FULL PAPER
WWW.C-CHEM.ORG
How Does a Hydrocarbon Staple Affect Peptide Hydrophobicity? Adelene Y. L. Sim*[a] and Chandra Verma[a,b,c] Water is essential for the proper folding of proteins and the assembly of protein–protein/ligand complexes. How water regulates complex formation depends on the chemical and topological details of the interface. The dynamics of water in the interdomain region between an E3 ubiquitin ligase (MDM2) and three different peptides derived from the tumor suppressor protein p53 are studied using molecular dynamics. The peptides show bimodal distributions of interdomain water densities across a range of distances. The addition of a hydrocarbon chain to rigidify the peptides (in a process known as
stapling) results in an increase in average hydrophobicity of the peptide–protein interface. Additionally, the hydrophobic staple shields a network of water molecules, kinetically stabilizing a water chain hydrogen-bonded between the peptide and MDM2. These properties could result in a decrease in the energy barrier associated with dehydrating the peptide–protein interface, thereby regulating the kinetics of peptide bindC 2015 Wiley Periodicals, Inc. ing. V
Introduction
other.[9–12] A protein–protein interface combines both hydrophobic and hydrophilic characteristics: polar NAH and CAO bonds in the backbone, and different chemical properties (polar, hydrophobic, and charged) of side chains. The polar atoms around a predominantly hydrophobic interface can engage waters in enthalpically favorable hydrogen bonds, thereby disfavoring perfect dewetting of the interdomain region (except if the interdomain distance is so small that waters are sterically excluded). Electrostatic interactions and van der Waals’ attractions contribute to the resultant water structure,[10] and these forces depend on the local hydrophobic/hydrophilic and topological features[13] of the interface. Understanding the structure of water around protein surfaces and in interdomain regions therefore yields insights into how water molecules could mediate protein–protein recognition,[11,14] aggregation,[12,15] and binding kinetics.[9,10,16–18] The dewetting of interdomain space prior to binding was found to be the dominant energetic barrier for ligand binding in some cases.[19,20] This energetic barrier during dewetting involves a delicate balance between the change in enthalpy and entropy. The dewetting phenomenon has length-scale dependent effects, because the enthalpy–entropy compensation depends on the size of the interface, as well as the strength of the solvent–solute interaction (i.e., the chemical properties of the
Water is a vital component in all-living systems, constituting about 70–80% by mass of most living cells.[1,2] Its unique physical properties[3] are critical to how biomolecular macromolecules fold and function and interact with each other.[4,5] The OAH bond of water is highly polar, and because each water molecule has two OAH bonds angled at 104.5 from each other, water has a strong net dipole. As a result, water molecules like to orient themselves around each other or around small apolar solute molecules to form as many hydrogen bonds as possible[6]: the ordering of water around small hydrophobic solutes is associated with the formation of highly intricate structures ([7] and citations within). These highly ordered water patterns break down around large and flat hydrophobic solutes, as it is harder to form highly structured networks of hydrogen bonds around such surfaces.[6] As two hydrophobic solutes or interfaces come into close proximity, the water molecules that lie in between the approaching surfaces adjust to optimize their own hydrogen bonds and accommodate both solutes’ topologies and hydrophobicities. For large hydrophobic solutes, water tends to be expelled from the intersolute region to maximize hydrogen bond contacts amongst the water molecules.[8] This distancedependent dewetting effect generates an intersolute vacuum that induces solute aggregation. In the limiting case, when the solutes are far apart, the intersolute waters behave as in bulk (except for waters close to the solute–solvent interface). That is, we can expect little or no intersolute dewetting. Below a “critical distance”, the dewetting phenomenon sets in and waters are ordered away from the intersolute region.[6] This critical distance depends on the extent of hydrophobicity, as well as on the area of the interface.[3,5,9] The situation is less straightforward for protein–protein interfaces, where examples of wet and dewetted interdomain spaces have been observed as two domains approach each
DOI: 10.1002/jcc.23859
[a] A. Y. L. Sim, C. Verma Bioinformatics Institute (A*STAR), 30 Biopolis Street #07-01, Matrix, 138671, Singapore E-mail: [email protected] [b] C. Verma School of Biological Sciences, Nanyang Technological University, 60 Nanyang Drive, 637551, Singapore [c] C. Verma Department of Biological Sciences, National University of Singapore, 14 Science Drive 4, Singapore117543 C 2015 Wiley Periodicals, Inc. V
Journal of Computational Chemistry 2015, 36, 773–784
773
FULL PAPER
WWW.C-CHEM.ORG
