Simulation Study of the Interaction of Some Albumin Subdomains with

We also define the interaction energy as Eint = (Efree + Eplanes) − Etot, where Efree is the energy of the free, isolated subdomain in the optimized...
2 downloads 0 Views 280KB Size
Langmuir 2003, 19, 3403-3412

3403

Simulation Study of the Interaction of Some Albumin Subdomains with a Flat Graphite Surface Giuseppina Raffaini and Fabio Ganazzoli* Dipartimento di Chimica, Materiali e Ingegneria Chimica “G. Natta”, Sez. Chimica, Politecnico di Milano, via L. Mancinelli 7, 20131 Milano, Italy Received November 15, 2002. In Final Form: January 15, 2003 We report atomistic computer simulations of some albumin subdomains on a hydrophobic graphite surface. The simulations are carried out in an effective dielectric medium by simple energy minimization and by long molecular dynamics (MD) runs. Further energy minimizations and shorter MD runs in the explicit presence of water are also performed to assess the stability of the geometries found and to describe the change of solvation of the adsorbed subdomains. We find that the initial adsorption is accompanied by significant rearrangements of the strands in contact with the surface, otherwise preserving the secondary structure and the overall globular shape. Much larger rearrangements take place at longer times during the MD runs, eventually yielding a thin layer of amino acids covering the surface as much as possible with complete denaturation. The interaction and strain energies of the adsorbed subdomains are discussed, together with their size and overall shape changes. The proposed adsorption mechanism is fully consistent with recent experimental findings for albumin on hydrophobic surfaces.

Introduction The interaction of macromolecules with heterogeneous surfaces is a relatively mature but still a very active research field (see, for instance, refs 1 and 2 and citations therein). In recent years, computer simulations have added significant new insights, complementing both theory and experiments. Coarse-grained models, either on a lattice or in continuum space, have often been used to integrate out the most local, irrelevant details or the fastest motions in order to describe the large-scale behavior, while atomistic models are often required to describe specific interactions with a surface at the nanometer scale.2 On the other hand, the interaction of biological macromolecules, in particular proteins, with a foreign surface is a closely related field that is not easily amenable to either theory or coarse-grained simulations because of the structural details at the atomistic level and of the denaturation process these molecules can undergo upon adsorption.3-10 Semimacroscopic (colloidal) models are fairly satisfactory from many viewpoints when electrostatic interactions with a charged surface are relevant,11,12 the * To whom correspondence should be addressed: e-mail [email protected]. (1) Israelachvili, J. Intermolecular and Surface Forces; Academic Press: London, 1992. (2) Yoon, D. Y.; Vacatello, M.; Smith, G. D. Simulation Studies of Polymer Melts at Interfaces, in Monte Carlo and Molecular Dynamics Simulations in Polymer Science; Binder, K., Ed.; Oxford University Press: New York, 1995; pp 433-475. (3) Proteins at interfaces II: Fundamentals and Applications; Brash, J. L., Horbett, T. A., Eds.; ACS Symposium Series 602; American Chemical Society: Washington, DC, 1995. (4) Wertz, C. F.; Santore, M. M. Langmuir 1999, 15, 8884-8894. (5) Wertz, C. F.; Santore, M. M. Langmuir 2001, 17, 3006-3016. (6) Buijs, J.; Norde, W.; Lichtenbelt, J. W. Th. Langmuir 1996, 12, 1605-1613. (7) Moulin, A. M.; O’Shea, S. J.; Bradley, R. A.; Doyle, P.; Welland, M. E. Langmuir 1999, 15, 8776-8779. (8) Kim, D. T.; Blanch, H. W.; Radke, C. J. Langmuir 2002, 18, 58415850. (9) Petrash, S.; Liebmann-Vinson, A.; Foster, M. D.; Lander, L. M.; Brittain, W. J. Biotechnol. Prog. 1997, 13, 635-639. (10) Sheller, N. B.; Petrash, S.; Foster, M. D.; Tsukruk, V. V. Langmuir 1998, 14, 4535-4544. (11) Roth, C. M.; Lenhoff, A. M. Langmuir 1995, 11, 3500-3509.

Coulombic interactions being typically described through the Poisson-Boltzmann equation and the van der Waals interactions through the Hamaker approach. These methods ignore all atomistic features of the proteins, including also the detailed pattern of the electrostatic potential at the molecular envelope. Therefore, more recent work explicitly considered the full protein structure to investigate the initial adsorption stages on a charged surface.13-15 However, these approaches still assume a rigid molecular structure. Accordingly, they cannot describe the rearrangements and possibly denaturation of proteins taking place upon adsorption on a neutral surface in order to maximize the interaction energy.4,5,10 Such features are of paramount interest both from an academic viewpoint and in many practical cases, for instance when assessing the biocompatibility of a foreign material in a body implant. Modeling such interactions for an ideal yet experimentally realistic case with atomistic simulations is the main goal of the present methodological study. Proteins can be viewed as copolymers consisting of different natural amino acids, connected into an apparently random sequence. The simplest amino acid characterization is in terms of their hydrophobic or hydrophilic nature, for instance through the Kyte-Doolittle hydropathy scale.16 Accordingly, theoretical models have been developed for amphiphilic copolymers made up of two unlike units, and the conformational features of these systems in selective solvents have been extensively studied, allowing for different degrees of hydropathy.17-19 Some interesting clues about the protein folding problem (12) Ståhlberg, J.; Jo¨nsson, B.; Horva´th, C. Anal. Chem. 1991, 63, 1867-1874; Ibid. 1992, 64, 3118-3124. (13) Noinville, V.; Vidal-Madjar, C.; Se´bille, B. J. Phys. Chem. 1995, 99, 1516-1522. (14) Asthagirl, D.; Lenhoff, A. M. Langmuir 1997, 13, 6761-6768. (15) Ravichandran, S.; Madura, J. D.; Talbot, J. J. Phys. Chem. B 2001, 105, 3610-3613. (16) Kyte, J.; Doolittle, R. F. J. Mol. Biol. 1982, 157, 105-132. (17) (a) Dill, K. A.; Bromberg, S.; Yue, K.; Fiebig, K. M.; Yee, D. P.; Thomas, P. D.; Chan, H. S. Protein Sci. 1995, 4, 561-602. (b) Pande, V. S.; Grosberg, A. Yu.; Tanaka, T. Rev. Mod. Phys. 2000, 72, 259-314. (18) Onuchic, J. N.; Luthey-Schulten, Z.; Wolynes, P. G. Annu. Rev. Phys. Chem. 1997, 48, 545-600. (19) Ganazzoli, F. J. Chem. Phys. 1998, 108, 9924-9932; Ibid. 2000, 112, 1547-1553.

