Molecular Dynamics Simulations of Peptide−Surface Interactions

Molecular Dynamics Simulations of Peptide-SurfaceInteractions

Vivek P. Raut,† Madhuri A. Agashe,† Steven J. Stuart,‡ and Robert A. Latour*,†

Department of Bioengineering and Department of Chemistry, Clemson University,Clemson, South Carolina 29634

Received September 1, 2004. In Final Form: November 14, 2004

Proteins, which are bioactive molecules, adsorb on implants placed in the body through complex andpoorly understood mechanisms and directly influence biocompatibility. Molecular dynamics modelingusing empirical force fields provides one of the most direct methods of theoretically analyzing the behaviorof complex molecular systems and is well-suited for the simulation of protein adsorption behavior. Toaccurately simulate protein adsorption behavior, a force field must correctly represent the thermodynamicdriving forces that govern peptide residue-surface interactions. However, since existing force fields weredeveloped without specific consideration of protein-surface interactions, they may not accurately representthis type of molecular behavior. To address this concern, we developed a host-guest peptide adsorptionmodel in the form of a G4-X-G4 peptide (G is glycine, X is a variable residue) to enable determination ofthe contributions to adsorption free energy of different X residues when adsorbed to functionalized Au-alkanethiol self-assembled monolayers (SAMs). We have previously reported experimental results usingsurface plasmon resonance (SPR) spectroscopy to measure the free energy of peptide adsorption for thispeptide model with X ) G and K (lysine) on OH and COOH functionalized SAMs. The objectives of thepresent research were the development and assessment of methods to calculate adsorption free energyusing molecular dynamics simulations with the GROMACS force field for these same peptide adsorptionsystems, with an oligoethylene oxide (OEG) functionalized SAM surface also being considered. By comparingsimulation results to the experimental results, the accuracy of the selected force field to represent thebehavior of these molecular systems can be evaluated. From our simulations, the G4-G-G4 and G4-K-G4peptides showed minimal to no adsorption to the OH SAM surfaces and the G4-K-G4 showed strong adsorptionto the COOH SAM surface, which is in agreement with our SPR experiments. Contrary to our experimentalresults, however, the simulations predicted a relatively strong adsorption of G4-G-G4 peptide to the COOHSAM surface. In addition, both peptides were unexpectedly predicted to adsorb to the OEG surface. Thesefindings demonstrate the need for GROMACS force field parameters to be rebalanced for the simulationof peptide adsorption behavior on SAM surfaces. The developed methods provide a direct means of assessing,modifying, and validating force field performance for the simulation of peptide and protein adsorption tosurfaces, without which little confidence can be placed in the simulation results that are generated withthese types of systems.

1. Introduction

When a medical implant is placed in the body, sur-rounding proteins rapidly adsorb to the implant surfaceat the solid-liquid interface. Cells present in the bodyfluid, which then come in contact with the implant, do notinteract with the actual implant material itself. Rather,they sense the implant by way of this adsorbed proteinlayer. Together with other interfacial processes, proteinadsorption thus regulates the biological response to anyimplant material in contact with a biological fluid, andhence implant biocompatibility. This phenomenon isattributed to the specific biochemical signaling potentialof a protein.1,2 For example, the adsorption of bloodproteins, such as fibrinogen, can influence the adhesionof platelets or macrophages and ultimately lead tothrombus formation or fibrous encapsulation.2-5 Thus thestudy of protein-surface interactions is extremely im-

portant in the field of biomaterials and in many otherareas of biotechnology.5-15

Because of the relevance of protein-surface interac-tions, much effort has gone into protein adsorp-tion experiments and models over the past severaldecades14,16-18 with an ultimate aim to quantitatively

* Corresponding author. Address: Department of Bioengineer-ing, 501 Rhodes Engineering Research Center, Clemson University,Clemson, SC 29634. E-mail: [email protected]. Phone: 864-656-5552. Fax: 864-656-4466.

† Department of Bioengineering.‡ Department of Chemistry.(1) Anderson, J. M.; Bonfield, T. L.; Ziats, N. Int. J. Artif. Organs

1990, 13, 375-382.(2) Ratner, B. D.; Hoffman, A. S.; Schoen, F. J.; Lemons, J. E. In

Biomaterials Science: An Introduction to Materials in Medicine;Academic Press: San Diego, 1996.

(3) Jahangir, A. R.; McClung, W. G.; Cornelius, R. M.; McCloskey,C. B.; Brash, J. L.; Santerre, J. P. J. Biomed. Mater. Res. 2002, 60,135-147.

(4) Shen, M.; Martinson, L.; Wagner, M. S.; Castner, D. G.; Ratner,B. D.; Horbett, T. A. J. Biomater. Sci., Polym. Ed. 2002, 13, 367-390.

(5) Ratner, B. D. J. Biomed. Mater. Res. 1993, 27, 837-850.(6) Texter, J.; Tirrell, M. AIChE J. 2001, 47, 1706-1710.(7) Nakanishi, K.; Sakiyama, T.; Imamura, K. J. Biosci. Bioeng. 2001,

91, 233-244.(8) Mulzer, S. R.; Brash, J. L. J. Biomed. Mater. Res. 1989, 23, 1483-

1504.(9) Baurmeister, U.; Vienken, J.; Grassmann, A. Nephrol. Dial.

Transplant. (Suppl.) 1991, 3, 17-21.(10) Linnola, R. J.; Werner, L.; Pandey, S. K.; Escobar-Gomez, M.;

Znoiko, S. L.; Apple, D. J. J. Cataract Refractive Surg. 2000, 26, 1792-1806.

(11) Widmer, M. R.; Heuberger, M.; Voros, J.; Spencer, N. D. Tribol.Lett. 2000, 10, 111-116.

(12) Gura, T. A.; Wright, K. L.; Veis, A.; Webb, C. L. J. Biomed.Mater. Res. 1997, 35, 483-495.