interface).[21–23] Traditionally, dewetting was thought to be driven by entropy: water molecules trapped within the interdomain region are released to bulk, resulting in a gain in entropy.[24,25] However, using a model system with concave cavity geometry, it was shown that dewetting can sometimes be driven by enthalpy instead of entropy; a critical determinant is whether the water is fluctuating more in the interdomain space than when released into bulk.[23] Hence to understand the effects of hydrophobicity on protein–protein/ ligand interactions, one has to consider system-specific features of each interface. We focus on the interface between the protein p53 and its negative regulator MDM2. In most human cancers, the tumorsuppressive function of p53 is disrupted[26] through disparate mechanisms, one of which is the upregulation of MDM2.[27] MDM2 is an E3 ubiquitin ligase that binds and controls the ubiquitination of p53: when MDM2 is upregulated, p53 undergoes increased ubiquitin-dependent proteolysis and its concentration in cells drops.[28] The primary interaction between p53 and MDM2 occurs in the N-terminal region of both proteins.[29] A short a-helix in p53 sits in a hydrophobic pocket of MDM2, with residues F19, W23, and L26 of p53 making critical contacts.[29] Inhibition of this interaction with small molecules ([30] and references therein) and peptides[10,31,32] have been shown to upregulate the levels of p53 and thus offers promise in treatment of cancers that harbor wild type p53 and overexpressed MDM2 (reviewed in [33–35]). The use of peptides as potential therapeutics has received a major boost with the development of stapled peptides.[36,37] These are short protein segments with hydrocarbon linkers (staples) that act as distance constraints between residues. The constraints reduce the conformational diversity of the peptides,[38] and in the case of p53-derived peptides, stapling the a-helical region that is known to interact with MDM2 helps maintain its secondary structure. There is a growing interest in stapled peptides as drugs[36,37,39,40] because staples stabilize peptides against proteolysis[41] and appear to promote cell permeability.[31,42] Additionally, optimizing the staple location such that the hydrocarbon linker becomes part of the peptide–protein interface enhances binding affinity[43–46]—these are the cases we will focus on in this study. Recently, Guo et al. [47] investigated the hydration properties around MDM2 and the p53 peptide as the latter was systematically displaced from its binding pocket. Interestingly, the interdomain space showed heterogeneous hydration, with regions of high and low water density. Because the hydrocarbon staple is hydrophobic, in cases where the staple becomes part of the peptide–protein interface, stapling the p53 peptide changes the overall hydrophobicity and area of the interface, which likely impacts the interdomain water dynamics and structure. These alterations in hydration properties could influence peptide recognition[11,12,14] and its binding dynamics[9,10,16–18] by modulating the topology of the energy landscape around the peptide-bound state. This picture is consistent with the notion that ligand–receptor dewetting can cause a significant energy barrier in the binding path774
Journal of Computational Chemistry 2015, 36, 773–784
way.[19,20,48] An increase in the average hydrophobicity of the peptide could increase dewetting propensity, resulting in a decrease in the transition state dewetting energy barrier, thereby altering the kon (binding association) and koff (binding dissociation) rate constants. For example, binding interfaces with hydrogen bonds shielded from water molecules by hydrophobic neighbors were found to be kinetically more stable than their unshielded counterparts.[49] Understanding these changes gives us general insights into hydrophobic effects in peptide–protein interfaces, and can also guide our design of MDM2-targeted stapled peptides with tuned binding kinetics. Binding kinetics has recently been found to be critical in determining drug efficacy,[50–53] especially in in vivo systems where binding equilibrium might not be reached; binding affinity is defined only for equilibrium conditions and is inapplicable for determining drug efficacy in such cases.[54] In earlier work,[55] we simulated the approach of peptides towards the hydrophobic MDM2 pocket, and found that the presence of a staple altered peptide-binding dynamics. We also observed that water molecules were structured around the peptide-MDM2 interdomain space and were released into bulk just prior to binding. We were curious to investigate whether it is beneficial to design peptides that increase the number of these prebound structured waters, to minimize the energy of the intermediate prebound state. To investigate this, and to understand the changes in interdomain hydration due to the addition of a hydrocarbon staple to the peptide-MDM2 binding interface, we conducted a series of restrained molecular dynamics (MD) simulations on three peptides: p53, a stapled and an unstapled peptide. Each peptide was restrained in increasing distances from MDM2, and we studied the corresponding interdomain water densities and patterns (Fig. 1). We focused on the water densities around the key hydrophobic residues F19, W23, and L26 (numbering based on the full p53 sequence; for simplicity, subsequently in the text, we will refer to this same residue numbering for all peptides). The sequences of these peptides are shown in Table 1. The stapled and unstapled peptides studied here were previously shown to activate p53 activity in the cell.[32] We omitted the study of small molecule hydration, as the interdomain volumes defined by small molecule drugs are more varied and hence harder to systematically compare. In addition, recent observations suggest that the small molecule Nutlin, that binds to the p53-binding pocket of MDM2 and displaces p53, was found to possibly first bind to a secondary binding site on MDM2, before maneuvering to the p53 binding site.[56] Similar observations have been made for small molecules binding to kinases,[19,57] suggesting initial secondary site binding might be a common occurrence for small molecules.
