Deviation in Supercritical States

P´artay and co-workers stated that the poor performance of energetic, density, and

structural properties of HF in the supercritical states H-K compared to experiments may be due to an overestimation of the pressure at which the maximum value of the

isothermal compressibility curve occurs [21]. This observation was made by noting that the deviation in these properties compared to experiments is much larger at lower

pressures than at higher pressures (see Tables 4.6 and 4.8). Shortly following that study, Baburao and Visco suggested that the isothermal compressibility of HF in the super-

understanding of HF in the supercritical states.

We estimated the isothermal compressibility κ for the supercritical states H-K by

using κ = 1 ρ  ∂ρ ∂P  T ≈ 1 ρ ∆ρ ∆P, (4.16)

where ρ and P denote the density and pressure, respectively, at each state for T = 473 K. Taking a one-sided finite difference for the end points (states H and K) and an average

of two one-sided finite differences for the middle points (states I and J), we found κ to be 0.0245, 0.0352, 0.007, and 0.003 bar−1 for states H, I, J, and K, respectively. This behavior of κ is in accord with the other models for HF (see Figure 5 of Ref. [21]), suggesting that the accuracy of the XPHF model in the supercritical states may also be

affected by this type of overestimation.

4.5

Conclusion

We have introduced a fluorine parameter into the PMOw model for use with X-Pol as a

QMFF for HF – the XPHF model. XPHF shows good agreement with experiments, and is comparable to the JV-P model and its PJV-P re-parameterization in terms of radial

distribution functions, energies per molecule, average dipole moments, and density profiles.

Although the results of the XPHF model are comparable to the PJV-P and JV-P models for most quantities tested, several of the same limitations of those models are

present in XPHF. P´artay and co-workers suggested that the r−12term in the Lennard- Jones potential of the PJV-P and JV-P models limits the accuracy of the predicted densi-

ties. The XPHF model could be improved by incorporating these exchange-correlation and dispersion effects directly into the Fock matrix using an approach similar to that

used by York and co-workers [84]. Further, a two-body correction to the X-Pol method which incorporates charge transfer effects into individual fragments by examining all

interactions of fragment pairs could be used [148]. In addition to including charge transfer effects, such an “XP2HF” model could immediately incorporate exchange-

correlation and dispersion effects through PMO’s pairwise D1 dispersion [95], though such treatment would be empirical. Finally, the semiempirical parameters could be

further optimized to give more accurate results, although as in the case of the JV-P and PJV-P models, the improvement would likely not be drastic.

The XPHF model is greatly improved over the previous attempt of modeling HF with X-Pol based on the AM1 model, and our results show that a two-site model for

HF can be as accurate as three-site, polarizable models such as PJV-P and JV-P. This fur- ther demonstrates the utility of the PMO/X-Pol/DPPC methods for use in force field

development beyond the XP3P model for liquid water, suggesting that this approach may be useful as a general framework for a polarizable force field.1

1 _{We thank Pal Jedlovszky for providing us with the total RDFs of the supercritical states for the}

TIPS, JV-NP, and JV-P models, and Sylvia McLain for the partial RDFs of the experiment at 296 K and 1.2 bar. We also thank Matthew McGrath and J. Ilja Siepmann for providing us with the BLYP-D2/TZV2P trajectory of 64 HF under the NPT ensemble at 300 K and 1 bar, allowing us to calculate the partial RDFs and make comparisons with that result. This work has been partially supported by National Institutes of Health grants GM46376 and RC1-GM091445. Computations were performed on an SGI Altix cluster

