Higher Accuracy Achieved in the Simulations of Protein Structure

Dec 5, 2018 - The accuracy of molecular mechanics force fields is of vital importance in biomolecular simulations. However, the admittedly more accura...
0 downloads 0 Views 1MB Size
Download PDF
Letter Cite This: J. Phys. Chem. Lett. 2018, 9, 7110−7116

pubs.acs.org/JPCL

Higher Accuracy Achieved in the Simulations of Protein Structure Refinement, Protein Folding, and Intrinsically Disordered Proteins Using Polarizable Force Fields Anhui Wang,†,‡ Zhichao Zhang,‡ and Guohui Li*,† †

Downloaded via TULANE UNIV on December 8, 2018 at 02:32:04 (UTC). See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

Laboratory of Molecular Modeling and Design, State Key Laboratory of Molecular Reaction Dynamics, Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian 116023, China ‡ State Key Laboratory of Fine Chemicals, School of Chemistry, Dalian University of Technology, Dalian 116024, China S Supporting Information *

ABSTRACT: The accuracy of molecular mechanics force fields is of vital importance in biomolecular simulations. However, the admittedly more accurate polarizable force fields were recently reported to be less able to reproduce the experimental properties in comparison to additive force fields in some cases. Here, we perform long-time-scale molecular dynamics simulations to systematically evaluate the effect of explicit electronic polarization in polarizable force fields. The results show that the inclusion of electrostatic polarization effect in polarizable force fields can improve their accuracies in protein structure refinement and generate conformational ensembles more approximate to experiments for intrinsically disordered proteins. In contrast, it is difficult for polarizable force fields to approach the native structure, let alone to predict the native state when it is unknown a priori in the real protein structure predictions. We speculate that these effects might be attributed to the preference of protein−water interactions in polarizable force fields.

W

PFFs show improved solvation of the peptide over CFFs, leading to relatively stronger protein−water interactions.21−24 As a consequence of these advances, it is of essential urgency to systematically evaluate the recently improved CFFs and PFFs, especially the effect of stronger protein−water interactions in the simulations where PFFs are rarely used. In this work, we report a systematic evaluation of CFFs and PFFs for their accuracies in a variety of systems, including protein structure refinement, protein folding, and conformational ensemble generation of intrinsically disordered proteins using long-time-scale MD simulations (∼40 μs for each force field). The widely used AMBER99SB7 (AMBER herein) and CHARMM36/CHARMM36m5,6 (CHARMM herein) are selected to represent CFFs, while AMOEBA-201311 (AMOEBA herein) and Drude-201310 (Drude herein) are used to represent PFFs. Protein Structure Refinement. Here, we first performed a baseline study on native state dynamics of globular proteins. Three independent 50 ns unrestrained simulations at 300 K for each of the 18 experimental structures from CASP1125 targets were carried out to see whether the force field can stabilize their native structures. The selected 18 targets with sequence length ranging from 106 to 450 residues share no more than 25% sequence identity and thus represent different fold-level

ith the rapid growth of computing power, molecular dynamics (MD) simulation has emerged as an indispensable tool for investigating complicated biological macromolecules as it can provide an atomistic description of their structures, dynamics, and functions. Despite its significant advantage of obtaining information not available via experimental techniques, MD’s predictive capabilities are still limited by the accuracy of the molecular mechanics force fields (FFs) which serve as the backbone of MD simulation.1 According to the difference in describing electrostatic properties, these force fields can be divided into two groups: classical additive force fields and polarizable force fields.2−4 In the widely used classical additive force fields (CFFs), such as AMBER, CHARMM, GROMOS, and OPLS, the electrostatic term is described with Coulombic interactions between fixed partial charges.5−8 In polarizable force fields (PFFs), such as AMOEBA and Drude, the induced dipole or Drude oscillator model is implemented to account for the electronic polarization effect during the simulation process.9−11 Polarizable force fields have proved to be computationally more accurate than CFFs in the simulations of highly polar versus hydrophobic environments for their explicit inclusion of charge redistribution due to the change in the surrounding electrostatic field.12−17 On the other hand, in addition to the application scenarios constrained by their massive computational cost, PFFs were recently reported to be less able to reproduce the experimental properties compared with CFFs in some cases.10,18−20 Furthermore, as revealed in previous contributions, the amino acid hydration free energies in © XXXX American Chemical Society