Materials and Methods Molecular dynamics simulations MD simulations were conducted using pmemd.cuda—the GPU the AMBER12 simulation accelerated version[58]—of WWW.CHEMISTRYVIEWS.COM
FULL PAPER
WWW.C-CHEM.ORG
Figure 1. The relative interdomain water densities as a function of simulation time. Ten independent simulations are concatenated together (and distinguished by black dashed lines) for the p53, stapled and unstapled peptides (black, blue and red, respectively). Raw density values are shown as thin semitransparent lines, while the time-smoothed values are shown in bold. The corresponding histograms of the relative interdomain water densities are shown on the right. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
package.[59] The starting coordinates for the p53 peptide bound to MDM2 were taken from the crystal structure 1YCR[29] while those for the stapled and unstapled peptides were obtained from prior equilibration simulations.[43] Peptides were capped at the N- and C-termini with acetyl and NH2 groups respectively, while MDM2 was capped at the N- and C-termini with acetyl and N-methyl groups, respectively. TIP3P waters[60] ˚ away were placed in a rectangular box with edges at least 8 A from MDM2 and peptide using the solvatebox command (with ˚ ) in the tleap module of Ambercloseness parameter 0.7 A [61] Tools12. Chloride ions were added to neutralize the systems (three for the p53-MDM2 system; four for the stapled peptideMDM2 and unstapled peptide-MDM2 systems respectively). Peptides were initially displaced with distances ranging from ˚ to 8 A˚ along the direction connecting the centers of 3.5 A masses of the peptide and MDM2. A two-step minimization routine was used. In the first step, all solutes were restrained while water and ion coordinates were minimized. 1000 cycles of steepest descent algorithm was used followed by another (at most) 1000 cycles of conjugate gradient minimization. In the next minimization phase, restraints were removed, and a maximum of 3990 conjugate gradient minimization cycles were performed after 10 cycles of steepest descent minimization. After minimization, production simulations were conducted, at constant temperature and pressure. Temperature was regulated around 300K using Langevin dynamics with collision frequency of 1.0 ps21. Pressure was kept around 1 bar using isotropic position scaling and constant pressure periodic boundary conditions. A pressure relaxation time of 2 ps was
used. In order to directly observe (de)wetting, no gradual heating or constant volume simulations were run prior to production simulations as has been recommended by Zhou et al.[62] in their work in probing hydration phenomena. Control simulations with a period of gradual heating and/or constant volume constraints were also attempted, and no significant hydration differences were observed (data not shown). All simulations were conducted with the AMBER-99SB forcefield[63] and electrostatics were handled with the particle-mesh Ewald[64] approach. All other nonbonded interactions are eval˚ . A time step of 2 fs was used, and uated with a cutoff of 12 A [65] SHAKE was implemented to constrain all bonds involving hydrogen atoms. Additionally, similar to previous studies,[9,47] all backbone heavy atoms (MDM2 and peptide) were restrained ˚ 2) to study the dynamics with a force constant of 10 kcal/(mol A of water molecules within the interdomain region. Side chain atoms were intentionally kept dynamic, so that hydrogenbonding interactions with water can be optimized. For each displacement distance, 10 independent simulations of 10 ns were run starting from the same minimized structure.
Table 1. Sequence of peptides studied.[a] Name p53 Stapled Unstapled
Sequence (*: staple) Ac-ETFSDLWKLLPEN-NH2 Ac-TSF*EYWALL*-NH2 Ac-TSFAEYWALLS-NH2
[a] In bold are the hydrophobic residues (F19, W23 and L26, based on full p53 sequence) that are critical for p53 binding to MDM2.
