Professional Documents
Culture Documents
Pim Schravendijk, Luca M. Ghiringhelli, Luigi Delle Site, and Nico F.A. van der Vegt*
Max-Planck-Institute for Polymer Research, Ackermannweg 10, D 55128 Mainz, Germany
ReceiVed: August 28, 2006; In Final Form: NoVember 7, 2006
We present a multiscale modeling procedure that offers the opportunity to study hydrated biomolecules at
metal surfaces. First principle DFT calculations and classical atomistic simulations are used interactively in
order to account for both quantum and statistical aspects of molecular conformations at the surface. We
present models for water, benzene, phenol, alanine, and phenylalanine at the (111) surface of nickel. These
models are subsequently used in classical atomistic simulations to study physical-chemical aspects of amino
acids at a Ni(111)/water interface. Application of this method to a larger set of “molecular building blocks”
opens a computational route for molecular engineering of bio/inorganic hybrid systems.
in ref 6, will be discussed more extensively in Section III as with a coupling constant of 1.0 ps) (300 K with a coupling
well. In Section IV, the newly developed models are used to constant of 0.5 ps), using Nose-Hoover temperature coupling16
study surface interactions under aqueous conditions at 300 K. and Parrinello-Rahman pressure coupling,17 but additionally
introducing the molecules described above and using the LINCS
II. Computational Details bond constraint algorithm.18 All simulations were performed
with a molecular dynamics (MD) integration time step of 2 fs.
We extend the “molecular building block” approach that has The solute-surface and solvent-surface potentials were treated
already been applied successfully to organic-inorganic inter- independent from the solute-solute, solute-solvent, and solvent-
faces.6,7 In this approach, macromolecule-surface interactions solvent intermolecular potentials, which were described by the
are described at two levels. First, building-block molecules are GROMOS 43a1 force field.19 As a solvent, the SPC/E water
chosen that describe the recurring parts in the macromolecule. model was used.20 The natural question arising at this point is
The small sizes of these building blocks allow for quantum if the use of a specific force field for water causes model
mechanical calculations of the building block-surface interac- dependencies. We address this question below and show that
tions, and the resulting data are used for parameterizing atomistic this is not the case. In addition to MD simulations, several
solute-surface interaction potentials. On the other, classical Langevin dynamics (LD) simulations (300 K) were performed
atomistic, level, this quantum-based parameterization describes in which the solute interacts with the metal surface in a vacuum
building block-surface interactions, not only for the single environment. The LD algorithm, available in Gromacs,15 was
building blocks but also for the structures such as (oligo)peptides used with a friction coefficient of 1 ps-1.
or proteins that can be constructed from combinations of these Solute-surface potentials of mean force (PMFs) in water
building blocks. In the current paper, we will model phenol and were obtained via constraint-biased simulation with force
the neutral forms of alanine, phenylalanine, and tyrosine and averaging21 as provided by the Gromacs package. The constraint
study their interaction with a Ni(111) surface from the bulk- direction was chosen perpendicular to the surface (i.e., in the z
hydrated state. The Ni(111) metal surface has been chosen direction). For each PMF, 142 defined constraint distances from
because its properties for similar systems have been well-tested the surface were chosen, in steps of 0.01 nm from 0.19 to 1.00
in previous works.6-11 Extension to other surfaces (e.g., Pt, Pd, nm, in 0.02 nm steps from 1.00 to 2.00 nm and in 0.10 nm
Au) is straightforward as will be discussed later on. As an ideal steps for distances up to 3.00 nm. In all cases, this proved to
model surface, any oxidation effects at the surface will be be far enough to reach a bulk-hydrated state of the solute
ignored. molecule as indicated by a constant value for the PMF at these
Quantum calculations of the amino acids alanine and phe- distances. Several constraint sites on the solute molecule were
nylalanine have been performed using the DFT based finite- tested, each separately, to see if any site dependence for the
electronic temperature method of Alavi et al.12 (FEMD), as PMF was present. The following interaction sites were chosen
implemented in the CPMD code.13 Note that amino acids under as constraint sites: for benzene and phenol the geometric center
physiological conditions exist mainly (>99%) in the zwitterionic of the ring; for alanine the carbonylic oxygen and the amine
state, whereas in the case of peptides the end groups are nitrogen; for phenylalanine the ring center, carbonylic oxygen,
condensed to form peptide bonds and are therefore of different and amine nitrogen. It is important that only the z component
chemical nature. The neutral amino acids used here are therefore of the constraint site was kept fixed. The constraint site could
not the exact representations of any of those two states but are move in the xy directions, and the remaining parts of the
still of interest because the neutral state (as well as the molecule could move in any direction as long as this movement
zwitterionic state) has been shown to bind to a Ni(111) surface.14 would not displace the z distance of the constraint site.