Received: November 15, 2018 Accepted: December 4, 2018 Published: December 5, 2018 7110

DOI: 10.1021/acs.jpclett.8b03471 J. Phys. Chem. Lett. 2018, 9, 7110−7116

Letter

The Journal of Physical Chemistry Letters

experimental structures. Next, three independent 50 ns standard MD simulations initiating from these model structures were carried out for each force field. Because the best structure could not be singled out on the basis of its RMSD to the experimental structure, which is missing in real protein structure refinement, a filtering and averaging approach proposed by Feig and co-workers26,27 was employed to pick out the representative structure from the MD simulations, and its RMSD compared to the experimental structure was calculated later. Then the ΔRMSD (representative structure’s RMSD minus model structure’s RMSD) was used to evaluate the refinement accuracy. During the three independent 50 ns MD simulations with different FFs at 300 K, AMBER, CHARMM, AMOEBA, and Drude force fields can produce refinement for 12, 11, 14, and 16 model structures, respectively (with ΔRMSD < 0 listed in Table S2). Because of some previous reports28,29 that unrestrained MD simulations starting from homology models at 300 K usually drift away from their experimental structures rapidly, we also checked the refinement performance of the initial three independent 5 ns MD simulations. From Figure 1B, AMBER and CHARMM CFFs can successfully refine most (16 and 14) of the 18 targets with average ΔRMSD values of −0.49 Å and −0.44 Å. Interestingly, both AMOEBA and Drude PFFs can successfully refine all the tested targets with better average ΔRMSD of −0.59 Å. Although for some models the overall structure improvements using PFFs are not so prominent with few of the targets reaching ΔRMSD over −1.0 Å, the simulation results are in good agreement with previous reports that MD simulations can produce moderate but consistent refinement for most systems.30−32 As mentioned earlier, the improved solvation of peptide in PFFs will lead to relatively weaker protein−protein interactions, in favor of the protein−water interactions. Consequently, the better refinement performance of PFFs might benefit from the weaker protein−protein interactions that facilitate a broader conformation search space in which the conformational states close to their experimental structures can be picked out by the latter scoring and filtering algorithm. Protein Folding. In addition to protein structure refinement, protein folding remains a long-standing challenge for the MD simulations of biomolecules in which the accurate reproduction of the folded-state conformation needs high-precision FFs.33−35 Fast-folding proteins of diverse secondary structures usually fold on microsecond time scales and thus serve as perfect models to study protein folding or to evaluate FF accuracy.36 We selected four fast-folding proteins that contain different secondary structureschignolin (β-hairpin or turn), Trp-cage (two α-helices), villin headpiece (three α-helices), and WW domain (three-stranded β-sheet)to check how the exper-

topologies. As shown in Figure 1A, almost all of the MD simulations with different FFs can keep the proteins close to

Figure 1. Average RMSD distribution of experimental structures (A) and ΔRMSD distribution of model structures (B) in MD simulations with different FFs. The mean values of each distribution are shown as stars, and the outliers are shown as plus signs. The average RMSD of each experimental structure is calculated from three independent trajectories. The detailed RMSD and ΔRMSD data of all targets are listed in Figures S1−S4 and Tables S1 and S3.

their experimental states (with average Cα RMSD < 3.0 Å). Among the 18 tested proteins, only the simulations of target T0841 with Drude PFF can reach a RMSD larger than 3.0 Å in which one trajectory (colored by red in Figure S4) drifts away from the experimental structure and shows striking fluctuation. The quite small average RMSDs and low fluctuations (RMSD averages and standard deviations for every single protein are listed in Table S1) both demonstrate experimental structures with various topologies can be stable throughout the simulations; thus, CFFs, as well as the PFFs mentioned above, are accurate enough to be applied to protein structure refinement. To test whether structural refinement can be achieved, a model structure was first constructed for each of the 18 CASP targets. This procedure was completed by our in-house deep learning algorithm based on the residue sequence of the target structure, and the generated model structures show a broad Cα RMSD distribution (2−9 Å, data listed in Table S3) to their

