Werner Ebeling Lutz Schimansky-Geier Yuri M Romanovsky
Stochastic Dynamics of
^0 'H a>
.=0
?
O
\ \
V^
CD O CD CO
This page is intentionally left blank
Stochastic Dynamics of rv
fr\
r\\
\\ / ? O i
%
\
SOfQ
I 1 Xtffm
i DO Werner Ebeling Humboldt University, Germany
Lutz Schimansky-Geier Humboldt University, Germany
Lomonossov University, Russia
w
World Scientific
New Jersey • London • Singapore • Hong Kong
CD O
ules
Yuri M Romanovsky
o
Published by World Scientific Publishing Co.. Pte. Ltd. P O Box 128, Farrer Road, Singapore 912805 USA office: Suite IB, 1060 Main Street, River Edge, NJ 07661 UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE
British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library.
STOCHASTIC DYNAMICS OF REACTING BIOMOLECULES Copyright © 2002 by World Scientific Publishing Co. Pte. Ltd. All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means, electronic or mechanical, including photocopying, recording or any information storage and retrieval system now known or to be invented, without written permission from the Publisher.
For photocopying of material in this volume, please pay a copying fee through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to photocopy is not required from the publisher.
ISBN 981-238-162-7
Printed by Fulsland Offset Printing (S) Pte Ltd, Singapore
to Yuri Churgin, Ruslan Stratonovich and Mikhail Volkenstein
This page is intentionally left blank
Preface
This is a book about the physical processes in reacting complex molecules, in particular in biomolecules. In the last decade scientists from different fields as medicine, biology, chemistry, and physics collected a huge amount of data about the structure, the dynamics and the functioning of biomolecules. Great progress has been achieved in exploring the structure of complex molecules. The knowledge of the structure of complex molecucules is of course a 'conditio sine qua non' for the understanding of their functioning, however the understanding of the dynamics is as important [Prauenfelder & Wolynes, 1985; Preissner, Goede & Froemmel, 1986; McCammon & Harvey, 1987; Havsteen, 1989; Froemmel & Sander, 1989]. Without a deep analysis of the physical mechanisms of the dynamics it seems to be impossible to understand the all details of the functioning of biological macromolecules. In particular this refers to the functioning of enzymes, which are the basic molecular machines working in living systems. Since this molecules operate on many thousands of degrees of freedom we have to start to analyse the physical mechanisms e.g. the dynamics of clusters consisting of a many atomic units. Further we have to study the dynamics of conformations, the dynamics of transitions between conformations etc.. In order to give an example, we want to understand the dynamics and the physical mechanism of enzyme-catalyzed bond breaking in substrate molecules. In particular we want to find out what determines the high rate of bond breaking in complex molecules. However to explore the dynamics of this or other complex processes we have to pay a price, only very simple structures allow a investigation of the dynamical phenomena. This is why we have to restrict our studies to rather simple models. In this context we will analyse simple mechanisms as the transitions beween two potential wells, the nonlinear coupling between oscillatory modes, the Fermi resonance, the excitation of solitons in chains of nonlinear springs etc.. The analysis of the complex processes developed in this book is based on methods of nonlinear dynamics, stochastics and molecular dynamics. In the first part of the book we start from the classical stochastic reaction theory. We intended to show how the famous Kramers expression for the chemical reaction rate is to be modified in
Vlll
Preface
the case of the more complicated processes occuring during the enzymatic catalysis. Kramers' classical reaction theory describes reactions as transitions over a potential barrier (activation processes) by studying Langevin equations and solving the corresponding Fokker-Planck equations. The basic assumption of Kramers model is that transitions over a potential barrier are due to stochastic forces. Kramers' model is based on the assumption of uncorrelated stochastic forces. For the case of reactions with simple molecules this model has been very successful [Hanggi, Talkner and Borkovec, 1991; Popielavski & Gorecki, 1991]. On the other hand there are specific reaction effects, which cannot be understood on the basis of Kramers model [Troe, 1991]. In particular this is true for enzymatic reactions which show reaction rates which are by orders of magnitude higher than the simple estimates provided by the Kramers theory [Chernavsky, Khurgin and Schnol, 1967; Volkenstein, 1981,; Somogyi, Welch and Damjanovich, 1984; Ebeling and Romanovsky, 1985; McCammon and Harvey, 1987; Havsteen, 1989, 1991]. Our motivation is to overcome the limitations of Kramers theory and to generalize it. Therefore we will develope in this work simple but more realistic microscopic models for transitions in different molecular environments. The physical effects leading to transitions are studied by means of theoretical models and molecular dynamics simulations. First we will study the effects of small damping as energy diffusion. We will show that in the generalized theory a maximum of the rates in dependence on the friction exists. Consequently, the optimization of effective friction at the reaction sites may be an important factor for the functioning of the reaction sites. Great impact we shall give to the role of nonlinear oscillations in complex molecules [Volkenstein et al., 1982; Pippard, 1983]. For example we will study the role of mode-coupling and nonlinear excitations. The investigations given in this book are mainly restricted to models operating in a two-dimensional physical space. Of course this is also a rather crude assumption, however as we will show, already in two dimensions there appear several new features which might be very helpful to understand the complex phenomena observed in enzymatic reactions, as e.g. the effect of Fermi resonance. Our special interest is devoted to local energy spots which may lead to an enhancement of reactive transitions. An idea expressed by several authors is that complex reactions as e.g. DNA denaturation [Dauxois et al., 1993] and the catalytic activity of enzymes [Ebeling, Jenssen and Romanovsky, 1989; Ebeling et al., 1994; Davydov, 1984] is supported by nonlinear excitations capable to localize energy at special reaction sites. The problem of the elementary excitations in biomolecules and their possible role with respect for functional relevant activation processes was studied in several of our earlier papers on the basis of simple models [Ebeling and Romanovsky, 1985; Ebeling, Jenssen and Romanovsky, 1989; Romanovsky, Tikhomirova, Khurgin, 1994; Ebeling et al., 1994; Netrebko et al., 1994]. Here we will summarize and generalize these approaches. Let us discuss now in some more detail the concrete physical and chemical
Preface
IX
phenomena which typically are connected with enzymatic reactions: As a rule, according to the Volkenstein school [Volkenstein, Golovanov, Sobolev, 1982], enzymesubstrate interaction is reduced to loosening or breaking of a certain bond in the substrate molecule. Thus, any two-atom molecule A-B or A-A can be considered as a model of the substrate. The problem of interaction of the simplest substrate with the enzyme can be formulated as the problem of A-B molecule in the field of several ligands. The authors of this book try to demonstrate the role of the physical processes at certain stages of the chemical reactions, in particular, at the stage of bond breaking in substrate molecules that is catalyzed by enzymes. In other words, we try to find out what determines the rate of bond breaking with the use of the methods of molecular dynamics. We intend to show how the famous Kramers expression for the rate of the chemical reaction is to be modified in the case of the complex processes of the enzymatic catalysis. Enzyme molecules consisting of hundreds and even thousands of atoms interact with substrates in water environment. The building blocks of proteins are the 20 amino acids. In this way the primary structure of a protein can be mapped to a linear string on an alphabet with 20 letters. The primary structure of proteins is rather complicated, showing a high degree of randomness but on the other hand some amount of local order [Ebeling &; Jimenez-Montano, 1980; Ebeling & Frommel, 1998; Jimenez-Montano et al., 2002]. This way already the primary structure contains many intriguing informations connected with the the function of proteins [Kolker & Trifonov, 1985; Ebeling & Frommel, 1998]. Much more complicated is the spatial structure of proteins and in particular the dynamics. Any comprehensive mathematical description of the dynamics of proteins implies the solution of systems of nonlinear differential equations modelling the motions in the 3-d physical space. However, not all the atoms and groups of atoms are involved in certain stages of the catalytic act. It is always possible to select some basic variables whereas the influence of all the others can be taken into account with substantial simplifications. This is not due to reduction of the complete systems of equations but due to the fact that the complex enzyme molecules always consist of subsystems each of which consists in turn of hundreds of atoms. Such subsystems or clusters can be described using only several variables. We discuss here several models which describe the transitions connected with the chemical reactions. In this book we do not intend to present a complete pattern of specific enzymatic reactions or to figure out the ways of their effective control. Instead we will concentrate on one of the most important aspects, the role of stochastic effects. Any theory of chemical reactions has to be stochastic because these transformations are determined by diffusion processes and thermal fluctuations. That is why we try to pose and solve new problems of stochastic nonlinear cluster dynamics. Let us discuss now the general structure of the book. The chapters were written
X
Preface
by different groups of authors. Most of them are recent or former collaborators of the Editors. Each chapter is devoted to a specific problem related to the solution of our general tasks. In the first chapter we introduce several basic concepts as the stochastic reaction theory of Arrhenius-Kramers and discuss the open problems of reactions in complex molecules in particular in biomolecules. For example we explain the difference between the reactions of spontaneous bond breaking in solution and those catalyzed by enzymes. We consider a cluster model of a-chymotrypsin functioning. The term "cluster dynamics" is introduced and several specific problems of nonlinear Brownian motion are formulated. The rates of the valence bonds breaking under the action of thermal fluctuations in aqueous solutions of proteins are estimated. Further we introduce the basic concepts of molecular dynamics and discuss several simple two-dimensional models of nonlinear oscillations. Special attention is devoted to Fermi resonances. The second Chapter is devoted to the fundamentals of the modern stochastic dynamics of complex systems. Here the mathematical tools and several applications are discussed. In particular we discuss theoretical methods for the calculation of the rates of simple chemical reactions. In the third Chapter we start with the analyis of a 2-d model of a chemical reaction. This chapter is devoted to the description the motion of one particle in a 2-d potential landscape with several minima. In detail we study the transitions of the particle from one minimum into another. The first treatment of this problem goes back to our original papers [Ebeling et al., 1994; Chikishev et al., 1998]. The potential landscapes under consideration simulate the force fields of clusters of molecules surrounding a reaction center. The parameters of these landscapes can be periodic or random functions of time because of cluster motion. Also considered is "stochastic resonance" for the transition of test particles from one potential minimum into another in the case of periodic changes of the distance between the minima. Even in a "frozen" landscape without damping a particle motion can be stochastic because such a landscape is similar to Sinai billiard of a complicated shape. In addition, the particles of finite size and specific shape can not be considered as point masses [Romanovsky, 1997; Chikishev et al., 1998]. Another paragraph of this chapter presents
an approximate method proposed by Ruslan L.Stratonovich in his last works on the determination of the characteristic time of a particle escape from reservoir through narrow clefts. The motion of such particles is described by either Langevin equations or obeys the rules of the Sinai billiard [Stratonovich, 1995; Stratonovich and Chichigina, 1996]. So far the analysis is mainly based on the assumption of white noise.
Preface
XI
In the fourth Chapter we take into account more realistic situations corresponding to the model of 'coloured noise'. This is based on an analysis of the microscopic dynamics of the molecules surrounding a reaction center. Further we investigate the role of entropic effects. We study dynamic models of interactions in an ensemble consisting of one reactive molecule (modelled here as a soft particle) imbedded into a bath of solvent molecules (modelled here as hard particles). It is demonstrated that a substantial amount of energy concentrates on the selected "soft" particle under certain conditions. This can facilitate its dissociation or provide a mechanism of accumulation of energy at the selected degrees of freedom in macromolecules. As one example we present a molecular dynamics simulation of the barrier crossing with the allowance of the influence of the surrounding molecules. In other words, the consideration of the simple models uses an atomistic approach based on the molecular dynamics instead of the phenomenological treatment of the stochastic forces. In the first part of the Chapter we develop a model employing a bistable 2-d Kramers potential to describe a reaction center embedded into a heat bath with a solution of molecules. Further discussion concentrates on the dissociation and recombination of the selected molecules in the heat bath. Thus, we take into account a real noise generated by atomic collisions instead of the phenomenological white noise. For this purpose we use standard methods of molecular dynamics [Allen and Tildesley, 1990; Norman et al., 1993]. Our calculations employ one of the variants of the Verlet algorithm [Norman et al., 1993]. We simulated 2-d systems containing 100 disc-shaped molecules. In particular, we demonstrate the concentration of a substantial amount of energy at a selected soft particle. Such a concentration can facilitate particle dissociation or provide energy accumulation in the selected degrees of freedom. The possible relevance of this mechanism for reactions in complex molecules is discussed. Chapter 5 is devoted to the investigation of nonlinear excitations in complex molecules. As a simple model we investigate the excitations in nonlinear ring chains. In particular we study Toda chains and investigate the solitonic excitations which are leading to relatively high energy concentrations on certain sites of the complex molecule. Such an approach allows the modelling of specific activation processes in enzymes. We start from a detailed analytical and numerical studies of Toda lattices, consisting of chains of masses with asymmetric nonlinear interactions, e.g. Toda forces. We demonstrated earlier [Ebeling & Jenssen, 1988; 1991; 1998], that soliton excitations in such non-uniform chains yields local energy spots at imbedded soft molecules. Similar processes take place in chains with Morse interactions [Ebeling, Jenssen & Romanovsky, 1989]. Further the existence of a very specific spatial excitation spectrum is derived [Ebeling, Chetverikov & Jenssen, 1999; Jenssen & Ebeling, 2000]. In Chapter 6 we analyze the the role of Fermi resonances between the vibrational modes with aliquot frequencies in 2 — d molecular systems of the active sites
Xll
Preface
of enzymes. This resonance was first studied in the classical but not so well-known theoretical work of Enrico Fermi (1931) on the Raman effect in carbondioxyd. Here we study the possible role of Fermi resonances in the functioning of enzymes. The late Mikhail Volkenstein was one of the first to consider the role of the Fermi resonance in the protein dynamics [Volkenstein, 1947]. For our model it is demonstrated that a redistribution of the energy between the modes takes place. This is shown for both the conservative case and for the dissipative system subjected to the action of periodic or stochastic forces. This new effect provides the decrease of the time of the reaction products escape from the active site pocket through narrow entrance or exit clefts. Kramers formula needs to be specified in the case when there is nonlinear interaction between the modes in different directions of 2 — d or 3 — d space [Romanovsky and Netrebko, 1998; Shidlovskaya, Schimansky-Geier and Romanovsky, 2000]. In Chapter 7 we discuss models of the molecular dynamics of acetylcholinesterase (ACE). We concentrate in this Chapter mainly on problems of the diffusion limitation of the operation rate of "molecular scissors". We discuss an important problem of the Brownian motion of dumbbell-shaped substrate molecules and disk-shaped product molecules in the ACE electric field. All of them interact with each other, with ACE molecule, and with water molecules. Special attention is paid to the interaction of the substrate (acetylcholine) molecule with the catalytic group of the ACE active site. In Chapter 8 we consider one of the key stages of the reaction catalyzed by serine proteinases - proton transfer in the hydrogen bond of the active site that precedes breaking of the substrate bond. Proton transfer takes place in a potential relief with two minima and a rather high potential barrier. As the length of the hydrogen bond (distance between the potential wells) and the parameters of the potential barrier vary in time, we solve a quantum mechanical problem of proton tunneling in a nonstationary potential (nonstationary Schrodinger equation). It is demonstrated that the variation of the parameters of the system (white noise, colored noise) results in substantial changes of the characteristic time of the proton transfer. The problem of the proton tunneling in the a-chymotrypsin active site with the allowance of the enzyme-substrate interaction was addressed earlier [Khurgin and Burstein, 1974; Romanovsky, Chikishev k Khurgin, 1988]. Such an interaction yields symmetrization of the potential profile and lowering of the potential barrier. This fact might be important for the nonstationary problem as well. Chapter 9 is devoted to the problem of damping of the oscillations of the different clusters (or subglobules) of macromolecules which are imbedded into liquid solutions. In the previous chapters it was already underlined several times that the value of the quality factor, or Q-Factor, for these oscillations determines the
Preface
xm
effectivity of energy transfer from one mode to another in Fermi-resonance. If Q » 1 the cluster oscillations due to the water molecules surrounding the oscillating molecule have the character of coloured noise. This leads to an amplification of the chemical transformations in the enzyme-subsrate complex [Romanovsky and Ebeling, 2000]. We will estimate here the Q-factors by means of the Stokes-LambLandau theory. Our estimations of the Q-factors for the globul oscillations with the typical frequencies 1012 - 1013Hz lead to the value Q ~ 1. For example for the enzyme chymotrypsin, it will be shown here by using methods of physical kinetics and molecular dynamics that , for the amplitude of oscillations less than 1A°, that the quality factor may reach values Q > 10. Finally in Chapter 10 several new methods for the functioning of molecular machines are considered. These methods are based on the investigation of the dynamics of transitions from one conformation (a state of protein molecule ) to another one [Shaitan,1994; Shaitan, Ermolaeva and Saraikin,1999]. This method includes also the concept to study the motion of the system on a landscape of the free energy of the system. In particular new results are obtained about the existence of correlations between the different degrees of freedom. It is shown that the diffusion motion of the system in conformation spase can be realized by different ways and is not necessary connected with chemical transformations. This motion can be concidered as an informationnal process which determines the changes of chemical states of different atomic groups and their functional activity. Note that most of the problems investigated in this book were modelled in the 2-d physical space (only a few could be modelled in the many-particle 3-d space). In contrast to the classical reaction theory which is based on models in the 1-d space - representing the reaction coordinate - the 2-d models allow one to reveal many new qualitative effects. We try to demonstrate also the differences between the approach based on the Langevin equations method and the "straightforward" one based on the models of molecular dynamics. Sometimes a combination of both methods might be appropriate. This book does not contain a comprehensive review on molecular dynamics of macromolecules and the corresponding experimental data, we concentrate here only on the problem of the dynamics at the active site. However, the authors refer to several modern monographs, reviews, and papers devoted to the general dynamics of macromolecules. This book summarizes and generalizes the results obtained in the 10-15 years of a close cooperation between the scientists from Humboldt University Berlin and the Lomonossov Moscow State University, which was sponsored in particular by DAAD ("Deutscher Akademischer Austauschdienst") and by the DFG (Deutsche Forschungsgemeinschaft) Sonderforschungsbereich 555. The research in Moscow was
Preface
XIV
supported by grants of the Russian Foundation for Basic Research (in particular this are the grants 98-04-48479, 98-03-33191, 01-03-33163, 01-04-49302 and the project 96-15-97782 and 00-15-97843 ("Scientific Schools of Russia")), by the Ministry of High Education of Russia, and for the year 2001 by the grants 1.1.144 and 1.2.45 of the Moscow government. Our work was supported also by the INTAS-grant 20010450 and by the Interdisciplinary Project of the Moscow State University "Molecular dynamics of Enzymes". Further we acknowledge supports by INES - International Network of Engineers and Scientists for Global Responsibility, and by the Ministery of Research and Technology (BMFT) of Germany (project no.BEO713-0311257). Finally, we would like to dedicate this book to the memory of three pioneers in the field of the dynamics of macromolecules, Yuri I. Khurgin, Ruslan L. Stratonovich and Mikhail V. Volkenstein. These great scientists and their way of thinking, working and teaching had a large personal impact on the present authors.
References M.R Allen, D.J. Tildesley (1990): "Computer Simulations of Liquids", Clarendon Press, Oxford. D.S. Chernavsky, Yu.I. Khurgin, S.I. Shnol (1967): "On elastic deformations of protein-enzymes" (in Russian), Molec. Biol. 1, 419. A.Yu. Chikishev, W. Ebeling, A.V. Netrebko, N.V. Netrebko, Yu.M. Romanovsky, L. Schimansky-Geier (1998): "Stochastic cluster dynamics of macromolecules", Int. Journal of Bifurcation & Chaos 8, 921-926. T. Dauxois, M. Peyrard, A.R. Bishop (1993): "Dynamics and thermodynamics of a nonlinear model for DNA denaturation", Phys. Rev. E 47, 648-695. A.S. Davydov (1984): "Solitons in molecular systems" (in Russian), Naukova Dumka, Kiev 1984. W. Ebeling, M.A. Jimenez-Montano (1980): "On grammars, complexity and information measures of biological macromolecules", Math. Bioscience 52, 53-60. W. Ebeling, Yu.M. Romanovsky (1985): "Energy Transfer and Chaotic Oscillations in Enzyme Catalysis", Z. Phys., Chem. Leipzig 266, 836-843.
Preface
xv
W. Ebeling, Yu. Romanovsky, Yu. Khurgin, A. Netrebko, N. Netrebko, E. Shidlovskaya (1994): "Complex regimes in the simple models of molecular dynamics of enzymes", Proc. SPIE 2370, 434-447. W. Ebeling, V. Podlipchuk & A.A. Valuev (1995): "Molecular Dynamics Simulation of the Activation of Soft Molecules Solved in Condensed Media", Physica A 217, 22-37. W. Ebeling, C. Frommel (1998): "Entropy and predictability of information carriers", BioSystems 46, 47-55. W. Ebeling, A. Chetverikov & M. Jenssen (1999): "Statistical thermodynamics and nonlinear excitations of Toda systems", Ukr. J. Phys.45, 479-487. W. Ebeling,W., M. Jenssen & Yu. M. Romanovskii (1989): "100 years Arrhenius law and recent developments in reaction theory", In: Irreversible Processes and Selforganization (eds. W. Ebeling and H. Ulbricht), Teubner, Leipzig, pp. 7-24. W. Ebeling, M. Jenssen (1988): "Soliton dynamics and energy trapping in enzyme catalysis", Z. Phys. Chem. 1, 269-279; Physica D 32, 183-193. W. Ebeling,W. & M. Jenssen (1991): "Soliton-Assisted Activation Processes", Ber. Bunsenges. Phys. Chem. 95, 356-362. E. Fermi (1932): "Ueber den Ramaneffekt des Kohlendioxids", Zeitschrift fur Physik 111, 250-259. H. Frauenfelder, P.G. Wolynes (1985): "Rate theories and the puzzles of hemoprotein kinetics". Science 229, 337-345. C. Froemmel, C. Sander (1989): Prot. Struct. Funct. Genet. 2, 1-10. B. Havsteen (1989): "A new principle of enzyme catalysis: coupled vibrations facilitate conformational changes", J. Theor. Biol. 140, 101-109. B. Havsteen (1991): "A stochastic attractor participates in chymotrypsin catalysis. A new facet of enzyme catalysis", J. Theor. Biol. 151, 557-571. P. Hanggi, P. Talkner, M. Borkovec (1991): "Reaction-rate theory: fifty years after Kramers", Rev. Mod. Phys. 62, 251-341.
XVI
Preface
M. Jenssen, W. Ebeling (2000): "Distribution functions and excitation spektra of Toda systems at intermediate temperatures", Physica D 141, 117-132. M.A. Jimenez-Montano, W. Ebeling, T. Pohl, P.E. Rapp (2002): "Entropy and complexity of finite sequences as fluctuationg quantities", BioSystems 64, 23-32. Yu. Khurgin, K. Burshtein (1974): "Proton transfer mechanism in acylation reactions of a- chymotrypsin" (in Russian), Dokl. AJcad. Nauk SSSR 217, 965-968. E. Kolker, E.N. Trifonov (1995): Proc. Natl. Acad. Sci. USA 92, 757-750. J.A. McCammon, S.C. Harvey (1987): "Dynamics of Proteins and protein acids", Cambridge University Press, Cambridge. A. Netrebko, N. Netrebko, Yu. Romanovsky, Yu. Khurgin, E. Shidlovskaya (1994): "Complex modulation regimes and vibration stochastization in cluster dynamics models of macromolecules" (In Russian), Izv. Vuzov: Prikladnaya Nelineinaya Dinamika 2, 26-43. G.E. Norman, V. Yu. Podlipchuk, A.A. Valuev: "Theory of Molecular Dynamics Method", Molecular Simulation 9, 417. J. Popielawski and J. Gorecki, eds. (1991): "Far-from-equilibrium dynamics of chemical systems", World Scientific, Singapore. R. Preissner, A. Goede, C. Froemmel, eds. (1986): "Workshop Theoretical Biophysics", Humboldt University Berlin, Berlin. Yu.M. Romanovsky, A.V. Netrebko (1998): "Some problems of cluster dynamics: models of molecular scissors", Izv. VUZ "AND" 6, 31-44. Yu.M. Romanovsky, W. Ebeling, eds. (2000): "Molecular dynamics of enzymes" (in Russian), Publ. Moscow University, Moscow 2000 Yu.M. Romanovsky, N.K. Tikhomirova, Yu.I. Khurgin (1979): "Electromechanical model of the enzyme-substrate complex"(in Russian), Biofizika 24, 442. Yu.M. Romanovsky (1997): "Some problems of cluster dynamics of biological macromolecules", in: Stochastic Dynamics, L. Schimansky-Geier, T. Poschel, Eds., Lee-
Preface
xvn
ture Notes on Physics, Springer Verlag, Berlin, pp. 1-13. Yu.M. Romanovsky, A.Yu. Chikishev, Yu.I. Khurgin (1988): "Subglobular motion and proton transfer model in a-chymotrypsin molecule", J. Mol. Catal. 47, 235-240 (1988). K.V. Shaitan (1994): "The Electron-Conformational Transition Dynamics and New Approach to the Physics of Biomacromolecular Functioning Mechanisms" (in Russian), Biofizika 39, 949-967. K.V. Shaitan, M.D. Ermolaeva, S.S. Saraikin (1999): "Nonlinear dynamics of the molecular systems and the correlations of internal motions in the oligopeptides", Ferroelectrics 220, 205-220. E. Shidlovskaya, L. Schimansky-Geier, Yu.M. Romanovsky (2000): "Nonlinear vibrations in 2-dimensional protein cluster model with linear bonds", Z. Phys. Chem. 214, 65-82. B. Somogyi, G.R. Welch, S. Damjanovich (1984): Biochim. Biophys. Acta. 768, 81. R.L. Stratonovich, O.A. Chichigina (1996): "Dynamical calculation of the spontaneous decay constant of a cluster of identical atoms". Soviet Phys JETP 83, 708-715. R.L. Stratonovich (1995): "On dynamical theory of spontaneous decay of complex molecules". JETP 81, 729-735. J. Troe (1991): "On the application of Kramer's theory to elementary chemical reactions", Ber. Bunsenges. Physik. Chem. 95, 228. M.V. Volkenstein (1947): "Structure of molecules" (in Russian). Izd. Akad. Nauk, Moscow. M.V. Volkenstein, LB. Golovanov, V.M. Sobolev (1982): "Molecular orbitals in enzymology" (in Russian). Nauka, Moscow.
This page is intentionally left blank
Contents
Preface
vii
Chapter 1 Introduction to the reaction theory and cluster dynamics of enzymes
1.1 1.2 1.3 1.4 1.5 1.6
W. Ebeling, A. Netrebko, Yu. Romanovsky Arrhenius law and basic ideas of reaction theory Breaking of the peptide and ester bonds Basic principles and methods of protein dynamics Effects of coupling and resonances on transition rates Basic variables. Block and cluster models The problems under consideration
Chapter 2
1 1 8 11 16 26 30
Tools of Stochastic Dynamics
L. Schimansky-Geier and P. Talkner Introduction Fluctuations in statistical physics 2.2.1 The canonical distribution 2.2.2 Einstein's formula 2.2.3 Fluctuations around equilibrium 2.2.4 Perrin's pendulum 2.2.5 General approach 2.3 Linear relaxation processes 2.4 Correlations and spectra 2.5 Linear response 2.5.1 Colored noise 2.5.2 Harmonic noise 2.5.3 Fluctuation dissipation theorem
2.1 2.2
xix
37 37 39 39 41 42 43 45 46 48 52 55 56 56
xx
Contents
2.5.4 Nyquist theorem. White noise 2.5.5 White noise and the Wiener process 2.6 Brownian Motion 2.6.1 Einstein's relation 2.6.2 Brownian motion as Markovian dynamics 2.6.3 Langevin's approach 2.6.4 The overdamped limit 2.6.5 Generalized Langevin equations 2.7 The Fokker-Planck equation 2.7.1 Kolmogorov's forward and backward equations 2.7.2 Moments of the transition probabilities 2.8 The bistable oscillator 2.9 The escape problem 2.9.1 Transition state theory 2.9.2 Kramers' rate formulae 2.9.2.1 Moderate to strong damping 2.9.2.2 Weak damping and energy diffusion 2.9.3 Transition rates in multidimensional landscapes 2.10 Pontryagin's equation 2.10.1 Boundary conditions for the forward and the backward equation 2.10.2 The first passage time distribution 2.10.3 Splitting probability 2.10.4 Examples 2.10.4.1 The splitting probability 2.10.4.2 The mean first passage time Chapter 3
58 61 62 62 64 66 67 68 71 71 76 77 81 84 85 86 88 89 89 90 92 93 94 94 96
Motion of test particles in a 2-d potential landscape
O.A.Chichigina, A.V.Netrebko, and N.V.Netrebko 103 Formulation of the mathematical model 103 Lyapunov spectra for the conservative system. Toda area for the landscape with two minima 106 3.3 Stratonovich method of calculating escape times in the chaotic regime and some applications. Dynamic model of the cluster dissociation . . . . 110 3.3.1 The role of a dynamic theory of cluster dissociation 110 3.3.2 The simplest dissociation model 113 3.3.3 The calculation of the rate of cluster dissociation using dynamic theory 115 3.3.4 Mean time of escape from a potential well under the action of noise. Metastable approximation 117 3.4 Test particle motion in a three-minima potential landscape 118
3.1 3.2
Contents
3.5
The problem of a test particle transition in the potential field with periodically changing parameters
Chapter 4
Chapter 5
145 145 148 151 154 161 170 175
Fermi resonance and Kramers problem in 2-d force field
S.V. Kroo, A.V. Netrebko, Yu.M. Romanovsky, L. Schimansky-Geier 181 2-d potential landscape and Fermi resonance 181 Basic 2-d cluster model 183 Analytical study 186 Numerical study 190 Stochastization of the vibrations 193 Basic model including damping and external harmonic action 193 Computer simulation of the nonautonomous system with damping . . . 195 Kramers problem for 2-d potential landscape 199
Chapter 7
7.1 7.2
125 125 130 134 135 137 140
Excitations on rings of molecules
A. Chetverikov, W. Ebeling, M. Jenssen, and Yu. Romanovsky 5.1 Solitary excitations in Toda systems 5.2 Statistical and stochastic theory of Toda rings 5.3 Energy accumulation at nonuniformities 5.4 Fluctuations in Toda rings and time correlations 5.5 Spatio-temporal excitations on rings 5.6 A ring model of enzymes 5.7 A polymer reaction model including entropy effects Chapter 6
120
Microscopic simulations of activation and dissociation
W. Ebeling, V. Yu. Podlipchuk, M.G. Sapeshinsky, and A. A. Valuev 4.1 Discussion of the Heat Bath Model 4.2 Molecular dynamics of transitions between potential wells 4.3 Dissociation of Morse Molecules 4.4 Dynamics of Recombination Reactions 4.5 Spectrum of atomistic collisional forces 4.6 Discussion of activation processes in an atomistic heat bath
6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8
xxi
Molecular scissors. Cluster model of acetylcholinesterase
A.Yu.Chikishev, S.V.Kroo, A.V.Netrebko, N.V.Netrebko, Yu.Romanovsky The role of acetylcholinesterase in the synaptic transfer ACE computer model based on X-ray data
209 209 211
xxii
Contents
7.3
Electrostatic field of ACE molecule 212 7.3.1 Charge distribution inside the molecule 213 7.3.2 Calculation of the potential 215 7.3.3 Determination of the dipole moment 216 Substrate enters the pocket: 2-d "toy" model 219 7.4.1 AC molecules enter ACE AS 220 7.4.2 The problem of the reaction products escape from ACE AS . . . 224 Kinetics of the enzymatic reaction of ester bond breaking 229 7.5.1 Michaelis-Menten equation 229 7.5.2 Mathematical model 232 7.5.3 Determination of the rate constant ki 235 7.5.4 Determination of constants/c_i and ^2 237 7.5.5 Determination of the constant ks 239 7.5.6 Substrate inhibition 240
7.4
7.5
Chapter 8 Dynamics of proton transfer in the active site of chymotrypsin A.Yu. Chikishev, B.A. Grishanin and E.V. Shuvalova 247 The basic model 247 Determination of the wave function by symmetrization of the evolution operator 248 8.3 Checking the results 250 8.4 Mathematical model of the active site of a-chymotrypsin and acetylcholinesterase 253 8.5 Proton transfer in the H-bond of the active site 255 8.6 Discussion 258
8.1 8.2
Chapter 9
On the damping of cluster oscillations in protein molecules
A.Yu.Chikishev, A.V.Netrebko, and Yu.M.Romanovsky 263 The estimate of damping of an oscillating ball by Stokes - Lamb - Landau theory 263 9.2 Estimating of Q-factor by the methods of statistical physics 267 9.3 Simulation by molecular dynamics 269 9.4 General discussion. Development of the model 279 9.1
Chapter 10 Protein dynamics and new approaches to the molecular mechanisms of protein functioning K. V. Shaitan 10.1 Topology of hypersurfaces of conformational energy levels 10.2 Dynamic correlation functions and free energy maps
285 285 289
Contents
xxiii
10.3 Restricted diffusion along a given pathway 10.4 The mechanism of non-Kramers kinetic effects in proteins and glass forming liquids under diffusion limited conditions 10.5 Mass transfer, energy transformation and control in structured media . 10.6 Conclusions
292 296 301 305
Chapter 11
311
Conclusions
List of authors
315
Index
317
Chapter 1
Introduction to the reaction theory and cluster dynamics of enzymes W. Ebeling, A. Netrebko, Yu. Romanovsky
1.1
Arrhenius law and basic ideas of reaction theory
The development of a theory of reaction rates begins in the 19th century with the observation of Arrhenius that the log of reaction rates k is proportional to the reciprocal temperature 1/T:
log. = B - ^ ,
(1.1)
where ks - is Boltzmann constant, A is an empirical constant with the dimension of an energy and B is the rate at high temperatures. According to this law, the characteristic transition time TW k~l should show a linear dependence on the reciprocal temperature 1/T with a positive slope proportional to the characteristic energy A. For example, the decay of hydrogen-iodid (HJ) follows rather precisely this law. Van't Hoff, Smoluchowski, Eyring and other workers have shown, that the constant A may be interpreted as an activation energy. In simple models of the type shown in Fig. 1.1 the activation energy A corresponds to the height of a threshold of the potential energy which has to be crossed on the reaction path . A reaction is then considered as a seldom event which occurs, if the reaction partners collected enough energy to cross the energetic barrier A = AU. In the following we will speak about Arrhenius behaviour of transition processes, if the linear relation AC/ logr = —— + const
, N (1.2)
holds. For Arrhenius processes the reaction rates are given by expressions of type k = T~l = vexp{-AU/kBT),
(1.3)
Here the prefactor v has the dimension of a frequency, it may be interpreted as the rate of attacks to cross the barrier. In the simplest approximation v is 1
2
Reaction theory and cluster
dynamics
equal to the frequency of oscillations a>o near to the bottom of the well v ~ o>o/27r. This corresponds to the so-called transition state theory (see Chapter 2) which overestimates the actual frequency. Indeed there are three main processes which are supporting transitions over the barrier AU: • Stochastic transitions over the barrier, • energy diffusion, • quantum tunneling. The rate of all these processes is proportional to the exponential factor exp (—AU/ksT) but give different contributions to the prefactor z/, the frequency of attacks to overcome the barrier. In some approximation the barrier crossing times corresponding to the three basic processes are additive
u-^v^+u^
+ u-1
(1.4)
We remember that the three terms correspond to the contributions of the stochastic transitions, the energy diffusion and the quantum tunneling processes respectively. We will discuss these terms in more detail below. The central theme of the theoretical treatment of reaction rates is the modelling of high energy events capable to barrier-crossing. Eyring, Kramers and others developed several models of thermal activation processes. A survey of the modern state of art may be found in (Hanggi, Talkner & Borkovec, 1991). As shown by Troe (1989) and other workers, there are many reactions which for various reasons do not follow the Arrhenius formula. In particular many enzyme reactions do not observe the Arrhenius law. One of the reasons for deviations are entropic effects. If the considered reactions occur in solutions then the solvents plays the role of a heat bath with given temperature T. According to the laws of thermodynamics then the free energy takes over the role of the energy. Consequently we find a different law which reads k = r~1 = const.exp(-AF/kBT),
(1.5)
Here AF denotes the change of the free energy in the reaction which consists of an energetic and an entropic contribution
AF = AU - TAS,
(1.6)
Macromolecular reactions are often accompagnied by conformational changes
what might lead to big changes of the entropy. This is an important point which
Arrhenius
Fig. 1.1
law and basic ideas of reaction theory
3
Two-dimensional bistable potential landscape.
has to be taken into account in a proper theory of enzyme reactions. Let us turn now to the basic ideas of the theory of stochastic transitions. The statistical theory of stochastic transitions is based on the model developed in 1940 by Kramers. The Kramers-theory of reaction rates makes use of the Fokker-Planck equation for the reactive molecule or on the corresponding Langevin equation with white noise sources. The Fokker-Planck equation may be derived from the microscopic equations of motion by a definite procedure which is explained in some detail in Chapter 2. However several input quantities of the statistical theory are not well defined. For example, the forces acting on a molecule imbedded into a bath of solvent molecules are not well determined. Further the dynamics in a real system may deviate from the Langevin model. It is also known from Gibbs statistical theory that the potential energy of an individual molecule does not always obey a Boltzmann distribution. Some facts that cannot be interpreted within the framework of the classic reaction theory [Popielawski, 1989; Frauenfelder and Wolynes, 1985]. Therefore what we urgently need is a more accurate consideration of the microscopic forces acting up on the reacting molecules which might be quite different from the other molecules of the ensemble. Strictly speaking the Kramers theory is a model which reflects only a few of the important aspects of real reaction. Nevertheless it is of basic importance for the understanding of the basic features of reactions. Therefore let us consider now briefly the main ideas of Kramers theory for the transitions of a particles imbedded into a stochastic heat bath. As an example we show in Fig. 1.1 a test particle moving in a 2-d potential well.
4
Reaction theory and cluster
dynamics
In order to get a physical picture about the individual acts of transitions we start from the Langevin equation for the motion along the reaction coordinate. A reaction coordinate is an abstraction (a model) which represents a generalized physical coordinate along which the reaction proceeds. According to Kramers the dynamics of the reaction is modelled by the following classical equation d2q
dq
dU (q)
/—„, ,
where m is the mass of the particles, 7 is the friction constant (with the dimension of a frequency), D is the diffusion constant, and £(£) stands for a (^-correlated noise. The potential U(q) is assumed to have two minima, therefore the system is bistable, similar as the potential shown in Fig. 1.1. The corresponding FokkerPlanck equation, which in equilibrium is solved by a Boltzmann distribution, was treated by Kramers with a special boundary condition modelling the transition over a potential barrier. The result for the transition rate, which will be derived in detail in Chapter 2, is for the case of moderate and large values of the friction given by [Hanggi, Talkner and Borkovec, 1991] k = r~1 = v3exp(-AU/kBT),
(1.8)
where the frequency of stochastic transitions v8 is after Kramers given by
vs =
i\
2y
/2
7'
^ - -
(1-9)
Here WQ is the angular frequency inside the minimum, where the transition starts and uib is the angular frequency of the potential maximum (the transition state). We see, that (1.8) is in full agreement with the Arrhenius law. In the limit 7 = 0 the Kramers expression converges to the result of the transition state theory VTST = Uo/2ir. For the case of strong friction we may use the approximation
vs ~
(1.10) 4-7T7
In the limit of very weak friction the Kramers solution presented above is not valid. Under the condition 7 < UJQ the rate is controlled by energy diffusion. We find approximately for the corresponding frequency (see Chapter 4) 7AE/ V^IT 2k-BT^ .
(. n
s (1-11)
Arrhenius
law and basic ideas of reaction
theory
5
3000.0
-1.0
o.o Ig(gamma)
Fig. 1.2 Transition time (in units l/wo as a function of the friction constant 7/u;o for U>Q = u}\, = V2, &U/kBT = 2 and kBT/h = 10.
In this case the constant in (1.1) shows a weak logarithmic dependence on A U/ks T, this leads to weak deviations from the Arrhenius law (see Chapter 4). We will not go here into the details of the theory of quantum transitions (Chernavsky & Chernavskaya, 1999). An estimate gives kBT
(1.12)
In this way assuming additivity of the three contributions we arrive at the estimate for transitions rates including stochastic transition, energy diffusion and quantum tunneling _ / ~
2TT7
\<jj0ub
2kBT
h
7AC/
k~B~f
-l
exp (-AJ7/fc B T) •
(1.13)
A more detailed analysis shows that the prefactor of the exponential has a maximum in dependence on the friction 7 which usually is located in the region 7 < wo- The corresponding transition time r = v~x shows a minimum as demonstrated in Fig.1.2. According to our estimate the minimum of the transition time is located at
lmin=[-^Mr)
•
(1.14)
We have to underline, that the formulae combining the three effects are esti-
6
Reaction theory and cluster
dynamics
mates only which give the qualitative overall behaviour but not quantitative values in the region around the minimum. The existence of a minimum is however correctly represented. We will point out in later sections that the optimization of the friction constant by building an appropriate sourounding of the reaction site may play an important role for biological catalysis. In one of the next sections we will show that resonance phenomena between longitudinal and transversal oscillations may lead to a considerable shift and lowering of the transition time. We will show that in systems with two or more degrees of freedom the exchange of energy between the degrees of freedom may lead to beating phenomena. Due to this new effect, much more energy may be concentrated on the reaction path and we may find strong enhancement effects. One of the main ideas we will develope in this work is to look for such and other affects of reaction enhancement in comparision to the standard theory explained above. Under what circumstances we still may expect strong deviations from the standard Kramers-type laws? Evidently one of the crucial point in Kramers derivation is the assumption about a white noise character of the forces acting on the active site. In real molecular systems and in particular under the influence of nonlinear collision effects, the 'real noise' may have a very complicated spectrum. In the fourth chapter we will show by MD-simulations that the spectrum of the forces acting on molecule in a bath has a maximum at finite frequencies and shows some similarity to coloured noise (harmonic noise). Simple model of systems with harmonic noise were studied by several workers [Straub & Berne, 1986; Ebeling & Schimansky-Geier, 1989]. Strictly speaking not all of these models do apply to a molecule imbedded into a thermal bath in equilibrium since some of these models do not obey detailed balance. However these easily tractable model reflect at least a few qualitative properties of the stochastic forces in more realistic systems. We will come back to these problems in Chapter 2, in Chapter 4, and in Chapter 10. We mention that the transition rates in real molecular systems are also influenced by the changes of the average forces, which are due to the equilibrium correlations. As well known from the modern theory of liquids (in particular we have in mind the BBGKY-theory), the interaction potentials are in dense thermal systems to be replaced by potentials of average forces, which are defined by the equilibrium distribution functions. Since the potential landscapes are changed by this equilibrium effect, this is another reason to expect that for real molecular systems the Arrhenius factor exp (—U/ksT) is to be replaced by an effective Arrhenius factor exp {-Uefj/kBT). Evidently the physical assumptions leading to a Fokker-Planck equation are rather severe, since in real systems the forces acting on a molecule do not correspond to the Langevin model and further we know from Gibbs statistics that the potential energy of a molecule must not obey an individual Boltzmann distribution. Having in mind that in spite of the great success of the classical reaction theory, several phenomena remain unexplained [Popielawski, 1989; Frauenfelder & Wolynes], a more careful study of the microscopic forces acting at special molecules seems to be
Arrhenius
law and basic ideas of reaction
theory
7
necessary. The standard form of the equipartition theorem of statistical mechanics is saying that any quadratic contribution AH — aq2 to the Hamiltonian leads in thermal equilibrium to a contribution &s T/ 2 to the internal energy. This follows immediately from the Boltzmann distribution which is the stationary solution of the Fokker-Planck equation.
P{q) = const. exp(--j-j;)
.
(1.15)
In other words, there is no way to accumulate in average more energy than k,B T/2 on a linear degree of freedom. However this is not true for nonlinear excitations. Therefore we have to look for nonlinear effects capable to accumulate energy at certain sites. In section 1.4 section 1.4 of this chapter and in chapter 6 we will study Fermi resonance phenomena and in chapter 5 we will investigate in detail the nonlinear excitations and the statistical thermodynamics of nonlinear 1-d systems with Toda interactions [Ebeling and Jenssen, 1988, 1991, 1992]. We will analyze analytically and numerically the basic nonlinear dynamical effects, and in particular the soliton fusion, which was shown to be responsible for an energy localization at definite sites. In earlier work we investigated already 1-d, 2-d and 3-d systems with nonlinear finite-size interactions by simulations [Ebeling et al., 1995]. As well known in such more realistic systems, soliton excitations strictly speaking do not exist. However one can show theoretically and numerically that nevertheless hard excitations, which are soliton-like do exist. An example are the hard pulses, which we can excite by knocking a piece of matter with a hammer. These excitations have several properties in common with solitons, in particular they are localiced in space.It was shown that the effects of the fusion of such nonlinear soliton-like excitations persist and show analogies to the fusion of solitons [Ebeling et al., 1995]. Especially we studied the influence of this dynamical effect on the energy distribution of soft sites imbedded into the system. The analytical results for Toda-systems showed rather strong energy accumulations in certain window of the temperature [Ebeling and Jenssen, 1991]. These effects were confirmed by MD-simulations for systems of 32-200 molecules Lennard-Jones-like finite range interactions (FLJ-potentials). Distribution functions of the energy were calculated and studied especially in the region of the high-energy tails. We found that in thermal equilibrium there exists a characteristic region of temperatures and densities where energy is mainly concentrated on soft sites. We observed also rather long tails in the energy distribution making high-energy events much more probable than expected from a simple Boltzmann distribution [Ebeling et al., 1995]. Further we have shown that the spectra of the excitations is similar to coloured noise spectra [Ebeling and Podlipchuk, 1996]. We will come back to most of the points discussed here in later chapters. For example a more advanced discussion of coloured noise effects will be given in Chapter
8
Reaction theory and cluster
dynamics
2, concrete 2-d landscapes will be studied in Chapter 3, the microscopic effects of atomistic collisions will be discussed in Chapter 4, soliton effects will be investigated in Chapter 5 and the transitions in real proteins will be discussed in Chapter 10.
1.2
Breaking of the peptide and ester bonds
The reaction of breaking of the chemical bond in a molecule A-B (not necessarily two-atomic) is a fundamental chemical process in living systems [Fersht, 1977, 1999; Keleti, 1986; Branden & Tooze, 1999]. In vivo, such a chemical process is in general assisted by enzymes. However in the liqid phase it can take place even without support by enzymes, when the molecules collide with the solvent molecules. But the probability of this process is rather low. Let us explain these processes in more detail, using as the examples the peptide bond (PB) breaking in a protein molecule and the ester bond breaking in the neuromediator acetylcholine (ACh) in water at room temperature. After that we will consider an effective "cutting" of these bonds by "molecular scissors" (hydrolytic enzymes). The peptid bond has the following chemical structure -CHR1
-NH~CO-
CHR2-
The molecule acetylcholin has the chemical composition (CH3)3 N+ [CH2)2 O ~ COCH3 We denoted in both cases the critical binding site which is subject to breaking by a wavy bond ~ . The breaking of the peptid bond plays a very important role in many processes in living systems. For example in the course of digestion PB must be broken and the protein chains split down to amino acids that get into blood to reach all cells of the living organism. The reaction of peptide bond breaking may involve rather long peptides or their fragments. Let us consider for example the interaction of chymotrypsin (CT) with a polypeptide chain. The CT active site effectively binds aromatic amino acids (e.g. phenylalanine) whereas small residues easily get through the AS pocket. What is the mechanism of this binding? A mechanical analogy may be a thread drawn through a needle eye? This problem can be considered based on the PDB data and the theory developed recently for estimating the rate of a polymer thread drawing through a hole in a membrane (see, for example, [Sung k Park, 1996; Park & Sung, 1998]). This is related, in particular, to drawing of polymer substrates through the pores of the artificial vesicles containing CT molecules [Raryi, 1995]. Another process of primary importance for life phenomena is the breaking of esther bonds. For example the "cutting" of the ester bond in acetylcholine molecules
Breaking of the peptide and ester bonds
O
O"
II 1
9
I 2
1
CHR - NH~C - CHR -...<-> ...CHR - NH~C - CHR 2 -...<->
I
o
o+
/\
/ \
HH
HH
O...CHR'-NH2+OH-C-CHR2-...
II O
with the formation of choline and acetate must take place continuously in the intercellular space of the neurons for maintaining normal nerve functioning. We choose these typical and, at the same time, remarkable enzymes because their structure (geometry) in the crystalline state is well known. Also well studied are chemical transformations in the enzyme-substrate complex (ESC) inside the active sites of these enzymes. However, the physical principles of some stages of the catalytic act are still under discussion. It was mentioned in the Preface that one of the aims of this work is the formulation and solution of some specific problems of nonlinear dynamics related to functioning of the molecular scissors. We hope to provide a better understanding of the physical principles of the enzymatic catalysis. The reaction of the peptide bond breaking can take place in aqueous solution of the protein. However this is a very slow process. It takes place in a way as follows: ... - CHR1 -NH-CO-
CHR2 -... + H-OH
«• ... - CHR1 - NH2 + HOOC - CHR2 - ...
++
(1.16)
Here R \ R 2 are the residues of the two neighbor amino acids, the peptide bond between which (wave line) must be broken. In a more detailed way this reaction can be represented as: The intermediate complex is formed under participation of the ions O H - . The role of the hydroxyl ions in the intermediate complex is reduced to loosening of the double bond C = 0 of the carbonyl group. In other words, the ion O H - must approach the substrate and bind in the "proper place". Note that the description of this process is a special problem of molecular dynamics.
10
Reaction theory and cluster
dynamics
The reaction of ACh hydrolysis leading to the formation of acetate (CHsCOO - ) and choline (CH 3 ) 3 N + (CH2)20H is as follows: (CH3)3 N+ {CH2)2 O ~ COCH3 + H20 o
<-> (CH3)3 N+ (CH2)2 OH + CHzCOO'
+ H+
(1.17)
or ACh+ + H20 o Ch+ + A~ + H+
(1.18)
Note that ACh and the products of its hydrolysis (A and Ch) are ions. Assuming that quantum transitions dominate and using the expressions given in the first section we find for the rate of the bond breaking (or the formation of the product P) the general formula: 1
dt
= / i 0
/
A P\
„[Oir]-exp(- —
j.
(1.19)
Here AE is the bond energy, [P] and [OH - ] are the product and hydroxyl concentrations, respectively. The prefactor v is determined by eq. (1.4). In the quantum case we find v ~ ksT/h as an estimate of the frequency of quantum transitions due to thermal fluctuations (?i=Planck's constant. This way ksT/h determines the mean number of thermal fluctuations in the bond being broken. The factor /x0 takes into account the [OH~] sorption time. The term n in the denominator stands for the number of configurations were the sorption can occur. Therefore 1/n represents the probability of the attack and sorption of the [OH~] ion in the "proper place". We call 1/n the "entropy" factor and write it in the form n -
6XP
fAS\ \^)
=
6XP
/TAS\ V ksf) '
where AS = —ks Inn. Then in agreement with the relations given in the first section (for quantumdetermined transitions) the "absolute rate" may be expressed by the change of free energy:
kA = ^ e x p f - - ^ - ^ J .
(1.20)
Here AF = AE - TAS is the free energy of the bond breaking and v ~ ^ ^ is an estimate for the frequency of attacks (for quantum determined transitions). Let us make some estimations assuming T — 300 K, fc^T = 0.025 eV = 0.6 kcal =
Basic principles and methods of protein
dynamics
11
4 x l 0 - 1 3 erg. Then our estimate for the frequency of attacks is v ~ fc^T/Ti ~ 1014 Hz. Assume that fio ~ 1 , then it follows from (1.19) and (1.20) that the exponential terms provide the main input to the rates of the processes. Further we recall that for the chemical bond we can make the estimate AE ~ lOOfc^T The results for these estimates are summarized in the following Table that shows the dependence of kabs on the free energy difference AF:
AF,eV
- P ( - ^ )
Kabsi '-Is
Comments
0.4 0.8 1.2
lO" 7 10-14
106 IO- 1
lO" 2 1
io- 8
this is a very high rate rate reached by enzymes typical rates without enzymes
What happens if an enzyme works as a catalyst instead of an OH~ group? At first it may reduce the entropy factor 1/n = exp [AS/kg]. At second it may diminish the energy barrier AE. In this case even the thermal fluctuation energy and the sorption energy may be sufficient for bond breaking. Sometimes a large and complex enzyme molecule is treated as a "structurized solvent" for the substrate. Going back to the epigraph one can say that the molecule A-B gets into a specially organized force field and is fixed therein. In this reaction "pocket" the reaction dynamics may occur under a very low or even optimized effective friction. Further the parameters of the potential landscape are fluctuating. Some oscillating atoms or groups of atoms of the enzyme active site play the role of the hydroxyl group and the potential landsape inside the active site is such that the effective values of AE (and AF) get lower. The charged groups attack the bond not randomly (like OH~ in water) but in a special way. Another effect which may play an essential role is the nonlinear coupling between different degrees of freedom leading to some concentration of energy at the active site. The mechanisms of such complex processes can be studied by the methods of molecular dynamics which is one of the main methods used in this book. In the following section 1.3 we will discuss some of the basic aspects of this methods and will analyze in section 1.4 a rather simple model in order to demonstrate how the optimization of the conditions may lead to an considerable enhancement of transition rates.
1.3
Basic principles and methods of protein dynamics
Several important aspects of the modern studies of protein dynamics can be outlined:
12
Reaction theory and cluster
dynamics
1. Modern roentgenography makes it possible to determine the coordinates of thousands of heavy atoms of protein globules in the crystalline state with the accuracy of 1-1.5 Angstroms. Hundreds of corresponding data files can be found in Protein Data Banks. 2. Various spectroscopic techniques allow one to trace the dynamics of individual atoms and atomic groups of proteins both in the solid state and in solutions. 3. Based on the methods of molecular dynamics one can write the system of equations for the motion of all the atoms in the protein molecule [Weiner et al., 1984]. The interaction with solvent and substrate molecules can also be taken into account [Gunsteren and Berendsen, 1990]. Below we will explain the MD-methods which will be used here. 4. Modern computers are capable of solving such systems of equations. It is possible, for example, to determine the coordinates of all atoms for stable conformations of the protein molecule. It is possible also to obtain realizations of the stochastic trajectories of individual atoms for the time intervals up to several picoseconds. Such problems can be solved at present using e.g. computer codes as GAUSSIAN, GAMESS, or JAGUAR. 5. The progress in modern quantum biochemistry allows one to determine the mechanisms of enzyme-substrate interactions in the active sites of enzymes. 6. The theory of propagation of solitons and excitons in protein structures is being developed along with the theory of nonlinear oscillations and waves in DNA. 7. The methods developed for the study of the nonlinear Brownian motion can be applied for the study of some stages of the catalytic act. Thus, it is possible to get an insight into the main physical and chemical regularities of the catalytic act on the basis of stochastic and quantum dynamics of all atoms, atomic groups, and molecules that take part in this act. A comprehensive solution of this global problem implies formulating a mathematical model comprising thousands of nonlinear differential equations. This point of view is quite possible because the belief in the capabilities of modern computers is virtually unlimited. That is why it is not expedient to simplify the problem. One should better achieve maximal completeness of the model. Such an approach is also possible in solving the problems of biochemical kinetics where the objects are also described by the systems of hundreds of nonlinear equations of the "reactiondiffusion" type. Not that similar problems are posed in mathematical ecology and the theory of chemical and biological evolution [Romanovsky, Stepanova, and Chernavsky, 1975, 1984]. Let us consider now the dynamical equations which will be used in this book. In most cases we will apply the method of molecular dynamics (MD) based on the classical Newton equations for the particles. Assuming that the system consists of N particles i = l,2,...,i,...,N with the masses m; the coordinates r; the velocities Vi we define an interaction potential
Basic principles and methods of protein
U(fi,...,fi,...,fN)
=
dynamics
13
(1.21)
v^Uij *J
and an external force Fi acting on particle i. For generality we include noise and friction terms and write down the Langevin equations in the form: dvi dU m ; — = Fi - — + (2A) 1 / 2 &(*) dt or;
m
Wi
(1.22)
In order to guarantee the existence of a thermal equilibrium we assume an Einstein relation
(1.23) 7t
"*i
which defines the temperature of the heat bath. If we are interested in the dynamics in a heat bath without friction and external noise, then we start the simulations with eqs.(1.22) and after a few thousand or ten thousands of steps we reduce the noise strength Z); and the friction 7J to a very small value but observing the Einstein relation (1.23) or switch off completely the noise and the friction term. For the interactions we will assume throughout the book that they are pairwise additive, which is an approximation of course. For the pair potentials describing the interaction energy of two particles several models will be used. For simplicity we will apply the notations U(r) — Uij and r = |F; — fj\ . • Standard Lennard-Jones potential (LJ): U(r) = 4e
(?)"-(?)'
(1.24)
We notice the properties U(r = a) = 0 . The minimum with U'(r) = 0 occurs for r — rmin = 21/6cr and has the (negative) value C/m,n = —e. • Finite range Lennard-Jones including only repulsion (FLJr): U(r) = 4e
(?)"-(;)'
(1.25)
for r < rmin and U(r) — —c for r > r m ;„. This potential is purely repulsive. By a shift we may bring it to a a form where the energy disappears at r —> oo:
U{r) = 4e
U
(F) -C)
6
1
+-
(1.26)
14
Reaction theory and cluster
dynamics
with U(r = a) = +e and U(r) = 0 if r>rmin.
(1.27)
We notice that the computer time necessary for simulations with finite-range potentials may be by a factor of ten lower than for the relatively long-range standard LJ-potential. Further we notice that the stiffness of this potential may be adapted to the properties of the molecule by changing the value of U(r = a) = e. For several problems the attraction is essential and we may include it by the following finite-range construction. • Finite size Lennard-Jones potentials including attraction (FLJa): TO
= Ae((^)»-l)exp((^-|)^) U(r) = 0 if r > 1.5a
(1.28) (1.29)
For n = 12 the steepness corresponds to the standard Lennard-Jones potential. The values of A, rj may be adapted in such a way that the zero-crossing occurs as for the LJ-potential at r = a and that the minimum is located at r = 21/6er. For example if n = 8 this requires the choice of the parameters A = 28.05, 77 = 1.028. • Exponential potential (Toda potential): £/(r) = £/(0)exp(-6r)
(1.30)
This is a two-parameter potential similar as the LJ-potential which however is more realistic with respect to the description of the repulsion between the molecules. The parameter b determines the stiffness of the potential. The diameter a is not well defined for this potential. It may be introduced e.g. by the condition U(a) = ksT which leads to kBT = 17(0) exp (-ba)
(1.31)
The exponential potential is not singular at r = 0 and is in close correspondence to the results of quantum-mechanical calculations for the repulsive forces. The repulsion is heavily based on the Pauli principle which prevents any overlap of the electron clouds of the molecules. • Morse potential: U(r) = e [exp(-6(r - rmin) - 2 exp(-6(r - rmin)\
(1.32)
The Morse potential is so to say the difference of two exponential potentials with different range. We may bring it to a similar shape as the LJ-potential by using the form
Basic principles
U(T
and methods of protein
15
dynamics
(exp(-6(r - 2 1 / 6 CT)) - l )
- 1
(1.33)
The Morse potential is quite realistic with respect to the repulsion and gives a good approximation for the attraction. The free parameter b may be used to adapt the stiffness to the properties of the molecules which are studied. With the special choice of the stiffness
we may reach U{r = a) = 0. This means that zero-crossing and minimum are at the same place as for the Lennard-Jones potential what leads to a nearly full agreement with the LJ-potential. The choice of the interaction potential is arbitrary to some extent. In this book we chose for solving concrete problems that of the potentials introduced above which is numerically simple and still realistic enough. We believe, that at least at the present stage of the theory, the choice of a realistic potential is not the most urgent problem. In most cases we have investigated the results are not sensitive with respect to details. This will be demonstrated below for several examples. So far we discussed the MD-method for point particles with two or three Cartesian coordinates. In the more general case, which is typical just for the simulation of complex molecules, the 'particles' itself may have a 'form' and several coordinates describing the internal motion. The simplest example is the water molecule which consists of three atoms, two hydrogen and one oxygen atom. The role of the internal coordinates play the distances between the protons and the oxygen nucleus which are determined by the Coulombic forces and the chemical bonds [Franks, 1979; Sazepina, 1998]. Of course at special conditions, e.g. in bulk water or at large velocities these structures may be broken and other bonds, e.g. hydrogen bond, may be formed. The general structure of the MD-equations still remains the same and we have e.g. in the simplest case without friction and noise the dynamical equations
7 7 2 ; — = Fk(xi,X2,...,Xk,...,Xn)
(1.35)
with k = l...n were the number of the degrees of freedom n includes the Cartesian as well as the internal ones. In many cases however the internal dynamics has to be described by quantum mechanics. The typical 'fluctuation' frequency to — 1/r above which quantum effects play an important role may be estimated by the inequality
HUJ
< kBT
(1.36)
16
Reaction theory and cluster
dynamics
If the temperature is about T ~ 30QK then we get the characteristic 'quantum frequency u)qu ~ 1 0 1 3 s - 1 . The equations of motion have to be supplemented by boundary conditions, external forces and interaction forces. In general one can construct potentials which describe the internal forces
Fk =
dU{x1,x2,...,xn)
(L37)
&Tk
As an example we write down a potential often used for the modelling of protein molecules
U(x1,x2,...,xn)
=
-^2h{b
- b0)2 +-^keiO
- 60)2 +-k^l
+ cos{n(j> - 5)]
+E(;£-2 + ^)+£(;£-£)
a*)
Here the summation extends first over all valency bonds, the corresponding angles and over the torsion angles. Next the sum runs over the non-valency bonds as the van der Waals and Coulomb interactions. Finally the sum is extended over all hydrogen bonds, which are often described alternatively by Morse potentials. The constants depend not only on the kind of the bonds but also on the kind of the particles. The meaning of the parameters in eq.(1.38) is the following: b denotes the length of the valency bond, 0 is the corresponding angle, (j> is the angle of torsion, r is the distance between the particles. Evidently the first two terms in eq.(1.38) may be valid only for rather small deviations. Details of these problems are out of the scope of this book. However the interested reader may find an enourmous material in the literature including computer codes for special calculations [Bernstein et al., 1977; Weiner et al., 1984; Karplus, 1988; Balabaev & Lemak, 1995; Karplus 1998] 1.4
Effects of coupling and resonances on transition rates
As we have discussed in the preface and in section 1.2 the enzymatic reaction take place at a specific site of the enzyme which often has the form of a "pocket" under very specific conditions. At the reaction site the local "reactive dynamics" connected with "barrier crossing" is coupled to other oscillating degrees of freedom. The mechanism of such couplings will be analyzed in more detail in section 1.5. Further we may imagine that the surrounding of the active site forms a kind of optimized "heat bath". In particular the "effective friction" acting on the reaction dynamics at the active site is much lower than in the solution itself due to the shielding of the reaction site by other molecular groups of the enzyme. In the following we will develope a very simple model of these effects which is based on the following assumptions
Effects of coupling and resonances on transition
rates
17
• the reaction oscillator is coupled to an "optimized" second oscillator, • the oscillators are imbedded into an "optimized" heat bath. At the end of our model study it will be clear what the vague term "optimized" means. For simplicity we study here only the coupling of two oscillators which is decribed in a two-dimensional coordinate space and a two-dimensional momentum space. We will show that this is the minimal requirement for obtaining inetesting coupling effects. Of course this is a rather crude assumption since the real dynamics occurs in rather high-dimensional spaces. However as we will show, already in two dimensions there appear several interesting new features which might be very helpful to understand the complex phenomena observed in enzymatic reactions. In particular we have in mind the coupling between oscillatory modes, e.g. coupling between modes of the same frequency 1 : 1 and the coupling between modes with frequency relation 1 : 2 . As one very important example of the second type we mention here the Fermi resonance which has been detected first in 1931 in Fermi's famous paper on a 2-d model of the Raman effect of CO2 [Fermi, 1931]. This theory was orked out in the book of Pippard on the theory of oscillations [Pippard, 197?]. In fact Volkenstein was the first who discussed the Fermi resonance in the peptid binding. In section 1.2 we showed the principal structure of a peptid bond. The place were Fermi resonance may take place is the binding N — H (See Fig. 1.3 and Volkenstein, 1981). The longitudinal and transversal oscillations of the N — H binding are approximately in 1 : 2 resonance. Therefore Fermi resonance may play a role in peptide breaking. Let us develope now the basic elements of the theory of resonance phenomena in 2 — d systems. Beside the Fermi resonance, which is in fact a switch of oscillation energy between two coupled oscillatory modes we will discuss a few other simple models of typical two-dimensional effects, without pretending on any completeness. Let us consider the Langevin equations of motion of a test particle (TP) in a two-dimensional potential U(x,y) under the action of noise and friction (hereafter all the variables are dimensionless; the mass was assumed to be equal to unity, kB = 1, and D = 7T:
£__«£*) _7| + ^ ( 0 g,_«^)_ 7 | + ^^w.
(L39) (L40)
Reaction theory and cluster
dynamics
Fig. 1.3 The structure of peptid bonds according to Volkenstein, 1981. Fermi resonance may occur at the N — H binding place.
Here x — X\ and y — x? are the Cartesian coordinates; t is time; T is the temperature (noise amplitude); £ x ,£ y represent random quantities with white noise characteristics; 7 is the friction coefficient. The potential U(x, y) models the force acting on the test particle (the reacting particle) as a constant or slowly varying external field. Let us first study the simplest 2d-dynamics assuming that the noise strength is zero T = 0 and that the potential is parabolic in normal representation
U(x,y) = 2 W i a;2 + 2Wlv2
(1.41)
In this most simple case we find two uncoupled modes. The solution is given by two sinusoidal oscillations
xk = Aksin(uJkt + 5k)
(1.42)
In the case that x and y are Cartesian coordinates the trajectories are enclosed into a rectangular box with the edges A\,A2- For the case of equal frequencies the orbits are ellipses or lines along the diagonal of the box. For curiosity we mention that there exist special nonlinear problems which have similar simple orbits as e.g. the Yang-Mills equation known from the theory of elementary particles [Landa, 2001].
Effects of coupling and resonances on transition
rates
19
Let us consider now the case of linear oscillations with different frequencies. Depending on the relation of the frequencies we observe periodic or quasiperiodic Lissajous orbits. For rational relations of the frequencies e.g u>2 = 2wi we observe closed orbits and for irrational relations the whole box is filled by the quasiperiodic orbits. Let us study now nonlinear mode coupling effects in particular for the case of 1 : 1 and 1 : 2— resonances. For a deeper study we refer to the literature [Rabinovich & Trubetskov, 1984; Anishchenko, 1995; Landa, 2001; Anishchenko et al., 2002]. The simplest case is the coupling of a nonlinear oscillator in a;—direction with a linear oscillator in y—direction through a mixed quadratic and cubic coupling term. The energy transfer between the modes has been proposed as a simple model for the processes in enzyme catalysis [Ebeling, 1985; Ebeling & Romanovsky, 1985]. The hamiltonian dynamics of the systems is defined by the potential U{x, y) = \kx2 + \k'x3
+ ]k"x* - exy + \KV2
+ rjxy2
(1.43)
with K > 0 and k" > 0. The dynamics (without noise and friction) is decribed by d2x m-yY+kx
M
+ k'x2+ k"x3 = ey-T]y2, d2v
~di> +Ky
= ex~
2r x
>y
(1.44)
(L45)
We observe here a mode coupling of the linear modes of the second oscillator on the modes of the first oscillator [Ebeling & Romanovsky, 1985; Ebeling & Jenssen, 1988; Hesse &; Schimansky-Geier, 1991]. Of special interest is the case k < 0, k' — 0, k" > 0 corresponding to a the bistable potential shown in Fig. 1.1. An enhancement of transitions between the wells was investigated in detail (for the case r] = 0) by Hesse-Schimansky-Geier & Ziilicke [Hesse & Schimansky-Geier, 1991]. The effects of coupling between the modes of oscillations for thsi model will be investigated in more detail in Chapter 2 and in Section 4.1 of this book. In particular it will be shown that the energy transfer between the modes of the oscillations may lead to an enhancement of the transitions. Another case of special interest for this book is the Fermi resonance between 2 modes with 1 : 2 relation between the frequencies. This problem goes back to a study of the Raman effect of CO2- molecule. According to Fermi the 3 atomes in the molecule are in one line in the rest position and have 3 oscillatory modes (symmetric and unsymmetric linear stretching and bending) corresponding to the frequencies V\ ~ 1230,1^2 = 673,^3 = 2350. Consequently we find for this molecule v\ ~ IviNonlinear coupling of these modes leads a very specific resonance phenomenon, which is called Fermi-resonance. Later it was found that Fermi-resonances appear
20
Reaction theory and cluster
dynamics
in many systems, e.g. in the NaCl- lattice [Fermi & Rasetti, 1931] and also in biomolecules [Volkenstein, 1946; Shidlovskaya et al., 2000]. Fermi proposed in 1931 a concrete potential model which introduces a third order coupling term into the potential of type yx2. A closely related dynamical model which includes a third order coupling between the x— and the y— oscillations may be found in the well-known book of Pippard. The Pippard-model is defined by the potential [Pippard, 1983] U(x, y) = -kx2 + 2k(y - ex2)2
(1.46)
This potential is symmetric with respect to x but unsymmetrical with respect to y. The minimum in y— direction is located at the parabola y = ex2. The x— and y— oscillations in the Pippard potential are strongly coupled and follow the dynamics. d2x m—Y + kx + 8c2kx3 = 8ckxy,
(1.47)
d2v m—f + Aky = 4cA;z2
(1.48)
We see that the two linear frequencies of the oscillators are in the relation 1 : 2. This leads to rather complex fading oscillations and the so-called Fermi-resonance. In the following we will present several illustrations for the case m = l,fc = l , c = l . In Fig. 1.4 we represented some part of the trajectory x{t). A typical phenomenon for Fermi resonance is the fading of the energy between the x— and y— oscillations. Due to this fading phenomenon we see from time to time rather large x— amplitudes (see Fig. 1.4). This can be observed also on projections of the trajectories in the 4-d phase space on 2-d subspaces. In Fig. 1.5 we see a typical phase portrait projected on the x — y— plane. The following two figures (Figs. 1.6 - 1.7) show projections on planes of containing one coordinate and one velocity (the x — vx— plane and the x — vy— plane). As will be investigated in detail in Chapter 6 the Fermi resonance of perpendicular oscillations may lead to an enhancement of the transitions. We will show this here directly on a combination of two Pippard oscillators to a double well. On the x— axis the double well is defined by:
This double-well potential is parabolic around the two wells and the two relevant frequencies are defined by WQ = /z, w£ = /i/4. The potential barrier has the height
Effects of coupling and resonances on transition
Fig. 1.4 Typical oscillations of the x(t)—coordinate resonance between the two degrees of freedom.
rates
21
of the 2 — d Pippard system showing Fermi
AU u
= * = kb
(1.50)
We couple now the dynamics in a;— direction in such a way with the dynamics in y— direction that around the wells the condition of Fermi-resonance is observed:
U(x,y) = U0(x) +2n[y-
2cU0{x)
(1.51)
This potential is symmetric and bistable with respect to x. On the other hand it is unsymmetrical with respect to y and has the the typical Fermi-banana shape around each of the wells. The minimum in y— direction is located at the fourth order curve
y = 2cU0(x).