(13) Kasemo, B. Surf. Sci. 2002, 500, 656-677.(14) Hlady, V.; Buijs, J. Curr. Opin. Biotechnol. 1996, 7, 72-77.(15) Lange, R.; Bergbauer, M.; Szewzyk, U.; Reitner, J. Facies 2001,

45, 195-202.(16) Brash, J. L.; Horbett, T. A. In Proteins At Interfaces II:

Fundamentals and Applications; American Chemical Society: Wash-ington, DC, 1995; pp 1-23.

1629Langmuir 2005, 21, 1629-1639

10.1021/la047807f CCC: $30.25 © 2005 American Chemical SocietyPublished on Web 01/14/2005

measure, predict, and understand the details of protein-surface interactions. As described by Norde et al.19 andSigal et al.,20 proteins typically adsorb strongly to hy-drophobic surfaces and weakly to neutral hydrophilicsurfaces. Charged surfaces are generally found to be moreadsorptive for oppositely charged proteins, and the degreeof adsorption is typically lower for similarly chargedsurfaces and proteins.1,19,20 While these trends are easilyconceptualized, the numerous simultaneous interactionsoccurring between the functional groups of proteins,material surfaces, and surrounding body fluids are verycomplex in nature and the actual submolecular-levelmechanisms and structural rearrangements involved inthese reactions are not well understood. As a result, thepostadsorption state of a protein (its conformation,orientation, and bioactivity) cannot currently be accuratelypredicted. Partly as a result of this problem, it has beenwidely accepted that nonspecific protein adsorption shouldbe prevented altogether and peptide ligands, such asarginine-glycine-aspartic acid (RGD), should instead beused in an attempt to control cellular response at thebiomaterials-biofluid interface. While we agree thatuncontrolled protein adsorption should be avoided, webelieve that great potential still exists for the use of surfacechemistry to control adsorbed protein orientation andconformation and to then use the natural bioactivity ofproteins to direct cellular response. This type of specificsurface design, however, will be possible only if submo-lecular-level protein-surface interactions can be quan-titatively understood and predicted.

Molecular simulation provides one of the most directmethods to theoretically investigate molecular behaviorin complex systems, such as in a system involving proteinadsorption. Because of the size of the molecular systemsinvolved, empirical force field based molecular modelingmethods, such as Monte Carlo (MC) and moleculardynamics (MD), are required for the simulation of thesetypes of processes.21 These methods employ a potentialenergy function (referred to as a force field) that calculatesthe overall potential energy of the system based on thesummation of individual atom-atom pair interactions.The force field equation takes into account the contribu-tions due to bonded interactions (bond stretching, bending,and torsion) as well as nonbonded interactions (van derWaals and electrostatic). These energy contributions aredetermined by a set of empirical parameters that are usedby the force field to calculate the energy contribution foreach type of interaction for the atom types that are definedin the molecular system being considered.21

To accurately simulate molecular behavior using anempirical force field, parameters of the force field mustbe balanced and tuned together to appropriately representthe behavior of the given molecular system being ad-dressed. As a consequence of this, a force field designedfor one type of application cannot be confidently appliedto other applications without separate validation; thisissue is known as force field transferability.21 For example,a problem of transferability was shown by Schuler et al.22

with the GROMOS96 force field, which was initially

parametrized to represent the behavior of proteins inaqueous solutions. When GROMOS96 was applied torepresent the behavior of condensed phase hydrocarbons,it was found to result in substantial errors and requirednew parameters to be developed for this particular typeof molecular system even though prior parametrizationincluded similar types of alkane functional groups, but ina different molecular setting.

Over the past few decades, several force fields havebeen developed and validated for the simulation of proteinsin an aqueous solution (e.g., AMBER,23 CHARMM,24

GROMACS,25,26 and OPLS27). However, none of these forcefields have been specifically parametrized with consid-eration of the adsorption behavior of proteins to syntheticmaterial surfaces. Protein-surface interactions are domi-nated by nonbonded interactions between the functionalgroups presented by amino acids, various functionalgroups presented by material surfaces (e.g., polymers),and water. For a force field to accurately represent thistype of system, it must be parametrized to properly balanceall of these types of interactions with one another. As afurther complication, however, when it comes to peptide-surface adsorption behavior, very little is known aboutthe actual molecular behavior that a properly param-etrized force field should reproduce. Therefore, before agiven empirical force field can be used confidently in anMC or MD simulation to accurately represent the adsorp-tion behavior of a protein to a surface, experimentalmethods must first be developed to allow validation of thesimulations. These experiments should probe a funda-mental characteristic of the adsorption behavior betweenspecific peptide residues and surface functional groupsfor a system that can also be readily represented bymolecular simulation. Simulation results with a givenforce field can then be compared with the experimentaldata in order to evaluate and validate or modify forcefields for specific molecular systems.

To directly address these needs, Vernekar and Latourpreviously developed an experimental method usingsurface plasmon resonance (SPR) spectroscopy28 to mea-sure the adsorption free energy of individual peptideresidues on functionalized alkanethiol self-assembledmonolayer (SAM) surfaces on gold using a host-guestpeptide system of the form GGGG-X-GGGG (G ) glycine,X ) variable residue). These methods were designed sothat the effects of single midchain residue substitutionson the adsorption behavior of this small peptide onfunctionalized surfaces could be measured using a systemthat was readily amenable to MD simulation. Thesemethods were applied to determine the free energy ofadsorption (∆Gads) for the peptides G4-G-G4 and G4-K-G4(G ) glycine, K ) lysine) on hydroxyl (OH) and carboxyl(COOH) functionalized SAM surfaces over a range oftemperatures from 288 to 310 K. From these studies, nopeptide adsorption (i.e., ∆Gads ) 0) was observed for either

(17) Sadana, A. Chem. Rev. 1992, 92, 1799-1818.(18) Claesson, P. M.; Blomberg, E.; Froberg, J. C.; Nylander, T.;

Arnebrant, T. Adv. Colloid Interface Sci. 1995, 57, 161-227.(19) Norde, W. In Adhesion and Adsorption of Polymers, Vol. 2;

Plenum Press: New York, 1980; pp 801-826.(20) Sigal, G. B.; Mrksich, M.; Whitesides, G. M. J. Am. Chem. Soc.