Journal of Computational Chemistry 2015, 36, 773–784
775
FULL PAPER
WWW.C-CHEM.ORG
Figure 2. The structures of a) p53 (black), b) stapled (blue, with staple highlighted in yellow), and c) unstapled (red) peptides. The surface electrostatics of the peptides (bottom panel) and MDM2 (right) are shown with a color scale going from red to blue (ranging from 21 kT/e to 1 kT/e). The F19, W23 and L26 pockets in MDM2 are indicated.
The results discussed here were obtained from combining statistics of all of these runs. When considering equilibrium behavior, only frames from 500 ps were considered, as the water density equilibrated within 500 ps. The corresponding equilibrated peptide-MDM2 distances were also evaluated. Electrostatic calculations Electrostatics were computed using the linearized Poisson– Boltzmann equation implemented in the Adaptive Poisson– Boltzmann software [66–68] Tools plugin in PyMOL.[69] Protein charges were assigned based on the Amber force-field implemented in PDB2PQR.[70,71] The system temperature was 300K and monovalent salt concentration was set at 150 mM. All other settings were set to default values. For better visualization in Figure 2, the electrostatic potentials on the solvent accessible surface were mapped onto the van der Waals surface. Identifying interdomain waters Prior to identifying interdomain waters, heavy atoms interacting in the peptide–protein interface were first determined using a peptide–protein heavy atom distance cutoff of 5 A˚. ˚ of a That is, any protein heavy atom that comes within 5 A peptide heavy atom (and vice versa) is identified as an “interface atom.” Following work by Zhou et al., [62] a water molecule was considered to be interdomain if its minimum distance to a protein interface atom (d1) and its minimum distance to a peptide interface atom (d2) satisfied the condition d1 1 d2 D 1 d, where D is the peptide–protein distance, ˚ to account for atomic radii. This choice of d is and d 5 5 A ˚ used by Zhou et al. [62] as we found that larger than the 2.5 A for our system, due to the relatively (compared with protein– protein interfaces) small surface area of the interface, a choice of 2.5 A˚ resulted in too many important interdomain waters 776
Journal of Computational Chemistry 2015, 36, 773–784
being omitted. Using a d of 5 A˚ gave a good balance between capturing important interdomain waters while minimizing the number of peripheral waters (essentially noise). When considering interdomain waters around F19 and L26, only MDM2 heavy atoms that interact with the respective residues in the bound conformation were considered to be part of the interface. We ignored the interdomain waters around W23 to prevent double counting waters that were considered in the F19only and L26-only volumes; W23 is located between F19 and L26 and splits the interdomain volume into two. If waters that were found around W23 were not present around F19 or L26, they were counted into the L26-only space. Recent work to study the MDM2-p53 interdomain hydration properties used a grid-based approach to determine water density,[72] which required that the interface remained rigid for cleaner classification of the interdomain space. However, to more realistically capture interdomain hydration properties as a peptide approaches MDM2, we allowed protein/peptide side chains to move and adjust to maximize hydrogen bonding with waters. The distance-based approach to determine interdomain waters is more appropriate for our analysis. We also avoid any issues of grid edge effects by using the distancebased classification. Evaluating interdomain water density To estimate the interdomain water density relative to bulk, we first determined the number of bulk waters that can fit into ˚ )3 TIP3P[60] the same interdomain space. We set up a (60 A water-box simulation for 200 ps with parameters as described above, and used the last simulation frame as the reference water box. The waters in each peptide-MDM2 simulation frame were then removed, and the peptide-MDM2 complex was placed inside the reference water box. Waters that resulted in steric clashes with MDM2 or the peptide were removed. The WWW.CHEMISTRYVIEWS.COM
FULL PAPER
WWW.C-CHEM.ORG
numbers of bulk waters in the interdomain space were then calculated, as before, using these constructed frames. The interdomain water density can be estimated with the relative number of waters (from simulation) to those from the bulk (using the reference water box). However, we found that this approach resulted in significant noise in the density distribution, as the difference of a single water molecule in the bulk calculation leads to significant changes in the evaluated relative density. To circumvent this, we determined the dependence of the number of bulk waters on the interdomain space. The interdomain space is proportional to the distance of the displaced peptide to MDM2. We linearly interpolated the number of bulk waters at the displacements to estimate the expected number of bulk waters for a given space. This was repeated for all peptides, and for the different volume spaces (i.e., full interface space, F19-only space and L26-only space). When considering the full interface space, it was found that a single line was not adequate to describe the dependence, and two piece-wise linear fits (data for distance
WWW.C-CHEM.ORG