(1.52)
A typical landscape of the bistable Pippard-type potentials which we study now is shown in Fig. 1.8. In more general cases the potential may be expressed by higher order polynoms in x and y. The bifurcations in our system are mostly determined by the catastrophes of the potential, i.e. the changes of the number of stationary points (maxima, minima, saddle points). According to Thom, there exist alltogether 7 elementary
Reaction theory and cluster
22
dynamics
30.0
Fig. 1.5
Projection of the trajectory of the Pippard system on the x — y— plane.
lis.
i sv •
(11-": ir
1V''
''5 <|. ' " i i l r' l
'j "i-fl . • li ! ' ( i: ' . I
Fig. 1.6
Projection of the trajectory of the Pippard system on the x — vx— plane.
catastrophes, corresponding to qualitative different changes of the potential. Including noise into the system according to eqs. (1.39-40) we may find a stationary solution of the Fokker-Planck equation in the form
Effects of coupling and resonances on transition
Fig. 1.7
23
rates
Projection of the trajectory of the Pippard system on the x — vy— plane.
P (x,y,vx,vy)
= const, exp
m{v2x + v2y) 2kBT
U{x,y) kRT
(1.53)
As we see, the probability density in x and y is fully determined by the potential U(x,y). Therefore the bifurcations of this probability density are the same as the bifurcations of the potential (Ebeling, Engel & Feistel, 1990, 2001). This underlines the role of the concrete form of the nonlinear potential for the details of the two-dimensional oscillations. In the course of this book we will demonstrate this on many concrete examples modelling molecular phenomena. The following investigation is concentrated on the transitions between the two wells of the double well Pippard oscillator defined by eq.(1.46. We will study in detail the two special parameter sets demonstrated in Fig. 1.8. First we investigate the transition behaviour for the case that the second oscillator is flat i.e. fj, — 0. This corresponds to the one dimensional case
U(x,y) = U0(x)
(1.54)
In this case an analytical estimate has been developed already in section 1.2. We have shown that the corresponding transition times r — v~l have a minimum in dependence on the friction parameter 7. This was demonstrated in Fig. 1.2.
24
Reaction theory and cluster
dynamics
0.50
'-I50-I.00-O.50 0.00 0.50 1.00 1.50 0.50'^ - - - 0.00 -1.50-1.00-050 0.00 0.50 1.00 1.50 Fig. 1.8 Landscape of a 2 — d bistable potential showing Fermi resonance phenomena between longitudinal and transversal transversal oscillations. Two parameter sets (a) ji = 1/2, b — 1/4, U* = 1/16 and (b) fj, = 1,6 = 1/2, U* = 1/8 with different asymmetry in y— direction are shown.
In Fig. 1.9 we presented numerical calculations of the transition time between the bistable wells first for the case that there is no coupling to the y— oscillations. We clearly see the existence of a minimum with respect to the friction 7. Further we have given in Fig. 1.9 a comparision with the case of Fermi resonance. We see that the dynamics of the transitions in the bistable potential with Fermi resonance is quite different from the case of no coupling between the modes. The switching of the the energy between the modes in x— and y—direction may leads to an enhancement of transitions between the wells. The minimum is much lower and is shifted to smaller values of the friction. An application of this effect will be demonstrated in Chapter 6. We will come back to the investigation of couplings of double well transitions to transversal oscillations in Chapter 4. There we will investigate the coupling of transitions between two wells to harmonic oscillations in y— direction (corresponding to k = l,c = 1,/J, = 1/2) including a coupling through a second order mixed term
xy. The model which will be studied corresponds in our notation to:
Effects of coupling and resonances on transition
8000 -i
5000-|
6000
4000-
4000 ^
3000
2000-
2000 A
00.0001 0.001
- i i IIIM|'
0.01
0.1
1
I OCX)
—
1
1 i • IMII|
rates
rii
0.0001 0.001
iiinj
25
r~ri
0.01
11 ni|
0.1
Y T—r i I I I I I I '
1
Fig. 1.9 Transition time as a function of the friction constant, obtained from simulations. Comparision of the transition in a 1 — d double well (dashed lines) with the transition time in a 2 — d well with Fermi resonance. The parameters are (a) \i = 1/2, b = 1/4,(7* = 1/16 and (i = l,b = 1/2; U = 1/8.
V U(x,y) = U0{x) + —
-exy
(1.55)
The dynamics of transitions between the two wells is at least in some approximation described by the Kramers theory for e — 0 and by the Hesse-SchimanskyGeier-Ziilicke theory for e > 0. This will be discussed in detail in Chapter 4. At the end of this section we will summarize the main results again: We started from the idea that enzymatic reaction take place at a specific "optimized site" of the enzyme. This site may have the form of a "pocket" were the reaction takes place under very specific conditions. At the reaction site the local "reactive dynamics" connected with "barrier crossing" is coupled to other oscillating degrees of freedom. In our 2—d model we have only one coupled oscillator. The nonlinear coupling to the second oscillator leads to resonance phenomena. For the "optimal" 1 : 2 relation of the frequncies we obtained Fermi resonance leading to beating of the energy between the reaction oscillator and the bath oscillator. Further we introduced a surrounding of the active site by a kind of optimized "heat bath". In particular the "effective friction" acting on the reaction dynamics at the active site is much lower than in the solution itself due to the shielding of the reaction site by other molecular groups of the enzyme. We have shown for our simple model that there exist an optimal value of the "effective friction" where the transition rates are maximal. In this way for our simple model "optimization" of the coupling to an oscillator bath and a heat bath means: • The relation between the frequency of the reaction oscillator and the bath
26
Reaction theory and cluster
dynamics
oscillator is 1 : 2 what leads to Fermi resonance and energy beating between the modes. • The heat bath realizes a small but not zero "effective friction" were the transition times have a minimum. This way we have made clear at least in the framework of our model what the vague term "optimization" of the surrounding of a reaction site may mean.
1.5
Basic variables. Block and cluster models
The reduction of the system of equations of the high order and the diminishing of the number of the parameters are always appreciated in computational studies and in presentation of the results of computations. Indeed, the methods of molecular dynamics [Mc Cammon and Harvey, 1987] deal with a large number of the internal degrees of freedom of the molecule which results in different computational complications leading to the losses of information on the selected degrees of freedom and, hence, to the loss of the regularities of the molecular dynamics related to the selected degrees of freedom. In course of evolution the biological molecules acquired quite definite structural features allowing energy transfer and redistribution over the degrees of freedom in the most effective way. These structural features allow one to introduce "selected" and "frozen" degrees of freedom which is the basis of the block (or cluster) model of the protein and the concept "protein-machine" [Chernavsky, Khurgin & Shnol, 1987; Chernavsky & Chernavskaya, 1999; Blumenfeld & Tikhonov, 1994]. Protein is a complex nonlinear vibrational system with many thousands degrees of freedom [Karplus, 1982; Havsteen, 1989; Netrebko et al., 1994; Ebeling et al., 1994, Nolting, 2000], that can be considered as several rigid blocks (clusters) bound to each other by soft nonlinear "springs". Thus, the consideration of the block model is not only a simplifying method but also a natural consequence of the structure and dynamics of the protein molecule. This is confirmed also by numerous experimental data (X-ray analysis, Raman and fluorescence spectroscopy, etc. [Karplus, 1982; Birktoft and Blow, 1972; Tsukada and Blow, 1985; Brown et al., 1972]. Note that the processes taking place in the active site (i.e. the catalytic act itself) and involving a relatively small number of atoms must be considered with the allowance of the quantum effects. But it is hardly expedient to apply these methods in the study of the molecule as a whole because of the computational difficulties and because of the fact that the motion of relatively large and rigid clusters can be adequately described by classical mechanics. Below we assume that the enzyme molecule and ESC are complicated nonlinear vibrational systems working as energy transforming and redistributing molecular machines. Such transformations of energy lead to breaking or formation of the
Basic variables. Block and cluster
models
27
chemical bonds in the substrate molecule. "Basic" variables correspond to a small number of the "selected" degrees of freedom taking part in this act. The selection of the basic variables for molecular machines must take into account the facts as follows. If the protein molecule is in equilibrium with the environment then the selected degrees of freedom do not differ from the others. However, they play the main role in the case of inequilibrium situations: formation of the enzyme-substrate complex, sorption of a charged ion, absorption of a light quantum, etc. Typical and well-known example is given by the conformational changes of myosine molecule upon acto-myosine interaction (rotation of the "head" and its return to the initial position). "Macro" changes are possible only if a series of interactions takes place in which the sorption of ATP molecule (substrate of the reaction) and calcium (controls the muscle contraction) are the key stages. Light absorption leads to an inequilibrium situation in such important process as photosynthesis. In this case the typical distances of electron migration are several orders of magnitude larger than the amplitudes of the thermal fluctuations. The authors of the "protein-machine" concept [Chernavsky, Khurgin, Shnol 1967, 1987; Chernavsky, Chernavskaya, 1999], stress that in the molecular machines the mechanical motions along the selected degrees of freedom are accompanied by the displacements of the charges. Below we present the corresponding examples. Consider the block model of a-chymotrypsin (CT) that is one of the most well studied enzymes. According to the X-ray data [Birktoft, Blow, 1972] its globule consists of two parts (subglobules or domains) A and B with a small fragment of a polypeptide chain between them. Aminoacid residues of the binding chain (37-38 and 204-205) play the role of a hinge in the vibrational motions of the subglobules. The active site is located between the subglobules. The catalytic triade of the active site consists of the aminoacid residues Serl95, His57 and Aspl02. Serl95 and His57 are located at different subglobules (Fig. 1.10 (a,b)). There are 250 aminoacid residues in CT molecule and its molecular weight is 24800 D. Each of the subglobules is a stable "structure" the backbone of which consists of six elongated /3-sheet fragments. The subglobules represent two "barrels" connected with each other by a fragment of a polypeptide chain and several H-bonds (Fig 1.10 (b,c,d)). In the block mechanical model taking into account the tertiary structure each of the parallel fragments of the polypeptide chain is considered as a "rod" and the H-bonds between the "rods" are simulated by nonlinear springs (Figs. 1.10 (c,d)). It is assumed that the connecting fragments of the polypeptide chain (wavy solid lines in the Figures) move freely being independent on the vibrations of the clusters ("rods"). Thus, the motion of these fragments does not influence the dynamics of the process. These fragments must be more mobile in comparison with the clusters (rigid "rods"). This assumption agrees well with the results on motility of the aminoacid residues in the CT polypeptide chain [Havsteen, 1989,1991]. All
28
Reaction theory and cluster
dynamics
aminoacid residues of the connecting loops exhibit high motility. All the motions result from random collisions of the protein molecule with the surrounding small water molecules and larger substrate molecules. In order to reveal the degrees of freedom and corresponding motions most important for adequate description of the events in the active site, consider ESC scheme consisting of CT molecule and peptide chain (Fig. 1.11). Consider the behavior of the rigid cluster (unchanged fragment of the substrate), e.g. block A in Fig. 1.11 at the entrance to the AS pocket and inside it with the allowance of the interaction with water molecules. All atomic groups surrounding AS contribute to a 3-d potential landscape inside AS and in the area near its entrance. Substrate molecules, the fragments of substrate, the products of hydrolysis, water molecules, and, possibly, some ions move in this potential landscape. Note that phenylalanine residue is the most complementary one to the CT active site [Volkenstein, Golovanov & Sobolev, 1982]. Substrate binding in the active site is frequently related to the action of the electrostatic field generated by the total charge of the aminoacid residues of the protein globule. When inside the AS pocket the substrate does not go immediately to the deepest potential well in the vicinity of the catalytic group which corresponds to the complementary binding but goes first to the potential wells of smaller depths. Therefore, we consider the problem of a "test particle" (cluster or an unchanged part of the substrate) transfer from one potential well into another [Netrebko et al., 1994; Ebeling et al., 1994; Netrebko et al., 1996]. It is likely that in the general case the diffusion of ligands in proteins is related to such sequential transitions [Shaitan, 1994]. The transitions take place under the conditions of fluctuating potential landscape that is determined by the motions of the subglobules and clusters of the enzyme. Now we consider the problem of proton transfer in the OH...N H-bond schematically shown in Fig. 1.10 (see [Fersht, 1977, 2000; Volkenstein, Golovanov and Sobolev, 1982; Popov, 2000; Khurgin, Burshtein, 1974] for details). One minimum is situated near oxygen atom and another near nitrogen atom. Proton can approach also the minimum near the atoms N and C of the cluster A (Fig. 1.11). This process is one of the most important steps in the chain of transformations that lead to peptide bond breaking. Indeed, if the proton leaves the oxygen atom, then the latter acquires the charge —e. Negatively charged oxygen attacks (due to random fluctuations) the peptide bond and loosens it which results in its breaking. Such a process is several orders of magnitude more probable than the spontaneous breaking of the peptide bond in water in the absence of the enzyme. Such a scheme of the proton transfer is valid also for splitting of acetylcholine in the active site of acetylcholinesterase [Fersht, 1977; Quinn, 1987; Zundel, 1992, 1997]. Both an overbarrier proton transfer (classical effect) and proton tunnelling (quantum-mechanical effect) are possible in the general case in a 3-d potential landscape [Romanovsky, Chikishev & Khurgin, 1988; Romanovsky, Khurgin, Chikishev, 1997]. The parameters of the potential landscape, in particular, the height and
Basic variables. Block and cluster
c)
models
d)
Fig. 1.10 (a) A "ribbon" presentation of chymotrypsin molecule (light and dark ribbons denote b-sheets belonging to different subglobules; (b) a simplified block model of a two-domain chymotrypsin molecule (a dot designates a "hinge" axis, His57 and Serl95 are the amino acids of the active site); (c), (d) each domain represents a spatial structure consisting of six "rods" (H-bonds are schematically shown only for the A domain; the dots indicate the position of the " hinge").
Fig. 1.11 (a) schematic presentation of chymotrypsin molecule consisting of two subglobules A and B and the substrate (polypeptide chain). N+H-O hydrogen bond is an important element of the active site. Note that nitrogen and oxygen belong to different subglobules (the distances are given in nm). Dashed square shows the substrate fragment under attack, (b) The unchanged part of the substrate is complementary to the active site. Point O shows the position of the "hinge".
width of the barrier, fluctuate due to fluctuations of the clusters and the unchanged part of the substrate. As the potential barrier is rather high (more than 40 kcal/moleunder the assumption of the "linear" H-bond) the probability of the direct transfer of proton is rather low. That is why it is important to consider an alternative way related with the appearance of the third minimum due to interaction with the substrate [Khurgin, Burshtein, 1974; Netrebko et al., 1994; Ebeling et al., 1994; Netrebko et al.,1996]. Note that the problem of proton transfer in a series of potential wells is discussed also in membrane biophysics. The soliton type of such a process
30
Reaction theory and cluster
dynamics
is demonstrated in [Manevich et al., 1994]. The problem of the proton transfer in CT AS is discussed in further details in Chapters 6 and 8. In Chapter 7 we consider also ACE structure and functioning. Thus, the catalytic act can be considered as consisting of several stages each of which can be described by the methods of molecular dynamics. At each stage the number of the "particles" under consideration is much less than that in the full model. Each "sub-model" has its own peculiarities and is of special interest for stochastic dynamics.
1.6
The problems under consideration
Below we list the most important problems of molecular dynamics that are discussed hereafter. All of them are somehow related to the enzymatic catalysis. Note that all such problems must be solved in 2-d or 3-d space because of the importance of the steric factors. The problems considered in 1-d space are only auxiliary ones. 1. The problem of energy distribution over the degrees of freedom in a thermal "bath" inside which a number of "solid" and one "soft" molecule interact with each other (Lennard-Jones potentials). 2. The problem of energy concentration in the "soft" element under the conditions of a quasi-soliton propagation in a chain of masses connected with each other by nonlinear springs. 3. Specification of the Arrhenius formula for the chemical reaction rate in the case of the allowance of the color noise in the Kramers problem. 4. Kramers problem in a bistable 2-d system considered in item 1. 5. Kramers problem in a bistable 2-d system for the case when the Fermi resonance condition is met for orthogonal vibrational modes. 6. The problem of the particle transfer from one potential well into another when the parameters of the 2-d potential landscape experience periodic changes and for the case when there are two or three minima in the landscape. 7. The role of stochastic resonance in the case of a particle transfer in 2-d potential landscape with two minima. 8. The problem of quantum-mechanical transitions (tunnelling) in 2-d and 1-d potential relieves with time-dependent parameters. The problem of proton migration in the system of H-bonds of CT AS. 9. The problem of diffusion limitation of the rate of ester bond breaking in acetylcholine in the case of its hydrolysis by ACE. The problem of penetration of the interacting Brownian particles of complex shape into the fluctuating cleft of AS and the problem of the reaction products escape through the entrance and the possible role of the "back door". 10. The question of the damping of the oscillations of subglobuls and clusters of enzymes in water. 11. Several problems of transitions between conformations in protein molecules.
The problems under
consideration
31
Therefore, we discuss and solve several problems of molecular dynamics. In each case we use the results for interpretation of certain stages of the AC and ACE catalyzed reactions. Of course, similar problems were discussed by many authors. We shall refer to their works in the discussion of the results. Here we mention two well-known theoretical models [Davydov, 1984, 1986; Yakushevich, 1998] of solitons and excitons in helical structures of proteins and nucleic acids.
References V.S. Anishchenko (1995): "Dynamical Chaos - Models and Experiments", World Scientific, Singapore 1995. V. Anishchenko, A. Neiman, T. Vadivasova, V. Astakhov, L. Schimansky-Geier (2002): "Dynamics of chaotic and stochastic systems", Springer, Berlin. N.K. Balabaev, A.S. Lemak et al. (1995): "PUMA CD - computer codes for the simulation of the molecular and collision dynamics of biological macromolecules", NIVTs Russian Academy of Sciences, Pushchino 1995. F.C. Bernstein, T.F. Koetzle, B.G.J. Williams, E.F. Meyer, M.D. Brice, J.R. Rodgers, 0 . Kennard, T. Shimanouchi, M. Tasumi (1977): "The protein data bank: a computer-based archival file for macromolecular structures", J. Mol. Biol. 112, 535-542. J.J. Birktoft, D.N. Blow (1972): "Structure of crystalline a-chymotrypsin", J.Mol.Biol. 68, 187-240. L.A. Blumenfeld, A.N. Tikhonov (1994): "Biophysical Thermodynamics of Intracellular Processes. Molecular Machines of the Living Cell", Springer, New York, Berlin, Heidelberg. C. Branden, J. Tooze (1999): "Introduction to protein structure", Garland Publ., N. Y. London H. Brown, S. Erfus, E. Small, W.L. Peticolas (1972): "Conformationally dependent low-frequency motion of proteins by laser RAMAN-spectroscopy", PNAS USA 69, 1467. L. Charles Brooks III, Martin Karplus and B. Montgomery Petitt (1988): "Proteins. Theoretical Perspective of Dynamics, Structure, and Thermodynamics", Series Advances in Chemical Physics Vol. 71, John Wiley and Sons, New York.
32
Reaction theory and cluster
dynamics
D.S. Chernavsky, Yu.I. Khurgin, S.E. Shnol (1987): "The Concept of 'proteinmachine' and its consequences" (in Russian), Molek. Biol. 20, 1356-1368. D.S. Chernavsky, N.M. Chernavskaya (1999): "Protein-machine. Biological and macromolecular constructions" (in Russian), Izd. Mosk. Universiteta, Moscow. W. Ebeling (1985): "Thermodynamics of selforganization and evolution", Biomed. Biochi. Acta 44, 831-838. W. Ebeling, Yu.M. Romanovsky (1985): "Energy Transfer and Chaotic Oscillations in Enzyme Catalysis", Z. Phys., Chem. Leipzig 266, 836-843. W. Ebeling, M. Jenssen (1988): "Soliton dynamics and energy trapping in enzyme catalysis", Z. Phys. Chem. 1, 269-279; Physica D 32, 183-193. W. Ebeling, A. Engel, R. Feistel: (1990): "Physik der Evolutionsprozesse", AkademieVerlag, Berlin, Russ. TransL: URSS, Moscow (2001). W. Ebeling, M. Jenssen (1992): "Trapping and fusion of solitons in a nonuniform Toda lattice", Physica D 32, 183-193; Physica A 188, 350. W. Ebeling,W. & M. Jenssen (1991) "Soliton-Assisted Activation Processes", Ber. Bunsenges. Phys. Chem. 95, 356-362. W. Ebeling, Yu. Romanovsky, Yu. Khurgin, A. Netrebko, N. Netrebko, E. Shidlovskaya (1994): "Complex regimes in the simple models of molecular dynamics of enzymes", Proc. SPIE 2370, 434-447. W. Ebeling, V. Podlipchuk k A.A. Valuev (1995): "Molecular Dynamics Simulation of the Activation of Soft Molecules Solved in Condensed Media", Physica A 217, 22-37. W. Ebeling & V.Yu. Podlipchuk (1996): "Molecular Dynamics of Time-Correlations in Solutions", Z.physik.Chem 193, 207 - 212. W. Ebeling,W. & M. Jenssen (1991): "Soliton-Assisted Activation Processes", Ber. Bunsenges. Phys. Chem. 95, 356-362. A. Fersht (1977): "Enzyme Structure and Mechanism", Freeman & Co., Reading and San Francisco A. Fersht (2000): "Structure and Mechanism in Protein Science", Freeman & Co.,
The problems under
consideration
33
New York H. Frauenfelder k P.G. Wolynes (1985): "Rate Theories and the Puzzles of Hemoprotein Kinetics", Science 229, 337-345. P. Hanggi, P. Talkner, M. Borkovec (1991): "Reaction-rate theory: fifty years after Kramers" ,Rev. Mod. Phys. 62, 251-341. B. Havsteen (1989): "A new principle of enzyme catalysis: coupled vibrations facilitate conformational changes", J. Theor. Biol. 140, 101-109. B. Havsteen (1991): "A stochastic attractor participates in chymotrypsin catalysis. A new facet of enzyme catalysis", J. Theor. Biol. 151, 557-571. J.J. Hesse, L. Schimansky-Geier (1991): "Inversion in Harmonic Noise driven bistable oscillators", Z.Phys. B 84, 467-470. M. Karplus (1982): "Dynamics of proteins", Ber. Bunsenges. Phys. Chem.86, 386-240. T. Keleti (1986): "Basic Enzyme Kinetics", Akademiai Kiado, Budapest. Yu. I. Khurgin, K.Yu. Burstein (1974): "Mechanism of proton transfer in reactions of alphachymotrypsin" (in Russian), Doklady Akademii Nauk SSSR 217 965-976. P.S. Landa (2001): "Regular and chaotic oscillations", Springer, Berlin 2001. L.I. Manevich, A.V. Slavin, V.V. Smirnov, S.N. Volkov (1994): "Solitons in nongenerated sytems" Physics Uspekhi 164, No. 8, p. 937-958. J.A. McCammon, S.C. Harvey (1987): "Dynamics of Proteins and protein acids", Cambridge University Press. A. Netrebko, N. Netrebko, Yu. Romanovsky, Yu. Khurgin, E. Shidlovskaya (1994) "Complex modulation regimes and vibration stochastization in cluster dynamics models of macromolecules" (in Russian), Izv. Vuzov: Prikladnaya Nelineinaya Dinamika, 2 26-43. A. Netrebko, N. Netrebko, Yu. Romanovsky, Yu. Khurgin, W. Ebeling (1996): "Stochastic cluster dynamics of enzyme-substrate complex" (in Russian), Izv. Vuzov: Prikladnaya Nelineynaya Dinamika 3 53-66.
34
Reaction theory and cluster
dynamics
B. Nolting (2000): "Protein folding kinetics, biophysical methods", Springer, Berlin P.J. Park, W. Sung (1998): "Polymer release out of a spherical vesicle through a pore", Phys. Rev. E 57, 730-734. A.B. Pippard (1983): "The Physics of Vibration. The simple vibrator in quantum mechanics", Cambridge University Press, Cambridge, London, New York, Melbourne, Sidney J. Popielawski, Ed. (1989): "The dynamics of systems with chemical reactions", World Scientific, Singapore 1989. E.M. Popov (2000): "Structure and Function of the Protein" (in Russian), Nauka, Moscow. D.M.Quinn (1987): "Acetylcholinesterase: Ensyme structure, reaction dynamics, and virtual transition states", Chem. Rev. 87 955-979. M.I. Rabinovich, M.I. Trubetskov (1984) "Introduction into the theory of oscillations and waves" (in Russian), Nauka, Moscow. R.V. Raryi, N.L. Klyachko, E.A. Borisova, Cortes C.J. Penagos (1999): "Lectin like center in the molecule of chymotrypsin", Biochem. Mol. Biol. 36, 31-37. Yu.M. Romanovsky, A.Yu. Chikishev, Yu.I. Khurgin (1988): "Subglobular motion and proton transfer model in a -chymotrypsin molecule", J. Mol. Catal. 47, 235-240. Yu.M. Romanovsky, N.V. Stepanova, D.S. Chernavsky (1975): "Mathematical models in biophysics" (in Russian), Nauka, Moscow Yu.M. Romanovsky, N.V. Stepanova, D.S. Chernavsky (1984): "Mathematical biophysics" (in Russian), Nauka, Moscow
K.V. Shaitan (1994): "Dynamics of electrone-conformational transitions and new point of view on physic mechanisms of molecule functioning" (in Russian), Biophysics 39 949. J.E. Straub, B.J. Berne (1986) "Energy diffusion in many-dimensional Markovian systems", J. Chem. Phys. 85, 2999-3006.
The problems under
consideration
35
W. Sung, P.J. Park (1996): "Polymer translocation through a pore in a membrane", Phys. Rev. Lett. 77, 783-786. J. Troe (1991): "On the application of Kramers' theory to elementary chemical reactions", Ber. Bunsenges. Phys. Chem. 95, 228-232. M.V. Volkenstein (1981): "Biophysics" (in Russian), Nauka, Moscow M.V. Volkenstein, LB. Golovanov, V.M. Sobolev (1982): "Molecular orbitals in enzymology" (in Russian), Nauka, Moscow. S. Weiner, P. Kollman, D. Case, U. Singh, C. Chio, G. Alagona, S. Profeta, Jr., P. Weiner (1984): "A new force field for molecular mechanical simulation of nucleic acids and proteins", J. Am. Chem.Soc. 106, 765-784. L.V. Yakushevich (1998): "Nonlinear physics of DNA", Wiley, Chichester, New York 1998. G.N Zatsepina (1998): "Physical properties and the structure of water", Izd. Mosk. Universiteta, Moscow. G.Zundel (1992): " Proton polarizabilite and proton transfer processes in hydrogen bonds and cation polarizabilities of other cation bonds - their importance to understand processes in electrochemistry and biology", Trends Phys. Chem 3, 129-156. G. Zundel (1997): " Proton polarizability of hydrogen bonds and proton transfer processes, their role in electrochemistry and biology", Miinchen University Press, Munich
This page is intentionally left blank
Chapter 2
Tools of Stochastic D y n a m i c s L. Schimansky-Geier and P. Talkner
2.1
Introduction
For most topics treated in this book it is of fundamental importance to understand the motion of classical particles in complex potential landscapes and, in particular, the escape of particles from locally stable states. An escape process only is possible if the particle can exchange energy with its environment in order to trespass intermediate states with high potential energy. These events can be treated by the help of various mathematical and computational tools some of which will be introduced in this section. Molecular dynamics provides a most powerful approach to gain information about the dynamical properties of large assemblies of particles with known interactions. The primary objects of this methodology are the trajectories of the considered system, say a macromolecule swimming in water. The mutual interactions of the macromolecule and the water molecules require to simulate the time evolution of the full system. From the resulting huge amount of detailed information about the state of each single molecule the interesting properties of the macromolecule, as for example its diffusional properties, then have still to be extracted. A less detailed description that also captures the relevant aspects of the problem under study can often be achieved by means of the theory of stochastic processes. There, the uninteresting part of the problem, the dynamics of the water molecules in the above example, is modeled in some crude manner that correctly reflects the statistical behavior of many collisions between the macromolecule and water molecules but not the detailed process of a single molecular encounter. In the mentioned example the force of the water molecules on the macromolecule is split into an average friction force and a fluctuating force. For small velocities of the macromolecule relative to the average motion of the water molecules the Stokes' law describes the friction force. The fluctuating force then is often modeled by Gaussian white noise the strength of which is related to the temperature of the water and to the friction constant according to Einstein's formula. The simplifications of the stochastic method compared to molecular dynamics 37
38
Tools of Stochastic
Dynamics
pay off in a smaller computational effort needed to investigate the relevant part of the system. In a simulation only a few stochastic differential equations describing the dynamics of the relevant degrees of freedom have to be solved in contrast to a huge number of Newton's equations of motion for all microscopic degrees of freedom. Both methods rely on statistical mechanics: In molecular dynamics the initial conditions usually are drawn from a distribution of points in phase space describing the system in a particular thermodynamic state. In the stochastic framework the statistical properties of the random forces and their relations to transport coefficients are dictated by relations from statistical mechanics, as for example the Einstein relation in the case of a diffusing macromolecule. In general, the final results of both methods are averaged quantities such as mean values of macroscopic quantities and correlations of fluctuations about the mean values and their Fourier transforms giving the according power spectra. In the stochastic framework, the most probable fluctuations about the mean values of macroscopic quantities turn out to be very small. However, there are important exceptions from this rule where large deviations play a crucial role. The escape from a metastable state is a typical example. In this process the system has to visit remote regions in phase space where it has a much higher energy than in its initial and final states and where therefore it is found only rarely. This chapter gives an introduction to stochastic methods and their underpinnings by statistical mechanics (excellent enlarged presentations can be found, for example, in: Becker, 1956; Landau Lifshitz 1971; Klimontovich, 1975, 1982, 1994). We will always deal with a few relevant degrees of freedom which we will refer to as velocities and coordinates of a single or a few interacting particles which are in contact with a thermal bath. We will study their dynamics in equilibrium and their relaxation to equilibrium. For a particle with metastable states the waiting times in these states and the transition rates between these states will be of particular interest. This chapter is a tutorial into stochastic dynamics. There are a large number of exciting monographs, reviews and collections of articles which show the vitality of applications of stochastic methods in physics. We have listed only a smaller number in the reference list (Stratonovich, 1963,1967; Kuznetsov et al., 1965; Tikhonov, Mironov, 1979; Gardiner , 1982; Horsthemke, Lefever, 1983; Risken 1984; Malchow, Schimansky-Geier, 1985; Montroll, Lebowitz, 1987; Moss, McClintock, 1990; van Kampen, 1992; Hanggi Jung, 1995; Hanggi, Talkner, 1995; Kogan, 1996; Schimansky-Geier, Poschel, 1997; Anishchenko et al., 1999; Garcia-Ojalvo, Sancho,1999; Freund, Poschel, 2000; San Miguel, Toral, 2000; Schimansky-Geier et al., 2000; Reimann, 2002; Anishchenko et al, 2002). Readers interested in detailed mathematical foundations and background are refereed to mathematical monographs (Wong, 1971; Karlin, Taylor, 1975; Arnold, 1992).
Fluctuations
2.2 2.2.1
in statistical
39
physics
Fluctuations in statistical physics The canonical
distribution
We will deal with a system being in thermal equilibrium with its bath and assume that the combined system plus bath is isolated. The total system then possesses a constant energy E0 = Hs + Hb and the volume V0 — Vs + Vb. The subscripts b and s stand for the bath and the system, respectively. Here we have neglected the contribution of the interaction between the system and the bath to the total energy. It indeed is small if both the system and the bath are macroscopic and the interaction is short ranged. Microstates of the bath and the system are labelled as Tb and T s referring to points in the respective phase spaces. For an isolated system the equilibrium probability to find the system and the bath in an infinitesimal volume of size dTbdTs at the phase space point (r s ,rt,) has the important property that it maximizes the entropy. As a consequence, the resulting microcanonical distribution is given by the equipartition of the microstates at the given energy: dp = constant6{HS{TS) + Hb(Tb) - E0)dTsdTb.
(2.1)
If the phase space volume is measured in units of Planck's constant taken to the power of the total number N of degrees of freedom, the normalization constant coincides with the inverse number of quantum states per unit energy of the total system at the energy EQ and volume Vo:
constant
= sukwr
(2 2)
'
Multiplying the density of states by an energy 5E which is small on the macroscopic scale we obtain the number of states W0{E0, Vb) = n0(E0, V0)5E in the phase space shell of width 8E at the energy EQ. With Boltzmann's formula this number can be expressed in terms of the thermodynamic entropy of the bath and system So{Eo,V0) = kB lnW0(E0,V0)
(2.3)
where k-g denotes the Boltzmann-constant which is the fundamental constant of statistical mechanics of equilibrium systems. It turns out that the actual value of the width of the energy shell has practically no influence on the entropy of a macroscopic system and therefore presents no real arbitrariness in the fundamental relation between the density of states and the entropy. Our next goal is to determine the probability with which one finds the system in the volume Vs at the energy E3 if the isolated total system plus bath has the energy EQ and the volume VQ. This probability is given by the sum of all those phase points weighted by the microcanonical distribution that fulfill the required
40
Tools of Stochastic
Dynamics
condition that the system has the energy E: P{E,VS)
=
fd{Hs-E)dp
= W.V.)0**-™-*)
(2.4)
ilo[E0, V0) Here we used the microcanonical distribution and introduced the density of states of the isolated bath and system at given respective energies and volumes: :(E,V)
=
,{E,V)
=
J5{Hs-E)dFs fs{Hb-E)dTb
(2.5)
The first factor in the second line of eq. (2.4) gives the number of states of the isolated system at energy E. Therefore, the second term represents the probability p{E, Vs) to find the system at energy E. Multiplying the denominator and numerator of this factor by the same energy 5E we obtain the ratio of numbers of states in the bath and in the total system at their energies and volumes. According to eq. (2.3) these numbers can be expressed in terms of the respective entropies:
P(E, Va) = exp (J- AS0(E, V.)\
(2.6)
where ASo{E, Vs) denotes the entropy difference of the bath and the total system: AS0(E, Va) = Sb(E0 - E, Vb) - S0(EQ, V0)
(2.7)
For a macroscopic bath the entropy is again related to the number of states Wb(E, V) Clb(E, V)SEb in the form of eq. (2.3): Sb(E0 -E,V0-Vs)
= kBInWb(E0
-E,V0-
Vs))
(2.8)
Introducing the free energies of the total system and the bath So(E0, Vb) = ^ (E0 - F0(T, V0)) ; Sb(E0 - E, Vb) = ±(E0-E-
(2.9)
Fb(T, Vb))
we find for the entropy difference AS 0 = ~ (Ft(T,Vs)
- E)
(2.10)
where Fs = Fo - Fb is the free energy of the system. This result combined with eq. (2.4) yields the canonical distribution of a system in thermal contact with a bath of temperature T P(E,VS)
= ils{E,Vs)exp(j^(Fs(T,Vs)
- £))
(2-n)
Fluctuations
in statistical
41
physics
From normalization one obtains the fundamental relation Ft(T,V.) = -kBT
\nj
Qs(E,Vs)exp (-j^)
*E.
(2.12)
that expresses the free energy of the systems in terms of the bath temperature and the density of states of the isolated system. We note that the distribution is determined by the difference of the free energy and its value in the constrained case, FS(T,VS) - E. Below, we will recover this particular structure for the distributions of more general observables. 2.2.2
Einstein's
formula
In 1905 Albert Einstein derived the probability distribution for arbitrary observables of a thermodynamic system. In thermodynamic equilibrium an observable which we denote by x in general will fluctuate. A simple example is given by the end to end distance of a linear polymer chain with N bonds immersed in a fluid. Another example is the position of an adiabatic piston between two gases at equal temperature. Both quantities fluctuate and are characterized by probability distributions. For the sake of simplicity we will assume that x can be expressed as a phase space function of the microscopic variables T s
x = x(r s ).
(2.13)
Further we restrict ourselves to isolated thermodynamic systems, i.e. Es, Ns, Vs are constants. So we look for the probability distribution P(x)dx of the fluctuating value x in the interval [x, x + dx] in this isoenergetic case. Because the condition x = const represents a cross-section at the hypersurface Es = const the probability to find the system at x will be proportional to the number of states that lie in this cross-section. Einstein solved this problem by using the statistical definition of the thermodynamic entropy. Under the assumption of a conditioned equilibrium he introduced the conditional entropy S (x\Es,Ns,Vs) defined by the number of microscopic states fl(x\Es,Ns,Vs) that realize this certain value of x. One again uses Boltzmann's formula S{x\Es,N3,Vs)
= kBlnn(x\Es,Na,Vs)
.
(2.14)
The probability distribution density of finding x at a certain value follows from its geometric definition ^[x\^s,Vs,^s)-
n(x\Es,Ns,Vs) n(E,N,V)
[2Ab)
42
Tools of Stochastic
Dynamics
or, respectively, according to (2.13) P[x) = exp J - (S(x\Es, NS,VS) - S(ES, Ns, V.))
(2.16)
KB
where S(ES,NS, V3) is the unconstrained thermodynamic entropy. Again the probability of the conditioned state is expressed by the reduction of entropy caused by the fixation of x. By virtue of eq.(2.15) one easily proves that normalization is obeyed. As a result one need not calculate the full thermodynamic entropy S(ES,NS,VS) which some times may be difficult. Obviously, it holds exp
-JTS(X)
Jdxexp-±S(x) where we write S(x\Es,Ns,Vs) — S(x) for short. Only that part of entropy enters which functionally depends on the considered degree of freedom whereas constant distributions drop due to normalization. 2.2.3
Fluctuations
around
equilibrium
Let us list some general properties of the probability distribution near the equilibrium value xeq. First of all, at its equilibrium value xeq thermodynamic potentials have extremal properties with respect to a;. As a statement of the second law in the isoenergetic case the entropy has a maximum there, S(x) —> Smax — S(xeq) if xeq is reached. Therefore, in the neighborhood of a stable equilibrium state the entropy can be expanded as 1 S(x) = S{xeq)--kBg{x-x 2 eq)
2
(2.18)
with
If higher terms in the expansion can be neglected it follows that fluctuations around the equilibrium state are well approximated by Gaussian distributions. If the majority of probability is in the neighborhood of xeq one may extend the range of the Gaussian till infinite values. The normalized probability density to find the state x is given by P{x) = \j — exp
1
i
-g(x-xeq)
\>
(2.20)
Fluctuations
in statistical
43
physics
The standard deviation is determined by the second derivative of the entropy = - . 9
(2.21)
In case of thermodynamic variables, i.e. for x being temperature, pressure, volume or particle numbers, the second derivative of entropy is a susceptibility like the heat capacity, compressibility or derivations of the chemical potentials. These susceptibilities describe the elasticity of thermodynamic systems in response to applied changes of the system and define the stability of thermodynamic states. According to (2.21) these functions determine the width of the probability distributions around equilibrium. For a set of n fluctuating variables (n
det(q) (X) = \l T E ^ f (2*)»
eX
x xx x x „ / ]9i,j{ P -^} i-iieq)(xieq) j- jeq)\xj J 9i,j \ ^ i,j
x
j
eq)
•
(2.22)
Therein 9hl
~
I f d2S \ kB [dx.dxj)s=Seq
(2.23)
ments are given by the inverse matrix i(xi~
' xieq) \xj
x
jeq)) — 9iJ •
(2.24)
Analogously, for other thermodynamic constraints (for example, isothermic or isobaric closed systems) one obtains Gaussian distributions where g is the positive second derivative of the corresponding thermodynamic potentials, F(x\Ts,Vs,Ns) or G{x\T„Ps,Na). 2.2A
Perrin's
pendulum
As an example we discuss the motion of a sensitive torsion pendulum in a gas at temperature T. The gas together with the pendulum form an isolated thermal system with constant overall energy E0. The single degree of freedom describing the angular motion of the pendulum is the angle <j>. The surrounding gas with a finite number of particles permanently hits on the pendulum causing fluctuations of the pendulum around a rest position (f>eq at which the entropy is maximal with respect to 0. What then is the probability to find deviations from the equilibrated rest position 4>eq ? For simplicity we set
eq = 0. The advantage of Einstein's approach consists in the calculation of the entropy reduction AS based on thermodynamic balance equations. Hence, one may use the
44 1
1
i
•
i
i
S(E)
-
S(x,E) .
/ '
/
/
S=const i
Fig. 2.1
I
.
I
i
.
Determination of the entropy of an occurring
fluctuation
Gibbsian fundamental equation
dS(E,x1,...,xn)
= — I dE - 22
x dx
i i
(2.25)
with X\ = . In fig.(2.2.4 the unconstrained entropy S(E) is plotted with respect to energy E. Deviations from the equilibrium state of cj> decreases the entropy along E = const which is an irreversible process (see vertical line in Fig.(2.2.4)). S(x,E) can be found from the concept of the reversible "Ersatzprozefi". It means that we calculate the entropy shift performing reversible work with 5 = const which is called the minimal work (horizontal line in Fig.(2.2.4). Applied to the pendulum the internal energy changes as dE{4>)
dRr,
•D(f>d
(2.26)
with D being the modulo of torsion. For small deviations (see Fig.2.2.4) the needed entropy shift relates as
d5
^ =(§)/*
(2.27)
and one gets
S(d>) = S(0) - — Df
(2.28)
Therefore we find P{4>)
D exp 2nkBT
D 2kBT
(2.29)
Perrin measured in experiments the standard deviation (2.30)
Fluctuations
in statistical
45
physics
and therefrom he found Boltzmann's constant k&- From this result he was able to find Avargado's number by NA = £
(2.31)
with R being the universal gas constant. It was one of the first findings where the consideration of fluctuations gave a strong support to prove the molecular structure of matter. 2.2.5
General
approach
Here, we give a more formal derivation of the probability density of a fluctuating variable x = x(Ts). Using Dirac's 5-function one obtains P(x) = j 5{x - x{Ts)) p ( r s ) d r s .
(2.32)
Therein p(Ts) is the probability of a microstate in a certain thermodynamic contact. The integration of p multiplied with the ^-function just collects the probability of those microstates which realize the given value x for the function x(Ta). In case of an isolated system the conditional entropy (2.14) then reads S{x\E,N,V)
= kB In f S {x - x(Ts)) S{H{TS) - E) ATa .
(2.33)
The probability distribution follows accordingly P{x) = exp ( - ^ j ^ )
JHX- x(ra)) 5(H(TS) - E) drs.
(2.34)
Immediately one finds the distribution (2.16). Formally this approach can also be extended to other distributions describing different thermodynamic constraints. For example if a; is a fluctuating value inside an isothermic-isochoric many-particle system the projection procedure yields
where F(x\T,V) =-kBTln f S{x-x(TB))exp(-^^\
dTs
(2.36)
is the conditional free energy of the isothermic-isochoric system for a fixed value of x.
46
2.3
Tools of Stochastic
Dynamics
Linear relaxation processes
In statistical physics macroscopic values £ of a thermodynamic system permanently fluctuate around their equilibrium state with maximal entropy. The probability distribution for these fluctuations was given in the last chapter. The actual sample path of x(t) most of the time visits nonequilibrium states. In this section we are concerned with the time evolution of the mean value of sample paths brought out of equilibrium by an initial impact at to. That might be realized by a thermal fluctuation as well as by an external perturbation. We will look at the relaxation of the mean (x(t))x(to)=Xo conditioned that it exactly starts at XQ at t — to = 0. Prom the 2nd law of thermodynamics one knows that the equilibrium state is a stable attractor of the dynamics of the mean value. Entropy is a Lyapunov function with respect to perturbations around xeq. The mean value as function of time obeys a dissipative law. Taking the entropy as function of the actual value and derivating with respect to time gives dS = — x > 0.
S(x)
(2.37)
Two new important values of non-equilibrium thermodynamics appear in this equation,
x
= W,
{2M}
is called in analogy to mechanics the thermodynamic force, where the entropy plays the role of a mechanical potential. The second term on the right hand side (2.37) J = x
(2.39)
is the thermodynamic flux. Lars Onsager gave a general connection between both values. Referring to the circumstance that in dissipative systems there is no flux without a force he concluded that near to equilibrium the linear connection J = L X
(2.40)
must hold with L being a constant. Obviously, the 2nd law (2.37) requires that L is strictly positive, i.e., S(x)
= L X2 > 0 .
(2.41)
This first of Onsager's hypothesizes about the linear connection between thermodynamic forces and fluxes led to the formulation of the thermodynamics of linear dissipative system. The kinetic coefficients L are called Onsager-coefHcients.
Linear relaxation
processes
47
In the neighborhood of the equilibrium state it holds with the approximated entropy from eq. (2.18) dS X = — = -kBgx.
(2.42)
With (2.39) and (2.40) we find the linear relaxational dynamics for non-equilibrium states x — -LkBgx
= -Xx
(2-43)
where X = LkBg
(2.44)
denotes the relaxation coefficient of the variable x. This value which occurs here as a result of our consideration describes the macroscopic behavior of a thermodynamic system brought out of equilibrium. Eq. (2.43) is readily integrated to yield for the relaxation of the variable x starting from the nonequilibrium value XQ M<))*(o)=x0 = zo exp (-A (t - t0)) .
(2.45)
Therefore, A - 1 plays the role of the decay time of the initial state. On the other hand, A stands in close relation to the fluctuational properties. Indeed, inserting (2.21) we obtain the static fluctuation-dissipation relation usually called Einstein's relation
A = kB J L .
(2.46)
It connects the standard deviations of the fluctuations with the relaxation coefficient of the dissipative dynamics for the mean value (comp.also eq. (2.30)). This relation is plausible because if the standard deviation of the distribution is small the relaxation of possible deviations has to be fast, as well. Both, the restoration of equilibrium and the deviation from equilibrium are counteracting but strongly connected in equilibrium. One can proceed in a quite similar way if dealing with several thermodynamic values. Indeed, taking the entropy as function of X{ (i = 1 . . . n) the entropy production reads S((Xl),...,(xn))=
g^(ii)
(2-47)
We agree, here and furtheron, to sum over repeating indices. According to Onsager, for the thermodynamic forces and fluxes X i =
dx~-'
Ji
=
il
(2-48)
48
Tools of Stochastic
Dynamics
a linear relation exists Ji = LijXj
(2.49)
with a symmetric and positive definite matrix of transport coefficients Lij. We later will come back to the symmetry. The positivity is an immediate consequence of the second law of thermodynamics. Performing the derivative in eq. (2.48) with the Taylor expanded entropy we obtain equations of motion for the variable x\ ±i - -kBLijgjtkxk.
(2.50)
The relaxational coefficients of the linear processes near equilibrium states are given by K,j = kBLitkgkj
.
(2.51)
Since g determines the stationary standard deviation we again meet the deep relation between fluctuations and and counteracting dissipation. 2.4
Correlations and spectra
Fluctuating processes can be quantified by their stationary or time dependent moments. Between them the first and second achieve central interest and are often sufficient to investigate the behavior of fluctuations. In equilibrated systems the first moment of the probability distribution (x) = f xP{x)dx
.
(2.52)
gives a stationary value. It need not coincide with the equilibrium state of the system but with no loss of generality we furtheron suppose (x) = 0. In the following we consider time correlation functions and their relations to spectra. These functions contain important informations about the dynamics of the many particle systems. A correlation function measures the dependence of the variable x taken at different times. Conventionally it is given by the average of the product of x(t) and x(t + T) CXlX(t,T)
= (x(t)x(t
+ r))=
f xlx2P(xr,t;x2,t
+ r)dx1dx2
;
r>0
(2.53)
where P(xi,ti; x2,t2)dx1dx2 with t2 < t2 is the joint probability that x(t) is inside [xx,xi + da?:] at time tx and, respectively in [x2,x2 + dx2] at t2. By definition the joint density can be expressed through the transition probability density as P(x2,t2\xi,ti) P(xi,h;x2,h)
= P{x2,t2|a:i,*i)P(a;i,ti)
•
(2-54)
Correlations
49
and spectra
In equilibrium the fluctuations of the variables x form a stationary process which means that P(x, t) is independent of time and the transition density depends on time differences r = ti —1\, only. Therefore the autocorrelation function is independent of the running time J |anP(n) j
x2P{x2,T\x1)dx2J
dxi •
(2.55)
The second integral defines the conditional mean value. Hence one might write compactly CX,X(T)
= J xx (x(r))x{0)=xi
P{Xl)dx!
.
(2.56)
Thus the time correlation function is defined as the average over the stationary probability distribution P(x) which has been established at time t = 0. For a multivariate stationary process Xi(t) (i = 1,.. .n) the correlations form a matrix with elements Ci,j(T) =
%i
P(x!...xn)
(XJ{T))X.(0)=X,
dx!...dxn.
(2.57)
From stationarity it follows immediately: (i) Since Cij(r) does not depend on the actual time we find by shifting t —> t' — r dj(T)
= (xi{0)xj(T))
= {xii-^XjiO))
= CJ:i(-r)
(2.58)
At r = 0 it coincides with the matrix of second moments Cij — (xiXj). The behaviour for large r is determined by the ergodic properties, of the considered stationary process. Under mixing conditions the correlation function vanishes if T goes to infinity. (ii)The Fourier-transform of the time correlation function + oo
% H =
I
Citj(T)e'WTdT
(2.59)
— oo
is called the spectrum of the fluctuating values. It can be calculated either by finding the time correlation function via Fourier transform (2.59) or by the analysis of the dynamics of the fluctuating values. Indeed, if the Fourier-components of Xi(t) dtxj(i)e l w t
(2.60)
are multiplied by Xju and averaged over the stationary distribution -f oo +oo
(xiuxju>)=
f
f dtdt'(xi{t)xj{t'))e[^+^t+iu1'^-^
(2.61)
50
Tools of Stochastic Dynamics
it yields due to the stationarity } = 2nd (u + u') Sij(u)
(2.62)
For real processes X{w = x\_u this relation expresses the fact that only the complex conjugated values are correlated and contribute to the Fourier transform of the correlation function. For each ui it forms a nonnegative definite matrix, (iii) It is easy to show by virtue of (2.62) that
[ \x{t)\2dt = i - f\xM\2du>
(2.63)
(Parseval theorem) where xu is defined in (2.60). (iv) A physical meaning for the Fourier transform of the correlation function was found by Wiener and Khinchin and formulated in a theorem named after them. Let us look on finite sequences of the real stochastic trajectory over the length T with xT(t) = x(t) for \t\ < T/2 and xT(t) = 0 otherwise and Fourier transform T/2
x%=
dtx(t)exp(iwt).
(2.64)
-T/2
As T —> co the original full stochastic process is recovered. Following Parseval's theorem the expression +oo
T/2
T
+oo
2
i J (x {t)) dt = i J x\t)At = JL J |^ |2dw — oo
—T/2
(265)
—oo
allows a spectral decomposition. For real processes xl (xlY = xlxlu
= |^|2,
(2.66)
is symmetrical for positive and negative frequencies and, hence, the integration is restricted to positive frequencies. The left and right hand sides of Parseval's relation stand for the summed energies of the deviations from the mean ((x) = 0), either as integrals over the trajectory or as integrals over the amplitudes of the Fourier components. Then the power contained in the frequency band [u, u + dw] with u > 0 can be assigned as PT(w)dw=
||^|2da;
with PT(UJ) being the spectral power density of the considered time series Averaging over an ensemble of different realizations yields
(2.67) x(t)T.
T/2
(PT("))
= |<|a£| 2 > = |
/ -T/2
{xMxfoVexpiiwih-hVdtidtz.
(2.68)
51
Correlations and spectra
With the stationary correlation function inserted one obtains with T — t\—ti T T
(P (UJ))
= 2 J (l - M )
CxtX(r)
exp(iu,T) AT .
(2.69)
-T
If the integral oo
l
TCX>X{T)6.T
(2.70)
0
converges the long time limit relates the spectral power density of the process x(t) P(u) = lim (PT(u)))
(2.71)
T-foo
to the Fourier transform of the correlation function as follows: oo
P(w) = 2 f C X i X (r)exp(iwr)dr = 25 x>I (w).
(2.72)
— oo
Lateron we will omit the difference between both values and refer the Fourier transform of the correlation function as to the power spectral density or, respectively, the power spectrum. (v) As an example we consider a diffusing particle with a second moment spreading in time according to (x2(t)-x2(0))
= 2Dt.
(2.73)
The l.h.s. can be replaced by the velocity correlation function as (a: 2 (<)-a: 2 (0)) = / f x{s)ds
f x{s')ds'\
.
(2.74)
Therefore it holds t
D = ^ J
t
f Cv,v{s - a') dsds'
(2.75)
o o In the limit t —• oo one finds + 0O
D = 1/2 J Cv,v(T)dT .
(2.76)
— oo
For the relaxation of a heavy particle with mass m in a viscous medium one has (comp.eq. (2.108)) C7„,„(T) =
^exp(-^|r|), m m
(2.77)
52
Tools of Stochastic
Dynamics
where 7 is Stoke's friction coefficient. Insertion into eq. (2.76) yields the well known Einstein formula D = ^
(2.78) 7
which will be met and discussed several times later on.
2.5
Linear response
In the section (2.2) we derived the linear relaxational dynamics for the mean values taking into account the thermodynamic force but neglecting fluctuating forces. Arriving at this conclusion Onsager made a second long ranging Ansatz with reading consequences. In the derivation of dynamics we did not specify whether the initial non-equilibrium state was prepared as a result of an external force like usually by considering the dynamics of mean values or in result of a fluctuation permanently generated in thermodynamic systems. Onsager postulated that there is no difference in the dynamics of the regression of a fluctuating value and the behavior of the mean value perturbed from its marginal equilibrium state by an external force. In consequence this regression hypothesis means that the behavior of the fluctuations are determined by the relaxational coefficients. Also the time correlation function and the spectrums are obtained from the kinetic laws for the mean values. We will see later that a more profound description of the fluctuation will lead us to the Langevin-equations where fluctuating forces have to be taken into the description. It turns out that this is equivalent to Onsagers's regression hypothesis, saying that the dynamics of the conditioned averages that enter the expression (2.57) for the correlation function is given by the relaxational dynamics of the mean values. To prove the regression hypothesis of Onsager we consider the relaxation of the mean value (x(t)) for t > 0 if the thermodynamic system was driven out of equilibrium by an external force F(t) with properties
«*)-{?
HI
e-">
We assume that the force has acted for a sufficiently long time thus that the stationary probability distribution is reached at t = 0 when the force is switched off. For t = 0 we find for the average value of x (x0)Fo = f x0P {x0, F0) dx
(2.80)
with P(x0,F0)~e
S(«n)
«(»n)
"B
»BT
(2.81)
Linear
53
response
where H(XQ) is the Hamilton-function H ~ -XQFQ associated to the force F(t). In the absence of a force at t > 0, the mean value of x will relax to its unconstrained equilibrium value (x)eq = 0. Obviously, the evolution of mean (x(t))x^=Xo at t > 0 is conditioned to its fluctuating initial value XQ which is distributed according to (2.81). Then, the relaxation process averaged over initial values reads
lx(t))p
=
/ d z 0 <»(t)>,(o,=,0 exp ( 5 j a l - i g a l ) 7 >N . / d x o exp ( ^ 1 - ^ 1 )
*>0-
2
-82
For small forces the averaging in (2.82) can be performed by the unperturbed distribution without force. Indeed, expanding in the numerator and denominator as
we find eventually (x(t))Fo = { ^ j
<*(*)>*(0)=xo zo P{x0) dx0 .
(2.84)
The integral on the r.h.s. defines just the time correlation function in (2.53). For small forces it holds {x(t))Fo = -^Cx,x(t).
(2-85)
Thus, in case of small perturbations the time correlation function is proportional to the mean value of the relaxational process. In conclusion we are able to write down the kinetic equation for the time correlation function which are identically to the the linear relaxation dynamics. For the autocorrelation function of x(t) we formulate frCx<x{T)
=
-XCX,X{T)
(2.86)
with the initial conditions C*,*(T = 0 ) = (x2)
=g~1.
(2.87)
Integrating (2.86) yields Cx,x(T)=g-1e-xW
(2.88)
by virtue of (2.58). The generalization to several fluctuating variables is straightforward. The application of Onsager's regression hypothesis leads to the kinetic equations frCiAT)
= -KkCkA-r)
(2-89)
54
Tools of Stochastic
Dynamics
that has to be solved with the initial conditions Ci,j(0) = (xiXj)
= gr] .
(2.90)
This can be done by means of the one-side Fourier-transformation of the time correlation functions oo
S^(w) = J^Citj(T)dT.
(2.91)
0
It gives an algebraic equation for 5^"(w) {-iu}6itk + \i,k) S^(u) = (xiXj)
(2.92)
The expression (2.91) together with the complex conjugated value oo
5
w
[ i( )]
,=
e
/ ""
oo
TC
T
w( )
0
dT
=/
dTe
0
"
iUIC
T
M(- )= /
0
e^dji^dr
(2.93)
-oo
determines the spectrum of the considered process
S i > ) = S+(a,)+[S+( W )]*
(2.94)
where we have used (2.58). Thus we find Sij(w) = (-iuS i i k + Ai,fc)_1 g^j + g~l (-iojSk,j + A f c ) i ) _ 1
(2.95)
or in a more compact form s
i,j(v)
= [(-iw^t.fc + Ai,fc) (iuSjti + \j,i)]~l (LkJ + Lij)
(2.96)
with Lij being the Onsager coefficients. The correlation function Cij(r) follows from the inverse Fourier transform. At the end of this chapter we will come back to Onsager's coefficient Lij that establish the linear relation between thermodynamic fluxes and forces. Empirically it was found that the coefficients are symmetric Lij = ±Lhi
(2.97)
This macroscopic property means that if a force Xj causes the thermodynamic flux Jj in turn the thermodynamic force Xi gives rise to a flux Jj. In this general form this law was first pointed out by Onsager. Special cases had be known long before. The reason for the symmetry of the transport coefficients lies in the stationarity of the process and the reversibility of the underlying microscopic Hamiltonian motion. As a consequence the motions forward and backward in time are related by Xi(t) = eiXi(-t)
(2.98)
Linear
55
response
where e* is the parity under time reversal which is + 1 for even variables and — 1 for odd ones. For the correlation function this yields C
i,j(T)
=
(Xi{t)Xj(t
+ T)) = CiCj (Xi(-t)Xj(-t
= dejCiji-r)
- T))
(2.99)
and with the stationarity (2.58) eventually Citj{T) = eitjC^T).
(2.100)
Further we write down the dynamical equations for both correlation functions ^Qj
= -Xi,kCk,j(T)
^Cjti
= -Xj,kCkli(T)
(2.101) .
(2.102)
Hence we obtain 5 > i i f c C M ( r ) = eiej^^kCkAr) k
(2.103)
k
For T = 0 the correlation function yield the standard deviation with Cij(0) — g^} and with the definition of the relaxational coefficients (2.51) one finds the OnsagerCasimir relation Lij = ei€jLj:i 2.5.1
Colored
(2.104)
noise
As an example we consider the motion of a heavy particle with mass m in a viscous liquid. For the mean velocity conditioned to its initial value v(t) we have md(v{t
+ r))
=
dr and hence for the stationary correlation function m— (v(t + T)v(t)) = - 7 (v(t + T)v(t)) dr with the initial condition s2\ kBT (v(tf) = " ^
(2.106)
(2-107)
It gives the decaying correlation function knT v(t + T)v{t)) = —exp(--!-\T\)
(2.108)
56
Tools of Stochastic
Dynamics
with the spectrum „ , , kBT 27/m SVtV(w) = -2'' (2.109) m (7/m) 2 + uj1 The considered stochastic process v(t) is called the Ornstein-Uhlenbeck process (more detailed information about colored noise, its relation with other noise and its effect on nonlinear dynamics can be found in (van den Broeck, 1983; Hanggi, Jung, 1995). Since not all frequencies are excited with equal strength the velocity v(t) is referred to as a colored process. In particular, for the present prevailing low frequency excitations one finds it also is termed red noise. 2.5.2
Harmonic
noise
A richer spectrum for a colored process is obtained if a harmonic force additionally acts on the particle. The system of equations for the velocity autocorrelation and the velocity position correlation function reads CX,V(T)
=
CV,V{T)
(2.110)
CvA-r)
=
-^CV,V{T)-U%CX,V{T)
(2.111)
Following the approach (2.89-2.95) one simply finds the spectrum *,,(«) = ^
m
W™
(2.112)
(7/m)» + («-£)
which now possesses a peak at a finite frequency. 2.5.3
Fluctuation
dissipation
theorem
We come back to the reaction of the system on a small external force that has been adiabatically switched on at t —¥ — 00. This problem usually is described within the linear response theory by introducing the response function \(t) of a many particle system. It describes the response of the mean value of x(t) on the external force F(t) related by t
<*(*))= J
x(t-t')F(t')dt'.
(2.113)
—00
The response function coincides with the Green's-function up to a multiplicative constants. Its Fourier transform Xu> is known as the generalized susceptibility. Because of causality, the response function vanishes for negative times, x(T) = 0 if r <0. It is obvious that (2.113) holds only in for sufficiently small forces. As a consequence of the linear approximation, in the frequency regime there is only a response
Linear
57
response
at the frequency of the force. The Fourier transform convolute simply (x^) with Fu as W = X M F »
(2-114)
On the other hand the mean (x(t)) and the correlation function are strongly related by (2.85). Therefore, it comes out that the response function is related to C(r) as well. Indeed combining (2.84) and (2.113) and using the expression for the force F(t) we get (now already independent of i*o as result of the linear approximation) oo
CX,X{T)
= kBT
fX(s)ds
with boundary condition J x(s)ds — {x2) /k^T. o gives ^
(2.115)
Differentiation with respect to time
= -ferX(r).
(2.116)
The expression (2.116) connects two fundamental physical processes and is know as the fluctuation-dissipation theorem. On the l.h.s. the correlation function describes the regression of fluctuations which is related to the response function upon external forces of a many particle system. The origin of this connection is in the common source of fluctuations and dissipation that is caused by the microscopic motion of the particles. In the frequency domain it is sufficient to know the imaginary part of the susceptibility x(w) in order to determine the power spectrum: + oo
— oo
Therefore, the spectrum of the considered value x(t) is determined by the imaginary part of the Xw S*,xM = ^ 3 x .
(2.118)
UJ
It is this relation which gives the connection between fluctuations and dissipation in the frequency domain. To show that 5%(w) is directly related to dissipation we consider the energy which is dissipated during one period if a system is driven by a periodic force (Landau, Lifshitz, 1971) f(t) = R ( / o e ^ ' ) = \ {F0e-^
+ / 0 *e^)
(2.119)
58
Tools of Stochastic
Dynamics
The energy change dE dt
df *> dt
(2 12
' °)
with (x) = \ (xHfoe-^
+ x(-W)/0*e^)
(2.121)
during one period T stands for the dissipation rate Q of the mechanical work in the many particle system. It is determined by the imaginary part of the susceptibility only T d
Q = ff dJdt= o
|l/o|2SxH
(2.122)
It makes clear the meaning of eq. (2.118). Just the part responsible for the dissipation in many particle system determines the spectrum of the fluctuations. As an example we again consider harmonic noise. The susceptibility is easily found
x« = - - ! — r i r - — T
( 2 - 123 )
The spectrum becomes after insertion into (2.118)
5...H = ^ m
= «~2svA")
J™(,-y2 2
2
{LOQ - u ) + w ( 7 / m )
(2-124)
2
with SViV(w) as defined in (2.112). 2.5.4
Nyquist
theorem.
White
noise
Another question was posed in the work of Nyquist who considered current fluctuations in electronic circuits. He asked about the spectrum of the pondermotoric forces which originate the permanent fluctuations from the rest states. We consider an one dimensional conductor of length / and area A in which charged move under the influence of friction described by Stoke's law. The conductivity of this conductor can be determines in linear response by applying a constant external electric field E. For the dissipative motion of the averaged velocities of N carriers we assume m(vi) = -'y{vi) + eE The stationary current density follows as
(t = 1 . . . N).
(2.125)
Linear
59
response
Therefore, the conductivity becomes ne2
(2.127)
7 and the resistance of the considered conductor is given by
The conductor is embedded in a bath with temperature T. This leads to Gaussian distributed fluctuations (ace. to sect.(2.3)) of the velocities of the independent carriers. Its velocity spectrum is according to (2.109) given by Svi.vjiu) = dij
—2 •
(2.129
The permanent motion of the particles will also induce a permanently fluctuating current N
/(*) = f 5 > .
(2-13°)
Its spectrum is found from (2.129) SlM
= N^.?{™ (2.131) 2. z 2 m l {~f/my + w Coming back to Nyquist's problem we ask what the origin of the permanent motion of the carriers is that result in the current fluctuations. The linear dynamics (2.43,2.50) can not answer this question, because it describes the relaxation to the equilibrium states of vanishing mean values. Earlier Langevin had answered a related question for Brownian motion which will be considered in the next section. In order to obtain fluctuations of the current noisy electric fields have to be incorporated into the description of the motion of the carriers. Thus we will add to the equation of the fluctuating velocities time dependent electromotoric forces eEi(t) mvi = -1Vi + eEi(t) (2.132) where Ei denotes the electric field at the position of the i-th charge. On average the electric fields vanish; Otherwise we would have non-vanishing mean velocities. For the current it results in an additional time dependent voltage over the resistance which induces current fluctuations
*(*) = ^ X > w
(2-133)
60
Tools of Stochastic
Dynamics
in KirchhofFs law LI = -RI + t(t).
(2.134)
where L denotes the inductance l2m Ne2 '
Rm 7
(2.135)
and the After Fourier transform of eq. (2.134) we find j
SkJ
(2.136)
where Z(w) is the complex impedance of the conductor. It is the inverse of the susceptibility x( w ) = 1/Z{U)- The correlation of the Fourier transformed currents becomes ^
= (iuL + RHML + R)
{
^ '
}
•
(2 137)
'
Using eq. (2.62 one can express the current spectrum in terms of the voltage spectrum SJM
= lxM|2S«M = -tlTjTjp SteM)
(2-138)
where S^^(w) is the spectrum of the spontaneous voltage changes that drive the current fluctuations. Comparison with (2.131) leads to Nyquist's result Su(u)
= 2RkBT.
(2.139)
The spectrum of the spontaneously occurring voltage drops in the resistance does not depend on the frequency. It is called white noise. Because 7(f) is Gaussian £(f) is Gaussian as well. Obviously we have <£(f)> = 0.
(2.140)
It is interesting to have a look at the correlation function of the voltage drops. Backward Fourier transformation results in mat'))
= 2e8(t-t>)
(2.141)
with e = RkBT
(2.142)
where S(t) is Dirac's-delta function. Hence white noise is uncorrelated in time. The factor e denotes the noise intensity. Its particular form as the product of thermal energy U-QT and the resistance R, that characterizes the dissipation, is in accordance
Linear
response
61
with the Einstein relation. We note that specifying the first and second moment (2.140,2.141) is sufficient to characterize the Gaussian process £(£). The fact that £(£) is uncorrelated at different times, and that its variance is infinite renders this process highly irregular jumping between plus and minus infinity. As a result it is nowhere continuous in time. Obviously, in physics white noise is an idealization. It is a good approximation for processes whose correlation time is much smaller than all other characteristic time scales. One also observes that white noise forces acting in the linear dynamics (2.134) exhibit a fluctuating current with a red spectrum (2.138). The resistance acts as low pass filter for the applied white noise excitations. The current fluctuations are exponentially correlated (I(t)I(t + r)} = ^ e x p ( - | | r | )
(2.143)
with correlation time re — L/R. Analogously, a RCL circuit would generate fluctuations of the current with a spectrum given by eq. (2.112) and LUQ is the eigenfrequency of the circuit 1/LC. In case of a strong resistance the correlation time in (2.143) becomes vanishingly small. The exponential function converges to a (5-function and the current becomes white noise with intensity 2/CBT'JR. 2.5.5
White noise and the Wiener
process
Though white noise is an idealization in physics it is widely used in the modern literature for modeling dynamical behavior under the influence of noise. We will list several properties of these systems in the next sections. The reason why it is often employed lies in the simpler mathematical treatment of processes originated by white noise. The important property of a process driven by white noise is that the change of the system within a small time step only depends on the actual state of the process but not on its history. Hence, the response on the white noise source is without memory which is not the case for correlated noise as will be seen in the section of the generalized Langevin equation. A dynamical variable of the system driven by white noise (as in the considered example, the current) is still a Markovian processes. In a differential shape I(t + dt) is given in probability as a result of the value of the current at time t and the random forces within the interval [t, t + dt, independently from the former history. In the mathematical literature white noise is defined in terms of its increment over a finite time step dt after time t. Assigning the increment t+dt
dW(t) = dWdt = J Z(s)ds t
(2.144)
62
Tools of Stochastic
Dynamics
the process is stationary and Gaussian with zero mean (dWdt) = 0.