The zwitterionic state of amino acids as well as the peptide The starting conformations at the 142 distances mentioned
form are currently under investigation14 and will be a matter of above were generated by first solvating the molecule in the
following publications. middle of the slit, following by pulling it to an adsorbed state
We used the Gromacs 3.3 code,15 adapted to allow for any (at z ) 0.19 nm). From there, the z distance was increased in a
combination of atom-wall potentials divided over the molecule, sequence of short pull runs. Once all starting points were
thus enabling a convenient parameterization of both configu- generated, 3 ns production runs were performed for each z
rational and energetic data from quantum calculations. Surfaces distance, enabling the calculation of distance-dependent mean
are introduced by atom-wall potentials acting in the z direction forces and mean surface interaction energies.
at both z borders of the box, and an empty space larger than
any cutoff length is added to remove periodicity in the z III. Modeling Surface Interactions
direction. With surfaces at two sides of the box, the system The quantum-atomistic multiscale modeling of amino acid-
actually represents a slit; however, the distance between the metal surface interactions is more complex than our previous
surfaces was chosen large enough (>7 nm) to enable a layer of multiscale modeling of benzene/polycarbonate adsorption.6,7
bulk-like water in the center, several nanometers thick. The box This is largely due to the low symmetry and relatively large
size in the x and y direction extended 3.4 nm. Pressure coupling number of interaction sites in amino acids and the fact that these
and particle mesh Ewald (PME) treatment of electrostatic interaction sites are located too close to each other to be treated
interactions is performed by (semi) two-dimensional schemes separately. The low symmetry greatly affects the number of
that are part of the Gromacs 3.3 package. Pressure coupling quantum calculations necessary: even though alanine has fewer
was chosen to be coupled only to the box size in the x and y atoms than benzene, there are many more nonequivalent
directions, parallel to the surface, and the box size in the z geometries in which it can be oriented relative to the surface.
direction was set constant. The PME Fourier spacing was set A complete quantum analysis of all of these configurations at
at 0.12 nm, the real space and neighbor search cutoff was set at all possible metal lattice sites would require a very expensive
0.9 nm, and van der Waals interactions were cut off at 1.4 nm. computational effort. An extension to our multiscale modeling
Simulations were carried out in a system similar to the one procedure6 is introduced here to cope with the before-mentioned
described in the previous paper,6 consisting of 3000 water problems in a way that actually enhances the procedure’s general
molecules at constant pressure and temperature (NPT) (1 atm validity and applicability.
Hydrated Amino Acids at Metal Surfaces J. Phys. Chem. C, Vol. 111, No. 6, 2007 2633
We start by emphasizing that the attention lies on finding locally at the surface water conformations must consist of full
global minimum energy conformations because these will be or half tetrahedra (see Figure 1) because of the confinement of
the dominant contribution to the sampling statistics during long the surface. Among possible arrangements like the ones shown
runs at the atomistic level. Calculations on intermediate, higher- in Figure 1, some are not allowed according to data available
energy states would provide only superfluous information at from quantum calculations, that is, those conformation display-
the current modeling precision. Therefore, the starting point of ing hydrogen down-like structures. Thus, we have a first
the modeling will be a selection of initial configurations that screening of possible conformations that must be reproduced
aims to give us the extreme binding conformations. The by our modeling. The second important point is that, because
electronic properties of the metal binding of these configurations we want to explore many relevant configurations, we should
were calculated by a series of quantum density functional carry out many calculations. Because first principles quantum
calculations (geometry optimization),14 considering per config- calculations are computationally very demanding, a solution to
uration the various possible positions on the metal lattice. The this problem comes from the observation that the tetrahedral
data retrieved from these calculations give us the parameters structure of Figure 1 can be described well by different
(interaction sites, interaction strength, optimal distances), based combinations of substructures consisting of monomers, dimers,
on which an atomistic model is constructed. However, because and trimers in different conformations. As shown in Figure 1,
we are studying a molecule with a large number of orientational we then make what we would call “the first layer approxima-
degrees of freedom, the molecule is likely to get trapped in local tion’’; that is, only the molecule close to the surface and those
energy minima during quantum calculations. We resolve this directly bonded to that participate to the adsorption strength.
by performing LD simulations of the molecule in vacuum, The adsorption energy per water molecule in these allowed
applying the modeled molecule-surface interaction parameters, configurations is calculated using a quantum-based approach10
and sample its configuration space at a range of distances (see the caption of Figure 1 for the calculation used), and the
perpendicular to the surface, using the constraint method and results are used to parameterize a water-surface potential for
sites described in the Methodology section. This set of LD runs classical simulations. We have chosen an attractive 10-4
can sample a larger amount of the phase space than could be potential to describe the interaction between water oxygen and
done by quantum optimization runs, in a fraction of the time. the surface (see eq 1). No interaction was applied between water
If the LD simulation generates (only) the surface adsorbing hydrogen and the surface. See Table 1 for exact values of the
conformations that were already found in our initial quantum parameters used.
