The Flexible Polyelectrolyte Hypothesis of Protein−Biomineral

pubs.acs.org/Langmuir © 2010 American Chemical Society

The Flexible Polyelectrolyte Hypothesis of Protein-Biomineral Interaction Graeme K. Hunter,*,† Jason O’Young,† Bernd Grohe,† Mikko Karttunen,‡ and Harvey A. Goldberg† †

School of Dentistry and ‡Department of Applied Mathematics, University of Western Ontario, London, Ontario, Canada Received January 27, 2010. Revised Manuscript Received May 7, 2010

Biomineralization is characterized by a high degree of control over the location, nature, size, shape, and orientation of the crystals formed. For many years, it has been widely believed that the exquisitely precise nature of crystal formation in biological tissues is the result of stereochemically specific interactions between growing crystals and extracellular matrix proteins. That is, the ability of many mineralized tissue proteins to adsorb to particular faces of biominerals has been attributed to a steric and electrical complementarity between periodic regions of the polypeptide chain and arrays of ions on the crystal face. In recent years, however, evidence has accumulated that many mineral-associated proteins lack periodic structure even when adsorbed to crystals. It also appears that protein-crystal interactions involve a general electrostatic attraction rather than arrays of complementary charges. In the present work, we review these studies and present some relevant new findings involving the mineral-modulating phosphoprotein osteopontin. Using molecular dynamics simulations, we show that the adsorption of osteopontin peptides to hydroxyapatite crystals does not involve a unique conformation of the peptide molecule, and that the adsorbed peptides are not aligned with rows of Ca2þ ions on the crystal face. Further, we show that the interface between osteopontin peptides and calcium oxalate monohydrate crystals consists of peptide regions of high electronegativity and crystal faces of high electropositivity. Collectively, the above-mentioned studies suggest that interactions between mineral-modulating proteins and biologically relevant crystals are primarily electrostatic in nature, and that molecular disorder assists these proteins in forming multiple bonds with cations of the crystal face.

Introduction Biomineralization is the process by which living organisms produce hard tissues such as bones, teeth, and shells.1 Although the mechanisms by which organisms generate mineral crystals are not well understood, there is widespread belief that proteins play important roles.2-4 Thus, biomineralization can be seen as an interfacial phenomenon in which (usually extracellular) proteins interact with nascent crystals in order to control their growth. Although the proteins involved are often either not known or insufficiently characterized, the existence of a highly specific protein-crystal interaction seems necessary to account for phenomena such as the different growth habit of hydroxyapatite crystals in dental enamel and the adjacent dentin, the brick wall-like structure of mollusk nacre and the elaborate single crystals of coccolith exoskeletons. In early studies, most attention was paid to the question of how crystal formation is initiated in mineralized tissues. As physiological fluids generally will not support spontaneous precipitation of biominerals, a nucleating agent is required to overcome the freeenergy barrier to crystal formation.5 The isolation of highly anionic proteins from bone, tooth, and shell suggested a possible means by which proteins could induce biomineral formation under physiological conditions. As early as 1967, Veis and Perry proposed that the highly phosphorylated protein they had isolated *Corresponding author. Dr. G. K. Hunter, School of Dentistry, University of Western Ontario, London, ON, N6A 5C1, Canada. Tel: 519 661 2185. E-mail: [email protected]. (1) Mann, S. Biomineralization: Principles and Concepts in Bioinorganic Materials Chemistry; Oxford University Press: Oxford, 2001. (2) Addadi, L.; Weiner, S.; Geva, M. Z. Kardiology 2001, 90(Suppl 3), 92–8. (3) De Yoreo, J. J.; Dove, P. M. Science 2004, 306(5700), 1301–2. (4) George, A.; Veis, A. Chem. Rev. 2008, 108(11), 4670–4693. (5) Williams, R. J. P. Philos. Trans. R. Soc. London, Sect. B 1984, 304, 411–424.

Langmuir 2010, 26(24), 18639–18646

from dentin “may provide the sites for the epitactic nucleation of mineralization of the matrix”.6 By drawing an analogy with epitaxy (the nucleation of a crystal by a crystal with related lattice structure), Veis and Perry implied that the dentin phosphoprotein nucleates hydroxyapatite (HA) because the distribution of its negatively charged side-chains approximates the spacing of Ca2þ ions in the crystal. As acidic proteins were isolated and purified from bone, enamel, shells, and other mineralized tissues, the concept of proteins as epitactic nucleators became widely adopted. In 1989, Glimcher wrote: “It is now generally accepted that the underlying physicochemical basis and mechanism for the formation of crystals of Ca-P in bone and other tissues, and in biological mineralization, in general, is a heterogeneous nucleation by one or more of the organic constituents in the tissue”.7 Studies on the effects of mineralized-tissue proteins on the growth habit (size and shape) of biominerals led to the proposal of a more specific mechanism of protein-crystal interactions. Using the principle that a protein adsorbing to the {hkl} faces of a crystal will inhibit growth of the crystal in the Æhklæ directions, Addadi and Weiner were able to identify the faces of calcium dicarboxylate crystals to which a preparation of acidic proteins from the bivalve mollusk Mytilus californianus adsorb. For each crystal studied, the carboxylate groups of the anion were oriented perpendicular to the face of interaction. If the mollusk proteins primarily adopted the β-sheet conformation, as was believed, carboxylate groups of protein and crystal would therefore be parallel to one another. Addadi and Weiner believed that this facilitated protein-crystal interaction because these groups were optimally oriented to complete coordination polyhedra around Ca2þ ions that formed the interface between the two components. (6) Veis, A.; Perry, A. Biochemistry 1967, 6, 2409–2416. (7) Glimcher, M. J. Anat. Rec. 1989, 224, 139–153.