10.1021/la026853h CCC: $25.00 © 2003 American Chemical Society Published on Web 03/06/2003

3404

Langmuir, Vol. 19, No. 8, 2003

have emerged,17 and insights about the kinetics of absorption, denaturation,20 and possibly refolding21 on a surface were gained through computer simulations on a lattice. However, it is clear that such description provides an oversimplified view of some important details of protein behavior. In particular, the local atomistic features may be highly relevant in many contexts. In these instances, one must account for the peptide moiety, which allows for the backbone hydrogen bonds (or H-bonds, in the following), hence for the secondary structure, and for the specific features of the hydrophobic or hydrophilic side groups. In particular, such detailed description is required for studying the adsorption of proteins on heterogeneous surfaces within a specific medium. Only atomistic computer simulations can cope with this task and provide a realistic description of the adsorption process, although the size and the complexity of the problem is often technically daunting. In the present paper, we investigate the interaction of some protein subdomains with a hydrophobic graphite surface through molecular mechanics and molecular dynamics (MD) simulations. The graphite surface was chosen because of its relative simplicity and rigidity, so that it can be treated as a fully rigid body for all purposes. Also, this surface may form an idealized, zeroth-order approximation of pyrolytic carbon, widely used in certain implants such as cardiac valves, where it interacts with blood.3,22 It should be remembered, however, that true pyrolytic carbon is a more complex material, being formed by randomly oriented graphite microcrystals with a size on the order of a few nanometers. Thus, different crystallographic planes may be exposed at the surface, together with truly amorphous carbon domains, producing additional edges and grooves that greatly modify protein adsorption. Moreover, partial oxidation of the surface may also modify its behavior, making it less hydrophobic. Such features, and of course any surface treatment (polishing, for instance), do also affect to various degrees the interactions of blood proteins with pyrolytic carbon. Interestingly, such interactions greatly enhance the biocompatibility of pyrolytic carbon despite its rigidity and hydrophobicity, largely due to the fast adsorption of albumin, the most abundant blood protein. Accordingly, in this preliminary study we investigate the interaction and adsorption of specific subdomains of human serum albumin onto the flat [0001] graphite surface, modeled through two large planes of carbon atoms. The very large size of albumin prevents accounting for the whole molecule, hence our choice of the subdomains. Also, we consider all the side groups in their nonionized, neutral state. Although this could be a serious limitation for a quantitative comparison with experiments, we believe that interesting features of the adsorption process over a hydrophobic surface can still be assessed.5 Moreover, the present study will provide a reference case for more realistic simulations at the physiological pH, where significant, or even complete, ionization takes place. We carried out the simulations according to the following two-step strategy: (i) direct energy minimizations of the subdomains close to the graphite surface, considering at first an effective dielectric medium affecting the dipolar interactions; (ii) MD runs of selected, optimized geometries in this dielectric medium and subsequent energy mini(20) Zhdanov, V. P.; Kasemo, B. Proteins: Struct., Funct., Genet. 1998, 30, 168-176 and 177-182. (21) (a) Zhdanov, V. P.; Kasemo, B. Proteins: Struct., Funct., Genet. 2001, 42, 481-494. (b) Catells, V.; Yang, S.; Van Tassel, P. R. Phys. Rev. E 2002, 65, 031912. (22) Feng, L.; Andrade, J. D. J. Biomater. Sci. 1995, 7, 439-452.

Raffaini and Ganazzoli

mization of the instantaneous snapshots. The first procedure corresponds to the initial adsorption stage, but it can also yield the preferred conformations if the adsorption process is either under kinetic control or at large surface coverage. Conversely, the second procedure yields the best overall conformation on a clean surface with the largest interaction energy under thermodynamic control and leads to the formation of a stable thin layer of amino acids, basically a monolayer. The obvious computational advantages of carrying out the simulations in the dielectric medium shall also be discussed in comparison with additional simulations carried out by explicitly including a large number of water molecules. Although the latter procedure might allow for the detection of possible conformational changes due to the solvent, it shall be mainly used to analyze the hydration of the subdomains in the various adsorption stages. In the following, after summarizing the simulation method, we describe our results for the two steps described above, first through simulation in the dielectric medium and then in water. In the final section, we summarize the results and discuss their implications for albumin adsorption on a graphite surface. Simulation Method All simulations were performed with the InsightII/Discover 2000 package, distributed by Molecular Simulations Inc., now Accelrys Inc.23 (San Diego, CA), using throughout the consistent valence force field24 CVFF with a Morse potential for the bonded atoms. The coordinates of the non-hydrogen atoms used for the initial trial geometries were taken from the experimental results obtained by single-crystal X-ray analysis and deposited with the Protein Data Bank (human serum albumin, 1AO6, ref 25). The selected subdomains (see later) were terminated with -COOand -NH3+ groups in a zwitterionic form, and the hydrogen atoms were added in the calculated positions. For simplicity, all the ionizable side groups were considered in their neutral state. The subdomains were then fully optimized up to an energy gradient lower than 4 × 10-3 kJ mol-1 Å-1. The graphite planes were prepared from scratch. We first built a single plane using appropriate templates of the simulation package and saturating with hydrogen atoms the dangling bonds at the edges. After full optimization in vacuo, we duplicated this plane and optimized the geometry and the position of the second plane by keeping fixed the first one. The protein fragments were then placed close to the surface and minimized with the assumption of an effective dielectric medium with a distance-dependent dielectric constant asymptotically yielding the correct water value ( ) 78). The energy minimizations in water were carried out by adding a few thousands of water molecules with periodic boundary conditions, adjusting the density to 1 g cm-3, and setting the dielectric constant to  ) 1. The MD simulations were performed in the dielectric medium or in water with periodic boundary conditions at a constant temperature (T ) 300 K), controlled through the Berendsen thermostat. Integration of the dynamical equations was carried out with the Verlet algorithm with a time step of 1 fs, and the instantaneous coordinates (or frames) were periodically saved for further analysis or geometry optimization. In all cases, the long MD runs in the dielectric medium lasted for 1 ns. Within these runs, the total and potential energy showed an initial decrease, possibly with a few separate kinetic stages, and then fluctuated around a constant value, indicating that an equilibrium state was eventually reached. Each frame collected during the MD run was then minimized up to a gradient of less (23) Accelrys Inc. InsightII 2000; Accelrys Inc.: San Diego, CA, 2000. See also the URL http://www.accelrys.com. (24) Dauber-Osguthorpe, P.; Roberts, V. A.; Osguthorpe, D. J.; Wolff, J.; Genest, M.; Hagler, A. T. Proteins: Struct., Funct., Genet. 1988, 4, 31-47. (25) Berman, H. M.; Westbrook, J.; Feng, Z.; Gilliland, G.; Bhat, T. N.; Weissig, H.; Shindyalov, I. N.; Bourne, P. E. The Protein Data Bank. Nucleic Acids Res. 2000, 28, 235-242. See also the URL http:// www.rcsb.org/pdb/.