{
calculations, then our classical model is considered complete.
[52(σz ) - (σz ) ] z e z
10 4
If, however, the LD simulation generates adsorbing (low-energy) 2π10-4
conformations that were not previously found in quantum UAttr.10-4 ) cutoff
(1)
calculations, then these conformations are used as starting 0 z > zcutoff
configurations in a new series of quantum-based structure
optimizations. If this second set of quantum calculations results In the classical simulation of the metal-water interface, water-
in interaction properties that differ from the LD simulations, water interactions are accounted for by a classical force field,
then the atomistic modeling is adjusted. Then, the procedure is and therefore the energy of water-water hydrogen bonding
repeated (LD run with new modeling, quantum calculation of should not be included in the water-surface interaction energy.
LD output), until consistency between the quantum and the b. Phenol. Phenol on Ni(111) has a maximum surface
atomistic level is reached and all LD lowest energy conforma- interaction energy that is nearly 15% lower than the maximum
tions correspond to stable conformations from quantum calcula- surface interaction energy of benzene (see Table 2 and ref 14).
tions with respect to the energy and the configuration. Its conformation at the surface has the hydroxyl oxygen lifted
The newly modeled molecules are introduced below, starting by 0.5 Å, which has some effect on the adjacent carbon atom
with phenol which forms a logical extension to the benzene as well. The hydroxyl hydrogen is pointing down, ending up at
modeling we performed previously.6 First, however, we describe the same height as the phenylic carbon atoms (see Figure 2).
the modeling of water at the metal surface. As a basis of the atomistic modeling of phenol, the benzene
model described previously6 is chosen. To describe the interac-
a. Water. A large amount of research of either experimental
tion with the surface, Morse potentials are used on the aromatic
or theoretical nature exists in the field of water-surface
ring carbon atoms (Cr) (eq 2), and repulsive 10-4 potentials
interaction.22-27,29 Recent development in this field focuses on
(eq 3) are used on the ring hydrogens (Hr). The model
the aqueous-solid interface.23,25,26 The basics of the atomistic
parameters are given in Table 1, and atom types are defined in
modeling of the interaction of water with metal surfaces applied
Figure 3. For all attractive atom types, a zcutoff of 1.4 nm was
by us has been introduced in a previous publication.6 However,
used.
in that paper we considered only one classical water model.
{
Here we briefly repeat how the water-surface interaction is
M(1 - e-a(z-σ))2 - M z e zcutoff
modeled and then proceed further by showing that the relevant UAttr.Morse ) (2)
interfacial hydration properties are independent of the classical 0 z > zcutoff
{
water model chosen to describe water-water interactions. To
[52(σz ) - (σz ) + 53] z e σ
10 4
model the water-surface interaction, we make use of quantum 2π10-4
density functional calculations of small clusters of water URep.10-4 ) (3)
molecules at the metal surface. The water cluster configurations 0 z>σ
are chosen by considering all relevant ways for liquid water to Besides the substitution of one of the hydrogens of benzene
interact with the surface. Examples are shown schematically in by a hydroxyl group, only a reduction of interaction energy was
Figure 1. needed, which was done by scaling down all carbon-wall
We base our modeling idea on a well accepted and by now interactions. The carbon next to the hydroxyl group (Cp) was
proven statistical property of liquid water; that is, it is locally scaled down to half the interaction energy of the other carbon
and instantaneously tetrahedral. This allows us to imagine that atoms (Cr). A weak 10-4 repulsion was put on the hydroxyl
2634 J. Phys. Chem. C, Vol. 111, No. 6, 2007 Schravendijk et al.
TABLE 1: Overview of Molecule-Ni(111) Surface Force TABLE 3: Properties of Four Neutral Alanine
Field Parametersa Conformations That Were Evaluated in Quantum
Mechanical DFT Calculations14 a
interaction potential type (kJ/mol) σ (nm)
conformation Eads (eV) N (nm) Oc (nm) Cβ (nm)
Water
Ow-Ni attractive 10-4 6.40 0.24 NdownCHup -0.57 0.22 0.32 0.46
Hw-Ni no interaction NdownCHdown -0.32 0.24 0.33 0.36
OdownCHup -0.37 0.46 0.25 0.57
Benzene
OdownCHdown nonbonding 0.39 0.22 0.27
Cr-Ni Morse (a ) 35 nm-1) 17.5 0.20
Hr-Ni repulsive 10-4 4.27 0.20 a
Shown are the adsorption energy on the Ni(111) surface, and the
Phenol atom-to-top Ni layer distance of three of the atoms whose surface
Cr-Ni Morse (a ) 35 nm-1) 15.8 0.20 distance most strongly affects the surface interaction.