Table 1. Summary of the MD Simulations for the Four Fast-Folding Proteins minRMSDc (Å) system

Nres

chignolin Trp-cage villin WW domain

10 20 35 38

a

SS

b

β-hairpin two α-helices three α-helices three-stranded β-sheet

avgRgd (Å)

AMBER

CHARMM

AMOEBA

Drude

AMBER

CHARMM

AMOEBA

Drude

0.17 0.34 0.32 3.49

0.13 1.74 2.55 3.61

1.13 2.85 4.34 4.48

1.76 2.35 3.54 2.71

5.82 7.77 9.84 9.90

6.25 8.84 11.30 11.80

6.88 10.65 18.38 16.52

6.62 10.45 14.60 16.90

a

Number of residues in each protein. bSecondary structures in each protein. cMinimal Cα RMSD relative to the native structure during simulations. dReweighted average radius of gyration, Rg, during the simulation with different FFs. 7111

DOI: 10.1021/acs.jpclett.8b03471 J. Phys. Chem. Lett. 2018, 9, 7110−7116

Letter

The Journal of Physical Chemistry Letters imentally determined native structure can be properly achieved by the MD simulations initiating from extended conformation using CFFs or PFFs. A fully extended structure was constructed based on the residue sequence for each of the four selected proteins; then, for every force field, four independent 1 μs accelerated molecular dynamics37,38 (aMD) enhanced sampling simulations starting from the extended conformation were separately carried out. The minimal Cα RMSD values relative to the corresponding native structure during simulations of different FFs are listed in Table 1. For the smaller chignolin and Trp-cage with simple secondary structure elements, both CFFs and PFFs can obtain near-native structures with Cα RMSD ranging from 0.1 to 2.9 Å. For villin headpiece with three α-helices, AMBER and CHARMM CFFs can sample conformations with Cα RMSD within 2.5 Å, whereas PFFs can search conformations only over 3.5 Å. For the WW domain with a three-stranded β-sheet, neither CFFs nor PFFs can sample conformations very close to native structures, although the Drude PFF can produce conformations of 2.7 Å. Therefore, the microsecond time scale enhanced sampling simulations with CFFs can sample conformations much closer to the native structures as indicated by the lower minimum Cα RMSD values (corresponding structures shown in Figure S5). To further confirm whether these FFs can predict a qualitatively reasonable structure when the native structure is unknown a priori, we checked whether the near-native structure is the most favorable during the simulations using the reweighted distribution of RMSDs shown in Figure 2. For chignolin and Trp-cage, both AMBER and CHARMM CFFs can locate the maximum of the distribution close to the native state (RMSD below 0.7 or 2.5 Å for chignolin and Trp-cage, respectively). For proteins with more complicated secondary structures, only AMBER CFF can locate the secondary

maximum of the distribution close to the native state (RMSD around 1.2 Å) for the villin headpiece. In contrast, for the four proteins studied here, neither AMOEBA nor Drude PFF favors the native states as indicated by the location of the maximum of the RMSD distribution away from native states. To summarize the simulations of protein folding, CFFs can successfully predict as well as favor the native states for chignolin, Trp-cage, and villin headpiece, and conversely, PFFs can fold only the smaller chignolin and Trp-cage with no preference toward their native states. We speculate that the slightly worse accuracy of PFFs in predicting native structures of the four fast-folding proteins studied here might be the result of the stronger protein−water interactions which weaken the protein−protein interaction in PFFs. To confirm this assumption, we calculated the reweighted distribution of the radius of gyration (Rg) and its average value during each simulation. It can be seen in Figure 3 that the Rg

Figure 3. Probability density distribution of radius of gyration, Rg, for the conformations of (A) chignolin, (B) Trp-cage, (C) villin headpiece, and (D) WW domain from simulations with different FFs. The color scheme is consistent with Figure 2.