1998, 120, 3464-3473.(21) Leach, A. R. In Molecular modeling: Principles and applications;

Addison-Wesley Longman: Essex, U.K., 1996.(22) Schuler, L. D.; Daura, X.; van Gunsteren, W. F. J. Comput. Chem.

2001, 22, 1205-1218.

(23) Pearlman, D. A.; Case, D. A.; Caldwell, J. W.; Ross, W. R.;Cheatham, T. E.; DeBolt, S.; Ferguson, D.; Seibel, G.; Kollman, P.Comput. Phys. Commun. 1995, 91, 1-41.

(24) Brooks, B. R.; Bruccoleri, R. E.; Olafson, B. D.; States, D. J.;Swaminathan, S.; Karplus, M. J. Comput. Chem. 1983, 4, 187-217.

(25) Berendsen, H. J. C.; van der Spoel, D.; van Drunen, R. Comput.Phys. Commun. 1995, 91, 43-56.

(26) van der Spoel, D.; van Buuren, A. R.; Apol, E.; Meulenhoff, P.J.; Tieleman, D. P.; Sijbers, A. L. T. M.; Hess, B.; Feenstra, K. A.; Lindahl,E.; van Drunen, R.; Berendsen, H. J. C. In GROMACS User Manual,version 3.1; University of Groningen: Groningen, The Netherlands,2003.

(27) Kaminski, G. A.; Friesner, R. A.; Tirado-Rives, J.; Jorgensen,W. L. J. Phys. Chem. B 2001, 105, 6474-6487.

(28) Vernekar, V. N.; Latour, R. A., Jr. Mater. Res. Innovations, inpress.

1630 Langmuir, Vol. 21, No. 4, 2005 Raut et al.

peptide on the OH SAM surface or for the G4-G-G4 peptideon the COOH SAM surface, while a strong adsorptionresponse, e.g., ∆Gads ) -7.0 ( 0.1 kcal/mol (mean ( stddeviation) at 300 K, was measured for the G4-K-G4 peptideon the COOH SAM surface.

The objective of our present research was therefore todevelop and assess MD simulation methods that could beused to calculate the adsorption free energy for these host-guest peptide-SAM systems for comparison with theseexperimental results, using GROMACS as an exemplaryprotein simulation force field. Peptide adsorption to anoligoethylene oxide (OEG) functionalized SAM surface wasalso simulated due to the expected nonadsorption behaviorfor this type of surface.29,30

The results of these simulations showed distinct dif-ferences in peptide adsorption behavior as a function ofthe guest peptide and surface functional group combina-tions,with thepredicted trends being in general agreementwith the experimental results of Vernekar and Latour.28

Several areas were also identified, however, where thesimulation results show what is considered to be unre-alistic adsorption behavior, which demonstrates the needfor force field parameter modification if GROMACS is tobe used to accurately simulate peptide or protein adsorp-tion to these types of surfaces.

2. Materials2.1. Computational Environment. The initial peptide

models as well as the functionalized SAM surface models weredesigned using the InsightII [Accelrys Inc., San Diego, CA] andWeb Lab Viewer Pro (version 3.20) software [Accelrys Inc.].Molecular dynamics simulations were performed on a 26-nodedual-processor 2.66 GHz Linux cluster with GROMACS software(version 3.1.3).26 Each simulation was run on a dedicated node,with a calculation time of about 1.5 days of CPU time pernanosecond of simulation.

2.2. Peptide Models. The peptide residues were designed aszwitterionic host-guest peptide models with a peptide sequenceof GGGG-X-GGGG, where G ) glycine and X represents anyresidue type. This peptide design enables the middle peptideresidue X (the guest residue) to impart specific characteristicsto the peptide. Two types of guest peptide residues were modeled

to compare with our prior experimental studies: X ) lysine (K)and glycine (G). The structures of these peptides are shown inFigure 1. The glycine residue, with its simple -H side-group, isa relatively polar residue with a hydropathy index of -0.4,31

while the side-group of lysine {-[CH2]4-NH3+} is positively

charged at a pH of 7.4.2.3. SAM Surfaces. Alkanethiol SAM surfaces on gold [Au-

S(CH2)n-X] with X ) OH, COOH, and (O-CH2-CH2)2-OH)(OEG) functionalities were selected for the model surfaces forthis study. The OH and COOH SAMs were selected to correspondto the surfaces used in the prior experimental studies by Vernekarand Latour.28 The OEG surface was used because of its expectednonadsorptive behavior for peptides and proteins.29,30 Two typesof OEG structures were evaluated: an all-trans structure, whichrepresents the minimum energy conformation of the OEGfunctional group in vacuum or air32 and a helical conformation,which is believed to represent the minimum energy structure ofOEG in aqueous solution.29,30 Figure 2 shows molecular modelsof the structure of each type of alkanethiol chain used to constructthe SAM surfaces. The lower portion of each chain has beentruncated to provide a surface of approximately 10 Å thickness;

(29) Kreuzer, H. J.; Wang, R. L. C.; Grunze, M. J. Am. Chem. Soc.2003, 125, 8384-8389.

(30) Wang, R. L. C.; Kreuzer, H. J.; Grunze, M.; Pertisn, A. J. Phys.Chem. Chem. Phys. 2002, 2, 1721-1727.

(31) Kyte, J.; Doolittle, R. F. J. Mol. Biol. 1982, 157, 105-132.(32) Rigby, D.; Sun, H.; Eichinger, B. E. Polym. Int. 1997, 44, 311-

330.

Figure 1. Molecular model of host-guest peptide design in the form of GGGG-X-GGGG. Two types of guest residues wereevaluated: X ) lysine and glycine.

Figure 2. Molecular structures of the functionalized alkanechains used to construct alkanethiol SAM surfaces: (A) OHfunctionalization, (B) COOH functionalization, (C) COO-

functionalization, (D) trans OEG functionalization, and (E)helical OEG functionalization (torsional angles set as follows:C-O-C-C ) 70°, C-C-O-C ) 70°, O-C-C-O ) 60° (refs30 and 32)).

