Calibration and Testing of a Water Model for Simulation of the Molecular Dynamics of
Proteins and Nucleic Acids in Solution
Michael Levitt*
Beckman Laboratory for Structural Biology, Department of Structural Biology, Stanford School of Medicine,
Stanford, California 94305
Miriam Hirshberg
Protein Structure Group, National Institute for Medical Research, The Ridgeway, Mill Hill,
London NW7 1AA, U.K.
Ruth Sharon
Department of Structural Biology, Weizmann Institute of Science, RehoVot, 7600, Israel
Keith E. Laidig and Valerie Daggett*
Department of Medicinal Chemistry, UniVersity of Washington, Seattle, Washington 98195-7610
ReceiVed: December 9, 1996; In Final Form: April 3, 1997
X
The objective of this work is to obtain a water model for simulations of biological macromolecules in solution.
A pragmatic approach is taken in which we use the same type of force field for the water as used for the
solute and derive the water potential as an integral part of the ENCAD macromolecular potential.
1,2
Here we
describe a flexible three-centered water model (F3C), which has already been used for many large-scale
biological simulations, and compare it with other water models. The model is further tested by comparing
calculated energetic, structural, and dynamic properties of liquid water, at several temperatures and pressures,
with experiment. The F3C model is extremely simple and fits experimental data well for different temperatures,
pressures, system sizes, and integration time steps. Because the F3C model works well with short-range
truncation, it is well-suited to high-speed computation of long molecular dynamics trajectories of
macromolecules in solution.
Introduction
Liquid water is the universal solvent in biological systems.
Biological macromolecules, such as proteins and nucleic acids,
adopt their structures and carry out their catalytic roles while
interacting with thousands of surrounding water molecules.
Water enables ionizable groups to retain net charges and is often
an intermediary in substrate binding and catalysis. Our focus
is on the development and use of computer simulation methods
to investigate complicated biological phenomena. Earlier studies
have demonstrated that one achieves much more realistic
dynamic behavior of proteins when simulations include water
molecules explicitly, as opposed to continuum dielectric screen-
ing terms.
3,4
Consequently, we have undertaken the develop-
ment of a simple, robust, and realistic model of water for use
in computer simulation of macromolecules in aqueous solutions
under a variety of conditions.
Our primary selection criterion for an appropriate water model
is that it have the same degrees of freedom and employ the
same potential form as used for the macromolecule. Without
such an underlying consistency, unexpected and undetected
systematic errors could be introduced into the simulations. The
need for such coherence was first emphasized by Lifson in the
consistent force field used for small organic molecules,
5
and
this is even more important in biological macromolecules, where
fewer experimental results are available for calibration. Our
macromolecular potential function and methods, which have
been developed over the last 25 years,
1,6,7
have been presented
in full detail elsewhere.
2
The water model presented here is
extremely simple, using only the atoms as interaction centers,
avoiding any constraints on bond lengths or bond angles, and
using potentials that are completely transferable to macro-
molecules.
Early emphasis on analytical treatments of water was
redirected to computer simulation by Rahman and Stillinger’s
pioneering simulation of liquid water done over 25 years ago.
8
That work, which has served as a prototype for hundreds of
subsequent studies, is characterized by (a) use of molecular
dynamics of a simple water model to simulate a trajectory (in
their case, lasting only 2 ps), (b) comprehensive analysis of the
trajectory to extract a wide range of energetic, structural, and
dynamic properties, (c) careful comparison of these properties
to experiment with the aim of improving the molecular
representation of water, and (d) relation of the simulation back
to theory and simple conceptual models.
As more work has been done on models of liquid water, the
representations of water interactions have developed in two
directions. On the one hand, the simple pairwise van der Waals
and Coulombic interactions between atom centers and lone-
pairs used in Rahman and Stillinger’s work have been extended
to models that incorporate polarization.
9-20
But, one confronts
problems using the more complicated water models together
with simple potentials that do not include lone-pairs and
polarization necessary for the macromolecules. Having ad-
* Address correspondence to either author.
X
Abstract published in AdVance ACS Abstracts, June 1, 1997.
5051 J. Phys. Chem. B 1997, 101, 5051-5061
S1089-5647(96)04020-5 CCC: $14.00 © 1997 American Chemical Society