distributions of AMBER and CHARMM CFFs are quite similar, and they are much narrower than those of PFFs. Consistently, the average Rg of CFFs is also smaller than that of PFFs (data listed in Table 1). It is in agreement with previous contributions reporting that unfolded states are over collapsed in CFFs and that more extended conformations can be sampled in simulations of Drude PFF.23,34,39,40 Conformational Ensemble Generation of Intrinsically Disordered Proteins. However, it is highlighted in recent studies that better balanced protein−water interactions can notably improve the agreement between simulations and experiments in the investigation of intrinsically disordered proteins (IDPs) which show high conformational heterogeneity and do not adopt a well-defined 3D structure under physiological conditions.41−44 Because PFFs strengthen the protein−water interactions and show a tendency to sample more extended conformations in

Figure 2. Probability density distribution of Cα RMSD of the conformations of (A) chignolin, (B) Trp-cage, (C) villin headpiece, and (D) WW domain from simulations with different FFs. Simulations with AMBER, CHARMM, AMOEBA, and Drude are colored in red, purple, blue, and green, respectively. 7112

DOI: 10.1021/acs.jpclett.8b03471 J. Phys. Chem. Lett. 2018, 9, 7110−7116

Letter

The Journal of Physical Chemistry Letters

Figure 4. Comparison of structural measures calculated from the conformational ensembles generated by simulations with each FF (the color scheme is consistent with Figure 2). (A) Average radius of gyration, Rg. The black line and gray band show the power-law relationship estimated by SARW model Rg = R0Nν with R0 = 1.927 Å and ν = 0.598 ± 0.028. (B) Average number of intrapeptide hydrogen bonds (HBs). (C) Average unsigned error compared to experimentally measured chemical shifts. (D) Average 3JHNHA couplings for RS. The experimental 3JHNHA couplings are colored in black. For panels A−C, pepG, p53, NTL9, and RS are shown as circles, stars, pentagons, and triangles, respectively.

shown in Figure 4B also demonstrate the weaker protein− protein interactions in PFFs. For all IDPs, AMBER CFF has the maximum number of intrapeptide hydrogen bonds, indicating the most compact conformational ensembles. To distinguish the structural differences of ensembles obtained from different FFs, the secondary structure contents were also calculated. As shown in Figure 5, AMBER CFF samples the fewest coils together with a comparable amount of helices, whereas PFFs, as well as CHARMM36m, can reach at least 70% population of coil conformations. The only exception is that the simulation of RS using Drude PFF samples almost equal populations for helix and coil and simultaneously produces more intrapeptide hydrogen bonds (shown in Figure 4B). This indicates that Drude PFF might suffer from the same drawback as CHARMM36, in which the population of α-helix in RS is overestimated.6,39 Because the structural properties of conformational ensembles depend heavily on the FFs, we further calculated the trajectory averaged chemical shifts and compared them with experimental data to determine which FF can generate more accurate conformational ensembles. For pepG, p53, and RS, the CA chemical shifts were calculated, whereas the HA chemical shifts of NTL9 were calculated instead for their experimental data availability. From the average unsigned errors (AUEs) compared to experimentally measured chemical shifts shown in Figure 4C, AMBER CFF produces larger AUE over CHARMM36m or PFFs for pepG and p53, while both CFFs and PFFs can produce comparable AUEs for NTL9. Like

protein-folding simulations, we instinctively want to check their feasibility for the simulation of IDPs. Four IDPs that have been extensively characterized by experiments, including pepG,45 p53,46 NTL9,40 and RS,39 were selected to compare the conformational ensembles generated by CFFs or PFFs. One in particular is that the CHARMM36 CFF used in the aforementioned simulations of protein structure refinement and protein folding was replaced by its newly refined version CHARMM36m,6 which could generate significantly better backbone conformational ensembles for IDPs. Using the same simulation protocol as protein folding, a fully extended structure was constructed for each protein; then, four independent 1 μs aMD simulations initiating from the extended conformation were separately carried out for each force field. The resulting conformational ensembles generated by different FFs were subsequently examined by calculating the reweighted radius of gyration, intrapeptide hydrogen bonds, secondary structure contents, chemical shifts, and 3JHNHA couplings. As shown in Figure 4A, the stronger protein−water interaction of PFFs can be reflected by the chain dimensions. For all IDPs studied here, PFFs show larger average Rg values than AMBER CFF, yielding a more approximate chain dimensions to the theoretically estimated values by SARW model47 (displayed as the black line). As is expected, the newly refined CHARMM36m CFF can produce moderate Rg values between that of AMBER and PFFs. Similar to the chain dimensions, the numbers of intrapeptide hydrogen bonds 7113