Interaction of Albumin Subdomains with Graphite

Langmuir, Vol. 19, No. 8, 2003 3405 Table 1. Amino Acids Contained within the A and E Subdomains and r-Helices, Together with the Corresponding Hydropathy Index

subdomain

R-helix

A1-A2-A3 A1 A2 A3 E2-E3-E4 E2 E3 E4

Figure 1. Position of the A and E subdomains within the whole albumin molecule. For simplicity, the remainder of the molecule is only shown through cylinders for the R-helices and ribbons for the random strands, while the two subdomains are shown in detail.

first amino acid

last amino acid

no. of amino acidsa

Ser5 Ser5 Glu16 Pro35 Tyr401 Tyr401 Ser419 Lys444

Lys64 Asp13 Leu31 Asp56 Thr467 Val415 Cys438 Thr467

60 9 16 22 67 15 20 24

hydropathy indexb -10.6 8.5 -2.7 -5.3 -0.3 -0.3

a The number of amino acids of the subdomains exceeds the sum of those in the R-helices because of the presence of regular turns and random strands. b The hydropathy index was simply calculated as the sum of the values for the individual amino acids, from ref 16.

than 4 × 10-3 kJ mol-1 Å-1. Additional short MD runs with a few thousands of water molecules were used to analyze the distribution of the solvent around the backbone of each subdomain through the radial distribution function (see below). As before, the density was set to 1 g cm-3, and periodic boundary conditions were used. These runs lasted in all cases for 10 ps, and the frames were saved every 80 fs. Data Analysis. The geometries periodically sampled in the MD runs can be usefully analyzed through the radial distribution function gij(r) (or RDF for short). This function gives the probability density of finding atoms j at a distance r from atoms i, and is defined as

gij(r) )

d〈Nij(r)〉 Fj dV(r)

where d〈Nij(r)〉 is the average number of times the j atoms are comprised in a spherical shell of thickness dr at a distance r from atoms i within an MD run, and Fj is their bulk density. Thus, gij(r) yields the average local density of atoms j (normalized by their average bulk value) in the shell volume dV(r) comprised at a distance between r and r + dr from atoms i. Therefore, the RDF provides an immediate picture of an increased or depleted density of j atoms due to specific interactions, for instance in a coordination shell. In the following, the ij subscripts will be dropped for simplicity, since no ambiguity between the two sets of atoms may arise.

Results and Discussion The albumin molecule consists of two main parts, denoted as A and B, that are equivalent and basically symmetrical, though not crystallographically. In turn, each part is hierarchically divided into six domains, labeled from A through F, that repeat in three pairs, namely, A-B, C-D, and E-F. Finally the A, C, and E domains comprise six R-helices, while the B, D, and F domains are formed by four R-helices. These helices, hence also the domains, are connected by a few amino acids present as regular turns or in a random conformation. Two subdomains belonging to the A and E domains of part A were selected for the present study. They are formed by three connected R-helices and short random sections and consist of the first, second, and third helix of domain A and of the second, third, and fourth helix of domain E. In the following, the subdomains will be denoted as A1-A2-A3 and E2-E3-E4, or more simply as A and E subdomains. We note that in both cases a disulfide bridge is retained throughout the simulations. The location of the subdomains within the whole albumin molecule is shown in Figure 1. More details on the amino acids they comprise

Figure 2. Structure of the individual A and E subdomains. Hydrogen atoms have been omitted for clarity. Cylinders indicate R-helices, while ribbons denote random strands and regular turns.

and their R-helices are reported in Table 1, together with the hydropathy index according to the Kyte-Doolittle scale.16 The latter index was calculated by simply summing the values of the constituent amino acids (more precisely, of their side groups). A positive hydropathy value indicates hydrophobicity, and a negative one, hydrophilicity. Thus, helices A2 and A1 form the two extreme cases within the chosen subdomains. However, we note that some amino acids of unlike hydropathy index are anyway present in all the R-helices, irrespective of the overall index. The structure of the selected A and E subdomains is shown more clearly in Figure 2, where the secondary structure is highlighted by cylinders denoting the R-helices and by ribbons that indicate random strands and regular turns. Note that in the E subdomain the long E3 and E4 helices are apparently split in two and in three sections due to local deviations from the ideal geometry.

3406

Langmuir, Vol. 19, No. 8, 2003

As for the heterogeneous surface, we generated two graphite domains having quite different sizes. The first domain measured 52 Å × 30 Å and was used for the initial simulations, while the second one had a 3 times larger surface, measuring 84 Å × 59 Å. The latter surface was used to check for possible edge effects of the planes in the initial simulations and in the MD runs when significant denaturation is present, with a large spreading of the subdomains on the surface. In both cases, the graphite material consisted of two carbon planes only, no additional planes being deemed necessary, on the basis of both literature results26 and our exploratory results about the interaction of single R-helices with up to six small planes.27 Preliminary optimization of isolated single R-helices showed a relatively large influence of the surrounding medium. In fact, the conformations obtained in the dielectric medium do essentially coincide with those found in the presence of water molecules, both being very similar to the experimental geometries. Conversely, the energy minimizations carried out in vacuo strongly modify the conformation of the hydrophilic helices, though they have only a minor influence on the hydrophobic ones. This effect is in part related to the small size of the isolated helices, being most relevant at the two ends. Nevertheless, because of these results all subsequent simulations were only carried out in the dielectric medium or in water. In this way, we may also assess the possible influence of the H-bonds between the subdomains and the water molecules on the overall conformation. In this context we point out that the energy minimizations carried out in the presence of a large number of water molecules are adequate to investigate the intramolecular conformation, in particular of the side groups, the solvation of the subdomains, and the intra- and intermolecular H-bonds. However, they cannot be used to assess the relative stability of different geometries. In fact, in the presence of the solvent molecules the overwhelming majority of atoms belongs to water, which therefore dominates the total energy. Moreover, from the thermodynamic viewpoint the energy minimizations are equivalent to freezing the system at 0 K. Therefore, they lead to the optimization of water molecules in some local glasslike configuration of the phase space, while the solvated groups are but a minute fraction of the whole system. Such procedures are then plagued by the problem of the huge number of local minima typical of the glassy state. In conclusion, the optimizations in water can be used to address the conformational effects due to the solvent and the hydration pattern but not the stability of the geometries under study. 1. Initial Adsorption by Direct Energy Minimizations in the Dielectric Medium. We first optimized the geometry of the isolated subdomains for the later determination of the interaction energy with the graphite surface. In the dielectric medium, the minimized energies amount to 2.47 and 2.61 MJ mol-1 for the A and E subdomains, respectively. In both cases, we found little difference between the optimized and the experimental geometries of the backbone but different orientations of the side groups, mostly due to the lack of hydrogen bonds with the solvent. Afterward, we optimized the geometries of the subdomains close to the graphite surface, keeping fixed throughout the two carbon planes. In this section we report the results obtained on the smaller planes (52 Å × 30 Å) (26) Hentschke, R. Macromol. Theory Simul. 1997, 6, 287-316. (27) Mantero, S.; Piuri, D.; Montevecchi, F. M.; Vesentini, S.; Ganazzoli, F.; Raffaini, G. J. Biomed. Mater. Res. 2002, 59, 329-339.