Published on Web 06/08/2010

DOI: 10.1021/la100401r

18639

Invited Feature Article

In the case of calcite crystals, the protein-adsorbing face was shown to be {001}, to which the (planar) carbonate ions are parallel. If the mollusk-shell proteins were adsorbed to the surface of a plastic dish, calcite crystals were nucleated from an {001} face. Again, this suggested a role for stereochemistry, as the optimal orientation of a carbonate ion (as opposed to a carboxylate group) for coordination of Ca2þ ions bound to a β-sheet is parallel to the protein-adsorbing face. Importantly, Addadi and Weiner noted that neither the charge density of the crystal face nor the spacing of Ca2þ ions in the lattice could account for the observed specificity of mollusk-protein interaction with calcite and calcium dicarboxylate crystals. That is to say, the preferred faces for protein adsorption were not necessarily those of highest positive-charge density and often had Ca2þ-Ca2þ spacings similar to those of non-preferred faces.8 The stereochemical theory was supported by other studies showing that ordered arrays of charged groups can nucleate crystals from specific faces. For example, Mann and co-workers showed in 1988 that monolayers of the fatty acid stearic acid can cause the formation of vaterite crystals in solutions that favor the formation of calcite, a more stable calcium carbonate polymorph. The vaterite crystals were nucleated from (001) faces, which are entirely composed of Ca2þ ions. Further, the carbonate ions of vaterite are perpendicular to the (001) face, and therefore parallel to the carboxylate groups of the stearic acid monolayer.9 On the basis of studies of this sort, Mann concluded that “organized organic surfaces can control the nucleation of inorganic materials by geometric, electrostatic, and stereochemical complementarity between incipient nuclei and functionalized substrates”.10 Because phosphate groups are tetrahedral, the stereochemistry of protein interaction with calcium phosphate crystals would be expected to differ from that of the calcium carbonates. By studying the interaction of mollusk shell proteins and dentin phosphoprotein with calcium phosphate ester crystals, Moradian-Oldak et al. found that the preferred faces for adsorption of protein had an orientation of phosphate groups in which two oxygen atoms are in the plane of the face. It was pointed out that this characteristic is shared by the {100} faces of HA.11 Collectively, the above-mentioned studies and others like them12,13 provide a general mechanism of protein-crystal interaction or at least one that describes face-specific nucleation and inhibition phenomena. This mechanism, which we will refer to as the “strong” stereochemical theory, has three postulates: (1) The crystal-binding region of the protein has a planar surface with charged groups projecting perpendicular to the plane (essentially this defines a β-pleated sheet). (2) The spacing of the charged groups of the protein is commensurable with the spacing of cations in a lattice plane of the crystal. (3) The protein-binding face of the crystal has a particular orientation of anionic groups that optimizes the coordination of the cations shared with the protein. More recently, other workers have proposed mechanisms of protein-crystal interaction that do not meet all the requirements of the strong stereochemical theory, either because the crystalbinding domain of the protein is not in a β-pleated sheet conformation or because the orientation of anionic groups in (8) Addadi, L.; Weiner, S. Proc. Natl. Acad. Sci. U.S.A. 1985, 82, 4110–4114. (9) Mann, S.; Heywood, B. R.; Rajam, S.; Birchall, J. D. Nature 1988, 334(6184), 692–695. (10) Mann, S.; Archibald, D. D.; Didymus, J. M.; Douglas, T.; Heywood, B. R.; Meldrum, F. C.; Reeves, N. J. Science 1993, 261, 1286–1292. (11) Moradian-Oldak, J.; Frolow, F.; Addadi, L.; Weiner, S. Proc. R. Soc. London, Sect. B 1992, 247, 47–55. (12) Sicheri, F.; Yang, D. S. Nature 1995, 375(6530), 427–31. (13) Teng, H. H.; Dove, P. M. Am. Mineral. 1997, 82, 878–887.

18640 DOI: 10.1021/la100401r

Hunter et al.

the crystal face is not optimal for coordination of shared cations, or both.14-17 These studies are, however, consistent with a “weak” version of the stereochemical theory, which retains postulate 2 of the strong theory but discards postulates 1 and 3. For the purposes of the present discussion, we will define the weak stereochemical theory as follows: proteins that interact with specific faces of biomineral crystals have arrays of anionic groups that are complementary to arrays of cations on those faces. However, the past decade has produced a large number of findings that are incompatible with either version of the stereochemical theory. These studies found that many mineral-modulating proteins do not have the periodic structures required to match a crystal lattice either in solution or when adsorbed to the crystal. Further, evidence has accumulated that face-specific adsorption of proteins to crystals is dictated by the electrostatic potential of the face rather than its orientation of anionic groups, as proposed in the strong stereochemical theory, or its cationcation spacing, as proposed by both versions of the theory. Thus, the current state of knowledge suggests a mechanism of proteincrystal interaction in which the key features are the lack of folded structure in the crystal-binding region of the protein and the electrostatic potential difference between this region and the crystal face. We will refer to this as the flexible polyelectrolyte hypothesis.