DOI: 10.1021/acs.jpclett.8b03471 J. Phys. Chem. Lett. 2018, 9, 7110−7116

Letter

The Journal of Physical Chemistry Letters

from the development of high-precision force fields. It should be noted that the different accuracy between CFFs and PFFs observed here might be the influence of more complicated factors, not just the interactions of protein−water or protein− protein discussed in this work. Limited by the massive computational cost of PFFs, our simulations of folding and IDPs include merely several cases which should be expanded in further studies. Despite this, the quantitative comparisons in this work may lay the foundation for future developments of both CFFs and PFFs.

■

ASSOCIATED CONTENT

S Supporting Information *

The Supporting Information is available free of charge on the ACS Publications website at DOI: 10.1021/acs.jpclett.8b03471. Detailed simulation methods, trajectory analyses, supporting figures and tables that list the trajectory RMSDs of protein structure refinement, and a supporting figure that shows the conformations with minimal Cα RMSD sampled during MD simulations of protein folding (PDF)

■

Figure 5. Secondary structure contents of (A) pepG, (B) p53, (C) NTL9, and (D) RS calculated from the conformational ensembles generated by simulations with each FFs. The color scheme is consistent with Figure 2.

AUTHOR INFORMATION

Corresponding Author

*E-mail: [email protected]. ORCID

Anhui Wang: 0000-0003-4041-9095 Guohui Li: 0000-0001-8223-705X

its discrepancy in characterizing intrapeptide hydrogen bond and secondary structure content, the simulation of RS using Drude PFF shows even larger AUE than AMBER CFF. To roundly reveal this discrepancy, we report in Figure 4D the 3 JHNHA couplings for RS residues, in which Drude PFF shows the maximum difference from the experimental values (black line), whereas AMOEBA and CHARMM36m can achieve better consistency with experiments. This again suggests the overestimation of the α-helix population for RS simulation in Drude PFF. Apart from Drude’s simulation of RS, the larger chain dimensions and more extended conformational ensembles sampled by PFFs lead to better or at least comparable agreement with experiments, demonstrating they are more suitable for the simulation of IDPs studied here. In summary, we performed long-time-scale MD simulations to systematically assess the accuracies of CFFs and PFFs for the simulations of protein structure refinement, protein folding, and conformational ensemble generation of IDPs. The results show that the inclusion of electrostatic polarization effect alters the preference of protein−water interactions in PFFs and causes accuracy differences in the simulation of different systems. For protein structure refinement, both AMOEBA and Drude PFFs can successfully refine all the targets with moderate improvements. Benefiting from the stronger protein−water interactions, PFFs can also sample more extended conformational ensembles which are more approximate to experiments in the simulations of IDPs. However, it is difficult for PFFs to approach the native structure, let alone to predict the native state when it is unknown a priori in the real protein structure predictions. For the folding of the WW domain, the aMD enhanced sampling of all FFs used in our work could not achieve the near-native conformation. This suggests more efficient sampling methods are also needed in the simulation of complicated systems apart

Notes

The authors declare no competing financial interest.

■

ACKNOWLEDGMENTS This work is supported by grants from the National Natural Science Foundation of China under Contract No. 31700647, No. 21625302, and No. 21573217.