Raffaini and Ganazzoli

Figure 3. Directions of close approach to the flat graphite surface of the initial geometries for the A subdomain in a cylinder and ribbon representation. Similar directions were also adopted for the E subdomain. The two carbon planes are shown in a side view.

for a comparison with the geometry opimizations in water (see later). We checked on larger planes that possible finitesize effects of the surface are entirely absent. Eight different starting orientations were chosen for both subdomains. These arrangements are depicted in Figure 3 for the A subdomain (analogous arrangements were adopted for the E subdomains) and may be qualitatively described as follows: (i) three orientations with the ends of two different helices approaching the surface, together with the adjacent random strands, when present (positions 1, 3, and 5); (ii) three orientations with each R-helix parallel to the surface (positions 2, 4, and 6); and (iii) two orientations with either side of the plane through the midpoint of the three helices facing the surface (positions 7 and 8). In all cases we found a significant initial adsorption on the graphite surface, usually accompanied by relatively large conformational rearrangements of the interacting strands. These rearrangements involve a partial loss of secondary structure in the neighborhood of the surface, particularly pronounced for the hydrophobic, or at least the less hydrophilic, R-helices, which show the strongest interactions. Before describing in detail the most relevant conformational changes of the adsorbed subdomains, we anticipate that they always correspond to local energy minima, as it will be shown later through long MD runs. The results of these initial energy minimizations are reported in Table 2. The lowest energy minima, corresponding to the most stable states found by this procedure, amount to 0.92 and 0.85 MJ mol-1 for the A and E subdomains, respectively. These values are henceforth taken as our zero values for either domain. All other total energy minima (Etot) are reported relative to these reference values (Erel in Table 2). We also define the interaction energy as Eint ) (Efree + Eplanes) - Etot, where Efree is the energy of the free, isolated subdomain in the optimized geometry. According to this definition, Eint > 0 is the energy released by the subdomains upon adsorption. Note that, since in our simulations the planes are kept

Interaction of Albumin Subdomains with Graphite

Langmuir, Vol. 19, No. 8, 2003 3407

Table 2. Results of the Initial Energy Minimizations and of the Long MD Runs and Subsequent Optimizations in the Dielectric Mediuma subdomain A1-A2-A3

subdomain E2-E3-E4

Erel

Eint

Estrain

n5Å

broken Hintra

1 2 3 4 5 6 7 8

0.56 0 0.25 0.10 0.82 0.21 0.14 0.01

1.00 1.55 1.31 1.46 0.73 1.34 1.41 1.54

0.08 0.35 0.42 0.24 0.17 0.32 0.30 0.44

12 20 20 18 12 15 15 23

7 15 18 16 1 13 11 12

I II

-1.83 -1.66

3.44 3.26

0.95 0.69

59 52

pos

Eint

Estrain

n5Å

broken Hintra

RAA

0.79 1.52 0.20 1.56 0.83 1.00 1.76 1.59

0.19 0.45 0.08 0.26 0.15 0.25 0.52 0.43

12 18 2 21 13 15 32 21

2 9 -2b 13 4 3 16 16

33 32 46 20 37 41 22 30

After MD Runs and Minimization 29 -1.98 3.71 30 -1.67 3.39

1.14 0.72

65 56

34 24

RAA

Erel

Initial minimization 33 0.97 23 0.24 27 1.56 17 0.20 35 0.93 22 0.75 31 0 17 0.17

a We report the total energy relative to the lowest-energy state found in the initial minimizations, E ; the interaction energy of the rel subdomains with the carbon planes, Eint; the strain energy of the subdomains, Estrain; the number of amino acids at a distance less than broken 5 Å from the plane, n5Å; the number of intramolecular H-bonds broken after adsorption, Hintra ; and the number of amino acids still found in R-helices, RAA. All energies are in megajoules per mole (see text). b A negative value actually indicates formation of additional H-bonds.

Figure 4. Geometries of the most stable A and E subdomains adsorbed on the graphite surface, obtained by direct energy minimization in the dielectric medium (initial adsorption stage). The conformational changes, in particular the loss of secondary structure close to the surface, may be gauged by the short length of the R-helices (red cylinders) in comparison with their initial size in Figure 2. The random strands are shown as green ribbons, and the regular turns are shown in blue. The detailed chemical structure is also shown, omitting the hydrogen atoms for clarity. Color codes for the atoms are carbon, green; oxygen, red; nitrogen, blue; and sulfur, yellow.

fixed, we have Eplanes ≡ 0, although of course all their carbon atoms do correctly interact with the subdomains. We also define Estrain ) Efrozen - Efree, where Efrozen is the energy of the isolated subdomains in the frozen geometry they adopt upon adsorption. Let us describe the subdomain rearrangements found through these initial minimizations. In all cases, the strands close to the surface locally optimize the interactions with the carbon planes by loosing in part the secondary R-helix structure, as shown in Figure 4 for the most stable arrangements. In particular, within the A subdomain the hydrophobic helix A2 is largely unwound, thanks to its large interactions with the graphite surface. However, even the less hydrophilic helices (E3 and E4, or

A3, for instance) are significantly shortened when close to the surface, in order again to maximize the amino acidsurface interactions through a partial loss of secondary structure. Actually, some helicoidal features are retained but with large distortions from the typical conformation of the R-helices, so that the H-bonds between neighboring turns are absent. Therefore, these helicoidal strands cannot be classified as genuine R-helices. The driving force for unwinding the helices consists of the favorable van der Waals interactions with the carbon plane. Thus, Eint can be as large as 1.55 MJ mol-1 for the A subdomain that contains the most hydrophobic helix A2, while the E subdomain, having more hydrophobic amino acids in helices E3 and E4, shows a larger interaction energy, up to 1.76 MJ mol-1. The above values correspond in either case to an average of about 25-26 kJ mol-1 per amino acid. Interestingly, two almost equivalent energy minima are found in the A subdomain, which only differ by 8 kJ mol-1 but show different adsorption geometries. This is also shown by the different strain energy, amounting to more than 90 kJ mol-1 (see Table 2). The larger deformation of the A subdomain in position 8 allows stronger interactions with the surface that largely compensate the greater strain energy. As a general feature, upon adsorption the subdomains achieve a lower energy, hence a larger Eint, when more amino acids interact with the surface, irrespective of their hydropathy index, thanks to the apparently random distribution of hydrophobic and hydrophilic residues. In fact, there is a significant positive correlation between Eint and the number of amino acids at an interaction distance with the surface. Such correlation is shown in Figure 5a, where Eint is plotted as a function of the number of amino acids being at less than 5 Å from the surface, n5Å. No significant difference between the two subdomains is evident, so that we are confident that the correlation is quite general. Considering the data points for both subdomains, the best-fit line through the origin is given by