How Does a Hydrocarbon Staple Affect Peptide Hydrophobicity? Adelene Y. L. Sim*[a] and Chandra Verma[a,b,c] Water is essential for the proper folding of proteins and the assembly of protein–protein/ligand complexes. How water regulates complex formation depends on the chemical and topological details of the interface. The dynamics of water in the interdomain region between an E3 ubiquitin ligase (MDM2) and three different peptides derived from the tumor suppressor protein p53 are studied using molecular dynamics. The peptides show bimodal distributions of interdomain water densities across a range of distances. The addition of a hydrocarbon chain to rigidify the peptides (in a process known as
stapling) results in an increase in average hydrophobicity of the peptide–protein interface. Additionally, the hydrophobic staple shields a network of water molecules, kinetically stabilizing a water chain hydrogen-bonded between the peptide and MDM2. These properties could result in a decrease in the energy barrier associated with dehydrating the peptide–protein interface, thereby regulating the kinetics of peptide bindC 2015 Wiley Periodicals, Inc. ing. V
Introduction
other.[9–12] A protein–protein interface combines both hydrophobic and hydrophilic characteristics: polar NAH and CAO bonds in the backbone, and different chemical properties (polar, hydrophobic, and charged) of side chains. The polar atoms around a predominantly hydrophobic interface can engage waters in enthalpically favorable hydrogen bonds, thereby disfavoring perfect dewetting of the interdomain region (except if the interdomain distance is so small that waters are sterically excluded). Electrostatic interactions and van der Waals’ attractions contribute to the resultant water structure,[10] and these forces depend on the local hydrophobic/hydrophilic and topological features[13] of the interface. Understanding the structure of water around protein surfaces and in interdomain regions therefore yields insights into how water molecules could mediate protein–protein recognition,[11,14] aggregation,[12,15] and binding kinetics.[9,10,16–18] The dewetting of interdomain space prior to binding was found to be the dominant energetic barrier for ligand binding in some cases.[19,20] This energetic barrier during dewetting involves a delicate balance between the change in enthalpy and entropy. The dewetting phenomenon has length-scale dependent effects, because the enthalpy–entropy compensation depends on the size of the interface, as well as the strength of the solvent–solute interaction (i.e., the chemical properties of the
Water is a vital component in all-living systems, constituting about 70–80% by mass of most living cells.[1,2] Its unique physical properties[3] are critical to how biomolecular macromolecules fold and function and interact with each other.[4,5] The OAH bond of water is highly polar, and because each water molecule has two OAH bonds angled at 104.5 from each other, water has a strong net dipole. As a result, water molecules like to orient themselves around each other or around small apolar solute molecules to form as many hydrogen bonds as possible[6]: the ordering of water around small hydrophobic solutes is associated with the formation of highly intricate structures ([7] and citations within). These highly ordered water patterns break down around large and flat hydrophobic solutes, as it is harder to form highly structured networks of hydrogen bonds around such surfaces.[6] As two hydrophobic solutes or interfaces come into close proximity, the water molecules that lie in between the approaching surfaces adjust to optimize their own hydrogen bonds and accommodate both solutes’ topologies and hydrophobicities. For large hydrophobic solutes, water tends to be expelled from the intersolute region to maximize hydrogen bond contacts amongst the water molecules.[8] This distancedependent dewetting effect generates an intersolute vacuum that induces solute aggregation. In the limiting case, when the solutes are far apart, the intersolute waters behave as in bulk (except for waters close to the solute–solvent interface). That is, we can expect little or no intersolute dewetting. Below a “critical distance”, the dewetting phenomenon sets in and waters are ordered away from the intersolute region.[6] This critical distance depends on the extent of hydrophobicity, as well as on the area of the interface.[3,5,9] The situation is less straightforward for protein–protein interfaces, where examples of wet and dewetted interdomain spaces have been observed as two domains approach each
DOI: 10.1002/jcc.23859
[a] A. Y. L. Sim, C. Verma Bioinformatics Institute (A*STAR), 30 Biopolis Street #07-01, Matrix, 138671, Singapore E-mail: [email protected] [b] C. Verma School of Biological Sciences, Nanyang Technological University, 60 Nanyang Drive, 637551, Singapore [c] C. Verma Department of Biological Sciences, National University of Singapore, 14 Science Drive 4, Singapore117543 C 2015 Wiley Periodicals, Inc. V
Journal of Computational Chemistry 2015, 36, 773–784
773
FULL PAPER
WWW.C-CHEM.ORG