■

REFERENCES

(1) Nerenberg, P. S.; Head-Gordon, T. New developments in force fields for biomolecular simulations. Curr. Opin. Struct. Biol. 2018, 49, 129−138. (2) Baker, C. M. Polarizable force fields for molecular dynamics simulations of biomolecules. Wiley Interdisciplinary Reviews: Computational Molecular Science 2015, 5 (2), 241−254. (3) Chu, H.; Cao, L.; Peng, X.; Li, G. Polarizable force field development for lipids and their efficient applications in membrane proteins. Wiley Interdisciplinary Reviews: Computational Molecular Science 2017, 7 (5), e1312. (4) Huang, J.; Lopes, P. E.; Roux, B.; MacKerell, A. D., Jr. Recent Advances in Polarizable Force Fields for Macromolecules: Microsecond Simulations of Proteins Using the Classical Drude Oscillator Model. J. Phys. Chem. Lett. 2014, 5 (18), 3144−3150. (5) Best, R. B.; Zhu, X.; Shim, J.; Lopes, P. E.; Mittal, J.; Feig, M.; Mackerell, A. D., Jr. Optimization of the additive CHARMM all-atom protein force field targeting improved sampling of the backbone phi, psi and side-chain chi(1) and chi(2) dihedral angles. J. Chem. Theory Comput. 2012, 8 (9), 3257−3273. (6) Huang, J.; Rauscher, S.; Nawrocki, G.; Ran, T.; Feig, M.; de Groot, B. L.; Grubmuller, H.; MacKerell, A. D., Jr. CHARMM36m: an improved force field for folded and intrinsically disordered proteins. Nat. Methods 2017, 14 (1), 71−73. (7) Klepeis, J. L.; Lindorff-Larsen, K.; Dror, R. O.; Shaw, D. E. Longtimescale molecular dynamics simulations of protein structure and function. Curr. Opin. Struct. Biol. 2009, 19 (2), 120−7. 7114

DOI: 10.1021/acs.jpclett.8b03471 J. Phys. Chem. Lett. 2018, 9, 7110−7116

Letter

The Journal of Physical Chemistry Letters (8) Schmid, N.; Eichenberger, A. P.; Choutko, A.; Riniker, S.; Winger, M.; Mark, A. E.; van Gunsteren, W. F. Definition and testing of the GROMOS force-field versions 54A7 and 54B7. Eur. Biophys. J. 2011, 40 (7), 843−56. (9) Jiang, W.; Hardy, D. J.; Phillips, J. C.; Mackerell, A. D., Jr.; Schulten, K.; Roux, B. High-performance scalable molecular dynamics simulations of a polarizable force field based on classical Drude oscillators in NAMD. J. Phys. Chem. Lett. 2011, 2 (2), 87−92. (10) Lopes, P. E.; Huang, J.; Shim, J.; Luo, Y.; Li, H.; Roux, B.; Mackerell, A. D., Jr. Force Field for Peptides and Proteins based on the Classical Drude Oscillator. J. Chem. Theory Comput. 2013, 9 (12), 5430−5449. (11) Shi, Y.; Xia, Z.; Zhang, J.; Best, R.; Wu, C.; Ponder, J. W.; Ren, P. The Polarizable Atomic Multipole-based AMOEBA Force Field for Proteins. J. Chem. Theory Comput. 