Molecular dynamics simulations of helical antimicrobial ...Molecular dynamics simulations of helical antimicrobial peptides ... We report long time scale simulations of the 18-residue

Peptides 26 (2005) 2037–2049

Molecular dynamics simulations of helical antimicrobial peptidesin SDS micelles: What do point mutations achieve?

Himanshu Khandelia, Yiannis N. Kaznessis∗

Department of Chemical Engineering and Materials Science and The Digital Technology Center, University of Minnesota,421, Washington Avenue SE, Minneapolis, MN 55455, USA

Received 14 January 2005; received in revised form 16 March 2005; accepted 18 March 2005Available online 24 June 2005

Abstract

We report long time scale simulations of the 18-residue helical antimicrobial peptide ovispirin-1 and its analogs novispirin-G10 andnovispirin-T7 in SDS micelles. The SDS micelle serves as an economical and effective model for a cellular membrane. Ovispirin, which isinitially placed along a micelle diameter, diffuses out to the water–SDS interface and stabilizes to an interface-bound steady state in 16.35 nsof simulation. The final conformation, orientation, and the structure of ovispirin are in good agreement with the experimentally observed

und peptide. Althoughhich make

/hinge at theattenuate

imi-ains.ro-celltionhasotics

ine-is

tent

ides.cantam-ions.tKI

properties of the peptide in presence of lipid bilayers. The simulation succeeds in capturing subtle differences of the membrane-bostructure as predicted by solid state NMR. The novispirins also undergo identical diffusion patterns and similar final conformationsthe final interface-bound states are similar, the simulations illuminate the structural and binding properties of the mutant peptides wthem less toxic compared to ovispirin. Based on previous data and the current simulations, we propose that introduction of a bendcenter of helical antimicrobial peptides (containing a specific C-terminal motif), without disrupting the helicity of the peptides mighthost-cell toxicity as well as improve membrane binding properties to bacterial cellular envelopes.© 2005 Elsevier Inc. All rights reserved.

Keywords: Ovispirin; SDS micelle; Molecular dynamics simulations; Antimicrobial peptides; Peptide–membrane interaction

1. Introduction

All animals possess evolutionarily ancient mechanisms forrecognizing and resisting attack by microorganisms. Muchof the immune resistance comes without the delay associ-ated with the development of a complete specific immuneresponse. Gene encoded antimicrobial peptides (AMPs) arenow well known to be a pervasive component of the immunedefense system all across the plant and animal kingdom.Research in this area has been galvanized by increasinglyworrying reports of bacterial resistance to conventionalantibiotics over the past two decades[23,25,31,34]. Theemergence of multi-drug resistant bacteria can lead to severeconsequences like treatment failure and increased mortality.AMPs are being investigated as potential substitutes to con-

∗ Corresponding author. Tel.: +1 612 624 4197; fax: +1 612 626 7246.E-mail address: [email protected] (Y.N. Kaznessis).

ventional antibiotics in the hope of concocting novel antcrobial agents to control antibiotic-resistant bacterial strAlthough it is fairly well established that AMPs lyse micbial cells by direct attack on the lipid bilayer of the innermembrane[6,9], the detailed molecular mechanism of acof most AMPs remains unclear. This lack of know-howlargely hampered the design of effective peptide antibiof therapeutic value.

Mammalian AMPs have been classified into cysterich (� or �) defensins and the cathelicidins. SMAP-29a 29-residue, helical bovine cathelicidin peptide with poantimicrobial activity[12]. Kalfa and coworkers[12] sys-tematically altered SMAP-29 to produce 23 related peptOne of the peptides, ovispirin-1, was found to have signifiantimicrobial activity against both Gram-positive and Grnegative bacteria for a wide range of salt concentratOvispirin-1 (KNLRR IIRKI IHIIK KYG) resembles the firsN-terminal 18 amino acids of SMAP-29 (RGLRR LGR

0196-9781/$ – see front matter © 2005 Elsevier Inc. All rights reserved.doi:10.1016/j.peptides.2005.03.058

2038 H. Khandelia, Y.N. Kaznessis / Peptides 26 (2005) 2037–2049

Table 1Representative functional properties of OVIS, G10 and T7

Toxicity Ovispirin-1 Novispirin-G10 Novispirin-T7

% Hemolysis (human RBCs) 70% at 80�g/ml 2.5% at 80�g/ml 10% at 80�g/ml% Cytotoxicity (cervical cells) 93.6% at 50�g/ml 5% at 50�g/ml 28.4% at 50�g/ml

Antimicrobial activity Minimum effective concentration (�g/ml)

P. aeruginosa 1.66 (mean) 2.97 (mean) 4.64 (mean)S. aureus 1.0± 0.13 4.6± 1.7 3.3± 1.5L. monocytogenes 1.5± 0.3 1.4± 0.4 2.4± 1.5

Data obtained from the work of Sawai et al.[28].

AHGVK KYG) [12]. However, ovispirin-1, like SMAP-29,is unsuitable for therapeutic use owing to its hemolytic prop-erties, which cause large-scale damage to human erythrocytesand epithelial cells. Sawai et al.[28] carried out single residuemutations in ovispirin-1 in an attempt to improve antimicro-bial activity and subdue hemolytic properties. Residue 10 ofovispirin-1 was changed from isoleucine to glycine to makenovispirin-G10 (KNLRR IIRKG IHIIK KYG). Residue 7 ofovispirin-1 was changed from isoleucine to threonine to makenovispirin-T7 (KNLRR ITRKI IHIIK KYG). From now on,we will refer to the three peptides as OVIS, G10 and T7.Antimicrobial activity of the peptides was measured in termsof a minimum effective concentration using a radial diffusionassay. Hemolytic activity was measured in terms of percenthemoglobin released after a fixed-time incubation with thepeptide.Table 1 lists the properties of the three peptides.It was found that both analogs of OVIS had reduced toxic-ity towards human cells. G10 is least toxic and has the bestantibacterial properties[14].