Eint ) 71(3)n5Å (kJ mol-1)

(1)

with the figure in parentheses giving the standard error on the last significant digit. In a way, the overall hydropathy of the helices is qualitatively less important. In fact, even the hydrophilic A1 helix shows a significant Eint when close to the surface (position 7 of Table 2), though

3408

Langmuir, Vol. 19, No. 8, 2003

Raffaini and Ganazzoli

Figure 5. Interaction energy Eint (a, lower panel) and strain energy Estrain (b, upper panel) of the A and E subdomains over the carbon planes after the initial energy minimizations in various orientations plotted as a function of n5Å, the number of amino acids being in contact with the surface (i.e., at a distance less than 5 Å). The straight lines are the best-fit lines through the origin given by eqs 1 and 2. The correlation coefficients are R ) 0.903 (a) and R ) 0.856 (b).

not as large as in the case of the hydrophobic A2 helix. With hindsight, the latter finding is not unexpected, due to the presence of hydrophobic residues in all helices and to the cooperative nature of the adsorption process of a complex copolymer such as a protein fragment. In conclusion, a low Etot, hence a large Eint, is achieved when more amino acids, hence also more hydrophobic residues, are in contact with the graphite surface. The subdomains undergo a large strain in order to optimize the interactions with the surface. Therefore, we expect that a larger interaction should usually be accompanied by a greater strain energy of the adsorbed subdomains, hence a larger deformation, so as to allow more amino acids to be in contact with the surface. Indeed, we see in Figure 5b that Estrain is well correlated with n5Å, just as Eint. The best-fit line through the origin is

Estrain ) 17(1)n5Å (kJ mol-1)

(2)

Therefore, we also find a positive correlation between Eint and Estrain, though with a larger scatter than shown in Figure 5. From eqs 1 and 2 we get an interesting result: Eint increases with n5Å about 4 times faster than Estrain. By extrapolation, we expect that the subdomains may also undergo much larger deformations than reported in Figure 4 because of the increasingly stronger interactions with the surface. We will show later through MD simulations that this is indeed the case. The strain energy is mainly (though not exclusively) due to the energy cost of breaking the intramolecular H-bonds of the R-helices. Thus, Estrain is also roughly correlated with the number of broken H-bonds (see Table 2). Additionally, a larger Estrain is also accompanied by a smaller number of amino acids still structured in R-helices (see again Table 2). Conversely, the higher energy states show only minor conformational changes of the adsorbed

subdomains. Therefore, they show a smaller strain energy but also a smaller Eint, with fewer broken H-bonds and more amino acids keeping the original R-helical structure. 2. Final Adsorption by MD Runs and Energy Minimizations in the Dielectric Medium. Selected geometries obtained in the previous section were subjected to MD runs in search of the best adsorption geometry with the overall energy minimum. For both subdomains we chose as starting points both the lowest- and the highest-energy geometries found by direct energy minimizations. These initial geometries respectively correspond to positions 2 and 5 for the A subdomain and to positions 7 and 3 for the E subdomain (see Table 2). Preliminary trial runs indicated that the large carbon planes (84 Å × 59 Å) were required to avoid any influence of the edges, due to the large spreading of the subdomains. When starting with the most stable geometries previously found, the MD runs yielded a relatively fast and steady energy decrease lasting for about 150-200 ps for both subdomains. Soon afterward, an equilibrium state was quickly reached, and the final part of the MD simulation, carried out for 1 ns, showed only relatively minor readjustments, mainly involving some local features with modest energy changes. Further energy minimization of selected frames provided no further significant changes, showing these final adsorbed states to be equivalent to better than (20 kJ mol-1. Some details of these geometries are reported as entry I in Table 2 and shall be discussed below. The large rearrangements undergone by the subdomains within the initial stage of the MD runs can be qualitatively described as a fast liquidlike spreading of the adsorbed subdomains on the surface, possibly followed by small later readjustments of minor importance. The whole process leads to a very large contact area between the subdomains and the surface, in agreement with recent experimental results for albumin on a hydrophobic surface (a C16 self-assembled monolayer).4,5 A more detailed kinetic study of the adsorption and spreading of these subdomains on a graphite surface shall be reported elsewhere in comparison with the behavior of different proteins.28 For the albumin subdomains considered here, the whole process quickly leads to the formation of a monolayer of amino acids evenly coating the surface, as shown in Figure 6 for the A subdomain. Here, all the R-helices are fully denatured and all the amino acids are in contact with the carbon plane. In these final states, the whole pattern of intramolecular H-bonds is disrupted. Only a few hydrogen bonds are found, mostly involving the side groups of topologically adjacent amino acids. Additionally, only two H-bonds involving the amide moieties of different residues are present in each subdomain between topologically distant amino acids that are kept close by the disulfide bridges. A similar kinetic pattern is also roughly followed by the geometries that initially have the highest energy minima. In this case, however, the overall process is a bit more complicated in some details. In fact, at the beginning of these MD runs, lasting again for 1 ns, only a few amino acids are already in contact with the surface, and the projections of the residues on the plane largely overlap, unlike what happened in the previous case. Nevertheless, after some tilting of the subdomains toward the surface, the large liquidlike spreading discussed before sets in, soon becoming the dominant process. Actually, this spreading is slowed and somewhat hindered by the interactions among the residues, mostly due to H-bonds involving both the side groups and the backbone. Accord(28) Raffaini, G.; Ganazzoli, F. Manuscript in preparation.