2013, 9 (9), 4046−4063. (12) McDaniel, J. G.; Yethiraj, A. Influence of Electronic Polarization on the Structure of Ionic Liquids. J. Phys. Chem. Lett. 2018, 9 (16), 4765−4770. (13) Peng, X.; Zhang, Y.; Chu, H.; Li, G. Free energy simulations with the AMOEBA polarizable force field and metadynamics on GPU platform. J. Comput. Chem. 2016, 37 (6), 614−22. (14) Qi, R.; Jing, Z.; Liu, C.; Piquemal, J. P.; Dalby, K. N.; Ren, P. Elucidating the Phosphate Binding Mode of Phosphate-Binding Protein: The Critical Effect of Buffer Solution. J. Phys. Chem. B 2018, 122 (24), 6371−6376. (15) Gokcan, H.; Kratz, E.; Darden, T. A.; Piquemal, J. P.; Cisneros, G. A. QM/MM Simulations with the Gaussian Electrostatic Model: A Density-based Polarizable Potential. J. Phys. Chem. Lett. 2018, 9 (11), 3062−3067. (16) Peng, X.; Zhang, Y.; Chu, H.; Li, Y.; Zhang, D.; Cao, L.; Li, G. Accurate Evaluation of Ion Conductivity of the Gramicidin A Channel Using a Polarizable Force Field without Any Corrections. J. Chem. Theory Comput. 2016, 12 (6), 2973−82. (17) Sun, R. N.; Gong, H. Simulating the Activation of Voltage Sensing Domain for a Voltage-Gated Sodium Channel Using Polarizable Force Field. J. Phys. Chem. Lett. 2017, 8 (5), 901−908. (18) Bradshaw, R. T.; Essex, J. W. Evaluating Parametrization Protocols for Hydration Free Energy Calculations with the AMOEBA Polarizable Force Field. J. Chem. Theory Comput. 2016, 12 (8), 3871− 83. (19) Manzoni, F.; Soderhjelm, P. Prediction of hydration free energies for the SAMPL4 data set with the AMOEBA polarizable force field. J. Comput.-Aided Mol. Des. 2014, 28 (3), 235−44. (20) Ploetz, E. A.; Rustenburg, A. S.; Geerke, D. P.; Smith, P. E. To Polarize or Not to Polarize? Charge-on-Spring versus KBFF Models for Water and Methanol Bulk and Vapor-Liquid Interfacial Mixtures. J. Chem. Theory Comput. 2016, 12 (5), 2373−87. (21) Deng, Y. Q.; Roux, B. Hydration of amino acid side chains: Nonpolar and electrostatic contributions calculated from staged molecular dynamics free energy simulations with explicit water molecules. J. Phys. Chem. B 2004, 108 (42), 16567−16576. (22) Huang, J.; Mei, Y.; Konig, G.; Simmonett, A. C.; Pickard, F. C. t.; Wu, Q.; Wang, L. P.; MacKerell, A. D., Jr.; Brooks, B. R.; Shao, Y. An Estimation of Hybrid Quantum Mechanical Molecular Mechanical Polarization Energies for Small Molecules Using Polarizable ForceField Approaches. J. Chem. Theory Comput. 2017, 13 (2), 679−695. (23) Hazel, A. J.; Walters, E. T.; Rowley, C. N.; Gumbart, J. C. Folding free energy landscapes of beta-sheets with non-polarizable and polarizable CHARMM force fields. J. Chem. Phys. 2018, 149 (7), No. 072317. (24) Konig, G.; Pickard, F. C.; Huang, J.; Thiel, W.; MacKerell, A. D.; Brooks, B. R.; York, D. M. A Comparison of QM/MM Simulations with and without the Drude Oscillator Model Based on Hydration Free Energies of Simple Solutes. Molecules 2018, 23 (10), 2695. (25) Moult, J.; Fidelis, K.; Kryshtafovych, A.; Schwede, T.; Tramontano, A. Critical assessment of methods of protein structure prediction: Progress and new directions in round XI. Proteins: Struct., Funct., Genet. 2016, 84, 4−14.