The three dimensional structures of all three peptides inpresence of tri-fluoro-ethanol (TFE) were evaluated from 2-D NMR. The point mutations led to significant structuralchanges in the peptides, inducing a helix kink in G-10 andN-terminal flexibility in T7 (Fig. 1).

Notably, both mutations involve replacement of ahydrophobic residue with a residue with a non-hydrophobics bicityi er-t enf elyc toa the-s MPib lacko i-cm pho-b es.C dueso aliane cells.T ap-e their

undesirable toxicity towards host cells. Indeed, this remainsthe essential goal of the very many mutational studies beingcarried out on AMPs with the overall objective of fabricatingAMPs as antibiotics with potential clinical utility. However,despite the above knowledge, there are but rare instances ofdesigns of better peptides achieved by simply reducing thehydrophobic content of native peptides. Indeed, these muta-tions must induce many other significant changes in the bio-physical characteristics of the peptides, the molecular detailsof which are not well understood, and are not accessible byexperiments. The primary objective of the current computa-tional study is to evaluate the differential molecular interac-tions of the three peptides with model membrane assembliesin order to examine the overall comprehensive ramificationsof mutations. Indeed, we have discovered that the role of suchsingle residue mutations extends well beyond the superficial-ity of reducing overall hydrophobicity of the peptide.

We have carried out long time scale molecular dynamics(MD) simulations of each of the peptides in anionic sodiumdodecylsulfate (SDS) micelles. We report the results fromthese simulations in the current article. Both SDS and diphos-phocholine (DPC) micelles have been successfully modeled

F s haveb

ide chain. Several studies have indicated that hydrophos directly correlated to hemolytic and/or cytolytic propies in AMPs[15,24]. Mellitin, a bee venom toxin, has beound to interact strongly with both neutral and negativharged lipid bilayers[5]. This effect has been attributedhighly hydrophobic N-terminus. Matsuzaki et al. hypo

ized that magainin-2, a thoroughly examined helical Asolated from the skin of the frogXenopus leavis, did notind to zwitterionic mammalian membranes owing to thef hydrophobic residues[20,21]. The postulation was vindated by studies carried out by Tache et al.[30] where twoagainin analogs with increased position-specific hydroicity were found to bind strongly to zwitterionic membranonceivably, a higher presence of hydrophobic resin helical peptides enhances their binding to mammukaryotic membranes and leads to toxicity against hosthis is a potentially important hypothesis which can chrone the rational design of AMP analogs to attenuate

ig. 1. Sequence and structures of the three peptides. The structureeen reproduced from Sawai et al.[28].

H. Khandelia, Y.N. Kaznessis / Peptides 26 (2005) 2037–2049 2039

for MD simulations in CHARMM [19,36]. The primaryadvantage of using micelles as opposed to lipid bilayers arethe faster time scales of motion of SDS lipids. It has beenshown both experimentally[8] and by molecular dynamicssimulations[26,35,37], that the slowest relaxation times oflipids in SDS and DPC micellar solutions are of the order of500–1000 ps. The only exception to this is the slower relax-ation times of counterions in simulations[19]. On the otherhand, the area per lipid in model phospholipid bilayers relaxeson the order of 10s of nanoseconds, and the response of thebilayers to peptide insertion is relatively sluggish[29]. Thisis one of the principal reasons why there are only scatteredefforts of simulations of AMPs with lipid bilayers. Singlesimulations of peptides in bilayers consume immense compu-tational resources. The interaction of the peptide with the lipidmacromolecular assembly induces a much faster response inmicelles as opposed to bilayers. The micelle contains abouthalf the number of atoms of a typical 128-lipid peptide–waterbilayer simulation cell. This allows much longer simulationsand permits monitoring of biological phenomena of longertime scales. Additionally, sometimes the simulations resultscan be directly compared to the experimental structures ofthe peptide, which are often determined in the presence ofanionic and cationic detergents.

Simulations of micelles of various anionic and zwitteri-onic lipid micellar systems[2,16,19,26,29,32,35–38]haveb infor-m X-r y. Int M,A ter-i Thec tideb tiono dingh owl-e of as in am

2

aterc byM ivem ac f 60S boxo 3Ppi thea ainedf U6,Ti llel

to the interface. However, we did not place the peptide atthe SDS–water interface to avoid biasing the simulationtowards the correct final conformation. Instead, each peptidewas placed in the simulation box with its center of masscoinciding with that of the micelle. In this conformation, thepeptide helix lay along one of the diameters of the micelle,with only its termini exposed to the solvent interface (Fig. 2aand c). Owing to spherical symmetry of the micelle, theorientation of the peptide is unimportant. The fast dynamicaltime scales of SDS molecules ensured that the repositioningof the peptide to its final conformation would be computa-tionally tractable. The side chains of lysines and arginineswere protonated. Both the C-terminus and the N-terminuswere charged. Thus, the peptide had a total charge of +7.

To remove initial bad contacts between the peptide andthe SDS core, and prevent penetration of water duringequilibration, the peptide and bulk water were kept underweak harmonic constraints with spring constants of 10 and5 kcal/molA, respectively. The constraints were graduallyremoved in 20,000 steps of minimization (steepest descentmethod). The entire system was then minimized for 20,000more steps without any constraints. Thereafter, the systemconsisting of about∼16,000 atoms was gradually heated to303 K. The entire assembly was subjected to NPT dynamicsat pressureP = 1 atm and room temperatureT = 303.15 K forat least 16 ns. The constant pressure-temperature module ofC te-g s setae com-pT arti-ca rou lvingb outu eters n theM tingI msa rees l ofn ep-t tent.E t 2 nso er-t

3

3

esh For

een successfully used to interpret and supplement theation obtained from experimental methods like NMR,

ay diffraction, CD spectroscopy and FTIR spectroscophe past few years, simulation force fields like CHARMMBER and GROMACS have been fairly well parame

zed to accurately simulate hydrated lipid assemblies.urrent study focuses on the initial events involved in pepinding on the micellar surface, stabilization/destabilizaf the secondary structure, and the influence that the binas on the micellar structure and dynamics. To our kndge, this is the first set of long comparative simulationset of three related helical antimicrobial peptide analogsembrane-mimetic environment.