(2.145)
Further, looking at the variance of the increments one finds <(dW dt ) 2 ) = 2edi
(2.146)
which is Fick's second law. For nonoverlapping time intervals di and dt' the increments are uncorrelated. The sum over the increments at subsequent time intervals is know as the Wiener process N
Wt = Jim zy2dWdti AT—>oo —* x=l
(2.147)
with dti = ti — ti-\ and tff = t and to = 0. It is continuous in time, nowhere differentiable and Markovian because its increments are independent one from each other.
2.6 2.6.1
Brownian Motion Einstein's
relation
Stochastic methods became intrinsic part of the physical language in the context of the theoretical explanation of Brownian motion. This is the erratic motion of small particles immersed in a fluid. It is named of the botanist Robert Brown in memory of its discovery who observed highly random motion of pollen grains suspended in fluids in 1827. Montroll and West (1979) reported that the court physician to Maria Theresa whose name was Jan Ingenhousz observed erratic motion of finally powdered charcoal floating on alcohol surfaces earlier in 1785 which became not known to the non-Dutch speaking world. Brown's first intuition was that he had found the "Secret of Life". Later he repeated the experiments with mineral particles and still found the permanent motion and his first hypothesis failed. His observation became a serious physical problem. First of all it was the first experimental proof showing deviations from the continuum theory of matter. The motion of the particles is the result of the molecular agitation resulting from their interaction with a finite number of the surrounding liquid molecules. In consequence Brownian motion supported the ideas of the molecular structure of matter yielding the possibility of the determining Boltzmann's constant and Avargado's number, see eq. (2.31). The second deep conclusion concerns the diffusion coefficient D. It is responsible for two different physical processes. On the one hand it describes the relaxation of
Brownian
Motion
63
an inhomogeneous distribution of a density inducing a matter flow by Fick's law U = -D°£
(2.148)
The same quantity describes the random behavior of a single particle. The mean square displacement of a Brownian particle increases linear in time with a proportionality factor given by the diffusion coefficient (x(t) - x(0))2)
= 2Dt .
(2.149)
Therefore the continuity equation
£-"£
<2-™>
has an ambiguous character. On the one hand it models the homogenization of a macroscopic density n(x, t) and on the other hand it describes the evolution of probability density of single particles. In statistical mechanics the motion of the particle is expressed by mechanical laws. Connecting the deterministic diffusion with the fluctuating Brownian motion gives the possibility to determine the transport coefficient in terms of the mechanical properties of particles immersed in a fluid at a certain temperature. We already have underlined this connection (eqs.(2.46),(2.78) and (2.139)). Here, we look at different fluxes which compensate each other in equilibrium. The distribution of gases in the gravitational field of the earth is due to the barometric formula
te=-l£«x)
(2 151)
-
The establishment of a stationary distribution is the compromise of two tendencies. Due to their weight the particles are attracted by gravity i.e. they fall to the earth-surface whereas the diffusion counteracts to homogenize the distribution. In equilibrium the diffusional flux (2.148) and the gravity flux fr
= -n(x)^-
(2.152)
compensate each other 3d+j9r=0.
(2.153)
Together with the Einstein relation (2.78) D=^-
(2.154) 7
the barometric formula results.
Tools of Stochastic
64
2.6.2
Brownian
motion
as Markovian
Dynamics
dynamics
At the beginning of the theoretical treatment of Brownian motion stood the important insight that the erratic motion of the particles only could be interpreted within the frame of a probabilistic theory. It was Einstein and v. Smoluchowski (1906) and, later, Langevin who pursued this idea. Einstein was unfamiliar with the experimental facts. He predicted in his work the agitation of suspended particles as a clever approach to determine experimentally Avargado's number in which Perrin succeeded experimentally. Einstein saw the reason for the ongoing irregular motion in the molecular nature of the liquid surrounding the Brownian particles. Evidently from the molecular chaos follows an out-off-balance of the impacts of the liquid molecules upon the Brownian particles which results in the random motion. Einstein supposed that the total momentum that the particle obtains from the liquid is statistically independent at any instant of time. It means that the position of the Brownian particle after a short time step only depends on the actual positions of the particle. With this assumption the motion of the particle becomes a Markovian process for which Einstein was able to derive a kinetic law for the time evolution of the transition probability. If we label the displacement of a Brownian particle during the time r by A we find for the probability density P(x, t + r) of a particle to be in the position x at time t + T
P{x,t + r)=
f
P(x + A,t)$(A,T)dA
(2.155)
— oo
where <&(A,T) denotes the probability of performing a displacement of length A during r. It is symmetric $(A,r) = *(-A.T)
(2.156)
and normalized +oo
/"$(A,T)dA = l.
(2.157)
— oo
The expansion for small r and A dP P(x,t + r) P(x + A,t)
= =
P{x,t)+r-^ P(x,t)
r) P 1 + A— + 5A
(2.158) 2
2
B P —
(2.159)
Brownian
65
Motion
yields in the limit r —> 0 d2P
dP
+ 00
2 ,. 1 r A lim - / — $ ( A . r ) d A . T->OO T
J
(2.160)
2
Here the first derivative with respect to the position x does not contribute because of the symmetry of $(A, r ) , see eq. (2.156). If the limit D=
1 t A2 lim - / — - $ ( A , r ) d A T—>oo T J
(2.161)
2
exists, the kinetic law for the time evolution of the density of a Brownian particle becomes dP_ _
d2P
dt ~
dx2
(2.162)
This law can be written in the form of a continuity equation for the total probability 8P d_ •j(x,t) dt dx' For the current one recovers Fick's law
= 0.
(2.163)
.dp (2.164) j = -D dx If the total probability initially at t — to is concentrated at x = XQ the solution of eq. (2.162) becomes P(x,t\x0,t0)
(X -
y/4nD (t -10)
exp
XQ)
~4D{t-t0)
(2.165)
For the spreading of particle's position this implies {x - x0)2)
= 2D (t - t0)
(2.166)
This coincides with Fick's second law. We note that in this derivation of Fick's second law the probability density of a single Brownian particle is assumed to be proportional to the mass density of a swarm of particles. Thus Einstein and v. Smoluchowski outgoing from probabilistic point of view in the interpretation of Brownian motion found the main macroscopic results. The main result of their approach consists in the derivation of a kinetic equation for the probability density. This equation was later generalized by v. Smoluchowski, Fokker and Planck, Kolmogorov, Feller and others. The diffusion coefficient in Einstein's theory remains undetermined. It is expressed as the second moment of the jump width during unit time. Hence further information about the distribution $ ( A , r ) is required.
66
Tools of Stochastic
Dynamics
We mention that the Wiener process as defined previously has identical properties as the position of the Brownian particles. The increment W{t) — W(to) between times t and to is distributed accordingly to eq. (2.165). 2.6.3
Langevin's
approach
The second general approach is due to Paul Langevin and was developed three years after Einstein's paper. Langevin added a random force in Newton's second law to compensate for the energy loss resulting from Stoke's friction law (comp. sect.(2.5.4)) mv =-iv
+ £(t).
(2.167)
The reason of its occurrence can be seen in the unbalanced impacts of the liquid molecules, in the thermal agitation. Langevin stressed that his approach was "infinitely simpler" than Einstein's one. In mathematical terms equation (2.167) constitutes a stochastic differential equation. The Brownian particle's velocity becomes a stochastic process. On average we expect to find the deterministic damped motion. This requires obviously <£(<)> = 0.
(2.168)
We now multiply eq. (2.167) by x(t). Simple algebra gives + x2 + — x(t)£(t) m
^-(xx) = ~^xx at m Averaging over an ensemble we find fL LLL
{xx)
=
- l
{xx)
+
(i2)
+ ±.{x(t)£(t))
lit
(2.169)
•
(2-170)
lib
The second moment of the velocity can be determined from the kinetic energy in thermal equilibrium, yielding for the second term on the right hand side of eq. (2.170)
(2.171)
A further assumption concerns the stochastic force. Langevin postulated that the stochastic force changes much faster than the position of the Brownian particle does. The time scales of the random part r c and the friction TbTake — m/l should be separated (r c -C Tbrake) that there can not be an effect of the stochastic force £(£) on the position of the particle x{t) at the same time. It makes the £(£) equivalently to Einstein's independence of the jumps at different times. Therefore, the last term vanishes (x(t)at))
= {*(t)) (£(t)) = 0 •
(2-172)
Brownian
67
Motion
This means for the autocorrelation function of the random forces that it possesses a very short correlation time. For the sake of definiteness we assume an exponential decay of the autocorrelations: <£(* + r)£(t)> = - e x p ( - - y
(2.173)
The simplifying assumption here is that r c is much smaller than any other time scale of the Brownian particles so that the limit r c —> 0 yielding a <5-function for the correlations of the random force provides a good approximation
If the particle at t = 0 is located at the origin the integration of (2.170) results in (Ax*(t))
= 27
i
- " (
1
-
e X P
( - ^ ) )
For times much smaller that the brake time t C Tbrake = mj ballistically like a free particle
(x2) = ^ t * = (vl)t*.
(2.175) * the particle moves
(2.176)
For large-times t ^> Tbrake the variance of the position spreads linear in time ( i ! ) = 2
^ r
t
_ 2 ^ n
( J m )
The additive constant is the squared length of the brake path of a Brownian particle
krake = J ^ - -
= ^/W)~ •
(2-178)
V m 7 7 Diffusion with linear growth in time of the mean square displacement takes place at length scales larger the brake path where the additive constant can be neglected. 2.6.4
The overdamped
limit
For strong damping and times larger than the braking time Tbrake the description can be considerably simplified. The velocities v(t) at subsequent time steps that are sufficiently larger than the brake time are independent of each other. Therefore the velocity itself becomes a white process. This can also be readily seen from the equation of motion (2.167) in which the acceleration can be neglected. For large 7 it acts effectively on short time scales <x Tbrake- This results in the simple equation v(t) = x{t) = -£(£) 7
(2.179)
68
Tools of Stochastic
Dynamics
where £(t) should be due to (2.168) and (2.174). Equation (2.179) is readily integrated t
x(t)=x0+-
I £(t)dt'. o
(2.180)
For the mean square displacement one obtains:
(x(t)-Xof) = ±/j at')dt'j t(t")dA t
^f
T
t
f(Ht')Z(t"))dt'dt".
(2.181)
o o
Using the properties of the random force one finds: (x(t) - x0)2) = 2-^ t .
(2.182)
This result confirms the statistical assumption made about the stochastic force by (2.168) and (2.174) and we find convergence with (2.177) at the considered coarse grained length scale. We also obtain that the intensity of the noise e which was introduced in (2.174) relates as c = £> 7 2 = kBTj.
(2.183)
In summary, Langevin was able to find a dynamical approach basing on a mechanical description for the properties of the Brownian particle. The additionally acting random forces have statistical properties as a white noise process. Obviously the specified first and second moments do not completely characterize the random force. Only for Gaussian statistics of the random force the Maxwell distribution for the velocity is recovered. 2.6.5
Generalized
Langevin
equations
In this section we discuss a Hamiltonian model of a single degree of freedom, referred to as the particle, interacting with other degrees of freedom representing the environment of the particle. Initially, the environment is prepared in a thermal equilibrium state at temperature T. Because we will consider the thermodynamic limit for the environment it can be considered as a heat bath at the initial temperature T. We will show how for the particle a random and dissipative process results that can be characterized by a (generalized) Langevin equation (Zwanzig, 1973; Reimann, 2001).
Brownian
69
Motion
The particle with coordinate Q and momentum P moves in an external potential Vext{Q)- The bath is assumed to consist of linear oscillators that are be linearly coupled to the particle: H(Q,P,qn:Pn)
= HS{Q,P)
+
(2.184)
Hs,B(qn,Pn,Q),
where Hs(Q,P)
=
^
+ Vext(Q)
Hs,B(
=
E
^
+ ^ t e " - ^
2
)
'
(2J85)
In the absence of the external potential, the system is translation invariant, i.e. H(Q, P, qn,pn) = H(Q + a, P, qn + a,pn). The dynamics results from the canonical equations of motion Q
=
P
=
Qn
=
pn
=
(2.186)
dVKS
Q^
dQ
mnu)l + ^
mnu?nqn
Pn
(2.187) (2.188)
-mn(jOnqn
+
(2.189)
mnujnQ
Because the equations for the bath oscillator are linear, their solution can be expressed in terms of the trajectory of the particle: t
qn(t) = q„(0) cosw n i + ^ - i - sinw„i + wn /
sinw n (t - t') Q {t') dt'
(2.190)
where qn(0) and p n (0) denote the initial states at t = 0. Inserting these expressions into the dynamics of the particle we obtain MQ(t)
SVr...
J2 mnw2n)Q(t)
dQ
V. n
/
I
+ J dt'Y^
+ J2
m
nul
™ ^ sin ojn (t - t') Q {t') qn(0) cos (w„(i)) + — sin (w„f) Qit'). mnwn
(2.191)
70
Tools of Stochastic
Dynamics
A partial integration of the third term on the right hand side yields t
MQ
=
_ ^ i _ y dt'Qtf)J2mnujZcoBwn(t-t') n o ^2
m
nU2n\—
(2.192)
sinw„i+ [qn(0) - Q(0)]coscjnt\
.
There are two types of forces that the bath exerts on the particle: One force explicitly depends on the initial conditions of the bath whereas the other one describes the retarded back reaction of the particles motion mediated by the bath. If one were able to control the initial conditions of the individual bath oscillators the dynamics still would be deterministic and reversible. However, in practice it is only to prepare the initial conditions of the bath according to some statistical law. This introduces a randomness in the initial conditions and, via the third term of the right hand side of eq. (2.192), a random force: £(*) = ] £ mni02n\^^-smujnt+[qn(0)-Q(0)}cosunt\
.
(2.193)
Because it is practically impossible to precisely convert the initial bath conditions to their time reversed values, the reversibility of the motion also is lost. As already indicated, we assume the bath to be initially prepared in a thermal equilibrium state constrained to the initial position of the particle Hence the probability of the bath initial momenta and positions are given by: p(qn,Pn\Q)
= Z-i exp { -
H S A
1 ^
Q )
)
(2-194)
where T denotes the temperature and Z the partition function Z = / d J v g„d J > n exp I
fcBy
J •
(2.195)
This is a Gaussian probability distribution. It completely determines the statistical properties of the random force £(£). As a linear combination of Gaussian random variables it is a Gaussian process which is completely determined by its mean value and its autocorrelation function: <£(*)> ma*'))
=
0
(2.196)
=
^7(*-f)
(2-197)
where 1(t-t')
= kBTj2
mnuj2ncoscon{t-t')
.
(2.198)
The Fokker-Planck
71
equation
This function coincides with the kernel of the second term on the right hand side of eq. (2.192) and generally is referred to as the memory friction. Eq. (2.192) may now be written in the form of a generalized Langevin equation t
MQ(t) + J dt> j(t - t') Q («') + ^
=m
(2-199)
o with a Gaussian random force specified by eqs. (2.196). The fact that the correlation function of the random force is closely related to the memory friction is know as the fluctuation dissipation theorem. In the particular case of a 5-correlated random force (Z(m(t')) = 2kBT16(t-t')
(2.200)
we both recover the Langevin equation for a Markovian process of coordinate and velocity of the particle (2.167) and the Einstein relation (2.78). We note, that for countably many oscillators the memory friction is a quasiperiodic function of the time difference. In order to approach an instantaneous friction, the bath must consist of a continuum of oscillators. In the case that £(t) is colored noise the generalized Langevin equation possesses a friction term with memory correspondingly to the friction memory f(t — t'). Dependency on the model under consideration the exponentially correlated OrnsteinUhlenbeck process with memory friction (2.108) or harmonic noise with spectrum (2.124) are used to account for a temporal structure of the bath.
2.7 2.7.1
The Fokker-Planck equation Kolmogorov's
forward
and backward
equations
An alternative approach to describe a Markovian process is based on the fact, that the change of the process in time only depends on the present state but not on the history of the process. As a consequence for a Markovian process the change of the process over a finite time can be built up from changes in infinitesimal time steps. In particular, from the knowledge of the transition probability density P(x, t\xo, to) for infinitesimally closed times t > to, the transition probability density at finitely separated times t > to can be constructed. With a staring probability P(xo,to) all multitime probabilities then can be constructed. The time evolution of the transition probability density can be considered in two ways. One can either fix the condition XQ and propagate the time t in the forward direction, or, alternatively, one fixes x at t and propagates from there to backward in time to the past. Both equations of motion, the forward and the backward equation are based on the Chapman-Kolmogorov equation which represents a necessary condition for a process to be Markovian. It states that the transitions probability
72
Tools of Stochastic
Dynamics
density to reach x at time t from xo at time to can be split in two steps from XQ,
(2.201)
Fixing now XQ and to we calculate the change of transition probability density between t and t + dt, dt > 0 P{x,t+dt\x0,to)-P(x,t\x0,to)
=
dy[P{x,t
+ dt\y,t)
- 5(x -
y)}P(y,t\x0,t0).
(2.202) One can show under very general conditions that the right hand side is proportional to dt for small time steps. Hence, we obtain the forward equation —P{x,t\x0,t0) P{x,to\x0,to)
= =
Lx(t)P{x,t\x0,t0) S(x-x0)
(2.203)
where L(i) denotes the forward operator L(t) p(x) = Jim ^ (J dyP(x, t + dt\y, t) p(y) - p(x)J
(2.204)
We have affixed the index x to the forward operator in the eq. (2.203) to indicate that L here acts on the forward variable. Before we further evaluate possible forms of the forward operator we proceed in an analogous way and derive the backward equation. Considering now the increment P(x,t\x0,to) — P(x,t\xo,to — dto) we find -—P(x,t\x0,t0) P(x,t0\x0,to)
=
L+O(f0)-P(x,<|a:o,io)
=
5(x-x0)
(2.205)
where L + (t) is the backward operator: L+(t)f(x) = jirn i
(JdyP(y,t\x,t-dt)f(y)
- f(xfj .
(2.206)
Note that in the integral kernel of the forward and backward equation the arguments x and y are interchanged. Hence, the backward equation is the adjoint operator of the forward operator relative to the scalar product
(f, p) = J dx f(x) p(x)
(2.207)
i.e.
(f,L(t)p)
= (L+(t)f,p).
(2.208)
The Fokker-Planck
73
equation
A general property of the forward and backward equation is the conservation of total probability. In terms of the forward operator one thus has /•
dxL{t)p{x)
= 0
(2.209)
for all integrable functions p(x), and in terms of the backward operator L+(<)1 = 0.
(2.210)
Both equations follow from the definitions of the forward and the backward operators (2.204) and (2.206), respectively. Now we discuss the possible explicit forms of the forward and the backward operator. First we consider pure jump processes which sit for a random time in some state and then hop over some finite distance to another state. This process by definition has no continuous component. We denote the probability for a jump from x to y taking place in the interval [t, t + dt] by r(x, y, t) dt = P{x, t + dt\y, t). In the case of a pure jump process the jump rates r(x,y,t) the forward and the backward operator: L{t)p(x)
=
L+(t)f(x)
=
(2.211) completely determine
fdyr{x,y,t)p{y)-fdyr{y,x,t)p{x) j dyr{y,x,t)f{y)
- J dyr(y,x,t)f(x).
(2.212) (2.213)
The first terms in the equations for L and L + result from the first terms the eqs. (2.204,2.206) for x ^ y. The second, negative terms take care of the conservation of the total probability as given by the eqs.(2.209,2.210). The form of the forward equation for a pure jump process is analogous to a Master equation of a Markovian process with discrete phase space: the time rate of change of the probability density p(x, t) at time t consists of an increase resulting from jumps into the state x, J dyr(y,x,t)p(x,t), and a decrease because of jumps out of the state x, fdyr(y,x,t)p(x,t). In order to observe the decay law of the state x one has to prevent the process to again visit the state x after it has left it. I. e., the backward equation simplifies to -
-K-P(X,
t\x0, t0) =
-K(X0,
t0)P{x, t\x0, t0)
(2.214)
OTQ
where K(x0,t0) = / dyr(y,x0,t0)
(2.215)
74
Tools of Stochastic
Dynamics
denotes the total rate out of the state xo. The integral of P(x,t\xo,to) over an infinitesimally small region GE{X0) containing the point XQ gives the probability Px0 (t, to) = / dxP(x, t\xo, to) that the process has not left the state XQ up to time t once it has been there at time to- From the backward equation one finds an equation of motion for pXo(t, to) -gj-Px0(t,to)
= K(x0,t0)pX0(t,t0)
(2.216)
with the final condition pX0{t,t)
= l.
(2.217)
This equation is readily integrated to yield Px0(t,t0)
— exp
- / dsK(x0,s)
.
(2.218)
In the case of a time homogeneous process the transition rates r(x,y,t) = r(x,y) are independent of time and an exponential law for the lifetime of a state of a Markovian jump process results: Pxo(Mo) = exp(-«(a:o)(i-
(2.219)
where K(X0) = Jdyr(y,x0). We note, that the process once it has reached the state x will jump to another state y is given by 9(^>*) =
!
^ i r -
(2-220)
K(X,t)
The other extreme case of a Markovian process is one that moves continuously. For such a process the probability for a change of the process x in dt vanishes faster than At and consequently the jump rates r(x, y, t) for x ^ y are zero. In other words, the transition probability density P(x + u, t\x, t — dt) is a very narrow function in u for small dt and therefore we can represent the backward operator in the following way:
l+(t)f(x)
=
limj-
dttoo at
/ duP(x + u, t\x,t - dt) f(x + u) -
- £^(^)|^/(*)
f(x)
( 2 - 221 )
71=1
where we have expanded the function f(x + u)about i in a Taylor series. The coefficients Kn(x, t) are the nth conditioned moments per time of the distance that
The Fokker-Planck
equation
75
the process covers in the infinitesimal time dt: Kn(x,t)
=
lim — / duunP(x K + u.tlx.t dt-M) At J ' lim <**">'(«>=' dt-+0
- dt); (2.222)
dt
where the second line is a shorthand. Because of the backward operator is the adjoint of the forward operator one immediately obtains for L
L(t)p(x) = J2 -J-
Q^ Kn{*, t) p(x)
(2.223)
n—l
Eqs. (2.221) and (2.223) are known as the Kramers-Moyal expansion of the backward and the forward operator, respectively and the conditional moments Kn (x, t) as the Kramers-Moyal moments. For a continuous process one may expect that only a finite number of the Kramers Moyal moments are different from zero because the existence of a high finite moment would indicate a large spreading of the process within a short time. Indeed, a theorem by Pawula states that is a Kramers Moyal moment of even order larger than 3 vanishes all Kramers Moyal moments except the first and the second ones vanish as well. Only those Markovian processes have continuous trajectories for which the forward and the backward operators are differential operators of at most second order. They are known as diffusion processes and the corresponding forward equation is a Fokker-Planck equation characterized by the Fokker-Planck operator
L{t) =
" Yx Kl(x' t) + H^K2{x't]-
(2 224)
'
By definition K^x^t) is a nonnegative function and is know as the diffusion coefficient, whereas Ki(x,t) is the drift. Before we discuss the connection of the Fokker-Planck equation and Markovian processes described by a Langevin equation we note, that a general forward operators of a Markovian process consists of a sum of a Fokker-Planck operator describing the continuous part of the motion and an integral operator as given in eq. (2.212) describing the contribution of the jumps. Finally we note that the generalization to processes in more than one dimension is straightforward: the integrations in forward and backward equations of jumpstype extend over the whole available n-dimensional state space. In diffusion equations the spatial derivatives became partial derivatives with respect to the different coordinates in phase space with the drift vector K(x,t) = {Ki{x\,... ,xn,t)} and the diffusion matrix D(x,t) = {Dij{x\,... ,xn,t)}. For example, the Fokker-Planck
76
Tools of Stochastic
Dynamics
operator in n dimensions becomes n
o
-
n
t=i
i,j—l
a2
Di)j(x1,...,xn).
dxidx.
(2.225)
the diffusion matrix is by definition nonnegative definite. The operator acts on the forward variables of the transition probability density of the n-dimensional stochastic process. 2.7.2
Moments
of the transition
probabilities
In this section we derive the Kramers-Moyal moments for a Langevin equation with Gaussian white noise as stochastic source term. In particular we show that all higher than the second order Kramers-Moyal moments vanish for this particular class of processes. Our starting point is the stochastic differential equation x = f{x,t)
+g(x,t)Z(t)
(2.226)
for a single variable x(t) and £(£) is Gaussian white noise with intensity e and the increments dW(t) during dt (see eqs.(2.144-2.146)). For a small time interval dt the increment of the process is given by t+dt
dx = x(t + dt) - x(t) =
/
t+dt
f(x(s),s)ds+
g{x(s),s)dW(t)
t
(2.227)
t
and the integral has to be understood as described after eq. (2.147). One only needs to know da; for short times di in linear order mdt because higher terms 0(dt2) do not contribute to the Kramers-Moyal moments. Because of the discontinuous behaviour of £(t) and the resulting non-differentiable increments of the Wiener process the second integral depends on the particular position of the points s in the interval [t, t + dt] even if dt approaches zero. To show this rather strange behaviour we write s in terms of t and di and a parameter q G [0,1]: s = t + qdt = q(t + dt) + (1
-q)t.
(2.228)
and express x(s) = x(t + qdt) in terms of x(t) and the increment dx: x(s) = x(t) +
qdx(t).
(2.229)
Now we are ready to evaluate the integrals up to order dt dx(t)
df(x(t),t)
f(x(t),t) g(x(t),t)
dx 1
dg(x{t),t) dx
dx(t) dx(t)
dt + dW(t).
(2.230)
The bistable
77
oscillator
We reinsert dx(t) on the right hand side and obtain in order dt: dx{t) =
f(x(t),t)
+
dg{x(t),t) dx
2eqg{x(t),t)
}dt + g{x{t),t)
dW(t).
(2.231)
The additional term of order dt that is proportional to q results from the squared increment of the Wiener process: (dW{t))2
= 2edt + o{dt).
(2.232)
In mathematics the choice q — 0 is common. It is know as the Ito-interpretation of a stochastic differential equation, whereas physicists often use the Stratonovich interpretation corresponding to q = 1/2 (Stratonovich, 1990). The conditioned moments of dx(t) follow as: ((dx(t))2) n
((dx{t)) )
=
f(x(t),t)
+
=
2eg2{x(t),t)
=
o{dt),
2eqg(x{t),t)
dg(x(t),t) dx
dt + o(dt) (2.233)
dt + o(dt)
(2.234)
n > 3.
(2.235)
This gives for the first Kramers-Moyal moment:
K^t)
= M
=
f{x,t)
+
2Dq^^g(x,t)
(2.236)
still depending on q. For the second moment one finds K2{x,t)
= ^
=
2Dg\x,t)
(2.237)
and the higher moments starting from n > 3 vanish, i.e. Kn(x,t) — 0. Hence to a stochastic differential equation with a prescribed interpretation, i.e. a fixed value of q, there corresponds a Fokker-Planck equation and vice versa to each Fokker-Planck equation a Langevin equation with a specified interpretation may be assigned.
2.8
The bistable oscillator
In the stochastic theory, the bistable oscillator coupled to a thermal bath is one of the best investigated nonlinear models. A mechanical realization is a Brownian particle of mass m moving in a potential U(x) with two local minima. The bath exerts a linear damping force and a Gaussian white random force. The dynamics of the bistable oscillator then is described by a Langevin equation which can be written as dx ~dl = v,
dv
dU(x)
m—
= -TV - — ^
/—-—--
+
, .
y/2ikBT£(t)
(2.238)
78
Tools of Stochastic
Dynamics
where £(t) is a normalized Gaussian white random force with (£(£)) = 0 and (£M£( S )) = 8{t - s). The strength of the random force is related to the damping constant 7 and the temperature T of the bath by the Einstein relation. As a particular example we take a symmetric quartic potential U{x) = - °-x2 + -x4,
a,b>0
(2.239)
that gives rise to the force
Fig. 2.2
Shape of the bistable potential with two minima.
f(x)
= - ^ ^
= ax-
bx3 .
(2.240)
If we neglect the random force but keep the damping, the resulting deterministic dynamics has two stable stationary points in phase space with coordinates (2.241) and a hyperbolic point at the origin: Vo
= 0,
xb = 0.
(2.242)
At weak damping, i.e. if 7 < 8a, the stable points are foci and else nodes. These points coincide with the stationary points of the energy of the particle, E — mv2/2+ U(x). The stable points correspond to the two equally high minima of the energy and the hyperbolic point to a saddle point. There, the energy barrier separating the two wells is lowest. The energy difference between the saddle point and the minima is AU — a2/(4b). In the deterministic approximation, the separatrix of the basins of attractions of the locally stable points is formed by the stable manifold of the hyperbolic point. The shape of the two basins of attraction is shown in fig.(2.3) for different values of 7. It resembles two tadpoles with infinitely long tails wrapped around each other.
The bistable
oscillator
79
20
(c)
10
V0
-10
-20
Fig. 2.3 The separatrix between the domains of attraction of the locally stable points v = 0, x = ± 1 for the motion of a particle of mass m = 1 in the potential U(x) = x 4 / 4 — x 2 / 2 for different values of the friction constant, 7 = 0.1,1, 3 in panels (a), (b) and (c), respectively. Note that the ratio of the velocity and friction scales is the same in the different panels but that the absolute scales differ.
Without the random force, the damping leads to a permanent loss of energy that is given by
±E=±
{^v2 + U(x)) =
-W<0.
(2.243)
dt dt It vanishes at the fixed points where v — 0. When the damping constant is large, 7 -> 00, or the particle is light, m —> 0, the inertial force becomes negligibly small compared to the damping and the potential force. Consequently, the velocity can be adiabatically eliminated and the particle moves according to dx ~di
m,
(2.244)
where we again took into account the fluctuating force and a general (bistable) potential. So far we have neglected the influence of the random force that is caused by the bath. It typically drives the particle out of the stable states. For small noise, however, the average typical deviation of the particle from the stable points is small compared to the distance between the stable points and the hyperbolic fixed point. On the other hand, there are realizations of the random force that drive the particle from one stable state over the saddle point to the other stable state. We will see that these events are rare if the thermal energy is small compared to the energy barrier separating the two stable states, but, also that they occur with certainty. Strictly speaking, an arbitrarily small white Gaussian random force destabilizes
80
Tools of Stochastic
Dynamics
the formerly stable states of a dynamical system and renders them metastable. The bistable oscillator is just one of the simplest examples of this widespread and important effect. It has many different applications in physics, chemistry, biology and technical sciences. To name an example we consider a molecule that may exist in two different configurations A and B. In a solvent, transitions between these two forms may occur: A^ B .
(2.245)
In the energy landscape of the molecule, the two configurations A and B correspond to two minima which are separated by a barrier. The interaction with solvent molecules provides the necessary energy to overcome this hindrance. The reaction coordinate that leads over the barrier where it is lowest can be interpreted as a particle's coordinate. Assuming that the solvent acts with many weak collisions on a fast time scale, the reaction can be described by a Langevin equation as given by eq. (2.238) where U(x) is a conveniently defined potential of mean force. Other examples of bistable behavior are more complicated chemical reactions, optical flipflop-devices, and optically bistable systems. Characteristic for all these cases and many others is the passage of a saddle point that acts as a bottleneck for the dynamics. Before we discuss the escape from a metastable state in more detail we give the Fokker-Planck equation for the probability density P(x, v, t) in phase space as it follows from the Langevin equations (2.238): \dt+Vdx
m dx dvj
nX,V,t)
\mdvV+
m* dv* f
nX,V t}
' ' (2.246) This equation is also named after Klein who derived it first and Kramers. The left hand side of the Klein-Kramers equation represents the reversible flow of probability in phase space while the right hand side describes the irreversible effects caused by the interaction with the heat bath. It has the form of a Stossintegral in the particular limit of infinitely frequent and vanishingly small collisions. For a potential that increases sufficiently fast for x —> ±oo the Klein Kramers equation has a uniquely defined stationary solution that can be normalized to unity on the total phase space. It coincides with the Maxwell Boltzmann distribution:
P^,„)= W -e x p(-^-^M).
( 2 . 2 47)
The fact that this probability density is approached from any initial distribution clearly demonstrates that transitions between the metastable states must take place. Otherwise the statistical weights that one can attribute to each metastable state (1/2 for a symmetric double well potential) could not reach their equilibrium values in accordance with the Maxwell-Boltzmann distribution (2.247). This also shows
The escape problem
81
that only the presence of noise brings into play the full nonlinearity of which otherwise the system would not pay much attention. Because the Langevin equation defines an ergodic process we can infer ratios of the times that the system dwells in different states from the corresponding ratios of the ensemble density. Therefore, the ratio of times tstabie and tsaddie that the system dwells in equally large phase space regions of the size dxdv at the stable point (VQ,XI) and the saddle point (vo,X2), respectively, equals the ratio of the Maxwell Boltzmann distribution at the corresponding locations: tsaddle _ P-{x tstabie
eq
P {x
= x2tv = v0) = X1,V
_exp(_*U\
= V0)
\
kBT
where AU = U{x2) — U(x\) = a2/(ib) denotes the hight of the energy barrier as seen from the bottom of either well. This corroborates our previous claim that for high barriers (AU
2.9
The escape problem
So far we have seen that Gaussian noise, how small it be, leads to the destabilization of deterministically stable states. It is quite natural to assume that the decay of a metastable state follows an exponential law which consequently is uniquely characterized by a rate constant k. This was verified theoretically, numerically and experimentally in a wide range of situations. Hereby it is assumed that once the system has left its initial metastable state this state will not be repopulated. This, of course, requires certain modifications of the process that will be discussed below. If we again consider the bistable oscillator as an example, we can ask how the probabilities develop in time to find it in either of its metastable states. In the case of an exponential decay, this coarse grained dynamics is given by a simple master equation describing the transitions between the two metastable states: p-(t)
=
-kp-(t)
+ kP+(t)
p+(t)
=
kp_(t) - kp+(t),
(2.249)
where P-{i) and p+(t) denote the probabilities to find the system in the domains of the attractors at (VQ,XQ) and (VQ,X\), respectively. Moreover, we assume a symmetric potential. Therefore the rates out of the metastable states are equal. They are denoted by k. Within this simplified picture, still different physical processes can be described which correspond to particular experiments from which the rate can be determined. The decay of a single metastable state, say the one at (i>o, £0), is an example that was already mentioned. It can be observed if the back flow from the other metastable
82
Tools of Stochastic
Dynamics
state is prevented, i.e if the second state is changed into an absorbing state. The same effect is reached by setting p+(t) = 0. The master equation then simplifies to pabs(t) = -kpabs(t),
(2.250)
yielding pa}s(t) = exp{-kt} for the decay of the population of the initial state. This population gives the probability that the system has not left the initial state up to the time t and therefore p" 6s (i) is also called the waiting time distribution. The probability density of exit times, p(t), follows as the negative derivative of the waiting time distribution, p(t) — kexp {—kt}, from which the moments of the exit time can be calculated: oo
n
(t ) = fdttnp{t) o
= n\k-n .
(2.251)
Of particular importance of course is the first moment, i.e. the mean exit time, which is given by the inverse rate. (t) = l/k.
(2.252)
In another experiment one allows for jumps in both directions and observes how a distribution initially localized in one state relaxes toward the equilibrium distribution. Prom the master equation the time dependence of this process readily follows: p T (t) = i ( l ± e - 2 f c t )
(2.253)
Note that in this case the relaxation is characterized by the sum of the rates of both states resulting in 2k which is just the negative non-vanishing eigenvalue of the coefficient matrix on the right hand side of the master equation. According to the regression theorem one finds the same relaxation law for equilibrium correlations of the population. This fact is used in the reactive flux method which is an effective way to numerically determine rates. One can also think of a non-equilibrium situation in which a steady current is maintained by a source that emits particles at a constant rate in one state, and a sink that instantly removes particles arriving at the other state. The probability current j that flows, say, from the left replenished to the right absorbing state, then follows as j = kp_ resulting in the flux-over-population formula for the rate: k = 3- , (2.254) n where n—p- denotes the population of the replenished state. These different situations can also be described in terms of the more detailed model of the Langevin equation, or the equivalent Fokker-Planck equation. We will shortly describe these methods before we discuss some of them in more detail in the following sections.
The escape problem
83
In order to determine the rate from the decay of the metastable state one again has to modify the process in such a way that back reactions are excluded. This conveniently is done by introducing an absorbing boundary enclosing the final state. The mean first passage time of the boundary then gives the inverse of the rate provided the absorbing boundary is close enough to the final state. The decay of a non-equilibrium initial state according to the Fokker-Planck dynamics seemingly is much more complicated than the same process described by the master equation because the Fokker-Planck operator has infinitely many eigenvalues each of which gives rise to an exponentially decaying component of the probability density. The very fact, however, that the transitions between the metastable states occur only rarely, also shows itself in the spectrum of the Fokker-Planck operator: It has a pronounced gap separating two eigenvalues from the rest of the spectrum. One of these two eigenvalues is zero, Ao = 0 corresponding to the equilibrium solution of the Fokker-Planck equation; the other one is negative and has a small absolute value. The corresponding eigenfunction has a node line leading through the saddle point and in the neighborhood of the stable states its absolute value coincides with the Maxwell-Boltzmann distribution. All eigenvalues belonging to the rest of the spectrum possess a negative real part with an absolute value that is much larger than the one of Ai: |5RAfc| 3> |Ai|, k > 1. This mathematically reflects the time scale separation: The "large" eigenvalues describe the fast relaxations that take place within each of the domains of attraction and the "small" one, Ai, is responsible for the transitions between the metastable states. As for the master equation the small eigenvalue Ai is the negative sum of the rates out of the two metastable states. There exist effective analytical and numerical methods to determine the small eigenvalue of a Fokker-Planck operator that describes a system with metastable states. At the first sight it seems very complicated to use the fiux-over-population method for the Fokker-Planck equation: In principle, one has to solve the stationary Fokker-Planck equation that is complemented by source and sink terms. This problem was circumvented by Kramers in a very elegant way: Rather than specifying particular sources and sinks he constructed a current carrying probability density that fulfills three requirements: First, that the sources maintain the Maxwell-Boltzmann distribution in the initial well, second, that the sinks lead to a vanishing distribution in the other well, and, third, that the region close to the saddle point is free from sources and sinks. In this region the potential force can be linearized and the resulting Fokker-Planck equation can be solved with asymptotic boundary conditions so that also the first two conditions are satisfied. Before we further discuss this approach we present a rough estimate of the rate that has the virtue of providing an upper bound for the rate.
84
2.9.1
Tools of Stochastic
Transition
state
Dynamics
theory
The transition state theory makes use of the flux-over-population formula of the rate. Flux and population are calculated from the Maxwell-Boltzmann distribution by only taking into account positive velocities at the barrier which for the sake of simplicity is assumed to be located at Xb = 0. With these assumptions one finds for the flux: oo
j
T S T
= / dvvPeq(0,«) J
= - ^ e mN
-
^
(2.255)
o and for the population of the initial (left) well: oo
n=
f -oo
0
,
dv
f
,
™„/
N
dxPeq(x,v)&
27rfcBT
-U(*P)
^ - e *B« ,
(2.256)
—oo
where we have used a parabolic approximation of the potential near the initial stable state at x — XQ and extended the integration to infinity. Both approximations are justified for low temperatures and for a symmetric potential. They are controlled by the smallness of the parameter U^(x0)kBT/(8(U"(x0))2) = SkBT/AU. lv Here U"(XQ) = mwg = 2a and U^ ^(x0) denote the second and fourth derivative, respectively. With the flux-over-population formula one obtains the rate of transition state theory (TST): kTST = ^ e ~ ^ .
(2.257)
This result has a simple interpretation: The frequency coo/2n gives the number of attempts per time to overcome the barrier and the Arrhenius factor the fraction of successful approaches. The TST rate is based on the two assumptions that, first, the positive velocities are thermally distributed even at the barrier and, second, that there are no particles with negative velocities there. The first assumption requires a rather strong interaction of the particle with the heat bath such that the thermal distribution can be maintained even though particles escape, whereas the second assumption requires that the interaction with the bath is sufficiently weak in order that a particle that already has passed the barrier is not scattered back by the random force exerted by the bath. Both assumptions overestimate the probability flux at the barrier. Therefore the transition state theory represents an upper bound of the true rate: kTST < k.
(2.258)
Transition state theory has widely been used in physics and chemistry. Often a considerable improvement of the rate can be achieved by taking into account other degrees of freedom than the reaction coordinate. The generalization of transition
The escape problem
85
state theory to multidimensional problems is also based on the flux over population expression for the rate. In the multidimensional phase space a dividing surface is introduced with the "reactants" on the one and the "products" on the other side of this surface, i.e. the reaction coordinate has to cross the dividing surface. The system is assumed to be in a thermodynamic equilibrium state fixing the probability distribution in phase space, as e.g. the canonical distribution. The population of reactants is determined by the integral over the respective part of phase space with the dividing surface as boundary, and the flux follows as the unidirectional probability current leading from reactants to products through the dividing surface. This again gives an upper bound for the rate that now may come very close to the true rate provided that those degrees of freedom are included that interact with the reaction coordinate and that the dividing surface is properly chosen. A variation of the dividing surface generally will lead to a change of the value of the TST rate. Together with the bounding property of the rate, eq. (2.258), one obtains a variational principle for the rate. This is the basis of the variational transition state theory. A related question which is of relevance in signal communication and for many other technical and biological processes concerns the frequency with which a signal x crosses a threshold Xs- Under the assumption that the signal is stationary and Gaussian distributed with average value zero and that the velocity of the process can be defined and has a finite variance, Rice (1944) determined the crossing rate of the threshold at x — x$ as: kRice
= J_ /SiMexpf--^|-V
(2.259)
where ipx,x(t) = (x{t)x) a n ( i i>v,v(t) = (v(t)v(Q)) — —d2i[)x,x(t)/dt2 denote the correlation function of the signal and its velocity, respectively. The exponential factor has its counterpart in the Arrhenius factor of the transition state rate. Moreover, Rice showed that the prefactor just coincides with the expected number of zero crossings per second of the signal x. Hence, also the prefactor has an analogous meaning as the one of transition state theory. It is remarkable that no further details of the process enter than the variances of the signal itself and its velocity. These quantities also can be expressed in terms of the spectrum of the process. 2.9.2
Kramers'
rate
formulae
In his pioneering work of 1940 Kramers derived expressions for the rate in the two cases of extremely weak damping and moderate to strong damping. It took almost half a century until Melnikov and Meshkov (1986) presented a crossover theory from the underdamped to a still weakly damped regime where transition state theory applies, and until Pollak, Grabert and Hanggi covered the full regime from weak
86
Tools of Stochastic
Dynamics
to strong damping. Here we will restrict ourselves to those cases originally treated by Kramers. For a discussion of the crossover theory we refer to the literature (Hanggi, Talkner, Borkovec, 1990; Melnikov, 1991). The model that was discussed by Kramers is that of the bistable oscillator, see eq. (2.238). 2.9.2.1
Moderate to strong damping
If the damping is sufficiently strong a typical escape path will cross the separatrix in the vicinity of the saddle point at (VQ = 0, xj,). Following Kramers we construct a probability density p(x, v) that in this region close to the saddle point is a solution of the stationary Fokker-Planck equation, Lp(x,v)=0,
(2.260)
and that approaches the Maxwell Boltzmann distribution in the reactant well and vanishes in the product well. As explained above this amounts to the construction of a stationary flux carrying solution maintained by sources repopulating the reactant state and sinks taking out the products. Moreover, neither the sources nor the sinks must come too close to the saddle point. In order to take into account the prescribed behavior in the two wells, we split off the Maxwell-Boltzmann distribution: p(x, v) = h(x, v)Peq(x, v)
(2.261)
where h(x, v) is a form function that approaches unity at the side of the reactants and zero at products: / 1 *<*.«) = { 0
at reactants, at products.
,„ „„„, '
(2 262)
Finally, as a consequence of eq. (2.260), in the region of the saddle point, the form function is a solution of the modified Fokker-Planck equation: r 2 7 I dh(x,v) -ykBT d2h{x,v) + - ^ 42 ^ 2 = 0(2.263) u>hx H v dv m dv m Because here the coordinate x is restricted to the close vicinity of the saddle point we have approximated the potential by a parabola: dh{x,v) —v- dx
U(x) a £7(0) - \mJix2
,
(2.264)
where mw2 = -d2U(xb)/dx2 is given by the second derivative of the potential at the barrier. This partial differential equation reduces to an ordinary one for £(u) = h(x, v) where u is a linear combination of position and velocity u = x
A T;V . 72
WiT
(2.265)
87
The escape problem
Here the coefficient A coincides with the positive Lyapunov exponent of the deterministic dynamics at the saddle point:
The differential equation for the auxiliary function ((u) takes the form:
jgu)
|
7Afc B rd 2 c(tt) =0j
(2267)
It is readily solved and, together with the boundary conditions (2.9.2.1), yields for the form function: oo
h{x,v) = -^=
dze~z2/2.
f
(2.268)
We note that one also can reduce eq. (2.263) to an ordinary differential equation by using the second (negative) Lyapunov exponent in eq. (2.267). With the resulting solutions, however, one cannot satisfy the boundary conditions (2.9.2.1). In contrast to transition state theory one now determines the net flux over the barrier rather than the unidirectional flux: oo
/
dvvp(0,v)
A ^ e " ^ . ujb mN
(2.269)
v
'
It differs from the flux obtained in the transition state theory by the " transmission" factor A/wfc. It is smaller than unity, approaches unity for •y/m —>• 0, zero for •y/m —> oo, and takes into account that particles that have crossed the boundary are possibly scattered back and therefore must not be counted as successful escape events. The population of the reactant well is given by the same expression as in the transition state theory because by construction the flux carrying probability density there coincides with the Maxwell-Boltzmann distribution. As a final result, one obtains for the rate: nKramers
=
"U)b
•
(2.270)
In the limit of weak damping the result of transition state theory is reproduced. We will come back to this limiting behavior. The stronger the damping is, the more important becomes the influence of backscattering on the rate which leads to an asymptotic decay of the rate for large damping proportional to 1/7:
88
Tools of Stochastic
Dynamics
This result also directly follows from the Smoluchowski equation as we will see below. The Kramers rate itself is an asymptotic result that becomes exact in the limit of high barriers, i.e. for AU/ksT —I oo. There are two types of corrections to this asymptotic behavior: Analytic ones in k&T/AU and non analytic ones of the particular Arrhenius form exp {- AU'/'(k^T)}. It turns out that, if the barrier height is so low that also the non-analytic corrections become relevant, the different approaches which we briefly discussed, will yield different results for the rate. This indicates that for too low barriers the rate picture starts failing. As a estimate when the rate description ceases to exist, one may consider a barrier height of AU/k-gT = 4 that introduces a non-analytical error of roughly 1%. For barriers lower than this value a more detailed analysis of the considered system is necessary. For higher barriers the rate picture yields an adequate description but the analytical corrections may still be relevant. They can systematically be taken into account both in transition state theory (Pollak, Talkner, 1993) and on the level of the Fokker-Planck equation (Talkner, 1994). 2.9.2.2
Weak damping and energy diffusion
In the previous section we revised the assumptions of transition state theory related to backscattering. For a sufficiently large damping strength this indeed is the main mechanism that modifies the rate. Because the damping strength is not only a measure how effective any excess energy is taken from the system to the bath but also how effective the bath supplies the system with energy, the maintenance of local equilibrium on the time scale of the escape is guaranteed if the damping is sufficiently large. If it is small, the supply of energy becomes the relevant rate determining step. For that reason Kramers determined a reduced diffusion equation for the energy of the particle holding in the limit of vanishing damping:
dP(E,t)
7
d 1 +
k
-
T
^
^lP(E,t),
(2.272)
where 1(E) — § ^2m(E - U(x)dx is the action and to(E) = 2wdE/dI the frequency of the undamped system at the energy E. For this one dimensional diffusion equation one can find the exact stationary solution carrying the constant probability current j . Using the flux-over population expression one obtains for the rate k
^
I E
0
=
i b)l j^TST tA^LL kTsr t mkBT
(2-273)
where the population has been evaluated to leading order in ksT/AU. Here I(Et) is the action at the energy of the barrier. In contrast to the rate expression (2.9.2.1), the energy diffusion controlled rate vanishes with the damping constant. Obviously, when the factor in front of the transition state rate becomes of the order of unity, the energy diffusion rate becomes larger than the TST rate and can no longer be
Pontryagin's
89
equation
valid. We will not discuss the rather complicated theories that describe the crossover between the two regimes. For practical purposes it is often sufficient to use a simple Pade like interpolation formula: hut = \kKramers
+ fc7->o) '
(2.274)
which may introduce an error of maximally 20%. 2.9.3
Transition
rates in multidimensional
landscapes
There are many situations which are described by a motion in a multidimensional energy landscape that cannot be reduced to a single coordinate. For systems that in the limit of long times approach a thermal equilibrium state and that obey the symmetry of detailed balance, transition rates between local minima of the energy can be determined in a way that is analogous to Kramers method described above. Landauer and Swanson (1961) determined the rates for a multidimensional overdamped system described by a Smoluchowski equation. Later, Langer (1969) also considered more general Fokker-Planck dynamics including the effect of inertia and also considered nucleation rates in field theories by performing a continuum limit. Here, we only will give general results for systems with a finite number of degrees of freedom. The transition rate out of a metastable state over a barrier with an energy AE above the initial well again is dominated by the Arrhenius factor exp{—AE/k#t}. The prefactor is given by the positive Lyapunov exponent A of the deterministic motion at the saddle point multiplied by the square root of the ratio of the determinants of the second derivatives of the energy (the Hessians) at the initial energy minimum and at the barrier, Ho = (d2E(xo)/dxidxj) and Hb — (d2E(xb)/dxidxj), respectively: hanger = — ^ J^~e"AE/ksT
.
(2.275)
Note that the absolute value of the Hessian at the barrier has to be taken because it is always negative as a consequence of the unstable direction. Here, we only have considered the most simple cases of a point like initial state and a single, point like barrier. For further details and more general cases we refer to the literature (Hanggi, Talkner, Borkovec, 1990). 2.10
Pontryagin's equation
Several more specialized tools exist for Markov processes that allow one to determine certain relevant aspects of the process without the need of knowing the full conditional probability as the solution of either the forward or the backward equation. Often the introduction of boundaries and the modifications of the process at
90
Tools of Stochastic
Dynamics
the boundaries are necessary for these methods. We first will introduce the most frequent types of boundary conditions and discuss how boundary conditions for the forward equation can be related to ones for the backward equation and vice versa. Then we will discuss first passage times and so-called splitting probabilities. 2.10.1
Boundary tion
conditions for the forward
and the backward
equa-
For the sake of simplicity we restrict ourselves to one dimensional processes. Generalizations of the concepts discussed in this section to processes in higher dimensions are straightforward. Most exact solutions known in one dimension unfortunately do not simply translate to higher dimensions, but much of the qualitative behavior of particular solutions found in one dimension also applies in higher dimensions and often can be used as an inspiration and motivation for approximate solutions and particular ansatze. We consider a time-homogeneous Markovian diffusion process x(t), i.e. the conditional probability P(x, t\y, s) of the process only depends on the time difference t — s and solves the forward and the backward equation: •^P(x,t\y)
=
jtP{x,t\y)
=
LxP(x,t\y), L+P(x,t\y),
(2.276)
where L and L+ denote the forward and backward Fokker-Planck operators, respectively. The indices of the respective operators indicate the variables on which they act: The state x at time t of observation in the forward and the condition y at the earlier time s = 0 in the backward equation. As we already have seen, the Fokker-Planck operator is a second order differential operator with coefficients K(x) and D(x) that characterize the drift and diffusion, respectively:
•+
_
In order to have a mathematically well denned problem the forward and backward equations have to be complemented by initial and boundary conditions. The initial condition is obvious: P(x,0\y) = S(x - y).
(2.278)
The boundary conditions depend on the particular physical situation under consideration and require somewhat more thought. An important aspect here is that the
Pontryagin's
91
equation
forward and the backward operators are adjoint operators relative to each other: [ dxf(x)Lp(x)
= fdxp{x)L+p{x)
G
(2.279)
G
This has to hold for all admissible functions f(x) and p{x). It does not only determine the relation between the forward and the backward operator as it is evident from the eq. (2.277) but also the behavior of the admissible functions f(x) at the boundaries of the domain G once the properties of p(x) are fixed and vice versa. As a first example for G we consider the interval [2/1, y2] with absorbing boundary conditions at y\ and y2. These are most naturally characterized in terms of the conditional variable of the transition probability by stating that no transitions may take place from the boundaries.This leads to the following conditions for the backward equation describing absorption: P{x,t\Vl)
= P{x,t\y2)
=0
forie(i/i,j/2).
(2.280)
In order to find the boundary conditions for the forward equation, one has to require that the condition (2.279) holds for all functions f(x) that vanish at the boundaries according to eq. (2.280). This is the case only if at the same time the function p(x) also vanishes at x = 2/1 and x — y2. Therefore, at the absorbing boundaries we find as boundary conditions for the forward equation: P{yi,t\y)
= P{y2,t\y)=0
for y G (2/1,2/2) •
(2.281)
As a second example we take the same interval [2/1,2/2] a s domain but now with reflecting boundaries at 2/1 and y2, i.e. every trajectory arriving from the interval at the boundary is send back to the interior of the interval. Hence, the probability fluxes through the boundaries vanish. This gives the reflecting boundary conditions for the forward operator: K(yi)P(yi,t\y)
- -^ [D{x)P(x,t\y)]x=yi
K(y2)P(y2,t\y)-~[D(x)P(x,t\y)}x=y2
= =
0
for y G (2/1,2/2) -(2.282)
Requiring such boundary conditions for the functions p(x) one finds the condition (2.279) satisfied only if the first derivative of the function f(x) vanishes at the boundaries. This leads to the reflecting boundary conditions for the backward equation: dP(x,t\yi) dP{x,t\y2) ,„ os„, \, = \ =0 for x G (2/1,2/2) • (2.283) dy dy So-called natural boundaries are never reached by a process in finite time. For the forward equation that amounts to the vanishing of the probability together with its first derivative. No condition follows for the backward equation in this case.
Tools of Stochastic
92
Dynamics
Finally we note that the left and the right boundary points may be of different character, e.g. y\ may be reflecting and 2/2 absorbing. 2.10.2
The first passage
time
distribution
The probability Vroh{x(s) S G, s < t,x(0) = y) = WG(t,y) that a process x(t) has not left a certain region G of its state space up to time t, in general depends on the starting point x(0) = y £ G. The probability WG(y,t) can be calculated from the transition probability PG(x,t\y) of a modified process that is stopped whenever the process has reached one of the boundaries of G. Before that happens it coincides with the original process. Consequently the conditional probability of the modified process fulfills the forward and backward equation of the original process on G with absorbing boundary conditions at the boundaries of G. The waiting time distribution WG(y,t) coincides with the total amount of probability found in G at time t:
WG(y,t) = J dxPG(x,t\y)
(2.284)
G
We differentiate both sides with respect to time, use the backward equation for the time rate of change of the conditional probability, interchange the backward operator (acting on y) with the integral over x, and finally obtain the backward equation as the equation of motion of the waiting time probability: =
L+WG(y,t)
WG{y,0)
=
1
WG{y,t)
=
0
^a(y,t)
for y e G, ioryeG,
ioryedG.
(2.285)
The initial and boundary conditions for WG(y,t) are direct consequences of the respective conditions for the conditional probability PG(x,t\y) in combination with the definition (2.284) of the waiting time distribution. Once the waiting time distribution is known, the probability density pG (y, t) of the exit times follows as the negative derivative of WG(y,t) with respect to time: pG(y,t)
= --WG(y,t).
(2.286)
Prom the probability of exit times poiv-, t) moments of the first passage time follow: 00
n
(t (y)) = JdttnpG(y,t).
(2.287)
0
Acting on both sides of this equation with the backward equation, using L+p(y, t) = dp(y,t)/dt and integrating by parts one finds the following hierarchy of equations
Pontryagin's
equation
93
for the nth moments of the first passage time: L+(t"(y)}
=
(tn{y))
=
-n{tn-\y)) 0 toryedG
(2.288)
where the absorbing boundary conditions follow from those of the waiting time distribution, see eq. (2.285). Moreover we have assumed that the moments exist. In particular, for the mean first passage time one obtains: L+(t(y))
=
-1
(t(y))
=
0 for y€dG
(2.289)
This equation was derived by Pontryagin, Andronov, and Witt (1933) and is known as Pontryagin equation. An extension to higher dimensional cases can be find in (Weiss, 1967). We mention that the same form of the Pontryagin equation also holds for Markovian processes in higher dimensions. The operator L+ then denotes the backward operator of the considered process. As a final comment we note that not all points of the boundary dG need to be absorbing. For example, in the case of a one dimensional process with an interval as domain, one endpoint of the interval may be reflecting and the other one absorbing. Before we discuss some examples for mean first passage times we introduce another quantity that characterizes a process in the presence of two absorbing boundaries.
2.10.3
Splitting
probability
We again consider a one dimensional Fokker-Planck process x(t) that starts at a point y within an interval [yi, 2/2] • We pose the question with which probability the process will leave the interval at the boundary point y\. This quantity is denoted by 7Ti (y) and is called the splitting probability. Obviously, it is sufficient to follow the process until it reaches either boundary and then to stop it. Hence, both boundaries are absorbing. The probability flowing out of the interval at x — y\ per time is given by the probability current j(Vi,t\v)
= -^[D(x)P{x,t\y)]x=Vl.
(2.290)
Here the drift term goes not contribute to the probability current because it is proportional to the conditional probability P(x,t\y) at the boundary x — y\ which vanishes there. The splitting probability results as the total probability flowing through x = y\ and, hence is given by the integral of the respective probability current over all
94
Tools of Stochastic
Dynamics
positive times: oo
7ri(y) = Jdtj(yut;y).
(2.291)
Acting on both sides of this equation with the backward operator, interchanching it on the right hand side with both the time integration and the derivative with respect to x, and using the backward equation one obtains the difference of the probability current j(yi,t;y) at infinite and zero time. For y £ (2/1,2/2) it vanishes at both times and, hence, one finds as an equation for the splitting probability:
^1(2/)
=
0
7Tb(2/l)
= =
1, 0.
7r6(2/2)
fory£
(2/1,2/2),
(2.292)
The boundary condition follow from the definition of the splitting probability in terms of the probability current (2.290) and its behavior at the boundaries. They have the simple interpretation that a particle that starts at the absorbing boundary 2/i will never reach the other boundary 2/2 and vice versa. For processes in higher dimensions that may leave a region G at different boundaries dGi the splitting probability K\{y) is defined as the relative frequency with which the particular boundary dG± is reached before any other boundary dGi, i i=- 1 has been crossed. It satisfies the analogous equation (2.292) with the corresponding multidimensional backward operator. The boundary conditions are 7Ti(j/) = 1 for 2/ E dGi and iri(y) — 0 for y G dGi, i ^ 1-
2.10.4
Examples
In one dimensions the equations (2.289) and (2.292) for the mean first passage time and the splitting probability can be solved for arbitrary drift and diffusion.
2.10.4.1
The splitting probability
The equation for the splitting probability out of the interval [2:1, £2] to the left boundary reads:
K
MTy+DM$h^ TTI(2/I) = 1,
= °>
7r1(y2) = 0.
(2.293)
Pontryagin's
95
equation
It readily is solved to read: y
fdzexp{V(z)} Mv) = Ti . /dzexp{V(z)}
(2-294)
2/1
where
v{y)
z
v
(2-295)
=-h W)' Vi
The other splitting probability to reach x/2 before y\ is just the complement of to one:
7TI(T/)
VI
fdzexp{V(z)} 7T2 (y) = Y2 = 1 - m (y). Jdzexp{V(z)}
(2.296)
yi
We specialize this general result to the two cases of pure diffusion and of the overdamped bistable oscillator of the previous section. In the first case we find for vanishing drift, K(y) = 0, and constant diffusion D(x) = D a linear dependence of the splitting probability on the starting point y: n*ff(y)=yiZJL.
(2.297)
2/2 -yi In the case of the overdamped oscillator we start the process in a point between the two metastable states and ask with which probability the right metastable state will be reached before the left one. This is given by the splitting probability
?dzexp{g} Mv) = ir1
r,
(2-298)
Jd,exp{fM} where, as denned above, x$ and x\ denote the metastable states to the left and the right of the barrier, respectively. The splitting probability connects the prescribed values at the boundaries in a monotonically decreasing way. For a symmetric potential it takes the value 1/2 at the barrier. This point from where the system reaches both metastable states with equal probability is also called the stochastic separatrix. If the potential is not symmetric about the barrier the stochastic separatrix in general does not coincide with the deterministic one sitting on top of the barrier. Only in the limit of weak noise the stochastic separatrix approaches the deterministic one.
96
Tools of Stochastic
Dynamics
In this limit one can approximate the potential in eq. (2.298) by a parabola, U(x) w U(xb) — mujl(x — Xb)2/2, extend the limits £0 and x-i to —00 and 00, respectively, and perform the resulting Gaussian integrals. As result one obtains: 7r 0 (»)»-erfc
W—±-y\
(2.299)
where erfc(z) denotes the complementary error function. Hence, the splitting probability is almost unity for all points left of the barrier except for a thin boundary layer of the thickness of the order of y/'k\,T'/'(muj) on which it rapidly falls to zero. That simply means that trajectories starting outside the thin layer at the barrier almost behave as being not influenced by the noise and go to the next local equilibrium point. Only within the thin layer at the barrier the noise is effective and may redirect a trajectory to the "wrong" metastable state. 2.10.4.2
The mean first passage time
In one dimension the equation for the mean first passage time becomes:
K +D{y)
iUm)
K
^-
= -1 {t(x2))=0
(2.300)
where we only consider exits through the right boundary point y = yi- Once the trajectory reaches the left boundary at y = y\ it is reflected. Other situations can similarly be treated but will not be discussed here. The solution of this boundary value readily can be found:
<«(,))./*,/d.aCMH» V
(2 .3 01 )
VI
where V(y) is defined by eq. (2.295) For a freely diffusing particle one obtains: (t(y))dlff
= ^
[(1/1 - y)2 - (2/1 - 2/2)2]
(2.302)
The parabolic profile reflects the typical behavior of diffusion. For the bistable oscillator we choose the metastable state at the right of the barrier as absorbing state, j/2 = xi and take the limit x\ -> - 0 0 for the left (reflecting) boundary point. The average time it takes to reach the metastable state Xi from a starting point on its left hand side then becomes:
m.£j«f^{m^m}. y
-00
(,303,
Pontryagin's
97
equation
For sufficiently weak noise the double integral is dominated by a single pronounced maximum of the integrand at u — Xb = 0 and v — xo. If the starting point y lies on the side of the barrier opposite to the final metastable state at x\ in a distance of the barrier that is large compared to the thermal length lth = ijk^T/(mu^) the integral becomes independent of y and takes the value T: xi
m) = T==
0
duexp
l^I
I dvexp{~^}
{l^}
iovy<Xb hh
~ -
—OO
XQ
(2.304) For low temperatures both integrals can be evaluated in Gaussian approximation to yield: mw0Wfc MU I K, ekBT
=
Vk^Zers-
(2-305)
If the starting point lies in the barrier region or on the side of the final metastable state, the two integrals in eq. (2.303) can be disentangled by taking zero as upper limit in the u-integral. This introduces an exponentially small error of the order exp < — Y^p >. The u-integral can be expressed in terms of the splitting probability. For the mean first passage time this yields the appealing result: (t{y)) = TMy)
for
x0 < y < xx.
(2.306)
It means that all the deterministic times it takes the system to move from the initial state to the next metastable state and from the barrier to the final metastable state are much shorter than the exponentially large waiting time in a metastable state. For y-values smaller than XQ the mean first passage time consequently also remains constant provided the potential goes sufficiently fast to infinity for x —> ±oo. The fact that there is a single mean waiting time for the whole domain of attraction apart from a thin layer at the boundary also justifies the coarse grained rate picture discussed above. A further quantitative justification of the rate picture is the agreement of the mean waiting time with the inverse Kramers rate, see eq. (2.305) (Reimann et al, 1999). References V.S. Anishchenko, A.B. Neiman, F. Moss, L. Schimansky-Geier (1999): "Stochastic Resonance: Noise Induced Order", Uspekhi Fiz. Nauk. 69, 7 [Engl. transl.:P/iys.Uspekhi 42, 7]. V. Anishchenko, A. Neiman, A. Astakhov, T. Vadiavasova, and L. SchimanskyGeier (2002), "Chaotic and Stochastic Processes in Dynamic Systems", Springer
98
Tools of Stochastic
Dynamics
Verlag, Berlin-Heidelberg-New York L. Arnold (1992), "Stochastic Differential Equations, Theory and Applications", Krieger, Malabar. R. Becker (1956), "Theorie der Warme", Springer, Berlin Heidelberg. J.A. Preund, T. Poschel T. (eds.) (2000), "Stochastic Processes in Physics, Chemistry and Biology", Lecture Notes in Physics, Vol. 557, Springer, Berlin, Heidelberg. J. Garcia-Ojalvo, J.M. Sancho (1999), "Noise in Spatially Extended Systems", Springer, New York, Berlin, Heidelberg C.W. Gardiner (1982), "Handbook of Stochastic Methods", Springer Series in Synergetics, Vol. 13, Springer, Berlin, Heidelberg. P. Hanggi, P. Jung (1995), "Colored Noise in Dynamical Systems", Adv. Chem. Phys. 89, 239. P. Hanggi, P. Talkner (eds.) (1995), "New Trends in Kramers Reaction Rate Theory", Kluwer, Boston. P. Hanggi, H. Thomas (1982), "Stochastic Processes: Time evolution, Symmetries and Linear Response", Phys. Rep. 88, 207. P. Hanggi, P. Talkner, M. Borkovec (1990), "Reaction Rate Theory: Fifty Years After Kramers", Rev. Mod. Phys. 62, 251. W. Horsthemke, R. Lefever (1983), "Noise Induced Transitions, Theory and Applications in Physics, Chemistry and Biology", Springer Series in Synergetics, Vol. 15, Springer, Berlin-Heidelberg. B. Karlin, M.M. Taylor (1975), "A First Course in Stochastic Processes", 2nd ed., Academic, New York. Yu.L. Klimontovich (1975, 1982), "Kinetic Theory of Non-ideal Gas and Non-ideal Plasma", Nauka, Moscow [Engl, transl.: Pergamon Press, Oxford, London, New York}. Yu.L. Klimontovich(1982, 1986), "Statistical Physics", Nauka, Moscow [Engl, transl.: Harwood Academic Publ., New York].
Pontryagin's
equation
99
Yu.L. Klimontovich (1994), "Statistical Theory of Open Systems", Kluwer, Dodrecht. Sh. Kogan (1996), "Electronic Noise and Fluctuations in Solids", Cambridge University Press, Cambridge. H.A. Kramers (1940), "Brownian Motion in a Field of Force and the Diffusion model of Chemical Reactions", Physica (Utrecht) 7, 284. P.I. Kuznetsov, R.L. Stratonovich, V.I. Tikhonov (1965), "Non-Linear Transformations of Stochastic Processes", Pergamon, Oxford. L.D. Landau, E.M. Lifshitz (1971), "Lehrbuch der Theoretischen Physik", Band 5, "Statistische Physik", 3rd ed., Akademie, Berlin. R. Landauer and J. A. Swanson (1961), "Frequency Factors in the Thermally Activated Process", Phys. Rev 121, 1668. J.S. Langer (1969), "Statistical Theory of the Decay of Metastable States", Ann. Phys. (N. Y.) 54, 258. H. Malchow, L. Schimansky-Geier (1985), "Noise and Diffusion in Bistable Nonequilibrium Systems", Teubner, Leipzig. V.I. Melnikov (1991) , "The Kramers Problem: Fifty Years of Development", Phys. Rep. 209, 1. V.I. Melnikov, S.V. Meshkov (1986), "Theory of Activated Rate Processes: Exact Solution of the Kramers Problem", J. Chem. Phys.85, 1018. E.W. Montroll, K.E. Shuler (1969), Adv. Chem. Phys. 1, 361. E. W. Montroll and J. L. Lebowitz, (eds.) (1987), "Fluctuation Phenomena", 2nd ed., North Holland., Amsterdam F. Moss, P.V.E. McClintock (eds.) (1990), "Noise in Nonlinear Dynamical Systems", Vol. 1-3, Cambridge University Press, Cambridge. E. Pollak, S.C. Tucker, B. J. Berne (1990), "Variational Transition State Theory for Reaction Rates in Dissipative Systems", Phys. Rev. Lett. 65, 1399. E. Pollak, P. Talkner (1993), "Activated Rate Processes: Finite Barriers Expansion
100
Tools of Stochastic
Dynamics
for the Rate in the Spatial-Diffusion Limit", Phys. Rev. E 47, 922. E. Pollak, H. Grabert, P. Hanggi (1989), "Theory of Activated Rate Processes for Arbitrary Frequency", J. Chem. Phys. 9 1 , 4073. L.S. Pontryagin, A.A. Andronov, A.A. Vitt (1933), "On the Statistical Treatment of Dynamical Systems", J. Exp. Theor. Phys. 3, 165 [Engl. Transl.: In: "Noise in Nonlinear Dynamical Systems", Vol. 1, ed. by F. Moss, P.V.E. McClintock, Cambridge University Press, Cambridge, p. 329]. P. Reimann (2002), "Brownian Motors: Noisy Transport far from Equilibrium", Phys. Rep. 361, 57. P. Reimann (2001), "A Uniqueness-Theorem for "Linear" Thermal Baths", Chem. Phys. 268, 337. P. Reimann, G.J. Schmid, P. Hanggi (1999), "Universal equivalence of mean firstpassage time and Kramers rate", Phys. Rev. E 60, 1. S.O. Rice (1954), "Mathematical Analysis of Random Noise", Bell Syst. Technol. J. 23, 1; , In: "Selected Papers on Noise", ed. by N. Wax, Dover, New York, p. 133. H. Risken (1984), "The Fokker-Planck Equation", Springer Series in Synergetics, Vol. 18, Springer, Berlin, Heidelberg. M. San Miguel, R. Toral (2000), "Stochastic Effects in Physical Systems", in: "Instabilities and Nonequilibrium Structures VI", ed. by E. Tirapegui, J. Martinez, R. Tiemann, Kluwer, Dordrecht, p. 35. L. Schimansky-Geier, V. Anishchenko, A. Neiman (2000), "Phase Synchronization: From Periodic to Chaotic and Noisy", in: "Neuro-informatics", ed. by S. Gielen, F. Moss, "Handbook of Biological Physics", Vol. 4, Series Editor A.J. Hoff, Elsevier, Amsterdam, p. 23. L. Schimansky-Geier, T. Poschel (eds.) (1997), "Stochastic Dynamics", Lecture Notes in Physics, Vol. 484, Springer, Berlin, Heidelberg. R.L. Stratonovich (1961, 1963,1967), "Selected Problems of Fluctuation Theory in Radiotechnics", Sov. Radio, Moscow (in Russian), "Selected Topics in the Theory of Random Noise", Vols. 1 and 2, Gordon and Breach, New York. R.L. Stratonovich (1990), "Some Markov Methods in the Theory of stochastic Pro-
Pontryagin's
equation
101
cesses in Nonlinear Dynamical Systems", in: "Noise in Nonlinear Dynamical Systems", ed. by F. Moss, P.V.E. McClintock, Cambridge University Press, Cambridge, p. 16. P. Talkner (1987), "Mean First Passage Time and the Lifetime of a Metastable State", Z. Phys. 5 68, 201. P. Talkner (1991), "Interrelations of Different Methods for the Determination of Rates: Flux over Population, Generalized Reactive Flux, the Lowest Eigenvalue and its Rayleigh Quotient", Ber. Bunsenges. Phys. Chem. 95, 327. P. Talkner(1994), "Finite Barrier Corrections for the Kramers Rate Problem in the Spatial Diffusion Regime", Chem. Phys. 180, 199. V.I. Tikhonov, M.A. Mironov (1979), "Markovian Processes", Sov. Radio, Moscow (in Russian). C. van den Broeck (1983), "On the Relation between White Shot Noise, Gaussianwhite Noise and the Dichotomic Markov Process", J. Stat Phys. 31, 467. N.G. van Kampen (1992), "Stochastic Processes in Physics and Chemistry", 2nd ed., North Holland, Amsterdam. N. Wax (ed.) (1954), "Selected Papers on Noise", Dover, New York. G.H. Weiss (1967), Adv. Chem. Phys. 13, 1. E. Wong (1971), "Stochastic Processes in Information and Dynamical Systems",McGraw Series in System Science, McGraw-Hill, New York. R. Zwanzig (1973), "Generalized Langevin Equation in Equilibrium", J. Stat. Phys. 9, 215.