Simulations of Peptide-Surface Interactions Langmuir, Vol. 21, No. 4, 2005 1631

this represents the SAM structure effectively while reducing thesize of the system to be simulated.

SAM surface models of each type were developed by replicatinga single truncated alkanethiol chain in a hexagonal close-packed(x3 × x3)R30° array with 4.97 Å lattice spacing to form a 36 Å× 40 Å surface plane, with chain orientation set according to thewell-established alkanethiol chain structure on a gold (111)plane33 that has been used in several of our prior SAM models.34-38

The COOH SAM surface was constructed with both COOH(protonated) and COO- (deprotonated) functional groups. Oneout of twenty (5%) of the COOH groups were deprotonated andevenly positioned throughout the SAM surface, as appropriatefor a surface pKa of about 8.7 in a solution with pH of 7.4. Thissurface pKa was obtained from work by Creager and Clarke.39

Figure 3 shows molecular models of each of the four SAMsurfaces: OH, COO-/COOH, trans OEG, and helical OEG.

The GROMACS library defines the residue topologies of allthe amino acids and some other atoms that are used to modellarger proteins. However, GROMACS does not define the residuetopologies of individual SAM chains. The topologies correspondingto the individual SAM chains therefore had to be defined in theresidue topology parameter (.rtp) file of the GROMACS standardffgmx2 force field.26 Table 1 summarizes the atom types used todefine these molecules. For consistency with the GROMACSmodel, all aliphatic hydrogens were treated as united atoms,

while the hydrogens associated with the functionalized surfacegroups were separately defined by modifying the hydrogendatabase (.hdb) file of the GROMACS standard ffgmx2 force field.

The partial charges assigned to each atom of these SAM surfaceresidues were based on residue topologies for the side-groups ofamino acids defined in GROMACS with corresponding functionalgroups. However, the partial charges associated with the OEGfunctional groups do not have corresponding predefined func-tionality in GROMACS. In order maintain self-consistency withthe GROMACS force field as much as possible, the partial chargedistribution on the OEG SAM chain was therefore borrowed fromthe partial charges for this functional group from another classI protein force field (AMBER23) as determined using the InsightIIsoftware program [Accelrys Inc.]. Table 2 summarizes the partialcharge distribution on the terminal functional groups of eachtype of SAM surface. As per GROMACS design rules, the unitedcarbon atoms of the SAM surface (i.e., the CH2 groups of thealkane chains of the SAM surface that were not part of the OEGsegments) were represented as having zero partial charge.

3. Methods

3.1. Assembly of Residue-Surface System. Beforeconducting the simulated adsorption studies, short MD

(33) Ulman, A.; Eilers, J. E.; Tillman, N. Langmuir 1989, 5, 1147-1152.

(34) Latour, R. A.; Hench, L. L. Biomaterials 2002, 23, 4633-4648.(35) Latour, R. A. Curr. Opin. Solid State Mater. Sci. 1999, 3, 413-

417.(36) Basalyga, D. M.; Latour, R. A. J. Biomed. Mater. Res. 2003, 64A,

120-130.(37) Latour, R. A.; Rini, C. J. J. Biomed. Mater. Res. 2002, 60, 564-

577(38) Wilson, K.; Stuart, S. J.; Garcia, A.; Latour, R. A. J. Biomed.

Mater. Res. 2004, 69A, 686-698(39) Creager, S. E.; Clarke, J. Langmuir 1994, 10, 3675-3683.

Figure 3. Molecular models of alkanethiol SAM surfaces: (A) hydroxyl (-OH) terminated SAM, (B) COO-/COOH terminatedSAM, (C) all-trans OEG terminated SAM, and (D) helical OEG terminated SAM. This structure is produced by repeating theoriented hydroxyl terminated alkanethiol chain in a hexagonal close-packed (x3 × x3)R30° array to form a surface of requireddimensions.

Table 1. GROMACS Atoms Types Employed to DefineNew Surface Topologies

atom name description mass (amu)

O carbonyl oxygen (CdO) 15.9994OM carboxyl oxygen (CO- and C-O-C) 15.9994OA hydroxyl oxygen (OH) 15.9994C bare carbon (CdO, C-N) 12.0110C1 aliphatic CH group 12.0110C2 aliphatic CH2 group 12.0110C3 aliphatic CH3 group 12.0110HO hydroxyl hydrogen 1.0080

Table 2. Charge Distribution on the Terminal SurfaceFunctionalities

terminalgroup (X)

partial chargesbased on:

atomtype

partial chargein |e|

chargegroup

OH serine (SER) C2 0.150 0OA -0.548 0HO 0.398 0

COO- aspartic acid C2 0.270 0(unprotonated) OM -0.635 0(ASP) OM -0.635 0

COOH aspartic acid C2 0.530 0protonated) O -0.380 0(ASPH) OA -0.548 0

HO 0.398 0

OEG C2 0.200 1OM -0.400 1C2 0.200 1C2 0.200 2OM -0.400 2C2 0.200 2C2 0.263 3OA -0.566 3HO 0.303 3


simulations were performed on each respective peptidein vacuum (300 K, 100 ps, 1 fs time step) in order togenerate a random configuration. The randomly struc-tured peptides were then placed over the alkanethiol SAMsurface models with the distance between the CR atom ofthe guest peptide and the plane of the terminal groups ofthe SAM surface (defined as the surface separationdistance, SSD) initially set to 7.5 Å. Three-dimensionalperiodic boundary conditions were defined, with the sizeof the periodic cell being 36 Å × 40 Å in the plane of thesurface and 65 Å perpendicular to the surface. The systemwas then solvated with simple point charge (SPC) watermolecules with Na+ and Cl- ions added to approximatea 0.150 M physiologic saline solution40 and to maintainoverall system neutrality. Water molecules falling withina 2 Å radius from the center of heavy atoms (i.e., non-hydrogen atoms) were then deleted to establish the finalmolecular assembly, with the resulting models containingapproximately 8900 atoms. Figure 4 illustrates the initialassembly of the complete peptide-surface system for aG4-K-G4 peptide over an OH SAM surface.