Interaction of Albumin Subdomains with Graphite

Langmuir, Vol. 19, No. 8, 2003 3409 Table 3. Radius of Gyration Rg, Principal Axes λi (i ) 1, 2, 3) Arranged in Decreasing Order, Asphericity ∆, and Prolateness S (see text) for the Isolated Subdomains (1) and for the Adsorbed States in the Lowest-Energy Minima after Direct Minimization (2) and in the Most Stable State Reached after the MD Runs in the Dielectric Medium (3)a subdomain A1-A2-A3 Rg

λ1

λ2

λ3

subdomain E2-E3-E4 S



Rg

λ1

λ2

λ3



S

1 12.1 9.6 5.6 4.9 0.19 0.16 12.8 10.3 6.5 3.8 0.25 0.16 2 12.8 10.5 6.3 4.0 0.26 0.20 14.0 11.2 7.8 3.3 0.26 0.06 3 21.5 19.3 9.3 1.0 0.54 0.63 21.0 17.4 11.7 1.0 0.35 0.07 a

All lengths are in angstroms (1 Å ) 0.1 nm).

stage, as reported in Table 3. However, a description of the molecular deformation at the surface is of greater interest. The molecular shape can be described through the principal axes λi (i ) 1, 2, 3) obtained upon diagonalization of the radius of gyration tensor and quantified through the asphericity ∆ and prolateness S also reported in Table 3. The latter quantities are defined as29 Figure 6. The A subdomain in its lowest-energy state after the MD runs and subsequent energy minimization (side and top views). The formation of a monolayer is evident. Note also the lack of any secondary structure (R-helices, shown before as cylinders) in both cases; the backbone consists only of a bidimensional random strand shown as a ribbon. The E subdomain achieves a similar monolayer, not shown for brevity.

ingly, some secondary structure (i.e., short R-helices) does persist for hundreds of picoseconds far from the surface, although it is eventually destroyed. Some details of the final geometries are reported as entry II in Table 2. These geometries mainly consist again of a monolayer of amino acids but show also a few overlapping residues, and therefore they are less stable. It is of course possible that even these geometries may eventually form a true monolayer, similar to that shown in Figure 6, in much longer MD simulations that unfortunately are not possible with our current computer resources. The very large stability of the final equilibrium geometries obtained after the MD simulations can be gauged by their energy relative to the lowest energy minima previously found (see entries I and II of Table 2). In the most stable state (entry I of Table 2), the interaction energy of the adsorbed A and E subdomains amounts to 3.44 and 3.71 MJ mol-1, respectively. These values, which are more than twice as large as the values found in the initial adsorption stage, correspond to 57 and 55 kJ mol-1 per amino acid, respectively, solely due to the dispersion (or van der Waals) forces. These interaction energies per amino acid should be compared with the value extrapolated from eq 1, namely, 71 kJ mol-1. The smaller figures actually found in the final stage should be attributed to the interaction of all the residues with the carbon surface, including also the hydrophilic ones. In fact, the interaction energy of the latter residues, though significant, is not as large as that shown by the hydrophobic ones, which dominate the initial stage. We also note that the strain energy of the adsorbed subdomains in the most stable state (see again entry I of Table 2) is much larger than what was found for the initial adsorption but is roughly consistent with eq 2. Anyway, it is not so large as to prevent the system from optimizing the interaction of all residues with the hydrophobic surface. The large rearrangements observed in the final adsorption stage greatly modify the size of the subdomains, and in particular their anisotropy. In fact, the radius of gyration Rg shows a pronounced increase compared to the isolated subdomains, unlike what happens in the initial

( )( ) ∏[ ( ) ]

∆)1-3

λi

∑ i*j R

g

S)

3

i

λi

Rg

2

λj

2

Rg

(3)

2

-1

(4)

The asphericity ∆ ranges from 0 for a sphere to 1 for a rod, while S ranges from -0.25 to 2, where a negative value indicates an oblate shape and a positive value a prolate one. In our case, the ratios λ1:λ2:λ3 of the principal axes are 2.0:1.2:1 and 2.7:1.7:1 for the isolated A and E subdomains, respectively (entry 1 of Table 3). Upon comparison with the asymptotic ratio of a random walk,30 3.48:1.65:1, we see that the original molecular shapes are quite isotropic, as expected for globular conformations. This feature is also indicated by the relatively small ∆ values. The initial adsorption on the surface modifies the molecular shape, but the change is not yet dramatic, as is shown by entry 2 of Table 3. Here in fact the ratios of the principal axes increase only to 2.6:1.6:1 and 3.4:2.4:1, and also the asphericity undergoes minor changes. On the other hand, after the MD runs the molecular anisotropy of the final adsorbed subdomains becomes huge, as indicated by the principal axes. In particular, λ3, perpendicular to the graphite surface, becomes negligibly small compared to the parallel axes λ1 and λ2, as shown by entry 3 of Table 3, the ratios λ1:λ2:λ3 becoming equal to 19.2: 9.2:1 and 17.3:11.6:1. Conversely, the asphericity and prolatness of the A subdomain become much larger than for the E subdomain. This fact is due to the more slender bidimensional shape of the former subdomain, as indicated by comparison of λ1 with λ2. We can also estimate the change in the contact surface area (the footprints of the subdomains) before and after the final adsorption stage. This contact area is most simply obtained through the difference between the area of the graphite surface accessible to the solvent before and after adsorption. In turn, the accessible surface, or Connolly surface, is evaluated through a spherical probe of radius 1.4 Å, mimicking a water molecule, rolling on the exposed van der Waals surface. Upon full spreading on the graphite plane, this area increases by a factor of about 2 in the A (29) Zifferer, G. J. Chem. Phys. 1998, 109, 3691-3698. (30) Zifferer, G. Macromol. Theory Simul. 1997, 6, 381-392.

3410

Langmuir, Vol. 19, No. 8, 2003

Raffaini and Ganazzoli

Table 4. Results after the Initial Energy Minimizations in Water According to the Water1 and Water2 Proceduresa subdomain A1- A2- A3 water1b pos 1 2 3 4 5 6 7 8

n5Å 12 20 20 13 11 15 15 21

broken Hintra

13 12 18 7 -5 9 4 12

Hinter 52 69 74 54 54 63 58 61

subdomain E2- E3- E4 water2c

RAA 32 27 25 17 38 20 30 13

n5Å 7 4 9 4 2 7 3

water1b

broken Hintra

Hinter

RAA

-2d

79 78 83 73 63 71 65 75

44 43 44 44 44 43 44 43

4 2 -3 -7 6

water2c

n5Å

broken Hintra

Hinter

RAA

12 16 2 21 14 14 29 17

3 14 2 10 6 -1 11 14

48 66 58 55 52 54 48 62

31 27 45 20 38 42 18 29

n5Å

broken Hintra

Hinter

RAA

5 4 1 1 2

-2 -4 -1 6 -2 -3

73 70 75 61 70 64 86 76

50 47 50 52 54 53 49 49

2 2

6

a

See text for details. We report the number of amino acids at a distance less than 5 Å from the plane, n5Å; the number of broken broken ; the number of intermolecular H-bonds with water molecules, Hinter; and the number of intramolecular H-bonds after adsorption, Hintra b amino acids still found in R-helices, RAA. Input geometry taken from the minimum found after energy minimization in the dielectric medium. c Input geometry taken from the experimental results. d A negative value actually means the formation of additional H-bonds.