. Methods

The starting coordinates of the SDS micelle–womplex were obtained from simulations carried outacKerell [19]. This structure was obtained after extensinimization and 120 ps of NVT and NPT simulation in

ubic simulation cell. We placed the micelle consisting oDS molecules and 4375 waters in a cubic simulationf cell size 54.15A. Water was modeled using the TIPotential[11]. 60 sodium counterions, and 5 Na+ and Cl−

ons as 0.15 mM electrolyte were randomly distributed inqueous phase. The initial structure of peptides was obt

rom the PDB databank (pdb ids: OVIS:1HU5, G10:1H7:1HU7). Solid state NMR experiments of OVIS[41]

n lipid bilayers suggest that the OVIS helix sits para

HARMM was used for the simulations with a leap-frog inrator. A time step of 2 fs was used. The temperature wat 303.15 K using Hoover temperature control[10]. For thextended system pressure algorithm employed, all theonents of the piston mass array were set to 500 amu[1,27].he electrostatic interactions were simulated using the ple mesh Ewald (PME) summation[3,7] without truncationnd a real space Gaussian width of 0.25A−1, a�-spline ordef 4, and a FFT grid of about one point perA. SHAKE wassed to eliminate the fastest degrees of freedom invoonds with hydrogen atoms. All simulations were carriedsing CHARMM version c29b2 with the param22 paramet. The simulations were run on parallel processors oarvel Tru64 Unix clusters at the Pittsburgh Supercompu

nstitute (PSC) and on the SGI Altix linux-based platfort the Minnesota Supercomputing Institute (MSI). All thimulations were carried out for 16.35 ns, which is a totaearly 50 ns of simulation. No previous simulations of p

ides in SDS micelles have been carried out to this exnsemble average properties were calculated for the lasf each simulation. For calculation of all dynamical prop

ies, trajectories were sampled every 5 ps.

. Results

.1. Peptide position

As described in the Section2, all ensemble averagave been taken for the last 2 ns of simulation.


Fig. 2. Initial (left) and the final (right) conformations of the simulation. (a and c) Side views; (b and d) top views. Snapshots were taken at thet = 0 and 16 ns.

dynamical properties, the trajectories were sampled every5 ps.

In each of the three simulations, the peptide diffuses fromthe core of the micelle to an interface bound conformationwhich is energetically more favorable. The diffusion of thepeptides occurs on the time scales of 8–12 ns. The initial andfinal conformations of OVIS w.r.t. the micelle are shown inFig. 2. G10 and T7 also have similar final binding confor-mations (seeFig. 6). Given that all three peptides are highlyamphipathic (see later), it was to be expected that an embed-ded state in the micelle would be energetically unfavorable.

Fig. 3 shows the temporal distance profile between thecenter of masses of the peptide and the micelle. In all three

F of thep

cases, the peptide traverses a distance of about 14A, whichclosely corresponds to the radius of gyration of the micelle.The peptide thereafter remains bound to the interface andshows no sign of adopting a new conformation, position ororientation for a further∼5 ns of simulation. We can safelyassume that the interface-bound conformation of the peptidecorresponds to its equilibrium state in the presence of theSDS membrane-mimetic (Fig. 4).

The observed final conformation of OVIS is in excellentagreement with that predicted from solid state NMR[40]and the proton inverse detected deuteron (PRIDE) NMR tech-nique[39]. In both experiments, the peptide was reconstitutedin lipid vesicles composed of POPC:POPG lipids in the ratio

tion.
ig. 3. Time profile of the distance between the center of masseseptide and SDS. Data was collected every 5 ps. Fig. 4. Average electron densities for the last 2 ns of the OVIS simula


Fig. 5. (a) The suggested binding orientation of ovispirin-1 to lipid bilay-ers from NMR. Hydrophobic residues are shown in red. Reproduced fromYagamuchi et al.[41] with permission. OVIS has a similar conformation inthe simulation (seeFig. 6). (b) Helical wheel diagram of OVIS. (For inter-pretation of the references to color in this figure legend, the reader is referredto the web version of this article.)

3:1. The latter is an anionic lipid. Both types of experimentsindicate that ovispirin prefers to bind nearly parallel to themembrane–water interface (Fig. 5a) (this corresponds to thefinal simulation conformation), and does not insert acrossthe bilayer (this corresponds to the starting simulation con-figuration). It was also suggested from the experiments thatthe non-polar face of the OVIS helix in embedded into thelipid core (Fig. 5a). Indeed, in the equilibrium conformationfrom simulations, the hydrophobic side chains of the pep-tide are anchored into the SDS core, while the polar face isexposed to the solvent interface (Fig. 6). Thus with respectto peptide’s equilibrium position and its orientation relativeto the membrane interface, the simulation results complywith experimental observation. There are various instancesin literature where peptides have been shown to orient at dif-ferent angles with respect to the interface from simulations.For example, gp41 fusion peptide and its mutants oriented atangles varying from 40◦ to 90◦ in POPE bilayers[13]. Thepeptides magainin-2 and MSI-78 have been shown to orientat different angles (parallel to perpendicular) at POPC inter-faces depending on different starting conformations of thepeptide[14].