3.2. Simulation Specifications. Once created, eachmolecular system was then energy-minimized withoutconstraints using the steepest descent integrator for 5000steps with the initial step size of 0.1 Å (the minimizationtolerance was set to 1000 kJ/(mol nm)).

Molecular dynamics simulations were then performedon this energy-minimized system using the leap-frogalgorithm in the NVT (canonical) ensemble. All bondlengths were constrained, which enabled a time step of 2fs to be used for the MD simulations.26 These conditionswere established based on preliminary studies that wereconducted to compare the trajectories for constrained andunconstrained simulations with the time step set at 0.5,1, and 2 fs. Similar peptide-surface interaction behaviorwas observed for each of these conditions, thus enablingbond length constraints and a 2 fs time step to be usedconfidently. Temperature coupling with a Berendsenthermostat was implemented with a time constant of 0.1ps at a temperature of 300 K. No pressure coupling was

applied to the system. Initial velocities were generatedaccording to the Maxwell distribution at 300 K. The cutoffdistances used for both the Columbic and the van derWaals interactions were 17 Å. All simulations were runfor 10 ns of simulated time, and the system configuration(atom coordinates, velocities, and energies) was savedevery 1.0 ps. In all, eight different molecular systems wereevaluated (G4-G-G4 and G4-K-G4 peptides; OH, COOH,trans OEG, and helical OEG SAM surfaces), with threeindependent simulations conducted for each system usingdifferent initial peptide orientations over the SAM surfaceand different initial velocities. The resulting trajectoryfileswere thenviewedandanalyzedusingVMDsoftware.41

3.3. Data Analysis. The trajectories for each molecularsystem were analyzed to calculate the adsorption freeenergy. This was accomplished using the probability ratiomethod.42,43 For this analysis, the GROMACS tool “g-traj”was used to track SSD as a function of time, which wasthen transformed into a probability density distributionwith respect to SSD. The probability density distributionswere then analyzed to calculate the relative free energyas a function of SSD, from which the overall adsorptionfree energy for the system was calculated.

The probability density distribution for each peptide-SAM surface system was determined by first dividing theSSD range of the simulation into 0.2 Å intervals (∆xi).The frequency of a given peptide being in a particularinterval of SSD (SSDi) was then calculated as the numberof times (Fi) that the peptide was positioned in a givenSSDi interval during the 10 ns simulation. The positionalprobability (Ai) of the peptide being in a given SSDi locationwas then calculated from these data as

Ai was then divided by the width of the interval, ∆xi,to calculate the normalized probability density (Pi) foreach SSDi, or

The normalized probability density (NPD) plot over theSSD range for each MD simulation was then used tocalculate the adsorption free energy as a function of SSD.Once the NPD values were determined, the relative freeenergy difference (∆Gi) between two SSDi positions wascalculated as

where R is the ideal gas constant constant,T is the absolutetemperature, and Pi and P0 represent the probabilitydensity of the peptide being at SSDi and SSD0, respectively,where SSD0 represents a defined reference state for thegiven molecular system.

Finally, the overall free energy of adsorption (∆Gads) forthe system was calculated as the weighted sum of therelative free energies, or

(40) West, J. B. In Best and Taylor’s Physiological Basis of MedicalPractice, 11th ed.; Williams and Wilkins: Baltimore, MD, 1985; pp441-442.

(41) Humphrey, W.; Dalke, A.; Schulten, K. J. Mol. Graphics 1996,14, 33-38.

(42) Mezei, M. Mol. Simul. 1989, 3, 301-313.(43) Gunsteren, W. F. V.; Weiner, P. K.; Computer Simulation of

Biomolecular Systems: Theoretical and Experimental Applications(Volume 1); ESCOM Science: Leiden, The Netherlands, 1989.

Figure 4. Typical molecular model constructed to depict aG4-K-G4 peptide over an OH SAM surface with explicitlyrepresented water.

Ai )Fi

∑Fi

with ∑(Ai) ) 1 (1)

Pi )Ai

∆xiwith ∑(Pi ∆xi) ) 1 (2)

∆Gi ) Gi - G0 ) -RT ln[Pi

P0] (3)


Following the above-described methods, plots of SSDvs time, NPD vs SSD, and ∆Gi vs SSD were generated foreach of three independent 10 ns simulations for eachpeptide-SAM surface system; the ∆Gi vs SSD plots werethen integrated to calculate ∆Gads for each run, from whichwere obtained the mean and standard deviation of ∆Gadsfor each system. These values were then compared withthe experimentally determined values of ∆Gads to assessthe performance of the GROMACS force field to representthis type of molecular system. In addition, the trajectorydata were graphically viewed to evaluate the specificfunctional group interactions predicted to occur betweenthe peptides and the SAM surfaces that were responsiblefor the observed adsorption behavior.

4. Results4.1. Peptide Adsorption on the OH SAM Surface.

In the initial simulations with the OH SAM, the lowertwo carbon atoms of the SAM surface were fixed in orderto maintain the orientation of each of its chains whileenabling the rest of each chain to freely move in responseto its environment. However, as shown in Figure 5, thisresulted in the peptides adsorbing strongly to the surface,which should not occur based on our previous experimentalresults.28 Upon inspection of the functional group interac-tions responsible for this behavior, it was observed thatthe adsorption response was largely due to hydrophobicinteractions between CH2 segments of the peptides andthe CH2 groups of the SAM that were exposed when theOH groups on the SAM surface separated due to the motionof the SAM chains (see Figure 6). To prevent this behavior,the coordinate positions of all atoms of the SAM werethen fixed except for the surface OH groups, which solvedthis problem, and the peptides no longer adsorbed to thissurface (Figure 5). Subsequent simulations for all of theSAM surfaces were conducted in this same manner, withthe atoms all fixed except for the top functional groups,which were free to respond to the surrounding atoms.This highlights one shortcoming of the GROMACSpotential for SAM surfaces, in that the parameters for theunited-atom CH2 groups result in overly mobile chains atthe correct packing densities, which also generates incor-rect surface heights and tilt angles without the impositionof additional constraints.