Conformation of Mineral-Modulating Proteins in Solution In a recent articulation of the strong stereochemical theory, Addadi and co-workers wrote: “In principle in this type of interaction, protein side-chain groups identical to one of the components of the crystal and spaced at the correct distance along the backbone can precisely match crystal lattice positions on one crystal surface”.2 The crystal binding “site” of the protein must therefore possess secondary structure, since only these structures, by definition, exhibit a periodic distribution of amino acid side-chains. In practice, the β-pleated sheet is the only type of secondary structure that fits the bill, as this presents a planar (or curved) surface with half the side-chains projecting perpendicular to each face of the sheet. In a β-sheet, side-chains are spaced 0.68 nm apart along the polypeptide backbone and 0.5 nm apart in neighboring polypeptide chain segments. If the side-chains have the required functional groups (carboxylate or phosphate) and their spacings match the distances between Ca2þ ions in a crystal face with the appropriate stereochemistry, it is easy to see how the protein could adsorb to the crystal face with high affinity and specificity. Note that free rotation around side-chain bonds means that the positions of charged groups are not exactly specified in a β-sheet. According to the weak stereochemical theory, however, any periodic (secondary) structure could provide a crystal binding site. For example, although the β-strand is less stable than the β-pleated sheet, it has the advantage that it can be oriented such that side-chains on both “sides” of the strand interact with crystal ions. Mechanisms involving β-strands have been proposed for the adsorption of dentin phosphoprotein and poly(L-aspartic acid) to hydroxyapatite and calcite, respectively (see below). Similarly, the R-helix could also present a series of charges complementary to a crystal lattice. However, because the R-helix has a nonintegral (14) Wierzbicki, A.; Sikes, C. S.; Madura, J. D.; Drake, B. Calc. Tissue Int. 1994, 54, 133–141. (15) Fujisawa, R.; Kuboki, Y.; Eur., J. Oral Sci. 1998, 106(Suppl 1), 249–53. (16) Hoang, Q. Q.; Sicheri, F.; Howard, A. J.; Yang, D. S. Nature 2003, 425 (6961), 977–80. (17) Ndao, M.; Ash, J. T.; Breen, N. F.; Goobes, G.; Stayton, P. S.; Drobny, G. P. Langmuir 2009, 25(20), 12136–12143.

Langmuir 2010, 26(24), 18639–18646

Hunter et al.

number of amino acids per turn, side-chains spaced equally along the polypeptide chain do not lie on the same face of the helix. Nonetheless, R-helical domains have been associated with the crystal-binding activities of osteocalcin, statherin, and the artificial peptide JAK1 (see below). Early studies reported that proteins from the shells of a variety of marine invertebrates not only exhibit extensive β-sheet in the presence of Ca2þ but also preferential orientation with respect to the crystallographic axes of the associated mineral phase.18 Because the proteins involved (together with the polysaccharide chitin) represent the “structural matrix” of the shell (analogous to the type I collagen fibrils of bone), the physiological significance of this coalignment is not clear. However, β-sheet structure was also found in “enamelins”, acidic proteins of dental enamel that are thought to interact directly with the mineral crystals.19 More recent investigations have shown that mineral-associated proteins exhibit a striking lack of secondary structure. Studies on osteopontin (OPN), a phosphoprotein that has been shown to inhibit the formation of calcium carbonates, phosphates, and oxalates, typify the evolution of ideas on mineralized-tissue protein structure. In 1989, Prince et al. analyzed the sequence of rat OPN using five different secondary structure prediction algorithms and found approximately 40% R-helix and 10% β-structure. Circular dichroism (CD) spectra of the rat bone OPN were typical of random coil, but this was attributed to denaturation of the protein by the solvents used in extraction and purification.20 CD analysis of the milk isoform of OPN, which can be purified without denaturants, found a high content (>50%) of β-pleated sheet and negligible R-helix.21 More recently, however, a study using nuclear magnetic resonance (NMR) spectroscopy reported that recombinant OPN is largely disordered.22 Although the discrepancies between the CD and NMR studies remain unresolved, it has become clear that secondary structure prediction tools are of limited utility for proteins like OPN, because these algorithms are based on globular proteins with amino acid compositions very different from those of mineralized tissue proteins. Lack of folded structure is also a feature of mineralized tissue proteins believed to nucleate biominerals, such as bone sialoprotein (BSP) and dentin phosphoprotein (DPP). Secondary structure prediction algorithms suggest that BSP contains a significant amount of R-helix and β-pleated sheet.23 However, no ordered structure was found by NMR spectroscopy,22,24 small-angle X-ray diffraction, or CD.25 Analysis of DPP by CD showed that the protein has a random-coil structure. Addition of Ca2þ “converts it to a more ordered structure, but...the transition does not go entirely to a β-sheet”.26 A study using NMR spectroscopy found that at very low pH DPP has a “condensed conformation”, but at physiological pH, the protein adopts a “global extended structure”.27 A more recent NMR study concluded that DPP “is an intrinsically unstructured or disordered protein”.28 (18) Worms, D.; Weiner, S. J. Exp. Zool. 1986, 237, 11–20. (19) Jodaikin, A.; Weiner, S. Int. J. Biol. Macromol. 1987, 9, 166–168. (20) Prince, C. W. Connect. Tissue Res. 1989, 21, 15–20. (21) Kaartinen, M. T.; Pirhonen, A.; Linnala-Kankkunen, A.; M€aenp€a€a, P. H. J. Biol. Chem. 1999, 274, 1729–1735. (22) Fisher, L. W.; Torchia, D. A.; Fohr, B.; Young, M. F.; Fedarko, N. S. Biochem. Biophys. Res. Commun. 2001, 280, 460–465. (23) Ganss, B.; Kim, R. H.; Sodek, J. Crit. Rev. Oral Biol. Med. 1999, 10(1), 79– 98. (24) Wuttke, M.; Muller, S.; Nitsche, D. P.; Paulsson, M.; Hanisch, F. G.; Maurer, P. J. Biol. Chem. 2001, 276(39), 36839–48. (25) Tye, C. E.; Rattray, K. R.; Warner, K. J.; Gordon, J. A.; Sodek, J.; Hunter, G. K.; Goldberg, H. A. J. Biol. Chem. 2003, 278(10), 7949–55. (26) Lee, S. L.; Veis, A.; Glonek, T. Biochemistry 1977, 16(13), 2971–9. (27) Evans, J. S.; Chan, S. I. Biopolymers 1994, 34(4), 507–27. (28) Cross, K. J.; Huq, N. L.; Reynolds, E. C. J. Pept. Res. 2005, 66(2), 59–67.

