Article pubs.acs.org/JPCB
Trp-Cage Folding on Organic Surfaces Zachary A. Levine,†,‡ Sean A. Fischer,§ Joan-Emma Shea,*,†,‡ and Jim Pfaendtner*,§ †
Department of Physics, University of California Santa Barbara, Santa Barbara, California 93106, United States Department of Chemistry and Biochemistry, University of California Santa Barbara, Santa Barbara, California 93106, United States § Department of Chemical Engineering, University of Washington, Seattle, Washington 98105, United States ‡
S Supporting Information *
ABSTRACT: Trp-cage is an artificial miniprotein that is small, stable, and fast folding due to concerted hydrophobic shielding of a Trp residue by polyproline helices. Simulations have extensively characterized Trp-cage; however, the interactions of Trp-cage with organic surfaces (e.g., membranes) and their effect on protein conformation are largely unknown. To better understand these interactions we utilized a combination of replica-exchange molecular dynamics (REMD) and metadynamics (MetaD), to investigate Trp-cage folding on self-assembled monolayers (SAMs). We found that, with REMD and MetaD, Trp-cage strongly binds to neutral CH3 surfaces (−25kT) and moderately adsorbs to anionic COOH interfaces (−7.6kT), with hydrophobic interactions driving CH3 adhesion and electrostatic attractions driving COOH adhesion. Similar to solid-state surfaces, SAMs facilitate a number of intermediate Trp-cage conformations between folded and unfolded states. Regarding Trp-cage’s aromatic groups in neutral CH3 systems, Tyr becomes oriented parallel to the surface in order to maximize hydrophobic interactions while Trp remains caged perpendicular to the surface; however, Trp can reorient itself parallel to the interface as the miniprotein more closely binds to the surface. In contrast, Tyr and Trp are both repelled from COOH surfaces, though the Trp-cage still adheres to the anionic interface via Lys and its N-terminated Asn residue.
■
stability of Trp-cage11,12 and its associated mutants13 through the construction of pressure−temperature profiles. In addition, direct molecular dynamics simulations have estimated the folding time of Trp-cage to be 5.5 μs at room temperature,14 in good agreement with experimental measurements at 4 μs.4 MetaD simulations, which bias sampling of phase space according to coarse descriptors (collective variables), reveal additional intermediate Trp-cage structures between the folded and unfolded states that resemble globular morphologies,15 similar to those seen in some experiments.5 Overall, many computational techniques have been employed to study Trpcage such as the use of metadynamics variants,16 MD with explicit water,17−19 MD with implicit water,20 Monte Carlo sampling,21,22 rate constant calculations,23 the use of multiple water models,24 the use of polarizable force fields,25 confinement between flat graphene sheets,26 and even confinement inside of fullerene.27 Despite the large body of literature that seeks to characterize Trp-cage in a variety of complex environments, very little is known about how it interacts with organic surfaces such as cellular membranes, which are commonplace in nearly every major biological system. In practice, however, it is difficult to simulate protein folding near organic surfaces due to the destabilization of soft interfaces during enhanced sampling protocols. In REMD, temperatures
INTRODUCTION The 20-residue miniprotein known as Trp-cage (AcNLYIQWLKDGGPSSGRPPPS-Nme) is an artificially synthesized protein which utilizes stable binding motifs from the 39residue extendin-4 peptide, where researchers observed that polyproline caging of a single hydrophobic Trp residue resulted in protein stabilization over a large range of osmotic pressures and temperatures.1 Trp-cage is further stabilized by the presence of Arg-Asp salt bridges,2 and unfolds only above 44 °C according to differential scanning calorimetry and circular dichroism,3 where after 4 μs4 it is 95% folded at room temperature and physiological pH.1 Initial characterizations of Trp-cage indicated that it could only populate two distinct states (folded and unfolded);3 however, later experiments revealed a multitude of intermediate states which ranged from globules5 to partially denatured structures,6 all of which expose Trp to varying amounts of water. Overall, Trp-cage provides a unique testing bench to investigate protein folding over a wide variety of physiological and nonphysiological environments, varying from individual amino acid substitutions7 to ureafacilitated denaturation.8 Molecular dynamics (MD) simulations, augmented with enhanced sampling protocols such as replica-exchange molecular dynamics (REMD)9 and metadynamics10 (MetaD), is well-suited for investigating the atomic mechanisms of Trpcage folding and stabilization at spatial and temporal scales not currently accessible through experiments. Simulations have revealed detailed information about the thermodynamic © XXXX American Chemical Society