Hr-Ni repulsive 10-4 4.27 0.20
Cp-Ni Morse (a ) 35 nm-1) 7.96 0.20 cantly higher than the error margin of the quantum calculation.
Op-Ni repulsive 10-4 1.00 0.25 Currently, no special modeling considerations were made to
Hp-Ni attractive 10-4 0.70 0.22 account for this orientation in the classical atomistic simulation.
Alanine c. Neutral Alanine. Four distinct conformations were used
Oc-Ni attractive 10-4 8.90 0.23 in the initial quantum calculations of neutral alanine, as
N-Ni attractive 10-4 15.0 0.22 described in ref 14. As an example, the strongest binding
CRβ-Ni rep. Morse (a ) 6.0 nm-1) 4.0 0.58
Cv-Ni repulsive 10-4 10.0 0.38 conformation (0.57 eV) is shown in Figure 4; the energies and
R-Ni repulsive 10-4 4.27 0.20 optimal surface interaction site distances of all configurations
Phenylalanine are shown in Table 3. The amine nitrogen (N) and the carbonyl
Oc-Ni attractive 10-4 8.90 0.23 oxygen atom (Oc) interact attractively with the surface. Surface
N-Ni attractive 10-4 15.0 0.22 attraction at these sites is of the order of one to two hydrogen
Hv-Ni repulsive 10-4 4.27 0.17 bond strengths and is modeled with attractive 10-4 potentials.
Cv-Ni repulsive 10-4 10.0 0.38 An interesting feature is the influence of the methyl side-group.
Cr-Ni Morse (a ) 35 nm-1) 17.5 0.20
Even when not directly interacting with the surface, the methyl-
Hr-Ni repulsive 10-4 4.27 0.20
R-Ni repulsive 10-4 4.27 0.20 surface distance is found to influence the total surface interaction
energy. The origin of this influence might be an effect of the
Tyrosine
Oc-Ni attractive 10-4 8.90 0.23 position of the methyl group on the surface orientation of the
N-Ni attractive 10-4 15.0 0.22 interacting amine and carboxyl groups. A simple scheme to
Hv-Ni repulsive 10-4 4.27 0.17 model this effect is chosen that includes a weak repulsive Morse
Cv-Ni repulsive 10-4 10.0 0.38 potential (eq 4) on the center of mass of the CR and Cβ atoms:
Cr-Ni Morse (a ) 35 nm-1) 15.8 0.20 this additional interaction site is referred to as CRβ (see Figure
Hr-Ni repulsive 10-4 4.27 0.20
Cp-Ni Morse (a ) 35 nm-1) 7.96 0.20 3).
{
Op-Ni repulsive 10-4 1.00 0.25
Hp-Ni attractive 10-4 0.70 0.22 M(1 - e-a(z-σ))2 z e σ
R-Ni repulsive 10-4 4.27 0.20 URep.Morse ) (4)
0 z>σ
a
See Figure 3 for the corresponding structures and eqs 1-4 for the
potential functions used. The “R” covers all interactions with no specific The optimal distances (σ values) for these potentials are
interaction with the surface (unlabeled atoms in Figure 3). retrieved from the binding conformations in our quantum
TABLE 2: Ab Initio Interaction Energies and Optimal calculations. Binding energies for the potentials ( values) are
Distances Used to Model Interaction Potentials for Classical chosen such that the sum of all site-surface potentials (eqs 1-4)
MDa acting on the molecule will reproduce the total surface interac-
interaction Eads (eV) dopt (nm) tion energy as found in the quantum calculation, for every
conformation tested. Atoms with no specific surface interaction
water-Ni 10
-0.25 0.24
benzene-Ni 44 -1.05 0.20 were given a simple repulsive 10-4 potential (with parameters
phenol-Ni 44 -0.9 0.20 (ring) based on the repulsive hydrogens in the benzene ring, see group
Ala-Ni -0.57 0.22 (N) R in Table 1) to prevent the occurrence of unphysical conforma-
Phe-Ni 9 -1.1 0.20 (ring) tions corresponding to a penetration of the surface. The first
a
For neutral phenylalanine (Phe-Ni), only the optimal configuration iteration of this modeling resulted in minimal energy LD
was evaluated. For neutral alanine (Ala-Ni), the energy of the conformations that had both the amino nitrogen and the carbonyl
configuration with highest surface interaction is given. A complete oxygen interaction sites at optimum distance. The total surface
overview is given in Table 3. interaction energy corresponded to the sum of both interactions.