Langmuir 2010, 26(24), 18639–18646

Invited Feature Article

Proteins from calcium carbonate-containing tissues tend to be less well-characterized than those of mammalian bones and teeth. However, the studies of Evans and colleagues have shown an absence of secondary structure in a variety of such proteins. Using CD and NMR spectroscopy, it was shown that the mineralbinding domains of AP7, AP24, and lustrin, proteins from the nacreous layer of red abalone, are highly disordered.29,30 Using a bioinformatics approach, it was shown that all ten members of the asprich (aspartic acid-rich) family of mollusk shell proteins are disordered.31 In a 2003 review, Evans wrote: “In those biomineralization proteins where a mineral interaction domain has been identified, the mineral interaction domain apparently adopts an extended structure or ‘random coil’ in solution”.32 Therefore, theoretical and experimental considerations appear to converge in concluding that mineralized tissue proteins are exceptionally disordered, at least in their crystal binding domains. In fact, many of these proteins, including BSP, OPN, DPP, and asprich, have all the features of intrinsically unordered proteins: high content of charged amino acids, low content of hydrophobic residues, and high sequence redundancy.33

Conformations of Crystal-Bound Proteins As described in the previous section, a large body of data suggests that mineral-modulating proteins from hard and soft tissues are largely or completely unordered in solution. Another possibility is that that these proteins are unordered in the nonmineralized extracellular matrix but, upon adsorbing to a crystal, adopt a periodic structure such as a β-pleated sheet. Strictly speaking, this is incompatible with both strong and weak versions of the stereochemical theory, which require that a conformation matching the crystal lattice is present prior to adsorption. However, conformational analysis of mineral-bound proteins does provide a means of distinguishing between stereochemical and flexible polyelectrolyte mechanisms. The main experimental approach used to address these issues is solid-state NMR, which permits the measurement of distances between specific C or N atoms in proteins adsorbed to crystals. As secondary structures are characterized by specific distances between backbone atoms, insight into protein conformation can be obtained from this approach. The salivary protein statherin is amenable for such studies because it adsorbs tightly to HA through its highly anionic N-terminus (DpSpSEE). In earlier studies, Drobny, Stayton, and co-workers found that this sequence exhibits little or no order, either in solution or on the crystal surface.34 Subsequently, however, these workers reported that HA-binding domain of statherin is either R-helical35 or a distorted R-helix.17 Chen et al. described two modes of statherin-HA interaction, the principal one involving an R-helical N-terminus.36 Another protein amenable to solid-state NMR analysis is leucine-rich amelogenin protein (LRAP), a splice variant of the enamel protein amelogenin consisting of the relatively hydrophilic (29) Michenfelder, M.; Fu, G.; Lawrence, C.; Weaver, J. C.; Wustman, B. A.; Taranto, L.; Evans, J. S.; Morse, D. E. Biopolymers 2003, 70(4), 522–33. (30) Wustman, B. A.; Weaver, J. C.; Morse, D. E.; Evans, J. S. Connect. Tissue Res. 2003, 44(Suppl 1), 10–5. (31) Delak, K.; Collino, S.; Evans, J. S. Biochemistry 2009, 48(16), 3669–77. (32) Evans, J. S. Curr. Opin. Colloid Interface Sci. 2003, 8, 48–54. (33) Dyson, H. J.; Wright, P. E. Nat. Rev. Mol. Cell Biol. 2005, 6(3), 197–208. (34) Long, J. R.; Dindot, J. L.; Zebroski, H.; Kihne, S.; Clark, R. H.; Campbell, A. A.; Stayton, P. S.; Drobny, G. P. Proc. Natl. Acad. Sci. U.S.A. 1998, 95, 12083– 12087. (35) Long, J. R.; Shaw, W. J.; Stayton, P. S.; Drobny, G. P. Biochemistry 2001, 40(51), 15451–5. (36) Chen, P. H.; Tseng, Y. H.; Mou, Y.; Tsai, Y. L.; Guo, S. M.; Huang, S. J.; Yu, S. S. F.; Chan, J. C. C. J. Am. Chem. Soc. 2008, 130(9), 2862–2868.