interface).[21–23] Traditionally, dewetting was thought to be driven by entropy: water molecules trapped within the interdomain region are released to bulk, resulting in a gain in entropy.[24,25] However, using a model system with concave cavity geometry, it was shown that dewetting can sometimes be driven by enthalpy instead of entropy; a critical determinant is whether the water is fluctuating more in the interdomain space than when released into bulk.[23] Hence to understand the effects of hydrophobicity on protein–protein/ ligand interactions, one has to consider system-specific features of each interface. We focus on the interface between the protein p53 and its negative regulator MDM2. In most human cancers, the tumorsuppressive function of p53 is disrupted[26] through disparate mechanisms, one of which is the upregulation of MDM2.[27] MDM2 is an E3 ubiquitin ligase that binds and controls the ubiquitination of p53: when MDM2 is upregulated, p53 undergoes increased ubiquitin-dependent proteolysis and its concentration in cells drops.[28] The primary interaction between p53 and MDM2 occurs in the N-terminal region of both proteins.[29] A short a-helix in p53 sits in a hydrophobic pocket of MDM2, with residues F19, W23, and L26 of p53 making critical contacts.[29] Inhibition of this interaction with small molecules ([30] and references therein) and peptides[10,31,32] have been shown to upregulate the levels of p53 and thus offers promise in treatment of cancers that harbor wild type p53 and overexpressed MDM2 (reviewed in [33–35]). The use of peptides as potential therapeutics has received a major boost with the development of stapled peptides.[36,37] These are short protein segments with hydrocarbon linkers (staples) that act as distance constraints between residues. The constraints reduce the conformational diversity of the peptides,[38] and in the case of p53-derived peptides, stapling the a-helical region that is known to interact with MDM2 helps maintain its secondary structure. There is a growing interest in stapled peptides as drugs[36,37,39,40] because staples stabilize peptides against proteolysis[41] and appear to promote cell permeability.[31,42] Additionally, optimizing the staple location such that the hydrocarbon linker becomes part of the peptide–protein interface enhances binding affinity[43–46]—these are the cases we will focus on in this study. Recently, Guo et al. [47] investigated the hydration properties around MDM2 and the p53 peptide as the latter was systematically displaced from its binding pocket. Interestingly, the interdomain space showed heterogeneous hydration, with regions of high and low water density. Because the hydrocarbon staple is hydrophobic, in cases where the staple becomes part of the peptide–protein interface, stapling the p53 peptide changes the overall hydrophobicity and area of the interface, which likely impacts the interdomain water dynamics and structure. These alterations in hydration properties could influence peptide recognition[11,12,14] and its binding dynamics[9,10,16–18] by modulating the topology of the energy landscape around the peptide-bound state. This picture is consistent with the notion that ligand–receptor dewetting can cause a significant energy barrier in the binding path774
Journal of Computational Chemistry 2015, 36, 773–784
way.[19,20,48] An increase in the average hydrophobicity of the peptide could increase dewetting propensity, resulting in a decrease in the transition state dewetting energy barrier, thereby altering the kon (binding association) and koff (binding dissociation) rate constants. For example, binding interfaces with hydrogen bonds shielded from water molecules by hydrophobic neighbors were found to be kinetically more stable than their unshielded counterparts.[49] Understanding these changes gives us general insights into hydrophobic effects in peptide–protein interfaces, and can also guide our design of MDM2-targeted stapled peptides with tuned binding kinetics. Binding kinetics has recently been found to be critical in determining drug efficacy,[50–53] especially in in vivo systems where binding equilibrium might not be reached; binding affinity is defined only for equilibrium conditions and is inapplicable for determining drug efficacy in such cases.[54] In earlier work,[55] we simulated the approach of peptides towards the hydrophobic MDM2 pocket, and found that the presence of a staple altered peptide-binding dynamics. We also observed that water molecules were structured around the peptide-MDM2 interdomain space and were released into bulk just prior to binding. We were curious to investigate whether it is beneficial to design peptides that increase the number of these prebound structured waters, to minimize the energy of the intermediate prebound state. To investigate this, and to understand the changes in interdomain hydration due to the addition of a hydrocarbon staple to the peptide-MDM2 binding interface, we conducted a series of restrained molecular dynamics (MD) simulations on three peptides: p53, a stapled and an unstapled peptide. Each peptide was restrained in increasing distances from MDM2, and we studied the corresponding interdomain water densities and patterns (Fig. 1). We focused on the water densities around the key hydrophobic residues F19, W23, and L26 (numbering based on the full p53 sequence; for simplicity, subsequently in the text, we will refer to this same residue numbering for all peptides). The sequences of these peptides are shown in Table 1. The stapled and unstapled peptides studied here were previously shown to activate p53 activity in the cell.[32] We omitted the study of small molecule hydration, as the interdomain volumes defined by small molecule drugs are more varied and hence harder to systematically compare. In addition, recent observations suggest that the small molecule Nutlin, that binds to the p53-binding pocket of MDM2 and displaces p53, was found to possibly first bind to a secondary binding site on MDM2, before maneuvering to the p53 binding site.[56] Similar observations have been made for small molecules binding to kinases,[19,57] suggesting initial secondary site binding might be a common occurrence for small molecules.