Consecutive quantum calculations of the LD output structures
oxygen (Op), and an attractive 10-4 potential (see eq 1) was showed that this was actually a nonbonding conformation.
put on the hydroxyl hydrogen (Hp). Consequently, a modification of our initial modeling was needed
Configurations of phenol with the OH group pointing away to ensure that only one interaction site at a time will be able to
from the surface did not need to be taken into account in QM bind with optimal interaction energy. This was reached by
calculations, because the perturbing effect of the OH group will introducing a repulsive site (Cv) at a position between the amino
be lower the further it is away from the surface, and any nitrogen and the carbonylic oxygen (at 40% of the bond length
interaction would be similar to the inclination dependence of between carbonylic carbon and CR, see Figure 3), leading to a
benzene modeled previously.6 The configuration where phenol seesaw-like mechanism that allows for the binding of either the
is pointing with the OH group toward the Ni(111) surface is amino or the carbonyl group. LD runs with this modeling lead
currently being investigated by quantum calculations.30 Pre- to optimal configurations close to the previous quantum
liminary calculations indicated a weak interaction, not signifi- calculations, see Figure 5, and this modeling was therefore
Hydrated Amino Acids at Metal Surfaces J. Phys. Chem. C, Vol. 111, No. 6, 2007 2635
Figure 1. Graphical aid showing the rationale behind the ab initio modeling. The water configurations studied by quantum mechanic DFT calculations
were chosen by considering all relevant ways for liquid water to interact with the surface. Some examples are schematically shown here. In white
circles are tetrahedral water substructures at the water-metal surface. I: A possible tetrahedral water structure with two waters interacting with the
metal surface. II: A trimer, the upper part of a tetrahedral water structure, with one water interacting with the metal surface and two additional
hydrogen-bonded water molecules. III: One of many configurations that could be discarded immediately because no electronic hydrogen-metal
interaction exists. The isolated water tetrahedron is an unstable structure, both in vacuo and metal-bound. Therefore, ab initio calculations are done
with smaller subunits. In DFT calculations, we can represent the tetrahedral structures shown in I and II by using all relevant water structures
consisting of (A) a water monomer, representing one of the symmetry axes of the tetrahedron; (B) a water dimer, representing the metal-bound
water with its first hydrogen-bonded neighbor (the tetrahedral center); (C) a water trimer, directly representing structure II. The bare water-metal
interaction energies (excluding contributions of H2O-H2O hydrogen bonding) are calculated as Eads ) E(surf+Nmol) - Esurf - ENmol, where Nmol refers
to the number of water molecules in the system, E(surf+Nmol) denotes the QM energy of the combined surface + water substructure, Esurf is the QM
energy of the isolated metal surface, and ENmol is the energy of the isolated water substructure. Eads/NH2O equals 0.25 ( 0.05 eV on Ni(111) independent
of NH2O ()1, 2, or 3). The complete overview of all conformations evaluated in quantum density functional calculations is given in ref 10.
a first water layer around z ) 0.2-0.3 nm, a second water layer the surface in Figure 6.
around z ) 0.5-0.6 nm, and a very weak third water layering
around z ) 0.8-0.9 nm. hydrated (z > 3 nm) and surface-adsorbed state (z ) 0.2 nm).
Our previous paper6 showed that the potential of mean force These variations are partly accounted for by the error margins
(PMF) for displacing a benzene molecule from bulk water to of these PMF calculations (between 4 and 8 kJ/mol at z ) 0.2
the surface not only gives an idea about adsorption/desorption nm), see the caption of Figure 7, and small differences in water
energies and intermediate energy barriers but also helps in densities at the surface. In addition, differences in bulk water
understanding the critical steps in the adsorption/desorption density and solvation free energies for the various water
mechanism.6 Any crucial effect that the choice of classical water models36-40 are likely to be a source for variations in the
model would have on the adsorption mechanism would therefore resulting free-energy difference.
be visible in the corresponding PMF. Figure 7 shows that with In Figure 8A, two-dimensional PMFs for benzene and phenol
all classical water models similar features are observed in the are shown as a function of z and the inclination angle between
PMF, with a major free-energy barrier (50-60 kJ/mol high) the aromatic ring- and surface normals. A selection of snapshots
starting almost directly at the surface (0.20 nm), ranging until of benzene at various surface distances is provided in Figure
ca. 0.60 nm. Additional oscillations occur further away from 8B. In this two-dimensional free-energy landscape, a more
the surface up to ca. 1.50 nm. Differences are found in the detailed picture is shown, and we can subdivide several zones
barrier heights and the free-energy difference between the bulk- when approaching the surface (z ) 0.2 nm) from the bulk