Received: May 1, 2015 Revised: July 22, 2015
A
DOI: 10.1021/acs.jpcb.5b04213 J. Phys. Chem. B XXXX, XXX, XXX−XXX
Article
The Journal of Physical Chemistry B
anionic systems, sodium counterions were also inserted to ensure that each system was net neutral. Box dimensions were on the order of 4.5 × 5 × 7 nm3, and periodic boundary conditions were implemented to mitigate system size effects and reduce computation. To avoid periodic interactions at the top of the simulation box with the bottom of the sulfur substrate, a repulsive harmonic potential was implemented at 6 nm, which only acted on protein atoms, and not solvents or electrolytes that were added later. The system was then hydrated at room temperature where a semi-isotropic barostat held the xy box dimensions fixed, while the z dimension was allowed to fluctuate in order to maintain 1 bar of pressure. A natively folded Trp-cage protein1 (with +1e net charge) was then added to bulk solution above the SAM, alongside a chloride counterion. After bulk water density was re-established with 1 bar of pressure in z, the z box dimension was fixed, and the system was ready to be simulated under an NVT ensemble. Replica-Exchange Molecular Dynamics Simulations. REMD simulations were performed using GROMACS 4.6.136 on the Stampede supercomputer at the Texas Advanced Computing Cluster. Trp-cage and SAM topologies were derived from the AMBER99SB-ILDN34 force field which contains updated side-chain potentials for the I, L, D, and N amino acids in comparison to AMBER99SB.31 TIP3P37 rigid water was used to hydrate each system, resulting in about 4000 water molecules per simulation box. Systems were first equilibrated using a weakly coupled semi-isotropic Berendsen38 barostat at 1 bar with a isothermal compressibility of 4.5 × 10−5 bar−1, and a velocity-rescaled39 thermostat which held temperature constant at 300 K. When the z box dimensions converged, the box volume was subsequently fixed, and the system was switched over to an NVT ensemble for the remainder of the study. A leapfrog algorithm was used to integrate Newton’s equations of motion with an integration time step of 2 fs. Trp-cage and SAM molecular bonds were constrained using the LINCS algorithm,40 and water bonds were constrained using the SETTLE algorithm.41 Short-range van der Waals and Coulombic forces were truncated at 1 nm, after which long-range interactions were tabulated using the particle mesh Ewald (PME) algorithm42 which utilizes fast Fourier transformations. To further enhance the sampling frequency of Trp-cage in molecular dynamics simulations, we employed the replica-exchange technique43 across 50 replicas that ranged in temperature from 288 to 506 K, where we enforced an exchange rate of 25% between replicas. Exchanges between replicas were attempted every 3 ps to allow an adequate mixing of states to occur. Replicas were initially heated for 10 ns followed by a 200 ns production run at constant temperature; however, only the final 100 ns were analyzed and summarized below. Metadynamics Simulations. Parameters for the MetaD simulations10,15 were the same as those listed above for REMD simulations unless otherwise noted. MetaD was implemented in GROMACS on the Hyak supercomputing cluster at the University of Washington, using the PLUMED44 module. Our approach combines MetaD45,46 with replica-exchange47 within the well-tempered ensemble (WTE).48,49 For brevity we will refer to the combined approach simply as “MetaD” in order to distinguish it from the REMD simulations. The WTE enhances fluctuations in the potential energy of the system, thereby allowing a reduction in the number of replicas needed. Additionally since the WTE maintains the average energy of the original ensemble, canonical averages of properties can be