In helical form, ovispirin demonstrates a conspicuous seg-regation of polar and non-polar residues into two topologicalhalves.Fig. 5b shows a helical wheel diagram of the peptide.The polar half of the helix has a large number (seven) of pos-

itively charged residues. All lysines and arginines lie on thesame half of the ovispirin helix. A subtle yet vital constituentof the electrostatic interactive forces is the cooperativity of thecharged amino acid residues. Although the membrane surfaceis negatively charged, the accumulation of oppositely chargedions at the interface significantly attenuates the attractiveelectrostatic potential near a membrane[22]. In presence ofa dielectric, the ion screening can reduce the electrostaticenergy between two ions to as low as 3 kJ/mol, which iscomparable to thermal fluctuation energies (kT) at room tem-perature. Thus, even in the vicinity of a negatively chargedmembrane, the kinetics of binding of a positively chargedpeptide will be dampened by thermal fluctuation. Yet, mostAMPs can lyse bacterial cells on the time scales of minutes,which indicates that binding must be swift. The collectiveelectrostatic attraction of all seven positively charged residuespotentially accelerates binding kinetics, and allows the pep-tide to rapidly approach the membrane. The other helix faceis hydrophobic, and after initial binding is complete, this partof the peptide will preferentially exclude solvent, and inter-act with the membrane hydrophobic core. The polar regionis now exposed to the aqueous/interfacial phase.

T7 and G10 also share these amphipathic structural fea-tures with ovispirin. Indeed, their binding conformations aresimilar to ovispirin. In both cases, the hydrophobic region ofthe peptide is lodged into the SDS hydrophobic core. It is ofn bacte-r nnerw n ofa ns,t hreep roleo veda

3

dG ,O s ar romr ptidel tidew ptidei uralf g thec ucess ureso

heb thep esist er tofT im-u (

ote here, that all the three peptides possess strong antiial activity, and are expected to interact in the same maith membrane assemblies containing a high proportionionic lipids. However, as a result of the point mutatio

here are subtle differences in the interaction of the teptides with the SDS core which shed light on thef mutations towards the reduced toxicity and/or improntibacterial activity of the mutants.

.2. Peptide structure in the simulations

The NMR-derived structures (Fig. 1) of mutants T7 an10 are different from OVIS[28]. In the presence of TFEVIS is a uniform helix from residues 4 to 16. T7 ha

andom-coil N-terminus and a reduced helix length: fesidues 7 to 17. The presence of glycine in the G10 peeads to the formation of a hinge in the middle of the pephich separates helices 5–11 and 14–16. The G10 pe

s conformationally most flexible. Although these structeatures of the peptides remain mostly conserved durinourse of the simulations, the presence of the micelle indubtle adjustments in structure. The initial and final structf all three peptides have been shown inFig. 7.

In TFE solutions, OVIS is a single helix all along tackbone[28]. However, this structure is not conserved inresence of lipid bilayers. NMR spectra led to the hypoth

hat the peptide might be bent at the C-terminus is ordacilitate better insertion into the lipid bilayer at that end[41].he dihedral angle profiles of OVIS obtained from the slations agree very well with the above hypothesis. Theϕ,


Fig. 6. Final binding conformations of the three peptides. Snapshots were taken from two lateral perspectives after 16 ns of simulation. Hydrophobicresidueshave been shown in red, and basic residues have been shown in blue. SDS is in cyan wireframe. The binding conformation of ovispirin (a and b) are in excellentagreement with the experimentally predicted structure (Fig. 5a). (For interpretation of the references to color in this figure legend, the reader is referred to theweb version of this article.)

ψ) pair at positions 13 and 14 conform to the standard valuesfor helices (−50,−60) for the initial structure of ovispirin inTFE as submitted in the pdb databank. However, the average(ϕ,ψ) values at these positions over the last 2 ns of simulationare (−70, 71) and (−70, 140): clearly deviant from a helicalmotif. Thus, the final membrane bound conformation of thepeptide has a bend in the helix near the C-terminus, as pre-dicted from NMR spectra. OVIS has a favorable hydrophobicgradient towards the C-terminus, and that end of the peptideis thus likely to be deeply inserted into the bilayer[41]. Thebend in the molecule facilitates the C-terminal residues of

the peptide to insert deeper into the bilayer, as hypothesizedfrom experiments. Unfortunately, the spherical curvature ofthe SDS micelle does not allow us to visualize this prefer-ential insertion explicitly. However, the simulations confirmthat the helical bend predicted from NMR is indeed formedwhen the peptide encounters a membrane-like environment(Fig. 8).

Interestingly, the TFE solution structure of G10 alreadyhas a bent C-terminus, courtesy of the glycine-induced hingeat positions 12 and 13. In OVIS, the bend is induced (partly)when the peptide encounters a membrane. It appears that the


Fig. 7. Initial (left) and final (right) structures of the peptides. The final snapshots were taken att = 16 ns. Basic residues are shown in blue, hydrophobic residuesare shown in red. Polar residues are in green. The C-terminal bend in the OVIS helix is discernible in (b). (For interpretation of the references to color in thisfigure legend, the reader is referred to the web version of this article.)

presence of the bend aids peptide binding to the membranein either case. In case of G10, the bend is more pronouncedand also leads to a significant angular offset in the two heli-cal axes. During the course of simulation, the bend in G10extends to positions 14 and 15 (Fig. 9). In the bound state, thesection of the C-terminus which is purely helical is the regioncomprised of residues 16 (lysine) and 17 (tyrosine). The influ-ences of the hinge region on peptide function and on micelledynamics are discussed in more detail in later Section3.3.

The structure of T7 is mostly conserved from the begin-ning to the end of simulation. It continues to have a flexibleN-terminus and a stable helix from positions 7 to 17 (Fig. 10).

F ea inateso r thel

3.3. Pair distribution functions: Role of individualamino acids

All three peptides have seven positively charges residues:lysines in positions 1, 9, 15 and 16 and arginines in positions4, 5 and 8. The positively charged centers on both side chainsget associated with sulfate ions on SDS. InFig. 11, pair dis-tribution functions (pdfs) between negatively charged centerson SDS (sulfur atom) and positively charged centers on thearginine and lysine residues have been constructed.

