable number of carbon atoms, double aromatic rings, etc.

Ten million sequences were randomly generated. Taking into account the mean peptide hydrophobicity, the helix orientation toward a membrane interface, the conformational stability of the structure in membrane, and the presence of a charged residue, only four peptides met the criteria; one was synthesized.

Biophysical assays on liposomes clearly showed that the peptide is fusogenic and destabilizes lipid vesicles with an efficacy slightly superior to the SIV tilted peptide, the paradigm in the domain. Peptide 97,848 is mainly helical in water solution and in lipids, in agreement with its stability during the angular dynamics.

The systematic test of the peptide insertion in IMPALA clearly demonstrates that the peptide’s center of mass might easily fluctuate between ±13Å and ±5Å close to the phospholipid headgroups/acyl-chain interface. This suggest that the peptide has easy access to a wide range of membrane levels and the existence of such metastable positions was already suggested for other tilted peptides 28 and should contribute to their destabilizing capacity.

The stability of the peptide was investigated by 10,000 steps of angular dynamics. The main benefit of angular versus molecular dynamics lies in its efficacy to investigate large molecular movement within a reasonable delay 34. For instance, the angular dynamics procedure required 1/100 the number of calculation steps that the molecular dynamics needed to explore the same conformational space of rotamer movements. Investigation of large movement is facilitated by the use of the Levitt’s modification of Lennard Jones equation for the Van der Waals energy term. Indeed, one major problem in dynamics studies is due to the fact that all movements are discrete steps, not soft continuous movement. Hence, if steps are very short, steric clashes can be avoided, but the equilibrium will be reached after very long computations. As soon as movements are larger, steric clashes and thus, Van de Waals energy bursts, create problems in the computation of energy minimization. This is why Levitt’s modification of the Van der Waals energy, which limits the energy burst of a steric clash to a value, one that we fixed to 1,000kcal in this case, is helpful. In other experiments, the value was set to range between 100 and 10,000kcal, according to the size of the molecule. The goal is to leave room for a balancing effect of the other energy terms for the entire course of the dynamics. The angular dynamics is interesting, because the movement is distributed in the structure, and the rotation of a ϕ- or ψ-angle in the middle of the structure will have more structural consequences than the rotation of an equivalent axis on the N- or the C-side or than the rotation of χ-angles of side chain. Cooperative stability of folded domains is, by this way, clearly evidenced. In the hydrophobic domain of IMPALA membrane, the peptide 97,848 backbone remains stable; ϕ/ψ-values are little changed while side chains are moving, creating intramolecular steric contacts that disappear after 2,000 steps—supporting the conclusion that the structure has found a local energy minimum.

Those results suggest that the peptide is stable in the membrane and that its insertion might generate conditions where lipid perturbation can occur and lead to a fusion process.

From this study, we support the idea that peptides with specific lipid-interacting properties can be de novo-designed using bioinformatic tools. The experimental tests validate the approach since a de novo peptide was able to destabilize liposomes and to adopt a stable helical conformation. By enabling the easy use of modified residues, the methods described in this article should be widely interesting even at a therapeutic design point of view. Indeed, non-natural amino acids may be more resistant to protease degradation and present a more decreased immunogenicity than natural ones 10, and specific lipid binding capacities could be useful to target virosomes and liposomes to specific cell types.