This page is intentionally left blank
Chapter 3
Motion of test particles in a 2-d potential landscape O.A.Chichigina, A.V.Netrebko, and N.V.Netrebko 3.1
Formulation of the mathematical model
Before going to the discussion of the problems of penetration and escape of ligands, transitions from one conformation of the polypeptide chain to another, and proton transfers in the system of hydrogen bonds we will study simple models. In this chapter we will consider several problems of the motion of a "test particle" in a 2-d potential landscape with several minima. A "test" particle (further simply denoted by TP) is a particle that plays a special role in the dynamics, it differs from all the others. In Chapters 1-2 we considered already examples of the dynamics of reacting particles in a bistable potential. Here study a physical model of a 2-d bistable potential in more detail. We aim to study the motion of a test particle with a relatively large mass, which is much larger than the masses of the surrounding molecules (e.g., substrate and solvent molecules). Thus, we can use the model of Langevin equations introduced in Chapters 1-2. The model of a 2-d potential that we are going to investigate, exhibits two or three minima (see Fig.3.1) and we include the case that the parameters of this potential are time-dependent. Several standard methods of classical stochastic mechanics will be used for the studies of the following problems which are relevant for the molecular dynamics of proteins: 1) stochastization of vibrations (complication of the vibrational spectrum, changes of Lyapunov index, appearance of the selected frequencies; peculiarities of distribution of the vibrational amplitudes); 2) lifetime of a particle in the area of the minimum of the potential field under the action of noise and friction (theory and numerical experiment, the case of Fermi resonance will be discussed in Chapter 6); 3) motion of a particle in a relief with a "bypass" through the third minimum; 4) the influence of the periodic changes of the parameters of potential relieves on these processes. The problem of penetration of a particle of a certain shape and finite dimensions into a potential well with narrow entrance that plays a special role will be considered in chapter 7. 103
104
Motion of test particles in a 2-d potential
landscape
We restrict ourselves here to classical approximations. Let us consider the equations of motion of a test particle (TP) in the potential field U(x,y) under the action of noise and friction (hereafter all the variables are dimensionless; the mass was assumed to be equal to unity):
Here x,y are the Cartesian coordinates; t is time; T is the noise amplitude (corresponding to a temperature); £1,^2 are random quantities. In our simulations we have made the assumptions that they are uniformly distributed in the interval [-1,1]. Further h is the friction coefficient (we note that in chapter 1 we denoted this quantity by 7). The potential is defined by U = Uo (arctg (ri — b) — arctg ( n + b) + arctg (r2 — b) — arctg (r2 + b)) + cr2. (3.2) In some cases a Lennard-Jones potential was added:
(3-3) Here we used the abbreviations ri = yky2
+ {x-g)2,
r = y/ky2 + x2,
r2 = y/'ky2 + (x + g)2,
rz = \J(x - xc)2 + (y -
(3.4)
yc)2,
The quantities Uo,c,b,g,xc,yc,d,ro,k are constants. The sum of the potentials (3.2) and (3.3) has three minima in the (x,y) plane. Let us comment on the representation of the potential (3.2). In the earlier works devoted to the problem of motion of a TP in ID potential landscape with several minima the potential function U was considered as fourth order polynomial: U (x) = ctox4 + a\xz + a2x2 + 03a; + 04. However, in this case the variation of the parameters leads to rather limited changes in the shape of the potential landscape (relative positions of the wells, their depths, the heights of the barriers). In addition, the parameters of one potential well influence the parameters of the others. In the case when each well is described by its own polynomial one has to deal with the problem of discontinuity. These problems are even more severe in the case of 2D potential relieves with several minima. Choosing the potential we make assumptions as follows: the potential landscape must have two minima (two potential wells) with a smooth barrier between them;
Formulation
of the mathematical
model
105
Fig. 3.1 2D potential landscape defined by eq.(3.2) (parameters of the potential field: U = 20/7r; 6 = 5; c = 0.05; g = 6.5; k = 1).
Fig. 3.2
2D potential landscape defined by eq.(7.3) and eq.(3.3) (parameters of the potential
field: U = 20/TT; b - 5; c = 0.05; g = 6.5; d = 100; r 0 = 10; xc = 0; yc = 20; k = 1)
potential wells must be the surfaces of rotation; there must be no influence of one potential well on the other; the variation of parameters of the function U (x, y) must provide the possibility of sharp changes in the shape of the potential landscape. In the vicinity of minima the potential (3.2) coincides with Morse potential with an accuracy of the terms of the third order of smallness. In the infinity the potential is close to Lennard Jones potential. Figures 3.1 and 3.2 present the topograms and 3D patterns of the potential landscape (3.2) and the sums of the potentials (3.2) and (3.3). The values of the constants are given in Figure captions. Such a potential simplifies the consideration of the particular problems. The real 2D and 3D potential wells are formed by summation of the individual potentials of atoms in protein molecules (Weiner et al., 1984).
106
3.2
Motion of test particles in a 2-d potential
landscape
Lyapunov spectra for the conservative system. Toda area for the landscape with two minima
The possible areas of stochastization for different potential relieves for the conservative case can be calculated by the Toda method (Toda, 1974). This method allows one to make an approximate analysis of the stochastization of vibrations in the system. Assume that h = 0, then (3.1) can be split into four differential equations of the first order:
* E - yx dt
'
^k__^ dt
*V-V
dx ' dt ~
dV y
'
v -
dt
dU
dy'
Consider two trajectories that are infinitely close at a certain instant and study the evolution of the quantities Sx and 5y (deviation of the trajectories) on the basis of the linearized system: ^ 1 - W dt
^
d5y
d5Vy
-^
._,
= 5v
dt
»'^r
= - — 2 6x - d2U 5 • dx dx dy d2U .
=
d2U .
6x
-dx-dy -wSy-
The characteristic equation for this system is: A4 + (a + c) A2 - (b2 - ac) = 0, fp'Tj
ri2TJ
f)2JT
2
where a = ^ 5 - , b = X jL , c = ^-TJ- ; its discriminant D = (a — c) + 462 is always positive, i.e. the eigenvalues of the characteristic equation are always real. Consider three cases: 1) all eigenvalues are real which corresponds to two positive and two negative eigenvalues; 2) two eigenvalues are real (one positive an one negative) and two are imaginary; 3) all eigenvalues are imaginary. Stochastization is possible only for the areas where at least one eigenvalue is positive. The boundary of such areas is given by an equation: ac — b2. The process of TP stochastization in a conservative system was studied for different initial conditions on the basis of two-well model (3.2) in the absence of noise and friction (h = 0, £1 = £2 = 0 in the system of equations (3.1)). The question arises if TP gets inside Toda area. In calculations the numerical values of the coefficients were as follows (unless specified otherwise): UQ = 20/7r; 6 = 5; c = 0,05; g = 6,5; k = 1. Thus, the potential field given by (3.2) was symmetrical relative to Y axis of the Cartesian coordinates. The points of minima belong to X axis and the lowest point of the potential barrier is in the origin.
Lyapunov spectra and Toda area
107
We studied the process of stochastization of TP vibrations under different initial conditions. The initial position of the particle at the potential surface was determined by the coordinates xo>2/o- We considered several variants of the initial conditions. In all cases x$ = —5 and yo was varied from 0 to 5. At t = 0 the initial velocity of the particle was Vo (vxo,vy0) (here vx,vy are the components of the velocity vector). In all the cases |V| = 5. This initial velocity is high enough to ensure overcoming the potential barrier between the wells. The system of differential equations (3.1) was integrated numerically according to the fourth order Runge-Kutta method with automatic choice of the optimum step of integration. Figure 3.3 shows the Toda area for the potential landscape defined by eq.(3.2) and represented in Fig. 3.1.
•20
-10
0
10
20
Fig. 3.3 Topogram of the potential surface (solid lines) in the area of the possible stochastization (shaded). All roots of the characteristic equation are real in the area 1; two of them are real and two are imaginary in the area 2.
The following results were obtained. Depending on the initial position and the initial velocity the particle can either overcome the potential barrier (and, hence, pass the Toda area) and appear in the vicinity of the other minimum, or not over-
108
Motion of test particles in a 2-d potential
landscape
come it. Initially the particle is outside the Toda area, however, it can get inside it. If one transition occurs the vibrations become stochastic (cease to be quasiharmonic). Changes in the vibrational spectrum (disappearance of the typical isolated peaks always present in the spectral curves in the case of harmonic and quasi-harmonic vibrations) can be an indication of stochastization. However, the time-dependence of the Lyapunov index clearly shows if the stochastization takes place (tendency to zero for the harmonic vibrations and tendency to constant for the stochastic vibrations). The possibility of the overbarrier transition depends both on the initial position of the particle and on the direction of its initial velocity. Note that if the trajectory passes through the Toda area the vibrations become stochastic and the particle goes to another minimum. However, this is not the general rule. Table shows the variants of the initial conditions and the results of numerical studies.
1 2 3 4 5 6 7 8 9 10 11 12 13
x0 -5 -5 -5 -5 -5 -5 -5 -5 -5 -5 -5 -5 -5
VxO
0 0 0 0 0 0 -5 -5 -5 -5 -5 -3 -4
Vo -5 -4 -3 -2 -1 0 -4 -3 -2 1 0 -4 -4
vy0 -5 -5 -5 -5 -5 -5 0 0 0 0 0 -4 -3
stochastization yes no no no no no no no yes yes no yes yes
Note that there is no stochastization for the case 11, although the transition from one well into another really takes place (one-dimensional oscillations along X axis). In all the cases when the trajectory does not touch Toda area the stochastization is not observed. However, there are some opposite cases when the trajectory touches the Toda area but no stochastization takes place. In Figs. 3.4, 3.5, and 3.6 we present TP trajectories, spectra and time dependencies of the Lyapunov index for the cases when there is no stochastization. It follows from these results that the frequency of the natural vibrations in the potential field (3.2) varies from 0.7 to 1.1 depending on the initial position of the particle. Figure 3.7 shows the trajectory of a particle, spectrum, and time dependence of the Lyapunov index for the case of stochastic vibrations (spectrum is complicated,
109
Lyapunov spectra and Toda area yd)
1
^ J 0.50
'wj
8.75
U.
1.00
^ 1.25
CO
1.50
b)
1 |A % ^WA W V M ^ VAA^IW
0
500
1000
""^AJ^^,
7500
Cl
Fig. 3.4 a) T P phase trajectory and the equipotential lines; b) a spectrum of T P oscillations along the X-axis; and c) Lyapunov exponents for the initial conditions: xo = —5; j/o = —4; vx = 0; vv = - 5 .
Lyapunov index does not tend to zero, the particle travels from one minimum to the other, and the trajectory passes through the Toda area). In the case when we deal with the "protected" degrees of freedom in which the damping is low due to the absence of the direct interaction with water molecules (Netrebko et al., 1991), the stochastization in 2D and 3D systems takes place in a way similar to that in Sinai billiard. The vibrational spectra become more complicated. That is why the "isolation" of a degree of freedom becomes weaker due to possible resonance interactions with the surrounding clusters and atoms. On the other hand, the motion along special "selected" trajectories is possible under certain initial conditions (e.g., in the case of substrate binding in the active site). The purpose of the further studies is to reveal which motions facilitate the catalytic act. Stochastization in conservative systems (clusters) consisting of several atoms is extensively studied in the recent years (Stratonovich, 1995; Stratonovich, Chichig-
Motion of test particles in a 2-d potential
110
-12
-9
-6
-3
0
0.50
0.75
landscape
1.00
a.
1.15
1.50
b.
X
K^ 0
**—••*-„.
1000
2000
3000
C.
Fig. 3.5 a) T P phase trajectory and the equipotential lines; b) a spectrum of T P oscillations along the i-axis; and c) Lyapunov exponents for the initial conditions: XQ = —5; j/o = — 3; vx = 0; vy = - 5 .
ina, 1996). The decay of such clusters is possible if the energy is concentrated in one of the atoms. Another similar problem is discussed in Chapter 4: how the molecular dissociation takes place and what is the distribution of energy between the interacting particles if one of them is "soft".
3.3
3.3.1
Stratonovich method of calculating escape times in the chaotic regime and some applications. Dynamic model of the cluster dissociation The role of a dynamic
theory of cluster
dissociation
Consider the process of a cluster dissociation in a metastable state. We use the word "cluster" but not, e.g. molecule, in order to stress that the interaction between the
Stratonovich
• 1!
-9
-6
-1
method
0.50
0
111
0.75
/.00
a)
1.25
1.50
b) X 0.25 -i
1
1
1000
2000
1
0.20
0.J5
0.10
0.05 - +
0.00 -I 0
I
^^*H
1 ' 5000
C)
Fig. 3.6 a) T P phase trajectory and the equipotential lines; b) a spectrum of T P oscillations along the X-axis; and c) Lyapunov exponents for the initial conditions: xo = —5; yo = —3; vx = —5; vy = 0 .
atoms is described within the frames of the classical theory. The results of the diffusion Markovian theory of cluster dissociation (Stratonovich, 1963) agree well with the experimental data. According to this theory the action of noise is described by introducing a random force into the dynamic equations - the so-called Langevin source of delta-correlated (i.e. maximally random) force. Such an approach yields an exponentially decaying probability of the cluster "nondissociation". Mean dissociation time rav is the time in which the probability decreases e times. However, the shortcoming of this theory lies in the fact that the use of Langevin forces is not well substantiated. If they are considered as external random action on the dissociating cluster, its dissociation is not spontaneous. Another interpretation of the forces £ a (t) presumes that the variables x are the part of the dynamic variables of the molecule and £Q depends on the other variables. However, the forces £ a (t) can not be exactly delta-correlated under the finite number of the de-
Motion of test particles in a 2-d potential
112
landscape
y(t)
\\ 0
w
-5
m-12
-6
0
12
6
m
0.50
Iwk, 1.00
CO
l.SO
b
a
I 0.12
\
AA
V
1
V- V\M
V-A
r 200
300
400
$00
Fig. 3.7 a) T P phase trajectory and the equipotential lines; b) a spectrum of T P oscillations along the X-axis; and c) Lyapunov exponents for the initial conditions: XQ = —5; j/o = —2; vx = —5; Vy=0.
grees of freedom of the cluster. In addition, this interpretation implies also the consideration of the equations for the other variables and the theory must be much more complicated. Thus, the purely dynamic theory of spontaneous dissociation became necessary and it was proposed in (Stratonovich, 1995). This theory does not take into account external random interactions and uses nonfluctuative dynamic equations:
xa
—j
a
\xj.
In this case the dissociation process is not exponential any more. It resembles an exponential one if a stable system related to the initial one exhibits a dynamic chaos and if an inequality \rav ~S> 1 is met (long lifetime), where A is the maximum Lyapunov index of the stable system that characterises the rate of divergence of trajectories starting from the points that are close to each other. Inequality \rav ^> 1 means that although the cluster is unstable its lifetime is rather large and the dynamic process in it does not differ strongly from the dynamic process in an absolutely stable cluster. Thus, we define a relationship be-
Stratonovich
method
113
tween an unstable (metastable) cluster and the stable cluster with slightly changed interaction potential. It allows one to introduce microcanonical distribution for the description of the statistical properties of such a system. It is evident t h a t the mentioned conditions are met if the energy of the system is slightly larger t h a n the maximum energy of the stable state. T h e thing is t h a t the possibility of dissociation makes the thermodynamic system opened which prohibits the application of the microcanonical distribution.
3.3.2
The simplest
dissociation
model
Consider the simplest and most descriptive model of the cluster t h a t illustrates the main ideas of the dynamic approach. In this model of t h e cluster the in-plane motion is described. All t h e atoms apart from the given a t o m are assumed t o b e stationary. If the interaction between the atoms is reduced t o the repulsion between rigid spheres, the given a t o m can be considered as a mass point, and the radii of t h e other atoms are accordingly increased two times. T h e a t t r a c t i o n of t h e a t o m s is replaced by a condition of nondeparture (in the case of a stable molecule) from a square LMNO with a side I. For further simplification assume t h a t there is only one stationary a t o m with an effective radius r\ apart from the mobile subject atom. T h u s we obtain the well known Sinai billiard. Finally, for simplicity we assume t h a t a periodic continuation is carried out beyond the lateral walls OL a n d NM, i.e., the square is bent into a cylinder by identifying the segment OL with NM. T h e molecule is unstable if the mobile mass point can escape through the opening F G with the length IQ in the wall O M (a model of dissociation) (Fig. 3.8.). Sinai demonstrated the existence of dynamic chaos in such a billiard (Sinai, 1963). T h e condition of fast (in comparison with the cluster lifetime) establishing of almost equilibrium distribution of probabilities is met due to the smallness of the opening l0/l and (p, where angle
Wst(q,tp)
= 271"-1 ( / 2 - 7 r r i ) ~
(3-5)
R is the direct product of the area of the interior of the square (minus the
114
Motion of test particles in a 2-d potential
Fig. 3.8
landscape
Billiard with a small opening.
area of the interior of the circle) and the interval [0;27r). In other words, this is a rectangular parallelepiped (l,l,2n) with a cylindrical (radius r\ and length 2ir) hollow in it. This equation follows from the microcanonical distribution wst (q,p) = C05 ( -p2 - -pi j = Cop^S (p - p0), which indicates that the distribution in R should be uniform. When the opening FG is sufficiently small, the lifetime of the cluster is large and specified as
•K
where v = (0,1). Therefore, when (3.5) is taken into account, we have ir (I2 - irrl) hvo
(3.6)
It is seen that rav —> oo if l0 —>• 0, i.e. gets closer to the case of the stable state. The dissociation process in this model was calculated numerically for the values n/l = 0.2; l0/l = 0.034. We assume a uniform initial distribution in the area 0.45 < qi/l < 0.6, 0.73 < q^/l < 0.76, 0 <
Stratonovich
115
method
i.o
0,8
0,2
0,0 i
0
•
'
•
SO
•
100
i
i
190
i
i
i
200
i
230
i
i—
300
y
Fig. 3.9
Cluster decay.
The empirical distribution function FN {y) = P (T/rav > y) for the lifetime was obtained for N trajectories (Fig.3.9). It was approximated by the exponential function. The empirical mean lifetime r'av, which was obtained by direct averaging of the lifetimes, was VQT^/1 = 84, while (3.6) gives the value V0T^V/1 — 80.8. It can be assumed that the relative disparity between the above values gets smaller with decreasing length of the opening. The decay curve is nearly exponential because the system quickly forgets the initial conditions due to discontinuity of transformation of angles at the moment of impact. 3.3.3
The calculation namic theory
of the rate of cluster
dissociation
using
dy-
The dynamic method proposed earlier for investigating the escape from a potential well is elaborated for calculating the spontaneous decay constant of a cluster or a metastable molecule. The purely dynamic method uses the fact that the dynamic process is a fluctuation consequence of self-stochastization, i.e., dynamic chaos. If the escape (decay) mean time is much longer than the characteristic time in which a stationary or quasistationary probability distribution is established in the phase space, then it can be assumed that almost all the time the system exhibits a distribution close to an equilibrium distribution in a stable system obtained by slightly deforming the initial distribution. Here we assume that the equilibrium distribution in the deformed system represents a microcanonical distribution: w(q,v) = CS{T(v) +
U(q)-E),
where T (v) = T\ (p(v)) - is the kinetic energy expressed in terms of the velocity
116
Motion of test particles in a 2-d potential
landscape
v = q. It is equivalent to the distribution wx (q,p) — C\8 (H (q,p) — E), since the Jacobian of the transformation p = p (v) is constant and can be included into the normalization constant. A stationary distribution in the deformed system can be used to find the decay constant a of the initial system, if this constant is small. The constant a is the escape probability per unit time given by:
w2 (x) fa (x) dTa
o=
(3.7)
r
Here x = (q,p) or x = (q,v), fa (x) — x (x), T is a closed hypersurface in the a
phase space, T_ is the part of the surface that corresponds to the condition of escape from a region lying inside T: fadTa > 0. Consider dissociation of a cluster consisting of atoms with their own coordinates and velocities. Then (3.7) can be rewritten as:
a = / w (q,v)r](vjnj)vjnjdSodv,
(3-8)
So
where So is a closed hypersurface in the coordinate space, n^-is the unit vector of the outer normal to So, VjTij is the scalar product of multidimensional vectors, and 77 (y) = (1 + signy) /2. Introduction of the factor r\ (VJUJ) selecting the space points where £) • VjTij > 0, is analogous to selecting the exit part T_ of the surface T in (3.7), and VjJijdSodv corresponds to fadTa. Thus we define a hypersurface So- If one of the atoms goes beyond this surface, the cluster is assumed to be dissociated. Then the probability of such an escape per unit time is given by the product of the velocity vector and the outer normal to this surface integrated with the probability distribution over all the velocities and S0. It is natural to place the hypersurface So, the crossing of which by an image point symbolizes the cluster dissociation, on top of the potential barrier. Computer experiment [Stratonovich, Chichigina, 1996; Chichigina, 1997] used a cluster consisting of three equal atoms interacting by Lennard-Jones potentials. The calculated cluster mean lifetime agrees well with the theoretical value calculated using (3.8). It means that the closed system consisting of only three particles can be described by thermodynamic methods. Such a possibility is related to the dynamic instability of the motion.
Stratonovich
3.3.4
117
method
Mean time of escape from a potential of noise. Metastable approximation
well under the
action
Formula (3.8) can be used not only in the absence of the external noise but for any metastable state inside the potential well U(x) to which the probability distribution w (x, v) can be related. This method is much simpler than the solution of FockerPlanck equation. For example, Gibbs distribution can be used as w (x, v) in the case of the external noise characterised by the temperature T. Assume that at the edge of the potential well U = oo everywhere apart from the area in the coordinate space that corresponds to the exit where Uout — 0. Large escape time Tav, i.e. metastability of the state can be achieved due to smallness of the opening (exit). The smallness of the opening allows one to neglect its geometry which substantially simplifies the problem. Consider 2D motion in the potential well U (xi,x2) in the area Cl. The escape takes place if the particle crosses a small fragment of the length 2A that is perpendicular to Oxi axis. As we consider the first escape from the area f2, v\ > 0 and we can restrict ourselves to integration over only positive values of v\ and omit 77-function in (3.8). In this case the final expression for a is: OO
OO
A
2
/
1
a=
[ A [A f A I dx2Viexp C / / / I -00
0
- A
2 \
Uout+^t±^t\ Ty
^
( ^
'
where Uout = 0, and the normalizing constant for Gibbs distribution is determined by the relationship: OO
C=
OO
f dv2 f dVl f dx1dx2exp -00
(
U
^
x
* ) + ^
+
-2X j
(310)
0
Integrations for a and C can be divided into two parts: the first related tp the coordinate the second related to the velocities: a = a'a", where A
"hi
dx2 =
-
-A -A
C = / dx1dx2expl
*'
-n and also 2kBT mix The final expression for the decay constant is:
2
j ,
118
Motion of test particles in a 2-d potential
landscape
2A J2kBT where C" can be sometimes calculated analytically. For example, C = kBTn/n the symmetrical parabolic well U = K (X\ + x\) and
3.4
for
Test particle motion in a three-minima potential landscape
We studied the transition of a particle from one potential well into another under the influence of random forces. The particle is initially located in the origin (XQ = 2/0 — 0) corresponding to the minimum of the barrier. Assume that at certain moments (from 20 to 100 times during the period of free oscillations determined by the frequency spectrum - see Section 3.2) the X and Y components of the velocity change by the values vx£, vy£. Here £ is a random quantity uniformly distributed in the interval [—0.5,0.5], vx, vy are additional velocity amplitudes (it was assumed that vx = vy = 1, which is insufficient for overcoming the potential barrier if the particle starts from the potential minimum). Due to noise action (such a noise can be considered as "white") the system is nonconservative, and we have to introduce friction. Coefficient h in (3.1) was determined by a numerical experiment based on conservation of the mean total energy of the system during the time of experiment. In calculations h = 0.05. The Lennard-Jones potential (3.3) was superimposed on the potential field (3.2). The dependence of the mean time (averaged over 25 realizations) of the first ten transitions from one well to another (tlO) on the depth and characteristic radius of the potential ULJ was determined. We considered five variants:
N 1 2 3 4 5
d 0 100 50 100 50
ro 0 10 10 13 13
Figure 3.10 shows phase trajectories of a particle and topograms of the potential landscape for the cases 1-3. The transition from one well into another can be either direct (via the potential barrier between the wells (3.2)) or indirect (via the bypass of the potential (3.3)).
Test particle motion in a three-minima
potential
119
landscape
y(t)
m.n -
x(t)
• IS
-10
-5
0
5
10
IS
b)
f% • IS
Fig. 3.10 vx = l;vy
-10
-5
0
5
/
0
15
m
Phase trajectories of a T P influenced by noise and friction, a) h\ = 0.1; /12 = 0.1; = l\d = 0; r0 = 0; b) hi = 0.1; h2 = 0.1; vx = 1; vy = 1; d = 100; r 0 = 10;
In the second case the particle can stay at the bypass for a long time or come back to the well from which it started. It was demonstrated that the TP lifetimes in each well are nearly equal. The distortion of the potential (3.2) by the potential (3.3) is the stronger the larger is the characteristic radius and the relative depth at the fixed position of the center of the latter. This influences, in turn, the time often transitions and can both increase and decrease this time in comparison with the absence of the potential (3.3). Note that the depth of the potential is the decisive factor. Specifically, a relatively deep potential (3.3) can diminish substantially the potential barrier separating it from the wells (3.2) and, thus, facilitate the transition from one well into another via the bypass. However, if the bypass is rather far, the particle can stay there for a long time accumulating the energy for escape. Eventually the time of ten transitions increases substantially. On the other hand, shallow potential can not influence the potential barriers and the straight way becomes more preferable. Therefore, deep and wide (large value of ro) Lennard-Jones potential decreases (variant 4) and narrow one increases (variant 2) the time of the first ten transitions
120
Motion of test particles in a 2-d potential
landscape
in comparison with the case of the absence of the potential (3.3) (variant 1). Shallow wide and narrow potentials (cases 3 and 4) do not change the time substantially. In our opinion the problem of three minima is directly related to the proton transfer in the system of hydrogen bonds disturbed by substrate binding (see Chapter 7). In Chapter 7 we consider the motion of TPs of different shape in the potential relieves with several minima for a specific case of the ACE force field.
3.5
The problem of a test particle transition in the potential field with periodically changing parameters
Below we consider two cases of the periodic changes of the parameters of three-well and two-well potential relieves. a) Assume that the center of the Lennard- Jones potential oscillates harmonically along the Y axis so that yc = 20 + 3 sin(fct) and the particle is initially placed in the center of one of the two wells (4.2). We determined the dependence of the time when the particle leaves this well for the first time (moves from the centre of the well by the distance equal to half of the distance between well) on the oscillation frequency k. Figure 3.11 shows the results of calculations: the calculated time is plotted versus the parameter 2nk. In practical calculations we considered the times that were less than 1000 (calculations were terminated if by that time the particle was still in the same well). The curves exhibit two pronounced minima. (This result is predicted by the theoretical analysis of the resonance vibrational frequencies under the conditions of linearization.) If the frequency of the potential center oscillations is close to the frequencies of the free oscillations (see problem 2), the escape time is the shortest. The second minimum (longer escape times) corresponds to double free oscillation frequency (Fig. 3.11). Two intervals of resonance frequencies correspond to two mechanisms of excitation of TP vibrations: additive (with the frequency close to the natural one) and parametric (with the frequency close to double natural frequency). If we introduce noise the result does not change dramatically. We can find the same ranges of frequencies where the escape times are minimal. If we introduce only friction it appears that a) there is a threshold value of the friction coefficient under which the particle does not leave the well at any frequency from the considered range during the observation time; b) if the friction coefficient does not reach the threshold value the ranges of resonance frequencies become more narrow (the range of double resonance frequencies may disappear) and the corresponding times increase proportionally to the increase of the friction coefficient. It is the noise but not the variation of Lennard-Jones potential that influences the time of the first escape. (In the case of random noise the escape time is the mean time determined
Periodically changing
parameters
121
in
IT ;i
2nk
* 13
b
Fig. 3.11 Plots of the T P escape times versus the frequency of oscillations of Lennard-Jones potential at different noise intensities (vx,vy) and friction coefficients (hi,hi): a) vx = 0; vy = 0; hi = h2 = 0; b) vx = 0.1; vy = 0.1; hi = h2 = 0.01; c) vx = 0.2; vy = 0.2; hi = h2 = 0.015; d) vx = 0.5; vy = 0.5; hi = h2 = 0.025;
by means of 25 realizations of a random process. A scatter of values of these times was less then 10% for the frequencies close to resonance ones and increased far from them). b) The problem similar to the previous one was solved for the case of the oscillating potential field (3.2) in the absence of the potential (3.3). The initial position of the particle in the proximity of one of the minima was XQ — 6 + £, yo = £, (where £ is a random quantity uniformly distributed at the interval [—1,1]); the initial velocity was equal to zero. It was assumed that the centers of the potential wells oscillate in counterphase (one of the constants in (3.2) was oscillating as g — 7,5±sin (kt)). Thus, the value and direction of the force acting up on the particle varied in time and were determined by the current position of the particle. The time of the first escape (crossing Y axis) depending on the oscillation frequency was determined. The calculations at each value of the potential oscillation frequency used 100 random initial positions of the particle and the results were averaged. In this case we can select several ranges (not only two as in the previous case) of the resonance frequencies being aliquot to the natural one (see Fig. 3.12a). The escape
122
Motion of test particles in a 2-d potential
landscape
-—t—'
r TfT•
i. -1
i
\ r\ I L tf i .
as
l.s
is
3.5
ft5
/.5
2.5
3.5
b
Fig. 3.12 Plots of the escape times versus the frequency of oscillations of the centers of the potential wells: a) xo = —5; j/o = 0; vx = 0; vy = 0; hi = h2 = 0; b) 10 = — 5; yo = 0; vx = 0; Vy - 0; h\ = h2 = 0.01
time quadratically increases with increasing frequency. If friction and (or) noise are introduced the results are similar to those obtained in Section 3 (see Fig. 3.12b). Note that the escape times larger than 500 were not determined because of the limited computational time. Thus, the horizontal fragments of the curves in Fig. 3.12 do not correspond to the real escape times (there were no transitions during the computation time at the corresponding frequencies). Thus, the harmonic variation of the parameters of the potential landscape on both the frequency close to one of the T P natural frequencies and the frequencies close to the main parametric resonance result in abrupt (by two, two and a half orders of magnitude) decrease of the particle lifetime in the area of one of the minimums. The relative variation of the parameter or its amplitude (relative change of the distance between the minimums) is only 0.1 How can we relate these effects to the processes taking place in the active site of an enzyme? The vibrational frequency of the proton in H-bond between donor A and acceptor B is 7 • 10 13 — 1014Hz (Sokolov, 1981) and depends on the distance between A and B atoms. The vibrational frequencies of large clusters that form the potential landscape in A-H-B system are much lower (less than 10 12 Hz). Thus, one can hardly expect any resonances. On the other hand, time T can get shorter due to slow decrease in the distance between the minima and the corresponding lowering of the potential barrier in the potential landscape. The problem of resonance with quantum-mechanical transitions is discussed in Chapter 8. In any case harmonic or quasi-harmonic parametric processes influence substantially the spectrum of proton oscillations (broadening of the spectrum). As the proton moves in 2D or 3D space, the spectrum of oscillations can be influenced also by the low-frequency modulations due to Fermi resonance. This possibility is mentioned by Sokolov (1981) and the corresponding mathematical models are presented in (Netrebko et al., 1994, 1996;
Periodically changing
parameters
123
Chikishev et al., 1996). On the other hand, the resonance phenomena can be expected for the vibrations of ligands bound by weak (e.g., hydrogen) bonds in the pocket of the active site. In particular, the natural frequencies of tryptophan bound by several H-bonds to the atoms of CT active site belong to the range 2 — 4 • 1012Hz. Therefore, the oscillations of groups of atoms and clusters that influence the AS potential field can be the sources of coloured additive or parametric noise. We mention that a similar situation holds for the calculation of conformational jumps of the angles $ and $ in helical peptide structures (see e.g. Fig. 6.2).
References O.A. Chichigina(1997): "Computer simulation of processes of spontaneous dissociation of 3-particle clusters", Vestnik Moskowsk. Univ., Ser. Fiz. 3. Fizika, Astronomiya 5, 6-9. A.Yu. Chikishev, W. Ebeling, A.V. Netrebko, N.V. Netrebko, Yu.M. Romanovsky, L. Schimansky-Geier (1996): "Stochastic cluster dynamics of macromolecules", Nonlinear Dynamics and Structures in Biology and Medicine: Optical and Laser Technologies, V. V. Tuchin, Editor, Proc. SPIE 3053, 54-70. N.V. Netrebko, Yu.M. Romanovsky, E.G. Shidlovskaya, V.M. Tereshko(1990): "Damping in the models for molecular dynamics", Proc.SPIE 1403, 512- 514. A. Netrebko, N. Netrebko, Yu. Romanovsky, Yu. Khurgin, E. Shidlovskaya (1994): "Complex modulation regimes and vibration stochastization in cluster dynamics models of macromolecules", Izv. Vuzov: Prikladnaya Nelineinaya Dinamika, 2, 2643. (In Russian) A. Netrebko, N. Netrebko, Yu. Romanovsky, Yu. Khurgin, W. Ebeling (1996): "Stochastic cluster dynamics of enzyme-substrate complex" (in Russian), Izv. Vuzov: Prikladnaya Nelineinaya Dinamika. 3, 53-64. Y. Sinai (1963): Dokl. Akad. Nauk SSSR 153, 1261-1264. N.D. Sokolov (ed.) (1981): "Hydrogen bond" (in Russian), Nauka, Moscow R.L. Stratonovich (1963): "Topics in the theory of random noise", Vol. I,II, Gordon & Breach, New York/London R.L. Stratonovich (1995): " On dynamical theory of spontaneous decay of complex molecules", JETP 8 1 , 729-735.
124
Motion of test particles in a 2-d potential
landscape
R.L. Stratonovich, O.A. Chichigina (1996): " Dynamical calculation of the spontaneous decay constant of a cluster of identical atoms". Soviet Phys JETP 83, 708-715. M. Toda (1974): "Instability of trajectories of the lattice with cubic nonlinearity". Phys. Lett. A 48, 335-345. S. Weiner, P. Kollman, D. Case, U. Singh, C. Chio, G. Alagona, S. Profeta, Jr., P. Weiner (1984): "A new force field for molecular mechanical simulation of nucleic acids and proteins", J. Am. Chem. Soc., 106, 765-784.
Chapter 4
Microscopic simulations of activation and dissociation W. Ebeling, V. Yu. Podlipchuk, M.G. Sapeshinsky, and A.A. Valuev 4.1
Discussion of the Heat Bath Model
The stochastic reaction theory as represented in brief in Chapters 1 — 2 concentrates mainly on the explanation of the high-energy events leading to overcoming the potential barriers between wells. Eyring, Kramers et al. developed several reaction theories interpreting thermal activation processes. For recent reviews of this problem see e.g. (Hanggi, Talkner, and Borkovec, 1991; Agudov and Malakhov, 1993; Talkner & Hanggi, 1995). The model developed by Kramers in 1940 represents the simplest model of the statistical reaction theory. As discussed already in Chapters 1 — 2 it is based on special solutions of the Fokker-Planck equation for the reacting molecules or on the corresponding Langevin equation with a white noise source. The present chapter aims to investigate reacting sites on a more microscopic basis. We will study reacting molecules which are imbedded into an atomistic fluid heat bath. We start with the hypothesis that the effects of hard collisions which act on the site may have - under certain conditions - a drastic effect on transition rates. In this way we want to make another step from a phenomenological theory of reactions at active site of complex molecules in solutions to a microscopic theory. In this way we want to contribute to the development of an appropriate statistical theory of reactive events at complex molecules. In some sense the microscopic approach which models the surrounding of an active center by an ensemble of discrete molecules instead by a heat bath is equivalent to introducing a coloured noise source into the Langevin equation instead of a white noise source. This corresponds to the fact that the spectrum of the realistic molecular forces acting on the reacting site is not white but has a much more complicated spectrum (Jenssen and Ebeling, 2000). Thus, before coming to the microscopic study we formulate here again our problem in the context of a coloured noise heat bath. According to Kramers phenomenological approach, which was explained in some detail in Chapters 1 — 2, the dynamics of the reacting molecule is modelled by a Langevin equation for the active site
125
126
Microscopic simulations
d2q
dq
of activation and
dU (q)
dissociation
/—- , ,
,
x
where m is the mass of the active particle (in the following m — 1), 7 is the effective friction constant (having the dimension of a frequency), D the diffusion constant. In Kramers original approach £(£) corresponds to (^-correlated noise (white noise). The potential U(q) is assumed to have two minima, therefore the system is bistable. The corresponding Fokker-Planck equation was solved by Kramers with a special boundary condition modelling the transition over a potential barrier. As we have shown in Chapters 1 — 2, the result for the log of the transition rate is (assuming moderate or large values of the friction)
ln* = l n , - ^ ,
(4.2)
where following Kramers (for moderate and for large 7) the frequency v is given by -Y 2
+
\
V 2
T -0
"i
ss-
(")
Here u>o is the angular frequency inside the minimum, where the transition starts and Wfc is the angular frequency at the potential maximum (the transition state). Equations (4.2-4.3) are in agreement with the Arrhenius law. In the limit of very weak friction (7 < < wo) the theory is to be generalized including energy diffusion (Straub & Berne, 1986; Hanggi, Talkner, & Borkovec, 1991). In the generalized Kramers theory the rate is given by a different expression (see chapter 1) since at small 7 the transition process is controlled by energy diffusion (Straub & Berne, 1986). As a consequence log A; shows in addititon to the linear Arrhenius term also a weak logarithmic dependence on AU /kg'T. According to chapter 1 we find in some approximation (additivity of the characteristic times) which includes energy diffusion as well as as quantum effects ,
/2TT7
2kBT
h \
AU
/A A.
We see that energy diffusion which comes into play at small values of the friction and small values of the Arrhenius term AU/ksT leads to deviations of the Arrhenius plot from a straight line what is demonstrated in the following figure. We see in Fig. 4.1 that the generalized Kramers theory (including energy diffusion) shows at small 7 an Arrhenius plot deviating from a straight line. At
7/wo = 0.1 we find a strongly expressed minimum of the transition time, at
Discussion
of the Heat Bath Model
4.0 6.0 (Delta U / k T)
Fig. 4.1 Arrhenius plot for the transition time log(r) = — log(fe) according to the theoretical expression eq.(4.4) including stochastic transitions and energy diffusion for 3 different values of the friction (7 = 0.1; 1.0; 10;w 0 = ub = l;h = 0).
7/wo = 1-0 this minimum is still weakly expressed and only at large friction 7/wo = 10 we find a straight line. Summarizing so far these results we may state that energy diffusion effects, observed at small friction in the generalized Kramers theory, already lead to deviations from the Arrhenius behaviour. Now we have to look for other effects. The crucial point in the Kramers theory of the transition rates, and its generalizations, is the assumption about a white noise character of the forces acting on the active site. In real molecular systems, and in particular under the influence of nonlinear collision effects, the 'real noise' may have a very complicated spectrum. In the next section we will show in detail by MD-simulations that the spectrum of the forces acting on a molecule in a bath has a maximum at finite frequencies and shows some similarity to coloured noise (harmonic noise). Here we will discuss only the question what are the consequences of introducing coloured noise sources into the phenomenological theory. Simple models of systems with harmonic noise were studied by several workers [Straub k Berne, 1986; Ebeling k Schimansky-Geier, 1989]. Straub and Berne studied the following generalization of (4.1) [Straub k Berne, 1986]:
m
cPq dt2
dU (q) dq
mey (t).
(4.5)
128
Microscopic simulations
of activation
and
dissociation
Here it is assumed that the bath acts directly only on the nonreactive auxiliary variable y. Schimansky-Geier, Ziilicke and Hesse investigated a similar model neglecting the term of order m/M [Schimansky-Geier & Ziilicke, 1990; Hesse & SchimanskyGeier, 1991]. It is easy to show, that the spectrum of the y-coordinate is harmonic [Schimansky-Geier & Ziilicke, 1990]
s
m M =
2Dft 2 „2 , . , , 2 , w r + (w
^ 22) -n
(4-7)
The peak of the spectral intensity is located at
ujmax = V« 2 - r 2 /2,
(4.8)
2DQ.A Syy {Umax) = p2(Q2 _ ( l / 4 ) p 2 ) '
^4'9^
and its height is
We consider now the oscillations of q around one minimum of the potential U(q), where it can be approximated by a harmonic force
U(q) = \u% (q ~ qof •
(4.10)
As shown by Hesse and Schimansky, the energy distribution for the oscillations around qo is approximately given by [Hesse & Schimansky-Geier, 1991]
P (E) - Z-^exp ( - ^ f - )
(4- 11 )
with the effective temperature
Teff ~ Syy (wo) •
(4.12)
We see that in this approach the effective temperature is no more a constant as for the white noise case but through WQ and Syy a function of the potential parameters itself. For the effective temperature and for the mean value of the energy, a resonance at uJma.x— wo- is found. The resonance condition is
Discussion
of the Heat Bath Model
n2-T2/2
= ujl.
129
(4.13)
In the vicinity of resonance large amplitudes and therefore possibly also enhanced transitions are observed. In this way it was shown by Schimansky et al. (1990, 1991), that coloured noise may drastically change the transition rates. Strictly speaking the Schimansky-Geier model does not apply to a molecule imbedded into a thermal bath in equilibrium, since the model does not obey detailed balance. Anyhow we may expect that this easily tractable model will reflect the qualitative properties of the stochastic forces in more realistic systems. We mention that the transition rates in real molecular systems are also influenced by the changes of the average forces, which are due to the equilibrium correlations. As well known from the modern theory of liquids (in particular we have in mind the BBGKYtheory), the interaction potentials are in dense thermal systems to be replaced by potentials of average forces, which are defined by the equilibrium distribution functions. Since the potential landscapes are changed by this equilibrium effect, this is another reason to expect that for real molecular systems the Arrhenius factor AU/kgT is to be replaced by an effective Arrhenius factor (AU/kBT)eff. Evidently the physical assumptions leading to a Fokker-Planck equation are rather severe, since in real systems the forces acting on a molecule do not correspond to the Langevin model and further we know from Gibbs statistics that the potential energy of a molecule must not obey an individual Boltzmann distribution. Having in mind that in spite of the great success of the classical reaction theory, several phenomena remain unexplained [Popielawski, 1989; Frauenfelder & Wolynes], a more careful study of the microscopic forces acting at special molecules seems to be necessary. The standard form of the equipartition theorem of statistical mechanics is saying that any quadratic contribution AH — aq2 to the Hamiltonian leads in thermal equilibrium to a contribution ksT/2 to the internal energy. This follows immediately from the Boltzmann distribution
P (q) = const exp ( —j^-f ) •
( 4 - 14 )
In other words, there is no way to accumulate in average more energy than ksT/2 on a linear degree of freedom. However this is not true for nonlinear excitations. Therefore we have to look for nonlinear effects capable to accumulate energy at certain sites. In the fifth chapter we will study in detail the nonlinear excitations and the statistical thermodynamics of nonlinear 1-d systems with Toda interactions. We will analyze in the following analytically and numerically the basic nonlinear dynamical effects, and in particular the soliton fusion, which is one of the effects supporting energy localization at definite sites.
130
Microscopic simulations
of activation
and
dissociation
In order to study the effects of nonlinear excitations, correlations and "real noise" in microscopic models, we simulated microscopic collisions on an active site by methods of molecular dynamics. The system we are going to study is a Hamiltonian system; the active site is coupled to a thermal bath consisting of molecules in equilibrium. For a special case, if the bath consists of an infinite set of harmonic oscillators, the method of Zwanzig allows the derivation of a generalized Langevin equation [Hanggi, Talkner & Borkovec, 1991]. This new Langevin equation obeys a fluctuation-dissipation theorem. The random force is stationary and Gaussian, it contains harmonic contributions as well as the memory kernel of the friction term [Hanggi, Talkner & Borkovec, 1991]. For systems with a bath consisting of hard molecules the situation is much more complicated and a complete theoretical treatment is still out of range. In our empirical MD-study we will concentrate on a special physical effect which might be specific for hard interactions. As pointed out already, we have observed in earlier work rather long tails in the energy distribution due to hard excitations and a harmonic structure of the force-spectrum [Ebeling & Podlipchuk, 1996]. These findings suggested, that under special conditions a considerable reaction rate enhancement could be possible. For simplicity we will restrict ourselves to the investigation of classical 2-d models of molecular systems. The rather simple model which we aim to study here is a system consisting of one "soft" molecule simulating the reactive site. This reactive molecule consists either of one soft atom, possessing two stable states in an external field, or of two soft atoms bound together by strong chemical bonds. The reactive molecule is imbedded into a thermal bath consisting of hard molecules forming a solvent. Note that we pay special attention to the areas of energy localization and highenergy "tails" of the distributions. It is assumed that the complicated reactions, such as DNA denaturation [Dauxois et al., 1993] and the enzymatic activity [Ebeling et al., 1989; Romanovsky et al., 1994], are related to the nonlinear excitations leading to energy localization in the special reaction centers. We studied the problem of elementary excitations in biomolecules and their possible role in the activation processes based on the simple models [Ebeling et al., 1989; Romanovsky et al., 1994].
4.2
Molecular dynamics of transitions between potential wells
In order to model a monomolecular reaction of Kramers-type we assume a soft 2 — d molecule (the active site) imbedded into a molecular bath of rather hard 2 — d molecules. We further assume, that the active molecule is moving in an external bistable 2 — d potential. Our main task is to study the stochastic transitions of the active molecule between the two wells in the 2 — d potential landscape (see Fig. 1.1 in Chapter 1). The external potential acting on the active molecule is denned as a 2 — d generalization of the Kramers potential
Molecular dynamics
of transitions
V(x,y,z)
between potential
= ex(x4-2x2)
131
wells
+ eyy2.
(4.15)
We see that ex corresponds to the height of the barrier between the wells and ey to the width of the cleft. In our model the active molecule which is wandering between the two wells is imbedded into a 2 — d molecular system consisting of N relatively hard compressible disks. In other words we have a kind of 2 — d solution of an active molecule in a solvent. The active molecule is modelled as a soft disk. The disks interact with each other by collisions due to thermal motions. The interaction energies are described in an appropriate way by finite-range LennardJones potentials [Hafskjold & Ikehoshi, 1995; Ebeling et al., 1995]. These potentials are obtained by a modification of the standard Lennard-Jones potentials: y L (r) = 4 e ( ( ^ ) 1 2 - ( ^ ) 6 ) .
(4.16)
Here e is the debth of the attraction well, this quantity will be used in the following as the energy unit. Further erj, will be used as the length unit. The Lennard-Jones potential is infinite-range. For the purpose of molecular dynamics, it is quite useful to work with potentials of finite range [Ebeling et al., 1995]. We have used as in our previous work for the hard interaction between solvent molecules the potential:
Vhh(r)
=
Vhh(r)
= 0ifr>1.5ahh.
\
Ahhe((^r-l)exP((^-h-A,
r
/
\ Ohh
*
'
(4.17)
Further we assumed that the interactions of the active molecule with the surrounding bath molecules are rather weak. This weak interaction is described by a potential with a weak r~2 singularity: Vwh(r) = Vwh{r)
\
= Oifr>
Awhe((^f-l)exp((—-h-i), r / V crwh 2, / 1.5awh.
(4.18)
The shape of the potential (4.18) is shown in Fig. 4.2 in comparison with a standard Lennard-Jones potential. In the following we shall call these special potentials FLJ-potentials ("finite range Lennard-Jones-potentials"). The parameters were chosen in such a way that the minimum is at exactly the same place
V(r) = - e
at r = 2*aL,
(4.19)
as for the Lennard-Jones case. This condition corresponds to the choice A^ = 28.05, <Jhh — 1.028
132 2 Morse potential 1.5
Lennard-Jones potential
FRJ, n=2
1
FRJ, n=8
u °5 o -0.5 -1 0.9
1
1.1
1.2
1.3
1.4
1.5
Fig. 4.2 Shape of the Morse potential, finite-range potentials for solvent-solvent interactions (n=8), for solut-solvent interactions (n=2) and for the corresponding Lennard-Jones interaction.
The potential parameters for the interaction of soft with hard molecules are chosen as, Awh — 69.75, crwh — 1.0024o,i . This choice guarantees that the minimum is at the same place as the minimum of the potential given by eq. (4.19) (see Fig. 4.2). The potential (4.17-4.18 ) has a finite range with a cutoff at 3/2, at this distance the potential itself and its derivative disappear. The finite range is of some advantage in MD-calculations for a smaller number of molecules in a cell. In most of our simulations we considered N = 100 i.e. 99 hard and 1 soft molecules. The density p was chosen always in such a way that in the static equilibrium configuration the distances correspond to the minimum of the potential. In the further calculations we used the zero point of the Lennard-Jones potential (4.19) as length unit and the potential depth e of an argon molecule as the energy unit. The corresponding temperature unit is T = 119-ftT. In these units we studied 2 — d molecular systems with the density p = 0.7. The dimensionless temperatures T were varied in the range 5 < T < 375. For this case we carried out molecular dynamical simulations for 100 particles in a box with periodic boundary conditions. A typical trajectory of the active molecule is shown in Fig. 4.3. We observe irregular transitions between the two wells. After a sufficiently long run the average time between transitions was obtained. The result is shown in Fig. 4.4. We represented the log of the average time between transitions from one of the Kramers wells to the other one, as a function of the Arrhenius exponent (AU/T). The temperature was in all runs fixed at the value T = 5. The heigth of the the potential well was changed between AU = 10 and AU — 50. Since in logarithmic scale the average time in the Arrhenius and also in the first (large friction) Kramers approximation should be a straight line, we have shown that there are significant deviations from the linear behaviour. In some sense the shape of the curves reminds the behaviour of the generalized Kramers theory represented in Fig. 4.1. This would mean that the transition processes are not
Dynamics
of Recombination
Reactions
133
Fig. 4.3 Transitions between the 2 wells of the bistable Kramers potential caused by molecular collisions.
Fig. 4.4 The log of the average time (j/-axis) log(r) = — log(fc) between transitions from one of the Kramers wells to the other one, as a function of the Arrhenius exponent AU/ksT. Line 1 - MD simulation; line 2 - theoretical expression eqs.(4.2-4.3); 7 Ri uo ~ ^b] line 3 - theoretical expression eq.(4.4); 7 3> <*>o; the value for 7 = 12.3 was taken from the MD simulation.
overdampded as often assumed but operate in a moderate or small friction regime. Further we see from the results of the simulations that the effective temperature corresponding to the local slope depends on the potential itself. This is what we observed already for a harmonic noise. In other words, the realistic microscopic noise is similar to a coloured noise.
134
Microscopic simulations
of activation
and
dissociation
100-
50-
o-50-
-100-100
-50
0
50
IOC
Fig. 4.5 Typical trajectories in the space of the relative distances x\ — X2 and j/i — j/2 between two Morse molecules. The arrows show events which lead to a dissociation of the molecule.
4.3
Dissociation of Morse Molecules
Our second model is a 2 — d diatomic soft Morse-molecule imbedded into a 2 — d solution consisting of rather hard atoms. Similar problems for low pressure systems were studied by Borkovec and Berne (see Hanggi, Borkovec & Talknerl). We study here the case of systems with liquid densities and assume that the two atoms forming the molecule are bound together by the Morse potential VM(r) = e((exp(-6(r - aL • 2*)) - l ) 2 - l ) .
(4.20)
The interactions between the atoms forming the molecule and the solvent molecule are described in the same way as above by the FLJ-potential (4.18). The molecular forces between the hard solvent molecules were approximated again by the FLJ-potential (4.17). The shape of the potentials is shown in Fig. 4.2 (Morse potential = dotted line, the finite-range potential for hard repulsion = full line, for the weak repulsion = dash-dotted lines). For comparision also the corresponding Lennard-Jones interaction is represented (by another dashed line). For this model we carried out again molecular dynamical simulations for N = 100. particles in a box with periodic boundary conditions. In Fig. 4.5 several typical trajectories in the space of the relative distances Xi — x-i and j/i — j/2 between two Morse molecules are shown. The arrows correspond to events which lead to a dissociation of the molecule. By averaging over many dissociation events we obtained the mean dissociation times, which are represented in Fig. 4.6. Here the logarithm of the dissociation time is represented as a function of the Arrhenius exponent. The full line corresponds to a fixed value of the potential barrier AU — 50 and the dashed line to fixed temperature T = 7.5 and variable potential barriers AU.
Dynamics
of Recombination
135
Reactions
.3 I
1
1
1
1
1
1
1
0
1
2
3
4
5
6
7
Fig. 4.6 The log of the dissociation time for a Morse molecule as a function of the Arrhenius exponent. The full line corrsponds to a fixed value of the potential barrier AU = 50 and the dashed line to fixed temperature T = 7.5 and variable potential barriers AU. All energies and temperatures are given in the Argon-units used in this paper (i.e. T = 1 corresponds to 119 K).
Again we do not find an appropriate description by the Arrhenius law which would require that both curves are identical and have the slope one. Again we could interprete this by coloured noise effects. The microscopic picture we have here is more rich and allows a more detailled study. The price to pay is however, that microscopic simulations require a lot of computer time, so that the coloured noise picture seems to be a reasonable compromise.
4.4
Dynamics of Recombination Reactions
In order to model the fusion reaction of two molecules we defined a potential with a well inside a high wall VF(r)
=
^ e x p ( - ^ ) ( ( ^ - l ) e x VF{r) = 0
if
r > 1.5
aF.
P
( ( ^ - ^ ) , (4.21)
The parameters aoF,crF and AF are choosen in such a way that the minimum of the potential (4.21) has the same depth and coordinate as for Lennard-Jones potential and the hight of the potential is equal to a given value (see below). We notice some similarity with the potentials used by Henderson et al. (1995) for the description of association phenomena. These authors modelled association by an attracting Gaussian potential inside a Lennard-Jones core. Our model corresponds to an association where the atoms merge completely, this is what we call 'fusion'. The shape of our potential is shown in Fig. 4.7 for different values of the reaction
136
Microscopic simulations
of activation
and
dissociation
60 50 40 30 U
20 10 0 -10 0.2
0.4
0.6
r
0.8
1
1.2
Fig. 4.7 Potential model for a fusion reaction with different 10,20,30,40,50, again measured in the Argon units).
1.4
reaction barriers (AC/
=
Fig. 4.8 The log of the average fusion time (time for crossing the barrier). The lower full line shows the time for reaching the maximum. The upper full line shows the time up to the end of one oscillation inside the well. The dashed line shows the time up to the moment where 100 time-steps are performed. Shot-dashed line shows the average life-time of the soft molecule inside the potential well.
barriers (AU — 10,20,30,40,50). All simulations were started in the position where the two atoms are initially at a distance corresponding to the outer well. The temperature was fixed at the value T = 5, the heigth of the barrier was changed (AU = 10,20,30,40). The average time was recorded for the events:
Spectrum of atomistic
collisional
forces
137
(i) reaching the top of the barrier, (ii) reaching the inner well and finishing at least one oscillation inside the well, (iii) after 100 time-steps inside the inner well. We see that the criteria (ii) and (iii) are more or less equivalent and may be used for defining the characteristic reaction time. The log of the average fusion time (with respect to the criteria given above) was represented in Fig. 4.8. The lower full line corresponds to the condition (i) i.e. it shows the time for reaching the maximum. The upper full line shows the time corresponding to the condition (ii), i.e. it represents the time up to the end of one oscillation inside the well. The dashed line shows the time up tho the moment where 100 time-steps are performed corresponding to the condition (iii). Already a first inspection shows, that the curves show a larger deviation from the Arrhenius law which increases with increasing height of the potential barrier.
4.5
Spectrum of atomistic collisional forces
We have shown in the previous sections that the molecular collisions acting on a reacting molecule generate a kind of coloured noise. In order to make this picture more precise we will study in this section the time correlations by means of molecular dynamics simulations. In particular we will calculate within our models the correlation functions as coordinate-coordinate, velocity-velocity and force-force for soft molecules solved in a solvent consisting of hard molecules. We simulate the thermal equilibrium of one soft molecule with r~2 - repulsion imbedded into a bath of molecules with r~ 8 - repulsion. Here the soft molecule should model an activation center in a solution formed of hard molecules. Following an earlier work [Ebeling & Podlipchuk] we will show that in thermal equilibrium the spectrum is similar to that of a coloured noise. Further we will show that a region of densities and temperatures exists, where the correlation functions of the solute molecules are quite different from those of the hard molecules of the solvent. Finally we will show that the diffusion coefficient of the soft molecules is much higher than that of the solvent molecules. In the previous sections we developed rather simple models of reactions in solutions which is rather well suited for MD-investigations. Here we aim to work out the model of coloured noise. Further we want to contribute to the study of the kinetic properties starting with the investigation of the time-correlation functions. Our present simulation refers to 3 — d molecules interacting via the finite-size Lennard-Jones model introduce above. For this case we carried out molecular dynamical simulations for 32-108 particles in a box with periodic boundary conditions. The molecular forces between hard molecules were approximated by the finite-range Lennard-Jones potentials introduced in the previous sections. We remember that the shape of this potential was shown in Fig. 4.2 for the exponent n = 8 in com-
138
Microscopic simulations
of activation and
dissociation
parison with a standard Lennard-Jones potential. As above we we shall call these special potentials finite - range Lennard-Jones-potentials (FLJ-potentials). The parameters were chosen in such a way that the minimum is at exactly the same place
Vmin = - c
at
r = 21/6L
(4.22)
as for the Lennard-Jones case. This condition corresponds to the choice n = 8, A = 28.05e, and a = 1.028L for the interaction of hard molecules with hard molecules. The potential parameters for the interaction of soft with hard molecules are chosen as n = 2, A = 69.75e, a = 1.0024L. This choice guarantees that the minimum is always at the same place. Our potentials have a finite range with a cutoff at 3L/2, at this distance the potential itself and its derivative disappear. The finite range is of some advantage in MD-calculations for a smaller number of molecules in a cell. In most of our simulations we considered N = 32 i.e. 31 hard and 1 soft molecules, a few calculations were made for N = 108. The density was chosen always in such a way that in the static equilibrium configuration the distances correspond to the minimum of the potential. This might correspond to a liquid-like phase. In the present calculations we used the zero point of the Lennard-Jones potential as length unit and the potential depth e as the energy unit. In these units we studied 3-dimensional molecular systems with the dimensionless densities p = n/L3 = 1, which correspond to the liquid state. The dimensionless temperatures T were in the simulations fixed at T = 10 which is not far above the characteristic temperature kT = 10/D (here D = 1,2,3 according to the dimension of the coordinate space) where the maximum of the energy localization effect is observed [Ebeling et al., 1995]. In the mentioned work we discussed the main physical excitations of the system which are expected to occur, as phonon- and soliton-like excitations. In Chapter 5 we will consider these effects again for Toda chains. In the following we will show that in the transition region also pecularities in the time correlation functions occur. For the simulations we have used an energyconserving algorithm [Norman et al., 1993; Ebeling et al., 1995, 1997]. In our earlier work [Ebeling et al., 1993] the distribution functions of the energy were calculated and studied especially in the region of the high-energy tails. We found that in thermal equilibrium there exists a characteristic region of temperatures and densities where energy is mainly concentrated on soft sites. Here we concentrate on the temperature-density region where long tails in the energy distribution were observed. For physical reasons one expects that high energy events could lead to special noise effects and kinetic properties. The result of our MD-simulations is demonstrated in Figs. 4.9-4.11. Figure 4.9 shows the time-dependence of the force-force correlation functions. What we observe is a typical well at about t — 0.5—1.5 in dimensionless followed by a slow sinusoidal damping. The spectral density (the Fourier transform) of the force-force correlations (Fig. 4.10) shows a peak at a frequency of about 25 (in
Spectrum of atomistic
collisional forces
139
Fig. 4.9 Autocorrelation functions force-force < f(t),f(t + At) > . 1-solute (potential-eq.(4.18)); 2-solvent (potential-eq.(4.17)). Two 2-dimensional Morse molecule. AU = 1; T = 1.8; p = 0.8.
dimensionless units). The behaviour of solute and solvent molecules is distinctly different. In Fig. 4.11 the time dependence of the mean square deviations ((Ar) 2 ) is represented as a function of time. According to Einstein we expect
<(r(t) - r(0)) 2 ) = 2dDt,
(4.23)
where D is the coefficient of diffusion of the solute molecules (or self-diffusion of the solvent molecules). We see that the diffusion coefficient of the solute molecules is by more than a factor 1.3 higher than that of the solvent molecules. Let us summarize: By MD-simulations for solutions of molecules with finitesize Lennard-Jones potentials we have considered a single soft molecule embedded into a surrounding, otherwise uniform hard system. It turns out that the soft molecule is able to trap and superpose narrow pulse-like excitations impinging from all directions within a characteristic excitation time. This energy superposition may lead not only to considerable concentrations of potential energy at the soft molecule as shown elsewhere (Ebeling et al, 1995) but also to larger diffusion coefficients and to a rather narrow-peaked power density of the force correlations.
140
Microscopic simulations
of activation and
dissociation
l«W
Fig. 4.10 Spectrum of force-force < / / > w time correlations. 1-solute (potential-eq.(4.18)); 2solvent (potential-eq.(4.17)). Two 2-dimensional Morse molecule .AC/ = 1; T = 1.8; p = 0.8.
4.6
Discussion of activation processes in an atomistic heat bath
We have shown in this chapter by molecular dynamical simulations most for 2—d and also for a few 3 — d systems, that larger deviations from a white noise spectrum and from an Arrhenius law occur. At first we observe some curvature of the Arrhenius slopes in particular at small values of the threshold similar as shown in Fig. 4.1 for the generalized Kramers theory. We may conclude from this that the transition processes in liquids are not overdamped processes. Further we observe that the slope of the Arrhenius curve is in general smaller than one (see Figs. 4.4, 4.6, 4.8) and shows in all regions some curvature. In other words, the effective Arrhenius factor (At7/fcfiT) e // if different from the original one and is not a constant but a potential-dependent function. This might be due to several reasons. At first we mention the effects of energy diffusion typical for small friction as demonstrated in Fig. 4.1. Another reason are deviations of the potential of average force from the vacuum potential or the deviations of the effective temperature from the real temperature in the solution. The latter is eventually caused by the fact, that the noise generated by molecular collisions is not of white noise type. I has been shown by Schimansky-Geier et al. (1990, 1991) that harmonic noise leads to an effective temperature, which is determined by the spectrum of the noise. In the previous section based on earlier work [Ebeling &Podlipchuk, 1996] we have investigated
Discussion
of activation processes in an atomistic heat bath
141
Fig. 4.11 The mean square deviation as a function of time. Line 1 - solvent (potential-eq.(4.17)); line 2 - solute (potential-eq.(4.21), AU = 30); line 3 - solute (Kramers potential, AU = 30); line 4 - solute (Morse potential, AU = 30); line 5 - solute (Morse potential, AU = 1); T = 1.8; p = 0.8. The mean slope of the curves corresponds to the effective diffusion constants.
the spectrum of the force-force correlations; we obtained a distribution which is similar to a harmonic noise distribution. This is a point which needs a more careful investigation including the study of the average forces, but in any case one reason for the the change of the slope could be the deviation from a white noise spectrum. Another observation which seems to be relevant is, that the log of the transition time is not a straight line and depends in a different way on AU and on T (see Figs. 4.4, 4.6 and 4.8). A conclusive interpretation of this quite complicated behaviour of the transition times is not possible so far. However our previous results on the energy distribution lead us to the hypothesis, that the reasons for the deviations from the Arrhenius-type behaviour are in part also due to the existence of hard (soliton-like) excitations in the system. In order to clarify this point, let us discuss in brief the main physical excitations of the molecules which are expected to occur in a dense system: In a thermal regime we have reasons to expect a whole spectrum of excitations, as phonon- and soliton-like excitations. In the case of a purely linear coupling we should find sinusoidal oscillations and waves, acoustical and optical phonons etc.. In nonlinear systems we will find also local excitations with local peaks of the energy density. From the theory of infinite chains of molecules with special interactions (as e.g. the Toda interactions, see for details Chapter 5), we know that there exist soliton-solutions. Solitonic excitations are absolutely stable local
142
Microscopic simulations
of activation and
dissociation
excitations. In realistic systems of molecules such excitations cannot persist already for purely mathematical reasons. However there might be hard localized excitations which remind a soliton behaviour. We consider such excitations as candidates for local energy spots and eventually as to be responsible for an enhancement of the rate of chemical reactions. At the present moment this interpretation is just a plausible hypothesis. However we have every reason to expect that further theoretical, numerical and experimental work will contribute to a solution of the problems, which were raised here. In Chapter 5 we will come back to this problem investigating in detail the properties of chains (rings) of masses with exponential repulsion (Toda chains).
References N.V. Agudov, A.N. Malakhov (1993): "Nonstationary diffusion through a piecewise continuous profile. Exact solutions and time characteristics", Izv. Vuzov Radiophysics 36, 148-165. T. Dauxois, M. Peyrard, A. Bishop (1993): "Dynamics and thermodynamics of a nonlinear model for DNA denaturation", Phys. Rev 47, 648-695. W. Ebeling, & L. Schimansky-Geier (1998): in "Noise in Nonlinear Dynamical Systems", (F. Moss, P.E.V. McClintock, eds.), Cambridge, Cambridge University Press. W. Ebeling, M. Jenssen & Yu. M. Romanovskii (1989): "100 years Arrhenius law and recent developments in reaction theory" , in: Irreversible Processes and Selforganization, eds. W. Ebeling and H. Ulbricht, Teubner, Leipzig, p. 7-24. W. Ebeling & V.Yu. Podlipchuk (1996): "Molecular Dynamics of Time - Correlations in Solutions", Z.physik.Chem. 193, 207-212. W. Ebeling, Yu. Romanovsky, Yu. Khurgin, A. Netrebko, N. Netrebko, E. Shidlovskaya (1994): "Complex regimes in the simple models of molecular dynamics of enzymes", Proc. SPIE 2370, 434-447. W. Ebeling, V. Podlipchuk & A.A. Valuev (1995): "Molecular Dynamics Simulation of the Activation of Soft Molecules Solved in Condensed Media", Physica A 217, 22-37. W. Ebeling, A. A. Valuev, V.Yu. Podlipchuk (1997): "Microscopic models and simulations of local activation processes", J. Molec. Liquids 73,74, 445-455. W. Ebeling, V.Yu. Podlipchuk, M.G. Sapeshinsky (1998): "Microscopic models
Discussion
of activation processes in an atomistic heat bath
143
and simulations of local activation processes", Int. J. Bifurc. & Chaos, 8, 755-760. W. Ebeling, M.G. Sapeshinsky (2001): "Microscopic models and simulations of local activation processes", Prog. XXV ICPIG, 255-256. W. Ebeling, M.G. Sapeshinsky (2001): "Microscopic models and simulations of local activation processes", Prog. IUVSTA IVC-15, AVS-48, ICSS-11, 160. B. Hafskjold, T. Ikehoshi (1995): "Partial specific quantities computed by nonequilibrium molecular dynamics", Fluid Phase Equilibria 104, 173-184. P. Hanggi, P. Talkner & M. Borkovec (1991): "Reaction-rate theory: fifty years after Kramers", Rev. Mod. Phys. 62, 251-341. J. Hesse, L. Schimansky-Geier (1991): "Inversion in harmonic noise driven bistable oscillators", Z. Phys. B 84, 467-470. D. Henderson, S. Sokolowski, A. Trokhimchuk (1995): "Association in a LennardJones fluid from a second-order Percus-Yevick equation", Phys. Rev. E 52, 32603262. N.D. Sokolov (ed.) (1981): "Hydrogen bond" (in Russian), Nauka, Moscow. M. Jenssen, W. Ebeling (2000): "Distribution functions and excitation spectra of Toda systems at intermediate temperatures", Physica D 141, 117-132. G.E. Norman, V.Yu. Podlipchuk, A.A. Valuev (1992): "Molecular Dynamics Method: Theory and Applications", J. Moscow Phys. Soc. 2, 7-17. G.E. Norman, V.Yu. Podlipchuk, A.A. Valuev (1993): "Theory of molecular dynamics method", Molecular simulation 9, 417-427. L. Schimansky-Geier & Ch. Ziilicke (1990): "Effect of Harmonic Noise on Bistable Systems", Z.Phys. B 79, 451-458. J.E. Straub, B.J. Berne (1986): "Energy diffusion in many-dimensional Markovian systems", J. Chem. Phys. 85, 2999-3006. P. Talkner, P. Hanggi (1995): "New trends in Kramer reaction theory", Kluwer Academic Publ., Dordrecht.