The strong electrostatic attraction between SDS and theamino acid side chains is manifest in the coincident sharp

F ea inateso e last2

ig. 8. Dihedral anglesϕ andψ of OVIS. The initial dihedral values for thngles (open boxes and circles) were obtained from the initial coordf the OVIS pdb. The final values were obtained from averaging ove

ast 2 ns. Error bars have been eliminated for clarity.

ig. 9. Dihedral anglesϕ andψ of G10. The initial dihedral values for thngles (open boxes and circles) were obtained from the initial coordf the G10 pdb. The final values were obtained from averaging over thns. Error bars have been eliminated for clarity.


Fig. 10. Dihedral anglesϕ andψ of T7. The initial dihedral values for theangles (open boxes and circles) were obtained from the initial coordinatesof the T7 pdb. The final values were obtained from averaging over the last2 ns. The error bars have been eliminated for clarity.

spikes in the pdfs. The pdfs unequivocally demonstrate thatpositively charged residues drive the binding of the peptideto the micellar surface. Although there are minor differencesin the heights of the peaks for the first shell, the bindingproperties of the positively charged residues are similar forall three peptides, with the exception of lysine at position 15.Unlike in T7 or OVIS, the pdf of lysine in position 15 in G10has a high peak. The insertion of a glycine residue in G10results in the formation of two helices separated by a hingeat position 12–13. The hinge results in a twist in the G10

helix which allows the C-terminal Lys-15 to be significantlyinvolved in electrostatic binding to the membrane. This isnot possible with either T7 or with OVIS because no suchexplicit bend is formed. Additionally, the hinge also allowsthe aromatic ring of the tyrosine residue at position 17 toassociate strongly with the micellar core as is evident fromthe pronounced peak in the pdfs constructed inFig. 12.

The implication of the insertion of the glycine residue inposition 10 is thus not confined to reducing the hydrophobic-ity of the ovispirin helix (which was the primary objectiveof Sawai et al.[28] when making the mutation in the firstplace). It plays a more significant role in altering the struc-ture of G10 so that the C-terminus of the peptide can bindstrongly to the SDS core by means of both electrostatic andhydrophobic interactions. Other workers have found the sametype of mutation leading to less toxic peptides. Recently,Lee et al.[18] modified IsCT (ILGKI WEGIK SLF), a non-cell-selective 13 amino acid helical scorpion AMP by threemutations, one of which involved the introduction of pro-line for a glycine at position 8. The resulting peptide IsCT-1(ILGKI WKPIK KLF) had a bend similar to G10. Indeed, thismutation did not involve removal of a hydrophobic residue,yet had much lower hemolytic and better antibacterial activ-ity. Like the ovispirins, IsCT and its analog also have the samemotif at the C-terminal helical turn: two positively chargedresidues (lysines) followed by an aromatic ring containingr evi-d m to

F d the polar residues on the peptides. Thex-axis is distance inA. For lysines and arginines,tn

ig. 11. Pair distribution functions between the sulfate head groups an
he positively charged centers on the side chains were used to draw the pdfsormalized by a density of 0.01 and by the number of atoms. Lys-15 is well a
esidue (phenylalanine). Remarkably, the simulationsence that the helix hinge in G10 (at position 12) see

. For histidine and asparagine, all side chain heavy atoms were used. The pdfs weressociated with the sulfates in G10, but not so much in OVIS or T7.


Fig. 12. Pair distribution functions between the SDS core hydrocarbon atoms and the hydrophobic residues on the peptides. All side chain carbon atomsonthe amino acids were used to draw the pdfs. The pdfs were normalized by a density of 0.01 and by the number of atoms. Tyr-17 in G10 has a strong preferenceto interact with the SDS core. Note the differenty-axis scale for Tyr-17.

serve the same purpose as the hinge in IsCT (at position 9)observed experimentally: improve binding of the peptide’sC-terminus to the membrane by a rotation of the helical axismediated by a hinge. The improved binding of the peptideto the membrane is directly correlated to the improved activ-ity of IsCT-1 and G10 compared to their parent peptides.Interestingly, Tossi et al.’s[33] 7-year-old design of a helicalAMP ‘PGG‘based on sequence homology (GLLRR LRKKIGEIFK KYG) also carries the exact same motif: a glycine inthe middle of the molecule, and positively charged residuesfollowed by an aromatic amino acid at the C-terminus. It ishighly probable that Tossi’s consensus sequence functionsthe same way as G10 and IsCT. Apparently, in OVIS, theC-terminal bend only partially achieves this objective, as thetyrosine at position 17 or the lysine at position 15 do notexhibit a strong association with SDS (Fig. 13).

The pair distribution functions at position 7 reveal someof the reasons behind the reduced toxicity of T7. As antici-pated, it turns out that this is directly related to lesser numberof hydrophobic residues present in T7 compared to OVIS.Thr-7 in T7 has a peak almost as sharp as other positivelycharged residues, indicating that the sulfates form stableH-bonds with the hydroxyl moiety of threonine at the inter-face. Position 7 in OVIS, on the other hand, is occupied byisoleucine, which prefers to localize inside the SDS core.Certainly, the replacement of Ile-7 by Thr-7 diminishes thea e oft ppo-

site viewpoint. If we mutate T7 by the replacement of a polar(THR-7) residue for a hydrophobic (ILE) residue, the result-ing peptide (OVIS) associates better with the hydrophobiccore of a membrane. These non-polar binding forces becomeespecially significant when the peptide binds to non-anionicmammalian membranes in the absence of direct cationic-peptide–anionic-membrane electrostatic forces.

Fig. 13. Pair distribution function of the SDS head group with peptideresidues at position 7. For isoleucine, all non-backbone carbons were used.F wereu ber ofa

bility of the peptide to interact with the hydrophobic corhe micelle. This argument becomes definitive from an o

or threonine, all non-backbone carbons, and the hydroxyl oxygensed. The pdfs were normalized by a density of 0.01 and by the numtoms.