Materials and Methods Molecular dynamics simulations MD simulations were conducted using pmemd.cuda—the GPU the AMBER12 simulation accelerated version[58]—of WWW.CHEMISTRYVIEWS.COM
FULL PAPER
WWW.C-CHEM.ORG
Figure 1. The relative interdomain water densities as a function of simulation time. Ten independent simulations are concatenated together (and distinguished by black dashed lines) for the p53, stapled and unstapled peptides (black, blue and red, respectively). Raw density values are shown as thin semitransparent lines, while the time-smoothed values are shown in bold. The corresponding histograms of the relative interdomain water densities are shown on the right. [Color figure can be viewed in the online issue, which is available at wileyonlinelibrary.com.]
package.[59] The starting coordinates for the p53 peptide bound to MDM2 were taken from the crystal structure 1YCR[29] while those for the stapled and unstapled peptides were obtained from prior equilibration simulations.[43] Peptides were capped at the N- and C-termini with acetyl and NH2 groups respectively, while MDM2 was capped at the N- and C-termini with acetyl and N-methyl groups, respectively. TIP3P waters[60] ˚ away were placed in a rectangular box with edges at least 8 A from MDM2 and peptide using the solvatebox command (with ˚ ) in the tleap module of Ambercloseness parameter 0.7 A [61] Tools12. Chloride ions were added to neutralize the systems (three for the p53-MDM2 system; four for the stapled peptideMDM2 and unstapled peptide-MDM2 systems respectively). Peptides were initially displaced with distances ranging from ˚ to 8 A˚ along the direction connecting the centers of 3.5 A masses of the peptide and MDM2. A two-step minimization routine was used. In the first step, all solutes were restrained while water and ion coordinates were minimized. 1000 cycles of steepest descent algorithm was used followed by another (at most) 1000 cycles of conjugate gradient minimization. In the next minimization phase, restraints were removed, and a maximum of 3990 conjugate gradient minimization cycles were performed after 10 cycles of steepest descent minimization. After minimization, production simulations were conducted, at constant temperature and pressure. Temperature was regulated around 300K using Langevin dynamics with collision frequency of 1.0 ps21. Pressure was kept around 1 bar using isotropic position scaling and constant pressure periodic boundary conditions. A pressure relaxation time of 2 ps was
used. In order to directly observe (de)wetting, no gradual heating or constant volume simulations were run prior to production simulations as has been recommended by Zhou et al.[62] in their work in probing hydration phenomena. Control simulations with a period of gradual heating and/or constant volume constraints were also attempted, and no significant hydration differences were observed (data not shown). All simulations were conducted with the AMBER-99SB forcefield[63] and electrostatics were handled with the particle-mesh Ewald[64] approach. All other nonbonded interactions are eval˚ . A time step of 2 fs was used, and uated with a cutoff of 12 A [65] SHAKE was implemented to constrain all bonds involving hydrogen atoms. Additionally, similar to previous studies,[9,47] all backbone heavy atoms (MDM2 and peptide) were restrained ˚ 2) to study the dynamics with a force constant of 10 kcal/(mol A of water molecules within the interdomain region. Side chain atoms were intentionally kept dynamic, so that hydrogenbonding interactions with water can be optimized. For each displacement distance, 10 independent simulations of 10 ns were run starting from the same minimized structure.
Table 1. Sequence of peptides studied.[a] Name p53 Stapled Unstapled
Sequence (*: staple) Ac-ETFSDLWKLLPEN-NH2 Ac-TSF*EYWALL*-NH2 Ac-TSFAEYWALLS-NH2
[a] In bold are the hydrophobic residues (F19, W23 and L26, based on full p53 sequence) that are critical for p53 binding to MDM2.