DOI: 10.1021/la100401r

18641

Invited Feature Article

N- and C-termini of the parent molecule. NMR studies showed that LRAP has no detectable secondary structure, either in solution or when adsorbed to HA. The authors concluded: “the current results are most consistent with an electrostatic interaction mechanism...rather than a lattice-matching mechanism”.37 Although it provides less specific information, CD can also be used to probe the conformations of mineral-bound proteins. Capriotti et al. described JAK1, a peptide designed to adsorb to the {100} face of HA in an R-helical conformation. The key to this activity was the presence of γ-carboxyglutamic acid at positions n, n þ 4, and n þ 7, which lie on the same “side” of the R-helix. JAK1 is disordered in solution, but on addition to HA becomes R-helical.38 Because of the paucity of physical techniques to characterize protein-crystal interaction, molecular modeling has been extensively employed. The major modeling approaches used are: static docking, in which a known (or assumed) molecular conformation is juxtaposed with a crystal lattice of interest;15,16 energy minimization, in which numerous possible conformations of the polypeptide are sampled in an attempt to find the mode of adsorption of lowest energy;39,40 and classical molecular dynamics simulations, in which the system is allowed to evolve, under the influence of a force field, to a stable state.41,42 Ab initio/ density functional computations have been used to study specific interactions between selected atoms,43 and structure predictionbased approaches, such as RosettaSurface,44 have also been applied.45 In some cases, information from several of these methods has been combined for a more complete picture.46 There are several issues that make computational modeling of protein adsorption to crystal surfaces a challenge. First, it is difficult to model most proteins due to their large size in comparison with computationally accessible system sizes; in classical molecular dynamics, systems up to roughly 10  10  10 nm3 can be studied.47 However, larger molecules can be “coarse-grained”, that is, simulated at lower resolution.48 Second, simulations of biominerals rarely include imperfections, such as step-edges, dislocations, and vacancies, that are present in real crystals. Third, the time scales needed in simulations of proteincrystal interactions should be long enough, preferably at least 100 ns, for the system to reach at least quasi-equilibrium. Of course, even this is much shorter than crystal growth processes. Fourth, electrostatic interactions must be treated carefully; even in membranes, where one may naı¨ vely expect strong screening, both static and dynamic properties may be very strongly influenced.49 (37) Shaw, W. J.; Ferris, K.; Tarasevich, B.; Larson, J. L. Biophys. J. 2008, 94(8), 3247–57. (38) Capriotti, L. A.; Beebe, T. P., Jr.; Schneider, J. P. J. Am. Chem. Soc. 2007, 129(16), 5281–7. (39) De Leeuw, N. H.; Cooper, T. G. Cryst. Growth Des. 2004, 4, 123–133. (40) Chien, Y. C.; Masica, D. L.; Gray, J. J.; Nguyen, S.; Vali, H.; Mckee, M. D. J. Biol. Chem. 2009, 284(35), 23491–23501. (41) Grohe, B.; O’Young, J.; Ionescu, A.; Lajoie, G.; Rogers, K. A.; Karttunen, M.; Goldberg, H. A.; Hunter, G. K. J. Am. Chem. Soc. 2007, 129, 14946–14951. (42) Zhang, H. P., Lu, X., Fang, L. M., Qu, S. X., Feng, B., Weng, J. Biomed. Mater. 2008, 3, art. number 044110. (43) Bhowmik, R.; Katti, K. S.; Venna, D.; Katti, D. R. Mater. Sci. Eng. C 2007, 27(3), 352–371. (44) Makrodimitris, K.; Masica, D. L.; Kim, E. T.; Gray, J. J. J. Am. Chem. Soc. 2007, 129(44), 13713–13722. (45) Addison, W. N.; Masica, D. L.; Gray, J. J.; McKee, M. D. J. Bone Miner. Res. 2010, 25, 695–705. (46) Freeman, C. L.; Asteriadis, I.; Yang, M. J.; Harding, J. H. J. Phys. Chem. C 2009, 113(9), 3666–3673. (47) Murtola, T.; Bunker, A.; Vattulainen, I.; Deserno, M.; Karttunen, M. Phys. Chem. Chem. Phys. 2009, 11(12), 1869–1892. (48) Vattulainen, I.; Karttunen, M. In Handbook of Theoretical and Computational Nanotechnology, Rieth, M., Schommers, W., Eds.; American Scientific Publishers: New York, 2004. (49) Patra, M.; Karttunen, M.; Hyvonen, M. T.; Lindqvist, P.; Falck, E.; Vattulainen, I. Biophys. J. 2003, 84, 3636–3645.

18642 DOI: 10.1021/la100401r

Hunter et al.

A detailed overview of the topic, including the treatment of nonperiodic systems, is given in ref 50. Several reviews on computational issues in biomineralization have recently appeared.51,52 Despite their limitations, computational methods offer several advantages. First and foremost, they enable studies of the fundamental interactions without disturbing the system, as one knows all the positions and momenta of all particles at all times. This has proven to be very valuable, for example, in studies of the dynamics of biological systems.53 Docking Studies. In an important early study, static docking was used to interpret the effects of mineralized tissue proteins on the growth of octacalcium phosphate (OCP) crystals. It was concluded that aspartic acid-rich mollusk shell proteins interact with the {100} faces of OCP, whereas dentin phosphoprotein interacts with the {010} faces of OCP (which has similar structure to the {100} face of HA).54 In an even earlier study, Hauschka and Wians noted that the predicted structure of the bone protein osteocalcin contains two R-helical sequences with carboxylatecontaining side-chains projecting to the same side of the molecule. The spacing between these side-chains was found to be “highly complementary” to that of Ca2þ ions in the {001} plane of HA.55 The presence of the two R-helices was confirmed when the structure of osteocalcin was solved by X-ray diffractometry. However, docking analysis found that the crystal face of HA best matching this region of osteocalcin was {100}, not {001}.16 On the basis of a solid-state NMR study showing that dentin phosphoprotein adsorbed to HA was in an extended, β-type structure, Fujisawa and Kuboki presented a model in which aspartic acid-rich sequences of the protein interact with the {100} face of HA.15 According to the model proposed, the polypeptide is in a “nonplanar folded trans-extended chain” (i.e., a β-strand) parallel to the crystallographic c-axis. Aspartic acids projecting from opposite sides of the chain can bond to two rows of Ca2þ ions. In this conformation, the side-chain carboxylates are at an acute angle to the crystal face, and therefore do not fulfill the requirements of the strong stereochemical theory, in which the carboxylates should be perpendicular to the face. Energy-Minimization Studies. An energy-minimization study of the interaction between poly(L-aspartic acid) (poly asp; a model compound for aspartic acid-rich mollusk-shell proteins) and the (110) face of calcite found favorable modes of interaction with the polypeptide oriented either parallel or perpendicular to the crystallographic c-axis. In the former case, every fifth sidechain carboxylate group was driven away from the crystal surface; in the latter case, carboxylate groups oriented to one side of the polypeptide formed closer contacts with crystal Ca2þ ions than those on the opposite side.14 For both orientations, the calcitebound poly asp molecule was in an extended conformation. As in the case of the DPP-HA interaction described above, the orientation of the side-chains does not fulfill the requirements of the strong stereochemical theory. In addition, the orientation of carbonate ions in the (110) face of calcite is not consistent with this version of the theory. Also using energy minimization, Huq and co-workers found that the most stable conformations of a phosphopeptide adsorbed (50) Karttunen, M.; Rottler, J.; Vattulainen, I.; Sagui, C. Curr. Top. Membr. 2008, 60, 49–89. (51) Raffaini, G.; Ganazzoli, F. Macromol. Biosci. 2007, 7(5), 552–66. (52) Harding, J. H.; Duffy, D. M.; Sushko, M. L.; Rodger, P. M.; Quigley, D.; Elliott, J. A. Chem. Rev. 2008, 108, 4823–4854. (53) Repakova, J.; Holopainen, J. M.; Karttunen, M.; Vattulainen, I. J. Phys. Chem. B 2006, 110(31), 15403–15410. (54) Furedi-Milhofer, H.; Moradian-Oldak, J.; Weiner, S.; Veis, A.; Mintz, K. P.; Addadi, L. Connect. Tissue Res. 1994, 30, 251–264. (55) Hauschka, P. V.; Wians, F. H., Jr. Anat. Rec. 1989, 224(2), 180–8.