Fig. 14. Varying structure of the micelle over time. SDS is shown in CPK cyan. The peptide is shown in dark blue. The top view is shown at all times. Thebulk rotation of the system was nullified before taking snapshots. (For interpretation of the references to color in this figure legend, the reader is referred to theweb version of this article.)

This is an important inference, because it directlyaddresses and corroborates abundant experimental evidencewherein increased presence of hydrophobic residues hasbeen shown to enhance host-cell (eukaryotic) toxicity and/orinduce increased leakage from zwitterionic lipid vesicles[4,5,15,20,21,24,30].

The same argument would also apply to position 10 incase of G10, where ILE-10 is replaced by GLY-10. However,in that case, the reduced hydrophobic binding at position 10is largely offset by the increased association of the phenylgroup at position 17.

3.4. Micelle structure and behavior

The placement of the peptide in the core of the SDS micellein the starting conformation leads to numerous bad contacts.These steric overlaps disturb the structure and dynamics ofthe micelle in the initial phase of the simulations. In all threesimulations, the micelle remains in a perturbed non-sphericalstate during the best of the first 9 ns of simulation. This is thetime when the peptide diffuses out of the micelle interiorto the interface. However, once the peptide approaches thesteady state at the interface, the micelle structure equilibratesin the final 5–6 ns of simulation (Fig. 14).

In order to monitor the equilibration, sphericity and struc-ture of the micelle, we calculated the eccentricity of them

e

ww et of 0,a canbt thes s oft lvess r-m es hent fter,

the micelle aggregates back to a more spherical shape in thefinal conformation. In all three cases, the final value ofehovers around 0.15. The ratios of moments of inertia in thex, y and z directions are also good estimates for monitor-ing micelle structure. As with the eccentricity, the ratios arehigh initially (indicating a more ellipsoidal shape) and sta-bilize to about 1.2 at the end of the simulations. Moment ofinertia ratios ranging from 1.02 to 1.15 and eccentricities of0.03 have been reported in previous simulations of pure SDSmicelles carried out for different lengths of simulations timeand different force fields (CHARMM[19], AMBER [2,26]).Indeed, an eccentricity of 0.15 and an inertia ratio of 1.2in our simulations indicate that the micelle deviates signifi-cantly from a perfectly spherical arrangement. This is owingto the presence of the peptide which forms strong hydro-gen bonds with the SDS head groups, and also interacts withthe micellar core by way of non-polar interactions. The twotypes of interactions apparently lead to localized rigidity inthe micelle in the region abutting the peptide, causing a dis-continuity in the symmetry of the biophysical forces whichotherwise enable formation of a purely spherical micelle(Fig. 15).

The radius of gyration (Rg) of the micelle also stabilizesonce the peptide diffuses out. A lowerRg corresponds to amore spherical micelle. The equilibrium value in all threesimulations is 15.5A. For pure SDS micelles, the valuesr1

tt in theG theh wsi sserd

oma lvente hards[ .4a aread ng toa icals

icelle as[2]:

= 1 − Imin

Iavg

hereImin is the moment of inertia along thex, y or z axesith the smallest magnitude, andIavg is the average of th

hree. For a perfect sphere, the eccentricity has a valuend the deviation of the structure from perfect sphericitye measured by monitoring the value ofe. Fig. 15a shows

he eccentricity of the micelle during the full course ofimulations. The presence of the peptide in the first 3 nhe simulations causes the micelle to split into two haeparated by the peptide (Fig. 14). This results in abnoally high values ofe. However, the fluctuations becom

maller and steadier for the last 5 ns of the simulation, whe peptide has diffused out of the micelle core. Therea

eported in earlier simulations are 16A [19], 16.2A [2] and6.4A [26].

A careful look at the profiles for theRg ande reveal thahe micelle remains more spherical and less perturbed10 simulation, compared to T7 and OVIS. Apparently,igher conformational flexibility of the G10 peptide allo

t to emerge from the core of the micelle while causing leisarrangement of the micellar lipids.

The hydration of the micelle, as well as its deviation frspherical structure can also be measured by its so

xposed surface area. We calculated the Lee and Ric17] surface area for the micelle with a probe radius of 1And an accuracy of 0.05. As anticipated, the surfaceecreases during the course of the simulation, equilibrativalue of 10,900A2. Again, this represents a more spherhape of the micelle as the simulations proceeds.


Fig. 15. Structural properties of the micelle. (a) Time profile of the eccentricity; (b) ratio of the moment of inertiaIyy/Ixx. Izz/Ixx (not shown) also has a similarprofile; (c) radius of gyration; (d) solvent exposed surface area. The Lee and Richards area was calculated with a probe radius of 1.4A. Data was collectedevery 5 ps for each of the above.

All of the above results pertaining to the structure of themicelle independently attest that the micelle has reached itsequilibrium conformation for all three simulations. Longersimulations are not necessary to observe any of the biophys-ical phenomena we are interested in, namely: binding of thepeptides to the membrane mimetic, their final conformationand orientation, and the influence of the peptide on the struc-ture of the micelle and vice versa.

4. Discussion

In the current work, we have reported long timescalemolecular dynamics simulations of three similar helicalantimicrobial peptides with SDS micelles. Despite start-ing from a peptide position far from the experimentallyobserved binding-conformation of OVIS in bilayers, the sim-ulations converged to the correct final position, and theresults are in excellent agreement with experimental data.Additional simulations with the mutant peptides G10 andT7 allow us to compare the biophysics of their differentialpeptide–membrane interactions, to help understand the deter-minants of their less toxic behavior. Amongst other things,we discover that the reduced hydrophobicity and the alteredstructure of these peptides are primarily responsible for alower toxicity level compared to ovispirin. The mutation ofa pep-t . Thes them wop due)o imic.T struc-

tural motif has been present, but has passed undetected. Thecurrent computational study helps uncover the function ofthis sequence and structural motif in AMPs which might beresponsible for reduced toxicity of many helical AMPs.