This page is intentionally left blank
Chapter 5
Excitations on rings of molecules A. Chetverikov, W. Ebeling, M. Jenssen, and Yu. Romanovsky
5.1
Solitary excitations in Toda systems
In the present chapter we investigate the effect of solitary excitations on reaction rates. Solitary waves as excitations of nonlinear chains of molecules found a remarkable interest in the last twenty years (Toda, 1983; Toda & Saitoh, 1983; Bolterauer & Opper, 1985; Trullinger et al., 1986; Mertens & Biittner, 1979, 1986; Theodorakopoulos, 1984; Schneider & Stoll, 1986; Ebeling & Jenssen, 1988, 1991; Jenssen, 1991; Jenssen & Ebeling, 2000). Of special interest for the study of Toda systems is the existence of exact solutions for the dynamics and the statistical thermodynamics. On this basis it was shown in the papers (Bolterauer & Opper, 1982; Schneider & Stoll, 1986; Theodorakopoulos & Bacalis, 1992; Jenssen & Ebeling, 2000) that phonon excitations determine the spectrum at low temperatures and strongly localized soliton excitations are the most relevant at high temperatures. In this chapter we will study the dynamics and activation processes on rings of masses connected by Toda or Morse interactions. Such configurations may be considered as special models of macromolecules, which allow a full theoretical treatment. We postulate a dissipative dynamics which includes white noise and friction. The noise strength is chosen such that it corresponds to an intermediate temperature region where both phonon and soliton excitations are present. In the Toda model such excitations are adequately described in terms of mutually interacting cnoidal waves. We will show here, that there exists a temperature region around a transition temperature TtT between the phonon and the ideal soliton regime, where the interaction of nonlinear excitations has a remarkable influence on several physical phenomena, a first one being energy localization at special sites, and another one the excitation of a broadband coloured noise spectrum with an 1//— region at low frequencies. This way we aim to give a contribution towards a new theory of energy activation processes as, e.g., chemical reactions in biomolecules. Here we will concentrate first on the investigation of classical Toda models of molecular systems. Toda forces (exponential repulsion) were introduced in section 1.3. The extension to more realistic forces (e.g. Morse models) follows in the last paragraph of the present chapter. 145
146
Excitations
on rings of molecules
1.0
0.9 0.8 cv/kB 0.7
0.6 t_T/c
0.5
10'10
10'5
'"I
105
1010
1015
Fig. 5.1 Specific heat per molecule in a Toda chain as a function of the temperature. At low temperatures the phonons dominate and at high temperatures the solitonic excitations. At the transition temperature with cy ~ 3fes/4 the most interesting properties are observed.
In our earlier work we have shown that soft Toda springs are able to trap and superpose pulse-like excitations (solitons) impinging from the hard host lattice (Ebeling & Jenssen, 1988; Ebeling et al., 1989; Jenssen, 1991; Ebeling & Jenssen, 1991, 1992; Jenssen & Ebeling, 1996, 2000). In thermal equilibrium there exists a finite temperature T(oc where the fusion of solitons leads to a maximal localization of potential energy at soft springs (Ebeling & Jenssen, 1991, 1992). We will show in this chapter that a system consisting of few soft springs which are imbedded into a bath of hard springs might be considered as a very rough model of an active reaction site imbedded into a bath of relatively passive atoms. The physical reason is that based on nonlinear excitations energy localization is possible at the soft sites (springs or molecules). In this and in the next paragraphs we concentrate on the study of systems with Toda interactions in thermal equilibrium including noise and friction effects following in large the work (Jenssen & Ebeling, 2000). At first we introduce the model and discuss its general properties. Then the properties of the equilibrium distribution function of the stochastic forces are investigated in particular regarding the transition region. Energy localization at soft sites will be shown to be a special consequence of the properties of the host system in the transition region. By numerical integration of Langevin equations we will study the time correlations of the equilibrium fluctuations of the stochastic forces. The influence of system size and varying degree of thermal coupling will be discussed. Our 1 — d model of a macromolecule consists of N point masses m; connected to the next neighbours at both sides by Toda springs. The actual distance between the mass i and the mass i + 1 is i?j, the equilibrium distance is assumed to be c;, therefore the spring elongation reads r, = Ri - (j{. The spring energy is described
Solitary excitations
in Toda systems
147
by Toda potentials
V{ri;ui,bi)
muj,
= —^-(exp ( - 6 ^ ) - 1 + 6;r;)
(5.1)
Here 6j is the stiffness of the spring i and w, is the linear oscillation frequency around the equilibrium position. We assume then the following equations of motion
Pi =
mi-iUi-! bt-i
c_bi_,ri_r
_
Pi+i
Pi
mi+i
rrii
rmujtc_biri h
1/2
+ 11
(5.2)
{2kBTmi)1l\i{t)-1\'2Pi
(5.3) The second term on the r.h.s. describes noise and damping generated by a surrounding heat bath with
(5.4)
The validity of an Einstein relation between noise strength and damping strength is assumed which guarantees the existence of a thermal equilibrium, independent on the friction parameter 7; > 0. We note that 7; = 0 corresponds to the conservative case. Toda springs are exponentially hard with respect to compression and rather weak with respect to expansion. Because molecules have similar properties we can consider the mass m, together with the spring r* fixed to the right side as a crude 1 — d model of a single molecule. For simplicity let us assume in the following that all the masses mi — m, all friction constants 7i = 7 and all equilibrium distances a; = a are equal. Further we assume that in the rest state at T — 0 the average distance of the particles is Ri = a, i.e. r* = 0. The linear term in (1) describes the dilatation interaction and implies the artificial assumption of an infinite range potential. However, interaction is confined to the next neighbours and consequently mu)}cr £i =
(5.5)
can be considered as the equivalent for the depth of the potential well in the Toda system. This quantity will be used as energy unit in the following. We will differ between uniform Toda chains where all the spring parameters bi = bo and un — UJQ are equal, and nonuniform chains where different spring parameters will occur. In the following paragraph we restrict our study to uniform systems and omit for simplicity all the indices.
148
Excitations
on rings of molecules
5.2
Statistical and stochastic theory of Toda rings
Let us begin with the discussion of the statistical properties of a canonical ensemble of "Toda rings" with pressure P and temperature T. With the abbreviations X
abkBT
Y =
P bkBT
the distribution function of a single molecule reads
f(p,r) = Zl exp(^ z-i._. 1
—
j
bx*+y y/2irnikBT exp (X) T(X + Y)
l
'
;
with the effective potential Veff(r)
= V(r) + Pr
Here kB denotes the Boltzmann constant. V(r) is the Toda potential as defined above. All thermodynamical functions follow by derivation of the partition function with respect to the corresponding parameters. Due to the Einstein relation the equilibrium distribution function does not depend on the friction strength 7. The mean specific volume is given by
v = j(lnX-y{X
+ Y)) + a
(5.7)
0
From this the pressure P is determined by: P = 6feBT*-1(lnX-6(v-(T))-e/a
(5.8)
where the digamma-function * is defined as usual by the logarithmic derivative of the T-function. The specific internal energy of the system is given by
e = E/N = ^kBT
+ (u)
(5.9)
The mean potential energy reads
(u) = ^ + ±(\nX-^(X 0 oa
+ Y))
(5.10)
Statistical
and stochastic theory of Toda rings
149
where the pressure P is given above. At low temperatures the equipartition theorem holds and we find
<«> = \kBT
(5.11)
In the high-temperature limit the "Toda rings" behave like rings of hard spheres and the potential energy disappears. The specific heat per molecule at constant volume reads
cv = kB(\+X
+
Y--^Y))
(5.12)
where the trigamma-function & is defined as usual by the second logarithmic derivative of the T-function.This function tends to ks in the low-temperature regime due to the thermal energy ksT of each phonon. On the other hand, cy shows the properties of a 1-d hard-sphere gas at high temperatures, i.e. cy tends to l/2fcs. For constant total length all the thermodynamic functions can be obtained from the phenomenology of an ideal soliton gas in the high-temperature limit, each soliton possessing a mean thermal energy of fcsT/2. However, all attempts to reconstruct the thermal properties of Toda systems from a phonon/soliton phenomenology failed in the transition region between the phonon regime and the regime of non-interacting solitons. Here we introduce a transition temperature Ttr corresponding to cy = 3A;B/4 (Fig.5.1). In the region around Ttr thermal excitations should be described more generally as cnoidal waves which contain both phonons and solitons as special cases in the low- and high- amplitude limit, respectively. Loosely speaking, cnoidal or solitary waves can be considered as deformed phonons with more or less steep compression humps and shallow dilatation valleys between them. In the transition region the interaction between solitary-wave excitations may lead to interesting physical effects as will be demonstrated in the following sections. In the remainder of this section we will discuss the time-independent properties of the stochastic forces. According to the definition of the Toda potential the force F acting on a molecule is given by
F=-(exp(-6r)-l) a The equilibrium distribution follows as tl„.
a
Xx+Y
/
aFY^-1
(5.13)
( aXF\
,C1^
150
Excitations
f(F)*E/o
on rings of molecules
10
F*a/e
Fig. 5.2 Distribution of the force acting on one molecule in a Toda chain for 3 different temperatures corresponding to the phonon regime (dotted line), the transition regime (solid line) and the solitonic regime (dashed line).
In the low-temperature limit we find the well-known symmetrical shape of the distribution function around the mean value F = 0 which is obtained for harmonic lattices (Fig. 5.2). With increasing temperature the distribution becomes biased and the maximum is shifted towards the left. As can be seen from the distribution, for X + Y < 1 the probability density diverges at F = — e/a whereas the integral over the F axis yields 1. The value F = —e/a corresponds to an infinite dilatation r. This behaviour reflects the tendency of the molecules of the chain to drift apart in the result of collisions, i.e., due to the presence of thermal solitons. As can be seen from our expression for cy, this behaviour is observed for cy < 0.89/CB, i.e., both in the transition region and in the region of noninteracting solitons. In the latter case, if T tends to infinity, the distribution function tends to get a rectangular shape consisting of a narrow peak at F = —e/a and a shallow but long-range tail for large F values which is typical for hard collisions (Fig. 5.2). In contrast, at the transition temperature we observe a relatively broad distribution with an exponential decay for large F. This behaviour reflects the dominance of "soft" collisions with relatively long interaction range. By integration we find for the mean value of the force F
(5.15)
(F) = P and the relative mean square deviation reads
52 =
(Ff
p2
+?I*BT
(5.16)
Energy accumulation
at nonuniformities
151
10 5
10 4
10 3
5
2
10 2
10
kBT/e
1 10- 10
T
10"5
1
10 5
1
10 10
1
1015
Fig. 5.3 Dependence of the mean square fluctuation of the molecular forces on the temperature; the minimum corresponds to the transition temperature.
Representing this quantity as a function of temperature we see, that the relative mean square deviation has a minimum near to the transition temperature Ttr (Fig. 5.3). Finally we notice that in the case of fixed total length the distribution functions given above are exactly valid only in the thermodynamic limit, i.e., for an infinite number of molecules. This is due to the fact, that we fixed the volume only in the thermal average by eq.(6). Hence the nature of fluctuations around the mean values may be modified for decreasing number of molecules in the chain. In particular, in a finite Toda chain the probability density of the force F cannot diverge at F — —e/a. With the constraint of periodic boundary conditions the expansion tendency of a finite Toda ring around Ttr may lead to a strong coherence of the molecular motion affecting the time correlations of the fluctuations, as will be shown in section 5.4.
5.3
Energy accumulation at nonuniformities
Now we will extend our model to the case of nonuniform rings which are considered as dilute solutions of n chemical sorts each represented by Ns soft "Toda atoms" with stiffness ba(s = l,...,n) in a bath of iVo hard "Toda atoms" (bs < bo) at constant volume. In the limit 60 —> 00 the bath becomes a l - d hard-core gas. In the following we assume Ns << No- Then the pressure is determined by the bath of hard molecules and remains the same as before (eq. (5.7-5.8)). The internal energy of the system is given by
E=^kBTlNo + J2Ns)+N(u°)+J2Ns(u°)
( 5 - 17 )
152
Excitations
on rings of molecules
5
4
3
/k B T 2
1
kBT/E 0 10"5
1
10
5
10
10
10 1 5
Fig. 5.4 At the transition temperature the mean potential energy of weak Toda springs imbedded into a bath of hard springs shows a pronounced maximum.
The mean potential energies of hard and soft molecules, respectively, read (Jenssen, 1991)
(us) = ?- + ^(\nXs-y(Xs
(5.18)
+ Ys))
(5.19)
where the pressure P is given by eq. (5.8). At low temperatures the equipartition theorem holds and we find
(u0) = (us) =
l
-kBT
(5.20)
In the high-temperature limit the "Toda molecules" behave like hard spheres and the potential energies disappear. Especially we find for the case v = a
{us) = ^ K )
(5-21)
Us
In previous work this exact result was reconstructed from a gas of noninteracting solitons which dominate the dynamics at high temperatures (Jenssen, 1991). There was also shown that at intermediate temperature the mean potential energy of soft molecules reaches a maximum in units of thermal energy fcgT (Fig.5.4). This effect was explained by superposition of thermal excitations at the soft molecules corresponding to multiple elastic collisions. Now we will study this maximum in more detail. From the localization condition
Energy accumulation
»
at
153
nonuniformities
=
.
(5-22)
we find using eqs. (5.10, 5.21)
lnXs - * [X. + Y.) - bh°Z{*Sty\
+, ' ,
osW'(A0 + YQ)
(5-23)
Y)
eaW(Xo + Y0)
+ <E±—El (l-X,*'{Xa+
¥.)) = 0
Equation (5.23) can be solved numerically yielding the temperature T\oc defining the maximum localization point. According to eq.(5.5) es can be interpreted as the depth of the potential well or binding energy of the soft molecules. For a given stiffness bs it defines the linear oscillation frequency of the soft molecules. With decreasing es the localization point T\oc is shifted to lower temperatures but (us(Tioc)) /kBTi0C increases slightly (cf. curves in Fig. 5.4). This means that mainly low-amplitude or phonon-like excitations will be localized at the soft sites with decreasing linear oscillation frequency. However, the time scale for high-amplitude or soliton-like excitations is given by the stiffness parameter. Hence the amount of collimated energy (us(Tioc)) /ksTioc increases significantly with decreasing bs (cf. upper curve in Fig. 5.4) due to the enhanced superposition of thermal solitons (Ebeling and Jenssen, 1988, Jenssen, 1991, Jenssen and Ebeling, 2000). In order to study this effect in more detail we will set es = e in the following. For sufficiently low stiffness b3 of the diluted molecules the localization temperature Tioc depends on the properties of the surrounding hard Toda chain only (Fig. 5.5). This behaviour was already observed in molecular-dynamics simulations of more realistic 3 — d models of dense fluids (see Chapter 4). In particular, for a vanishing ratio of stiffness parameters bs/bo the localization condition (5.22) can be simplified yielding
I n X . - ^ . + y
^
j
^
l
^
^
-
^
.
)
^
(5.24)
Solving this equation together with eq.(5.7) and inserting the solution in eq.(5.12) we obtain for the specific heat at the localization point
cv{Tloc)
~ 0.73
(5.25)
yielding a good agreement with the definition of the transition temperature Ttr given above. This result confirms the statement that energy localization at soft
154
Excitations
on rings of molecules
sites is a special consequence of solitary-wave interaction in the transition region. Finally we obtain from eqs.(5.17, 5.18, 5.23) for the mean potential energy of soft molecules at localization temperature a linear relation to the ratio of stiffness parameters
(us(Tioc)) = const.-±kBTloc
(5.26)
Energy localization at soft sites in thermal equilibrium is due to the pressure P produced by the bath of hard molecules in the transition region and can be attributed to the superposition of solitary waves corresponding to multiple elastic collisions. According to eqn.(5.15) the pressure P acts as a random force with a non-zero mean on the soft molecules. Hence the distribution functions (5.6) of the soft molecules are raised up to higher energies. If we interpret the soft molecule as a reactive site with a certain activation barrier, this force P may lead to a considerable enhancement of the transition rate in the region of Tj oc as was shown within the frame of elementary transition-state theory (Ebeling & Jenssen, 1991).
5.4
Fluctuations in Toda rings and time correlations
Beside the static localization effect discussed in the last section we expect novel effects concerning the nature of fluctuations around the equilibrium functions due to the interaction of nonlinear excitations in the transition region. So we will now proceed to the investigation of time correlation functions (ACF) and the derived spectra. In this section we restrict ourselves to the investigation of uniform Toda systems. Since the analytical theory of the ACF is restricted to the harmonic case, we rely on computer experiments in the following. Correlation functions from molecular-dynamic simulations of Toda systems were calculated, e.g., by Schneider (1986), mainly in order to identify thermally activated solitons in the spectra. Here we ask the question if Toda systems may transform the uncorrelated fluctuations of the surrounding heat bath, which are modelled by the Gaussian white noise (,(t) in eq.(5.4), into some kind of coloured noise that may favour low-frequency activation processes. This question is of central importance especially for the understanding of transition processes in complex biomolecules (Welch et al., 1982; Havsteen, 1989; Ebeling et al., 1989; Chikishev et al., 1998; Regeida et al., 1999). In contrast to the equilibrium properties following from the distribution function (5.6), the time correlation functions depend on the friction parameter 7, because its inverse determines the characteristic relaxation time of the system (5.2-5.3). In the conservative case 7 = 0 the system follows the Hamiltonian dynamics. On the other hand, for large 7 the Hamiltonian dynamics is modified and the "Toda atoms"
Fluctuations
in Toda rings and time
155
correlations
0.2
kBTlx/e 0.1
/ /
1
b/b s .
i
10
,
i
i
100
i
i
.I
1000
Fig. 5.5 Temperature of maximal energy localization at weak Toda sites imbedded into a bath of hard Toda springs. The temperature Tioc ~ 0.26e corresponds to the transition, since cv(Tioc) ~ 0.73fcB.
reproduce the white noise of the surrounding heat bath. So we performed our simulations of the Langevin eqs. (5.2,5.3) for the case of weak thermal coupling with 7 = 10_3wo- For integration we used a Runge-Kutta algorithm of 4th order using a constant stepsize of 10_3o;o including white noise sources according to eq.(5.4). First the system was heated up to a temperature around Ttr. After attaining the thermal equilibrium we calculated the one-particle ACF of the forces (5.13) from the trajectory
ACF(t) = (6F(t0 + t)5F(to))
(5.27)
From this we calculated the spectrum (FF) defined as the Fourier transform ofeq.(5.27). For low temperatures the Toda system behaves like a harmonic lattice and the dynamics is determined by the phonons. Fig. 5.6 shows the force spectrum calculated from a long-term simulation of a molecular ring consisting of N = 10 identical "harmonic molecules" subject to periodic boundary conditions. As expected, the excitation spectrum shows 5 distinct peaks at the well-known normal-mode frequen-. cies. Naturally, the white noise of the surrounding heat bath is reproduced at low frequencies. For N —> oo the thermal energy will be distributed equally among an infinite number of phonons, the distinct peaks disappear and we observe a whitenoise spectrum over the entire frequency range. On the other hand, for high temperatures the Toda spectrum behaves like an ideal hard-core gas characterized by the simple white noise again. Let us study now the characteristic features of nonlinear excitations. Fig. 5.7 shows typical realizations of the force acting on a Toda site in dependence on time and
156
Excitations
on rings of molecules
(FFVmax((FFX„) 1-3
0.001-
(l)/COo
0.00010.01
"1 10
Fig. 5.6 The spectrum of the force-force time-correlations for a uniform ring of 10 harmonic oscillators. We observe 5 phonon peaks at u = 2u>osin(iir/10) for i = 1,2, ...5 and a white noise tail at small frequencies.
Fig. 5.8 shows the corresponding excitation spectrum of the one-particle ACF. The curves were obtained from a simulation of a uniform Toda ring made up of TV = 10 "Toda atoms" in the transition-temperature region. The trajectory of the molecular force F shows excitations on many time scales. The high-energy events corresponding to high-compression peaks or solitary waves occur most likely in clumps as it is typical for beating phenomena (Fig. 5.7). The excitations on many time scales correspond to a very specific spectrum (Fig. 5.8). Most strikingly, in the double-logarithmic presentation we observe a straight line with a slope near to — 1 clearly indicating a broadband coloured noise of 1 / / type at low frequencies. In particular, the 1 / / spectrum implies a hierarchy of beatings where periods with more energetic compression pulses are more probable to appear at longer time intervals. The hierarchical order of the fluctuations is due to the Hamiltonian dynamics of the finite-size Toda ring. Finally let us try to get a hint on this underlying dynamics by comparison of the spectra of the harmonic and the Toda ring in more detail (Fig. 5.9). We notice the complete failure of the first phonon peak in the Toda spectrum. The time-dependence of the force-force ACF is shown in Fig. 5.10. In the harmonic ring the (linear) excitations of the first phonon correspond to a standing wave with two knots which can be decomposed into 2 waves of the same frequency but running in opposite directions. Indeed, on the Toda ring we observe during the simulations mostly 2 distinct soliton-like wave peaks running in opposite directions and merging into mainly 1 wave peak 2 times each turn (Fig. 5.10). The frequency of these solitary waves corresponds to the broad peak in Figs. 5.7 and 5.9 which has its maximum accurately at the second-phonon frequency. Furthermore, the speed of the nonlinear waves is varied with the fluctuations of their
Fluctuations
in Toda rings and time
157
correlations
200
F*o7e
t/10000*coo
200
F*o7e 100
t/10000*coo 1,165
Fig. 5.7 Timedependence of the force acting on one site in Toda chain with N = 10 for 2 time scales. We observe a hierarchy of fluctuations on different time scales and beating-like phenomena.
amplitudes and hence we do not have a fixed phase relation leading to standingwave phenomena. Instead, we observe a beating-like modulation of the amplitude of compression peaks due to the strong interaction of mainly 2 solitary or cnoidal waves each of wavelength N and a frequency fluctuating around that of the 2nd harmonic normal mode. The internal dynamics of the Toda ring and the hierarchy of nonlinear excitations is accompanied by long-term correlated random rotations of the whole ring (Fig. 5.11). These coherent fluctuations of the ring correspond to a diffusion regime. With respect to the spectrum of fluctuations there is some analogy to the traffic-jam models which are classical examples of 1 / / noise (Helbing, 1997; Helbing et al., 2000).
158
Excitations
on rings of molecules
(FFX/aV 0,01 —
TTT]
0.01
0.1
oVcoo
100
Fig. 5.8 Spectrum of the force-force correlation function for Toda rings with N = 10 and N = 20. The temperature is in the transition regime, the friction is rather small 7 = O.OOlwo- We observe a 1 / / tail at small frequencies.
(FF)„,/max((FF)„) 1
(0/(Oo
0,1
10
Fig. 5.9 Comparison of the force-force spectra of an N— particle Toda ring with the corresponding ring of harmonic oscillators. We observe the complete failure of the first-phonon peak in the Toda system and instead a soliton-determined peak at the frequency of the second (harmonic) phonon peak.
In this analogy the "Toda atoms" correspond to "cars" moving around a circular highway and the compression peaks make up the jams of different size. However, our "Toda atoms" do not possess a "fuel tank" and are driven only by the Brownian motion of the surrounding heat bath. Consequently, the long-term average of the velocities tends to zero. The case of driven Toda atoms which may use internal energy sources was treated by Erdmann et al. (2000). We stress that the passive Toda rings treated in the present work are in thermal equilibrium and the 1 / / tail of the spectrum reflects the character of equilibrium fluctuations due to nonlinear excitations.
Fluctuations
in Toda rings and time
correlations
159
O
<
Fig. 5.10 The autocorrelation functions of the force acting on a Toda site in rings of 10 particles (full line) and 20 particles (dotted line). We observe a distinct soliton peak corresponding to two solitary waves running in opposite directions around the ring. The time at which the maximum appears corresponds to the time needed for one full rotation, which nearly linearly increases with the size of the ring N.
Prom the above considerations it follows not only a dependence of the effect from the temperature and the degree of thermal coupling, but we presume also a great importance of the boundary conditions and hence a dependence on the size of the Toda ring. We already speculated about coherent molecular motion as a result of the repulsion tendency of the molecules within a finite Toda ring when discussing the distribution function of molecular forces. Clearly, such a coherence gets lost with increasing number of molecules. It follows from Fig. 5.10 that the distinct solitarywave peak corresponding to the second-phonon frequency will not only shift to the right on the time axis but also gets less pronounced with increasing N. Hence the thermal energy gets more equally distributed over a broad spectrum of cnoidal waves with varying wavelength and frequencies instead of being contained in mainly 2 strongly interacting pulses. Therefore in the infinite Toda ring we expect a white noise at low frequencies again, the noise effects we discuss here are typical finite-size effects. The most important observation is a kind of l//-noise. The two main types of l//-type noise systems discussed so far in the literature are: (i) flicker-noise observed in many simple physical systems (Klimontovich, 1995), (ii) long-correlated noise observed in many complex systems in nature and society (Bak, 1996). As shown by Klimontovich, flicker-noise may be interpreted as a diffusion regime
160
Excitations
on rings of molecules
Total moment * a(0(/£ 40
t/10000*0)o
Fig. 5.11 The total momentum of the Toda ring corresponding to the simulations shown in the previous figures fluctuates slowly what corresponds to slow stochastic rotations of the ring.
of the dynamics of finite systems (Klimontovich, 1995). Typical properties of the flicker-noise of Klimontovich-type are: (i) This type of flicker noise appears at frequencies uifi bounded from above by the diffusion time L2/D and from below by the observation time t0bs
(1/ioto) < wfi <
(D/L2)
(5.28)
where D is the diffusion constant (corresponding here to the rotational stochastic motion shown in Fig. 5.11) and L is the length of the system (corresponding here to 10a). (ii) the amplitude of the flicker noise is proportional to l/(Nw), where N is the particle number. A quite different type of l//-noise has been observed in large and complex manyparticle systems far away from thermal equilibrium. This type of l//-noise is of central importance in the theory of self-organized criticality (Bak, 1996). In this section we presented a relatively simple dynamical system fluctuating in thermal equilibrium that transforms the uncorrelated, white noise of the surroundings into noise of l//-type. In many respect the type of noise we have observed corresponds to a flicker-noise of the type investigated by Klimontovich since it is
Spatio-temporal
excitations
on rings
161
clearly connected with a diffusion-type of dynamics as shown in Fig. 5.7 and since it seems to decrease with a dependence 1/N. A closer inspection of this point however requires more extensive simulations; therefore we have to leave this question to further investigations. Even at the present stage of the investigations we may draw some general conclusions: The l//-noise observed in Toda systems is not connected to a fine-tuning of temperature, structural parameters, particle number, or thermal coupling, but occurs in a wide range of these quantities with varying intensity. Furthermore, it can be expected to persist for a wide class of more realistic molecular potentials in the transition region. However, it was not observed in previous investigations for Lennard-Jones molecules at higher dimensions so far (see Chapter 4). This point needs further clarification. From the investigations carried out so far for our models we hypothetically derive the following preconditions for fluctuating equilibrium systems to transform white noise into l//-type noise: (i) Nonlinear, asymmetrical interactions between molecular units with a steep repulsive and a flat attractive branch. (ii) Quasi-Id configurations of a finite many-body system with periodic boundary conditions (ring). (iii) Weak thermal coupling to a surrounding heat bath in the transition-temperature region. We hope that the further detailed investigation of the effects demonstrated in this section paper will support the understanding of complex molecular motions and energy activation processes. A first application in this direction is presented below.
5.5
Spatio-temporal excitations on rings
The volume/density fluctuations discussed in the previous section are closely related to the structure factor which is a central quantity for the description of excitations (Schneider & Stoll, 1986; Mertens & Biittner, 1986; Ortner, Schautz & Ebeling, 1997; Ebeling & Ortner, 1998; Ebeling, Chetverikov and Jenssen, 2000). We will use here the definition (Ebeling and Ortner, 1998): 1 r°° •wt S(w, k) = ^ J e (p(k, t)p(-k, 0)> dt,
(5.29)
P(*'') = ^ E e x p H M * ) )
(5-30)
where
162
Excitations
on rings of molecules
is the Fourier component of the density of masses in the chain. The static structure factor is defined as the integral over u> which is denned by
s
w = jj E <exp Wri - r*))>
(5-31)
This quantity may be expressed by the partition function. Several analytical and numerical approaches to estimate the structure factor in the volume edited by Trullinger, Zakharov & Pokrovsky (1986). The dynamical structure factor allows judging time behaviour of spatial (collective) structures of specific scales, generated on a ring. Setting quantity k, defining a frequency composition S(w, k) and estimating breadth of spectrum peaks, it is possible to judge a stability of structures (nonlinear waves) and velocity of its motion on a ring. We study here rings of N masses with Toda interactions. Most of our simulations were carried out for rings with N — 10. This has the advantage that the number of excitations is still rather small and so we can observe their shifts in some detail. Further we are interested in finite size effects, which play an important role in the soliton-determined region. Let us first discuss the linear case of only N phonons i = —N/2 + 1-7- N/2 with wave numbers and frequencies ki0 = (2Tri/N(T),
wi0 = 2w0|sin(fcio(7/2)|
(5.32)
For simplicity it is supposed that N = 2M is an even number. The positive and negative signs denote waves traveling either to the right or to the left, the excitation with k = 0 corresponds to very slow rotation of the ring as a whole, and k — n denotes the non-traveling wave. The motion of any mass is strictly sinusoidal. The speed of the longest traveling wave of i — 1 is near to the long wave limit vs — UWQ which is the sound velocity. The group velocity of the phonon excitations is always smaller than the sound velocity
vig = vs cos ——
(5.33)
Including nonlinear interactions, in the simplest case only quadratic terms, the phenomena due to nonlinear multiwave interactions in a medium with dispersion are very specific. Suppose that N/2 waves are propagating left/right on the ring in accordance with the dispersion relation D(u>i,ki), including two components of the standing wave with a wave number ki — ir. Then due to the nonlinearity combination waves will arise with frequencies and wave numbers
kc = ^2niki,
wc = ^ n i W i
(5.34)
Spatio-temporal
excitations
on rings
163
Wave numbers kc coincide with wave numbers ki always because of equally spaced components of a wave numbers spectrum on a ring. But combination frequencies OJC do not coincide with Wi in a general case due to dispersion and may be distinguished in the frequency spectrum. If nonlinear interactions cause the waves to have forms very different from sinusoidal, the quasiharmonic approximation will not work. In the sufficiently (but not very strongly) nonlinear case e.g. for the Toda ring the situation is the following. In the unlimited chain we could observe N cnoidal waves which might be described by elliptic functions. In some sense the cnoidal waves are nonlinear modifications of the phonons, they are still periodic functions but are not sinusoidal. A wave length of cnoidal waves grows when their energy increases and nonlinearity develops, and then they convert into N proper solitons. The velocity of the solitons depends on their energy but it is always larger than the sound velocity. However, the wave length of nonlinear waves on a ring is limited by the ring length, so they are not cnoidal waves, strongly speaking, when the energy becomes sufficiently high. Thus, each phonon converts to the nonlinear ("like cnoidal") wave with only one hump on the ring finally. But in relatively long rings they may be considered as solitons if their widths are much less in comparison with the ring length. Specific role is played by the first phonon with the longest wavelength. It has the space structure topologically like the one hump nonlinear wave from the very beginning. Therefore it is transformed to the soliton at lower temperature than the other "many hump" phonons. We now consider these linear and nonlinear excitations in thermal systems: In accordance with section 3 and refs. [4-7], the excitations at low temperatures are determined by the phonons and the excitations at high temperatures are soliton-like. Between these limits there is a temperature, where the specific heat turns from ks to ks/2. Around the transition (or localization) temperature we observed in earlier work very specific physical properties and therefore we will pay here special attention to it. We calculate the dynamic structure factor (DSF), since this quantity contains a rich information about spatial and temporal excitations as well. We simulated a dynamics by a Langevin equation including a white noise at given temperature, then we switched off the noise and the friction (7 = 0) and started the calculation of the dynamic structure factor under Newtonian dynamics. This procedure ensures that the dynamical effects studied refer to a Hamiltonian system in thermal equilibrium at temperature T. In our numerical calculations we use the same dimensionless variables as in previous sections. In particular, we suppose m = 1 and use dimensionless time T = t(cj0/2n) and coordinates r' = R/a0. Our choice for w0 is the linear oscillation frequency around the minimum and a0 is the equilibrium distance. The dynamic structure factor of a ring of N identical linear oscillators may be estimated analytically. Taking into account that the time of integration Tint of the integrals defined above are limited in numerical modeling we get for this case the analytic expression
164
S{u,k) _
Excitations
1
JZ
on rings of molecules
{sm(±(k-km0)N)\
/|sin(i(W-Wroo)Tint)|
\±(k-km0)N\ J \
||(w-« m o)r int | J
(5.35) Here Smax is the maximum of S(uj,k) in the range of (u>,k) considered, am — const/w„0. The last follows from the fact that the dynamical structure factor is determinated by averaged displacements of particles < rf > and each phonon gives contribution to < rf > proportional to l / w ^ 0 because of all phonons have the same potential energy < um > = | w ^ 0 < r*hm >= \kBT. Here < r*hm > denotes the averaged displacement in the m—th phonon. The phonon with k = LJ = 0 ("zero" phonon) corresponding to rotation of the ring as a whole should be discarded in our consideration because of am=o —> oo. As it is supposed that N = 2M = 10 is an even number, then M linear phonons traveling to the right and M linear phonons traveling to the left with frequencies ujio may be excited in the chain taking into account that the standing wave with k = IT may be represented as a superposition of two opposite traveling waves. As it follows from the above equation, the dynamical structure factor of a linear oscillators ring as a function of a frequency looks as a single spectral line of the frequency w^o for the case k = fcjo- But for k ^ k^ we can see 5 peaks corresponding to the 5 different frequencies w;o- It follows that the initiation of excitations with wave numbers near to ki may be observed if A; = ki. Further we may observe a shift of wave numbers ki which could be caused e.g. by nonlinear effects. At low temperatures when nonlinear effects are weak, the interaction of the weakly deformed phonons may be described by the quadratic terms in the forces (the cubic term in the potential). This approximation leads to a potential of Fermi-Pasta-Ulam type (Ebeling, Chetverikov and Jenssen, 2000). The Fig. 5.12 shows the result of simulations for this approximation. Due to the quadratic nonlinearity adjacent phonons having frequencies uiio and Wi_|_i,o give rise to combination components, i.e. excitations with frequencies Au>i — u)i+ito — w;o and intensities S(Auii) ~ ajOi+i. Because of the dispersion relation of the ring u — oj(k) the lowest component of them is A W M - I , and the highest one is Awi which is near to woi. The wave numbers of the excitations A/c, are approximately equal to k\ and they are very much pronounced in the spectrum S(w, k = kit0), especially if N is not big. In this case there are only ./V = 2M traveling phonons with M frequencies on the ring and we are able to study their shifts in detail. As phonons convert to cnoidal waves, it is reasonable to suppose that under increasing of temperatures first two soliton-like structures with only one maximum, traveling in opposite directions, will be excited on the ring appearing from two lowest traveling long wave phonons. They are equivalent and we will consider for definition the one with a positive wave number. For simplicity we call this onehump cnoidal wave the "soliton". Its energy is small and its velocity is near to 1 in
Spatio-temporal
excitations
0.01
0.1
on rings
165
o.i
S E u,
a to a; 3
0.01
0.001
0.0001
le-05 1
10
w
Fig. 5.12 Dynamic structure factor under weak nonlinearity for the first phonon wave number k — k\fl at T = 0.5 * 1 0 - 3 e . We observe a spectral peak of the first phonon and four combination components.
accordance with the dispersion relation but always a bit larger than 1. Consequently, the recurrence frequency of the soliton on the ring is near to the basic frequency tOio but grows when energy is increased. At transient temperature the frequency corresponding to the soliton wso; is already twice the initial value uji0 (Fig.5.13) and continues to grow further as well the velocity of the soliton. Simultaneously a shift of the wave numbers and an increase of the frequencies of higher cnoidal waves in the range W20 -=- w50 — 2 take place. Near Ttr all of them including the soliton" interact intensively giving rise to combination components with frequencies lower than the frequency of the soliton wso; = W\ due to dispersion effects (Fig. 5.13). Also the harmonics 2u>soi, Zu>soi and others become important. However, higher waves are less stable already than the soliton as they begin to change their space structures to convert into nonlinear waves with the only hump at high temperatures. It is proved by results of the study of a ring chain with relatively strong connection to surrounding (7 = 10~3w0, Fig.5.14). The noise spectrum corresponding to coloured noise is observed at T « Ttr with pronounced frequency peak of the soliton, a decaying tail corresponding to higher waves and harmonics and the only distinguishable combination. So structures corresponding to higher waves are easily destroyed under influence of noise. This is always observed by analyzing the dynamical structure factor under increasing coupling of the chain to the surrounding. With increasing AT the process of forming of the high energetic soliton as result of synchronization and degradation of some low frequency phonons becomes not so distinct because of density increasing of the spectrum of phonons. Here however we did not study in details rings with big JV. When temperature increases in the region T > Ttr, energy, velocity and fre-
166
Excitations
on rings of molecules
Fig. 5.13 Dynamic structure factor for the wave number k = /s^o near at T = 0.13e near the transition temperature. We observe spectral peaks of the soliton (w s o ;), its harmonics and combinations with w < u/soi.
Fig. 5.14 Dynamic structure factor for the wave number k = kio at T = 0.13e of the ring imbedded into a heat bath with 7 = 10 -3 u>o.
quency UJSOI of the soliton grow too. Furthermore high cnoidal waves convert into soliton like structures also. The soliton becomes more narrow, hence its harmonics in the spectrum S(w) grow as well and combination components decrease owing to moderation of the waves interaction (Fig.5.15). The results described refer to the behaviour of ring excitations on a space scale corresponding to the main resonance scale of the ring defined by its length. We
Spatio-temporal
excitations
on rings
167
Fig. 5.15 Dynamic structure factor for the wave number k = fcio at high temperature T = 24.5e. Spectral peaks of both the soliton and its harmonics are observed pronounced. Combinations do not disappear still.
highlight them due to the exclusive role of the soliton as an asymptotic structure of all nonlinear excitations on the ring at high temperatures. However the range of small frequencies and wave numbers is a field of specific interest because of the phenomena of 1 / / noise and flicker noise found in earlier work (Klimontovich, 1995, Jenssen and Ebeling, 2000). In Fig.5.16 we represent the dynamical structure factor for the case of small wave number k = 0.1 for the Hamiltonian system (7 = 0) at the transient temperature. We observe both a spectrum with 1 / / tail at low frequencies and the typical high frequency spectrum of nonlinear waves and combinations described above. But the like spectrum in the small frequencies region is found at small temperatures as well, in spite of the fact that the average velocity of particles is set to zero at the beginning of the equilibrium stage and that the average shift of all particles excludes a rotation of the ring as a whole. It means that we can not exclude even the possibility that the weak numerical noise in calculations of the dynamical structure factor plays a role in virtue of < r% > ~ 1/w2 —>• 00 for excitations with w « k « 0 even if they have very small energy. Moreover it might be that chaos phenomena in this Hamiltonian system gives a contribution to a low frequency part of the DSF. But if a ring contacting with a surrounding is considered (7 ^ 0), the most essential contribution to the 1 / / part of S(LJ) is provided by a very slow rotation of the ring as whole (Fig.5.17.). It should be associated definitely with excitations with a frequency and wave number near to zero that usually are excluded in thermodynamical considerations. Besides the frequency dependence l / w a , a > 1 is observed at the low frequency edge of this part of the spectrum. It may be supposed that it is due to a diffusion
168
Excitations
on rings of molecules
Fig. 5.16 Dynamic structure factor for the small wave number k = 0.1 at T = 0.13e. We observe both peaks in the region of traveling nonlinear waves and 1 / / spectrum at small frequencies
Fig. 5.17 Dynamic structure factor for the small wave number k — 0.1 at T = 0.13e of the ring imbedded into a heat bath with 7 = 10~3W0'
type flicker noise predicted by Klimontovich (1995) as mentioned above. It takes
place in the frequency range defined by (5.28). In accordance with the work of Klimontovich the dynamic structure factor as a function of a frequency and wave number is expressed at small ui and k by Dk2
S{u,k) = C 1 w + (Dk2)'
<
D k
\
(5.36)
Spatio-temporal
S
excitations
on rings
169
o.oi -
Fig. 5.18 Dynamic structure factor for the small wave number k = 0.1 at X « 24.5e of the ring imbedded into a heat bath with 7 = 10—3CJQ-
Fig. 5.19 Dynamic structure factor for the small wave number k = 0.2 at T R S 24.5e of the ring imbedded into a heat bath with 7 10 Ju>o- (Narrow dips in the range considered are not of physical nature)
where C is a constant. The maximum value of dynamic structure factor is realized at w = ujextr ~ 0.725Dk2. As it follows from an estimation, cjextr is small enough at T w Ttr. Therefore first the phenomenon has been studied at high temperature T = 24.5 « 190T tr . Indeed we find nonmonotonic function S(u) with the maximum shifted as k2 (Figs. 15.18 and 15.19). Returning to the system under the transient temperature, we can see a slope
170
Excitations
on rings of molecules
corresponding to nicker noise which is shifted as k2 indeed. But much further studies are required to confirm the relation S(CJ, k) w Du>2/k2 at small Dk2 and its connection with the flicker-noise of Klimontovich-type. 5.6
A ring model of enzymes
As we have underlined already several times in the previous Chapters, the detailed investigation of the dynamics of molecular processes is of central importance to a better understanding of the activation processes in enzymes. In fact many activation phenomena in complex molecular systems as, e.g., enzymatic reactions or protein-folding processes remained unexplained so far. We summarized also several approaches to extend the original Kramers theory to more complicated reactions and to treat the coupling of non-reactive molecular systems to the reactive site explicitely. As a particular simple 1-d models of excitations in molecular systems we investigated in the previous section Toda systems. We have shown that it is the special form of the Toda interactions which admits exact solutions for the dynamics and statistical thermodynamics. In respect to activation problems the most important result is, that in the transition-temperature region the strong interaction between solitary waves may lead to an enhancement of transition rates. We have studied several mechanisms in Chapter 3 and 4. The most effective one is the energy localization at "soft" reaction sites embedded in a hard Toda chain (investigated the present Chapter) or in a two-dimensional bath of hard molecules (Chapter 4). A complete theory of activation processes assisted by nonlinear excitations is not yet available. In some earlier work (Ebeling, Sapeshinsky and Valuev, 1998) as well as here in the forth chapter we estimated reaction rates by means of molecular dynamics simulations. Further we gave analytical estimates according to transition-state theory assuming equilibrium distributions of coupled reaction systems (Ebeling & Jenssen, 1991). Several estimates resulted in considerable rate enhancements at intermediate temperatures. One of the factors leading to enhancement follows from the non-zero average force (pressure) which acts on a soft reactive molecule and is of static character. This force affects the exponential factor of the rate constant significantly by lowering the activation barrier. Another dynamic effect of enhancement can be interpreted by a superposition of solitary waves which becomes possible at "soft" molecules at intermediate temperatures. It is not restricted to 1-d Toda lattices but persists also in more realistic 2-d and 3-d models of dense fluids consisting of solvent and solute molecules with Morse- or Lennard-Jones interactions. Superposition of solitons corresponds to multiple collisions in these systems. In higher dimensions a weak localization of potential energy was observed also at the bindings of the bath molecules and was connected to a transition between different lattice configurations(Ebeling et al., 1995).
A ring model of enzymes
171
Beside this static effect a considerable rate enhancement can be expected from a transformation of the uncorrected noise of a surrounding heat bath into a broadband coloured noise with a long tail at low frequencies. This behavior was proven above for a finite-size Toda ring with weak thermal coupling in the transitiontemperature range. Such a system seems to be an ideal host for the excitation of special active sites possessing resonance frequencies inside this low frequency band. The finite-size Toda ring with moderate coupling to a surrounding heat bath in the transition-temperature region is perhaps the simplest classical model of a ring-shaped biomolecule in solution. In the following investigation based on an earlier work (Ebeling, Jenssen and Romanovsky, 1989) the real molecular forces in proteins and DNA will be fitted to Morse potentials. We will show that the transition-temperature region corresponds to the range of physiological temperatures (Muto et al., 1989). Indeed, statistical analysis of experimentally observed single-molecule trajectories of the enzyme cholesterol oxidase revealed significant and slow fluctuations in the reaction rate (Lu et al., 1998). This effect was described as a molecular memory phenomenon, in which an enzymatic turnover is not independent of its previous turnovers because of slow fluctuations of protein conformation. Such long-term correlated conformational changes which are relevant in context with protein-folding processes and enzyme reactions might be explained by dynamical mechanisms similar to those observed in our simple Toda model. In the models discussed before we found coherent molecular motions, in particular mainly 2 soliton-like waves of similar amplitude and frequency running in opposite directions. The fluctuations of their amplitudes and frequencies lead to some kind of nonlinear beating phenomenon that is connected to a region of the spectrum that is similar to l//-noise. In particular we consider a molecular ring in the diffusion regime as an ideal host for the excitation of special active sites possessing resonance frequencies inside this low frequency band. A possible field of applications is the dynamics of biomolecules with encymatic activity (Welch et al., 1982; Havsteen, 1991; Chikichev et al., 1998; Lu et al., 1998). Consider a particular enzyme molecule. Cluster models of enzymes sometimes give examples of the chains of masses interconnected by H-bonds. One of this models was discussed in Chapter 1 (see Fig. 1.2). Recall that the successful operation of the molecular scissors (CT) depends on the proton transfer in the hydrogen bond from oxygen (of Serl95) to nitrogen (of His57). Note that these amino acids are located on different subglobules. It is seen from Fig. 1.2 (see Chapter 1) that each subglobule consists of 6 rods interacting with each other by H-bonds. Therefore, the vibrations of the rods and the subglobules may change the parameters of the selected H-bond ( N...H-0) in the enzyme active site and substantially influence the probability of the proton transfer. This problem is discussed in details in Chapter 8. Here we make an attempt of estimating the statistical regularities of the cluster (rod) vibrations using a simple model. We consider a chain of masses forming a
Excitations
on rings of molecules
V(r0<0), kJ/mole
a)
30
* .
20 10
••
r.
.1
0 10
20
t, ps O.l-i
00.1-
r0, nm
: x*.: :-S ^ /A M r*. r**x^*\f~y/~fi^ v v './
..
•;.•
:j
v'
v
•;
V
v
1/
20
10 t, ps
b)
V(r 0 <0), kJ/mole 30 20 H 10
V .A ft
>*-* 10
S\
-A20
t, ps
r0, nm O.H
:ywyvvwvvv^ A
'•
T
10
t, ps
20
Fig. 5.20 Simulations for the Morse potentials (a) and for quadratic potentials (b). Represented is the potential energy V and the elongation ro of the active spring after an initial excitation of the weak spring.
ring with H-bonds described by the Morse potentials. Based on the ideas and results discussed above, we propose the following ring model for an enzyme: T h e dynamic systems consists of n interacting spring interaction by forces with the potential V(rn, where rn denotes the deviation from the equilibrium position. We will use as model cases the Toda potential
A ring model of enzymes
V(r„) = {an/bn) [exp(-6„r n ) - 1 + brn]
173
(5.37)
and in most cases the more realistic Morse potential V{rn) = Dn [(exp(-A„r„) - l ) 2 - l]
(5.38)
Both potentials reduced after expansion up to quadratic terms into harmonic ones. For the Toda case the frequency is given by
ul = anbn
(5.39)
Ji = 2DnA2n
(5.40)
and in the Morse case we have
The equations of motion are defined by
—j- =vn-
u„-i
(5.41)
nil
^ = - 7 n « n + [V'(rn+1 - V'(rn)} + Fn(t) at n = 0,l,...N-l
(5.42)
(5.43)
Here Fn(t) is an external force acting on the mass n which will be used to model external impacts. For the simulations which we will discuss now we assumed N = 12 masses connected by Morse springs, where n = 0 corresponded to a weak spring and n = 1 — 11 to hard springs. In order to be as realistic as possible we assumed Dn = D — 17kJ/mole,An = 0.5nm _ 1 for n = 1 — 11. These parameters should be a rough approximation for the situation of hydrogen bonds. For the weak spring which should model the active site we assumed AQ — A/50 = O.Olnm -1 and WQ = w 2 /10. The masses were adapted to the case of a-chymotrypsin corresponding to M„ = 2083a.u. + / - 10%. The motion of the active spring is exposed to the solvent modelled by a friction 70 = 2 . 8 6 1 0 n s - 1 . The internal motions of the springs n = 1 — 11 are assumed to be undamped 7„ = 0. An initial impact modeling an adsorption event is different from zero only for a small moment
Fu = -F0{t) = C, ifT <= 2TT/UJ0
(5.44)
Excitations
174
on rings of molecules
V(r 0 <0), kJ/mole
a)
30 20 10
i / V / V / w A . . ^ /\
J^
-.
20 t, ps 0.1-
r0, nm
.^yV^Vw^vv^^ i
10
20 t, ps
V(r 0 <0), kJ/mole
b)
30 20 10
-~T—•
r-
20
10 t, ps r0, nm O.l-i
0-
VV>Wy
0.1-
-r 10 t, ps
20
Fig. 5.21 Simulations for the Morse potentials (a) and for quadratic potentials (b). Represented is the potential energy V and the elongation ro of the active spring after an initial excitation of the hard spring.
The height of the pulse is adjusted in such a way to provide a maximum total energy of 65kJ/mole. Fig. 5.20 shows the result of the simulations for a ring of Morse springs in comparison with the case of linear springs. We see that in the case of Morse springs the initial compression energy is rather weakly damped and survives at least 10 ps. Opposite to this the pulse decays in a linear chain very quickly. In Fig. 5.21 the same experiment is demonstrated for the case that the initial compression refers to the hard spring which connects the
A polymer reaction model including entropy
effects
175
masses 3 and 4. This situation could possibly model an excitation by the absorption of an energyrich molecule (ATP). In the linear case the initial energy never reaches the active site. In the nonlinear Morse model the energy is transferred always to the soft site. In this way we have demonstrated that the nonlinearities of interactions in a macromolecule may provide a localization and repeated recurrence of compressional energy at a functionally relevant part. In a recent paper a systematic investigation of the coherent motions and the cluster formation in Morse rings including the action of negative friction has been undertaken (Dunkel et al., 2001,2002). Note that all the effects discussed above are clearly seen at low decay decrements. In Chapter 8 we will estimate the Q-factors and damping coefficients for the oscillations of the elements or clusters at the amplitudes of tenths of angstroms.
5.7
A polymer reaction model including entropy effects
The Kramers theory describes reactions as stochastic transitions over a barrier of potential energy AU (activation processes). In reality the reaction processes occur in a heat bath of given temperature. Then as we have explained already in Chapter 1, the role of the energy barrier is taken over by the free energy barrier AF = AU — TAS. In this way the entropic changes along the raction path are included. Since macromolecular reactions are mostly connected with conformational changes this point may play a very important role. In order to elucidate the possible role of entropy changes let us consider in the following a toy model for the microscopic mechanism of reactive transitions in a polymer which are coupled to entropic changes. The physical effects which we would like to study is the coupling between a simple Kramers transition over a barrier with entropic transitions inside the reacting polymer. The model is closely related to recent studies of the activation of polymers (Sung and Park, 1998; Park and Sung, 1968) and also to the model explained in section 4.1. At first we develop a special microscopic model for the coupling of a Kramers-type bistable reaction site to a polymer conformation. We repeat here again that our special interest is devoted to the microscopic dynamics of the transitions in enzymes and in particular in the exceptionally high transition rates. Let us introduce now the model: We consider a reacting molecule with the coordinates XQ and the momenta po with the internal reaction coordinate q following an overdamped dynamics. The reacting corresponds to the breaking of a bond which is modelled as a motion in the cubic potential
U(q) = \aq2 - \aq*
(5.45)
176
Excitations
on rings of molecules
The bond is breaking, if q reaches the value of the barrier qbar = a/b. The center of mass of our molecule is moving freely in a liquid. Further we assume, that a linear polymer is swimming in the liquid with a Gaussian distribution of the end-to-end-distance 2/3
3L 2 \
exp(-^j
(5-46)
We assume that the free energy of the polymer is a known function of the end to end distance F(T, L) and the value corresponding to the thermodyamic equilibrium is the average over L.
F{T) = f dLF(T, L)W(L) * G(L)
(5.47)
where G(L) is a geometric factor. Let us assume now that the reacting molecule can be bound to the two ends of the polymer. We may imagine that the molecule is hold between the ends of the polymer like a tool in a lathe; the binding energy is Eb- In difference to the ring model explained in the previous chapter, the ring has now a free form, this causes strong entropy effects. In the free state where the reacting particle is far from the polymer, the potential barrier is rather high, so that transitions are practically excluded. In the bound state the reacting molecule is coupled to the polymer and its barrier depends on the end-to-end distance L of the polymer. We may assume that the constants a, b are dependent on the endto-end distance: a(L),b((L) with the condition a(L) —» a,b(L) —> b. We call our model in the following a Kramers polymer. In the bound configuration of polymer and reacting molecule the latter is hold between the ends of the polymer like a tool in a lathe. Due to the parameter dependence on L the potential is getting L—dependent
C/(g;L) = i 6 ( L ) g 4 - i a ( L ) g 2
(5.48)
This may lead to a lowering of the potential barrier
«-W
(M,)
Correspondingly we may find an enhancement of the rates. The main effect however is of entropic nature. In the state with the end-to-end distance L the entropy of the polymer is much smaller than in the thermal equilibrium and its free energy is higher F(T,L) > F(T). In this way we obtain the free energy barrier
A polymer reaction model including entropy
effects
SF(T, L) = [F(L) - F{T, L) + Eb] + AU(L)
177
(5.50)
In the case that the inequality [F(L) — F(T, L) + Eb] < 0 is fulfilled, this corresponds to an effective lowering of the barrier by entropic effects. Cutting the bond gives freedom to the polymer which will assume in a short time the equlibrium distribution of L and the equilibrium value of the free energy. The bound polymer has so to say an reservoir of free energy which can be used for overcoming the barrier. As a first estimate we get for the rates in the strongly damped case
If the polymer is sufficiently large the dependence on the end-to-end distance may be rather strong and the reservoir of free energy may be high. It was the aim of this section to study entropic effects on a simple artificial model. We have seen that under certain circumstances we may expect strong effects of the polymer confirmation changes on rates of bond breaking in an associated reacting molecule.
References P. Bak (1996): "How Nature Works", Springer, New York. J. Bernasconi, T. Schneider, Eds. (1981): "Physics in One Dimension", SpringerVerlag, Berlin-Heidelberg-New York. H. Bolterauer, M. Opper (1981): "Solitons in the Statistical Mechanics of the Toda Lattice", Z. Phys. B 42, 155-165. H. Buttner, F. Mertens (1979): Solid State Comm. 29, 663-673. A.Yu. Chikishev, W. Ebeling, A.V. Netrebko, N.V. Netrebko, Yu.M. Romanovsky, L. Schimansky-Geier (1996): "Stochastic cluster dynamics of macromolecules", in: Nonlinear Dynamics and Structures in Biology and Medicine: Optical and Laser Technologies, V.V. Tuchin, editor, Proc. SPIE 3053, 54-70. A.Yu. Chikishev, W. Ebeling, A.V. Netrebko, N.V. Netrebko, Yu.M. Romanovsky, L. Schimansky-Geier (1998): "Stochastic cluster dynamics of macromolecules". Int. Journal of Bifurcation & Chaos 8, 921-926.
178
Excitations
on rings of molecules
T. Dauxois, M. Peyrard & A.R. Bishop (1993): "Dynamics and Thermodynamics of a Nonlinear Model for DNA Denaturation", Phys. Rev E 47, 648-695. J. Dunkel, W. Ebeling, U. Erdmann (2001): "Thermodynamics and transport in an active Morse ring chain", Eur. Phys. J. 24, 511-524. J. Dunkel, W. Ebeling, U. Erdmann, V.A. Makarov (2002): "Coherent motions and clusters in a dissipative Morse ring", Int. J. Bifurcations & Chaos, in press (2002). W. Ebeling, Yu.M. Romanovsky (1985): "Energy Transfer and Chaotic Oscillations in Enzyme Catalysis". Z. Phys., Chem. Leipzig 266, 836-843. W. Ebeling, M. Jenssen (1988): "Soliton dynamics and energy trapping in enzyme catalysis", Z. Phys. Chem. 1, 269-279; Physica D 32, 183-193. W. Ebeling, M. Jenssen (1988): "Trapping and fusion of solitons in a nonuniform Toda lattice", Physica D 32, 183-193; Physica A 188, 350-355. W. Ebeling,W., M. Jenssen & Yu. M. Romanovskh (1989): "100 years Arrhenius law and recent developments in reaction theory" , in: In: Irreversible Processes and Selforganization, eds. W. Ebeling and H. Ulbricht, Teubner, Leipzig, p. 7-24. Ebeling,W. & Schimansky-Geier,L. (1989): In "Noise in Nonlinear Dynamical Systems", F. Moss, P.E.V. McClintock, eds., Cambridge, Cambridge University Press. W. Ebeling,W. & M. Jenssen (1991): "Soliton-Assisted Activation Processes" Ber. Bunsenges. Phys. Chem. 95, 356-362. W. Ebeling, J. Ortner (1998): "Quasiclassical theory and simulations of strongly coupled plasmas", Physica Scripta T75, 93-98. W. Ebeling, M.Jenssen (1999): "Brownian particles with Toda interactions - a model of nonlinear molecular excitations", in: Proc. Int. Conf. on Nonlinear Optics Saratov 1998, SPIE 3726, 112-123. W. Ebeling, A. Chetverikov & M. Jenssen (2000): "Statistical thermodynamics and nonlinear excitations of Toda systems" Ukrainian. J. Phys. 45, 479-487 W. Ebeling, U. Erdmann, J. Dunkel, and M. Jenssen (2000): "Nonlinear dynamics and fluctuations of dissipative Toda chains", J. Stat. Phys. A 101, 443-457. U. Erdmann, W. Ebeling, L. Schimansky-Geier, F. Schweitzer (2000): "Brownian
A polymer reaction model including entropy effects
179
particles far from equilibrium", Eur. Phys. J. 15, 105-113. H. Frauenfelder & P.G. Wolynes (1985): "Rate theories and the puzzles of protein kinetics", Science 229, 337-345. H. Frauenfelder, N.A. Alberding, A.Ansary et al., J.Phys. Chem. 94, 1024 (1990) P. Gruner-Bauer, F.G. Mertens (1988): Z. Phys. B 70, 435-445. B. Havsteen (1989): "A new principle of enzyme catalysis: coupled vibrations facilitate conformational changes". J. Theor. Biol. 140, 101-109. B. Havsteen (1991): "A stochastic attractor participates in chymotrypsin catalysis. A new facet of enzyme catalysis". J. Theor. Biol. 151, 557-571. D. Helbing (1997): "Verkehrsdynamik", Springer, Berlin. D. Helbing (2001): "Traffic and related self-driven many-particle systems", Rev. Mod. Phys. 73, 1067-1141. M. Jenssen (1991): Phys. Lett. A 159, 6-16. M. Jenssen, W. Ebeling (1991): In: "Far-from-Equilibrium Dynamics of Chemical Systems", eds. J. Popielawski and J. Gorecki, World Scientific, Singapore. M. Jenssen, W. Ebeling (2000): "Distribution functions and excitation spectra of Toda systems at intermediate temperatures", Physica D 141, 117-132. M. Karplus (1982): "Dynamics of proteins". Ber. Bunsenges. 386-240.
Phys. Chem. 86,
Yu. L. Klimontovich (1995): 'Statistical theory of open systems" Kluwer Academic Publ., Dordrecht. J.A. Krumhansl, J.R. Schrieffer (1975): Phys. Rev B 11, 3535-3545. H.P.Lu, L.Xun, X.S.Xie (1998): Science 282, 1877-1887. J.A. McCammon, S.C. Harvey (1987): "Dynamics of Proteins and protein acids", Cambridge University Press. F.G. Mertens, H. Biittner (1986): "Solitons on the Toda lattice", In: Trullinger et
180
Excitations
on rings of molecules
al., eds., I.e. V. Muto, A.C. Scott, P.L. Christiansen (1989): Phys. Lett. A 136, 33-43. J. Ortner, F. Schautz, W. Ebeling (1997): "Quasiclassical molecular dynamics simulations of the electron gas: dynamic properties", Phys. Rev. E 56, 4665-4670. P.J. Park, W. Sung (1998): "Polymer release out of a spherical vesicle through a pore", Phys. Rev. £ 5 7 , 730-734. R. Regeida, A. Romero, A. Sarmiento, K. Lindenberg (1999): J. Chem. Phys. I l l , 1373-1383. Yu.M. Romanovsky (1997): "Some problems of cluster dynamics of biological macromolecules". In: Stochastic Dynamics, L. Schimansky-Geier, T. Poeschel, Eds. Ser. Lecture Notes on Physics, Springer Verlag, Berlin, p. 140-152. T. Schneider (1986): "Classical statistical mechanics of lattice dynamic model systems", In: Trullinger et al., eds., I.e. W. Sung, P.J. Park (1996): "Polymer translocation through a pore in a membrane", Phys. Rev. Lett. 77, 783-786. N. Theodorakopoulos (1984): Phys. Rev. Lett. 53, 871-881. M. Toda (1983): "Nonlinear Waves and Solitons" Kluwer, Dordrecht. M. Toda, N. Saitoh (1983): J. Phys. Soc. Japan 52, 3703-3713. S.E. Trullinger, V.E. Zakharov, V.L. Pokrovsky, Eds (1986): "Solitons", North Holland, Amsterdam. G.R. Welch, B. Somogyi, S. Damjanovich (1982): "The role of protein fluctuation in enzyme action". Prog. Biophys. Molec. Biol. 39, 109-146.
Chapter 6
Fermi resonance a n d K r a m e r s p r o b l e m in 2-d force field S.V. Kroo, A.V. Netrebko, Yu.M. Romanovsky, L. Schimansky-Geier 6.1
2-d potential landscape and Fermi resonance
In the previous Chapters we presented several problems related to the Kramers problem or specification of Arrhenius law. Among them we considered the problem of a "test particle" transfer from one potential well into another. In particular we studied in Chapter 6 a potential landscape with the property that the returning force in x— direction Fx = — ^ depends only on x, and the corresponding force Fy = — ^f- depends only on y. However, the configuration of the real potential landscapes may be quite different. In Chapters 1 and 4 we studied already in brief the effects of a nonlinear coupling between x and y. In particular we pointed out that Fermi resonance may lead to an enhancement of transition rates. Here we continue to study this problem in more detail. In a recent paper (Shidlovskaya, Schimansky-Geier & Romanovsky, 2000) the problem of the substrate oscillations in the active site of an enzyme (Figs. 1.3, 1.10, 1.11 and 6.1) was considered. Examples of proton oscillations in two-well and threewell potential relieves can be found in (McDonald, Thorson & Choi, 1993; Netrebko et al., 1994). In particular, the distortion of the symmetry of the two-well potential in the system of H-bonds in the active site of serine hydrolases (chymotrypsin and acetylcholinesterase) results from substrate binding. The allowance for the rotation of the cluster planes relative to each other leads to the relationship between the returning forces. In this case the potential landscapes can be asymmetric (Fig.6.2) (Rubin, 1987). Note the recent studies of the 2-d systems consisting of the chains of long molecules subjected to the impacts of the water molecules. The static characteristics of such systems were calculated in (Shaitan, Pustoshilov, 1999; Shaitan, Ermolaeva, Saraikin, 1999) (see also Chapter 10). The physical pendulum is a classical example of a mechanical system in which each component of the returning force depends on both coordinates due to nonlinear coupling of both degrees of freedom (Fig.6.3). Let us assume that there are limitations on the angle, then the potential land181
182
Fermi resonance and Kramers
problem
4) proton transfer 2.P5 X
His57-,^^N'&
•
r N2^
-H-0-Serl95 2.0 k.
_
.2.8 A
-C— N ~ d
S;o:
\ /
substrate \ /
\2.8 A\
\ N - /
V
Aspl02 \ 2) H-bonJ break \
0
\C-c-c>" I I I Ser214
5) peptide bond break
') "^placement
Fig. 6.1 Catalytic triad or "charge relay system" in serine proteases. Substrate is fixed in the active site by a system of H-bonds allowing oscillations of small amplitude.
4>
I
-60
0
VN
60
120
V
180
'•P(N-C=t), rpaa b) Fig. 6.2 (a) Plane conformation of methylamide-N-acetyl-L-alanine molecule; (b) Diagram of the potential surface of methylamide-N-acetyl-L-alanine molecule with conformation energy (in kJ/mole) level lines.
scape U(
183
Basic 2-d cluster model
////
\
\
/
\ 1 1
Fig. 6.3
y
y
X
Pendulum with two vibrational degrees of freedom.
analyzed the Raman spectra of gaseous carbon dioxide by means of perturbation method. In this molecule the fundamental frequency of one of the modes (1330 c m - 1 ) is close to the doubled frequency of another fundamental mode (667.5 c m - 1 ) which results in appearance of the additional bands in Raman spectra. The classical analogy of this effect (redistribution of energy in classical pendulum shown in Fig.6.3) was demonstrated by Volkenstein (1947). Several authors (e.g., (Volkenstein, 1975) showed that Fermi resonance is possible also in protein structures because the frequency of N-H stretching vibration coincides with the doubled frequency of amide II vibration. At present Fermi resonance is demonstrated for different systems with self-modulation (see, for example, (Pippard, 1985, 1989). Continuing the principal study of Fermi resonances in section 1.4. we consider now in detail an example of the a simple mechanical system (Fig.6.4) in which Fermi resonance takes place (Netrebko et al., 1994; Shidlovskaya, Schimansky-Geier & Romanovsky, 2000). The discussion follows the schema: 1. Fermi resonance in a conservative system. 2. Induced oscillations and Fermi resonance in a system with losses and external harmonic force. 3. System with losses under the action of noise. 4. Test particle (TP) escape from a 2-d potential well in the case of resonance redistribution of energy between the degrees of freedom.
6.2
Basic 2-d cluster model
Consider dynamics of a classical particle bound by four linear springs to four immobile walls (Fig.6.4) The dynamics of the system in y\, ?/2 coordinates is given by
m
tli.
-
dt
2
_
d U
(Vl'Vl)
dyi
_
d
Udist (2/1 , V2 )
dyi
•_ i o
(a i N
184
Fermi resonance and Kramers
problem
t*
Fig. 6.4 Schematics of the basic 2-d cluster model; m is the cluster mass; ki (i = 1...4) are the rigidities of the springs simulating the H-bonds under linear approximation.
where the second terms in the right-hand sides are determined by the local distortions of the potential landscape in the presence of rigid sphere. In the case of linear forces (springs) and the absence of distortions and external action the potential function of the system is given by:
U(yi,y2) = fy(y/vl + (l-V2)2-i)
uV / +k{\Jyl
+ fy (Jvl
V + {i + yi?-i)
+ {I ~ Vif - l) +
u\ I +!%{\/y22 + (i +
V
(6 2)
-
yi)2-i)
here 21 is the distance between the walls. Note that in spite of the linear forces the dynamics of the system is nonlinear. We use this simplest model for the study of the dynamical aspects of behaviour of the unchanged part of the substrate bound inside the pocket of the active site. Immobile walls in Fig.6.4 simulate the nearest massive immobile clusters of the active site. The force coefficients are determined by the number and the directions of i7-bonds. The distortions of the potential landscape in the area of small oscillations of the substrate result from subglobular or cluster motility. This simplest model can be used for the description of some dynamical features typical of more complicated cluster models. It can be considered as a part of two or several crossing cluster chains in models of /3-strands and /^-sheets (Fig. 1.2). The model can be used for the study of low-frequency processes in systems of more heavy elements of the secondary structure (12 /3-strands connected with each other form two /3-sheets in CT molecule) (Netrebko et al., 1996, Romanovsky, 1997). As a matter of fact we should consider not 2-d but 3-d case, in particular, possible rotating motions of aromatic rings of the substrate inside the CT active site (Shidlovskaya, Schimansky-Geier & Romanovsky, 1999). We assume that the behaviour of the system in 3-d (nonfractal) potential landscape is similar to that in 2-d landscape but the behaviour in 1-d system is essentially different.