Representative data plots for SSD vs time, the NPDdistribution of SSD, and ∆Gi as a function of SSD arepresented in Figure 7 for both the G4-G-G4 and G4-K-G4peptides on the constrained OH SAM surface. Similardata analyses were applied for each of the three inde-pendent simulations for each of these peptides on the OHSAM surface, but only one exemplary set of data plots isshown for conciseness. In producing the NPD plots, it wasnecessary to discard the SSD vs time data for SSD > 35Å because the use of the 3-D periodic boundary conditionscaused the peptides to interact with the image of thebottom of the SAM surface (with its fixed CH3 functional-ity) when the peptide moved within the cutoff distance ofthe top of the simulation cell. To plot ∆Gi as a functionof SSD, a reference state (SSD0) had to be defined to providethe value of P0 for the calculation of the free energy ateach SSDi position relative to SSD0 (see eq 3). Afterinspecting the NPD vs SSD plots, it was decided to usethe average value of Pi between the SSD of 25 and 35 Åfor P0 because this represents a region in the simulationwhere the peptide is beyond the cutoff distance fornonbonded interactions with the SAM surface at both the

bottom and top of the simulation cell. This assumption isequivalent to setting the zero of the free energy at an SSDof 30 Å. Based on these analyses, the adsorption freeenergies (mean ( std deviation) for the G4-G-G4 and G4-K-G4 peptides on the OH SAM surface were calculated tobe 0.12 ( 0.08 and -0.09 ( 0.39 kcal/mol, respectively.

As shown from the plots presented in Figure 7 and bythe near zero value of adsorption free energy, the peptidesdid not adsorb on the OH SAM surfaces, indicating thatwater molecules formed more thermodynamically favor-able interactions with the SAM surface and peptidefunctional groups than the SAM and the peptides did witheach other. This is in agreement with the experimentalstudies of Vernekar and Latour, which also indicate that

∆Gads ) ∫SSDiPi ∆Gi dxi ≈ ∑[Pi ∆Gi ∆xi] (4)

Figure 5. SSD vs time trajectory plot for the unconstrainedOH SAM surface compared to the constrained OH SAMsurface: (A) G4-G-G4 peptide and (B) G4-K-G4 peptide.

Figure 6. A snapshot from the trajectory file for the G4-K-G4peptide on the unconstrained OH SAM surface showing theCH2 segment of a glycine residue (*) hydrophobically interactingwith subsurface CH2 groups of the SAM that were exposedwhen the surface OH groups separated. Hydrogen bonds formedby the NH and CO of this glycine residue with the OH groupsof the SAM surface also helped stabilize this interaction.


neither the G4-G-G4 nor the G4-K-G4 peptide adsorbs toan OH SAM.28 From viewing the trajectory files, it wasobserved that each of these peptides translated away fromthe surface in an apparently random manner after a fewinitial interactions with the SAM surface in the form oftemporary hydrogen bonds formed between the variouscharged and polar functional groups of the peptide andthe OH functional groups of the SAM. These interactions,however, were readily displaced by the surrounding SPCwater molecules, which formed more stable hydrogenbonds with the OH functional groups on the SAM surface.Asaresult of thesemore favorable interactions,a relativelystable water layer was maintained over the SAM surfacewith minimal interactions between the surface and thepeptide functional groups. A representative configurationfrom the simulation of G4-G-G4 over the OH SAM surfaceis shown in Figure 8.

4.2. Peptide Adsorption on the COOH SAM Sur-face. Figure 9 shows the SSD versus time, NPD vs SSD,and ∆Gi vs SSD plots for the G4-G-G4 and G4-K-G4 peptidesover the COOH SAM surface. As clearly shown, and unlikethe OH SAM surface, both peptides adsorbed to the COOHSAM surface, with the positively charged G4-K-G4 peptideadsorbing most strongly. As shown in the SSD vs timetrajectory plot, the SSD value for each system wasmaintained at about 3.5 Å for most of the simulation,with further approach being prevented by van der Waalsforces of repulsion. As a result of the strong interactionsbetween the peptides and this surface, and as shown inthe NPD vs SSD plots (Figure 9), the 10 ns simulationsdid not provide adequate time to enable the peptide tosample the full range of SSD values between 0 and 35 Åover the SAM surface. Consequently, the probabilitydensity could not be calculated accurately in the un-sampled areas. Thus, the probability distribution doesnot properly account for the relative probability distribu-

tion of the peptide for the whole range of SSD, and thevalue of P0 that is needed for calculating ∆Gi could not bedetermined in the same manner as for the OH SAMsurface. To account for this problem, P0 was defined asthe NPD value of the SSDi location furthest from the SAM

Figure 7. SSD vs time, NPD (Pi) vs SSD, and ∆Gi vs SSD for each of the peptides on an OH SAM: (A) G4-G-G4 peptide and (B)G4-K-G4 peptide.

Figure 8. A snapshot from the trajectory file of the G4-G-G4peptide over the OH SAM surface. The sodium ions (darkspheres) and chloride ions (light spheres) are shown with aspace-filling model, while the water molecules are not shownfor the sake of clarity. In this system, the OH functional groupson the SAM surface interacted more strongly with the SPCwater molecules than with the peptide, thus resulting innegligible peptide adsorption to the SAM surface.


surface, and this value was then used to calculate andplot the normalized free energy as a function of SSD, asshown in Figure 9, with the resulting values of theadsorption free energy for the G4-G-G4 and G4-K-G4peptides on the COOH SAM surface calculated as -2.10( 1.61 and -3.75 ( 0.65 kcal/mol (mean ( std deviation),respectively. It must be realized, however, that the trueNPD ratios that would have been obtained if the systemhad been adequately sampled (i.e., with P0 representingthe actual average NPD value for SSD between 25 and 35Å) would reflect larger values of Pi /P0 and subsequentlymore negative values of ∆Gi and ∆Gads per eqs 3 and 4.Thus, from these simulations, it is more appropriate toexpress the adsorption free energy values for G4-G-G4 andG4-K-G4 peptides on the COOH SAM surface as an upperbound, with ∆Gads < -2.10 ( 1.61 kcal/mol and ∆Gads <-3.75 ( 0.65 kcal/mol, respectively.