subdomain, from 333 to 675 Å2, and by less than 50% in the E subdomain, from 419 to 604 Å2. These ratios should be compared with the experimental value of about 5 found for the initial spreading of albumin on a hydrophobic surface.4 Note that we might expect the whole protein to show an even larger increase of its footprint upon surface denaturation because of its overall size (see Figure 1). On the other hand, it should be realized that in addition to the intramolecular interactions still present (mainly hydrogen bonds), the network of intramolecular disulfide bridges hinders the full albumin spreading on the surface, thus limiting somewhat the contact area. 3. Simulations in the Explicit Presence of Water: (A) Initial Adsorption by Direct Energy Minimization. Let us first point out that in water the results of the energy minimizations always lead to relatively small changes of the starting geometries, in particular concerning the overall conformation. In fact, when the geometries previously obtained in the dielectric medium for the initial adsorption stage are further optimized in the presence of a large number of water molecules, they only show minor readjustments. These changes mainly involve the orientation of the side groups far from the carbon planes so as to maximize their solvation. This optimization procedure will be referred to as Water1 (see Table 4). Conversely, the analogous optimization of the experimental geometries close to the surface leads only to a very weak adsorption, if any, with little conformational changes. Accordingly, this alternative strategy, denoted as Water2, yields only higher-energy metastable states that maximize the intramolecular interactions within the subdomains and their hydration at the expenses of the interactions with the surface (see Table 4). Interestingly, the latter result is qualitatively consistent with recent theoretical estimates of the van der Waals interaction energy of proteins onto a surface in the presence of a very thin water layer separating them.31 Concerning the geometry of the adsorbed subdomains obtained according to the procedure above denoted as Water1, we found no significant conformational changes of the backbone induced by the solvent, as just said. Therefore, both the number of amino acids in contact with the surface, n5Å, and the number of amino acids still structured in R-helices, RAA, are essentially identical to what was found in the dielectric medium (see Tables 2 and 4). Thus, the size of the subdomains is basically unaffected by the presence of water molecules. In all cases Rg increases only by up to 0.2 Å in comparison with what found in the dielectric medium, and the principal axes (31) Zhdanov, V. P.; Kasemo, B. Langmuir 2001, 17, 5407-5409.

show a similar change. Therefore, the accessible surface area exposed to the solvent (the Connolly surface defined before) shows only a minor increase of about 35 Å2 for both subdomains. In particular, for the lowest-energy geometries the accessible surface area increases from 813 to 848 Å2 in the A subdomain and from 843 to 865 Å2 in the E subdomain. This area can be taken as proportional to the dispersive interaction energy with the solvent, but unfortunately it is not indicative of the solvation energy. In fact, it does not carry information about the electrostatic potential, so that it cannot account for the energy contribution due to the H-bonds. In fact, we did not find any satisfactory correlation between the accessible surface and the number of the H-bonds formed with the water molecules, although of course a larger surface still allows more room to the water molecules of the first hydration shell. Furthermore, the solvation and further energy minimization in water usually leads also to some changes in the number of the intramolecular H-bonds, that may correspond to either the formation or the disruption of a few H-bonds. Often, breaking some intramolecular Hbonds allows for more intermolecular H-bonds, i.e., for a better solvation. Therefore, to better describe the solvation of the subdomains, we carried out short MD runs of the most significant geometries in water and studied their hydration through the radial distribution function g(r) (or RDF for short) of the water molecules around the backbone. This function yields the density probability of finding water molecules at a distance r from the backbone, defined here through its NCOCR atoms. During the MD runs the frames used to calculate the RDF were saved every 80 fs. This time interval is much larger than the correlation time of the water molecules in the bulk (on the order of 10 fs), but it is short enough to prevent any largescale rearrangement of the backbone, though still allowing for minor readjustments of the R-helices induced by the solvent. We first show in Figure 7 the RDF calculated for the isolated subdomains in water. In panel a of the figure, we report the RDF of the oxygen atoms of the water molecules around the backbone for the whole subdomains. The first peak of the plot at about 2.7-2.8 Å corresponds to the first hydration shell: the similarity of the curves indicates a similar overall solvation, slightly more pronounced for the more hydrophilic A subdomain (see the higher peak in the lower panel). However, the three R-helices of each subdomain are differently hydrated, as shown in panel b of the same figure, because of their different hydrophilicity and their different exposure to water. Thus, the most hydrophilic helix A1 (see Table 1)

Interaction of Albumin Subdomains with Graphite

Figure 7. Radial distribution function g(r) of the water molecules around the backbone of the isolated subdomains in water. Solid symbols apply to the A subdomain, and open ones to the E subdomain. (a, lower panel) RDF of the oxygen atoms of the water molecules around the backbone of the whole subdomains. (b, upper panel) RDF of the oxygen atoms of the water molecules around the backbone for the individual R-helices indicated in the legend (see also Table 1). The plots for helices A2 and E4 are not shown for simplicity, their curves being intermediate between those displayed for the other helices of the corresponding subdomains. Note that the most hydrophilic helices A1 and E2 show the largest peaks.

Langmuir, Vol. 19, No. 8, 2003 3411

the first hydration shell show little changes compared to what was found for the isolated subdomains (see Figure 7a), both for their position and for their heights, apart from a small increase for the E subdomain. Therefore, even if part of the subdomains is not accessible to water, being in contact with the surface, the overall solvation shows only minor changes thanks to the deformations due to absorption, which expose more residues to the solvent. (B) Final Adsorbed State. Unfortunately, we could not run the long MD simulations leading to full spreading of the subdomains in water on large surfaces for simulation times of the order of 1 ns, due to the huge number of atoms to be accounted for. However, this is not a severe limitation in view of the overall similarity found in the dielectric medium and in water by the direct minimization procedure as discussed before. Instead, we performed shorter MD runs following the same protocol of the previous section in order to assess the hydration of the adsorbed monolayers in the most stable state depicted in Figure 6. The resulting RDF for both subdomains is reported in Figure 8 (upper curves). Upon comparison with the analogous plots for either the isolated subdomains (Figure 7) or the initial adsorbed states (Figure 8, lower curves), we see that the first peak, corresponding to the first hydration shell, keeps the same position but the amplitude is much higher. Accordingly, we conclude that the backbone is now fully exposed to solvent, so that the hydration of the subdomains is maximized. In other words, a larger number of water molecules are kept close to the backbone, though of course at the same distance as before. Moreover, the equal peak heights indicate a very similar overall solvation for the backbones of the two subdomains, even in the presence of a shoulder at 3.2 Å in the E subdomain due to a few small hydrophobic pockets that keep the water molecules farther away. In this context it is interesting to point out that while the peak heights are larger by about 60% than in the initial adsorption stage, the accessible (or Connolly) surface area shows a smaller increase by about 10-20%, to values of 885 and 1032 Å2 for the A and E subdomains. In keeping with what was said before, the Connolly surface may roughly measure the dispersive interactions with the solvent but is unable to accurately describe the intermolecular H-bonds, being insensitive to the electrostatic potential. Concluding Remarks