Langmuir 2010, 26(24), 18639–18646

Hunter et al.

to the {001} face of HA involved bends in the backbone. Most notably, constraining the peptide into a β-strand conformation greatly reduced the binding energy.56 The adsorption to calcium oxalate dihydrate crystals of an aspartic acid-rich peptide based on a sequence in OPN (DDLDDDDD) was examined using RosettaSurface. This peptide was predicted to be unstructured in solution and capable of adsorbing to the {110} face in multiple conformations.40 Molecular Dynamics Studies. Docking and energy minimization approaches are often predicated upon a particular conformation of the protein and therefore may not be appropriate for the study of mineralized tissue proteins, which, as described above, generally lack ordered structures. As computing resources have increased, dynamic modeling methods have become preferred, as these provide information about pathways to equilibria and multiple energy minima. Ideally, such simulations should be complemented by free-energy calculations. A molecular dynamics analysis of the adsorption to calcite of an undecapeptide of the protein lithostatine found that the peptide unfolded as it approached the crystal surface. The adsorbed conformations of the lithostatine peptide had kinked backbones. On the basis of these findings, it was concluded that “The ability of peptides to inhibit crystal growth is therefore essentially based on backbone flexibility”.57 Our molecular dynamics study on the adsorption of OPN peptides to the {100} face of COM indicated that peptide-crystal contacts form sequentially. A monophosphorylated peptide corresponding to amino acids 220-235 of rat OPN initially adsorbs to the {100} face of COM via its C-terminus and only subsequently at the N-terminus. At this point, the central part of the peptide is looped out from the crystal face and highly mobile. Eventually, the central region also adsorbs stably to the face (see Supporting Information to ref 41). As described below, we also used molecular dynamics to simulate the interaction between OPN peptides and the {100} face of HA.58 One of the virtual peptides interacting most strongly with HA corresponds to amino acids 65-80 of rat bone OPN (pSHDHMDDDDDDDDDGD). Its adsorption to the {100} face was subsequently studied in detail. To test the effect of peptide orientation on adsorption, OPN65-80 was rotated 60 around the polypeptide chain axis between successive simulations, for a total of six starting orientations. As shown in Figure 1, all six starting orientations result in different conformations of the peptide on the {100} face of HA. In none of these was the polypeptide backbone aligned with any row of Ca2þ ions on the crystal face. In fact, the backbone was always kinked in one or more places (Figure 1). This finding shows that there are many modes of adsorption that result in a stable interaction. From such studies, two aspects of the adsorption process appear clear. First, backbone flexibility assists the peptide in making enough contacts for a stable interaction. Second, there are multiple possible conformations of the peptide on the crystal surface that will result in strong adsorption.

Role of Electrostatics in Protein-Crystal Interactions It is generally assumed that interactions between mineralized tissue proteins and biominerals involve anionic groups of the (56) Huq, N. L.; Cross, K. J.; Reynolds, E. C. J. Mol. Modeling 2000, 6(2), 35– 47. (57) Gerbaud, V.; Pignol, D.; Loret, E.; Bertrand, J. A.; Berland, Y.; FontecillaCamps, J. C.; Canselier, J. P.; Gabas, N.; Verdier, J. M. J. Biol. Chem. 2000, 275(2), 1057–64. (58) Azzopardi, P. V.; O’Young, J.; Lajoie, G.; Karttunen, M.; Goldberg, H. A.; Hunter, G. K. PLoS One 2010, 5, e9330.