2638 J. Phys. Chem. C, Vol. 111, No. 6, 2007 Schravendijk et al.
Figure 7. Solute-surface PMFs for displacing the geometrical center of the phenyl ring of (A) benzene and (B) phenol perpendicular to a Ni(111)
surface, in liquid water (300 K) described with five classical water models. PMFs were obtained by integrating the average constraint force on the
solute center of mass along 140 discrete points between 0.20 and 3.00 nm from the metal surface. At each point, 3 ns MD runs were performed
to sample the mean constraint force. Solid black line, SPC water;33 dashed black line, SPC/E water;20 dashed/dotted black line, TIP3P water;34 solid
gray line, TIP4P water;34 dashed gray line, TIP5P water.35 The errors of these PMF calculations ranged between 4 and 8 kJ/mol at z ) 0.2 nm and
were estimated by calculating the block average error estimate45 for every constraint distance and taking the square root of the integral of the square
of all error estimates from bulk (z ) 3.0 nm) toward the closest distance to the surface that was sampled (z ) 0.20 nm).
solvent (z > 1 nm). In the bulk zone going from z > 1 nm show a similar profile to the solvated state only for short
toward z ) 0.8 nm the sampling is distributed evenly over all distances below 0.3 nm, because here the orientation of the ring
angles. For benzene (less clear for phenol), it can be seen that is governed by the presence of the surface.
around 0.7 nm (in between the third and second water layer;
see Figure 6) already some structural effects occur and the c. Amino Acids. To study surface-interacting amino acid
parallel conformation (cos(θ) ) 1) is slightly less sampled. At conformations, we performed several MD runs in which a single
distances corresponding to the second adsorbed water layer interaction site-surface distance (order parameter) was con-
(between z ) 0.6 nm and z ) 0.5 nm), benzene (phenol) strained and a PMF as a function of that order parameter was
preferentially orients parallel to the metal surface. (The distribu- obtained by integrating the mean force starting from z ) 3 nm
tion of cos(θ) is narrowed down to 0.9-1.0 in Figure 8A. A downward toward the surface. Using this method, the distances
snapshot of the parallel conformation can be seen in Figure 8B, with minimal free energy were determined. Snapshots of
at 0.49 nm.) This orientation is favored energetically because corresponding conformations as well as the average surface
of O-H‚‚‚π hydrogen bonding involving water molecules in interaction energies and relative free energies (as compared to
the first and second surface hydration layers. This type of weak z ) 3 nm) at these distances are shown in Figure 9.
water-aromatic hydrogen bonding is described properly by the
force field.41 In Figure 7, this causes a shoulder to appear in Comparison of the free energies in Figure 9A and B as well
the PMF at z ) 0.58 nm. A rather flat landscape with respect as Figure 9C and D shows that, independent of whether the
to the orientational degree of freedom is found in the region distance constraint is applied on the oxygen or nitrogen, similar
between the second and first adsorbed water layer (0.4 nm < z values are obtained, which is an indication that sufficient
< 0.5 nm). Although, toward z ) 0.4 nm, perpendicular sampling has been reached. The snapshots show that the amino
orientations are favored in comparison to parallel ones. In Figure acid conformations in Figure 9A and B as well as those in Figure
7 a shoulder is observed at these distances, and in Figure 8B 9C and D are similar. The free energies for phenylalanine
one can see how a perpendicular orientation minimizes the viewed from the amino N (Figure 9C) and carbonylic O (Figure
displacement of water molecules in both the first and second 9D) order parameters are lower than those for the alanine cases
adsorbed water layer. As the geometrical center of the ring (about 9 kJ/mol). This is rather counterintuitive because the
approaches the first adsorbed water layer, the benzene (phenol) bulky phenyl ring did not contribute to the surface interaction
ring normal gets significantly tilted with respect to the surface
energy for these order parameters, but intuitively its excluded
normal and finally lines up with the surface normal, driven by
volume can be assummed to displace more water molecules
an energetic stabilization of 1 eV due to the overlap of benzene-
(phenol)-π-orbitals with free electrons in the surface. In this from the surface as compared to the methyl group from alanine.
process, first layer water molecules are expelled from the An explanation could be that in the case of alanine the small
hydrophilic nickel surface, even before significant benzene- methyl group is able to come close to the surface, thereby weak-
(phenol)-surface binding interaction is present. It is clear that ening the interaction of amine and carbonyl groups. In contrast,
if the benzene would adsorb in a solvent-free environment then the bulky phenyl group of phenylalanine will not be able to
the z-dependent angular distribution plot would show a random pass the hydration layers present at the surface and therefore
orientation for distances of ca. 0.5 nm and higher and would will not hinder binding of the amine and carbonyl groups.