Chapter 5

MACROSHAKER: A

Coarse-Grained Force Field for

Crowded Systems of many Proteins

5.1

Introduction

Macromolecular crowding is a characteristic feature of living cells, due to the large

size of the confined macromolecules compared to the relatively small, finite cellular

compartments that hold them. Even at low concentrations of individual proteins, the excluded volume of the cellular compartments can be high, leading to exceptionally

large activity coefficients up to 106 [259]. Experimental studies [259–264] have exam- ined the effects of macromolecular crowding on numerous properties. For example,

reaction rates in vivo can be drastically different from those in vitro. According to El- lis [262], “it can be stated with some confidence that many estimates of reaction rates

and equilibria made with uncrowded solutions in the test tube differ by orders of mag- nitude from those of the same reactions operating under crowded conditions within

cells.” Therefore, it is of paramount importance to include the effects of macromolec- ular crowding when modeling the biochemical and biophysical processes occurring in

Figure 5.1: An artistic rendition of a cross-section of an E. coli cell clearly showing a crowded environment of macromolecules. Reproduced with the permission of D.S. Goodsell, Scripps Research Institute [27].

The vast majority of molecular dynamics (MD) simulations of biological systems to date have focused on the explicit modeling all atoms, which limits the integration

time step in the equations of motion to a value on the order of 1 fs. The conditions in such simulations are often close to in vitro experiments, making them convenient

for comparison with observed data at very low concentrations of proteins. However, such conditions are far from the reality of the crowded compartments in a living cell,

as vividly depicted by Goodsell’s artistic perception (Figure 5.1) [27]. Furthermore, all- atom MD simulations have prohibitively high computational cost for a system at even

a fraction of the size of a cell. Therefore, there is an urgent need for the development of a “mesoscopic” computational model for specific biological processes taking place in

the cell under conditions closely resembling the real system that includes all relevant macromolecular particles1 _.

The goal of this study is to develop a theoretical and computational model for the representation of macromolecular particles that can be conveniently used to model

a section of the cell, which can adequately describe intermolecular interactions and

the execution of the dynamic and reactive trajectories of cellular processes, such as metabolism and signal transduction. As an initial step towards this goal, we introduce

a model to reproduce concentration-dependent diffusion coefficients obtained from experiment.

One approach to model macromolecular diffusion and transport is to use Brownian dynamics. Studies such as those described in Refs. [265–273] have demonstrated that

Brownian dynamics can be useful for modeling macromolecular transport in cellular organelles. Consequently, our computational model is based on stochastic processes,

and uses the Brownian dynamics scheme of Ermak and McCammon [46]. Another

1 _{We refer to “large” molecules other than metabolites, enzyme cofactors, single molecules, ions and}

solute, or a small fragment of polypeptide as macromolecular particles. They include folded proteins, nucleic acids, protein-nucleic acid complexes, protein complexes, ribosomes, and lipid bilayers. For sim- plicity, at present time, we do not consider intrinsically disordered proteins.

approach we take is the reduced complexity of the macromolecules through the con- traction of several groups of atoms into single “beads”, or coarse-graining. Our model

incorporates a coarse-graining scheme using a superimposed “uniformly”-spaced grid similar to, but also different from, the approach of Byron [274]. The approximation

scheme is nearly volume-preserving, and preserves the center of mass of the original macromolecule. In addition to Brownian dynamics and coarse-graining, we treat the

macromolecules as rigid bodies, and use quaternions and the techniques of Bulgac and Adamut¸i-Trache for rotational dynamics [275].

Our model, called MACROSHAKER2 , has been implemented into a software package of the same name, and has a graphical user interface for the generation of con-

figuration files and an interactive visualization environment that renders a scene sim- ilar to the artistic works of Goodsell [27]. The remainder of this chapter describes our

implementation of dynamics for the diffusive process, our coarse-grained force field, and our graphical user interface with visualization environment. We also provide a

comparison of concentration-dependent diffusion coefficients for myoglobin obtained

from MACROSHAKER with those obtained from experiments.

2 _{MACROSHAKER is a portmanteau of “macromolecule” and “shaker”. The “macromolecule” part}

comes from the types of molecules we use in simulations, while the “shaker” part comes from the random movements of these particles while undergoing Brownian dynamics. The name MACROSHAKER is not an acronym, but rather is capitalized solely for dramatic effect, in the same fashion as many other scientific programs.