185
Basic 2-d cluster model
The system of equations can be linearized in the case of small oscillations 3/i) 2/2 < / (e < 0.1, where e = ao/Z, a 0 is the maximum vibrational amplitude and I is the length of H-bond). Two uncoupled modes with the natural frequencies w\ = \/(k2 + k4)/m , w2 — -\/{ki + k3)/m are observed. Nonlinear features of the model and coupling between two degrees of freedom can be considered analytically by solving the system (6.1) using the averaging method in the case of small vibrations e < 0.1. Under this approximation the resonance relation between two fundamental frequencies can be used. Even in this simplest case the motions can be quite complicated and redistribution of energy between the vibrational modes can be observed like in the classical model of the Fermi resonance in the unharmonic potential landscape. Let us substitute (6.2) in (6.1) and introduce the new variables: Vi V2
£2/2-
The expansion of the equations in the Taylor series in the point of equilibrium yields the expressions as follows (hereafter we omit asterisks in y\ and y^)•
dt2 +e2
+ co2yx
=
e (fci - k3) yiy2 + 2 (k2 - h)y2
- \ (fci + fc3) y\ + (fci +k2 + k3 + k4) 2/12/1
+
+ ... , (6.3)
^ 1 , 2 +62
=
e (k2 - k4) 2/12/2 + 5 (fci - k3) y\
- \ {k2 + kA) 2/| -f (fci + k2 + k3 + k4) y2y\
+
+ ... ,
where fej, i = 1...4 - are the force coefficients of springs, w\ = \Jk2 + k4 and w2 = y/ki + k3 are the natural frequencies. Equations (6.3) use the dimensionless units: I = 1 corresponds to the length of H-bond and m — 1 is the cluster mass. Below we consider the case k\ ^ £3 or k2 ^ k4 and the most interesting case of the internal (autoparametric) resonance when the value of one natural frequency coincides with the doubled value of another one: a>2 = 2wi or u\ = 2u)2. Note that under approximation of small vibrational amplitudes analogous equations can be written for nonlinear springs (for example, when the H-bonds are simulated by Morse or Lennard-Jones potentials). For example, if each spring in Fig.6.4 corresponds to one or several H-bonds described by Morse potential
Ui (r) = Doi ( l
-di(r-iy
Doi,
then the following system can be obtained instead of (6.3):
(6.4)
186
Fermi resonance and Kramers
Vi + (k2 + k4) yi=e +£Z
(fci - k3) yxy2 +
ki 2
*y - (a2 - aA) yj
2
+
2
M ^ i + it A yxy\ + (t <*t) ylV 2 - (ft +ft) + yf
j/2 + (Ai + k3) yi=e
+e<
hi2 -
problem
_«2.
Uk2 - k4) yxy2 +
fcl
k 2
3y2 - (a x - a 3 ) y | +
^ % 2 3 + ( E * i ) J/iIfc + (E<*) y\y2 - (ft + ft) y\
+ •
The terms with the factors a* and ft are missing new in the system (6.3). In this case 3 7 ki = 2dD0i, an - -kid, Pi — 7,kiQ, C = dl. The typical values for an H-bond are (see, for example, (Dashevskii, 1987)) I = 0.17 -r 0.18 nm, d = 30 -f 5 nm" 1 . That is why C ~ 10. As e = 0.1, only the terms of the order of e and e2 must be taken into account in (6.3) (the terms of higher orders must be taken into account in the case of large shifts). 6.3
Analytical study
We use approximate asymptotic methods (Bogolyubov, Mitropolskii, 1974) to demonstrate a periodic intermode transfer of the vibrational energy in the conservative system (6.3) at the frequency ft. Following the method of slowly varying amplitudes and phases we present yi (i=l, 2) in a way as follows: yt = Ai(t) cos ipi, Ipi = Wit + tfi (t) . Here Aj(t) and
((aAiA2 cos ipi cos ip2 + M 2 c o s 2 ^2) sini/ ) i) T ,
A'2 —
((cAiA2Cosij)icosil)2 + dA\cos2ipi)
sini/j2)T,
LO2
V>2 =
-
AiU)!
^2^2
((aAiA2
cos^i cos^2 + bA\ cos2 ^2) cosV , i) T ,
((c.Ai.^cosV'i cosV>2 + cL4 2 cos 2 ^i) cos^2) T ,
Analytical
study
187
where a = ki - k2, b = (k2 - k4)/2, c = k2 - k4, d = (ki - k3)/2. Brackets stand for time averaging. Averaging over the period T\ = 2TT/WI (in the case when UJ2 = 2wi) yields: ' dA1/dt = - C A i ^ 2 s i n $ , dA2/dt = | C A 2 s i n $ ,
(6.5)
d$/dt = 2C (-A2 + J ^ i J cos$, where e
C
k\ — 4
(6.6)
(6.7)
$ = 2(/>i - 02The following integral of the system (6.5) can be easily obtained:
Al + 4Al = Al
(6.8)
Multiply the last equation in (6.5) by cos $ in order to obtain the relationship between the phase $ and amplitudes A\ and A2:
dZ/dt
=2C-
\-z
8
4 l . + ^i 8A1
(6.9)
Z = sin $ As dA2 = 1 CA\Z, dt ~ 4 then ZdZ 1 - Z2
-8Al + A\ dA2. ^2A2
Using (6.8) to exclude A\ we obtain:
dQln(l-Z 2 ))
<M2
4dAl
A2 + AI-4AV
- - In (1 - sin 2 $ ) + InD = In A2 + In (A20 - 4A\) where D is the integration constant. Finally we have:
188
Fermi resonance and Kramers
D — A2A\
COS
problem
$.
(6.10)
Quadratures for A\(t), A2(t), and $(£) can be obtained but the corresponding expressions are quite cumbersome. That is why consider first the stability of the stationary solution of the system (6.5) in order to reveal the possible periodic changes of A±, A2, and $. Stationary values of <&,Ai,A2 are given by algebraic equalities: sin $ = 0
(cos $ = 1) ,
8Al = A\.
Using the integrals of the system (6.8) and (6.10), we obtain: M =^
,A2
$ = 0.
2V3'
Thus, the stationary solutions depend on AQ and, hence, on the initial conditions. Linearization of the system (6.5) near the values Ai, A2, $, yields the following system for small deviations ai, a2, $:
dai/dt — da2/dt =
-CAiA2$, \CA\&,
(6.11)
d$/dt = 2 C { | ^ 0 ! - 2a2\ . Characteristic equation for the complex frequency p follows from (6.11):
so that p —
P
0
0 -Cs/2
v AC
JCAQ.
CAl-± °3V2 -CAl\ p
= 0,
or p J + CMjS = 0
Substituting C from (6.6) we obtain:
fe
n^A.T1*1 4 sjk2 + ki
(6.12)
Resonance energy exchange is characterized by modulation frequency fii that is proportional to the difference between the force coefficients of the "high-frequency" mode \ki — k^\ and inverse proportional to the root of their sum v f c i + ^ 3 (i-e- ^ l is inverse proportional to w\ and u>2). This conclusion is valid at various initial conditions, in any degree of freedom, and at any values of the parameter k, corresponding to the resonance condition except for k\ = k%. In the latter case we have to take into account additional terms in the expansion in the powers of e.
Analytical
189
study
Let us calculate the modulation frequency fl in (6.5) without employing the linearization procedure. Using the system integral D we can write:
.
yJ{A\A2f-D* A\A2
Substituting this expression into the second equation of the system (6.5) with the allowance of (6.8) we obtain:
dA -^2
a
dt
= = _c_
4A2
r^77->. T^J ^AUAl-4Alf-D*.
The modulation period is given by the expression:
^
^AUAl-AAlf-D*
where A2min and A2max are the minimum and maximum values of the envelope A2. These quantities can be determined as follows. Since |cos$| < 1 and D — A2A\cos§, \D\ A*{Al-lAl)
< 1.
Therefore, _ A0 2min — ~^
(-K a\ COS I — + — I ,
A Ao 0
fir
a\
where cos a = 3\/^\D\/AQ. Going back to determining the modulation period and introducing the variable rj = A\ — 4A\ we obtain:
T •
2 C
\2 /"* 2 m i n J J-v3
I
dr) + AW
- 4L>2
2max
The denominator of the integrand equals zero at: „ 2 2 a0 Al B 0 = - - j 4 g c o s - ^ + -£, 2
A2
/7T
a0\
B2 = - ^ c o s ( - -
Al
y
)
+
^ ,
190
Fermi resonance and Kramers
problem
where
As a 0 = 2a, B1 T =
=
54£>2 cosa 0 = —75 1, B0 < B± < B2. A o A 2 - 4A22max a n d B2 = A 2 4A22min,
dy
2 f CJ
=
2_
2
C
^B2 - B0
V(£o - V) (Bi - V) {B2 - r?)
K [mh
where K(m) is the elliptic integral of the first type:
J y/(l-x2)(l-mx2) B2-Bi B2 - B0
2^oV 2 VZctgtf
m ;
+1
Finally, we have: T
=
2TT
^/(2n-l)!! = > A>C^cos(f-f)^oV 2«n! 7
ao = arccos
VVZctg{a0/Z) + 17 '
(T-0
When linearizing the system we assume that cos$ = 1, hence, D — A\A2 -^9= and a 0 = 0. In this case T = A and fi = .4oC = e A 0 i J * L ^ M , 3V3 ^oO 4 ^ 2 + /.4 which coincides with (6.12). 6.4
Numerical study
Below we present the results of numerical solution of the equations of the base system. There are two resonance regimes of effective energy exchange between two vibrational modes at 2 : 1(0.5 : 1) ratio of the natural frequencies in the given 2-d conservative system. Figures 6.5 and 6.6 show the time-dependencies of the coordinate yi and the envelopes of the coordinates y\ and y2 of the imaging point. It was assumed that the cluster mass is m — 100 a.u., k\ = k, k^ = 2k, k2 = 0, 5A;, ki = 0,25A;, where k = blN/m is the force coefficient of a single H-bond. Note that the cluster mass is close to that of the substrate fragment bound to the groups of the CT AS pocket. The values of fc; were chosen to meet the conditions of the intermode energy transfer. In fact, the real values of ki depend on
Numerical
study
191
yi 0,1-
—^fll m»-
i
-0,1-
100
200
Fig. 6.5 Vibrations in y\ axis at the carrier frequency ui\. Initial conditions: yi(0) = 0 and o J/2(0) = 0.1. Time scale: 1 = 0.1 ps; amplitude scale: 1 = 2 A-
0.12-1
0.08
0,04
4000
8000
Fig. 6.6 Positive envelopes of the amplitudes y\ and 2/2 (in dimensionless units). Initial conditions: 2/i(0) = 0 . 1 and 2/2(0) = 0. Carrier frequencies are not shown. Scales and parameters are the same as in Fig.6.5.
the plane in which the oscillations take place and the specific conformation of the enzyme-substrate complex. The latter can slowly (relative to the ligand oscillation period) vary in time. As the formula (6.12) is obtained using linearization, it remains valid only at small deviations from the stationary state. To illustrate this fact, let us find numerical solution of the system (6.5) setting the initial conditions in a way as follows. Let the initial phase $ be always equal to zero and the initial amplitudes change in such a way that the energy of the system (integral AQ) always remains constant. In this case the stationary values of A\, A2 , and $ do not depend on the initial conditions: Ax = A0
$ = 0. 2^3'
The deviation of the system from the stationary state can be characterized by the parameter d that is equal to the ratio of the initial and stationary value of one
192
Fermi resonance and Kramers
problem
l.l - i
l.o -
_>
0.9 -
./
0.8 -
l-^•
/
0.7 -
\
1
/
•
1
0.6-
°-H—'—i—' 0.0
0.2
i—'—i—'—i—'—i—'—i— 0.4
0.6 d
0.8
1.0
1.2
Fig. 6.7 A plot of the ratio fi/fi; versus the parameter d(fi) is the frequency determined by the numerical solution of the equations (6.5), H; is determined by the formula (6.12)). AQ = 0,5; £=0,01; fci = 1; fc2 = 0,5; k3 = 2; fc4 = 0,25.
of the amplitudes, e.g. Ai (d = 1 corresponds to the stationary state): d = Ax/A^l. Fig.6.7 shows the dependence of the ratio fl/fij on the parameter d(fl) is the frequency determined by the numerical solution of the equation 6.5); fi; is given by the formula (6.12). Thus, the frequency determined by the numerical solution of equations (6.5) under small deviations from the stationary value coincides with the frequency given by the formula (6.12). We compared the results of numerical solution of equations (6.1) and (6.5). A good coincidence of the redistribution frequencies is possibly limited by the accuracy of the numerical scheme. For the case of the Morse potentials we also observed the regimes of selfmodulated oscillations representing a particular case of the Fermi resonance (although at different frequencies) (Netrebko et al., 1994, Ebeling et al., 1994). It is demonstrated that there are three resonances instead of two corresponding to the case of linear springs. Their positions are substantially different from the positions of two peaks in the two-resonance case. Even if a
Stochastization
of the
193
vibrations
yi
yi
yi
a)
b)
c)
Fig. 6.8 Various trajectories of the imaging point and the boundaries of the areas of possible stochastization.
6.5
Stochastization of the vibrations
Note that the described strongly modulated regimes are observed in the case of very small vibrational amplitudes (in molecular dynamics the amplitudes of such processes do not exceed 10% of the H-bond length which corresponds to 0.1^-0.3/9^4). If the amplitudes are larger than 0.51 (or e > 0.5) the stochastic behaviour can be observed. The possible stochastization areas can be calculated for different potential relieves using the Toda method (see chapter 3). Figure 6.8 shows the areas of possible stochastization and the trajectories of the imaging point for the case of the Fermi resonance in the system (6.1) with linear bonds at large amplitudes (0.5/ — 1.0Z). The motion is regular if the imaging point does not cross the areas of possible stochastization (Fig.6.8a). Stochastization is possible if the imaging point passes through Toda area (Fig.6.8c). At the same time the latter condition is not sufficient for chaos formation (Fig.6.8b). Note that Toda areas for the system with Morse potentials (6.4) differ from those for the system with the corresponding quadratic potentials (Fig.6.9) (Nunez-Yepez et al., 1990). Chaotic behavior can be observed also in the case of the small amplitudes e < 0.1, if we introduce distortions into the potential landscape or introduce a solid cluster into the oscillation area (Netrebko et al., 1994).
6.6
Basic model including damping and external harmonic action
Up till now we studied a conservative system. However, the real oscillations of clusters decay in time and they are subjected to the action of external (in the general case, random) forces. Let us consider the system (6.3) in the case of external regular excitation and damping in only one degree of freedom y\. Leaving only the terms linear in e and assuming that m = l w e obtain:
194
Fermi resonance and Kramers
problem
0.5-
yi °°-D
0.0-
•0.5-
-1.0-
I CDC
) VJ
> )DD r yi c)
Fig. 6.9 Stochastization areas for a system with (a) quadratic and (b) Morse potentials. Exponential instability is possible for the shaded areas, (c) Superimposed boundaries of the stochastization areas for the same potentials.
d2 2/1 dt2
,
2
( « i - f c a ) 2/12/2 +
d2V2 ^
, 2 + ^ 2 / 2 =
£
(k2
—»—2/2
hi)
eF0cos(wt + <$>)-e5^,
2/12/2+2
( fc i
(6.13)
~ k3) 2/i
Here Fo and u are the amplitude and frequency of the external periodic force and 5 is the damping decrement. We take into consideration only the terms of the order of e. Thus, the following relationships for 8 and FQ are obtained:
|&i-fc3|4i42~.Fo~<MiWi,
(6.14)
where A\ and A\ are the maximum amplitudes of the envelopes y\ (t) and 2/2(0For ui = uii we search for a solution of (6.13) in the form: yt = Ai cos (cjit + ifi). In the resonance case u\ = w = W2/2 one can average the equations for slowly varying envelopes (6.13) and obtain an expression for the modulation frequency: \/3 \h-k3\F0 fii = e ' 4v^2 k2 + k4 S '
(6.15)
The maximum amplitudes of the envelopes are given by:
A?1 = ^ 5w
and
Al2 = ^ A \ . 2v^
(6.16)
Computer simulation
of the nonautonomous
system with
damping
195
Fig. 6.10 Positive envelopes of the amplitude y\ and 3/2 for the case FQ = 0.05, a = 0.5; initial conditions: yi(0) = 0 and 2/2(0) = 0. The scales and parameters are the same as in Fig.6.5. An intermode energy exchange at the frequency Cliis observed.
If fci —¥ &3 we have to take into consideration the second order approximation. On the other hand, if the external force is large, the relationship (6.14) is not valid because the nonlinear term becomes larger than the linear one: 1*1 -kz\A\ |*i-fc3|
»<5w, 1
F0
» 6w.
That is why one can expect complication of the modulation regime in the case of large FQ. 6.7
Computer simulation of the nonautonomous system with damping
Figures 6.10 and 6.11 show the solutions of the system (6.13) at small amplitudes of the external force under various initial conditions. It follows from these solutions that the induced oscillations with the envelope modulated at the frequency Oi are established in the system after a certain transient process. The modulation regime becomes more complicated if the value Fo increase 10 times: "double modulation" appears at the frequencies fix ~ e and O2 ~ e 2 (Fig.6.12). This result can be obtained by averaging with the allowance for the terms of the order of e2. The same result can be envisaged for the system in which the Morse potential is used instead the quadratic one. To specify the behavior of the system under large perturbations we solved the initial nonlinear equations (6.1) with the allowance of the periodic force and damping: d2y1 _ dt2
dU(yuy2) dyi
eF0 cos (ut + $ ) - eS
dyi dt '
196
Fermi resonance and Kramers
10.000
20,000
problem
30,000
40.000
50,000
Fig. 6.11 Positive envelopes of the amplitude y\ and 1/2 for the case Fo = 0.05, a = 0.5; initial conditions: 2/1 (0) = 0.3 and 2/2(0) = 0. In spite of the higher initial energy the type and parameters of oscillations are the same.
.
10-000
"b e a t i n g i " at Q -t
20.000
30.000
4 0 . 0 0 0
50.000
b) Fig. 6.12 Positive envelopes of the amplitude y\ and 2/2 for the case Fo = 0.5, a = 0.5; initial conditions: 2/1 (0) = 0 and 2/2(0) = 0. The scales and parameters are the same as in Fig.6.5. An intermode energy exchange at the frequency fii and modulation at the frequency 02 ~ e 2 are observed.
d2y2 dt2 where U{y\,y2)
=
dU(yi,y2) 0y2 '
is determined by the relationship (6.2).
Computer simulation
of the nonautonomous
system with
damping
197
y2ioo
T
T
-1
0.25
w/w,
0.00
w/w.
4
0 (C)
Fig. 6.13 Positive envelopes of the amplitudes j/i and j/2 (a), trajectory of the imaging point (b) and the normalized spectral power density of the time-dependence of coordinates (c). I — 100, 8 = 1, 3/i(0) = y 2 (0) = 0, F o = 3 0 .
If w = wi there are two regimes determined by the values of Fo, 6, and I. The first regime exhibits a regular energy exchange between the modes and is analogous to that considered above. Figures 6.13 and 6.14 show time-dependencies of the positive envelopes y\ and 2/2 at two values of Fo. Note that the maximal amplitude of oscillations along 7/2 axis can be larger than that for oscillations along y\ axis (Fig.6.14). There is no energy exchange between the modes under stationary conditions for the second regime (Fig.6.14). Further studies are needed to determine the criterion by which one can differ between the first and the second regimes. Note that there is no stochastization of motion in this system even for the deviations from the equilibrium that are larger than the length of the spring. Figure 6.16 shows the trajectories of the stationary motion of the imaging point for two values of Fo. If the system (6.13) uses noise instead of the harmonic force, one can also expect the regime of energy redistribution between the modes "j/i" and "2/2"- For the high quality system (5 ) we must use < F (wi) > ^
I/^FK)
2S
= \/5'F(WI)AW,
instead of F0 for estimating the statistical parameters A\, A\, and fii. Here SF (WI)
198
Fermi resonance and Kramers
problem
y 2 100-
S» 0.50
Fig. 6.14 Positive envelopes of the amplitudes y\ and J/2 (a), trajectory of the imaging point (b) and normalized spectral power density of the time-dependence of coordinates (c). I = 100, S = 1, yi(0) = y 2 (0) = 0, F 0 = 20.
is the spectral density of F(t) at the frequency uii and Awi — uii/Q = 28 is the spectral bandwidth of our vibrational system. Using (6.15) and (6.16) we obtain:
/fl~I,/2*H, u>
,A
°' ~ ^ v f • 2UJ
v^j,2 "
l|fci-fc 3 |./35F(a>i)
n
11
V o
£
4 A;2 + A;4
The increase of T/SF (U)/S leads to the proportional increase in RMS values of ^4°! -4°' a n d ^ i - The corresponding distributions must be estimated by Rayleigh formula. Figures 6.17 and 6.18 demonstrate the action of the white noise F(t) in only one direction (yi and y2, respectively). It is seen that a transient process is followed by the regime of stochastic modulations. The envelopes of the amplitudes y\ and 2/2
Kramers problem for 2-d potential
199
landscape
y,ioo 4U -
-
^ y
20-
fh
r i
i
0
i
I
'
I
200 0
1000
'
3000
-100
-50
50
0
100
(b)
(a) 1.00 -
1.00 —i
0.75 -
0.75 -
Syi0.50-
S y 0.500.25 -
0.25 -
w/w
'
1
1
1
'
w/w 1
1
1
1
1
1
I
(c)
Fig. 6.15 Positive envelopes of the amplitudes y\ and 3/2 (a), trajectory of the stationary motion of imaging point (b) and normalized spectral power density of the time-dependence of coordinates (c). I = 100, <5 = 1, j/i (0) = 2/2 (0) = 0, F 0 = 10.
change in opposite phase like in the case of the nonstochastic regimes considered above. It is likely that the real systems exhibit such types of motion. Note that the colour noise action seems to be more probable. To our opinion, the Fermi resonance in the systems with damping and external force action is important for molecular dynamics of macromolecules.
6.8
Kramers problem for 2-d potential landscape
We meet the problem of the particle escape from the minimum of 2-d potential landscape each time when it is necessary to estimate the time of transition of a ligand from one minimum to another or the time of the escape of the reaction products from the AS pocket. In 1-d case such estimates for noninteracting particles can be done using Kramers formula. This problem was discussed in details in Chapter 1. Substantial differences between 2-d and 3-d cases may result from the possible energy exchange between the vibrations in different directions. In the case of the 2d potential landscape the consideration is reduced to solving the problem of escape into 4-d space (see, for example, (Tikhonov, Mironov, 1977)). This problem was considered in Chapter 3 under the assumption that the Fermi resonance conditions
200
Fermi resonance and Kramers y,100
problem
y,100
-100
50
100
1.00
Sy 0.50 -
Sy 0.50
0.25
w/w,
0.00
w/w.
x_ui_
S„ 0.50
S» 0.50
'2
0.25
0.25 -
w/w,
0.00
w/w,
.I, M, 2
(a)
(b)
Fig. 6.16 Trajectories of the stationary motion and the normalized spectral power density of the time-dependence of coordinates at / = 100, S = 1, FQ = 80 (a) and FQ = 85 (b).
are not met. However, the equations for the escape probability and the mean escape time are rather complicated and can be solved only numerically. We estimated above the value of the mean amplitudes Ai and A^ and the beat frequency (see formulas (6.15) and (6.16)) for the case of the white noise in one ofthe coordinates (yi). This estimates show that Fermi resonance leads to oscillations in 2/2 even if £2 = 0. Thus, the particle can escape in the direction j / 2 - Rough estimate of the barrier U* = U(yl) (Fig.6.19) crossing in the direction 2/2 can be done in a way as follows. TP reaches the threshold value y\ in the case of the maximum and minimum values ofthe envelopes A2(t) and Ai(t), respectively, because Ai{t) and A%{t) change in the opposite phase. The number of escapes per unit time for the function
Kramers problem for 2-d potential
0
T ' 1 ' 1 ' 1 2000 3000 4000 5000
1000
201
landscape
1 ' 1 ' 1 ' 1 2000 3000 4000 5000
1000
Fig. 6.17 T P trajectory at the white noise in the y\ coordinate under the condition of the Fermi resonance (wi = 2^2).
0.4 •
0.2 •
0.0 .no-
-0.2-
-0.4 •
I 1000
'
I 2000
'
I 3000
'
I 4000
'
I 5000
T^
I 1000
'
I 2000
'
I 3000
'
I 4000
'
I 5000
Fig. 6.18 T P trajectory at the white noise in the j/2 coordinate under the condition of the Fermi resonance (a>i = 2u)2).
y2{t) = A2{t) cos u2{t) can be estimated by the formulas (1.14) and (1.15): w
2
(
yf
(6.17)
If the vibrational amplitude in the direction of the random force is (Ai) ~ 0,3pA, the escape is possible only in the perpendicular direction, and the boundary value is y2 ~ 0, 5pA, then according to (6.16): <^> = ^ ui2
n2 = — exp 2n
yf 2 (AD
W2
= ^=0,025, /
2TT 6 X P I
0,25
2-0,025
= - ^ e ~ 5 =0,001w 2 2TT
If the noise of the same intensity acts only along the y2 axis, we have:
202
Fermi resonance and Kramers
problem
n2 = 0,012w2. Thus, the escape provided by the energy redistribution is an order of magnitude less probable than that realized under the direct noise action. If noncorrelated noises act up in both directions, there is no correlation in the energy redistribution processes, A\{i) and A2{i) do not change in the opposite phase, and one can not use (6.17). In this case it is necessary to solve the problem in 2-d space. Note once more that the above estimates are valid only if the quality factors of the system are very high for both coordinates. The coincidence of the characteristic frequencies of the external action and the resonance frequencies is hardly possible but one should keep in mind that the force constants ki may change as a result of, e.g., slow modulation of the length and orientation of the il-bonds by which the ligands are bound in the enzyme AS. The estimates show that the resonance frequencies of the substrate bound in CT AS range from 1012 to 1013Hz and vary slowly with the motion of the clusters constituting CT subglobules. To specify Kramers conclusion on the dependence of the escape time r on h/u we carried out a computer experiment. The system of equations for the motion of a particle (m = 1) under the action of the delta-correlated noises £i, £2 in the potential field U(x, y) is:
We studied the dependence of the time that is necessary for a particle to reach a boundary G or to acquire a certain potential energy U* on the friction coefficient h that is unambiguously related to the noise amplitude. We considered five different variants of the shape of the potential profile and configurations of the boundary G (topograms of the potential surfaces are presented in Fig.6.19(a-d)): 1) 2) 3) 4)
U U U U
= = = =
x2 x2 x2 x2
+ y2, (G): U* = 25; i.e. x2 + y2 < 25. + y2 , (G): U* = 16; i.e. x2 + y2 < 16. + y2 , (G): U* = 25; or x < 4. + 2/2/4, (G): U* = 25; or x < 4.
5) U = (x2 + y2) * ( f | - ^
+ ^
+
9c
° 1 S f Q ) ) , 0 < a <2TT, (G):U* = 25;
or x < 2\/2. Figure 6.19e shows the plots of the mean time (averaging over 1000 realizations) versus the friction coefficient for the above variants.
Kramers problem for 2-d potential landscape
0.01
0.10
1.00
10.00
203
100.00
e)
Fig. 6.19 (a)-(d) Topograms of the potential relieves for the cases 1, 3-5: the escape is possible at (a) ( / > [ / * = 25; (b), (c) U > U* = 25 or x > 4; and (d) U > U* = 25 or x > 2%/2. (e) plots of the mean escape time versus the ratio h/uj (u; = 1).
The character of the escape time dependence on the friction coefficient agrees well with the Kramers' results for the 1-d case (see paragraph 1.1): minimum is achieved at the friction coefficient equal to the natural frequency of free oscillations, detuning from this value leads to exponentially growing escape time. Potential surfaces 3 and 4 represent elliptical paraboloids cut by the plane x = 4; curve in the cross section is parabola; the value of the potential energy in the minimum is 16. For these cases the two-dimensionality of the potential is critical because the point mass can reach the boundary at lower energy (in comparison with the case 1 - uncut potential). In the case 5 the shape of the potential surface is chosen in a special way and corresponds to the systems in which the conditions of the Fermi resonance are met.
204
Fermi resonance and Kramers
problem
However, the potential surface obtained as a result of summation of the potentials U = ki ((x — Xi) + (y — y{) J appears to be "inconvenient" for numerical calculations. Firstly, its type substantially depends on the length of the springs. It is natural that oscillations are quasi-harmonic only in the vicinity of the equilibrium and they are not such if the amplitude is larger than 0.1/j. Hence, one can hardly speak about energy redistribution. Secondly, the potential landscape is still symmetrical relative to X and Y axes but it looses symmetry in other directions. Thirdly, the cross section of this surface by a plane can give more than one minimum or the minimum can be far from X or Y axes. All these facts forced us to consider potential 5 that is obtained by means of approximating the sum potential landscape of four springs with the coefficients hi = 0,25, fc2 = 1, A3 = 0,5, and ki = 2, under the condition that they are either extended or compressed. The conditions of the energy redistribution (multiplicity of the natural frequencies in the directions X and Y) are maintained in this potential but it is free of the mentioned shortcomings. Figures 6.20a and 6.20b show the relieves for the potentials U5 and U2. Vibrational process and, consequently, energy redistribution resulting in fast approaching the boundary are observed in this case at small friction coefficient. Vibrational motion collapses at large friction coefficients and the time of approaching the boundary is comparable with that for the cases 3 and 4. Additional studies of the case 5 were performed: 5-1) 5-2) 5-3) 5-4)
U = (x2 + y2) * f(a), (G): x < 2yf{2). U = (x2 + y2) * / ( a ) , (G): x > - 4 . U=(x2+y2)*f (a), (G):y<8. U = (x2 + y2) * f(a), (G): y > - 4 ^ 2 ) .
Here there is an exit in only one direction. Figure 6.20c shows the results of the calculations. Here the curves with triangles correspond to the cases 5-1 and 5-3, the curves with circles correspond to the cases 5-2 and 5-4. It is seen that the escape time depends on the two-dimensionality of the potential landscape in the case of the vibrational process (the curves seize to be symmetrical relative to 1 - the frequency of natural oscillations). Moreover, we considered the case (curve without markers) of the boundary 5-1 and the external force acting up only in the Y direction. Under such conditions the particle never crosses the boundary provided that there is no energy transfer from one mode into another. In the considered case the vibrational process allows energy redistribution and at the small friction coefficients the escape time was nearly the same as in the cases 5-1 - 5-4. However, if the friction coefficient is rather high (the system is overdamped) there is no vibrational motion in the system and the escape time rapidly increases. Note that we studied qualitative effects, so in the last cases the amplitude of the noise action was increased four times and the number of realizations was decreased down to 100. Such changes lead to breaking curves and decreasing escape time.
Kramers problem for 2-d potential
Y
i
Y,
a)
b)
205
landscape
0.01
0.10
i.oo
10.00
100.00
c)
Fig. 6.20 (a) and (b) Topograms of the potential relieves for the case 5: the escape is possible only at (a) y\ > 2\/2 or j/i < - 4 and (b) y2 > 8 or 2/2 < —4y/2. (c) Plots of the mean escape time versus h/ui (w = 1) for the potential relieves (triangles) 5a and (circles) 5b; the curve without markers corresponds to the case when the noise acts up in Y2 direction and the escape is possible in Y\ direction.
However, the qualitative character of the results remains unchanged (it was proved by additional computations at certain values of h in which the number of realizations was increased to 1000 and the noise amplitude was decreased). In the computer experiments the inevitable quantization error leads to limiting time of the T P motion preceding its escape from the potential well. If the lifetime of the given state is rather large, one must use the theoretical description (see paragraph 3.3) valid at the low escape probability. In the notation of the paragraph 3.3 the dimensionless time is represented as:
^_KZI2R [k^T\ (Umin\ * - 2A \lUmin[eXP\kBT)
,' \ •
This result does not depend on the friction coefficient h. However, the particle moves slowly at large h which means the violation of the condition of smallness of the correlation time in comparison with the escape time. On the other hand, the noise action becomes negligible at small h and one can hardly apply the statistic approach instead of the dynamic one. Table shows escape times at several depths of the potential well.
Umin/ksT T
5 118
6 294
7 740
8 1183
Note that the results of the above calculations (see Fig.6.19e) agree well with the case when f/ m f n /fcsT = 5.
206
Fermi resonance and Kramers
problem
How can we use (at least qualitatively) these results for solving the problem of the products escape from the active site? It is likely that they can be used for evaluation of the situation in the CT active site (Shidlovskaya, Schimansky-Geier & Romanovsky, 2000) where only one ligand resides after peptide bond breaking. The situation is much more complicated in the case of ACE.
References N.N. Bogolyubov, Yu.A. Mitropolskii (1974): "Asymptotic methods in nonlinear vibration theory" (in Russian), Nauka, Moscow. V.G. Dashevskii (1987): "Conformational analysis of macromolecules" (in Russian), Nauka, Moscow. W. Ebeling, Yu. Romanovsky, Yu. Khurgin, A. Netrebko, N. Netrebko, E. Shidlovskaya (1994): "Complex regimes in the simple models of molecular dynamics of enzymes", Proc. SPIE 2370, 434-447. E. Fermi (1931): "Uber den Ramaneffekt des Kohlendioxids", Zeitschrift fur Physik, 250-259. K.M. McDonald, W.R. Thorson, J.H. Choi (1993): "Classical and quantum proton vibration in a nonharmonic strongly coupled system", J. Chem. Phys. 99, 46114621. A. Netrebko, N. Netrebko, Yu. Romanovsky, Yu. Khurgin, E. Shidlovskaya (1994): "Complex modulation regimes and vibration stochastization in cluster dynamics models of macromolecules" (in Russian), Izv. Vuzov: Prikladnaya Nelineinaya Dinamika 2, 26-43. A. Netrebko, N. Netrebko, Yu. Romanovsky, Yu. Khurgin, W. Ebeling (1996): "Stochastic cluster dynamics of enzyme-substrate complex" (in Russian), Izv. Vuzov: Prikladnaya Nelineynaya Dinamika 3, 53-64. H. N. Nunez-Yepez, A.L. Salas-Brito, C.A. Vargas, L. Vincente (1990): "Onset of chaos in an extensible pendulum", Phys. Lett. A145, 101. A.B. Pippard (1983): "The Physics of Vibration", Cambridge University Press, Cambridge, London, New York, Melbourne, Sidney Yu.M. Romanovsky (1997): "Some problems of cluster dynamics of biological macromolecules", In: Stochastic Dynamics, L. Schimansky-Geier, T. Poeschel, Eds. Ser.
Kramers problem for 2-d potential
landscape
207
Lecture Notes on Physics, Springer Verlag. Berlin, p. 140-152. A.B. Rubin (1987): "Biophysics" (in Russian), Vyshaya Shcola, Moscow. K.V. Shaitan , M.D. Ermolaeva, S.S. Saraikin (1999): "Nonlinear dynamics of the molecular systems and the correlations of internal motions in the oligopeptides", Ferroelectrics 220, 205-220. K.V. Shaitan , P.P. Pustoshilov (1999): "Molecular Dynamics of a Steric Acid Monolayer", Biophysics 44, 429-434. E. Shidlovskaya , L. Schimansky-Geier , Yu.M. Romanovsky (2000): "Nonlinear vibrations in 2-dimensional protein cluster model with linear bonds", Z. Phys. Chem. 214, No 1, 65-82. V.I. Tikhonov, M.A. Mironov (1977): "Markov's processes" (in Russian), Sovietskoe Radio, Moscow. M.V. Volkenstein (1974): "Molecular biophysics" (in Russian), Nauka, Moscow.
This page is intentionally left blank
Chapter 7
Molecular scissors. Cluster model of acetylcholinesterase A.Yu.Chikishev, S.V.Kroo, A.V.Netrebko, N. V.Netrebko, Yu.Romanovsky 7.1
The role of acetylcholinesterase in the synaptic transfer
To illustrate the possible applications of the problems discussed in the previous Chapters we consider the interaction of a hydrolytic enzyme acetylcholinesterase (ACE) with its substrates. ACE plays an important role in the synaptic transfer of the nerve pulses. In addition, ACE offers a good example of how diversified can be the problems of the Brownian motion as applied to functioning of the molecular machine. ACE, as well as CT, is a serine hydrolase. It catalyses ether bond breaking in the neuromediator molecule (AC). In Chapter 1 we discussed a possibility of the spontaneous bond breaking in the aqueous environment. Note that the synaptic (chemical) transfer of the electric excitation from one neuron to another or from the nerve terminus to the target cells that control the work of the muscle fibers is impossible without ACE. Consider in brief the process of the synaptic transfer (see, for example [Alberts et al., 1994]). A synapse represents an intermembrane contact of two excitable cells. According to the mechanism of the pulse transfer from neuron to neuron the synapses are classified as chemical, electric, and mixed. The chemical synapses are the dominating ones in the synaptic apparatus of the central nervous system of animals and human beings. Nerve-muscle transfer uses only chemical synapses. The synapses consist of the three main elements: presynaptic membrane, postsynaptic membrane, and the synaptic cleft. (Fig.7.1). Presynaptic membrane covers the nerve terminus that can be considered as a neurosecretory apparatus. Here the mediator is stored and released. The mediator excites or inhibits a target cell. Acetylcholine, adrenaline, noradrenaline, dofamine, and some other substances can be mediators in the interneuron synapses. Acetylcholine plays the role of the mediator in the skeletal muscles of all vertebrates and human beings. A presynaptic terminus accommodates 50-nm "bubbles" containing acetylcholine. The neuromediator is released from the bubbles under the action of the propagating potential and goes into the synaptic cleft. Each nerve pulse releases a certain 209
210
Molecular scissors.
Cluster model
1 7 8 2
Fig. 7.1 The scheme of the nerve pulse transfer in the chemical synapse. 1 presynaptic neuron terminus; 2 postsynaptic neurone; 3 presynaptic membrane; 4 postsynaptic membrane; 5 synaptic cleft; 6 bubbles with neuromediator; 7 released neuromediator; 8 acetylcholine receptors; 9 enzyme acetylcholinesterase.
amount of AC molecules. The repetition rate of the pulses is about 1 kHz. The width of the cleft is about 50 nm. Mediator diffuses quickly through the cleft and acts up on the membrane of the target cell. The part of the cell membrane that is most close to the nerve terminus is called postsynaptic. Acetylcholine receptors are located at the postsynaptic membrane. In response to the action of acetylcholine the receptors change the penetrability with respect to Na+ and K+ ions which results in generation of the action potential in the postsynaptic muscle fiber or in the body of another neuron. It is known that postsynaptic membranes contain large amounts of acetylcholinesterase. This enzyme represents a kind of molecular scissors "cutting" acetylcholine. Recall that under natural conditions the nerve pulses arrive at the postsynaptic cells at a rather high rate so that the postsynaptic membrane depolarized by the preceding portion of acetylcholine becomes insensitive to the action of the next portion. Normal exciting action of the subsequent pulses is possible only if the previous portion of the mediator is removed or deactivated. There are several pharmacological agents capable of inhibiting acetylcholinesterase activity. These inhibitors are used to avoid muscle relaxation under narcosis and in the case of such diseases as myastenia. On the other hand, it is known that people can be poisoned by insecticides based on these inhibitors. The spasms caused by these intoxications result from the prolonged activation of the acetylcholinergetic synapses especially in the vegetative nerve system. The major part of toxins acts on acetylcholinesterase. Thus the study of the mechanism of this molecular machine is an important
ACE computer model based on X-ray data
211
problem for medicine and for interpreting the principles of the enzymatic catalysis.
7.2
ACE computer model based on X-ray data
Consider the geometry of ACE. The coordinates of the heavy atoms 0 , C, N, and S were determined by X-ray analysis [Sussman et al., 1991] with the accuracy of 2.8pA. The results of the analysis can be found in the Protein Data Bank. File format is described in [Bernstein et al., 1977]. In this work we used the data file with the atomic coordinates of ACE from Torpedo californica. ACE represent an ellipsoid with the dimensions 45 x 60 x Q5pA. This molecule consists of 537 amino acids. The AS pocket represents a deep and narrow cleft that goes inside the protein globule. AS catalytic triade consists of the amino acid residues Ser 2 0 0 , Glu327, and His440 and is located at the bottom of the cleft. 14 aromatic residues form the walls of the cleft and prevent the positively charged substrate (AC) from the interaction with the negatively charged acidic residues. The calculations of the electrostatic field of the enzyme based on the X-ray data showed that an extremely large dipole moment of ACE is directed along the axis of the cleft. (There are contradictory data regarding the value of ACE dipole moment: 505 D [Ripoll et a l , 1993], 1500 D [Antosiewich, Gilson k McCammon, 1994], 1 D = 3.33564 • 10~ 30 C-m.) The positively charged substrate can be pulled into the cleft due to electrostatic interaction. In addition, the total charge of the enzyme is negative: it is about — 12e without allowance for the ionic surrounding that results from the interaction of the protein with the solvent. The problem of the release of the reaction products from the active site is mentioned in [Ripoll et al., 1993]. Indeed, one of the reaction products, choline, is positively charged and its diffusion out of the cleft is hindered by the electrostatic attraction to the enzyme. This difficulty gave birth to the hypothesis of the "back door" allowing the escape of the reaction products from the active site. As the indole ring of Trp84 is situated inside the cleft and the residue as a whole is located at the outer surface of the protein, the authors proposed a possibility of a conformational change opening an additional exit from the cleft of the active site. Gilson et al. [Gilson et al., 1994] studied the possibility of the "back door". The conformations of the protein molecule under which a water molecule could leave the active site were studied. Simulation of ACE dynamics in water showed that an exit really appears for a short time. The channel is formed in a thin wall of the active site in the vicinity of TrpM. Its entrance is near Gly441, Tyr442, and TrpM. The channel goes around Trp84 and appears at the surface near Glu445. Displacement of Trp84, Val129, and Gly441 residues at the mean deviation of their atoms of about 1.3pA opens the channel. No other channels were found in spite of the fact that all the conformations were analyzed. Fig.7.2 shows the entrance to the cleft, the catalytic triad, the bottom of the cleft, and the "back door". The calculation of the electrostatic potential around the enzyme showed that
212
Molecular scissors.
I- gorge entry
active site
I III IN'
Cluster model
|., :
| - gorge bottom
^ S S ~ 'back door'
Fig. 7.2 ACE molecule: (a) the axis of the cleft is perpendicular to the figure plane; (b) the axis of the cleft belongs to the figure plane.
the potential minimum is located near the bottom of the active site in the vicinity of He444. Thus, the electric field does not allow the positively charged choline to leave the cleft through the "back door". The problem of the escape of the reaction products remains unsolved. Let us consider in details the geometry of the enzyme and substrate molecules. X-ray data regarding both AC and ACE are available. Fig.7.3 demonstrates the layers of 5-angstrom thickness cut by the planes shown in Fig.7.2. Fig.7.4 shows two positions of the substrate molecule: near the cleft entrance and in the vicinity of the catalytic triad. It follows from Fig.7.2-7.4 that the cleft of the active site can accommodate only one AC molecule. Assume that the "back door" does not exist then it is evident that the substrate molecule can approach the catalytic triad only if the previous substrate molecule or its fragments have left the AS cleft. Note that diffusion of the positively charged molecule from the cleft is hindered by the electrostatic attraction to ACE molecule which can limit the rate of the enzymatic reaction.
7.3
Electrostatic field of ACE molecule
Consider the motion of the substrate and the reaction products in the vicinity of the enzyme. One has to take into account their interaction with solvent molecules, van-der-Waals interaction with each other and with the enzyme molecule, and the electrostatic interaction. In order to compare the influence of these three factors, let us calculate the electrostatic potential of the enzyme in the solvent.
Electrostatic field of ACE
Layer H
Layer G
Layer A Fig. 7.3
7.3.1
213
molecule
Layer F
Layer B
The layers of ACE molecule confined by the planes shown in Fig.7.2.
Charge distribution
inside the
molecule
Amino acids are classified as acidic (can have negative charge), alkali (can have positive charge), and neutral [Volkenstein, 1988]. There are three acidic (Asp, Glu, Tyr) and three alkali (His, Lys, Arg) amino acid residues. The charge of the amino acid residue depends on pH of the solvent. At pH 7.0 that corresponds to natural environment of the protein [Pasynsky, 1963] only four residues are ionized: Asp, Glu, Ly, and Arg. All the further calculations are performed for pH 7.0. Thus, we know the charges and coordinates of all the ionized atoms of the protein and can use them for calculating electrostatic parameters of the enzyme. Before going to the results let us discuss the choice of the coordinates. In this work the origin coincides with the atom CD1 of the residue He444 that is located at the bottom of the cleft. Z axis goes through the center of the entrance to the active site and, thus, represents the axis of the cleft. The center of the entrance to the active site is defined as a mean value of the coordinates of Glu73.CA, Asn280.CB, Asp285.CG, and Leu333.0. X and Y are oriented in such a way that the atom Glu73.CA belongs to the plane YZ. ^-coordinate of the center of entrance and
214
Molecular scissors.
Cluster model
a) Fig. 7.4
b)
(a) Substrate enters the cleft; (b) substrate is in the vicinity of the catalytic triad.
z, A
20-
o-
-20-
-16
-12
-8
-4
0
4
8
12
16
Integral charge density
Fig. 7.5
Distribution of the enzyme integral charge density along Z axis.
y-coordinate of Glu73.CA are positive. Fig.7.5 shows the integral charge density distribution along Z axis, i.e. along the axis of the cleft. The curve shows the sum charge between the planes that are perpendicular to the axis of the cleft; one of the planes is fixed and crosses Z at z = 0. The charge of the "upper" part of the enzyme is negative and equals -14e, the charge of the "lower" part is +2e. Thus, the total charge of ACE is negative and equals -12e. The number of the ionized atoms is 110. Asymmetry of distribution of the charge shows that the molecule must have a dipole moment. Below we determine the dipole moment for different ionic strengths of the solvent for comparison with the experimental data. The agreement of the results can prove indirectly the correctness of the calculated potential.
Electrostatic field of A CE molecule
7.3.2
Calculation
of the
215
potential
Different ions surround ACE molecule in the interstitial liquid. Thus, it is necessary to take into account the influence of the ionic atmosphere on each charge. Assume that the solvent contains different ions. Let their valences be Zi and the concentrations Co,. If a system of charges is introduced in bulk of the solvent the concentrations of ions Cj become spatially inhomogeneous. The concentrations are given by the Boltzmann distribution Ci \f) = c0i exp '
kBT
where
1 -
Zie
Charge density in the given point of the solvent can be obtained by means of summation of the charge densities of all sorts of ions: 2
c
P - ^2 CiZie = e ^2 oiZi ~ -j—f XI zi°0ii
i
i
As the solvent is electrically neutral, the first some equals zero. Then e2tp ^
2
2eV
where I is the ionic strength of the solvent,
1=
~J2CoiZ?Zje
216
Molecular scissors.
Cluster model
Table 7.1. Ionic composition of the interstitial liquid. Anions Concentration, mM Cathions Concentration, mM 144 114 Na+ Cl~ K+ 4 HCOz 30 2 Ca + 1 1 HPO'l~ Mg2+ 1 Organic anions 5
Assuming that the charge density induced by the ions is represented as: eifi
p = e c 0 + exp
\
I CQ-
k„T
exp
eip
kBT
where co+ is the concentration of the positive monovalent ions, CQ- is the concentration of the negative monovalent ions, and CQ+ = co_ = CQ (because of the solvent neutrality), we arrive at: el I exp I —
ef \
( eip
Poisson equation for the potential can be written as: A
££o
eeo
exp | - r - J - exp ( J * L ) ££o V V kBTJ
where pa is the charge density in the enzyme molecule. This equation was solved numerically at the grid 100x100x100 points at the grid step of 1 angstrom. Laplace operator was approximated by a difference operator and the system was solved by the method of simple iterations. The dielectric permittivity was equal to 80. Fig.7.6 shows equipotential lines for the electrostatic field at zero ionic strength. The energies correspond to the charge of + l e . Fig.7.7 shows enzyme potential at the cleft axis at the ionic strengths of 0, 10, and 20 mM. Going back to the applicability of the Debye-Huckel theory we can conclude that it can not be used in the potential minimum where 7.3.3
Determination
of the dipole
moment
Knowing not only the charges of the protein atoms but also the density of the induced charge one can calculate the dipole moment of the system "enzyme plus ionic atmosphere":
i
When considering the solvent we define qi as the product of the charge density in the given point and the volume of the quantization cell. Dipole moment of a
Electrostatic field of A CE molecule
/
^
T Lr -
A 40
w --4kT " \
- * MLL _ 'n^H
\
•2F
217
L r -4kT x \
m d °«5||
I
\
(Wttm MmLj ' 1
t^SfK^y
Kao^/
-40
40 -40
0 a)
40
A
-40
40 A b)
Fig. 7.6 ACE equipotential lines at zero ionic strength. The energies correspond to the charge of + l e : (a) "back door" is closed; (b) "back door" is opened.
•0.04
-0.08
-0.12
-0.16
-0.20 10
20
30
z,A
Fig. 7.7
ACE potential at the cleft axis at different ionic strengths.
system of charges the sum of which is different from zero depends on the choice of coordinates. We performed the calculations in the frame of the center of gravity:
d—Y^^i ~ -^y^Qi, where R is the radius-vector of the center of gravity. Fig.7.8 shows the results of calculations for several values of the ionic strength and the experimental data based on the measurements of the electric dichroism in the fields of different strength [Porschke et a l , 1996]. The differences in the results can be explained by the fact that the experiments yield the values of the dipole moment in the frame of the center of diffusion (not of the center of gravity). In addition, the experiments used ACE from urchin Bun-
218
Molecular scissors.
Cluster model
2UUU-1
1600-
a
1200-
•tf 800400-
010
20
Fig. 7.8 Plot of the dipole moment d versus the ionic strength / : (circles) experimental data and (squares) calculated values.
gams fasciatus, whereas the calculations used the PDB file for Torpedo californica. Relatively small differences allow us to use the calculated potential in the further calculations of the enzymatic reaction rate. It is interesting to compare dipole moment of AChE molecule with dipole moment of system, consisted of the same number of the charges, which distributed in the corresponding volume. Let us assume that we have a sphere of radius R, which contain N point unit charges of arbitrary signs. The density of charge distribution inside sphere is PN(ft,...,r$)
1 yN>
where V - volume of sphere. Then the probability density function of dipole moment is
W(d) = f 5(ct - ] T
=
yN(2
,3 /
Qi
rt)PN (ft,...,
dk exp(ikd
r$) dr\ ... dr$ =
-ik^2
qir$)drt •.. dr^.
In approximation eRk
2
W(P) = 1/2
where (3 = -4- and do = eR jk mean value of /? is equal to
. This function has a maximum at (3 = 2; the
f,2 .
Let's take into account that N — 110, R — 55pA. Then the most probable value of dipole moment is equal to 1750.D, the mean value of dipole moment is equal to 198QD. We calculated above a dipole moment which is equal to 185QD at zero ionic
Substrate enters the pocket: 2-d "toy" model
219
strength. So, asymmetry of spatial distribution of charges in AChE molecule is not too high. On the other hand, the orientation of the dipole moment relative to the axis of active site gorge can play an essential role in a catalytic process: the dipole can direct substrate molecules in a special way and provide a possibility of ESC formation.
7.4
Substrate enters the pocket: 2-d "toy" model
Prior to consideration of the problem of AC "entering" the AS pocket and the release of the reaction products with regard to the chemical stage of the ester bond breaking, we consider a "toy" model that is rather close to reality. In this paragraph we consider a substantially simplified 2-d model with the geometrical configuration and the profile of the potential landscape that are slightly different from real ones but still having some typical features. Thus, we shall consider some problems of the Brownian motion that can be met in the case of more accurate simulation. The problems are as follows. In course of penetration into the AS AC molecules interact with neutral water molecules, negatively charged A and positively charged C molecules. All of them interact with each other and with the walls of ACE molecule randomly deformed due to thermal fluctuations. 1. The most simple and effective method of simulation lies in consideration of the Langevene system of equations for the interacting AC, A, and C. The interaction with water molecules is simulated by a random force and damping and the walls of ACE are considered as nondeformable. That is approAC we used for 3D simulation. However, one should not forget that the mass of water molecules is comparable to those of the reaction products and that a part of the AC sorption energy can go back to A and C after ester bond breaking. It is also unclear whether we can neglect the inertia forces (the second derivatives in the model) within the frames of the Langevene approach. 2. Because of these reasons it is expedient to compare the results provided by this simple appro AC with the results that can be obtained in the model where the interactions between all the molecules (AC, A, C, and water) are taken into account. This is what the simplified 2-d "toy" model allows one to do. 3. To study the interaction of ACE with the substrate and the reaction products we: a) consider penetration of AC molecule into the AS pocket from the intercellular space as dependent on the number and concentration of AC molecules at the entrance to the AS; b) consider the necessity of the "back door" for removing of the fragments from the AS pocket; c) illustrate the influence of the frequency of oscillations of the AS entrance width on the rate of operation of the molecular scissors;
220
Molecular scissors.
Cluster model
d) consider the influence of the time during which the substrate molecule remains bound with the catalytic group on the rate of operation of the molecular scissors; e) consider the influence of the sorption energy transfer to reaction products after AC bond breaking on the diffusion limitation and on acceleration of the reaction products escape from the AS pocket (this could be expected according to the concept "protein-machine", see [Chernavsky, Chernavskaya, 1999; Chernavsky, Khurgin k Shnol, 1978]; f) make some calculations varying the configuration of ACE electrostatic field, in particular, the heights of the potential barriers. Note that in this case we do not plan to obtain real numerical estimates for the rate of the enzymatic reaction. We study only the possible influence of different physical factors. Some of these problems were considered in [Romanovsky, Netrebko, 1998; Romanovsky et al., 1999]. When constructing the model of the active site we made the assumptions proved by experimental data: a) the geometrical sizes of AC molecules are such that they can penetrate inside the AS pocket only in turn; AS accommodates not more than three AC molecules at a time; b) the distribution of charges inside AS pocket is such that an attracting force acts up on the AC molecules outside the pocket; c) AC molecule (positively charged) is decomposed inside the AS to form two fragments (positively and negatively charged); the positively charged fragment is smaller than the negatively charged one; d) AC fragments must be removed from AS pocket (not more than six fragments can be inside the pocket at a time); their quick removal suggests the existence of the "back door"; e) several water molecules can be located inside the pocket. We consider the case of the rather high substrate concentration in the vicinity of the AS. Enzyme operation rate is limited in this case by the processes taking place inside the enzyme molecule in the very proximity of the AS.
7.4.1
AC molecules
enter ACE
AS
It was demonstrated that in the case of rather high concentration of the substrate the excessive amount of AC molecules concentrate near the entrance to the AS pocket under the action of the dipole field. It happens so because the rate of transportation of the substrate molecules to the entrance is higher than the ACE operation rate (penetration of AC into AS pocket, cutting with the formation of A and C, escape of the reaction products from the AS pocket). We considered the problem of penetration of the dumbbell-shaped particles (two flat disks of the radius TQ — 0.5, the centers of which are at the distance of 2r 0 ) through the corridor formed by two Lennard-Jones potentials (ULJ = (Rp/R) —
Substrate enters the pocket: 2-d "toy" model
Fig. 7.9
221
Potential landscape in the vicinity of the AS gates. Lennard-Jones potentials are cut.
(Rp/R) ; R = J(x - xp) + (y - yp) ) with the characteristic radius Rp = 11, centered in (xp\ — —10; yp\ — 0) and (xP2 — 10; yP2 = 0). A force center (xp = 0; yp — 20) attracts one disk of the dumbbell and repulses the other one so that the repulsing force is twice as small as the attracting force. Such a center provides "proper" orientation of the dumbbell allowing its penetration inside AS. In reality only one of the disks is charged. The attracting center was determined in the first case by the potential Up = k\ I (x — xpc) + (y — ypc) J and in the second case it was determined by the Coulomb potential Up = kz/ J(x — XpC) + (y — ypc) . The relationship between fci and fc2 was chosen based on equality of forces acting up on the center of gravity of a particle passing through the gates of the potential landscape (the origin of the coordinates) for both types of the attracting potentials. Fig.7.9 shows the corresponding potential landscape. In the initial moment of time there are n randomly oriented dumbbells (in practical calculations n was varied within the range 1-100) inside the area —10 < x < 10; —20 < y < 0 (accumulating area). The initial velocity of all the dumbbells was equal to zero. In the case the center of gravity of any dumbbell crosses X axis or the particle leaves the accumulating area, the particle is excluded from the further consideration. Instead of this particle another one with zero kinetic energy is placed into the accumulating are. Thus, the constant concentration of the particles in the accumulating area was maintained in course of the calculations. We examined the mean operation rate of such a machine (number of dumbbells passing through the gates) as dependent on the concentration of the particles in the accumulating area, the type of the attracting force, possible external random interactions, and the frequency of the possible oscillations of the width of the potential cleft. The dumbbells could interact with each other. The interaction of particles was simulated in a way as follows. It was assumed that the center of each disk is the center of Lennard-Jones potential with the characteristic radius ro,besides that the center of Coulomb potential is located in the same point. The parameters of the
222
Molecular scissors.
Cluster model
Coulomb potential were chosen to compensate attraction of disks with equal charges due to Lennard-Jones interaction. For oppositely charged disks the parameters of the Coulomb potential had opposite signs which additionally increased the mutual attraction. However, such disks could not "stick" due to Lennard-Jones interaction. Thus, all the particles interact with each other. The system of equations describing the motion of a single dumbbell in the potential field U under the noise action is presented as: d2x
I-jjj
dU
, dx
Arrrr, , N
= r0 x (-gradC/i + gradf/ 2 )
Here x and y are coordinates of the dumbbell center of gravity, t is time, h is the friction coefficient, D is the amplitude of noise action, £iand £2 are deltacorrelated noises (random quantity, uniformly distributed in the interval [-1;1]), / is the moment of inertia of the dumbbell, ip is the angle between the dumbbell axis (the line connecting the centers of gravity of the disks) and X axis, and U is the sum potential of the forces acting up on the center of gravity of the dumbbell:
U = ULJI + ULJ2 + UP + Ucont,
(7.2)
U\, U2 are the sum potentials of the forces acting up on the centers of the first and second disks of the dumbbell, respectively (they are determined according to a similar formula), ULJI,UU2 are Lennard-Jones potentials that simulate the AS gates, Up is the sum potential of the forces acting up on the dumbbell from the AS, Ucont is the potential of the forces of interaction between the dumbbells. The parameters of the potential landscape were chosen in such a way that the particles could go through the cleft only in turn; the angle ip in the moment of the transition must have been close to 90°. Fig.7.10a and 7.10b show the time of the m-th penetration for different amounts of the dumbbells (n) in the accumulating area for the first and the second variants of the attracting potential. Time is plotted versus the number of particles (ra) that passed through the gates. The numbers near the curves correspond to the numbers of particles in the accumulating area (n). All the plots can be approximated with rather high accuracy by linear functions, i.e. the rate of penetration of the particles is constant on the average and depends only upon n. For some dumbbells the calculated time could be higher than the theoretical one by an order of magnitude (theoretical time is the mean time of the particle travel from the starting point
Substrate enters the pocket: 2-d "toy" model
223
a) 1/t (n=SO0)
Fig. 7.10 Plot of the mean time of the n-th penetration of the dumbbell into the AS pocket versus the number of the dumbbells in the accumulating area: (a) quadratic potential and (b) Coulomb potential; (c) plot of the penetration rate versus the number of the particles in the accumulating area.
to the AS under the action of only the potential Up). This is typical for small n {n — 1,2) or, vice versa, for large n (n > 15). It happens so because in the first case the particle can not find the gates and oscillates quasi-harmonically near them. In the second case several particles come together to the entrance and block it. Only the action of the other particles can destroy this block. The increase of n from 1 to 10 leads to increasing passing rate. If n > 10 the rate virtually does not depend on n. If we divide the passing time of each particle by the theoretical time of penetration of a single particle in the corresponding potential, the calculated curves for the cases (a) and (b) coincide. Thus, the penetration rate is constant and does not depend on the number of particles in the accunulating area for n > 10 and depends only on the value of the attracting force. This result is illustrated by Fig.7.10c showing the time of 500 entrances into AS for the Coulomb attracting potential plotted versus the number of particles in the accumulating area. In these calculations h = 0, i.e. there is neither noise action nor friction. Fig.7.11 shows the results of calculations for the case of noise action at low
224
Molecular scissors.
0
20
40
Cluster model
60
80
100
Fig. 7.11 Plot of the time of the n-th penetration of a dumbbell into the AS pocket in the case of the noise action with the corresponding friction coefficient h (two dumbbells in the accumulating area and quadratic attracting potential).
friction. The results did not change qualitatively. The increase of the friction coefficient leads to a certain slowing down of the process. (In these calculations we assume that there are at least two dumbbells in the accumulating area at a time and chose the potential of the first type as the attracting one to reduce the computational time). Fig.7.12 demonstrates the results of calculations for the case of the mobile "gate" Lennard-Jones potentials. It was assumed that their centers harmonically oscillate along the X axis near the initial position at the frequency w. The amplitude of oscillations (A) was aliquot to the geometrical size of the dumbbell disk ro- Thus, at certain moments of time the gates are closed (a dumbbell can not pass through them because of the insufficient energy) but sometimes two particles can go through them without mutual interferences (see also [Chikishev et al., 1998]). Note that for both values of the amplitudes of oscillations a certain "locking" frequency was observed {2-K/U* = 0.71). In the case of the higher frequency not a single particle can pass through the moving gates. However, there was no correlation of the penetration rate with the frequency w. 7.4.2
The problem
of the reaction
products
escape from ACE
AS
To solve this problem one has to simulate not only the AS gates but also the AS itself and the "back door". The simulation of the escape of the reaction products from the ACE AS pocket appears to be difficult because of the following reasons. Firstly, several particles can be found in the AS pocket at a time: positively charged AC and C and negatively charged A. It is necessary to take into account their interaction. Secondly, the potential landscape inside the AS pocket has a complicated shape and can hardly be determined accurately. In contrast to the problem of entrance the shape of the landscape is of great importance. Fig.7.7 shows the electrostatic component of the potential U(z) at the axis of ACE molecule at different values of the ionic strength inside AS pocket built on the
Substrate enters the pocket: 2-d "toy" model
225
I
280-
\
240-
200 -
\
\
\ 160-
y.A-0.2 V'
(2jl/w)»=0.71 0.1
/ A=0.1
1.0
\ . '
i*. 10.0
100.0 2TC/U1
Fig. 7.12 Plot of the time of 10 penetrations into AS pocket versus the frequency (u>) of the oscillations of the centers of the gate Lennard-Jones potentials at two oscillation amplitudes (A).
Fig. 7.13
The shape of the potential landscape inside AS along Y axis.
basis of PDB data. One can be sure that U(z) has a minimum at the distance of several angstroms from the bottom of the pocket. Therefore, the random interactions of the particles with each other and with the oscillating atoms of the walls of the AS pocket can lead to the escape of the positively charged AC and C. The negatively charged A can get into the minimum in the vicinity of the pocket bottom (dashed line in Fig.7.13). Based on the potential landscapes we chose a simplified scheme of charges in ACE globule for the escape simulation. The first experiments with this model showed that the parameters of the 2-d electrostatic potential can be chosen in such a way that small oscillations of the entrance Lennard-Jones potential at certain "resonance" frequencies influence the escape of both positively and negatively charged fragments. These frequencies vary within wide ranges under rather small variation of the parameters. In addition, the escaping fragment must not meet the attacking dumbbell that can just bring it back to the AS. Recall that there must be a large number of the attacking dumbbells in the accumulating area and, hence, the collision at the exit is highly probable. It seems that the continuous work of the
226
Molecular scissors.
a)
Cluster model
b)
Fig. 7.14 The shape of the potential landscape in AS and the accumulating area for (a) positively and (b) negatively charged particles.
molecular scissors is possible if the negatively charged fragment is removed through the "back door". There is another possibility: A~ associates with free H+, becomes neutral and leaves the AS. The probability of such an event depends upon the mean pH inside the AS pocket. In addition, A~ can "stick" to the positively charged ion X+. Our estimates show that in the latter case the bond energy is comparable to the mean kinetic energy of a molecule assuming that the energies of both X+ and A~ equal Z/2kBT. In practical computations the AS area was simulated with the help of four Lennard-Jones potentials with the characteristic radius Rp — 11. The centers of the potentials are located in the points with the coordinates xp\ — —10; yp\ — 0; xp2 = 10; yP2 — 0 (entrance to AS); xp3 = -10; yp3 = 8; xP4 = 10; yp\ = 8 ("back door"). Two negative charges are placed in the points with the coordinates Xpki = —3; ypki = 0; xpk2 — 3; ypk2 = 0 (attracting centers for the positively charged AC molecules). Two positive charges are placed in the points with the coordinates xpki = —3; ypki = —8; xpk2 = 3; ypk2 — —8 (potential barrier for the positively charged fragments). Thus, like in the previous case the entrance to AS is in the origin and the "back door" is at the Y axis. An accumulating area for AC ( - 8 < x < 8; - 8 < y < 0) and water (-10 < x < 10; - 1 0 < y < 0) molecules was in the lower half-plane. Water molecules are considered as disks with the radius ro/2, the mass of the water molecule is ten times smaller than the mass of the dumbbell. The mean kinetic energies of water molecules and dumbbells are assumed to be equal (thermal balance). Water molecules are electrically neutral. Up to 100 water molecules can be considered at a time. The resulting potential relieves for the positively and negatively charged particles are presented in Fig.7.14a and 7.14b. Fig.7.13 shows the potential landscape inside AS along the Y axis (solid line). It is seen that this profile reproduces in details the shape of the potential surface obtained using PDB data.
Substrate enters the pocket: 2-d "toy" model
227
It was assumed that the dumbbell breaks into two fragments (positively charged one with the radius 0.25 and the mass M / 4 and the negatively charged one with the radius 0.5 and the mass M/2, where M is the mass of the dumbbell) when its center of gravity passes the AS gates (y coordinate of the center of gravity is less than - 2 ) . It is seen (Fig.7.14) that in the potential under consideration the negatively charged fragment is removed easily through the "back door" (there is no potential barrier for it on this way). At the same time the "back door" is virtually closed for the positively charged particle by the potential barrier the height of which is comparable with that of the barrier at the entrance. Hence, the fragment can overcome the barrier only under the action of another dumbbell, fragment, or water molecule. The positively charged fragment can hardly be removed through the entrance door although this variant is more probable since the corresponding energy barrier is lower. At this way it virtually always meets a large positively charged AC particles that try to get inside the AS. Nevertheless, a certain number of the fragments leaves the AS through the entrance door due to their small geometrical size. In calculations we closed the "back door". In this case the AS zone was pretty soon saturated by the AC fragments that could leave it. Thus, the existence of the "back door" appears to be a necessary condition of the molecular machine operation under the mentioned assumptions on the shape of the AS potential landscape. In calculations we assumed that constant AC (10 molecules) and water (from 0 to 200 molecules) concentrations are maintained in the accumulating area. Maximal total concentration (the area of the accumulator occupied by disks and dumbbells divided by its total area) was equal to 0.5. The accumulating area was randomly filled with dumbbells (zero initial velocity) and disks (the initial kinetic energy corresponds to the mean total energy of a single moving dumbbell). A new particle is placed inside the accumulating area in the case when a dumbbell or a disk leaves the area. A fragment is removed if its center of gravity leaves the AS area through the "back door" or escapes through the entrance door. Fig.7.15 shows the results of calculations. The lower set of curves corresponds to the time of penetration of AC molecules inside AS pocket at various concentrations of water molecules inside the accumulating area (the entering time is defined in the same way as in the previous Section). The lowest curve corresponds to zero concentration of water molecules and the topmost curve corresponds to 100 water molecules in the accumulator. The upper set of curves correspond to the time of penetration of AC molecules inside AS pocket at various concentrations of water molecules for the case when the whole process is considered (entering, breaking, escape of fragments). The difference between these curves yields the "lifetime" of fragments in AS. (New particles can not go inside until the fragments are removed.) Note that according to the results of calculations 99% of the negatively charged fragments and 95% of the positively charged ones leave the AS through the "back door". Lifetime of fragments in AS can be infinitely large if they do not get ad-
228
Molecular scissors.
Cluster model
800
600
400
200
0 0
20
40
60
80
100
Fig. 7.15 The time of the n-th penetration into AS pocket at various numbers of water molecules in the accumulating area (lower and upper curves correspond to entering and escape, respectively).
ditional energy from water molecules or new fragments. Continuous work of thus constructed molecular machine is provided by the presence of a certain number of AC molecules inside the accumulating area. We tried to substitute the noise action and corresponding friction for the action of water molecules. The noise amplitude is chosen in such a way that a maximal single action on AC particle changes its energy by not more than 1/10 of its mean kinetic energy. The friction coefficient is chosen for each concentration of water molecules to ensure coincidence of the resulting curves for the penetration time. We also made an attempt at estimating the influence of the decay time of AC molecules on the working rate. We assume that AC molecule does not break immediately when entering AS but waits for a certain period after loosing its kinetic energy. By that time it can acquire a certain additional energy and even leave the AS (such cases were quite rear) due to interactions with the other dumbbells and fragments. Otherwise it splits into two fragments and their total kinetic energy equals the kinetic energy of the particle before its stop. The results of calculations show that if we simply subtract the total delay time from the resulting data the curves that correspond to zero and nonzero delays virtually coincide. A certain difference is observed for the case when the dumbbell really leaves the AS during the delay time. Based on the results of the calculations we arrive at the following conclusions. An effective model of the molecular machine can be built under the mentioned assumptions on the shape and type of the potential landscape in the vicinity and inside AS, on the type of interactions, and on the shape and geometrical sizes of the particles. If the concentration of AC molecules near the AS entrance is rather high, the operation rate of the machine does not depend on this concentration. If the assumptions on the shape of the potential landscape are valid then the existence of the "back door" is a necessary condition of the continuous work. However, it is understood that all these results are approximate because they are based on the assumptions that are in only qualitative agreement with the experimental data. We
Kinetics
of the enzymatic
reaction of ester bond breaking
229
used 2D model in our calculations. Calculations within the framework of a 3D model (that may yield different results) are possible but necessitate substantial increase of the computational time. On the other hand, 2D model allows one to take into account interactions with water molecules and to apply more accurate numerical methods. We did not take into account the real distribution of charges around AS. For example, if the centers of Coulomb potentials are moved inside AS closer to the center, the character of the potential landscape changes drastically (dashed line in Fig.7.13).