ranging from 290 to 520 K are often used to force proteins into energetically unfavorable conformations in order to overcome entropic free-energy barriers. Therefore, while it is favorable to subject proteins to very high temperatures in order to sample a large multiplicity of states, nearby lipid assemblies quickly break apart at high temperatures if they are heated beyond their melting points. Conversely, if lower temperatures are utilized in order to stabilize nearby membranes, then there is a reduction of available protein conformations that Trp-cage can populate, thus reducing the accuracy of REMD. To overcome these obstacles, the sampling of relevant phase space in the folding process is enhanced by targeting more localized collective variables such as the degree of protein helicity, the protein’s folding propensity, normal distance from a surface, and the magnitude of the protein’s Hamiltonian,28 or by artificially anchoring the tails of membrane lipids.29 The latter technique significantly dampens proper membrane motion and fluidity, thereby perturbing the conformational landscape of Trp-cage near a membrane’s surface. The former techniques, however, are the basis for MetaD simulations, though these additional degrees of freedom must be carefully selected to minimally disrupt nearby interfaces. Moreover, studies have shown that self-assembled monolayers30 (or SAMs) can be substituted for membranes in REMD simulations because they maintain their room-temperature orientations and long-range geometries under large thermal stresses, and can easily be constructed with hydrophobic, hydrophilic, neutral, or anionic headgroups. Other obstacles include adequately sampling the time scales required to observe convergent protein behaviors, which in the case of Trp-cage can take place on the order of microseconds,12 much longer than most REMD simulations, which typically span a few hundred nanoseconds per replica. All of these issues are further complicated by the choice of molecular force field,31 which can often yield disparate Trp-cage morphologies and α-helical propensities.11,12,32,33 In this study, we performed a combination of REMD and MetaD simulations on a prefolded Trp-cage miniprotein for 200 ns/replica, in the presence of CH3-terminated (neutral) and COOH-terminated (anionic) self-assembled monolayers. The goals of this study are to (1) ascertain how Trp-cage folding is affected on the surface of neutral and anionic membrane-mimics (SAMs), (2) demonstrate that for protein/ surface interactions that tend to stabilized the folded state that REMD and MetaD approaches provide very complementary data, (3) investigate the adhesion energies of Trp-cage to organic surfaces, (4) investigate how surfaces and temperature affect the Trp-caging motif, and (5) compare Trp-cage behaviors on the surface of SAMs across two popular molecular force fields, AMBER99SB-ILDN34 and AMBER03.35 Elucidating the interactions between Trp-cage and organic surfaces will inevitably help clarify how membranes and other physiological surfaces affect globular protein behavior, thus yielding a deeper understanding of protein−membrane interactions in biological systems.
■
METHODS Systems and Structures. SAM systems30 were constructed by creating a fixed 10 × 10 grid of sulfur atoms in a twodimensional diamond geometry at the bottom of a unit cell. Then, an 11-atom hydrocarbon chain (CH2)11 was attached to each anchor, terminated by either a CH3 group (for neutral interfaces) or a COOH group (for anionic interfaces). In B
DOI: 10.1021/acs.jpcb.5b04213 J. Phys. Chem. B XXXX, XXX, XXX−XXX
Article
The Journal of Physical Chemistry B Table 1. Metadynamics Collective Variables CV1 Trp-cage distance to surface
CV2 α-helical hydrogen bonds (r ≤ 2.5 Å)
CV3 hydrophobic core contacts (r ≥ 5.0 Å)
SAM/FF
min (nm)
max (nm)
min (#)
max (#)
min (#)
max (#)
CH3/A99 COOH/A99 CH3/A03
0.3 0.4 0.2
4.1 4.1 4.1
0 0 0
7.6 7.6 7.6
4 4 4
14.7 14.7 14.6