It is commonly taken for granted that the structures ofpeptides in alcoholic solvents are the same as when boundto membranes. Previous short simulations (<3 ns) have illus-trated that this may not be the case, and the change in structurehas often been attributed to peptide–headgroup interactions;for an example, see Wymore and Wong[37]. Our 40 ns sim-ulations also illustrate that the peptide structure from thesimulations is different compared to the structures observedexperimentally in organic solvents. As we have seen, thesubtle differences in structural detail can have a significantinfluence on peptide function and activity. We thus needto exercise caution when making structure–activity–toxicitycorrelations based entirely on the solved structures of AMPsin aqueous or non-aqueous solvents. A better approach wouldbe to solve the structures in better membrane-mimetic envi-ronments, and/or refine them with data available from simu-lations.

Although not as appropriate a model for a real membraneas a lipid bilayer, the SDS micelle captures the essential fea-tures which modulate peptide–membrane interaction. Theseare the presence of a strongly hydrophobic core, and a flex-ible polar interface capable of forming H-bonds and saltb ble ofr ions.T scaleo imu-l tidef n them ipid

n isoleucine to a threonine reduces the ability of theide to associate with the hydrophobic membrane coreimulations also indicate that a glycine-induced bend iniddle of helical AMPs enables the C-terminal motif (tositively charged residues followed by an aromatic resif the peptide to associate better with the membrane mhere are other instances in literature where the same

ridges with the solvent and the peptide, and also capaesponding dynamically to peptide–membrane interacthe obvious advantages of using SDS are the faster timef motion and smaller system size which enables longer s

ations. We expect that the remarkable diffusion of the peprom the membrane core to the interface as observed iicelle would not be observable in all atom models of l


bilayers in the same time frame, due to the relative rigidityof membrane phospholipids.

To our knowledge, this is the first exhaustive set of sim-ulations which addresses the molecular basis of innumerousattempts being made to design better antimicrobial peptidicmolecules by means of modification of existing peptides.The current work demonstrates that simulations of differ-ent peptides under identical conditions sets up the basis forgenerating comparative molecular scale information whichcan be of immense help in interpreting experimental data,and helping understand atomic-scale physical interactionswhich result in slightly different peptides having vastly dif-ferent cytotoxic and antibacterial activities. In the future,a rational comprehension of the role of these structural,biochemical and physical factors will be indispensable tothe goal of designing AMPs which can have therapeuticvalue.

Acknowledgements

This work was supported by grants from IBM (Facultyaward to Y.N.K) and 3 M (Young Faculty award to Y.N.K).H.K. has an Amundson Fellowship at the Department ofChemical Engineering and Materials Science, Universityo otaS ed.T pu-t izedt nter.

R

skylytic

m-llar

em B

anPhys

ofof

bialanes

heir.

der-Phys

lventPhys

ity

[10] Hoover WG. Canonical dynamics: equilibrium phase-space distribu-tions. Phys Rev A 1985;31:1695–7.

[11] Jorgensen WL, Chandrasekhar J, Medura JD, Impey RW, Klein ML.Comparison of simple potential function for simulating liquid water.J Chem Phys 1983;79:926–35.

[12] Kalfa VC, Jia HP, Kunkle RA, McCray Jr PB, Tack BF, BrogdenKA. Congeners of SMAP29 kill ovine pathogens and induce ultra-structural damage in bacterial cells. Antimicrob Agents Chemother2001;45:3256–61.

[13] Kamath S, Wong TC. Membrane structure of the human immunodefi-ciency virus gp41 fusion domain by molecular dynamics simulation.Biophys J 2002;83:135–43.

[14] Kandasamy SK, Larson RG. Binding and insertion of alpha-helicalanti-microbial peptides in POPC bilayers studied by moleculardynamics simulations. Chem Phys Lipids 2004;132:113–32.

[15] Kwon M, Hong S, Lee K. Structure–activity analysis of brevinin1E amide, an antimicrobial peptide fromRana esculenta. BiochimBiophys Acta 1998;1387:239–48.

[16] Langham A, Kaznessis Y. Simulation of the N-terminus of HIV-Iglycoprotein 41000 fusion peptide in micelles. J Pept Sci, 2005; 11(4): 215–24.

[17] Lee B, Richards FM. Interpretation of protein structures: estimationof static accessibility. J Mol Biol 1971;55:379–400.

[18] Lee K, Shin SY, Kim K, Lim SS, Hahm K-S, Kim Y. Antibioticactivity and structural analysis of the scorpion-derived antimicro-bial peptide IsCT and its analogs. Biochem Biophys Res Commun2004;323:712–9.

[19] MacKerell Jr AD. Molecular dynamics simulation analysis of asodium dodecyl sulfate micelle in aqueous solution: decreasedfluidity of the micelle hydrocarbon interior. J Phys Chem1995;99:1846–55.

[ r-andActa

[ mi-lipo-cteria.

[ nter-

[ sh)

[ er-fectsRes

[ iotict and4;7:

[ nicB

[ f theolec-

[ R,cturetein

[ theeptinm J

[ dentthe

f Minnesota. Computational support from the Minnesupercomputing Institute (MSI) is gratefully acknowledghis work was also partially supported by National Com

ational Science Alliance under MCB030027P and utilhe marvel cluster at the Pittsburgh Supercomputing Ce

eferences

[1] Anderson DH, Sawaya MR, Cascio D, Ernst W, Modlin R, KrenA, et al. Granulysin crystal structure and a structure-derivedmechanism. J Mol Biol 2003;325:355–65.

[2] Bruce CD, Berkowitz ML, Perera L, Forbes MDE. Molecular dynaics simulation of sodium dodecyl sulfate micelle in water: micestructural characteristics and counterion distribution. J Phys Ch2002;106:3788–93.

[3] Darden T, York D, Pedersen L. Particle mesh Ewald:N log(N) method for Ewald sums in large systems. J Chem1993;98:10089–92.