(26) Heo, L.; Feig, M. What makes it difficult to refine protein models further via molecular dynamics simulations? Proteins: Struct., Funct., Genet. 2018, 86, 177−188. (27) Mirjalili, V.; Feig, M. Protein Structure Refinement through Structure Selection and Averaging from Molecular Dynamics Ensembles. J. Chem. Theory Comput. 2013, 9 (2), 1294−1303. (28) Raval, A.; Piana, S.; Eastwood, M. P.; Dror, R. O.; Shaw, D. E. Refinement of protein structure homology models via long, all-atom molecular dynamics simulations. Proteins: Struct., Funct., Genet. 2012, 80 (8), 2071−2079. (29) Mirjalili, V.; Noyes, K.; Feig, M. Physics-based protein structure refinement through multiple molecular dynamics trajectories and structure averaging. Proteins: Struct., Funct., Genet. 2014, 82, 196−207. (30) Hovan, L.; Oleinikovas, V.; Yalinca, H.; Kryshtafovych, A.; Saladino, G.; Gervasio, F. L. Assessment of the model refinement category in CASP12. Proteins: Struct., Funct., Genet. 2018, 86, 152− 167. (31) Nugent, T.; Cozzetto, D.; Jones, D. T. Evaluation of predictions in the CASP10 model refinement category. Proteins: Struct., Funct., Genet. 2014, 82, 98−111. (32) Modi, V.; Dunbrack, R. L., Jr. Assessment of refinement of template-based models in CASP11. Proteins: Struct., Funct., Genet. 2016, 84, 260−281. (33) Lindorff-Larsen, K.; Piana, S.; Dror, R. O.; Shaw, D. E. How Fast-Folding Proteins Fold. Science 2011, 334 (6055), 517−520. (34) Wu, H. N.; Jiang, F.; Wu, Y. D. Significantly Improved Protein Folding Thermodynamics Using a Dispersion-Corrected Water Model and a New Residue-Specific Force Field. J. Phys. Chem. Lett. 2017, 8 (14), 3199−3205. (35) Robustelli, P.; Piana, S.; Shaw, D. E. Developing a molecular dynamics force field for both folded and disordered protein states. Proc. Natl. Acad. Sci. U. S. A. 2018, 115 (21), E4758−E4766. (36) Perez, A.; Morrone, J. A.; Simmerling, C.; Dill, K. A. Advances in free-energy-based simulations of protein folding and ligand binding. Curr. Opin. Struct. Biol. 2016, 36, 25−31. (37) Hamelberg, D.; Mongan, J.; McCammon, J. A. Accelerated molecular dynamics: a promising and efficient simulation method for biomolecules. J. Chem. Phys. 2004, 120 (24), 11919−29. (38) Miao, Y.; Feixas, F.; Eun, C.; McCammon, J. A. Accelerated molecular dynamics simulations of protein folding. J. Comput. Chem. 2015, 36 (20), 1536−49. (39) Rauscher, S.; Gapsys, V.; Gajda, M. J.; Zweckstetter, M.; de Groot, B. L.; Grubmuller, H. Structural Ensembles of Intrinsically Disordered Proteins Depend Strongly on Force Field: A Comparison to Experiment. J. Chem. Theory Comput. 2015, 11 (11), 5513−24. (40) Chen, W.; Shi, C.; MacKerell, A. D., Jr.; Shen, J. Conformational Dynamics of Two Natively Unfolded Fragment Peptides: Comparison of the AMBER and CHARMM Force Fields. J. Phys. Chem. B 2015, 119 (25), 7902−10. (41) Piana, S.; Donchev, A. G.; Robustelli, P.; Shaw, D. E. Water dispersion interactions strongly influence simulated structural properties of disordered protein states. J. Phys. Chem. B 2015, 119 (16), 5113−23. (42) Henriques, J.; Cragnell, C.; Skepo, M. Molecular Dynamics Simulations of Intrinsically Disordered Proteins: Force Field Evaluation and Comparison with Experiment. J. Chem. Theory Comput. 2015, 11 (7), 3420−31. (43) Best, R. B.; Zheng, W.; Mittal, J. Balanced Protein-Water Interactions Improve Properties of Disordered Proteins and NonSpecific Protein Association. J. Chem. Theory Comput. 2014, 10 (11), 5113−5124. (44) Henriques, J.; Skepo, M. Molecular Dynamics Simulations of Intrinsically Disordered Proteins: On the Accuracy of the TIP4P-D Water Model and the Representativeness of Protein Disorder Models. J. Chem. Theory Comput. 2016, 12 (7), 3407−15. (45) Dames, S. A.; Aregger, R.; Vajpai, N.; Bernado, P.; Blackledge, M.; Grzesiek, S. Residual dipolar couplings in short peptides reveal systematic conformational preferences of individual amino acids. J. Am. Chem. Soc. 2006, 128 (41), 13508−13514. 7115

DOI: 10.1021/acs.jpclett.8b03471 J. Phys. Chem. Lett. 2018, 9, 7110−7116

Letter

The Journal of Physical Chemistry Letters (46) Wang, W.; Ye, W.; Jiang, C.; Luo, R.; Chen, H. F. New force field on modeling intrinsically disordered proteins. Chem. Biol. Drug Des. 2014, 84 (3), 253−69. (47) Kohn, J. E.; Millett, I. S.; Jacob, J.; Zagrovic, B.; Dillon, T. M.; Cingel, N.; Dothager, R. S.; Seifert, S.; Thiyagarajan, P.; Sosnick, T. R.; Hasan, M. Z.; Pande, V. S.; Ruczinski, I.; Doniach, S.; Plaxco, K. W. Random-coil behavior and the dimensions of chemically unfolded proteins. Proc. Natl. Acad. Sci. U. S. A. 2004, 101 (34), 12491−6.

7116

DOI: 10.1021/acs.jpclett.8b03471 J. Phys. Chem. Lett. 2018, 9, 7110−7116