obtained through reweighing techniques.50,51 We chose to bias three collective variables (CVs) that are described in Table 1, which modulate the distances between the center of mass of Cα atoms from the SAM surface, hydrogen bonds related to protein secondary structure, and the hydrophobicity of the protein core. Details of the hydrogen bond and hydrophobic core CVs can be found in our previous work.49 Gaussians were deposited every 1 ps with an initial height of 2 kJ/mol and a bias factor of 12. The widths of the Gaussians for the CVs were 0.05, 0.1, and 0.2 kJ/mol, respectively. For the replica-exchange portion, we employed 20 replicas that spanned a temperature range between 300 and 450 K, with exchanges attempted every 0.5 ps. Systems were equilibrated for 1 ns under the NVT and NPT ensembles under the same parameters as the REMD simulations. Also, as we have previously reported,52 a 10 ns preproduction simulation (where only the potential energy was biased) was used to establish the WTE. This potential was subsequently used as a static contributor to the total potential energy during the production simulations that followed. Analysis and Tools. Molecular graphics were generated with Visual Molecular Dynamics (VMD) 1.9.1.53 The GROMACS analysis tools g_hbond, g_traj, g_gyration, and g_cluster were used to measure the probability of Trp-cage intramolecular hydrogen bonds, protein end-to-end distance (Ree), the radius of gyration (Rg), and clustering of dominant protein morphologies at room temperature. Hydrogen bonds were defined as bonds that had an O−H separation of 0.25 nm or smaller, along with an O−H−N angle of 30° or less. Ree was measured from the N-Asn center of mass to the C-Ser center of mass. Trp-cage conformations were clustered together according to the Daura criterion,54 which compares protein backbones (excluding terminal amino acids) and groups them together on the basis of a root-mean-square cutoff of 0.14 nm or less. When protein clusters were extracted from MetaD, a sparsely sampled trajectory (updated every 20 ps) was utilized and reweighted by correcting for the biases used to sample the free-energy surface (i.e., by assigning each frame to a grid point on the free-energy surface calculated from the bias potential, and then taking exp(−F/kT)/N, where F is the free energy at a grid point, kT is the usual thermodynamic constant at room temperature, and N is the number of frames assigned to that grid point). The fraction of native Trp-cage contacts was defined by first counting the number of Cα−Cα pairs which were no further than 6.5 Å in bulk solution (excluding nearest neighbors). Then, during the simulation we monitored the number of remaining Cα−Cα pairs, and normalized that amount by the initial number, which yielded a distribution between 0 and 1. Secondary structures were extracted using the DSSP tool,55 and RMSD comparisons to low-energy NMR structures were performed on both Trp-cage’s α-helical region (RMSDα‑helix) which encompassed the C, O, N, and H atoms in residues 2−8, while RMSDCα was calculated from Cα atoms in residues 9−19. Probability distributions for unbiased variables in the MetaD simulations were obtained with the commonly
used reweighting algorithm50 and associated code distributed with PLUMED.44 The convergence of the REMD and MetaD simulations was assessed by ensuring the relevant free energies were not changing at the end of the simulation. Data showing the convergence of simulations is provided in Figures S13 and 14.
■
RESULTS AND DISCUSSION Trp-Cage More Stable on −COOH Than −CH 3 Surfaces. The dominant Trp-cage conformations that emerge from REMD are morphologically similar to those observed in MetaD simulations. Figure 1 displays the most prevalent
Figure 1. REMD clusters of dominant Trp-cage backbones with select side chains displayed on the surface of SAMs. Numbers correspond to the relative time spent in each conformation, where they are lettered A−C. Purple backbones represent α-helical regions, blue backbones represent 3−10 helical regions, cyan backbones represent turns, and white backbones represent protein coils. Dark green side chains represent Tyr, gray represents Trp, brown represents Pro, bright green represents Lys, red represents Asp, yellow represents Leu, orange represents Gln, violet represents Arg, and black represents N-Asn.
structures at 300 K for the REMD simulations, and Figure S1 shows both REMD and MetaD structures. Overall the miniprotein remained relatively compact on both surfaces, retaining native-like folds that are similar to the NMR structure (Figure S2) in its most populated conformations. Free-energy surfaces obtained from both REMD and MetaD simulations can be found in Figure 2 and Figure S3. The Cα backbone diverges from its ideal NMR structure (PDB: 1L2Y) by 1−2 Å on CH3 surfaces, and by less than 1 Å on COOH surfaces. Bulk studies by Day12 show the existence of two, distinctly folded (
