1276
J. Chem. Inf. Comput. Sci. 2000, 40, 1276-1284
Source of the Ice-Binding Specificity of Antifreeze Protein Type I Pranav Dalal and Frank D. So¨nnichsen* Department of Physiology and Biophysics, Case Western Reserve University, 10900 Euclid Avenue, Cleveland, Ohio 44106-4970 Received March 20, 2000
Antifreeze proteins (AFPs) are a group of structurally very diverse proteins with the unique capability of binding to the surface of seed ice crystals and inhibiting ice crystal growth. The AFPs bind with high affinity to specific planes of the ice crystal. Previously, this affinity of AFPs has been ascribed to the formation of multiple hydrogen bonds across the protein-ice interface, but more recently van der Waals interactions have been suggested to be the dominant energetic factors for the adsorption. To determine whether van der Waals interactions are also responsible for the binding specificities of AFPs, the protein-ice interaction of the helical AFP Type I from winter flounder (HPLC6) was studied using a Monte Carlo rigid body docking approach. HPLC6 binds in the {11h02h} direction of the [202h1] plane, with the Thr-Ala-Asn surface comprising the protein’s binding face. The binding of HPLC6 to this ice plane is highly preferred, but the protein is also found to bind favorably to the [101h0] prism plane using a different protein surface comprised of Thr and Ala residues. The results show that van der Waals interactions, despite accounting for most of the intermolecular energy (>80%), are not sufficient to completely explain the AFP binding specificity. INTRODUCTION
Fish antifreeze proteins (AFPs) are a class of structurally diverse proteins that protect polar fish from fatally freezing in ice-laden environments, where the temperature of seawater is about 1 °C below the freezing point of the fish’s body fluids. The proteins are present in fish serum and certain tissues such as gills, fins, and skin. AFPs depress the freezing point of their solutions in a kinetic, noncolligative manner, brought about by AFP adsorption to specific surface planes of seed ice crystals. This adsorption inhibits further growth of the ice-seed1 and effectively lowers the freezing point of the fish’s body fluids from -0.8 °C to about -2 °C. Five different types of fish AFPs are known to date.2 They are very divergent in size and structure but seem to function similarly. Of these proteins, AFP type I is the most studied antifreeze protein, particularly the liver-isoform from winter flounder (HPLC6). This monomeric, alanine-rich protein is comprised of 37 amino acids, containing three 11-amino acid repeats (TA2NA7) and additional N- and C-terminal capping sequences. It folds into a single, fairly straight R-helix,3 in which all polar residues with the exception of one Glu-Lys salt bridge are located on one side of the helix. Several studies using residue replacements indicated the protein’s ice binding site and supported the involvement of hydrophilic residues in ice-binding.4 These Thr and Asn residues are regularly spaced with 16.5 Å separation.5 The distance almost ideally matches the ice structure periodicity of 16.7 Å in the protein’s adsorption direction in the adsorption plane (the {11h02h} direction in the [202h1] plane). [Numbers in square brackets represent the family of 12 symmetry related planes in the hexagonal, four axis system identified by the Miller indices of one selected plane. Directions of binding are given as vectors in curly brackets, again referring to all symmetry * Corresponding author. Phone: (216) 368-5405. Fax: (216) 368-1693. Email:
[email protected].
related directions. For consistency, the selected vector represents the absolute direction in the respective, selected plane and is defined as the vector along the helical axis of the HPLC6 from the N- to the C-terminus.] Both the direction and plane have been determined experimentally by iceetching studies.6 The distance match supported the hypothesis that AFP-ice binding is mediated primarily by the formation of multiple hydrogen bonds between the protein and ice surface.7 More recently, the presumed critical role of hydrogen bonds in the adsorption of AFPs to ice has been questioned. A detailed analysis of the solution structure of AFP type III and its ice-bound model8 suggested a significant role of hydrophobic interactions in the AFP-ice adsorption. Conclusive experimental evidence for the importance of van der Waals interactions was obtained by systematic Thr replacements in HPLC6.9 Isosteric Val analogues were found to be highly active, whereas Ser analogues were practically inactive despite the retained hydrogen bonding capacity in their side chain. These results and subsequent studies10,11 strongly suggested that van der Waals interactions are primary determinants of HPLC6-ice interaction, while hydrogen bonds energetically play a minimal role in the protein-ice adsorption. Furthermore, very recent systematic Ala to Leu replacements around the helix have proposed an alternate protein binding face comprised of the Thr residues and two equivalent Ala positions (Thr-Ala-Ala) along the helix.12 This surface is rotated by 90° relative to the putative Thr-AlaAsn adsorption face. These recent results demonstrate the limits of our current understanding of the HPLC6-ice binding mechanism and suggest the need for a reconsideration of mechanistic aspects, such as the role of hydrophobic versus hydrophilic interactions, the location of the protein’s ice binding surface, and the source of its orientational and planar specificity.