Figure 8. Radial distribution function g(r) of the water molecules around the backbone for the adsorbed subdomains. The two lower curves correspond to the initial adsorption stage obtained by direct energy minimization (Water1 procedure), and the two upper ones to the lowest energy minima found after long MD runs (1 ns) in the dielectric medium and subsequent short MD runs in water (see text).

shows the largest peak, followed by the slightly less hydrophilic helix E2. In full agreement with what was suggested by their hydropathy indices, these are the bestsolvated helices. The other helices, being more hydrophobic, show only weaker peaks, suggesting a poorer hydration due to a lower accessibility to solvent and/or to a larger mobility of the neighboring solvent molecules. We now turn to the adsorbed subdomains in the most stable geometry found by the direct minimization procedure and then optimized in water (Water1 procedure). The RDF plot of the oxygen of the water molecules around the whole backbone is shown in Figure 8 (lower curves). It is interesting to note that the peaks corresponding to

In the present paper we investigate the adsorption of two albumin subdomains on a graphite surface by atomistic computer simulations. The two subdomains were selected either because they contained strongly unlike hydrophilic and hydrophobic R-helices (A subdomain) or because they were more representative of the average hydropathy pattern of albumin (E subdomain). In both cases, we assumed all the ionizable side groups were in their electrically neutral form so as to simplify the present exploratory simulations. A first result of the present study is that the outcome of the simulations is affected by the adopted strategy, in particular concerning the medium. In fact, the direct energy minimization of the subdomains in water close to the carbon planes (the strategy called Water2) leads to very little adsorption on the surface, essentially with no conformational changes within the subdomains. Conversely, larger changes take place if the minimization is first carried out into a dielectric medium. Subsequent optimization in water of the resulting geometries (the Water1 procedure) does not bring about further changes,

3412

Langmuir, Vol. 19, No. 8, 2003

apart from some local readjustment of the side groups. In this case, the strands close to the surface lose the ideal R-helical geometry, no matter what their degree of hydropathy, even though often they keep some residual loose helicoidal features. This feature is quite general and involves even the hydrophilic R-helices interacting with the hydrophobic surface. However, as expected, the largest interaction energy is found for the adsorption of the hydrophobic helices (in particular helix A2). It is interesting to note that the initial optimizations in the dielectric medium show many widely different energy minima. Therefore, the configurational phase space of the adsorbed subdomains displays a rugged energy landscape, somehow reminiscent of a glassy state. However, the energy barriers separating these local minima from the lowest one are not prohibitively high and can be easily surmounted through a suitable kinetic energy input. Accordingly, the most stable state is readily found through long molecular dynamics runs at room temperature. In the global energy minimum, found after optimization of many selected snapshots, both the A and the E subdomains are fully denatured and form a monolayer on the carbon plane. Therefore, all amino acids do interact with the graphite surface. A similar extensive denaturation was recently reported for albumin on a hydrophobic surface (a C16 self-assembled monolayer) by different experimental techniques.4,10 Cooperative effects are important in bringing the hydrophilic residues as close to the surface as the hydrophobic ones. Note, however, that the latter residues show stronger interactions with the planes but even the hydrophilic residues do interact favorably. Additionally, they are still exposed to the solvent. The latter feature is clearly shown by the relatively large peaks at 2.7-2.8 Å of the radial distribution function g(r) reported in Figure 8. On the basis of the present simulations, we can attempt to extrapolate the present results to the behavior of the whole albumin onto the [0001] graphite surface. Thus, we propose that in the initial stage the protein adsorption is qualitatively independent from the nature of the approaching subdomain, i.e., of its hydropathy. This adsorption is accompanied by some loss of secondary structure on the surface, with relatively little rearrangement of the more distant strands, locally similar to the geometries found by our direct minimization procedure. After this initial stage, the whole molecule reorients and unfolds on the surface, spreading as much as possible with largescale rearrangements. In this way, it maximizes the number of amino acids in contact with the surface and eventually forms a thin layer, in fair agreement with recent experimental findings.4,9,10 This picture can also be carried one step further. The two-stage process we have just described corresponds to the adsorption on a clean surface. On the other hand, when most of the surface is already covered, new albumin molecules can only stick at the few

Raffaini and Ganazzoli

exposed graphite “islands” still available. Accordingly, they cannot spread to form a monomolecular layer32 and remain trapped with a loosely globular shape in the initial adsorption geometry, akin to what was found by our direct energy minimizations. The lower interaction energies of these states also suggests that they might show a reversible adsorption, unlike the fully denatured molecules.10 Note that the adsorption irreversibility is consistent with experimental results4,5,9,10 showing that the slow spreading of albumin over the surface (or relaxation, in the terminology of refs 4 and 5) eventually leads to a smaller coverage in terms of adsorbed mass of protein per unit surface due to the lower bare surface that remains available for other molecules. The latter feature can only be explained by irreversible adsorption of albumin. This adsorption pattern has obvious implications for albumin adhesion on the graphite surface under flow. With increasing shear rate, the loosely bound molecules can be easily removed by the applied stress, both because they show a weaker adsorption (a smaller Eint) and because they show a much larger cross section to flow. Conversely, the albumin molecules forming a monolayer do stick to the surface, even under large shear rates because of the larger Eint and the smaller cross section, and cannot be removed in practice. As a final comment, it should be remarked that in our description of the adsorption process of albumin on the graphite surface we implicitly ignored the possible adsorption and spreading of a second molecular layer on top of the first one. In this case the surface would be strongly modified, being made less hydrophobic by the first layer. Therefore, while we might expect the formation of the second layer to be of lesser importance, the issue should be addressed by appropriate simulations. Another relevant question is about the surface specificity. Even when we limit our attention to hydrophobic surfaces, the influence of their structure on the interaction energy and on the extent of the albumin spreading may be nontrivial, due for instance to the difference between a rigid graphite surface and a self-assembled monolayer exposing terminal methyl groups to the approaching protein. We believe that the main features should be essentially the same, although some quantitative differences may show up, for instance in the interaction energy. Appropriate experimental and simulation studies may help to answer such questions. Acknowledgment. This work was financially supported by Polytechnic of Milan, Project LSC (Large Scale Computing) and by MIUR (Italian Ministry for Instruction, University and Research). LA026853H (32) Talbot, J.; Tarjus, G.; Van Tassel, P. R.; Viot, P. Colloids Surf. A: Physicochem. Eng. Aspects 2000, 165, 287-324.