Langmuir 2010, 26(24), 18639–18646

Invited Feature Article

former and Ca2þ ions of the latter and thus are basically electrostatic in nature. However, interactions involving polyelectrolytes can manifest in unexpected ways, such as like-charge attractions between DNA and the lipid bilayers of biological membrane.59 In addition, Addadi and Weiner’s 1985 study on calcium dicarboxylate and calcite crystals found that the preferred face for adsorption of mollusk shell proteins rarely corresponded to the most Ca2þ-rich lattice plane.8 Therefore, the nature and specificity of electrostatic interactions between proteins and crystals is an important question in biomineralization. To address these questions, we examined the adsorption of differentially phosphorylated peptides to different faces of COM crystals. The peptides corresponded to amino acids 220-235 of rat bone OPN and contained phosphorylations at none, one, or all three of the serines that can be modified in vivo. These peptides were labeled with a fluorescent tag and added to preformed COM crystals that had {100}, {010}, and {121} faces developed. All three peptides adsorbed preferentially to {100} faces.41 In an attempt to understand the specificity of peptide adsorption to COM, we have generated electrostatic potential maps of the three crystal faces present (Figure 2, panels A-C). It can easily be seen that the {100} face is much more electropositive than the {010} or {121} face. Therefore, an obvious interpretation of the observed preference of anionic peptides for the {100} face is that the interaction is predominantly electrostatic. In agreement with this conclusion, Chien et al. recently reported that an aspartic acid-rich peptide of OPN adsorbs preferentially to the Ca2þ-rich {110} face of calcium oxalate dihydrate.40 Our study on phosphopeptide adsorption to COM provides an important test of the strong stereochemical theory, which postulates that crystal faces preferentially interacting with anionic proteins should have carboxylate groups oriented perpendicular to the face. COM contains two classes of oxalate ions: the carboxylates of oxalate 1 ions are perpendicular to the {010} face and parallel to the {100} face; the carboxylates of oxalate 2 ions are parallel to {010} and at an oblique angle to {100}.60 Therefore, {010} should be the preferred protein-binding face of COM. However, it is in fact the face to which OPN peptides adsorb least well.41 Figure 2 also shows electrostatic potential maps of the nonphosphorylated (P0), monophosphorylated (P1), and triphosphorylated (P3) peptides of OPN220-235 adsorbed to the {100} face of COM (panels D-F). In each case, the side of the peptide adjacent to the crystal has negative potential, while any areas of positive potential are on the opposite (solution) side of the peptide. This analysis again suggests that the interaction between OPN peptide 220-235 and the {100} face of COM is predominantly electrostatic in nature. Studies on the effects of OPN peptides on COM crystal growth also suggest that the interaction is electrostatic rather than stereochemical. P3 is a much more potent inhibitor of COM growth than P1, which in turn is more potent than P0.41 More recently, we have used molecular dynamics to study the adsorption to the {100} face of HA of 19 (virtual) peptides covering the entire sequence of rat bone OPN. The average peptide-crystal distance over the final 2 ns of the simulation showed a good correlation with peptide isoelectric point. That is to say, peptides with lowest isoelectric points were found to adsorb most strongly to the {100} face of HA. This relationship was validated by showing that the ability of synthetic peptides to (59) Radler, J. O.; Koltover, I.; Salditt, T.; Safinya, C. R. Science 1997, 275 (5301), 810–4. (60) Echigo, T.; Kimata, M.; Kyono, A.; Shimizu, M.; Hatta, T. Mineral. Mag. 2005, 69(1), 77–88.

DOI: 10.1021/la100401r

18643

Invited Feature Article

Hunter et al.

Figure 1. Effect of peptide OPN65-80 orientation on its crystal-bound conformation. Molecular dynamics simulations of OPN65-80

(pSHDHMDDDDDDDDDGD) adsorption to the {100} face of hydroxyapatite were performed as previously described.58 Explicit water (simple point charge) and ions were used. The peptide was rotated 60 around its long axis between each of panels A-F. Crystal: Ca - green, O - red, P - orange. Peptide: C - gray, H - white, O - pink, N - purple, P - orange, S - yellow.

inhibit the growth of HA seed crystals also correlated with isoelectric point.58 In a previous study, we demonstrated that synthetic OPN peptides with high negative charge density were better inhibitors of HA nucleation than peptides of lower negativity.61 Other workers have reached similar conclusions. Wang et al. reported that the interaction between a synthetic phosphopeptide corresponding to amino acids 93-106 of human OPN and the

{100} face of COM was “predominantly electrostatic”.62 Huq and co-workers used molecular modeling to study the adsorption of a phosphopeptide from DMP-1 to various faces of HA, concluding that “These preferences are principally governed by electrostatic interactions”.56 Likewise, Chen et al. suggested that the adsorption of the salivary protein statherin to HA was a “simple electrostatic interaction”.36

(61) Pampena, D. A.; Robertson, K. A.; Litvinova, O.; Lajoie, G.; Goldberg, H. A.; Hunter, G. K. Biochem. J. 2004, 378, 1083–1087.

(62) Wang, L.; Guan, X.; Tang, R.; Hoyer, J. R.; Wierzbicki, A.; De Yoreo, J. J.; Nancollas, G. H. J. Phys. Chem. B 2008, 112(30), 9151–7.

18644 DOI: 10.1021/la100401r

Langmuir 2010, 26(24), 18639–18646

Hunter et al.

Invited Feature Article

Figure 2. Electrostatic potential maps of osteopontin peptides and faces of calcium oxalate monohydrate. A. {010} face of COM. B. {121} face of COM. C. {100} face of COM. D. Peptide P0 (SHESTEQSDAIDSAEK) adsorbed to the {100} face of COM. E. Peptide P1 (SHESTEQSDAIDpSAEK) adsorbed to the {100} face of COM. F. Peptide P3 (pSHEpSTEQSDAIDpSAEK) adsorbed to the {100} face of COM. Simulations were created as previously described.41 For panels A-C, the isocontours are drawn at þ5 kT (blue) and -5 kT (red). For panels D-F, the isocontours are drawn at þ3 kT (blue) and -3 kT (red); the crystal face (not shown) is at the bottom of the panel. Table 1. Sequences of Osteopontin Peptides Predicted to Adsorb Most Strongly to the {100} Face of HA58 amino acids

sequence

simplified sequencea

49-64 EpTDDFKQETLPpSNpSNE CPCCxxxCxxxPxPxE 65-80 pSHDHMDDDDDDDDDGD PxCxxCCCCCCCCCxC 81-96 HAEpSEDSVNpSDEpSDES xxDPCCxxxPCCPCCx 97-112 HHpSDEpSDESFTASTQA xxPCCPCCxxxxxxxx 145-160 VpSDEQYPDApTDEDLTpS xPCCxxxCxPCCCxxP 193-208 pSHEpSSQLDEPpSVETHS PxCPxxxCCxPxCxxx 257-272 pSLEHQpSHEFHpSHEDKL PxCxxPxCxxPxCCxx a C = D or E, P = pS or pT, x = any other amino acid.