Hydrated Amino Acids at Metal Surfaces J. Phys. Chem. C, Vol. 111, No. 6, 2007 2639
Figure 8. (A) Two-dimensional PMFs for benzene (left) and phenol (right) at Ni(111) in SPC/E water (300 K). The solute (center-of-mass)-to-
surface distance is plotted vertically; the cosine of the angle between the solute- and surface normal vectors is plotted horizontally. The 2D-PMF,
G(z,θ) ) G(z) + G(θ|z), was calculated from constrained MD. By applying a constraint to keep the z coordinate fixed, G(z) (see Figure 7) is
obtained by integrating the mean constraint force along the z coordinate. G(θ|z) ) -kBT lnP(θ|z) is obtained from the conditional distribution
function, P(θ|z), sampled in the constrained MD runs. (B) Benzene snapshots at a selection of center-of-mass constraint distances from the surface.
The large positive free energy obtained by displacing the lanine are plotted against the distance of the constrained center
phenyl ring in phenylalanine toward the surface (Figure 9E) of mass of the phenyl ring. The contribution of water-surface
requires additional comments. This conformation has the lowest interactions decreases over the distance from bulk to optimal
total solute-surface interaction energy of all structures that adsorption with 140 kJ/mol, about 70 kJ/mol more than for the
interact with the surface as found here, but it is the least other cases, where, instead of the phenyl ring, the O or N
favorable from a free-energy point of view (∆Gads ) +32.6 interacts with the surface (data not shown). One can get an
kJ/mol ( 19.8 kJ/mol). A reasoning to explain this observation impression of this solvent effect when comparing the pheny-
would be to consider the fact that transferring the ring to the lalanine surface interaction in explicit solvent MD (Figure 9E)
surface requires a displacement of more water molecules than with LD runs in vacuum (Figure 9F), where we find free-energy
in all other cases. In Figure 10 the surface interaction energies minima at the same distance (geometrical center of the ring at
for water, phenylalanine, and the interaction sites of phenyla- 0.22 nm), similar interaction energies (-134.7 kJ/mol in explicit
2640 J. Phys. Chem. C, Vol. 111, No. 6, 2007 Schravendijk et al.
Figure 9. Snapshots of amino acids and parts of the surrounding water at the free-energy minimum distance of the constraint sites chosen, determined
by the PMF method as explained in the text. Eint denotes the sum of interaction energies for all solute interaction sites with the surface. ∆G denotes
the free energy of surface interaction, taken by the difference in the PMF between the distance given here, and the bulk state at z ) 3.0 nm. Error
bars are calculated by calculating the block average error estimate45 for every constraint distance and taking the square root of the integral of the
square of all error estimates from bulk (z ) 3.0 nm) toward the constraint distance mentioned here. ∆G for a given molecule is found to be
independent of the interaction site (within the error of the calculation) and dramatically dependent on the presence of water. (A) Alanine, amino
N constrained at z ) 0.24 nm. (B) Alanine, carbonyl O constrained at z ) 0.28 nm. (C) Phenylalanine, amino N constrained at z ) 0.24 nm. (D)
Phenylalanine, carbonyl O constrained at z ) 0.29 nm. (E) Phenylalanine, geometrical center-of-ring constrained at 0.22 nm. (F) As in part E, but
taken from an LD simulation in vacuum.
solvent MD and -137.8 kJ/mol in vacuum LD), but largely procedure to obtain a correct parameterization is presented in
different free-energy values (+32.6 kJ/mol in explicit solvent the current paper. The second factor is related to the competitive
MD vs -93.3 kJ/mol in vacuum LD). By looking at the water- adsorption energies of the solvent and solute, where one should
and solute-surface interaction energies in Figure 10, it is take into account the number of solvent molecules that have to
apparent that a competition between water and nitrogen plays be displaced by the solute.6,42 It has become apparent, however,
an important role: every increase in solute-surface interaction that taking these competitive effects alone does not provide the
energy is accompanied by a decrease of water-surface interac-
full description of the system because it does not properly
tion energy and vice versa.
The similarities in the surface interaction mechanism of account for solute hydration and the surface hydrophilicity.6 This
phenol and benzene (Figure 8) indicate that the tyrosine-surface is the third factor that should be taken into account: depending
interaction is similar to that of phenylalanine. on surface hydrophilicity, dense adsorbed water layers may exist
close to the surface. Intrusion of these mutually hydrogen-
V. Discussion bonded layers causes energy barriers for surface approach.
Four factors are necessary at the level of computer simulations Therefore, explicit solvent (atomistic) simulations are necessary,
in order to get a complete theoretical picture of peptide using timescales long enough to allow for solvent rearrange-
adsorption on metal surfaces. First, any molecule-metal surface ments. The fourth factor concerns the geometry and orientation.
interaction has to be parameterized by ab initio methods and Especially for longer molecules like polypeptides, many con-
cannot be represented by generalized force fields. An efficient formations exist next to the surface and a correct sampling has
Hydrated Amino Acids at Metal Surfaces J. Phys. Chem. C, Vol. 111, No. 6, 2007 2641
(10) Sebastiani, D.; Delle Site, L. J. Chem. Theory Comput. 2005, 1, (27) Verdaguer, A.; Sacha, G. M.; Bluhm, H.; Salmeron, M. Chem. ReV.