10.1021/ci000449b CCC: $19.00 © 2000 American Chemical Society Published on Web 08/06/2000
ICE-BINDING SPECIFICITY
OF
ANTIFREEZE PROTEIN
Due to the experimental difficulties in studying ice-bound AFP structures, various groups13-16 have modeled the AFPice interactions using computational approaches. Common features of these modeling studies are (a) manual placement of the protein in selected orientations, (b) a primary focus on hydrogen bonds, and (c) modeling to only the [202h1] ice plane with a predefined binding face. Wen and Laursen13 and Madura et al.15 proposed that HPLC6 favorably binds to ice in the {11h02h} direction on the [202h1] ice plane with the N-terminus pointing toward the c-axis apex ({11h02h} or N+), compared to the mirror symmetry direction ({011h2h}) and their respective 180° rotations ({1h102} and {01h12}). In their model, Thr and Asn residues bind to different ranks of oxygen on the ridge of the [202h1] plane, as opposed to a model proposed by Sicheri and Yang5 in which Thr and Asn bind to the same rank of oxygen. Recently, Cheng and Merz16 completed a study using molecular dynamics calculations including the effects of water. The authors confirmed Wen and Laursen’s model13 and further suggested a correlation between experimentally observed antifreeze activities and binding energies or the number of protein-ice hydrogen bonds. A model developed using a Monte Carlo approach to automatically identify the preferred binding position of HPLC614 agreed with the model by Wen and Laursen.13 Further, AFP binding to the [202h1] plane was favored relative to the basal plane ([0001]) binding. The source of this plane specificity originated in larger favorable van der Waals interactions in the [202h1] plane, whereas contributions from hydrogen bonds and charge interactions to the intermolecular energies were similar. The suggested importance of hydrophobic interactions also leads to the need to reevaluate the current protein-ice models. In the absence of hydrogen bonds, these models do not necessarily explain the protein’s ice affinity and specificity of binding. Further, the current studies have only investigated the Thr-Ala-Asn protein surface, and the properties of the alternative binding site adsorbing to ice have yet to be modeled. Thus, the focus of this study is to generate unbiased ice-bound protein models using a Monte Carlo rigid body docking procedure. The protein is modeled to several ice surface planes in order to identify the source of the protein’s directional and orientational specificity and to determine the preferred ice-adsorption surface of the protein. METHODS
Initial coordinates for HPLC6 were taken from the crystal structure.5 An analysis of these coordinates (PDB accession codes 1WFA and 1WFB) showed variable degrees of helix bending of the two AFP molecules in the unit cells. Therefore, idealized straight helices with φ and ψ angles of -65° and -41.6°, respectively, were generated, which reproduced the approximate 11-amino acid repeat observed in the crystal structures. Side chain conformations were obtained from the deposited coordinates. Alternative Thr side chain rotamers were generated manually in Insight II (MSI, San Diego, CA) from this template. The resulting protein structure was minimized for 1000 steps with fixed backbone. The details of the minimization are described below. An Ih unit cell was constructed with TIP3P water molecules. Ice lattices (approximate dimensions of 80 Å × 65
J. Chem. Inf. Comput. Sci., Vol. 40, No. 5, 2000 1277
Å × 10 Å) were then obtained by repetition of this Ih unit cell by a Fortran program and were subsequently cut to obtain the desired crystallographic surface. The choice of the exact location of the cut results in planes with identical indices but differing surface properties. Four different [202h1] planes can be generated, each with a different surface. After estimating the plane stability using the number of broken hydrogen bonds, and the number of hydrogen bonds per surface water molecule, two planes were selected. The plane with the smallest number of broken hydrogen bonds (5) was selected for this study ([202h1], different cut), although it has not been used for modeling studies so far. The only other plane ([202h1]), in which all surface water molecules are involved in at least two hydrogen bonds, is identical to the [202h1] plane used in previous modeling studies by Wen and Laursen.13 Another plane, parallel but slightly different from the ones utilized here, has been previously used in another study.15 This plane might be considered to be the ideal [202h1] plane, since