From viewing the trajectory files for these peptide-SAM systems, it was observed that this adsorptionbehavior was a result of several contributions fromdifferent types of nonbonded interactions between thefunctional groups of the peptides and the SAM surface.The dominant effect appeared to be the electrostaticattractions between the positively charged NH3

+ func-tional groups of the peptide (lysine side-chain for the G4-K-G4 peptides and the N-terminus of both peptides) andthe COO- functional group of the surface. These interac-tions, however, always occurred with two or three inter-vening water molecules between the NH3

+ and COO-

functional groups. The NH3+ functional groups of the

peptide were therefore not tightly held to the surface butunderwent a fair amount of motion over the surface whilemaintaining their position in general proximity to thecharged COO- functional groups of the SAM surface.Electrostatic repulsion was also generally observed be-tween the C-terminus of the peptides and the surface,

with the C-terminus generally being positioned furthestaway from the surface while the peptide was adsorbed. Inaddition to electrostatic effects, relatively stable hydrogenbonding interactions were also observed to occur betweenboth charged-polar and polar-polar functional groupcombinations between the peptide and the SAM surface.Unexpectedly, hydrophobic interactions were also fre-quently observed between the CH2 groups in the peptide(from both the glycine and lysine side-groups) and theCH2 groups immediately below the COO-/COOH groupson the SAM surface that were exposed when the COO-/COOH surface groups occasionally separated during thesimulation despite the fact that all other atoms of theSAM were fixed in position. Thus, the strong adsorptionof these peptides to the COOH SAM surface was due toa combined result of electrostatic, hydrogen bonding, andhydrophobic interactions.

The disagreement with experiment,28 along with theunexpected occurrence of hydrophobic contacts, indicatesthat perhaps the GROMACS force field overestimates thestrength of hydrophobic interactions. Figure 10 presentsa representative configuration from the simulation of theG4-K-G4 on the COOH SAM surface that illustrates eachof these types of interactions.

4.3. Peptide Adsorption on the OEG SAM Surfaces.Simulations of each peptide were performed on two typesof OEG SAM surfaces: trans and helical. It was readilyapparent from the simulations that the trans conformationwas much more stable than the helical one, with the transOEG segments retaining their conformation throughoutthe 10 ns simulations, while the helical OEG segmentsshifted into a trans conformation within about the first100 ps of the simulation and then maintained the transconformation for the remaining 9.9 ns. Because of this,the trajectory data were essentially similar for both typesof starting conditions, and only the trans OEG results

Figure 9. SSD vs time, NPD (Pi) vs SSD, and ∆Gi vs SSD for both peptides on the COOH SAM: (A) G4-G-G4 peptide and (B) G4-K-G4peptide.


will therefore be presented. Accordingly, Figure 11 showsthe SSD vs time, NPD vs SSD, and ∆Gi vs SSD plots forG4-G-G4 and G4-K-G4 over the trans OEG SAM surface.While it was anticipated that this surface would be verynonadsorptive like the OH SAM, as shown in Figure 11,surprisingly both peptides interacted more strongly withthis surface than either one did with the OH SAM surface,although less strongly than with the COOH SAM surface.While the G4-G-G4 peptide did sample the entire range ofSSD during the simulation, the G4-K-G4 did not sampleSSD values greater than 10 Å, remaining closely adsorbedto the surfaces during the whole 10 ns of the simulation.Thus similar methods had to be applied to calculate the

∆Gi and ∆Gads as were used for the COOH SAM surface,with these values again under-representing the absolutemagnitude of the free energy values. Accordingly, fromthe data generated by the simulations, the adsorptionfree energy values (mean ( std deviation) for the G4-G-G4and G4-K-G4 peptides on the OEG SAM surface werecalculated to be -1.37 ( 2.03 and < -2.73 ( 1.54 kcal/mol, respectively.

An inspection of the trajectory files for the adsorptionof these peptides on the OEG surface showed that whilethe interactions involved hydrogen bonding between thecharged and polar functional groups of the peptides withthe OH functional groups of the OEG SAM, by far the

Figure 10. A snapshot from the simulation of the G4-K-G4 peptide over the COOH SAM surface. The single atoms are sodiumions (dark spheres) and chloride ions (light spheres) contained in the solvent, with the water molecules not shown for clarity. Inthis system, the COO-/COOH functional groups on the SAM surface interacted with the peptide through a combination of electrostatic,hydrogen bond, and hydrophobic interactions resulting in strong adsorption of the peptide to the SAM surface.

Figure 11. SSD vs time, NPD (Pi) vs SSD, and ∆Gi vs SSD for both peptides on the OEG SAM: (A) G4-G-G4 peptide and (B) G4-K-G4peptide.


dominant cause of adsorption was due to hydrophobicinteraction between CH2 groups of the peptides and thetopmost CH2-CH2 segments of the OEG chain that wereexposed when the top OH functional groups of the OEGSAM separated during the simulation. It was also observedthat the water molecules did not penetrate the intermo-lecular spacing between OEG segments during thesimulation to form polar bonds with the inner oxygenspresent in the OEG segments, which has been proposedas the mechanism that stabilizes the experimentallydetermined helical conformation of the OEG chains.29,30

Figure 12 shows a representative configuration fromsimulation of the G4-K-G4 peptide over the OEG surfaceto illustrate the typical types of bonding that occurredbetween the peptides and the OEG surface.