7.5
Kinetics of the enzymatic reaction of ester bond breaking
Spontaneous breaking of ether bond in AC molecule in water is described in chapter 1. The description of chemical processes of bond breaking in the case of AC interaction with ACE AS can be found in many works (see, for example, [Quinn, 1987; Fuxreiter, Warshel, 1998]). Not going into the details of this chemical reaction consider only the simplified kinetic scheme that helps to understand the role of AC, A, and C motility in the ACE electric field for estimation of the rates of the reaction stages that are diffusion and sterically limited. Recall, at first, how the kinetics of the simplest enzymatic reaction is described ([Volkenstein, 1988]). 7.5.1
Michaelis-Menten
equation
Irreversible enzymatic reaction with one substrate can be schematically presented in a way as follows: fci
E + S ^ _ ES -^E
+ P.
(7.3)
fe-i
Here E is the enzyme, S is the substrate, P is the reaction product, and ES is the enzyme-substrate complex (ESC). In the case of complex formation the substrate is fixed near the active group of the enzyme due to electrostatic, hydrogen, and hydrophobic interactions, and also due to valence bonds. ESC can dissociate and can also undergo changes leading to the formation of the final reaction product and regeneration of the initial enzyme. This reaction goes in stages (in one stage in the simplest case as shown in (7.3)). The reaction rates are related to the concentrations of the components [E], [S], [ES]: = fci [E] [S] - k2 [ES] - fc_! [ES],
230
Molecular scissors.
Cluster model
^ 1 = -A* [E] [S] + k2 [ES] + fc_x [ES], d[S] dt
---hlBllSl
+
k-^ES],
If the concentration of the substrate is much higher than the concentration of the enzyme ([S] » [E]), the system quickly arrives at a stationary state where the ESC decay (reactions (2) and (-1)) is balanced by ESC formation (reaction (1)). Thus, in the stationary case: ^ M
= h [E] [S] - k2 [ES] - *_! [ES] = 0.
Let the total concentration of enzyme be [E]Q: [E]0 = [E] + [ES}. Then the rate of the enzymatic reaction (7.3) is given by the formula:
{7A)
"-KMTWV where KM
k-i + k2 *1
Equation (7.4) gives the dependence of the enzymatic reaction rate on the concentrations of the enzyme and substrate and is known as Michaelis-Menten equation. Constant KM is called Michaelis constant and is measured in the units of concentration. According to Michaelis-Menten equation the rate of the reaction is proportional to the concentration of the enzyme and inverse proportional to the concentration of the substrate. It is seen from the equation (7.4) that KM corresponds to such concentration of the substrate under which the rate is twice smaller than the maximal rate. These regularities are really typical for the major part of the enzymatic reactions. The thing is that Michaelis-Menten equation _ kcat [E}0 [S] " " K% + [S]
.
.
(7 5)
-
Kinetics
of the enzymatic
reaction of ester bond breaking
231
formally describes the kinetics of more complex enzymatic processes, e.g. multistage reactions [Berezin, Martinek, 1977]: fcl ^2,
ir
1
1
^3.
^n.
-w ^r,
£ + 5 < _X 1 AX 2 A...-^X„"-^£ + P, fc-i
for which 1 kcat
«2
1 «3
«n+l
fc_i + fc2
Tsim
^(4 +4+- + ^ ) Formally the same expression is valid for the rate of the multistage reaction in which all the stages are reversible. In this case, however, kcat and Kffl are rather complex functions of the rate constants of elementary stages [Heinrich, Schuster, 1996]. With regard to the aforesaid assume that the reaction scheme is as follows:
E + S~^_EAS
^
E*S <
ES-^E*P-^E + P.
(7.6)
fc-l fc-2 fc-3
Here EAS corresponds to such mutual position of the substrate and enzyme molecules under which the substrate is located in the vicinity of the entrance to the cleft of the active site and no other molecule can penetrate inside the cleft (Fig.7.4a); ES is enzyme-substrate complex; E*S and E*P correspond to the positions of substrate and product molecules near the catalytic triade (Fig.7.4b). ESC formation from E*S does not require the substrate diffusion to the catalytic triade; hydrogen, hydrophobic, and electrostatic interactions are repsonsible for this process. The system of differential equations for the scheme (7.6) is given by: ^ p i
=fci[E] [S] +fc_2[E*S]-
(k2 +fc_!)[EAS],
d [E * S] = k2 [EAS] +fc_3[ES] - (k3 +fc_2)[E*S\, dt d[ES] = k3[E*S\dt d [E * P] = k4 [ES] dt
(*4 +fc_3)[ES],
-k5[E*P}.
(7.7)
232
Molecular scissors.
Cluster model
Assume that the concentration of the substrate is much higher than the concentration of the enzyme and consider the stationary case under which the concentrations of EAS, E*S, ES, and E*P remain constant: d[EAS] dt
=
d[E*S] dt
=
d[ES] dt
d[E*P] dt
=
[
j
Let the total concentration of enzyme be [E]0: [E}0 = [E] + [EAS] + [E*S} + [ES] + [E*P].
(7.9)
The rate of the enzymatic reaction is u = k5 [E * P]. Determining [E*P] from the equation (7.7) under the conditions (7.8) and (7.9), we obtain Michaelis-Menten equation (7.5) for the reaction rate in which: 1 _ 1 , k-2 &2 k2k3 kcat
k-2k-3 k2k3k±
1 k3
A;_3 fc3fc4
kCat hk2 1 K™ fc_i + f e " j , * - 2 k3
1 k$
k-2k-3 k3k4
5 fe
(7.10)
(7.11)
The simulation of the diffusion of the substrate molecules and their "fragments" allows one to determine the rate constants ki, k_i, k2,and k$. Constants k_2, k3, k_ 3 , and k4 determine the processes of sorption, desorption, and hydrolysis of the substrate; these processes are not simulated in this work. It follows from Fig.7.6 that a molecule with a charge + l e must overcome the barrier with the height of several fc^T in order to leave the cleft; that is why we can not neglect electrostatic effects in comparison with the thermal motions. 7.5.2
Mathematical
model
Consider enzyme molecule as a rigid static construction in which the positions of atoms are determined from the X-ray data. Two models can be used for the comparison of the rates of the fragments escape from the cleft for the cases of the opened and closed "back door". The first model corresponds to ACE molecule with the closed "back door" and contains all the atoms from the corresponding PDB file. The second model corresponds to the enzyme with the opened "back door" and must have a channel by which the reaction products leave the AS cleft. That is why some atoms must be excluded. This can be done in a way as follows. Assume that the channel has a cylindrical shape and that its radius is equal to 5pA. We
Kinetics
of the enzymatic
reaction of ester bond breaking
233
determine the position of the channel with the minimal number of atoms inside by varying the position of the cylinder axis and the coordinates of a certain point at the axis of the "back door". Delete the atoms from the channel (belonging to the residues Trp 84 and Met 83 , which is in agreement with the results reported in [Gilson et al., 1994]) to obtain the second model. Substrate and its fragments can be also presented as rigid constructions. However, in this case acetylcholine and choline molecules can be presented as dumbbells and acetate molecule can be presented as a sphere. Thus, the computational time is reduced substantially. The sizes of dumbbells and sphere are determined on the basis of the X-ray data for the corresponding molecules. Each point element of the enzyme construction has mass and charge. Van der Waals interaction of two points that belong to different objects (e.g., enzyme and the reaction product) can be presented by Lennard-Jones potential:
where r is the distance between the points and the constants A and B for different atoms can be found in [Volkenstein, 1988] (see also chapters 4 and 9). Introduction of the random force according to Langevin method allows one to take into account the impacts of the solvent molecules (see chapter 2). Assume that the enzyme is fixed, then the translational motion of mobile objects can be described by coordinates and velocities of their centers of gravity. Moving coordinates related to the center of gravity of each object are used for the description of the rotational motion. The orientation of the coordinate axes is chosen in such a way that the tensor of inertia is of diagonal type. This can be done in a way as follows. Tensor of inertia and its eigenvectors are calculated in the frame of the center of gravity. The axes of the new coordinates are directed along the eigenvectors. Then the equations of motion for the l-th object are given by [Khalilov, Chizhov, 1993]: mixi - ^2
F
ix+lixi
= £ix (t),
miVi ~ Y^ Fiy+1iyi = &w (*) > i
mm - ] T Fi+^zt
= £lz (t),
(7.12)
i
J1&11 = (J12 - Ji3) ft/2^3 + Yl {yi^
- ziF,iiy)
- 9nton + Oi (t),
234
Molecular scissors.
Jl2^l2 = {JlZ ~ Jn) fijlfiB + E
Cluster
model
(ziF'^
- XiF'iz)
(^'l
- ytF'l)
- 912^2 + Cl2 (t) ,
i
Ji3^i3 = {Jn - J12) nnfi«2 + E
~ 913^13 + Cw (<) •
i
Here mj is the mass of the object; x;, yi, and 2;; are the Cartesian coordinates of the center of gravity; Jn, J12, and J o are the components of the tensor of inertia; fin, fii2, and 0(3 are the frequencies of rotation around the axes of the objectcentered coordinates; x\, y\, and z\ are the coordinates of the z-th point of the object in the object-centered coordinates; £;x, £;y, and £iz are the projections of the random force on the axes of the laboratory coordinates; Oi, C/2> and £(3 are the projections of the random moment of forces on the axes of the moving coordinates; 7; is the coefficient of viscous friction; gn, gi2, and gi% are the coefficients that take into account the force of viscous friction in the case of rotational motion;.?1^, F[ , and F\z are the projections of the force F\, acting up on the i-th point of the object on the axes of the laboratory coordinates; F'\x, F'\y, and F'\z are the projections of the same force on the axes of the moving coordinates. Force F{ takes into account van der Waals and electrostatic interactions: F
i = E *«+viz = - E s radf 4 - Qh™dip3
i
Here 4>| • is the van der Waals force acting up on the «-th point of the Z-th object from the j - t h point of another object (subscript j runs over all points of all objects besides the Z-th one); q\ is the charge of the i-th point of the Z-th object; E is the electric field intensity. The translational Brownian motion was simulated in a way as follows. After equal time intervals At the velocity of the center of gravity of the Z-th object was instantaneously changed by the value Viv , where Vi is the additional velocity amplitude and v is a random vector quantity uniformly distributed in the cube [1,1;-1,1;-1,1]. All the other time the body moves according to the equations (7.11) at zero random force. An expression for the noise amplitude: _. V6fe B T 7/ Ai Vi = mi follows from the theorem on equal distribution of energy over the degrees of freedom miii \ _ in the case of diffusion in a free space in the absence of electrostatic fields 3kpT In a similar way we simulated the rotational Brownian motion. After equal time intervals At the angular velocity of rotation around the i-th axis was changed by a
Kinetics
of the enzymatic
reaction of ester bond breaking
235
R
Fig. 7.16 An area in the vicinity of the active site entrance. The substrate is located at Z axis in the point z\.
value ^1^, where SlH is the amplitude of the additional angular velocity and rji is a random quantity uniformly distributed in the [-1,1] interval. The amplitudes of the additional angular velocities are given by: £lti = ^/6kBTguAt/Ju. 7.5.3
Determination
of the rate constant
fci
Constant k\ determines the rate of formation of the complex EAS: wi = fci[E][S]. This constant can be determined in a way as follows. Assume that there are cylindrical walls impermeable for substrate molecules near the entrance to the cleft of the active site (Fig.7.16). Consider diffusion of AC molecule inside this area in the field of the enzyme molecule. EAS complex is formed if z-coordinate of the substrate molecule is less than z\. Repeating computations allows one to determine the complex formation frequency v. Mean concentration of the substrate in the selected area is [S]i. Then the mean ratio of the time during which the substrate can be found in the selected area to the observation time T is given by:
(j)
= [S\iVi
,
where V\ is the volume of the selected area. This relationship is valid if [S]iVi
ki[S\ = "(f)
= v[S]iV1 •
Molecular scissors.
236
Cluster model
k,, 10" M V 40 >
0.0
0.1
0.2
0.3
I, 10"3 M
Simulation time, us
Fig. 7.17 a) Plot of the number of the EAS complexes N versus simulationtime at zero ionic strength, b) Plot of the rate constant fci versus the ionic strength / . The error bars correspond to 66% confidence interval.
Mean concentration of the substrate [S]i inside V\ is given by Boltzmann distribution:
_q(p_
[S]i = [ S ] * p r / e x p
dV,
(7.13)
Vi
where q = -t-le is substrate charge. Introduce notation: e x
P l
- ^ ] ^ ;
Vi
then
ki =
i/fiVi.
(7.14)
In computations the volume V\ was confined by a plane z2 = 35pA and a cylinder of the radius R = 8pA. Z axis is the axis of the cylinder. It was assumed that zi = 20pA. Euler method with the time step of 2 • 10" 14 s was used for integration of the equations for substrate motion. One of the realizations for zero ionic strength is presented in Fig.7.17 where the number of EAS complexes is plotted versus time. Based on the results of several realizations similar to those presented in Fig.7.17 one can determine the mean value and the boundaries of 66% confidence interval for the frequency v. Fig.7.18 shows the plot of fci versus ionic strength (see equation (7.14)).
Kinetics
7.5.4
of the enzymatic
Determination
reaction of ester bond breaking
of constants
fe_i
237
and k?
The decay rate for the EAS complex is given by: w_i=k_i[£; A 5]. The rate of transformation of EAS complex into E*S is determined as:
w2=k2[£AS]. Hence, the constants can be determined in a way as follows. Assume that during time interval T EAS complex dissociates N-\ times with the formation of enzyme and substrate and transforms iV2 times into E*S complex. Then fc_i = lim ——-, T->oo
fco =
T
h m -—. T->oo
T
Assume that the initial complex decays with the formation of independent molecules if the substrate goes away from the bottom of the cleft by the distance Ri. This distance must be large enough to allow penetration of another molecule inside the cleft, i.e. the enzyme must be free. E*S complex is formed if z-coordinate of the substrate is smaller than zcat. X-ray data provide the coordinates of the AC molecule bound to the catalytic triad of ACE. Hence, zcat can be assumed to be equal to z-component of these coordinates. The simulation scheme is as follows. Substrate is initially located in the point (0, 0, z{). If the substrate reaches the catalytic triad (z = zcat) or leaves the sphere with the center in the origin and the radius R\, the result is recorded and the substrate is brought back to the initial point. The division of the number of the formed E*S complexes by the time of simulation yields the constant A;2; and the division of the number of the dissociated EAS complexes by the simulation time yields the constant k-\. In calculations zcat = 6.5pA and R\ = 25pA. The following reaction rates were obtained for zero ionic strength (the errors correspond to the boundaries of 66% confidence interval): k-i = (2.24 ± 0.08) * 10 9 s _ 1 ; k2 = (1.54 ± 0.10) * l O ^ " 1 . Using these results and the value of k\ calculated in the previous paragraph for 1 = 0 rewrite the equation from (7.11) leaving unknown constants fc_2, fc3, fc_3, and k^: K$
•
(2,41±0,18)xl09 1+ ^ 1 + k3
(M_lg_1}_
(7
fc
-2fc-3 k3k4
In experiments on enzymatic splitting of acetylcholine by acetylcholinesterase the ratio kcat/Ki^ was determined to be 1.6 * 1 0 8 M _ 1 s _ 1 [Ferscht, 1980, Quinn,
238
Molecular scissors.
Cluster model
1987], 2 * 108 M-i-s-1 [Fuxreiter, Warshel, 1998], and 6.5 * K ^ M " 1 ^ 1 [Berezin, Martinek, 1977]. However, it is not clear under which ionic strength these experiments were carried out. Experimental dependence of kcat/Kl$ on ionic strength is obtained for another substrate (acetylthiocholine (ATCh)). This dependence was extrapolated by means of the least squares method to the zero ionic strength [Quinn, 1987]. Thus, the ratio kcat/K}^ was determined to be 4.2*109 M ' V 1 . Table 7.2. Substrates of acetylcholinesterase. Structure Substrate Acetylcholine CH 3 C(0)OCH 2 CH 2 N+Me3 Acetylthiocholine CH 3 C(0)SCH 2 CH 2 N+Me 3 It is seen from the Table 7.2 that ATCh molecule differs from ACh molecule by one atom only. Therefore, the rate constants k\, k-i, and fc2for these molecules must be nearly equal. The value of the ratio kcat/K1^ obtained by means of extrapolation of the experimental curve for ATCh to zero ionic strength is larger than numerator of the ratio in the right side of (7.14). Hence, assume that A;_2 — Os"1. Then the scheme (7.6) can be rewritten as: fc-
E +S
EAS
E*S
ES
E*P
E + P.
Accprding to this scheme the reaction necessarily takes place if the substrate gets into the area of the catalytic triad. For the simplicity reasons omit the stages of formation and splitting of ESC and assume that the transformation of E*S complex into E*P complex is an elementary stage: fci.
E +S ,
K«
EAS^hE*S-^rE*P^E
+ P.
Then 1
1
1
(7.16)
^cat
kik2 jy'tm
k.
h
One constant k% is introduced instead of three unknown constants k 3 , k_ 3 , and k4; the relationship between them follows from the equations (7.10) and (7.16): 1 fc3
k±
k^k^
Kinetics
of the enzymatic
reaction of ester bond breaking
239
Knowing the constants k§ and kcat one can determine which process occuring in the active site cleft (diffusion of the reaction products or the chemical process leading to bond breaking) limits the enzyme operation rate. 7.5.5
Determination
of the constant
k^
Consider the escape of the reaction products from the cleft of the active site under the assumption that choline and acetate do not experience collisions with substrate molecules that move towards the entrance to the cleft. Such a situation is realized in the case of rather low concentrations of the substrate. The scheme of calculation of the constant k$ follows from the formula
u6 = k5[E*P],
(7.17)
that determines the decay rate of E*P complex. Let choline and acetate molecules be located in the vicinity of the catalytic triad. We simulate their diffusion in the field of the enzyme molecule and determine time interval in which the fragment closest to the origin shows up at the distance R\ from the bottom of the cleft. Repeating this procedure several times yields the escape rate of the reaction products. It follows from (7.16) that the constant k$ equals the mean frequency determined in such a way. The simulation was performed for two models of the enzyme molecule: with opened and closed "back door". For zero ionic strength the rate constant was determined to be &5 = (2.90 ± 0.17) * 10 8 s _ 1 (additional channel is absent) and ks = (2.57 ± 0.14) * 10 8 s _ 1 ("back door" is opened). The errors correspond to the boundaries of 66% confidence interval. Estimate the range of substrate concentrations under which there are virtually no collisions between the reaction products that leave the cleft and the ongoing flow of the AC molecules. If there is a single free enzyme molecule the mean time of EAS complex formation is l/(fci[S]); the mean time of escape of the reaction products is l/k^. The case under consideration is realized if 1/&5 -C l/(ki[S]), i.e. if [S]
(7.18)
Consider again the experimental data on the kinetics of acetylcholine splitting by acetylcholinesterase. The rate constant k c a t was measured to be: 1.4*104 s _ 1 [Ferscht, 1980] and 1.6*104 _ 1 [Fuxreiter, Warshel, 1998]. Consider the first equation from (7.16). Diffusion limited rate constants are four orders of magnitude larger than the experimentally determined kcat. Hence, fcjj w kcat and the chemical processes of bond breaking, sorption and desorption of the substrate appear to be the slowest among all the processes taking place inside the cleft of the active site.
240
Molecular scissors.
Cluster model
The latter can be not true for high substrate concentrations (when the condition (7.17) is not met). Indeed, the ongoing flow of acetylcholine molecules can substantially slow down the diffusion of the reaction products from the cleft. This problem is discussed in the next paragraph. 7.5.6
Substrate
inhibition
It follows from Michaelis-Menten kinetics that the increase of the substrate concentration leads to the monotonous increase of the rate of enzymatic reaction to a certain limit. However, experimental data show that there is a threshold value [S]opt above which the reaction rate decreases. This phenomenon is known as substrate inhibition [Braunstein, 1964]. The reason of the substrate inhibition lies in the interaction of the intermediate substances with another substrate molecule (or with several substrate molecules, which is less probable), resulting in formation of an inactive compound (nondecaying complex). Such a mechanism can be schematically presented in a way as follows [Braunstein, 1964]: fci
E + S ^ _ ES -^E
+ P,
k-i
fc3
ES + S ^ _
ES2.
fe-3
The reaction rate is KM + [S] + [S?/K'S% where
K
s
~
* » •
Brik and Yakovlev [Brik, Yakovlev, 1962] experimentally studied the dependence of the rate of acetylcholine hydrolysis by acetylcholinesterase on substrate concentration (Fig.7.18). It was demonstrated that the optimal concentrations are 10~ 3 M (for AChE from lamprey, frog, and mouse) and 5.6*10 -4 M (for AChE from flies). However, there can be another reason of substrate inhibition for this enzyme: the decrease of the escape rate of the reaction products upon the increase of the substrate concentration. In connection with this it is necessary to determine the value of the rate constant fcs for rather high substrate concentrations. After that it must be possible to determine which of the two reasons leads to the decrease of the reaction rate.
Kinetics
of the enzymatic reaction of ester bond breaking
241
a), ^M/min 6
5
4
3
1
'
-t
10
-3
10
-2
10
-I
Kl l J M ~ ' "*
10
Fig. 7.18 Plot of the rate of acetylcholine hydrolysis by acetylcholinesterase versus substrate concentration [Brik, Yakovlev, 1962].
Consider enzyme model with the closed "back door" and the following scheme of the fcs determination. In a way similar to that we used for determination of k\, consider an area with impermeable walls in the proximity of the entrance to the cleft of the active site. Let there be a certain number of substrate molecules inside this area; reaction products are located near the catalytic triad. Knowing the escape time for the reaction products one can calculate the escape rate for the known substrate concentration in the area. Constant fcs is equal to thus calculated frequency (see previous paragraph). In order to determine the relationship between the substrate concentration inside the area [S]i and the mean concentration [S] (formula (7.13)), one has to set the ionic strength of the solvent. The results of calculations can be compared with the experimental data [Brik, Yakovlev, 1962] if the simulation is performed at the ionic strengths under which the experiments were carried out. Unfortunately the authors do not present these data, so for each value of [S]i we vary the concentration [S] within the interval corresponding to ionic strength variation from zero to infinity. The left boundary if the interval can be determined from the relationship (7.13) under zero ionic strength. The concentration of the substrate is assumed to be constant inside the volume ([S]=[S]i) under very high values of the ionic strength. In calculations we considered an area in the vicinity of the entrance confined by a plane z — 50pA and cylinder with the radius of 15pA, coaxial with Z axis. Maximal number of AC molecules moving inside thus defined volume amounts to ten. The results of simulation are presented in Table 7.3. [S]i,mM 0 536 1070
[S], mM 0 11-7-536 22-fl070
k5, l O S " 1 2.90±0.17 1.56±0.14 1.07±0.05
Table 7.3. The dependence of the rate constant k 5 (the error corresponds to 66% confidence interval) on the substrate concentration [S]i in the selected area. The
242
Molecular scissors.
Cluster model
-0.05-
// // / / / / / /
-0.10-
U, V
-0.15 0.15-
// -0.20'
/
/- /
/ -0.250
. , 50. —— , —r10 -.— 20 _,- -. 30 , 40
Z, A
Fig. 7.19 Potentials of the initial enzyme molecule (solid curve) and enzyme molecule with shifted atoms (dashed curve) at Z axis at zero ionic strength.
intervals of the substrate mean concentration [S] corresponds to the variation of the ionic strength from zero to infinity. According to the experimental data the rate of the enzymatic reaction decreases approximately two times (Fig.7.18) if the substrate concentration is larger than the optimal one [S]opt by an order of magnitude (i.e. for [S] wlO mM). Assume that the reason of substrate inhibition lies in the decrease of the escape rate of the reaction products. Then under such concentration one can expect two-fold decrease of the constant kcat (relative to the case of zero substrate concentration). Then it follows from the previous paragraph and formulas (7.16) that k$ « A;3=(1.4-^1.6)*104 s _1 for [S] «10 mM. However, it follows from the Table 7.3 that at the concentration under consideration the rate constant ks is still four orders of magnitude higher than the rate constant feg. Therefore, this mechanism does not account for the substrate inhibition. Its reasons must be related to the formation of the inactive complexes mentioned in the beginning of the paragraph. In conclusion of this paragraph consider the validity of the proposed model of the enzyme molecule. We assumed the atoms to be fixed in the points with the coordinates determined by X-ray diffraction. That is why the potential of ACE molecule used in simulations did not change in time. In reality there are thermal oscillations of atoms, besides that, their coordinates can change due to conformational transitions. Estimate how critical these factors are for the simulation. Assume that at a certain moment the coordinates of atoms change in such a way that all the positively charged atoms of the enzyme move by lpA from Z axis and all the negatively charged atoms come closer to Z axis by lpA. Fig.7.19 shows the potential along Z axis calculated for this case and the initial potential. We simulated the escape of the reaction products and the entrance of the substrate into the cleft of the active site under the changed potential. Table 7.4 shows the results and the earlier determined rate constants.
Kinetics
of the enzymatic
Potential Initial Changed
reaction of ester bond breaking
k l s 1010 M - ^ " 1 3,75±0,20 7,98±0,42
243
K5> 108 s- 1 3,36±0,29 3,06±0,11
Table 7.4. Reaction rate constants calculated for the initial potential and the potential of the molecule with the displaced atoms (at zero ionic strength) Among all the rate constants characterizing the diffusion-limited stages of the AChE-catalyzed reaction this is the constant fej that undergoes the strongest changes since it depends not only on the electric field strength but also on the potential in the vicinity of the active site cleft entrance. It is seen from Table 7.4 that the use of the "perturbed" potential does not lead to pronounced changes in the simulation results which allows us to make some conclusions based on the calculations. It was demonstrated that the "back door" slightly influences the escape rate of the reaction products upon small substrate concentrations. At the large substrate concentrations the enzyme operation rate is limited by the chemical processes taking place in the active site. The rate constant of the latter is less than the rate constant of the reaction, controlling the escape of the fragments from the active site, by four orders of magnitude. We have attempted at accounting for the effect of the substrate braking by the increase in the escape time of the reaction products. Such a consideration could have given an answer to the question if there is a "back door" in the AChE molecule. Indeed, if the calculated value of [S]opt for the enzyme without the "back door" is substantially smaller than the experimentally observed one, one can assume that there is an additional channel in the active site cleft. However, if the substrate concentration is larger than the experimentally determined [S]opt by an order of magnitude, the mean escape time is still four orders of magnitude smaller than the characteristic time of the chemical stage. Thus, the idea of the "back door" is not proved by our simulations. Based on the existing experimental data we can not state that there is an additional channel in the AChE active site cleft. However, the authors are far from insisting on the inconsistency of the "back door" hypothesis. It is clear that the proposed model employs various simplifications and assumptions which means that we do not take into account some effects. Here are some of them. The reaction rate can be influenced by the oscillations of atoms forming the entrance to the active site cleft because the motion of the AC molecule to the catalytic triad through a kind of a "bottle neck" and the escape rate depend on the amplitude and frequency of the cleft walls. We used the Langevin method to take into account the chaotic impacts of the solvent molecules. The diffusion coefficient for the molecules inside the cleft was assumed to be equal to that in the free space. However, the properties of water (in particular, its viscosity) near the surface of the protein globule can be different from those of the "normal" water. It is also possible that H-bonds influence the motion inside the active site cleft.
244
Molecular scissors.
Cluster model
The "back door" problem is not only typical for the acetylcholinesterase. One meets this problem when studying the functioning of carboxypeptidase and some other proteins, too.
References B. Alberts, D. Bray, J. Lewis, M. Raff, K. Roberts, J.D. Watson (1940): "Molecular Biology of the Cell". Garland Publishing Inc., New York, London. J. Antosiewich, M.K. Gilson , J.A. McCammon (1994): "Acetylcholinesterase: Effects of ionic strength and dimerization on the rate constants", Israel.J.Chem. a, 34, 151-158. I.V. Berezin, K. Martinek (1997): "The principles of physical chemistry of enzymatic catalysis" (in Russian), Vysshaya Shkola, Moscow. F.C. Bernstein, T.F. Koetzle, G.J.B. Williams, E.F. Meyer, M.D. Brice, J.R. Rodgers, O. Kennard, T. Shimanouchi, M. Tasumi (1977): "The Protein Data Bank: a computer-based archival file for macromolecular structures", J. Mol. Biol. 112, 535-542. A.E. Braunshtein, Ed. (1964): "Enzymes" (in Russian), Nauka, Moscow. I.L. Brik, V. A. Yakovlev (1962) "Investigation of the properties of Cholinesterase in the nerve system of animals and insects" (in Russian), Biokhimiya 27, 993. D.S. Chernavsky, N.M. Chernavskaya (1999): "Protein-machine. Biological and macromolecular constructions" (in Russian), Izd. Mosk. Universiteta, Moscow. D.S. Chernavsky, Yu. I. Khurgin, S.E. Shnol (1978): "Mathematical biology and medicine. Principles of biological catalysis" (in Russian), VINITI Moscow 1, 9-59. A.Yu. Chikishev, W. Ebeling, A.V. Netrebko, N.V. Netrebko, Yu.M. Romanovsky, L. Schimansky-Geier (1998): "Stochastic cluster dynamics of macromolecules", Int. Journal of Bifurcation & Chaos 8, 921-926. A. Fersht (1977): "Enzyme Structure and Mechanism", Freeman & Co., Reading and San Francisco. M. Fuxreiter, A. Warshel (1998): "Origin of the catalytic power of acetylcholinesterase: computer simulation studies", J. Am. Chem. Soc. 120, 183-194.
Kinetics
of the enzymatic
reaction of ester bond breaking
245
M.K. Gilson, T.P. Straatsma, J.A. McCommon, D.R. Ripoll, C.H. Faerman, P.H. Axelsen, I. Silman, J.L. Sussman (1994): "Open back door in a molecular dynamics simulation of acetylcholinesterase", Science 263, 1276-1278. R. Heinrich, S. Schuster (1996): "The regulation of cellular systems", Int. Publ, N. Y. Albany / Bonn.
Thomson
V.R. Khalilov, G.A.Chizhov (1993): "Dynamics of classical systems" (in Russian), Izd. Mosk. Universiteta, Moscow. J. Musil, O. Novakova, K. Kunz (1980): "Biochemistry in schematic perspective", Czechoslovak Medical Press, Prague. G.A. Pasynsky (1963): "Biophysical chemistry" (in Russian), Vysshaya Shkola, Moscow. D. Porschke, Ch. Creminon, X. Cousin, C. Bon, J. Sussman, I. Silman (1996): "Electrooptical measurements demonstrate a large permanent dipole moment associated with acetylcholinesterase", Biophys. J. 70, 1603-1608. D.M. Quinn (1987): "Acetylcholinesterase: Enzyme structure, reaction dynamics, and virtual transition states", Chem. Rev. 87, 955-979. D.R. Ripoll, C.H. Faerman, P.H. Axelsen, I. Silman, J.L. Sussman (1993): "An electrostatic mechanism for substrate guidance down the aromatic gorge of acetylcholinesterase", Biochemistry 90, 5128-5132. Yu. Romanovsky , A. Netrebko, N. Netrebko, S. Kroo, A. Chikishev, I. Sakodynskaya, V. Molodozhenya (1999): "Enzyme Molecule Control of the Substrate Microflows and Some Problems of Optical Diagnostics", In: Proc. SPIE, Optical Diagnostics of Biological Fluids IV, Eds.: A. V. Priezzhev, Asakura Toshimitzu, Proc. SPIE 3599, 167-179. Yu.M. Romanovsky, A.V. Netrebko (1998): "Some problems of cluster dynamics: models of molecular scissors", Izv. VUZ "AND" 6, No 4, 31-44. J. Sussman, M. Harel, F. Frolow, C. Oefner, A. Goldman, L. Toker, I. Silman (1991): "Atomic structure of acetylcholinesterase from torpedo californica: a prototypic acetylcholine-binding protein", Science 253, 872-878. M.V. Volkenstein (1988): "Biophysics" (in Russian), Nauka, Moscow.
This page is intentionally left blank
Chapter 8
Dynamics of proton transfer in the active site of chymotrypsin A.Yu. Chikishev, B.A. Grishanin and E. V. Shuvalova 8.1
The basic model
We consider now one of the typical problems of the stochastic dynamics of enzymes - the process of proton transfer in the H-bond between oxygen of serine and nitrogen of histidine in the active sites of CT and ACE (Figs. 1.9, 1.10 and 6.1). It follows from the results presented in the previous paragraphs that the rate of breaking of ether bond is limited by the chemical stages after AC fixing in the proximity of the catalytic group of the active site. Nearly the same is valid for breaking of peptide bond in the active site of CT (see chapter 1). As proton transfer process is necessary stage of bond breaking, the rate of it must be higher than the operation rate of the enzyme: V = 106 s _ 1 for CT and for ACE. However, if the considered H-bond is isolated (substrate is absent), the probability of the overbarrier proton transfer is low because the barrier is rather high (more than 40 /esT) [Fersht, 1980] (Fig. 8.1). The probability of quantum-mechanical tunneling is also low because of the asymmetry of the two-well potential profile. Substrate interacts directly with a small fragment of the enzyme molecule (tens of atoms) — its active site. Thus, we consider the enzyme molecule as a whole as a classical system whereas the active site is considered at the quantum level. In chapter 1 we considered cluster models of ACE and CT. Oscillations of the clusters after substrate sorption are determined by the collisions of the protein molecule with water molecules. Basing on the results of the classical model calculations and using the Raman data [Romanovsky, Tikhomirova, Khurgin, 1979] one can estimate frequency fio and amplitude £o of clusters oscillations: ft0 ~ 10 11 -^ 1012 Hz,
£0 ~ 0.1-=-0.2.
(8.1)
Slow conformational changes of enzyme structure and, hence, of the potential profile of H-bond are related with substrate sorption [Popov, Kashparov, Popov, 1998; Sawaya, Kraut, 1997]. These changes are much slower than the oscillations of the subglobules or vibrations of the valence bonds. Thus, in our model we do 247
248
Dynamics
of proton transfer in the active site
l/m ole
-20 -
CO
-40 -60 -80 -
o
^ -100Z> - 1 2 0 -
-140-160-I
1
0,8
1
1
1 ,0
1
1
1 ,2
1
1
1 ,4
•
1
1 ,6
1
1
1
1 ,8
1
2,0
r, A° Fig. 8.1 It is shown potential field of Ser 1 9 5 - His 5 7 hydrogen bond of a-chymotrypsin active site without substrate.
consider only the thermal fluctuations of the clusters. In this paragraph we present quantum-mechanical consideration of proton transfer in the active sites of CT and ACE with regard to fluctuations caused by oscillations of clusters. We studied, first of all, the probabilities and characteristic times of proton tunneling in stationary two-minimum potentials that simulate active sites of CT and ACE. Stationary potentials were obtained by means of PM3 method [Steward, 1993]. In addition we studied proton dynamics and characteristic times for the case of the nonstationary potential with regard to fluctuations caused by thermal oscillations of the clusters. Original computer codes based on the method of symmetrization of the evolution operator allow one to calculate the time of the proton transfer in stationary and nonstationary two-minimum potential and time evolution of the proton wave function. The method of symmetrization of the operator of evolution makes it possible to determine the wave function of the proton and, hence, the probability of proton localization in one of the wells, total energy of the proton, and time of tunneling in a two-minimum potential.
8.2
Determination of the wave function by symmetrization of the evolution operator.
Time dependence of the wave function in the system with Hamiltonian:
Determination
of the wave function
by symmetrization
H = V(r,at))
of the evolution operator.
+^ ,
249
(8.2)
where £(t) is noise, which can be determined with the use of the evolution operator S(t):
V>(r,i) = S(i)VM).
(8.3)
For short time intervals t = At evolution operator can be presented as[Kosloff, 1988, Fleck et al.,1982]: S (At) f» exp
iV (r) At exp 2h
ip1 At exp 2mh
iV (f) At 2h
(8.4)
Wave function at any instant t can be determined by means of repeating of this algorithm k — t/At times. Error of this method is due to the discrepancy between the two representations of the evolution exponent in the case of noncommutative kinetic and potential energy operators. Due to symmetrization, the error is order of ~ At 3 , which is much smaller than ~ At 2 , when no symmetrization is provided. The time step At must be chosen small enough to assure stability of numeric scheme and accurate computation of the time-dependent wave-function. The criterion of this selection is based on the width of the energy spectrum, determined by the Fourier transform of (tp (r, 0) | ip (r,t)). The sampling interval should be chosen to accommodate the entire spectrum of bound state energy levels [Feit et al., 1982, 1990, 1991] At < 7r/3Ay max , where AV max is the maximum excursion of the potential. Initial wave function "0(r)O) can be determined as Gaussian function with the maximum in one of the wells. Eigen functions of the corresponding level can be also used in order to increase the accuracy of calculations and to study the other tunnel levels. In our case the initial wave function was presented as a superposition of two eigen functions (symmetric and antisymmetric) of the first split tunnel level (Fig. 8.2). The square of the absolute value of the wave function IV' {r, t) | is the probability density. Thus the probability of the localization of particle in one of the wells is given by: (r2+rx)/2
P(t)=
J
|V(r,t)| 2 dr.
(8.5)
Knowing the wave function one can determine the total mean energy of the particle:
250
Dynamics
r
i
1
Fig. 8.2
r
of proton transfer in the active site
• , mini
min2
Initial wave function in a two-minimum potential V(r) at the segment [ri,r2J.
E(t) = V(f) + ^=Jv(r)\i/,(r,t)\2dr
+ J
n
L-\^(p,t)\2dp.
(8.6)
-Pmax
Here [ n , ^ ] is the segment at which the potential is defined (Fig. 8.2) and the wave function is defined in coordinate presentation $f(r, t); [-pmaxiPmax] is the segment at which the wave function is defined in the momentum presentation *(p, t). The energy conservation law is valid for a particle in the stationary potential, E (t) = Const.
8.3
Checking the results
The correctness of the proposed algorithm for calculation of the wave function was tested with the use of the linear harmonic oscillator model. Potential energy in the Hamiltonian (8.2) for an oscillating particle is given by: V(r) = -z-r2, where LJ is the oscillator frequency. The initial wave function of the particle (proton) was defined as: (r - r 0 ) I f — Vn
i\> (r, 0) = A exp a = y/h/muj, A is the normalizing constant.
2<J2
Checking the results
251
Wave function, being shifted at an initial time moment, oscillates with the time period To = 2TT/UJ near the equilibrium position determined by the potential minimum. The mean energy of the particle is given by [Landau, Liphshits, 1989]: p2
—
mui2
9
TIUJ
For the following parameters m = 1.672648 x 1 0 - 2 4 g, u = 1014 Hz, TQ = r t = 0 = rt=T0 = cr — 0.25 x 10~ 8 cm, L = 40 x 10~ 8 cm we compared the results obtained by the method of symmetrization of the evolution operator with the theoretical results. The results of comparison are presented in Table 8.1. The energy of the particle remained constant in time.
Eu
*(&) = 2irh/T0 (kcal/mole)
Exact theory 62.83 1.507044
Symmetrization method 62.81 1.50752
Table 8.1. Comparison of the results obtained by means of the method of symmetrization of the evolution operator for the harmonic oscillator with the theoretical results. In addition we calculated the probability of tunneling of a particle in a twominimum stationary potential. It is known [Feynman, Lepton, Sends, 1966] that for a two-minimum potential U = Ui + Ui with eigenfunctions ipi and ip2'2
- + Uk\
1>k = E°k, fc = l , 2 .
(8.7)
The Hamiltonian of the system as a whole is described by a 2x2 matrix: Hki=(il>k,Hrl>ty
k,l = 1,2.
(8.8)
The splitting of the levels corresponding to the matrix (8.8) is given by:
AE = yJ(H22 - tfn)2 + 4#21ff12.
(8.9)
The probability of a particle localization in one of the wells is:
fAEt
p{i)=
vw)-) .
(810)
The time of transition from the first well into another is determined as: •K%
T
=AE-
(8-H)
Comparison of the calculation results for the proton transfer in a two-minimum potential with the theoretical results (8.10), (8.11) is shown in Fig. 8.3. The transition time for the first tunnel level and for the barrier of 10 kcal/mole was
252
Dynamics
of proton transfer in the active site
+
Theory Method of symmetryzation of evolution operator
t(ps.) Fig. 8.3
Comparison of the dependence P{t) obtained by means of the evolution operator sym-
metrization method with the theoretical dependence P (t) = cos 2 I -
1
2%
) •
determined to be about 10 12 s (the splitting of the first tunnel level is about AE as 1(T 3 kcal/mole). We studied also the influence of parameters of the potential (asymmetry, barrier height, the distance between minimums) on the main characteristics of the process of the proton transfer. It was demonstrated that the increase of the height of the potential barrier results in the increase of the tunneling time. In the case of the asymmetric potentials the relationship (8.9) can be rewritten as:
AE = y/(&U + (t/ x ) 2 2 - {U2)irf
+ 4H12H2l
(8.12)
where (E® — E2) = AU is the difference between the energy levels and
(Ui)22 = f fa (r) Ui (r) fa (r) dr,
(U2)n = J fa (r) U2 (r) fa (r) dr.
ri
n
Off-diagonal matrix elements of the Hamiltonian in (8.12) are of the same order of magnitude as the splitting in a symmetrical potential (RS A£^AC/=O)- Wave functions fa and fa are exponentially small in the areas of the minimums 2 and 1, respectively. If AU > AEAU=O then the main input into the splitting AE is determined by the term AU and the equality (8.12) can be rewritten as AE « AU. The period of tunneling in this case can be determined as: 27r/i ACT
(8.13)
Mathematical
model of the active site of ct-chymotrypsin
and acetylcholinesterase
253
The amplitude of tunneling probability in two-minimum asymmetric potential decreases by approximately an order of magnitude in comparison with that in the symmetric potential if the absolute difference between the depths of two interacting potential wells is comparable with the value of splitting in a symmetrical twominimum potential: AU « AEAU=0 « 1CT3 < kT (kBT = 0.6 kcal/mole). The period of tunneling is given in this case by the formula (8.11). In the case of large asymmetry (AU 3> AEAU=O) the period of proton tunneling can be estimated with the use of the formula (8.13). The amplitude of probability decreases in this case by several orders of magnitude. Thus, the most effective proton tunneling takes place in the symmetrical potential.
8.4
Mathematical model of the active site of a-chymotrypsin and acetylcholinesterase
It is known that catalytic activity is impossible without conformational changes of protein molecule [Popov, Kashparov, Popov, 1998]. Proton transfer is one of the fastest stages of the catalytic act. In the absence of the substrate the potential profile of the H-bond in the active site is strongly asymmetrical (Fig. 8.1). It is assumed that the conformational changes of enzyme upon substrate binding result in symmetrization of the energy profile [Fersht, 1980]. For CT the sequence of transformations in the active site is as follows. Aminoacid residue Ser214 shifts towards the substrate due to the formation of the H-bond. This results in breaking of the H-bond between Ser214 and Aspl02. The latter causes proton transfer in the H-bond between Aspl02 and His57. Imidazole ring of His57 becomes positively charged. At the same time an H-bond between Serl95 and substrate is formed (Fig. 5.1). All these events lead to symmetrization of the H-bond between Serl95 and His57. Note that the real symmetrization is instantaneous and related with the slow modification of the potential profile. Proton transfer that is necessary for breaking of the peptide bond of the substrate takes place in the quasi-symmetrical potential. The characteristic time of dynamic symmetrization of the potential profile is much larger than the typical period of thermal fluctuations of the clusters. Thus, the simulation of the proton transfer is performed under adiabatic approximation in the symmetrical potential profile with regard to only thermal oscillations of the clusters. The height of the potential barrier in a model of the CT active site calculated by means of PM3 method is about 14 kcal/mole, the distance between the minimums is about 1 pA. The model two-minimum potential can be presented as:
V{r,t)
= U1(r,t) + U2{r,t) = U(r,t) + U( -r + Ar,t),
where U\ and U2 are two Morse potentials:
(8.14)
254
Dynamics
of proton transfer in the active site
U1=U(r),
U2 = U(-r + Ar),
U (r, t) = D0 {exp [-2a (r - (r 0 + £(*)))] - 2 exp [-a (r - (r 0 + £(*)))]} .
(8.15)
here DQ is height of barrier, a is empirical parameter, Ar is the distance between atoms O and N in the H-bond, ro is equilibrium station, £ (£) is noise. In earlier work [Grishanin et al., 2000] we have studied the process of proton transfer in more adopted for reality potential of CT active site. We assumed that cluster oscillations result in fluctuational changes of the distance between the atoms in H-bond. Hence, the evolution of the potential profile in time is determined by the changes of the distance between the atoms. Two types of noise were used for simulation of the fluctuational changes of the length of the Hbond: color noise (that is model of temporal oscillations of subglobules) and white noise (that is model of temporal oscillations of valence bonds). The deviation of the distance between the minimums from the equilibrium one in the case of white noise is given by
CM = 2£o£(i),
£(<)e
l
l
"2'2
where £o is the fluctuation amplitude. Spectral density of white noise power is constant: G^ (w) = 2£Q/7T = G 0 = Const. The transformation of the white noise £(t) into the color noise x(t) was performed with the use of the equation: d2x _
+
„rdx „, . 9 2 * - + W g * = *(*)•
The amplitude and frequency of the thermal fluctuations are determined according to (7.1). Frequency Wo and the decay constant 8 are the parameters in the model of the color noise. In our work constant 5 was determined from the relationship for the quality factor Q — wo/25 « 10 [Romanovsky, Khurgin, Chikishev, 1988]. Note, for detuning value Aw = \w$ — wr\ of the central frequency of noise from the frequency of the cluster vibrations u)r = 1012 Hz (the frequency of the first tunnel level) the following relation is carried out:
Aw < w 13 ,2 3 , ^ (8-16) n Here wis^s is the transfer frequencies from the first and second to the third level, Ub is the barrier height. Fulfilling of equation (8.16) transitions from two lower level to higher states is excepted. Spectral density of colored noise power is: g[u) =
^ . ( w 2 - w 2 ) 2 + 452w2
(8.17)
Proton transfer in the H-bond of the active site
8.5
255
Proton transfer in the H-bond of the active site
We studied the time dependence of the probability of tunneling in a two-minimum symmetrical potential with regard to fluctuational changes of the distance between the atoms in the H-bond. The parameters of the Morse potential (8.15) were as follows:
a = 3pA~1,
Do = 30 kcal/mole,
AT = 0.05 fs, (8.18) A T = lOOAt is a time interval after which the distance between the minimums changes under the influence of the noise. The results of the study of the proton dynamics in the potential (8.14) with the parameters (8.16) by means of the method of symmetrization of the evolution operator are presented as time dependencies of the probability of proton localization in one of the wells and the total energy of the proton. The period of probability oscillations for the first tunnel level of the stationary potential is T = 5, l x l O - 1 2 s (the corresponding frequency is « 0.2xl0 1 2 Hz). The results of the study of the proton dynamics in a nonstationary two-minimum potential fluctuating under the action of the white noise showed that there are two ways of the proton transfer from oxygen to nitrogen in the H-bond of the active site: tunnel and overbarrier. The observed increase of the proton energy is related with dephasing of the wave function (Fig. 8.4b). The probability of proton localization in one of the wells tends to a stationary value of 0.5 (Fig. 8.4a). It means that the proton can be in the left and right potential wells with equal probabilities and the increase of energy can be an indication of the overbarrier transfer. The probability can be approximated with the use of the formula [Lax,1974]:
P t
()
= \ +\
r0 = 1 pA,
cos
Ar = 3 pA,
( 2 ™i*) e x P ( " ^ ) >
( 8 - 19 )
where TO and u\ are the varied parameters. The results of approximation showed that V\ coincides with the frequency of the tunnel transition under study: v\ = 0.24 p s _ 1 , T0 = 4 ps. Studying the influence of the color noise on the dynamics of the proton transfer in the potential (8.14) with the parameters (8.16) we varied the detuning Aw = OJQ— ojr of the central frequency of the noise UIQ from the natural frequency of the system ur = 0.2 x 1012 Hz. Figure 8.5 shows the dynamics of proton in a nonstationary potential where the fluctuations of the distance between the atoms of the H-bond were introduced as color noise with the amplitude defined by Eq. (8.1) and the frequency UJQ at the exact resonance with the frequency of the tunnel transition (Aw = 0).
256
Dynamics
of proton transfer in the active site
a)
b) 70 35 0 -35
t*3
-70
d
-140
c*3
" (erg)
-105
1
-175 J-210
Fig. 8.4 Dynamics of the proton transfer in a nonstationary potential with white noise fluctuations ( A T = 0.5 fs): (a) the probability of proton localization in one of the potential wells; (b) proton energy (EB = 23.8 kcal/mole is the barrier level).
The dephasing of the wave function is accompanied by the increase of the proton energy (Fig. 8.5b) and the probability decrease down to the level 0.5. The set of local maximums of probability P(t) can be approximated as 1 1 ^approx \P)
/
Q i 7} e x p I
(8.20)
Time evolution of the probability P(t) can be approximated with the use of function
Proton transfer in the H-bond of the active site
257
b)
t,ps
0
- i — - —
kcal /rrde
5
-10
LLf
-20
i
•^^V
-5
-15
-25 -3D
Fig. 8.5 Dynamics of the proton in a nonstationary potential with zero detuning of the color noise from the natural frequency: (a) the probability of proton localization in one of the potential wells; (b) proton energy.
(8.17). The parameters were determined to be T0— 1 ps, v\ = 0.24 p s - 1 . The value of TQ is one and the same for different realizations. The action of color noise with amplitudes (8.1) and detunings Aw = —F, 5F (F — ojres/Q) from the natural frequency of the system resulted in the changes of the value of r 0 : Aw = 5F: r 0 = 0.8 ps, vx = 0.24 p s - 1 ; Aw = -F: r 0 = 1.1 ps, vx = 0.24 p s - 1 . If the amplitude of the noise is decreased down to £o ~ 10~ 2 , the behavior of P(t) is absolutely different. In the case of noise action with small amplitude the time dependence of probability is nearly the same as in the stationary case. Deformation of the wave function and, hence, of the probability, is not accompanied by the relaxation to the level 0.5.
258
Dynamics
of proton transfer in the active site
Behavior of the temporal probability dependence under different detunings, asymmetries of potential, and amplitudes is fulfilled in our work [Grishanin et al., 2000]. In this work all of accounts have been calculated in the potential of CT.
8.6
Discussion
1) The relaxation of probability to the level 0.5 accompanied by the increase of the total energy can be caused by interactions with upper levels. In the case of white noise such an interaction with upper levels is provided by wide spectral range of the noise that contains the frequencies of all transitions between the levels. Spectral interval of the color noise is close to the tunneling frequency (8.17). Therefore, direct transitions to upper levels are impossible under conditions (8.16). The interaction with upper levels in this case can be explained in a other way as follows. Time evolution of the wave function is determined by the formula (8.2). In the case of stationary potential the wave function is the eigen function of the potential with the known parameters. If the potential is nonstationary, the wave function is determined for a potential with certain parameters whereas the parameters of the potential in the operator of evolution (8.3) are different due to noise action. Thus, the wave function determined before noise action is not the eigen function for the changed potential and it can be expanded over the new eigen functions
rP = J2CWk-
(8.21)
k
The stronger is the variation of the parameters of the potential the larger is the dispersion of the coefficients, which means the involvement of many levels. The interaction with the upper states results in overbarrier transitions and relaxation to the level 0.5. 2) Proton transfer in the active site is an irreversible process. Hence, the describing theory mechanism had to build taking into account irreversibility. Processes of proton transfer including both of reversible tunneling and irreversible overbarrier transition are most probably under condition of energy potential symmetrization. Moreover, this process becomes irreversible in the case of the nonstationary potentials. Overbarrier transition can be realized in stationary potential because of fluctuations of proton potential energy only [Kramers, 1940]. Such process has less probability then tunneling. In addition tunneling there are irreversible noncoherent overbarrier transitions in nonstationary fluctuating potential. In our model we did not take into account the feedback providing the variation of the potential profile upon the changes of the wave function. That is why tunneling as reversible process does not describe proton transfers, unlike irreversible overbarrier transitions. 3) Many authors also considered transitions of particles in stationary and nonsta-
Discussion
259
tionary double-well potentials. The important conclusion is made in paper [Kagan, 1991] due to overbarrier transition determines mechanism of particle transfer under certain parameters of fluctuating barrier. Authors of work [Blumenfeld et al., 1967] Showed by theoretically methods the mechanism of overbarrier transition can be realized under nonadiabatic conditions. Effects of classical molecular motions influence on quantum transitions rate are considered in [Ermolaeva, Shaitan, 1997]. The article [Elyutin, Rogovenko, 1999] Is also interesting for our tasks. In this paper the analysis of autolocalizated particle transitions is carried out by means nonlinear dynamic technique in stationary and stochastic double-well potentials. We employed quantum theory to study the proton transfer in a stochastic double-well potential of H-bond. The choice of quantum consideration rather than classical one requires some notes. Particle can cross the barrier by either thermal hopping and tunneling. In each case, the escape mechanism depends on temperature, shape of the potential and dissipative mechanism [Gammaitoni et al., 1998, Tang et al., 1997]. If tunneling and thermal hopping have the same contribution to transfer process, an incoherent regime is established and an analytical description becomes rather complicated. It is the region, witch our parameters belong. Although we started from quantum description, we have achieved an incoherent regime. Studying the effect of the colored noise on the double-well system, we start from considering the action of a periodic force. In this case, there are physical phenomena such as "coherent destruction of tunneling" [Grossmann et al., 1991], the "stabilization of dissipative coherence" with the increasing temperature [Dittrich et al., 1991,1993, Oelschlagel et al., 1993], effect of driving-induced quantum coherence, dissipation and decoherence [Grifoni et al., 1993, 1999, 2000] etc. The special interest is the stochastic resonance (amplification of a weak coherent signal in a threshold system by noise). Lofstedt and Coppersmith [Lofstedt and Coppersmith, 1994] considered the quantum stochastic resonance under incoherent tunneling at adiabatic driving frequencies. The stochastic resonance in classical , semiclassical and quantum cases have been investigated in [Gammaitoni et al., 1998, Klimontovich,1999]. Zhou [Pu et al., 1999] considered phase transitions in quantum tunneling for a parameterized double-well potential. Stability of localized quantum states in specific case of symmetric double-well potentials has been studied in [ Adhikari et al., 1996]. 4) The results obtained here can be useful not only for bio-molecular dynamics but also for the studies of chiral configurations, theories of superconductivity, in low temperatures physics and in the general problem of charge transfer in threshold systems. In our further studies we will consider the role of interaction of the peptide chain oxygen with the oxygen of serl95 with a plenty of negative charge in peptide chain breaking. We plan also to take into account feedback in our system and look into 2D and 3-D problems.
260
Dynamics
of proton transfer in the active site
References S. Adhikari, S.P. Bhattachargia, P. Dutta (1996): "Stability of localized quantum states on the top of the barrier and some of its consequence: the specific case of a symmetric double-well potential", Chem. Phys. Lett. 248, 218-222. D.S. Chernavsky D.S., N.M. Chernavskaya (1999): "Proteine-machine. Biological macromolecular constructions" (in Russian). Moscow: Moscow University. F. Grossmann, T. Dittrich, P. Jung, P. Hanggi (1991): "Coherent destruction of tunneling", Phys. Rev. Lett. 67, 516-519. T. Dittrich, B. Oelschlagel, P.Hanggi (1993): "Driven tunneling with dissipation" Europhys. Lett. 22, 5. P.V. Elyutin, A.N. Rogovenko (1999): quant-ph/9912026. R. Feynman, R. Leighton, M. Sands (1966): "The Feynman lectures on physics, v. 8 Quantum Mechanics" (in Russian), Mir, Moscow, p. 271. M.D. Feit, J. A. Fleck, A. Steiger (1982): "Solution of the Schrodinger Equation by a Spectral Method", J. Comp. Phys. 47, 412-433. M.D. Feit, J. A. Fleck (1990): "Wave-optics description of laboratory soft-X-ray lasers" Jr. J. Opt. Soc. Am B 7, 2048. M.D. Feit, J. A. Fleck (1991): "Spatial coherence of laboratory soft-X-ray lasers" Jr. Opt. Lett, 16, 76. A. Fersht (1977) "Enzyme structure and mechanism", Freeman Reading, San Francisco. L. Gammaitoni, P. Hanggi, P. Jung, F. Marchesoni (1998):"Stochastic resonanse" Rev. Mod. Phys. 70, 223-283. M. Grifoni, M. Sasseti, Stockburger, U. Weiss (1993): "Nonlinear response of periodically driving damped two-state system", Phys. Rev. E 48, 3497-3509. M. Grifoni (1999) "Dissipation, Decoherence and Preparation Effect in the SpinBoson System", Europ. Phys. J. B 10, 719-729. L. Hartman, I. Goychuk, M. Grifoni (2000): "Driven tunneling dynamic: Bloch-
Discussion
261
Rodfield theory versus path-integral approach", Phys. Rev. E 6 1 , R-4687. B. A. Grishanin, A. Yu. Chikishev, Yu. M. Romanovsky, E. V. Shuvalova (2000): "Quantum-mechanical model of proton transfer in a fluctuating potential field of the active site of a-chymotrypski", in: Jan A. Preund and T.Poschel (Eds.) "Stochastic Processes in Physics, Chemistry and Biology. Lecture Notes in Physics" 57, Springer Verlag, Berlin, pp. 338-349. F. Grossman, T. Dittrich, P. Jung, P. Hanggi (1991): "Tunneling in a periodically driven bistable system" Z. Phys. B 84, 315-325. Yu. Kagan (1991): "The role of barrier fluctuations in the tunneling problem", Ber. Bunsenges. Phys. Chem. 95, 411-421. Yu.L. Klimontovich (1999): "What are stochastic filtering and stochastic resonance?", Physics- Uspekhi 42, 37-44. R. Kosloff (1988): "Time-dependent quantum-mechanical methods for molecular dynamics" J. Phys. Chem. 92, 2087-2100. H.A. Kramers (1940): "Brownian motion in field of force and the diffusion model of chemical reactions", Physica 7 284-304. L.D. Landau, E.M. Lifshitz (1973): "Quantum Mechanics" (in Russian), Nauka, Moscow. M. Lax (1968) "Fluctuation and Coherence Phenomena in Classical and Quantum Physics" Gordon&Breach, New York. R. Lofstedt, S.N. Coppersmith (1994): " Stochastic Resonace: Nonperturbative calculation of power spectra and residence-time distribution", Phys. Rev. E 49, 4821-4831. B. Oelschlagel, T. Dittrich, P. Hanggi (1993):"Damped periodically driven quamtum transport in bistable systems" Acta. Phys., Pol B 24, 845. M.E. Popov, I.V. Kashparov, E.M. Popov (1988): "Theory and Method of a priori computation of catalytic acts of aspartic and serin proteinases" Adv. Exp. Med. Biol. B 436 123-126. Yu.M. Romanovsky, Yu.I. Khurgin, A.Yu. Chikishev (1985): "The analysis of proton transition in dynamic model of active site of enzyme chymotrypsin" (in Russian),
262
Dynamics
of proton transfer in the active site
Zh. Fiz. Khim. 59 2021-2025. Yu.M. Romanovsky, N.K. Tikhomirova, Yu.I. Khurgin (1979): "Electromechanical model of the enzyme-substrate complex" (in Russian), Biofizika 24, 442. M.R. Sawaya, J. Kraut (1997): "Structure-function studies of DNA polymerase error specifity" Biochemistry 36, 586-603. K.V. Shaitan, M.D. Ermolaeva (1997) "Dynamical modulation of quantum beating and transition rates in three-level system" (in Russian), Izv. RAN, Fisicheskaya seria 1, 1673. J.J.P. Stewart (1993): "MOPAC93, Manual", Fujitsu Limited, Tokio. J. Tang, S. H. Lin (1997): "Quantum tunneling versus thermally activated electron transfer in Ohmic and non-Ohmic heat bath", J. Chem. Phys. 107, No 9. F.-C. Pu, J. Liang, S. Kou, Y. Zhang, H.J.W. Muller-Kirsten, Y. Nie (1999): "Periodic instanton and phase transition in quantum tunneling of spin systems", Physics Letters A 253, 345-353.
Chapter 9
On the damping of cluster oscillations in protein molecules A.Yu.Chikishev, A.V.Netrebko, and Yu. M. Romanovsky 9.1
The estimate of damping of an oscillating ball by Stokes - Lamb - Landau theory
In this Chapter we pose several problems that have not been resolved yet. It is demonstrated in the previous Chapters that if the motion of subglobules, clusters, and atomic possesses the features of the colour noise (which means that the oscillations of these subunits are not overdamped), the motions in the molecular machines can lead to increasing efficiency. The problem of oscillations of large clusters in subglobules of protein molecules in natural environment (water) remains open. Below we present a brief review of the spectroscopic studies of proteins in crystal state at the frequencies close to 1012 Hz. The interest in the simulation of the subglobular motions is related to the experimental data regarding the low-frequency bands in the Raman spectra of proteins and polypeptides. The first measurements in the spectral region below 200 wavenumbers date back to the 70s (Brown et al., 1972). For relatively recent data see, for example (Calaianni &; Nielsen, 1995; Urabe et al., 1998). Table 9.1 summarizes low-frequency Raman data on several proteins (Painter et al., 1982). (The linear frequency in Hz can be obtained by multiplying the values in the Table by the speed of light c = 3 • 10 10 cm/s; for example, the resonance frequency for chymotrypsin is 8.7- 1011Hz.) The corresonding linewidth are about 10 20 c m - 1 . Note that the successful measurements of the low-frequency Raman bands were carried out in crystalline samples. The bands disappear upon protein denaturation and can hardly be observed in solutions. Although there is no commonly accepted interpretation of these data, it is admitted that low-frequency vibrations can be related to the conformational dynamics of protein molecules. The characteristic values of the frequencies (tens of wavenumbers) agree well with the results of the simplest simulation using a " spring-mass" model. In these calculations the force constant of the spring is estimated based on the H-bond potential and the mass is assumed to be equal to half-mass of the protein globule (for the case of two subglobules). The estimation of the frequency by the formula of the charac263
264
On the damping of cluster oscillations in protein
molecules
teristic frequency of the oscillations of an elastic sphere with the sizes and Young modulus typical of the protein molecule yields nearly the same results (Painter et al., 1982). The analysis of low-frequency Raman spectra of dry, semi-dry, and wet lysozyme crystals (Urabe et al., 1998) used damped harmonic oscillator model and the representation of the dynamic susceptibility as follows:
X (w) = J2 ^ w 0 i W [(<4 - w2)2 + w272 where WOJ the z-th resonance frequency and 27; is the corresponding linewidth. Fitting Raman spectra with this formula shows that the ratio of 7 and UJ is close to unity and is virtually independent on the amount of water in the sample. This can be considered as an indication of very low Q-factor of the corresponding intramolecular oscillations. Virtually all the authors working in the field suggest that the disappearance of the low-frequency Raman band in solutions indicates the overdamping of the corresponding oscillations. However, the estimations of the damping coefficient usually employ Stokes formula that can hardly be used in the case when the amplitude of oscillations is comparable with the linear dimensions of water molecules forming "continuous" medium. Thus, the purpose of our work was to develop a model allowing more or less adequate simulation of the interaction of protein oscillations with water environment giving rise to damping in the equation of motion of subglobules. First of all let us consider the problem of the Q-factor in the case of cluster oscillations in a protein molecule. For simplicity, we assume that the cluster or subglobule represents a sphere oscillating in a certain direction being bound to the "walls" by a spring. The value of Q is determined as:
« = £=4
(91)
'
where u is the frequency, 7 is the decrement, M is the mass and /? is the friction coefficient of the system. At first, let us estimate Q-factor for an oscillating ball in a viscous continuous medium. The ball is bound by springs to two walls and can oscillate along the horizontal direction. The Stockes, Lamb and Landau theory (Landau and Lifshitz, 1986, p. 130) yields an expression for the force acting upon a ball of the radius R and moving at the velocity V in a medium with the viscosity 77 and the density p:
where
The estimate of damping of an oscillating ball by Stokes - Lamb - Landau theory
265
(9.3)
5=
is the thickness of a boundary layer. It depends on the frequency of the oscillations. Here the first term describes the friction force Ffr and the second term F(n corresponds to additional inertia force related to the assembled mass
Ribonuclease A Lysozyme /3-lactoglobulin a-chymotrypsin Pepsin Ovalbumin Concavalin A Bovine serum albumin Bovine immunoglobulineG Adolase Thyroglobulin
Monomer 5,800 Dimer 11,600 13,700 14,000 Monomer 18,000 Dimer 36,000 22,600 35,000 44,000 55,000 67,000 150,000
Not observed 25 25 25 29 20 22 20 14 28, 36 "shoulders". 32 17
158,000 669,000
SnR\l^-(l+2^
Mas =
to to
Insulin
to to
Table 9.1. Low-frequency bands in Raman spectra of proteins Observed Raman Molecular weight Protein line [cm-1) A - 1
(9.4)
The formula (9.2) is valid if 5
and
a « l
(9.5)
where / is the characteristic dimension of the oscillationg body (e.g., / = 2R) and a is the amplitude. The Reynolds number
Re =
laujp
(9.6)
266
On the damping of cluster oscillations in protein
molecules
must be not small. To make the estimates, let us consider the molecule of a-CT (see Chapter 1) consisting of two subglobules (the mass of each subglobule is M — 12.5fc.Da K, 4 • 10~ 20 o and the typical radius is about 20pA). The eigenfrequency of oscillations of two masses connected by a spring Wo is about 2w • 1012Hz (see Table 9.1). The solvation sheath increases the mass by 10-30% (Khurgin, 1974). The lifetime of water molecules in the solvation sheath is about 10ns which is substantially larger than the period of oscillations. Therefore, the total mass of water molecules can simply be added to the mass of the subglobule. Assume that the mass of the solvation sheath is MH — 0.1M. In the case of one globule, we can use expression (9.2). If w —> 0 this formula is reduced to Stokes formula provided that Re -C 1: F = 6irr]RV + -nR3p— (9.7) o cut Table 9.2 shows the values of 7, d, and Q for several values of R and w. Inequalities (9.5) are satisfied for the parameters presented. Table 9.2. Characteristics of damping of oscillations of a ball in water obtained by hydrodynamics of viscous medium
RpA w0
=
21.5
A. Model of subglobule of q-CT 5pA x Mas M/10- 2 0 g w 20 12 2TT10 HZ io- g 5.84 4.48 0.66 9.08
710
12
s-1
Q
1.02
2.04
0.98 0.77
2.19 3.19
2TT10 12 HZ
3.88 1.94
20 15
RpA w0
=
1.75
8.48 6.54
0.69 0.78
B. Model of ethane molecule Mas x M/10- 2 0 g w 20 12 2TT10 HZ io- g 11.1 4.8 7.3 1.3
SpA
710
12
s-1
Q
0.53
0.66
0.52
0.50
2TT10 12 HZ
2.5
10.0
13.0
8.3
The total mass equals the sum of the mass of the subglobule, assembled mass Mas, and the mass of the solvation sheath MH'Mf = M + Mas + MH
(9.8)
Estimating
of Q-j"actor by the methods of statistical
267
physics
It follows from (9.2) that
2*y = 6rinR(l + R/6)/Mf
and
Q = ojMf
(9.9)
The radius of the subglobule obeys the equation M = -irR3p o where p = lg/cm 3 . Therefore, we have
In addition, we calculated the parameters of oscillations of ethane molecule in water (see Table 9.2). We did so to compare the results of our simulation with the results of Shaitan and Saraikin (Shaitan and Saraikin, 2000). It is no wonder that the formula (9.3) overestimates the friction coefficient. The thing is that the very idea of water as a continuos viscous medium is hardly valid at the oscillation frequencies of about 1012 — 1013Hz and the amplitudes of less than lpA which are smaller than the size of the water molecule. The thickness of the boundary layer is comparable with the size of the water molecule which makes the estimates by Stokes-Lamb-Landau theory inaccurate. An estimate of Q-factor based on the losses related to generation of the acoustic waves is also incorrect since the wavelength at the frequencies under consideration is less than lpA. Thus, we have to restrict consideration to the estimates based on the principles of the physical kinetics. Simple estimates yield the Q-factor of 10-20 for oscillations of such a system in water (Romanovsky, Chikishev and Khurgin, 1988).
9.2
Estimating of Q-factor by the methods of statistical physics
For simplicity, consider a parallelepiped with the dimensions typical of the chymotrypsin globule that is moving in water at the velocity VQ (Fig.9.1). The mean velocity of a water molecule at 300°K is V = 400m/s (V^ < V). The impacts of the water molecules from the right and from the left sides of the subglobule yield the momentums:
P1 = 2m(V + V0)nS(V + V0)dt/6, Pl = 1m (V - V0) nS (V - V0) dt/6, Thus, the "friction" force is given by:
{
'
268
On the damping of cluster oscillations
in protein
V
molecules
-V
V, Fig. 9.1
A simplified model of the motion of a subglobule in water.
F = (3V0 =
-mVnSVo.