it intersects a unit cell exactly at the positions defined by the Miller indices or reciprocal coordinates. This plane, however, seems less realistic as one water molecule is bonded to the lattice only via a single hydrogen bond. Similar selections were made for the prism and the basal plane, while only one possible cut exists for the secondary prism plane. The docking algorithm used for these calculations is implemented in the Insight II (MSI) software. For energy evaluations in the docking calculations and the minimizations, the all-atom constant valence force field (CVFF) was used. A simplified energy expression was employed consisting of bonded interactions, and van der Waals interactions as the only nonbonded interactions. The van der Waals energies were calculated using the cell multipole method,17 with the Insight II implementation specific parameters set to single energy evaluation and coarse accuracy. Oxygen atom positions of the ice were kept fixed throughout the calculations using a subset definition. The intermolecular energy was calculated as the difference between the total energy of the system (protein and ice) and the sum of individual energies of ice and protein. The MC•Minimize docking algorithm used in these calculations is implemented in the fixedDocking command, part of the Affinity module in the Insight II software. The algorithm uses alternating random rigid body moves of a ligand and energy minimization of the system to identify the favorable conformations. The energy minimization step in this algorithm utilizes the Discover 95.0/3.00 program implemented in the Discover•3 module of Insight II. The docking algorithm was slightly modified, a R. M. S. check that was originally placed after the minimization was moved to before the minimization in order to avoid unnecessary minimizations, and thus saving on the computational cost. Details of the algorithm are given in Figure 1. For each step of the calculations, the protein alignment relative to the ice was altered by a Monte Carlo rigid body move of the protein. This move, which had 6 degrees of freedom, consisted of random translations of the protein in all three directions of the ice-crystal coordinate system and random rotations around the three principal axes of the protein. Subsequently, an energy and a root mean square check were performed. To avoid minima trapping, instead of a Metropolis energy acceptance criterion, a more generous
1278 J. Chem. Inf. Comput. Sci., Vol. 40, No. 5, 2000
Figure 1. Flow chart of the algorithm employed in the calculations. The MC•Minimize algorithm implemented in the Affinity module of the Insight II software was used for the calculations. It was slightly modified for these particular calculations. The modified algorithm is described in the text.
energy range criterion was used. New structures with energies up to a specified number (∆E) of kilocalories per mole less favorable than the best accepted structure were accepted. The root mean square check ensured that the root mean square (RMS) distance between the current and all previously accepted protein structures was greater than a specified number (∆RMS) of angstroms. After passing these two checks, a conjugate gradient energy minimization was performed. If the minimized structure passed the energy check, it was saved for analysis. Independent of this acceptance, the protein geometry was restored to the (idealized) starting protein geometry after each minimization before the next Monte Carlo move. A typical calculation used two phases: the purpose of the first phase was the identification of a global minimum by an extensive search of the alignment space. Rotational and translational moves were allowed on all three axes, and randomly selected from -180 to +180° and -6 to +6 Å. A root mean square criterion (∆RMS > 1.5 Å) and an energy criterion (∆E < 5 kcal/mol) were used to prescreen the new alignment, ensuring that only significantly different structures with better or similar energies were subjected to a 100 steps conjugate gradients minimization. After minimization, the energy criterion (∆E < 5 kcal/mol) was again employed for final acceptance of the alignment. The end of phase I, usually reached after about 250 000 iterations, was indicated by the repetitive nature of the solutions. At this point, accepted structures represented equivalent models, but with translated protein adsorption positions on the ice. To further optimize the protein alignment on the ice, the root mean square
DALAL
AND
SO¨NNICHSEN
criterion was removed, and its rotational and translational movements were restricted (