5. Discussion

The results of these simulations show areas where thesimulated behavior is both in agreement and in conflictwith experimental results. As presented above, to obtaina satisfactory simulation of peptide behavior over the OHSAM surface, it was necessary to constrain all the atomsof the SAM except for the surface OH group. Once thiswas done, the simulated behavior was in excellentagreement with the experimental results of Vernekar andLatour,28 with negligible interactions between the peptideand the surface and a negligible adsorption free energy.However, unexpected hydrophobic interactions betweenthe peptides and the subsurface CH2 groups of the SAMstill occurred with the COOH SAM surface even when itwas constrained. These interactions contributed to therelatively strong adsorption of the G4-G-G4 peptide to thissurface, which was contrary to the prior experimentalstudies, that found no adsorption for this system. Peptideadsorption was also unexpectedly observed for the OEGsurfaces, again largely due to hydrophobic interactions.

These observations underscore the problem of force fieldtransferability. The GROMACS force field was primarilyparametrized for the simulation of peptides and proteinsin aqueous solution and was not specifically designed tomodel the behavior of alkanes when arranged to form aSAM structure or OEG chain segments. This is also

reflected by the fact that new GROMACS residue-typedefinitions had to be defined in order to represent themolecules needed to construct the SAM surface. Theparameters for these molecular segments were definedbased on the parameter set for similar-type functionalgroups present in the GROMACS library for amino acidsor, for the case of OEG, another force field (i.e., AMBER).It is apparent that the resulting parametrization satis-factorily represents neither the behavior of the SAMsurface nor the adsorption of peptides to these SAMsurfaces. Further modification and verification work istherefore necessary before this force field can be confi-dently used to simulate peptide and protein adsorptionbehavior on these types of surfaces.

A second problem with the presented simulationmethods was clearly demonstrated with the G4-K-G4

peptide over the COOH SAM, in which the system clearlywas not sufficiently sampled in order to accuratelycalculate adsorption free energy. Referring to eqs 3 and4, to represent the experimentally measured adsorptionfree energy of -7.0 kcal/mol at 300 K, the probability ratioof the likelihood of the peptide being adsorbed to thesurface vs being far away from the surface in a nonad-sorbed state (i.e., Pi/P0) would have to average over100 000. However, a 10 ns simulation with data savedevery 1 ps only provides 10 000 data points. Thus it is notsurprising that the simulations did not sample SSDpositions far from the surface in the strongly boundsystems. To solve this problem, either much longersimulation times are needed (on the order of micro-seconds) or biased sampling methods, such as umbrellasampling,21,44-47 must be applied. We are currentlydeveloping the latter approach to address this issue forfuture simulations.

(44) Ryckaert, J. P.; Bellemans, A. Mol. Dyn. Liq. Alkanes, FaradayDiscuss. 1978, 20, 95-106.

(45) Torrie, G. M.; Valleau, J. P. J. Comput. Phys. 1977, 23, 187-199.

(46) Jorgensen, W. L.; Gao, J.; Ravimohan, C. J. Phys. Chem. 1985,89, 3470-3473

(47) Jorgensen, W. L.; Buckner, J. K. J. Phys. Chem. 1987, 91, 6083-6085.

Figure 12. A snapshot from the simulation of G4-K-G4 over the OEG SAM surface. Note that the OEG chains are generallystructured in a trans conformation. The peptide is strongly adsorbed to the OEG surface via a combination of hydrogen bonds withthe OH groups of the SAM surface and hydrophobic interactions between CH2 groups of the peptide and the CH2-CH2 segmentsof the OEG chain that became exposed by the separation of the surface OH groups during the simulation.


6. ConclusionsAs stated, the objectives of this research were to develop

and assess methods to simulate the adsorption of a host-guest peptide on functionalized alkanethiol SAM surfaceson gold using a selected molecular mechanics force fieldand to calculate the free energy of peptide-surfaceadsorption from the trajectory data for comparison withexperimentally determined adsorption free energies. TheGROMACS molecular simulation program and force fieldwere selected for our initial studies, and our results haveidentified several areas in need of further development,involving both the simulation methods and the param-etrization of GROMACS for this specific application.

These simulation results do not allow us to make firmstatements regarding whether the discrepancies betweensimulation and experiment are due to problems with theparametrization of the SAM surfaces, the peptide-SAMsurface interactions, or both. However, we can concludethat the parameter set used in these simulations did notprovide a satisfactory representation of the peptide-SAMmolecular system. Additional parameter fitting is requiredto accurately represent SAM surfaces in GROMACS,rather than transferring parameters from other parts ofGROMACS or other force fields. The alkane-alkaneinteractions will need to produce a SAM structure thatcan be simulated dynamically, without requiring un-physical constraints. It is also possible that new param-eters, or parameter scaling factors, will also be needed toadjust interactions between the peptides and the SAM atthe solid-solution interface in order to obtain the rightbalance of hydrophobic and hydrophilic interactions.Interactions at this interface are distinct from those within

the SAM surface and the aqueous peptide solutionthemselves. Due to physical effects such as differences ineffective polarizability between the bulk and interfacialregions, the interfacial interactions do not necessarilyfollow directly from using standard combining rules withclassical, nonpolarizable potentials.

Finally, we also conclude from our studies that biasedsampling methods, such as umbrella sampling, are neededin order to efficiently calculate the free energy of peptideadsorption. This is due to the very low probability ofsampling the nonadsorbed state for strongly adsorbingpeptide-SAM systems.

Once these issues are resolved, the developed simulationmethods for peptide adsorption to SAM surfaces, coupledwith the previously developed SPR experimental methods,should provide a very effective approach for the evaluation,modification, and validation of existing force fields for thesimulation of peptide and protein adsorption behavior tofunctionalized surfaces. Once validated, a protein adsorp-tion force field should have great potential for use as atool for surface design to predict and control adsorbedprotein orientation, conformation, and bioactivity.

Acknowledgment. We thank the National ScienceFoundation (NSF) (Award Number EPS-0296165), theState of South Carolina, and Clemson University forproviding the funding support for this project. We alsothank the NSF Center for Advanced Engineering Fibersand Films (CAEFF), Ms. Corey Ferrier, and Mr. TimShelling at Clemson University for computational re-sources and computer system support.

LA047807F


Molecular Dynamics Simulations of Peptide−Surface Interactions

Documents