78. 2006, 106, 1478.
(11) Murakhtina, T.; Delle Site, L.; Sebastiani, D. ChemPhysChem. 2006, (28) Shelley, J. C.; Patey, G. N.; Bérard, D. R.; Torrie, G. M. J. Chem.
7, 1215. Phys. 1997, 107, 2122.
(12) Alavi, A.; Kohanoff, J.; Parrinello, M.; Frenkel, D. Phys. ReV. Lett. (29) Shelley, J. C.; Bérard, D. R. ReV. Comput. Chem. 1998, 12, 137.
1994, 73, 2599. (30) Ghiringhelli, L.; Delle Site, L. In preparation.
(13) Hutter, J.; Alavi, A.; Deutsch, T.; Bernasconi, M.; Goedecker, S.; (31) Jorgensen, W. L.; Tirado-Rives, J. J. Am. Chem. Soc. 1988, 110,
Tuckerman, M.; Parrinello, M. CPMD V. 3.4.1, 1995-1999. 1657.
(14) Ghiringhelli, L.; Schravendijk, P.; Delle Site, L. Phys. ReV. B 2006, (32) The setup for the DFT calculation of this phenylalanine conforma-
74, 035437. tion is the same as that specified in ref 14, except that a 4 × 4 Ni(111)
(15) van der Spoel, D.; Lindahl, E.; Hess, B.; Groenhof, G.; Mark, A.; surface was always used.
Berendsen, H. J. C. J. Comput. Chem. 2005, 26, 1701. (33) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.;
(16) Nosé, S. Mol. Phys. 1984, 52, 255. Hermans, J. In Intermolecular Forces; Pullman, B., Ed.; D. Reidel
(17) Parrinello, M.; Rahman, A. J. Appl. Phys. 1981, 52, 7182. Publishing Company: Dordrecht, The Netherlands, 1981; pp 331-342.
(18) Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. G. E. M. J. (34) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.;
Comput. Chem. 1997, 18, 1463. Klein, M. L. J. Chem. Phys. 1983, 79, 926.
(19) van Gunsteren, W. F.; Billeter, S. R.; Eising, A. A.; Hünenberger, (35) Mahoney, M. W.; Jorgensen, W. L. J. Chem. Phys. 2000, 112, 8910.
P. H.; Krüger, P.; Mark, A. E.; Scott, W. R. P.; Tironi, I. G. Biomolecular (36) Guillot, B. J. Mol. Liq. 2002, 101, 219.
Simulation: The GROMOS96 Manual and User Guide; vdf Hochschulverlag (37) Vega, C.; Sanz, E.; Abascal, J. L. F. J. Chem. Phys. 2005, 122,
AG an der ETH Zürich, 1996. 114507.
(20) Berendsen, H. J. C.; Grigera, J. R.; Straatsma, T. P. J. Phys. Chem. (38) Hess, B.; van der Vegt, N. F. A. J. Phys. Chem. B 2006, 110, 17616.
1987, 91, 6269. (39) Shirts, M. R.; Pande, V. S. J. Chem. Phys. 2005, 122, 134508.
(21) McDonald, I. R.; Rasaiah, J. C. Chem. Phys. Lett. 1975, 34, 382. (40) Raschke, T. M.; Levitt, M. J. Phys. Chem. B 2004, 108, 13492.
(22) Thiel, P. A.; Madey, T. E. Surf. Sci. Rep. 1987, 7, 211. (41) Schravendijk, P.; van der Vegt, N. F. A. J. Chem. Theory Comput.
(23) Guidelli, R.; Schmickler, W. Electrochim. Acta 2000, 45, 2317. 2005, 1, 643.
(24) Henderson, M. A. Surf. Sci. Rep. 2002, 46, 1-308. (42) Cooper, T. G.; de Leeuw, N. H. Langmuir 2004, 20, 3984.
(25) Michot, L. J.; Viliéras, F.; François, M.; Bihannic, I.; Pelletier, M.; (43) Delle Site, L.; Sebastiani, D. Phys. ReV. B 2004, 70, 115401.
Cases, J.-M. Geoscience 2002, 334, 611. (44) Delle Site, L.; Alavi, A.; Abrams, C. F. Phys. ReV. B 2003, 67,
(26) Taylor, C. D.; Neurock, M. Curr. Opin. Solid State Mater. Sci. 193406.
2005, 9, 49. (45) Hess, B. J. Chem. Phys. 2002, 116, 209.