[4] Dathe M, Nikolenko H, Beyermann M, Bienert M. Optimizationthe antimicrobial activity of magainin peptides by modificationcharge. FEBS Lett 2001;501:146–50.

[5] Dathe M, Wieprecht T. Structural features of helical antimicropeptides: their potential to modulate activity on model membrand biological cells. Biochim Biophys Acta 1999;1462:71–87.

[6] Epand RM, Vogel HJ. Diversity of antimicrobial peptides and tmechanisms of action. Biochim Biophys Acta 1999;1462:11–28

[7] Essmann U, Perera L, Berkowitz ML, Darden T, Lee H, Pesen LG. A smooth particle mesh Ewald method. J Chem1995;103:8577–93.

[8] Fernandez P, Schroedle S, Buchner R, Kunz W. Micelle and sorelaxation in aqueous sodium dodecylsulfate solutions. ChemChem 2003;4:1065–72.

[9] Ganz T, Lehrer RI. Antimicrobial peptides. Innate Immun2003:287–303.

20] Matsuzaki K, Sugishita K, Harada M, Fujii N, Miyajima K. Inteactions of an antimicrobial peptide, magainin 2, with outerinner membranes of Gram-negative bacteria. Biochim Biophys1997;1327:119–30.

21] Matsuzaki K, Sugishita K, Miyajima K. Interactions of an anticrobial peptide, magainin 2, with lipopolysaccharide-containingsomes as a model for outer membranes of Gram-negative baFEBS Lett 1999;449:221–4.

22] McLaughlin S. Electrostatic potentials at membrane–solution ifaces. Curr Top Membr Transport 1977;9:71–144.

23] Morell V. Antibiotic resistance: road of no return. Science (Wa1997;278:575–6.

24] Ohmori N, Niidome T, Hatakeyama T, Mihara H, Aoyagi H. Intaction of a-helical peptides with phospholipid membrane: efof chain length and hydrophobicity of peptides. J Pept1998;51:103–9.

25] Payne D, Tomasz A. Antimicrobials the challenge of antibresistant bacterial pathogens: the medical need, the markeprospects for new antimicrobial agents. Curr Opin Microbiol 200435–8.

26] Rakitin AR, Pack GR. Molecular dynamics simulations of iointeractions with dodecyl sulfate micelles. J Phys Chem2004;108:2712–6.

27] Ryckaert JP, Ciccotti G, Berendsen HJC. Numerical integration oCartesian equations of motion of a system with constraints: mular dynamics ofn-alkanes. J Comput Phys 1977;23:327–41.

28] Sawai MV, Waring AJ, Kearney WR, McCray Jr PB, Forsyth WLehrer RI, et al. Impact of single-residue mutations on the struand function of ovispirin/novispirin antimicrobial peptides. ProEng 2002;15:225–32.

29] Shepherd CM, Vogel HJ, Tieleman DP. Interactions ofdesigned antimicrobial peptide MB21 and truncated dermasS3 with lipid bilayers: molecular-dynamics simulations. Bioche2003;370:233–43.

30] Tachi T, Epand RF, Epand RM, Matsuzaki K. Position-depenhydrophobicity of the antimicrobial magainin peptide affects


mode of peptide–lipid interactions and selective toxicity. Biochem-istry 2002;41:10723–31.

[31] Thomson CJ, Power E, Ruebsamen-Waigmann H, Labischinski H.Antibacterial research and development in the 21(st) century—anindustry perspective of the challenges. Curr Opin Microbiol2004;7:445–50.

[32] Tieleman DP, Spoel van der D, Berendsen HJC. Molecular dynam-ics simulations of dodecylphosphocholine micelles at three differentaggregate sizes: micellar structure and chain relaxation. J Phys ChemB 2000;104:6380–8.

[33] Tossi A, Tarantino C, Romeo D. Design of synthetic antimicro-bial peptides based on sequence analogy and amphipathicity. Eur JBiochem 1997;250:549–58.

[34] Walsh FM, Amyes SG. Microbiology and drug resistance mech-anisms of fully resistant pathogens. Curr Opin Microbiol 2004;7:439–44.

[35] Wong TC, Kamath S. Molecular dynamics simulations of adrenocor-ticotropin (1–24) peptide in a solvated dodecylphosphocholine (DPC)micelle and in a dimyistoylphosphatidylcholine (DMPC) bilayer. JBiomol Struct Dyn 2002;20:39–57.

[36] Wymore T, Gao XF, Wong TC. Molecular dynamics simulation ofthe structure and dynamics of a dodecylphosphocholine micelle inaqueous solution. J Mol Struct 1999;485–486:195–210.

[37] Wymore T, Wong TC. Molecular dynamics study of substance Ppeptides partitioned in a sodium dodecylsulfate micelle. Biophys J1999;76:1213–27.

[38] Wymore T, Wong TC. The structure and dynamics of ACTH (1–10)on the surface of a sodium dodecylsulfate (SDS) micelle: a moleculardynamics simulation study. J Biomol Struct Dyn 2000;18:461–76.

[39] Yamaguchi S, Hong M. Determination of membrane peptide ori-entation by 1H-detected 2H NMR spectroscopy. J Magn Reson1997;155:244–50.

[40] Yamaguchi S, Hong T, Waring A, Lehrer RI, Hong M. Solid-state NMR investigations of peptide–lipid interaction and orien-tation of a b-sheet antimicrobial peptide, protegrin. Biochemistry2001;41:9852–62.

[41] Yamaguchi S, Huster D, Waring A, Lehrer RI, Kearney W, TackBF, et al. Orientation and dynamics of an antimicrobial peptidein the lipid bilayer by solid-state NMR spectroscopy. Biophys J2001;81:2203–14.

Documents

Molecular dynamics simulations of helical antimicrobial ...Molecular dynamics simulations of helical antimicrobial peptides ... We report long time scale simulations of the 18-residue