Our molecular dynamics analysis of the OPN-HA interaction provides another test of the stereochemical theory. Shown in Table 1 are the sequences of the 7 virtual peptides found to adsorb best to the {100} face of HA.58 These sequences are also shown in simplified form in which aspartic acid and glutamic acid are represented as “C”, phosphoserine and phosphothreonine as “P”, and all other amino acids as “x”. According to the strong stereochemical theory, the anionic amino acids should ideally be spaced at every second position along the polypeptide chain in order that all will be on the same side of a β-sheet; according to the weak theory, anionic amino acids should be periodically distributed along the sequence. No such periodic distribution of charged Langmuir 2010, 26(24), 18639–18646

groups is seen in the sequences shown in Table 1. Peptide OPN257-272 is the most periodic, with a repeating motif of PxCxx. In general, however, the OPN sequences predicted to adsorb best to HA exhibit clustering of anionic amino acids. In a previous study, we showed that a synthetic peptide corresponding to amino acids 290-301 of rat bone OPN was a potent inhibitor of HA nucleation in an autotitration assay. The sequence of this peptide is SHELEpSpSpSSEVN (xxCxCPPPxCxx). A peptide with the same composition but more regular spacing of the anionic amino acids, pSSHELpSSEVpSNE (PxxCxPxCxPxC), was about one-third less effective an inhibitor.61 The studies described above indicate that the adsorption of proteins to biominerals is governed by electrostatic interactions. In general, high-affinity adsorption does not appear to require the distribution of charged amino acids suggested by the stereochemical theory.

Conclusion Many studies have demonstrated face-specific interactions between proteins and biominerals with concomitant modification of crystal growth habit. According to the stereochemical theory proposed by Addadi and Weiner and supported by the work of Mann and others, face-specific adsorption of proteins to crystals DOI: 10.1021/la100401r

18645

Invited Feature Article

involves anionic groups of both that are optimally oriented to coordinate shared Ca2þ ions. Subsequently, other workers proposed models of protein-crystal interaction that dispensed with the constraints on the orientations of Ca2þ-coordinating groups. What the former (strong) and latter (weak) mechanisms of protein-crystal interaction have in common is the concept of a structural complementarity (or identity) between protein conformation and crystal face. The finding by us and others that the preference of mineralized tissue proteins for particular crystal faces is essentially electrostatic;that is, adsorption is favored by more anionic proteins/peptides and more cationic faces;is not incompatible with the weak stereochemical mechanism. However, the accumulating evidence that the mineral binding domains of proteins are generally lacking in folded structure suggests a quite different view of protein-crystal interaction;one that we have called the flexible polyelectrolyte hypothesis. According to this view, the lack of folding in mineral binding proteins facilitates their adsorption by allowing the charged groups maximum conformational freedom to form electrostatic bonds with Ca2þ ions exposed on the crystal face. The hallmarks of the stereochemical mechanisms are conformational rigidity and chemical specificity; the hallmarks of the flexible polyelectrolyte mechanism are conformational freedom and chemical nonspecificity. High negative charge density and conformational flexibility are also hallmarks of the glycosaminoglycan (GAG) chains of proteoglycans. Numerous studies have shown that proteoglycans and GAGs can affect the growth of biominerals.63-65 Solid-state NMR studies indicate that sugar groups are closely associated with the mineral phase, although these sugars could also be from the oligosaccharide chains of glycoproteins such as OPN and BSP.66 Taken at face value, the flexible polyelectrolyte hypothesis suggests that the most potent inhibitor of crystal formation would be an unstructured polymer with as many anionic groups as (63) Chen, C. C.; Boskey, A. L.; Rosenberg, L. C. Calc. Tissue Int. 1984, 36(3), 285–90. (64) Hunter, G. K.; Szigety, S. K. Matrix 1992, 12, 362–368. (65) Shirane, Y.; Kurokawa, Y.; Sumiyoshi, Y.; Kagawa, S. Scanning Microsc. 1995, 9, (4), 1081-8; discussion 1088. (66) Jaeger, C.; Groom, N. S.; Bowe, E. A.; Horner, A.; Davies, M. E.; Murray, R. C.; Duer, M. J. Chem. Mater. 2005, 17(12), 3059–3061.

18646 DOI: 10.1021/la100401r

Hunter et al.

possible. However, there are already suggestions that the chemistry of protein-crystal interaction is more complicated than this. For example, the virtual OPN peptide VLDPKpSKEDDRYLKFR exhibits a tighter adsorption to the {100} face of HA than its isoelectric point would predict, apparently because its anionic groups are clustered within the molecule.58 Thus, the sequence of mineral binding peptides/domains is probably not irrelevant. Also, although phosphate groups are more electronegative, carboxylate groups may be better able to form bonds with Ca2þ ions of a crystal face.41 If this turns out to be generally true, stereochemistry may be a factor in protein-crystal interaction at the amino acid level rather than at the secondary-structure level. Studies involving synthetic peptides and site-directed mutagenesis of proteins will undoubtedly cast further light on this issue. Perhaps the best test of the flexible polyelectrolyte hypothesis is to determine the conformations of proteins bound to crystals. According to the stereochemical mechanisms, the protein domains in contact with the crystal face should be in a unique periodic conformation ( β-sheet, β-strand, or R-helix). According to the flexible polyelectrolyte mechanism, these domains should be in a variety of different nonperiodic conformations. These alternatives could be distinguished by solid-state NMR analysis of proteins such as osteopontin or asprich. As noted above, there are several cases of mineral binding proteins or peptides that are R-helical. It is not inconceivable that others have a β-pleated sheet conformation. However, in view of the disordered nature of most mineral-associated proteins, the flexible polyelectrolyte form of adsorption appears likely to predominate. Finally, it should be borne in mind that even the most sophisticated experimental and theoretical studies on proteincrystal interaction are only crude approximations of processes occurring in mineralizing tissues. In vivo, such interactions will be modified by a variety of ions, small molecules, and macromolecules. OPN, for example, binds to cell surface receptors, forms covalently cross-linked oligomers, and interacts with other extracellular matrix proteins. Thus, concepts such as the stereochemical mechanisms and flexible polyelectrolyte hypothesis can serve only as general principles of protein-crystal interaction rather than complete descriptions of biomineralization processes.

Langmuir 2010, 26(24), 18639–18646