Journal of Computational Chemistry 2015, 36, 773–784
775
FULL PAPER
WWW.C-CHEM.ORG
Figure 2. The structures of a) p53 (black), b) stapled (blue, with staple highlighted in yellow), and c) unstapled (red) peptides. The surface electrostatics of the peptides (bottom panel) and MDM2 (right) are shown with a color scale going from red to blue (ranging from 21 kT/e to 1 kT/e). The F19, W23 and L26 pockets in MDM2 are indicated.
The results discussed here were obtained from combining statistics of all of these runs. When considering equilibrium behavior, only frames from 500 ps were considered, as the water density equilibrated within 500 ps. The corresponding equilibrated peptide-MDM2 distances were also evaluated. Electrostatic calculations Electrostatics were computed using the linearized Poisson– Boltzmann equation implemented in the Adaptive Poisson– Boltzmann software [66–68] Tools plugin in PyMOL.[69] Protein charges were assigned based on the Amber force-field implemented in PDB2PQR.[70,71] The system temperature was 300K and monovalent salt concentration was set at 150 mM. All other settings were set to default values. For better visualization in Figure 2, the electrostatic potentials on the solvent accessible surface were mapped onto the van der Waals surface. Identifying interdomain waters Prior to identifying interdomain waters, heavy atoms interacting in the peptide–protein interface were first determined using a peptide–protein heavy atom distance cutoff of 5 A˚. ˚ of a That is, any protein heavy atom that comes within 5 A peptide heavy atom (and vice versa) is identified as an “interface atom.” Following work by Zhou et al., [62] a water molecule was considered to be interdomain if its minimum distance to a protein interface atom (d1) and its minimum distance to a peptide interface atom (d2) satisfied the condition d1 1 d2 D 1 d, where D is the peptide–protein distance, ˚ to account for atomic radii. This choice of d is and d 5 5 A ˚ used by Zhou et al. [62] as we found that larger than the 2.5 A for our system, due to the relatively (compared with protein– protein interfaces) small surface area of the interface, a choice of 2.5 A˚ resulted in too many important interdomain waters 776
Journal of Computational Chemistry 2015, 36, 773–784
being omitted. Using a d of 5 A˚ gave a good balance between capturing important interdomain waters while minimizing the number of peripheral waters (essentially noise). When considering interdomain waters around F19 and L26, only MDM2 heavy atoms that interact with the respective residues in the bound conformation were considered to be part of the interface. We ignored the interdomain waters around W23 to prevent double counting waters that were considered in the F19only and L26-only volumes; W23 is located between F19 and L26 and splits the interdomain volume into two. If waters that were found around W23 were not present around F19 or L26, they were counted into the L26-only space. Recent work to study the MDM2-p53 interdomain hydration properties used a grid-based approach to determine water density,[72] which required that the interface remained rigid for cleaner classification of the interdomain space. However, to more realistically capture interdomain hydration properties as a peptide approaches MDM2, we allowed protein/peptide side chains to move and adjust to maximize hydrogen bonding with waters. The distance-based approach to determine interdomain waters is more appropriate for our analysis. We also avoid any issues of grid edge effects by using the distancebased classification. Evaluating interdomain water density To estimate the interdomain water density relative to bulk, we first determined the number of bulk waters that can fit into ˚ )3 TIP3P[60] the same interdomain space. We set up a (60 A water-box simulation for 200 ps with parameters as described above, and used the last simulation frame as the reference water box. The waters in each peptide-MDM2 simulation frame were then removed, and the peptide-MDM2 complex was placed inside the reference water box. Waters that resulted in steric clashes with MDM2 or the peptide were removed. The WWW.CHEMISTRYVIEWS.COM
FULL PAPER
WWW.C-CHEM.ORG
numbers of bulk waters in the interdomain space were then calculated, as before, using these constructed frames. The interdomain water density can be estimated with the relative number of waters (from simulation) to those from the bulk (using the reference water box). However, we found that this approach resulted in significant noise in the density distribution, as the difference of a single water molecule in the bulk calculation leads to significant changes in the evaluated relative density. To circumvent this, we determined the dependence of the number of bulk waters on the interdomain space. The interdomain space is proportional to the distance of the displaced peptide to MDM2. We linearly interpolated the number of bulk waters at the displacements to estimate the expected number of bulk waters for a given space. This was repeated for all peptides, and for the different volume spaces (i.e., full interface space, F19-only space and L26-only space). When considering the full interface space, it was found that a single line was not adequate to describe the dependence, and two piece-wise linear fits (data for distance