(9.11)
where n is the number density of water molecules (~ 3 • 1 0 2 8 m - 3 ) , m is the mass 23 of a water molecule (18 Da or 3 • 110n ~ 23 g) and /3 is the friction coefficient. Then the damping decrement is:
27
M
4 m VnS. 3M
(9.12)
Since 5 = (40/oA)2 = 16 • lfr 1 4 cm 2 , 2 7 = 4 • 10 11 s" 1 and the Q-factor is Q = w / 2 7 « 10. If S = (20pA)2 = 4 • 10- 14 cm 2 , then Q « 40 and so on. It is well known that the H-bond network of water molecules allows the formation of quasi-crystals or clusters exhibiting the lifetime of 1 0 ~ u - 10~ 10 s (Wood, 1979). The typical size of such a cluster is 5 x 5 x 5 water molecules. As the cluster lifetime is much larger than the period of the subglobule oscillations, we can consider damping in the case when mi, V±, and n\ stand for the mass, velocity, and concentration of the water clusters rather than of isolated molecules. Water exhibits dense "packing" which means that vti\ = Nm and n\ — n/N, where N is the number of water molecules in one cluster. In addition, the mean kinetic energies of both the cluster and a single molecule are proportional to kBT. Therefore, V\ — V/N1/2. Thus, the damping decrement for the case of the interactions with clusters is smaller than that resulting from impacts of isolated molecules: 5\ = 5/N1/2. Hence, the Qfactor increases to 100. In reality, the Q-factor is smaller since a part of energy is additionally spent by excitation of oscillations inside the clusters.
Simulation
by molecular
dynamics
269
These results can by no means be considered as a proof of the possibility of resonance action on the rates of the enzymatic reactions. All the above considerations are related to a single protein molecule. The generalization of the results for the case of a molecular ensemble (real solutions of protein molecules) implies synchronizing the main stages of the catalytic act (e.g., complex formation) that can be just diffusion limited. Therefore, the resonance action at the frequencies corresponding to the subglobular oscillations is possible only in the systems with "coherent" state of the enzyme molecules (virtually all the molecules are in one and same stage of the reaction). Note that experimental realization of such states is possible in light-triggered systems. The action of the light pulses with the durations comparable to the characteristic times of the reaction stages provides the aforementioned synchronization. Let us estimate the values of 8 and Q by formulas (9.1) and (9.12) for the oscillations in ethane molecule. Consider ethane molecule as a dumbbell in which the masses are bound by a spring with the rigidity determined by the chemical bond between the carbon atoms. Assume that one of the carbon atoms is immobile (Shaitan and Saraikin made the same assumption). Effective cross section of CH3 group is comparable with that of water molecule. Assume that in (9.12) S — {bpA)2 = 2hpA2 and M = m. Then 28 = 4-10 12 and Q = 25 since u = 120-10 1 2 s _ 1 (Shaitan and Saraikin, 2000). If S = lOpA2, then Q = 68. Thus, the estimates by the simplest kinetic theory yield the values of Q that are larger than those estimated by hydrodynamics of viscous medium (see Table 9.2B) by two orders of magnitude. According to (Shaitan and Saraikin, 2000) the Q-factor is larger by one more order of magnitude: Q = 429.
9.3
Simulation by molecular dynamics
The main goals of the estimates by molecular dynamics are as follows: 1. To determine the validity of the estimates of the decrement 8 and the Q-factor obtained in the previous paragraph. 2. To determine the dependences of 5 and Q on the initial amplitude or the initial energy deposited to the vibrational system. 3. To determine the dependences of 8 and Q on the on the parameters of the potentials describing the interaction between water molecules and between water molecule and subglobule. 4. To determine the dependences of 8 and Q on the dimensionality of the model (2-d and 3-d cases). 5. To estimate 8 and Q for ethane molecule and compare the results with the above data. 6. To estimate the coefficient of self-diffusion of water based on the results for the system considered. Let us consider a numerical experiment employing the method of molecular
270
On the damping of cluster oscillations
Fig. 9.2
in protein
molecules
The shape of the potential defined by eqs. (9.13-9.14)
dynamics. A 2-d system shown in Fig. 6.4 was surrounded by water molecules simulated by flat disks of the diameter 2r = ZpA and the mass m. A model of the protein molecule was placed in the center of a square with the size L = QOpA. Such a 2-d area contains n — 312 water molecules. n=312 is the largest possible number of water molecules in such a monomolecular layer which corresponds to the filling factor of 0.8. Assume that the subglobule of a-CT represents a cylinder with the mass M. In these calculations Ms = M/s, where s is the ratio of the protein characteristic dimension (height of the cylinder) to that of a single water molecule. Assume also that the rigidities of all the springs k are divided by s. Hence the eigenfrequencies of oscillations equal the frequencies of the subglobules-cylinders 2 = M/k = Ms/ks. w Numerical experiments used the value Ms/m — 100, which corresponds to the dimensions and masses of the cluster and water molecule. The size of the square is such that at least five small disks are located between the globule and the wall provided that the large disk is in the center of the square (see Fig 9.2). The cylindrical subglobule was "cut" into s = 10 disks. The rigidities of the springs are such that the eigenfrequency of the horizontal oscillations equals 1012 Hz (in calculations fci = ki = k$ = k^ and the natural frequencies of the horizontal and vertical oscillations are equal). Lennard-Jones potential describes the interaction between the large and small disks and between the small disks.
Simulation
by molecular
271
dynamics
30 -20 -10
0
10
20
30
20
30
d
Fig. 9.3 Positions of water molecules at t = 0 (0) and t=200ps a) Ug = 25fc B X; Uw = 25kBT; b) Ug = 0.25fc B T; Uu = 25fc B T; c) Ug = 25fe B T; Uw = 0.25fc B T; d) Ug = 0.25fc B T; Uw = 0.25fc B T.
U** = AU0 U** =
-U0
fflar(*f
if
r < r\ =
if
r > ri
TQ\/2
(9.13)
or
U* = Wo
(?)"-(?)'
(9.14)
Here r is the distance between the centers of the particles; r\ is assumed to be equal to {R+r) for the "large-small" pair and (r+r) for the "small-small" pair. Figure 9.2 shows the shape of potentials (9.13) and (9.14) .
272
On the damping of cluster oscillations
30
30-i
20
20
10
10
OH
in protein
molecules
\ « ^ A ^
0
* -10
-10
-20 H
-20
-30
i
—
|
—
i
—
|
—
i
30 -20 -10 Fig. 9.4
—
|
—
i
—
0
|
—
10
i
—
|
—
i
20
—
|
30
-30
i
—
|
—
i
—
|
—
i
—
30 -20 -10
|
—
0
i
—
,
—
10
i
—
|
—
20
,
—
|
30
Trajectories of three water molecules for the cases (a) and (d) for the time t = 200 ps.
-1
50
'
1
100
'
1
150
'
1
200
'
~i—•—i—•—i—'—r
1
250
0
50
100
150
200
250
Fig. 9.5 Oscillations of a subglobule at OX and OY axes. At the moment t = 200 ps, the X component of the velocity experiences a stepwise change.
The values UQ = Uw (for the interaction between two water molecules) and UQ — Ug (for the interaction between a water molecule and the subglobule) were equal to 0.25/CBT (the case of "weak" interactions) or 25ksT (the case of "strong" interactions"). It is seen from (9.13) that the repulsing of the particles occurs at the intercenter distances smaller than the sum of the radii. There is no interaction at long distances which substantially reduces the computational time. Below we demonstrate that there is an insignificant difference between the results obtained with the potentials (9.13) and (9.14). If not specified otherwise, the temperature is T — 300 K. In this case kBT = 2.5kJ/M. Note that the determination of the parameter of rigidity or effective depth of the potential well UQ for the model of water molecules representing disks is a rather difficult problem. However, we can assume that in the case of formation of a "temporal" hydrogen bond HO-H UQ — WksT.
Simulation
by molecular
dynamics
273
In the case of the van-der-Waals interaction, UQ = 0.5fcsT. The effective e must be somewhere in between these two values. First, consider the results obtained for 2-d models. For clearness, we present all the data in the natural scales. Note that we did not try to establish one-toone relationships between the characteristics of the model and the subglobule. In particular, we assumed that the subglobules are equal and the density of the protein equals the density of water. Thus, the results are approximate and qualitatively describe the effect of the changes in the parameters. In the initial moment t = 0, all the water molecules possess randomly oriented velocities corresponding to the temperature T = 300K. The stochastization of the distribution of the velocities of water molecules and oscillations of the subglobule takes place in t = 200ps. Figure 9.2 shows the positions of water molecules at this moment of time upon various values of Uw and Ug for the case when the interactions are described by the potential (9.13). In the case of "weak" interactions, water molecules form small clusters or quasi-crystals. Figure 9.3 shows initial positions of water molecules at t — 0 and t — 200ps. Figure 9.4 shows trajectories of three water molecules corresponding to the same interactions as in Fig. 9.2a ("strong" potentials) and Fig. 9.2b ("weak" potentials). Figure 9.4 also shows the corresponding changes in the square of the distance. According to Einstein formula
<
2 r
> = 4Dwt
and it must be possible to determine the coefficient of self-diffusion of water molecules. Rough estimates yield the values 10~ 5 and 1 0 - 7 cm 2 /s for "weak" and "strong" potentials, respectively. Note that the values of Dw determined by the trajectories not ending up at the walls are nearly twice larger. The experimental value is Dw = 0.9 • 10~5m2/s for T — 292 K (Grigoriev and Meilikhov, 1991; Zatsepina, 1998). Hence, "weak" potentials provide better agreement with the experimental data. In the monograph (Wood, 1979), one can find the values of Dw obtained by molecular dynamics. For the room temperatures Dw = (2 — 4) • 10~ 5 cm 2 /s. The calculations employed a more complicated "threepoint" potentials of interaction between water molecules taking into account the directions of H-bonds. Figure 9.5 (for example shows oscillations of the subglobule in X and Y coordinates. At the moment t = 200ps, the subglobule acquires the initial velocity in the positive direction of the X axis with the absolute value equal to that of the water molecule so that the initial energy of the subglobule is 10 k^T. In addition to the time-dependences of the coordinates and energies of the subglobule, we present the spectral densities given by
274
On the damping of cluster oscillations in protein
2n*
molecules
0.2-i^ 0.1-
-0.1~ i — • — i — • — i — • — i — ' — i
200
204 208 212 216 220
200
-0.2
^ — ' — i — • — i — • — i — ' — i
200
204 208 212 216 220
205 210 215 220 225
5-i S/co)co
Fig. 9.6 Oscillations of a globule {"soft" spheres); Oscillations in X (a) and Y (b) coordinates after excitation at t = 200ps, Co = 251IBT\ (C) - Relaxation of the total energy of a globule (two cases); (d) - Kinetic eneggy of water molecule; (e)-(f) - Spectral densities for the oscillations of a globule (time interval 300 — 400ps)
Simulation
by molecular
dynamics
275
Fig. 9.7 Relaxation of the total energy of a globule after excitation at t = 200ps for the case: 1) Ug = 25kBT; Uw = 25fc B T; 2) Ug = 0.25fc B T; Uw = 25fc B T; 3) Ug = 25fc s T; Uw = 0.25fc B T; 4) Ug = 0.25fc B T; Uw = 0.25fc B T.
where a* and 6, are the coefficients in the expansion of the realization in sines and cosines with the frequencies wi = 27ri/5000. In some cases the spectrum was averaged over a few realizations. Note that we did not try to achieve maximum accuracy since the parameters of the models considered and the potentials of interaction were rather far from the real ones. We were mainly interested in the approximate estimates and qualitative dependences. First, consider the results for 2-d model. Figure 9.6 shows decaying oscillations in X and Y coordinates for the case when the initial energy of the subglobule is 100 ksT. The initial amplitude was as high as 1.5pA. Figures 9.6c and 9.6d show the relaxation of the total energy of the subglobule and the increase in the total energy of water owing to redistribution of the energy of the subglobule between 312 water molecules. As the initial energy of the subglobule is 100 fcsT, the mean energy of a single water molecule increases by 1/3. Figures 9.6e and 9.6d show the spectral densities built at the interval 300 — 400ps where there is virtually no effect of the initial impact at t = 200ps. The estimates by the mean decrement and the width of the spectrum yield the values Q — 12 and Q — 10, respectively. The difference is within the accuracy of our estimates. Figure 9.7 shows relaxation of the total energy of the subglobule (each curve is averaged over four realizations) for four types of water-water and water-subglobule interactions corresponding to the results presented in Fig. 9.2. The cases 2 and 3 exhibit two characteristic fragments in the decay curves. Fast decay (T\, 6I, Qi) takes place while the energy is larger than 5-8 KBT. It is changed by slow decay
276
On the damping of cluster oscillations
in protein
molecules
IUUU-
WOkT: V A
At 1*
-
V** -.2
IOOT
1
-
in
1U
200
VV
I
i
205
210
\ A i*\ * wi/v 1 1 i \# 1 * 11 ii ' i ' 11 *
215
220
i
t
225
Fig. 9.8 Relaxation of the total energy of a globule after excitation at t U0 = 1.25KBT: (1) "shorf'-range and (2) "long"-range potentials.
200ps for the case
(T"2I ^2, Q2) at low energy. The mean values were obtained by averaging over the entire interval. The results are summarized in Table 9.3. It follows from the data presented in Table 9.3 that: A) The Q-factor of the system increases with decreasing "strength" of the potentials. B) Upon "strong" potentials, the values of Q coincide with the results of calculations presented in paragraph 9.2. C) At small amplitudes of oscillations, the value of Q increases by almost an order of magnitude. Figure 9.8 and Table 9.4 show for comparison the results of calculations of the decay times for the "weak" potential (9.13) and the Lennard-Jones potential (9.14). The simulations were performed for soft water molecules and soft subglobule: Ug = Uw — 1.25fcsT. There is minor difference between the results which justifies the use of the potential (9.13) substantially decreasing the computational time. Table 9.5 shows the Q-factors of oscillations of the subglobule for various values of the eigenfrequency. It is seen that Q increases with frequency faster than linearly if the dimensions of the subglobule remain unchanged.
Simulation
by molecular
277
dynamics
Table 9.3. Characteristics of the decay of oscillations of the subglobule upon various parameters of the potentials of interaction and the eigenfrequency of oscillations u> = 2 • IP 12 . III IV I II Ug Ug ug 0.25kBT 25kBT 25kBT 0.2bk T
uw
25kBT
n
T
B
uw
uw
25kBT 3.7
0.25fcsT 3.7
0.25fcBT
37 6,25 0.25
20 6,0 0.25
0.025 0.16 4 40 6,0
0.05 0.16 4 20 6,0
uw
ps T-l
S = I ps-1
>
2.5
Si
12.5
r
Q=fs
s2
<5> Qx Q2
0.4
2.5
0.08
12.0
Table 9.4. Comparison of the characteristics of the decay of oscillations of the subglobule obtained for various types of the potential (9.13,9.14). U0 = Ug = Uw = 1.25fcBT rr** U* T (ps) 7.52 7.54 10.2 10.1 Q Table 9.5 Dependence of the Q-factor upon the eigenfrequency of oscillations of the subglobule. C/o T ( K ) w(l/ps) Q 2 9.09 22.62 2\/2 0.25 300 4 52.79 4\/2 122.84 2 2.76 2\/2 4.26 25 300 4 11.46 38.27 4V2 2 3.94 2\/2 8.74 0.25 600 4 12.95 4\/2 20.79
278
On the damping of cluster oscillations
in protein
molecules
Based on the rough 2-d model we can conclude that the Q-factor increases with decreasing density. Q increases with the dimensions and masses of the disks of the solvent. The experiments with 3-d model employed one-layer and two-layer models. In a one-layer model, water molecules were represented by spheres with the radius r whereas the subglobule was represented by a disk. As the height of the disk is slightly larger than 2r water molecules can move in the vertical direction. In a two-layer model the height of the disk was doubled as well as the number of the spheres (water molecules). The calculations for one-layer model yield the value of Q that is larger than that for the corresponding 2-d model. The value of Q is even larger for two-layer model. The reason for such a behavior of Q is the increase in the free space for the motion of molecules in 3-d model. Note that the values of Q determined by the decay time of the oscillations and the width of the spectral curve are virtually identical. Finally, we studied a 3-d model with the parameters corresponding to those employed in (Shaitan and Saraikin, 2000). A sphere simulating half of the ethane molecule was placed in the center of a cube and symmetrically bound to the walls by six equal springs. The mass and the radius of the subglobule were equal to those of water molecule. The 3-d volume considered contained 7 x 7 x 7 = 343 small spheres. The parameters of the potentials Ug = Uw = 0.25fcsT. The eigenfrequency of oscillations of the subglobule was equal to the frequency of C-C stretching vibration in ethane molecule (w = 1.4 • 1014 s _ 1 , period is T = 2.23-1013s, the rigidity of one spring is k = cj2m/2 = 520 kcal/mole 2 ). This is a rough model. A more adequate model employs a freely rotating dumbbell in which the masses are bound by a spring (valence bond). The energies of the subglobule remain virtually unchanged during 40 ps. The explanation is as follows. The amplitude of oscillations of the velocity of the subglobule is ten times larger than the mean velocity of water molecules (300 m/s for 2-d case). Water molecules have a very limited time for interaction with the subglobule when it has the maximum velocity. They meet it at the maximum displacement of the latter and simply change the direction of the motion upon head-on impact. We also calculated the oscillations of ethane molecule in the 3-d model with the allowance for the initial impact in the Ox direction. The corresponding plots are shown in Fig. 9.9. The decay curves also exhibit stepwise character.The left panel in Fig. 9.9 shows relaxation of oscillations of ethane in Ox, Oy, and Oz axes. The initial displacements of the subglobule in all axes are equal. The right panel shows the spectral densities for the velocities in these coordinates. It is seen that the oscillations represent undamped trains arising at the moments of impacts with water molecules. The mean Q of such oscillations is about 100 which is in agreement with the results presented in (Shaitan and Saraikin, 2000).
279
General discussion. Development of the model 0.8 -i x
0 50 25000- Ay(i>
100
150
200 250
20000 15000
I
10000 5000 H -0.8 0 50 16000 -,,40)
1.2
100
150
200
250
100
150
200
250
0.8 12000 04 0
h#iii|t/liiM> i««j
8000
-1)4 4000
-0.8 -1.2
n
'
1
2
i
'
i
3
'
i — > — i
4
5
0
50
Fig. 9.9 (left) 3-d oscillations of Ci?3-group in ethane and (right) the corresponding spectral densities.
9.4
General discussion. Development of the model
Let us briefly formulate the main conclusions. The method of molecular dynamics predicts vibrational motion of protein macromolecules in water. The calculations by
280
On the damping of cluster oscillations
in protein
molecules
hydrodynamics of viscous medium yield aperiodic motions. Stretching vibrations of small molecules (ethane) exhibit large Q-factors. The smaller are the sizes and mass of the molecules of solvent (or of their clusters) in comparison with those of the subglobule, the smaller is the Q-factor of the subglobular oscillations. The calculations based on hydrodynamics of viscous medium are incorrect since they substantially underestimate the (Q-factor. The simplified calculations for the model in which ethane molecule is represented by disks or spheres yield smaller values of the Q-factor than the calculations within the framework of more realistic models (Shaitan and Saraikin, 2000). The Q-factor substantially depends on the parameters of the potential of interaction. The Q-factor is larger for weak potentials which provide larger time of the water-subglobule interaction. Soft interactions between water molecules increase the free space in the system owing to the formation of short-living quasi-crystals of water. The coefficient of self-diffusion of water is close to the experimentally observed one. Let us formulate several problems posed by the studies presented. A) In the discussion of the models for simulation of oscillations of protein molecules in water we follow the idea of the cluster structure of the subglobules (see Chapter 1). Recall that the molecule of a-CT consists of two subglobules each of which contains six relatively rigid clusters (Fig. 1.2). We assume that the clusters are identical and the number of the degrees of freedom is reduced to 30-40. Such a model allows rough simulation of the substrate binding in the active site between the subglobules, evaluation of the conformational changes, and characterization of the related oscillations. Note that such a problem was considerd in (Romanovsky, Tikhomirova, Khurgin, 1979). As the members of the catalytic triad Serl95 and His57 belong to different subglobules, it is expedient to estimate the characteristics of the vibrational motions of the subglobules and clusters. B) The construction of the dynamic model of the solvation sheath of the subglobule is another interesting problem. It was demonstrated (Khurgin, 1976) that there are tens of sorption centers at the surface of the a-CT globule. These centers are randomly distributed and can not give rise to quasi-crystal growing. The construction of the model implies setting the network of the centers interacting with water molecules by Lennard-Jones and Morse potentials. Water molecules must be simulated by PUMA computer codes (Lemak, Balabaev, 1994) as it was done in (Shaitan, Saraikin, 2000). It is expedient to trace the formation of the quasi-clusters or quasi-crystals of water molecules. The initial velocities of water molecules must obey Maxwell distribution rather than a uniform one. When solving the problem of formation of the solvation sheath or cluster of small molecules, one must take into account the impacts with the loss in the kinetic energy that can be transferred into the energy of oscillations or atomic groups constituting the subglobule or cluster. C) It is necessary to more accurately determine the coefficients of self-diffusion of water in the models free of the effect of the walls, subglobules, and initial conditions.
General discussion.
Development
of the model
281
D) Estimating the damping for the "protected" degrees of freedom can be the next step in the simulations. In the simplest case a ID system consists of three masses and the water molecules interact only with the side ones. A similar problem was considered in [Netrebko et al., 1991]: only the end mass in the chain of 8 was interacting with the environment. It was demonstrated that the protected degrees of freedom exhibit larger Q-factor. In other words, the "hot" protected degrees of freedom can live rather long. Do not forget, however, that 2-d and 3-d systems exhibit nonlinear intermode interactions leading to stochastization and destroying the harmonic regimes. However, it was demonstrated in Chapters 3 and 5 that even in these cases there can be long-living periodic motions along special trajectories. As there are thermal noises in such systems, the motions in the protected degrees of freedom are color-noise-like. E) In Section 9.1 we already mentioned that the experimental determination of the resonance frequencies of proteins in the frequency range 1012 —1013 Hz (30 - 300 c m - 1 ) is rather difficult because of either Rayleigh scattering (Raman spectroscopy) or water absorption (IR absorption spectroscopy). Note that the Q-factor of an individual molecule must be larger than that determined from the bandwidth of the corresponding vibrational band since the band is additionally inhomogeneously broadened due to the conformational differences in the protein molecules in the ensemble under study. F) It is expedient to study the effect of the parameters of the interaction potentials on the distributions of the velocities of water molecules and subglobules.
282
On the damping of cluster oscillations
in protein
molecules
References C.L. Brooks, M. Karplus, B. Montgomery (1988): "Proteins. Theoretical Perspective of Dynamics, Structure, and Thermodynamics". Advances in Chemical Physics 71, Wiley, New York. K.G. Brown, S.S. Erfurth, E.M. Small, W.L. Peticolas (1972): "Conformational^ dependent low-frequency motions of proteins by laser Raman spectroscopy", Proc. Natl. Acad. Sci. USA 69, 1467-1469. S.E.M. Colaianni, O.F. Nielsen (1995): "Low-frequency Raman spectroscopy", J. Mol. Structure 347, 267-283. L. Genzel, F. Keilmann, T.P. Martin, G. Winterling, Y. Yacoby (1976): "Lowfrequency Raman spectra of lysozyme", Biopolymers 15, 219-225. Yu.I. Khurgin (1976): "Hydration of globular proteins", Mendeleev Journal 21, 684-690. L.D. Landau, E.M. Lifshits (1986): Hydrodynamics (in Russian). Nauka, Moscow. A.S. Lemak, N.K. Balabaev (1994): "On the Berendsen Thermostat", Molecular Simulation 13, 177-187. Yu. M. Romanovsky, W. Ebeling, eds. (2000): "Molecular Dynamics of Enzymes" (in Russian), Izd. Moskovskogo Universiteta, Moscow. N.V. Netrebko, Yu.M. Romanovsky, E.G. Shidlovskaya, V.M. Tereshko (1991): "Damping in the models for molecular dynamics", Proc. SPIE 1403, 512-514. P.C. Painter, L.E. Mosher, C. Rhoads (1982): "Low-frequency modes in the Raman spectra of proteins", Biopolymers, 21, 1469-1472. I.S. Grigor'ev, E.Z. Meilikhov, eds. (1991): Physical Quantities. Handbook, Energoatomizdat, Moscow. Yu.M. Romanovsky, N.K. Tikhomirova, Yu.I. Khurgin (1979): "Electromechanical model of the enzyme-substrate complex", Biofizika, 24, 442 (1979). Yu.M. Romanovsky, A.Yu. Chikishev, Yu.I. Khurgin (1988): "Subglobular motion and proton transfer model in a-chymotrypsin molecules", J. Mol. Catal. 4, 235-240 (1988).
General discussion.
Development
of the model
283
K.V. Shaitan, & S.S. Saraikin (2000): "On the effect of the amplitude of fluctuations on the friction coefficient of the Brownian oscillator in water", Biofizika, 45, 407-413. H. Urabe, Y. Sugawara, M. Ataka, A. Rupprecht (1998): "Low-frequency Raman spectra of lysozyme crystals and oriented DNA films: dynamics of crystal water", Biophysical J. 74, 1533-1540. D.W. Wood (1979): "Computer simulation of water and aqeous solutions". In : Water. A comprehensive treatise. Ed. Pranks F. Recent advances. Plenum Press, NY-London 6, 279-409. G.N. Zatsepina (1998): "Physical properties and structure of water". Izd. Moskovsk. Universita, Moscow.
This page is intentionally left blank
Chapter 10
P r o t e i n dynamics a n d new approaches t o t h e molecular mechanisms of p r o t e i n functioning K. V. Shaitan 10.1
Topology of hypersurfaces of conformational energy levels.
It is well known, and we have demonstrated this several times in the previous chapters, that functionally active biopolymers exhibit interesting and nontrivial dynamic properties. The ideas and concepts regarding protein dynamics undergo evolution in time. Instead of simple models, such as a model of conformational substates (Prauenfelder et al., 1979,1988; Goldanskii et al., 1986), the Brownian oscillator (Shaitan & Rubin, 1980; Knapp, Fisher & Parak, 1983), and the models explained in the first 9 chapters of this book, now a new more complicated picture will be developed. In terms of physics, despite of the well-ordered spatial arrangement of atoms, the' proteins often behave dynamically as disordered systems. In connection with this, the nonexponential kinetics of chemical processes and non-Arrhenius temperature dependencies of the relaxation times arises as we have discussed on many places of this book. At present, there are reasons to assume that premises for macromolecular level self-organization and regulation of biological systems are inherent in the structural-dynamic biopolymer organization. The problem can be divided into three closely related ones: mass transfer, energy transformation, and information transmission within biomacromolecular structures (Shaitan, 1992,1994,1996). Note, that biological systems appear to have physical preconditions for accomplishing all three processes simultaneously during the functional act. This has been quite reliably established for some steps of photosynthesis (Shaitan et al., 1991). From the viewpoint of classical physical chemistry this situation is rather unusual. For example, in solution, the vibration relaxation of the reaction products takes a significantly shorter time (10~ 12 s) than the consecutive collisions between the reacting molecules. The correlation radius in liquids is smaller than the average distance between the reagents. Thus, there is no spatio-temporal correlation among the elementary acts of particular chemical steps. The reaction energy transforms in the most primitive way into heat. There are also no physical preconditions for interactions between elementary acts of the processes not related directly to the chemical mechanism. Another situation arises in biomacromolecular systems form285
286 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
Fig. 10.1 Potential energy level hypersurface (3) for an ideal crystal (1). The second part (2) of the figure represents a multidimensional paraboloid.
ing microheterogeneous structured media (Shaitan & Rubin, 1983). It features the multiplicity of nonequivalent pathways on a hypersurface (HS) of potential energy level where a transition can occur from one state to another (Shaitan, 1992, 1994, 1996). This involves the distribution functions of activation energy, relaxation times and some other parameters characterizing a dynamic process. Prom the point of view of modern mathematics, the general properties of protein dynamics are caused by structure peculiarities of potential energy level hypersurfaces. The key notion here is the conformational energy hypersurface (EHS) U(q). If we have N generalized conformational coordinates q = (qi, ...,qu ), then the total (thermal) energy of the system is (in the harmonic approximation) E = NkBT. The points satisfying the equation E — U(q) belong to the conformational energy level hypersurface (ELHS). The system moves in the classical area of the configurational space over the entire hypersphere U(q) < E. It is possible to analyze the formation of such a HS. Specifically, in the case of an ideal crystal the potential energy function is quadratic in displacements of atoms and the potential ELHS is topologically equivalent to the hypersphere (Fig. 10.1). If there is a single defect in a crystal, we have two hyperspheres connected by a handle or a tube (Fig. 10.2). Thus, in ideal crystals with small energies E, the potential energy is a quadratic function of the relative atomic displacements from the equilibrium positions in the crystalline lattice and the ELHS is topologically equivalent to a hypersphere. We
Topology of hypersurfaces
of conformational
energy levels.
287
Fig. 10.2 Potential energy level hypersurface (4) for the crystal with single defect (1). Part 2 shows the potential energy profile along defect jumping coordinate and part 3 represents the potential energy hypersurface.
meet another situation in dynamically disordered systems, for instance, in proteins. The structure of the ELHS is very complicated here even at small energies E. It happens so because there are some degrees of freedom in proteins in which the potential energy has local maxima and minima, in addition to the absolute minimum. Such degrees of freedom are represented by hydrogen bonds and rotations around single bonds. In the presence of local minima and relatively high potential barriers, more phase space regions become available at ordinary temperatures. A cross section of a conformational EHS U(q) by an energy level E ~ NksT differs greatly in its topological structure from that of rigid molecules. There is a large number of disconnected areas that create conformational substates appearing as local minima in a one-dimensional picture. A general structure of the ELHS can be developed on the basis of the Morse theory. According to this theory the structure of a HS is determined by its properties near the critical points (these are the points in which all the first partial derivatives of the potential energy equal zero). These properties are determined by the Hessian (matrix of the second derivatives) of the function at the critical point. When all diagonal elements of the Hessian are positive, the situation is reduced to that of an ideal crystal. When in some degrees of freedom the potential energy has a local maximum at the critical point, the corresponding diagonal elements are negative. In the simplest case where only one diagonal element is negative, there is a saddle point. Figure 10.2 shows the projection of the HS onto the corresponding plane. In the case when there is a large number of negative and positive diagonal elements of the Hessian at the critical point, the projections can be presented only schematically (Fig. 10.3a). There are many attracting regions or basins in the configuration space. They are connected by handles or tubes with lower dimensions or at least with dimensions
288 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
Fig. 10.3 (a) Potential energy level hypersurface for a conformationally labile macromolecular system (a schematic drawing), (b) Map of the topologically non-equivalent paths of diffusion x(q) (broken lines) on the hypersurface of the level of conformational energy: 1 and 2 (2'), initial and final conformations (states 2 and 2' are indistinguishable with respect to the position of the group with the coordinate xy); 3- saddle point for the case of two measurements; 4- local minimum for the system without conformational degrees of freedom (the pathways between "A" and " B " are topologically equivalent).
less then the dimensions of the basins. There are a lot of directions of movement out of the multi-dimensional basin (Fig. 10.3b). This basin restricts the movement in the vicinity of the critical point. The trajectories of movement cannot intersect the total energy lines. The outgoing direction possesses low dimension that corresponds to the number of the negative diagonal elements. In a three-dimensional projection, the situation can be represented as a set of "octopuses" connected by "tentacles" (Fig. 10.3a). The structure of EHS is determined by the degrees of freedom with local maxima in the potentials. The situation is typical of crystals with defects, segnetoelectrics, glasses, hydration layers and other systems. The specificity of polypeptides and similar compounds featuring a large number of the degrees of freedom related to rotation around single bonds and hydrogen bonds lies in the fact that the energies of local maxima are scattered in a very narrow interval of a few kcal/mol. Thus, the network of bonds between basins becomes saturated and ramified. Note that even for tetrapeptides, the number of different saddle points is of the order of 104 (Czerwinsky & Elber, 1998). Such a topological structure of the energy level hypersurfaces results in a number of consequences for the dynamic behavior of the system and its functioning. This is important for organizing and controlling the functional processes. Without
Dynamic
correlation functions
289
and free energy maps
going into details of the topological properties of the HSs, we note that their specific property lies in the existence of a multiplicity of topologically nonequivalent pathways in the classically exposed region of the configurational space which binds substates 1 and 2 at room temperature (Fig. 10.3b).
10.2
Dynamic correlation functions and free energy maps
The basins (Fig. 10.3) are connected by many pathways that are not topologically equivalent. So, there is a web of possible but not equivalent pathways of conformational relaxation. What are the probabilities of these pathways? We have not got yet any general solution of this problem, but we have suggested an approach based on detailed analysis of the dynamical properties and corresponding conformational energy level hypersurfaces of protein fragments (Shaitan et al., 1997, 1999, 2000). We consider a series of dipeptides with modified groups at the ends. Side groups are varied with aminoacid sequence changing. The scanning of the accessible configuration space at given temperature by figurative point is carried out by molecular dynamics. All atom-atom interactions are included (Brooks, Karplus, Pettit, 1988, Balabaev, Lemak & Shaitan, 1996). The technical problem lies in selection of calculation conditions, at which the dynamic trajectory gives rather good representation of probabilities of all accessible states in a molecule. It means that the trajectories possess the ergodic properties. It was established, that it is possible only at rather long trajectories (about 5000 ps) at the temperature about 2000 K. We use the original method of collisional dynamics (Lemak & Balabaev, 1994; Shaitan, Balabaev et al., 1997) in which we introduce a model solute of low viscosity (~ 1 centipoise) allowing for an effective energy exchange between intramolecular degrees of freedom. Let us consider a typical 2D distribution function of dipeptide for a pair of dihedral angles (Fig. 10.4). This function was obtained by integrating the multidimensional distribution function
P(an,am)
= / . . . / P(a1,...,ai,...,aN)
JJ
da*
(10-1)
i^n,m
where P(ai,... a* ,...,a;v) is the probability density function. The peculiarities of the conformational energy level hypersurface structures were investigated by calculating the free energy maps. Note, that the free energy is related to the probability of the corresponding states by the well-known Boltzmann formula. These maps are strongly different from, for example, Ramachandran maps or potential energy maps since the free energy maps include the entropy factor too. Fortunately, we found only a few typical kinds of free energy level maps for all pairs of dihedral angles studied (about 400 variants of sequences with all twenty aminoacids have been investigated). It was obtained that the structures
290 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
0.2.
Xi
150
0 0 Fig. 10.4 2D probability function for dihedral angles of tyr residue in tyr-trp molecule. Angles vary from -360 to 360 degrees in linear scale.
IIc
\ CH3.
NH ) cr
o
M^
\ NH
j 0 - \ ^>c°yC~
-^CH2V^ V,
NHJ
^ ^
P-cr
^CH3
o
Fig. 10.5 Modified gly-asp dipeptide molecule and the conformational degrees of freedom (dihedral angles). R = CH 2 - COOH.
of the corresponding regions of the free energy level maps are directly related to the dynamic correlation of the conformational degrees of freedom. Consider typical examples. There are several typical maps of free energy levels for the pairs of dihedral angle. Let us consider pairs 2X2i-plane. Two of them are connected by a narrow bottleneck (Fig. 10.6.1). Free energy level map for the ip2, X21 pair looks like the map of a homogeneous rough surface (Fig. 10.6.2). Dynamic cross correlation function for the dihedral angles can be defined as (Shaitan et al., 1997,1999)
Dynamic correlation functions
.
f5 ?
'
and free energy maps
• i"'
, t # A 4 f^J. -&L ^ ^-ift :,'i * -;:v* ^ : W
u
^3S\ 5 - * 0
; «*
291 < V- V "
4
•A
KH-
F*&> O | . : ^ / s
.IX-' ' % 'I
*&
• -^ ? > " • -
^\
it™
f
' '.
2
1 •
^
„ ' * •
*^H -»^_
\
*' ""\ S**f
a
•
S 5 -s- •
.
5
*.* -"
is
*
»
,
'
•
f
•
5<5
-S...R.\.,A> .»
« «^ ,^°»A.l
" im
}
..3
Fig. 10.6 Free energy level maps of gly-asp: (1) ip2 X21 and (2) i/>3, X21 plots. The angles linearly vary from -360 t o 360 degrees.
0.12 0.1 O
I 0.08 o0.06
i
W I/ \ti m
^
v
i i
A' '
o
I 0.04 Q
0.02 IV \ /J I —V—Lu
A
1
206
AM A
^ ' 300 l /4Q0 / 500
600
T . DS Fig. 10.7 The real part of the cross correlation function for (1) 2, X21 dihedral angles.
F a / 3
(r)
=
^ ei[a(t)-a(t+T)}
e-i[0(t)-P(t+T)}
\_/ei[a(t)-a(t+T)]\/e-i[P(t)-0(t+T)V
(10.2) where a and /? are the values of the angles at the moments t or t + r . The real parts of these functions for the pairs of angles under consideration are substantially different (Fig. 10.7). There is a dynamic correlation for if2, X21 degrees of freedom and it is a result of a transition from one basin to another through the bottleneck (Fig.
292 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
10.6.1). On the contrary, the second pair of angles ip2, X21 exhibits no correlation in accordance with the corresponding free energy level map (Fig. 10.6.2). The analysis of the free energy level maps shows that the strongly correlated motions are possible if (1) there is a rather curved valley with great amplitude, (2) the basins are connected by a narrow tube, (3) there is a set of finger-like paths from one basins to another and (4) a long valley is formed around basin (Shaitan et a l , 1997, 1999, 2000). On the other hand, the uncorrelated degrees of freedom are observed in the same molecule if (1) there are no ways from one basin to another and (10.2) there is a broad, rectilinear and slightly rough valley only. We revealed another interesting phenomenon in the dynamics of amino acid residues and called it dynamic isomorphism. It was shown that autocorrelation functions of some aminoacid residues are virtually identical (Fig. 10.8). The autocorrelation functions are defined by equations (10.2) at a = j3. Figure 10.8 shows time-dependence of the autocorrelation functions for ipi angle in asp-asp and x n angle in tyr-trp. The similarity of the plots results from the peculiarities of the structure of the potential energy level hypersurface. It is likely that the amino acids exhibit dynamic rather than structural similarity (Shaitan et al., 1999). We investigate the dynamic symmetry effect by varying the chemical structure of the natural aminoacids. We demonstrate the effect of tyr modifications (Fig. 10.9) using another type of the autocorrelation function (compare with eq. (10.2)):
F M = ( e ia W
e-ia(t+r)
,ia{t)
(10.3)
Note that a minimum chemical modification of the natural amino acids leads to pronounced changes in the dynamic properties and breaking down the dynamic symmetry effects. It can be demonstrated in more details on the free energy maps for these modified molecules (Fig. 10.10). We select the
10.3
Restricted diffusion along a given pathway
Assume that each path x(q) (Fig. 10.3b) represents a certain Markovian processes. Figure 10.11a shows a typical potential energy profile for a given pathway x(q).
Restricted
diffusion along a given pathway
293
Fig. 10.8 The autocorrelation functions for the dihedral angles in (1) asp-asp and (2)-(6) tyr-trp: (1), (2) 2-
This function has a large number of minima and barriers. The dynamic properties of systems with such potentials have been discussed (Frauenfelder et al., 1980, 1990, Zwanzig, 1988; Shaitan k Rubin, 1980; Shaitan, 1992, 1994, 1996). It is well known that in biopolymers the characteristic times of conformational relaxations of atomic groups are much larger than the relaxation time of velocities. Under these conditions, let us consider the one-dimensional local diffusion. It can be described by the generalized Smoluchowski equation (see Chapter 2):
ip(*''>
d_D[x) dx
8P(x,t)+^dU»pM dx kBT
dx
(10.4)
Here P(x,t) is the probability density. The initial potential U(x) contains two terms (Fig. 10.11b): U{x) = U0(x) + e(x),
(10.5)
where U o is an envelope of the local minima and e(x) is a barrier function that determines the diffusion coefficient: D(x) =
D0exp[-e(x)/kBT}.
(10.6)
294 Protein dynamics
and new approaches to the molecular mechanisms
functioning
Tyrl
Tyr
F(x)
of protein
HO - ® - C H j - C H 2 -
HO-@-CH2-
0.8 0.6 0.4
{
-
Tyr3
Tyr4 1
. 4 OH
-....HO
0
Tyr2
3 -C: -
10
1 IS
0.2
20 x,ps
30
40
Fig. 10.9 The autocorrelation functions given by eq. (10.3) for dihedral angles
In experiments, we observe the coordinate of the group under study. The coordinate changes as the system moves via a multitude of nonequivalent pathways. Each region of the configurational space corresponds to a certain state of a given molecular group (Fig. 3b). In this case, the potential energy and the diffusion coefficient appear to be random functions describing changes in the position of the given group. Let us consider rotations in the functional groups. Let the coordinate x be the rotational angle (Fig. 10.3b). In experiment we detect a transition from angle x\ to angle x
(10.7)
where est (x) is the height of the barrier representing a random quantity characterized by the distribution function p(e). We do not make explicit calculations to determine p(s) and consider it as a model function only. We shall turn now to the formula for the mean first passage time characterizing the transition from one position to another. From literature follows (Zwanzig, 1988) that this time can be represented as:
Restricted diffusion along a given pathway
295
Fig. 10.10 Free energy level maps on the 'P2, X21 plane for the dipeptides containing modified tyrosine residues (see also Figs. 10.5 and 10.9). Angles vary from -90 to 90 degrees in linear scale ('P2 is in abscissa).
Xf2
< T> =
J < exp [e (x) /kBT]
> ~:~:1:)dx,
(10.8)
Xf!
! x
(7f(x)) =
-00
<exp[-e(y)/kBT]>Po(y)dy,
(10.9)
296 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
Fig. 10.11 (a) Potential energy profile along x(q). (b) Stochastic potential for the movement along the coordinate Xf(b) (compare to Fig. 10.3b).
oo
< exp [±e (xf) /kBT] > =
pXf (e)exp [±e/kBT]
de,
(10.10)
o where Po(x) ~ exp[—Uo(x) / kBT] and the random functions are averaged. Here we perform averaging with respect to the topologically nonequivalent pathways. The dynamic effects can be different, depending on the pattern of /9-distribution. The transition rate will have an Arrhenius plot only for very narrow distributions. It means that HS exhibits a relatively simple structure and a single pathway linking the states. In the dynamically disordered systems discussed here one can predict several temperature effects. For example, we may assume gamma-distribution functions for barrier energy (Rubin et al., 1989). In this case we can see the effect of localization of a group at the temperature below threshold. This effect is similar to the Andersen localization in quantum mechanics of disordered systems.
10.4
The mechanism of non-Kramers kinetic effects in proteins and glass forming liquids under diffusion limited conditions
In the condensed phase the rate constants are affected by the dynamics of the surrounding. In the disordered and microheterogeneous systems the Kramers theory is not correct and the rate constants depend on viscosity as follows:
kr ~ k,o
-exp(-e/kBT)
(0 < a < 1)
(10.11)
where rj is the viscosity and e is the barrier height of the reaction. Several approaches based on correlation of the fast fluctuations described by generalized Langevin equation have been discussed (Grote & Hynes, 1980; Bagchi & Oxtoby, 1983; Frauenfelder et a l , 1988; Zwanzig, 1992). In this we have to come back to one of the most essential problems investigated in this book: Why does the Kramers theory
The mechanism
of non-Kramers
kinetic
297
effects
fail to describe rather slow processes in the disordered media? A new approach to the theory of these effects will be developed here . We start from the equation for the probability density P(x,t) to find the conformational position x of the molecular group in the initial chemical state at the moment t:
d_ P{x,t) dt
d_
dx
D(x)
dP(x,t)
1 dU0
dx
ksT
dx
P(x,t)
- k{x)P.
(10.12)
Here D is the diffusion coefficient (D ~ kgT/r]), U(x) is the potential energy, and k(x) is the chemical rate constant depending on x. Following Kramers, we neglect the chemical term in the equation (10.12) and assume that an irreversible reaction occurs at the point on the top of the barrier (Fig. 10.12). In other words, we assume that k(x) = 0 if x < xr, and k{x) —> oo for x > xr . This can be taken into account by using the absorbing boundary condition P(xr,t)
= 0.
(10.13)
Note that in all these cases the Fokker-Planck (or Smoluchowski) equations lead to the Kramers formula at the strong viscosity limit. The modified viscosity dependencies of the reaction rate can be obtained by allowing for the memory effects (or correlation of fluctuations). In equation (10.12), these effects can be described by the high order derivatives. On the other hand, there is a simple reason of the non-Kramers effects for both fast and slow processes. We mean the nonequivalence of the particle positions with respect to the reaction rate (see the last term in equation (12)). There are many physical reasons for taking this function into consideration. For example, this function can describe the effects of microheterogeneity and disordering on chemical reaction rate. The parabolic dependence in k{x) can be used to describe the diffusion through fluctuating bottleneck (Zwanzig, 1992). We consider this problem in a simpler way and generalize this result by estimating k(x) in the general form (Shaitan, 1994):
k(x) ~fco+ v
x — XQ
where S is the characteristic length and v sionless variables:
7 > 0,
~ e elkBT.
We introduce the dimen-
S2 T £>' " and represent the probability density in the form: x — XQ z = — -5 — ;'
p z
{. ,t)
= y^2,o,nPn{z) exp
(10.14)
(10.15)
T
(10.16)
298 Protein dynamics and new approaches to the molecular mechanisms
of protein
functioning
The kinetics of this process is given by: oo
N(t) = f P{z,t)dz.
(10.17)
— oo
Let us start from the mathematically convenient situation: UQ = 0 and 7 = 2. In this case pn obeys the following equations: d2 •J^Pn + (A* - g4z2)Pn
=0
(10.18)
which formally correspond to the Schrodinger equation of a harmonic oscillator. It is well known that X2n = k0r + g2 (2n + 1); n = 0,1,2
(10.19)
and Pn(z)
=
(M^) 2 Hn {9z)
ex
P(- £ T")
(10-2°)
Now we consider the solution (10.16) with two initial conditions by using the following relations: A:
P(z,0) =8{z) = X ) P „ W P „ ( 0 )
(10.21)
n
B:
p(Z, o) = i = ^ Yl (- 1 )" p« (*) p» (°) "
(10-22)
n
These relations define the coefficients o„ in (10.16). To obtain the compact results for reaction kinetics we use the Mehler formula: A(
. ,
, ,
{2xya-{x2+y2)a2\
1
D4^-frB(*)F„(») =7
r
^-^ \
n=0
Representing a as exp (-2vt/g2) B), we obtain:
/
(10 23)
1-^
-
*
(in case A) and as i x exp (-2vt/g2)
^
/
,
1 + CT2
, ,\
(in case
The mechanism
of non-Kramers
kinetic
effects
299
B:
""> - ip-«*l-uM'*{-'«-,W
(10 25)
'
In the case B at t —> 0, N(t) diverge as t~x'2 in accordance with the initial condition (10.22). These formulas represent the exact solution. Thus we can see that kinetics are approximately exponential and the effective rate constants are given by:
keff = k0 + VvD62 = k0 + -^exp
(-gj^f J •
(10.26)
The physical pattern that corresponds to this result is as follows. In the beginning, the reaction takes place in the regions with large k(x) and the probability density drastically decreases in these points. Owing to fast reactions the sharp gradient of P arises and then there is a balance between diffusion and chemical reaction processes which leads to non-Kramers effects. Let us return to a more general dependence (10.14), that can be represented as: k{z) = k0 + i/|zf> 7 > 0 .
(10.27)
We can use the same mathematical techniques and obtain the equations for pn and A
n-
d2 j^Pn+
{X2n - 94 \z\y)pn
= 0.
(10.28)
The analog of the WKB method can be used to construct an asymptotic solution (10.28) under the diffusion limit if g4 > 1.
(10.29)
Under these conditions, the eigenvalues A^ can be obtained by using the well-known asymptotic relationship
/«-^r) S = f(» + ^);n = 0 , 1 , 2 , . . . ,
(10.30)
where z\ and z-i are the roots of the equation A^ - fl4 | Z |7
=
0_
(1Q-31)
300 Protein dynamics
and new approaches to the molecular mechanisms
a
of protein
functioning
c
"-o y 1
> Z 0 =0
A-o
/
k
^-**'r
•'
I I -Zo
0
Zo
Z0
0
Fig. 10.12 Determination of the reaction rate in a chemically heterogeneous medium at the diffusion limit VT > A^; (a) 0 and 1 are the boundaries of the region of diffusion; the coordinate of the minimum of the rate constant of the chemical reaction z0 lies within the potential box; (b) plot of the reaction rate constant in the case of diffusion through the fluctuating gap versus the gap width z; z0 is the van-der-Waals diameter of the ligand; (c) minimum of the function k(z) is close to the boundary of the region of diffusion (the broken line denotes the mirror reflection of the function k(z)); (d) minimum of the function k(z) lies outside the region of diffusion of the particle (the dashed lines denote (1) the mirror reflection of k(z) and (2) the linear approximation of k(z) in the vicinity of the boundary of diffusion.
The lowest eigenvalue AQ provides the main contribution to the kinetics: |i +
wr
; _ ^ ( 7 + 2)r(l
+
l) (10.32)
Then the formula for the effective rate constant is written as
heff
ko +
^eXP{'(,
+
2)kBT
7 a = 7+2
(10.33)
where e is the activation energy of the chemical reaction. This expression describes the non-Kramers effects for all values of the parameter a. Note that the threshold type of k{x) corresponds to the limit 7 ->• 00.
Mass transfer, energy transformation
and control in structured media
301
Let us change the physical situation and assume that the particle diffuses in a rectangular box (-zTtzr) (Fig. 10.12). How can we use the above results in this case? It is necessary to consider the effect of the additional reflecting boundary conditions on AQ. Under diffusion limited conditions (10.29) the modifications are rather simple: 1. If Z\ and z2 are within the box interval and AQ • oo) with the width of the order 2zQ (Fig. 10.12b). In this case a = l. 3. The left reflecting boundary lies at z — 0 (Fig. 10.12c). Only even pn functions obey the reflecting condition and hence the result is the same. 4. The minimum of the function k(x) lies outside the box interval (Fig. 10.12d). In this case k(z) can be expanded in the boundary point as k(z) ~ fc(0) + k'(0)z and the result will be described by expression (10.33) at fco = fc(0) and a = 1/3. Note, once more, that the most often value a = 1/2 arises, if k(x) has an ordinary local minimum in the diffusion region (7 = 2).
10.5
Mass transfer, energy transformation and control in structured media
The features of mass transfer in biomacromolecular systems as distinct from simple liquid and solid states stem from the heterogeneity of the medium and presence of relatively rigid bound structural elements, forming a strongly fluctuating framework as was considered in foregoing chapters and in (Shaitan, 1992, 1994). Diffusion within such structured media is related not only to general factors as in the case of liquids, but also to fluctuation opening of cavities and gaps formed by relatively rigid structural elements by a value exceeding the van-der-Waals diameter of ligand. Below this situation is considered for a model of diffusion through a fluctuating gap. The model employs a more graphic approach than the earlier work (Shaitan et al., 1985) and uses the idea of movement along a given pathway on the conformational energy hypersurface. Figure 10.13a shows the section of hypersurface U(q) in the system with ligand diffusion along the x coordinate through a fluctuating gap, the opening of which is characterized by the x coordinate. The broken line marks the optimal pathway (corresponding to the minimum values of T) of ligand transfer through the gap. In liquid there can be another optimal pathway
302 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
(dashed lines in Fig. 10.13a). Using (10.8), we obtain the characteristic time of the diffusion act. The integration pathway can therewith be naturally divided into three parts: < T > = <Ti>
+
2
>
+
3
>
(10.34)
In the situation presented in Fig. 10.13a, the gap opening is the limiting step (1) and T ~ T\ . In this case we obtain a well-known result: ~
Tcexp(j2f\
(10.35)
where r c ~ Tz is the characteristic time of the gap conformational relaxation, es is the gap tension energy, determined by the rigidity and size of ligand xo . Using the above method one can see how the diffusion act is affected by various factors, such as the matching between the gap and the ligand, the effect of the ligand on the gap walls, the interaction between the ligand surface and the inner surface of the gap, etc. The presence of relatively rigid fluctuating structural elements alongside with the structure of the conformational EHS discussed above create a clear physical pattern for controlling mass transfer in the systems considered. Thus, an alternating force balance resulting from changes in the chemical state of, for example, the charge of the groups (Fig. 10.13c), alters the potential energy surface. In the case shown in Fig. 10.13b, this significantly affects ligand diffusion because of a change either in the wall geometry of the equilibrium gap or in its rigidity, stipulating a decrease in effective activation energy. Therefore, within a structured and strongly fluctuating medium the elementary act of mass transfer takes place by a cooperative (self-coordinated) system rearrangement over many degrees of freedom. The change in the conditions of motion (equilibrium point, potential energy profile, energy distribution of potential barriers, etc.) even in one degree of freedom will give rise to a corresponding change in the rate of ligand diffusion, including changes in the optimal pathway in the configuration space. On the other hand, changes in energy characteristics of the conformational degrees of freedom are tightly connected with elementary processes of energy transformation in the chemical reactions in biomacromolecular systems. Thus, a change in the electron (chemical, charge) state of the functional groups gives rise to a change in not only the potential energy surface of low-amplitude intramolecular motion but also in the conformational energy hypersurface, U a (q) -» Ub(q) (Fig. 10.13). Therefore, the reaction energy is not immediately spent on the excitation of the vibration degrees of freedom (heat). A part of the energy AE(q)=Ufc(q) - U a (q), depending on the conformation q wherein the reaction took place, is stored in boosting the conformational degrees of freedom. Obviously, the heat effect AQ 0 (more correctly, AQ 0 = -AGo +TAS C where AGo is the change in the free energy of the reaction and ASC is the conformational contribution to the change in entropy) exceeds AE c (q) (Fig. 10.13d). If AQ 0 = AE C , then all the
Moss transfer, energy transformation
and control in structured media
303
reaction heat will transfer into the conformational stress. If AQ0 < AEC (q), then the reaction is practically impossible in the q conformation. It is known that the relaxation in conformational degrees of freedom is significantly slower than the vibrational one. However, in the former case, the main effect is not the delayed heating. The rearrangement of the conformational energy surface makes new regions of configuration space accessible, as the system moves along the given pathways. Since the reactivity of functional groups depends on conformation, we have a physical basis for diverse interaction mechanisms among various processes. Let us consider one of the simplest variants. Let two chemically independent reactions A —• B and C -> D run in the system. Consider the above example of the diffusion through a gap and assume that the A —• B reaction changes the charge and, consequently, the balance of forces which determines the gap opening (Figs. 10.13b, 10.13c). C —> D reaction is the mass transfer. We assume that the first reaction rate constant kaf, does not depend on conformation. The rate constant kc(j(q) depends on the gap opening (conformation) in a stepwise manner. The conformational energy surface changes in the course of the A —> B transition: U a (q) -> U&(q) (Fig. 10.13d). In this case, the system is characterized by two probability densities P a c (q,t) and Pf,c(q,t), where the subscripts correspond to the state of the functional groups. The dynamics of transformation is described by a set of simultaneous equations:
dPac
dt dPbc
dt
t^a*ac
~ kab* ac ~~ "-cd \Q) *ac
(10.36) AbPfcc
+ kabPac
- kcd (q) Pbc
Here the operators A a and Af, describe diffusion within the configuration space along surfaces U a and U& respectively. In the case discussed the rate constant kC(j(q) is different from zero within the configuration space region practically unreachable along the former surface U a (q) (large es): kcd(x0) » kcd(x°). Thus, the reaction will take place as "pumping" the BC state or the system transfer to the surface U& with subsequent relaxation (Fig. 10.13d). At the hypersurface structure considered, a strict chronological ordering of elementary reaction acts not directly related to the chemical mechanism takes place. The characteristic time of the C —• D transition is estimated as < r > ~ k~bl + < rp > + (fcc*d)-\
(10.37)
where < r p > is the mean time of the conformational transition along the surface Uj, to the q region corresponding to relatively large kcd(x0) ~ k*d (Fig. 10.13d). Thus, in the case considered, the energy of the A —> B chemical reaction partly transforms into the deformation of the conformational energy surface, which fi-
304 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
nally triggers the C —> D reaction. However, this event occurs not by a primitive (and practically impossible) transfer of the energy of the A —> B reaction to the C molecule, but by a fine mechanism closely related to the information transmission and processing in biomacromolecular structures. In the example considered, the information on the elementary act A —>• B is transmitted by relaxation of the system into a new (probably, metastable) state. This information is stored for the lifetime of the state. The information is processed in a relatively fast C -» D reaction. Thus, particle C acts as the Maxwell's demon, selecting the systems with the microscopic process A —>• B. More realistic is the case when the conformational evolution of C is not predetermined. Certainly, there is a great number of pathways along the conformational energy hypersurface not leading to the D state even if the gap is open (Fig. 10.13b, state 2). Particle C undergoes the required transition C —» D , if it receives information that the gap is open. This implies a certain interaction between C and the gap. Finally, this is associated with the system transfer to a more localized region of the configuration space 1 (Fig. 10.13b). The loss in entropy ASi exactly coincides with the amount of information transmitted in units of Boltzmann constant (Wiener, 1961). At constant temperature and pressure the decrease in entropy can be compensated only by useful work performed at the expense of the free energy of the chemical reaction A —> B : TASi > AGo- This is not a very strict condition, but it shows that within the framework of the mechanism discussed the absolute value of the decrease in the free energy during the reaction starting from the information transmission should not significantly be lower than 1 kcal/mole. As for the C —¥ D dynamics, a decrease in the particle C information entropy leads to a sharp drop in the number (or to selection) of the most probable diffusion pathways of C along the hypersurface Ub(q) (insets 1 and 2 in Fig. 10.13b). That is a common property of all the Markovian processes. The larger is -AGo , the harder is the selection. Whether the selection would promote or hamper the C —• D reaction depends on the particular structure of the potential energy HS. Note that the mechanism presented can not be realized on HSs with a developed energy minimum, as in the case of molecules without conformational degrees of freedom. The reason lies in basically unique result of the relaxation transition along such surfaces. However, in conformationally flexible systems, the transfer from, for example, state 1, is not predetermined (Fig. 10.3.b). There is a choice between many topologically nonequivalent pathways. The A ->• B reaction, bringing the system into a nonequilibrium state on the surface Ub{q), can obviously affect the corresponding probability distribution owing to a potential gradient at the transition point. The larger is the decrease in the free energy AGo during the reaction, the larger is the gradient (Fig. 10.13d), the more information is transmitted to particle C, and the higher is the probability of the direct C -» D transition.
Fig. 10.13 (a) Map of the levels of the potential energy for the pathway C —> D through the fluctuating gap (x°s is the equilibrium opening of the gap. (b) Map of the levels of the potential energy for the transition C -> D after the reaction A-»B leading to a new equilibrium conformation of the gap with opening XQ. Broken lines - relaxation pathways of the system after the act A—>B (see d); 1 and 2, windows in the region of the configuration space relating to the states of the particle C promoting (1) and deflecting (2) the transition C —^ D. (c) Reaction A —> B leading to opening of the gap. (d). A transfer from the surface of the conformational energy U a ( q ) to the surface U&(q) and subsequent relaxation; AEc(q) is the change in the conformational energy at the moment of A —> B transition at point q.
10.6
Conclusions
The ideas developed in this Chapter can be summarized as follows. A normal mode approach fails to describe the dynamic properties of molecules with conformational
306 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
mobility. The new approaches developed here are based on the analysis of the free energy maps and on the study of the correlation functions of dihedral angles. There are only a few typical types of the free energy maps for the aminoacids. If the map contains bottleneck structures or curved valleys, the dynamic correlation between the corresponding degrees of freedom takes place. On the other hand, there is no dynamic correlation for the motion along a uniform rough energy surface or in one narrow basin. Interestingly, there can be strong correlations between displacements of distant atoms in rather shot peptides. It is important that in peptides, the potential energy hypersurfaces apparently consist of a number of elements (like in a mosaic). This leads to the phenomenon of dynamic isomorphism. Each conformation of a biopolymer corresponds to a relatively narrow local minimum in its potential energy surface. Therefore, the systems with a large number of conformational degrees of freedom exhibit a complicated topology of the energy level hypersurfaces. The motion along the potential energy hypersurface represents a diffusion along a large number of topologically nonequivalent pathways. The uncertainty in the result of the conformational transition opens basically new physical possibilities for organizing the functional processes. It must be possible to organize a nontrivial transformation of energy with direct information exchange between subsystems involved in the chemical reaction but having no direct contacts. The change in the conformational energy hypersurface caused by a change in the chemical state of functional groups and the above factors unite at the molecular level such processes as mass transfer, transformation of energy, and information transmission. In the final analysis, this can be used as a basis for regulation and control in biological systems.
References B. Bagchi, D.W. Oxtoby (1983): "The effect of frequency dependent friction on isomerization dynamics in solutions". J.Chem. Phys. 78,2735-2741. N.K. Balabaev, A.S. Lemak, K.V. Shaitan (1996): "Molecular dynamics and electron conformational interactions in ferrodoxin". Molecular Biology 30, 812-817. C.L.Brooks, M. Karplus, B.M. Pettit (1988): "Proteins: A Theoretical Perspective of Dynamics, Structure, and Thermodynamics". Adv. Chem. Phys. (Eds. I. Prigogine & S.A. Rice. Wiley) 711-259. R. Czerminski, R. Elber (1989): "Reaction path study of conformational transitions and helix formation in a tetrapeptide". Proc. Natl. Acad. Sci. USA 86, 6963-6967. H. Frauenfelder, G.A. Petsko, D. Tsernoglou (1979): "Temperature-dependent Xray diffraction as a probe of protein structural dynamics". Nature 280, 558-563.
Conclusions
307
H. Prauenfelder, F. Parak, R.D. Young (1988): "Conformational substates in proteins". Ann. Rev. Biophys. Biophys. Chem. 17, 451-479. H.Frauenfelder, N.A.Alberding, A.Ansary et al. (1990): J.Phys. Chem. 94, 1024. V.I. Goldanskii, Yu.F. Krupyanskii, V.N. Flerov (1986): "Rayleigh scattering of Mossbauer radiation data, hydration effects and glass-like dynamical model of biopolymers". Physica Scripta. 33, 537-540. R.F. Grote, J.T. Hynes (1980): J.Chem. Phys. 73, 2715. E.E. Knapp, S.F. Fisher, F. Parak (1983): "The influence of protein dynamics on Mossbauer spectra". J. Chem. Phys. 78, 4701-4711. A.S. Lemak, N.K. Balabaev (1994): "On the Berendsen Thermostat". Molecular Simulation 13, 177-187. A.B Rubin, K.V. Shaitan, A.A. Kononenko, S.K. Chamorovskii (1989): "Temperature dependence of cytochrome photooxidation and conformational dynamics of Chromatium reaction center complexes". Photosynthesis 22, 219-231. K.V. Shaitan, A.B. Rubin (1989): "Conformational mobility and Mossbauer effect in biological system. Brownian motion damped oscillator model for conformation modes". Molecular biology (Transl. from Russian) 14, 1323-1335. K.V. Shaitan, I.V. Uporov, E.P. Lukashev, A.A. Kononenko, A.B. Rubin (1991): "Photo-conformational transition causes temperature and light effects during charge recombination in reaction centers of photosynthesizing bacteria". Molecular biology (Transl. from Russian) 25, 560- 569. K.V. Shaitan (1992): "Electron conformational transitions in proteins and physical mechanisms of biomacromolecular function" (Transl. from Russian). Molecular Biol. 26, 193-210. K.V. Shaitan (1994): "Conformational dynamics and new approaches to the physical mechanisms of elementary acts of mass transfer, energy transformation, and information transmission in biological macromolecular structures". Molecular Biol. (Transl. from Russian) 28, 444-449. K.V. Shaitan (1994): "Dynamics of electron-conformational transitions and new approaches to the physical mechanisms of functioning of biomacromolecules" (Transl.
308 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
from Russian). Biophysics 39, 993-1011. K.V. Shaitan (1996): "The topological structure of hypersurfaces of conformational energy levels and physical mechanisms of internal proteins mobility". Macromolecular Symp. 106, 321-335. K.V Shaitan, N.K. Balabaev, A.S. Lemak, M.D. Yermolaeva, A.G. Ivaikina, M.V. Orlov, Ye.V. Gelfand (1997): "Molecular dynamics of oligopeptides. 1. The use of long trajectories and high temperatures to determine the statistical weight of conformational substates" (Transl. from Russian) Biophysics 42, 45-51. K.V. Shaitan, A.V. Nemukhin, D.A. Firsov, T.V. Bogdan, LA. Topol (1997): "Importance of effective charges in the analysis of electron-conformational interactions in peptides" (Transl. from Russian) Molecular Biol. 3 1 , 108-117. K.V. Shaitan, M.D. Yermolaeva, N.K. Balabaev, A.S. Lemak, M.V. Orlov (1997): "Molecular dynamics of oligopeptides. 2. Correlations functions of the internal degrees of freedom of modified dipeptides". (Transl. from Russian) Biophysics 42, 547-555. K.V. Shaitan, M.D. Ermolaeva, S.S. Saraikin (1999): "Nonlinear dynamics of the molecular systems and the correlations of internal motions in the oligopeptides". Ferroelectrics 220, 205-220. K.V. Shaitan, M.D. Ermolaeva, S.S. Saraikin (1999): "Molecular dynamics of oligopeptides.3. Free Energy Maps and Dynamic Correlations in Modyfied Dipeptide Molecules". Biophysics 44, 14-17. K.V. Shaitan, P.P. Pustoshilov (1999): "Molecular Dynamics of a Stearic Acid Monolayer". Biophysics 44, 429-434. K.V. Shaitan, A.K. Vasil'ev, S.S. Saraikin, M.G. Mikhailyuk (1999): "Dynamic Properties, Electronic Structure and Functional Activity of Radioprotectors". Biophysics 44, 648-655. K.V Shaitan, A.B. Rubin (1989): "Conformational mobility and Mossbauer effect in biological system. Brownian motion damped oscillator model for conformation modes". Molecular biology 14, 1323-1335. K.V.Shaitan, A.B.Rubin (1983): "Bending fluctuations of alfa-helices and dynamics of enzyme-substrate interactions". Molecular biology 17, 1280. K.V Shaitan, I.V. Uporov, A.B. Rubin (1985): "To the theory of ligand migration
Conclusions
309
in bio-macromolecules". Molecular biology 19, 742-750. K.V. Shaitan, I.V. Uporov, E.P. Lukashev, A.A. Kononenko, A.B. Rubin (1991): "Photo-conformational transition causes temperature and light effects during charge recombination in reaction centers of photosynthesizing bacteria". Molecular biology (Transl. from Russian) 25, 560-569. K.V. Shaitan (1992): "Electron conformational transitions in proteins and physical mechanisms of biomacromolecular function". Molecular Biology (Transl. from Russian). 26, 193-210. K.V. Shaitan (1994): "Conformational dynamics and new approaches to the physical mechanisms of elementary acts of mass transfer, energy transformation, and information transmission in biological macromolecular structures. Molecular Biology (Transl. from Russian) 28, 444-449. K.V. Shaitan (1994):"Dynamics of electron-conformational transitions and new approaches to the physical mechanisms of functioning of biomacromolecules". Biophysics (Transl. from Russian) 39, 993-1011. K.V. Shaitan (1996): "The topological structure of hypersurfaces of conformational energy levels and physical mechanisms of internal proteins mobility". Macromolecular Symp. 106, 321-335. K.V. Shaitan, N.K. Balabaev, A.S. Lemak, M.D. Yermolaeva, A.G. Ivaikina, M.V. Orlov, Ye.V. Gelfand (1997): "Molecular dynamics of oligopeptides. 1. The use of long trajectories and high temperatures to determine the statistical weight of conformational substates". Biophysics. (Transl. from Russian), 42, 45-51. Shaitan K.V., Nemukhin A.V., Firsov D.A., Bogdan T.V., Topol LA. (1997): "Importance of effective charges in the analysis of electron-conformational interactions in peptides". Molecular Biol. (Translated from Russian). 3 1 , 108-117. Shaitan K.V., Ermolaeva M.D., Balabaev N.K., Lemak A.S., Orlov M.V. (1997): "Molecular dynamics of oligopeptides. 2. Correlations functions of the internal degrees of freedom of modified dipeptides". Biophysics. (Translated from Russian) 42, 547-555. Shaitan K.V., Ermolaeva M.D. Saraikin S.S. Nonlinear dynamics of the molecular systems and the correlations of internal motions in the oligopeptides. Ferroelectrics. 1999. V.220. pp.205-220.
310 Protein dynamics
and new approaches to the molecular mechanisms
of protein
functioning
Shaitan K.V., Ermolaeva M.D. Saraikin S.S. Molecular dynamics of oligopeptides.3. Free Energy Maps and Dynamic Correlations in Modyfied Dipeptide Molecules. Biophysics. (Translated from Russian). 1999. V.44. p.14-17. K.V. Shaitan, P.P. Pustoshilov (1999): "Molecular Dynamics of a Stearic Acid Monolayer". Biophysics (Transl. from Russian) 44, 429-434. K.V. Shaitan, A.K. Vasil'ev, S.S. Saraikin, M.G. Mikhailyuk (1999): "Dynamic Properties, Electronic Structure and Functional Activity of Radioprotectors". Biophysics (Transl. from Russian) 44, 648-655. K.V. Shaitan, A.Ya. Mukovskii, A.A. Beliakov, S.S. Saraikin (2000): "Dipeptids statistical distributions in protein structures and dynamic properties of some protein fragments". Biophysics 45, 399-406. K.V. Shaitan, S.S. Saraikin (2000): "The influence of fluctuation amplitude on factor of friction of the Brownian oscillator in water solutions". Biophysics 45, 407-413. N. Wiener (1961): "Cybernetics or Control and Communication in the Animal and the Machine", M.I.T. Press and John Wiley & Sons, Inc. New York-London, p. 342. R. Zwanzig (1988): "Diffusion in a rough potential". Proc. Natl. Acad. Sci. USA 85, 2029-2030. R. Zwanzig (1992): "Dynamical disorder: Passage through a fluctuating bottleneck". J.Chem.Phys. 97, 3587-3589.
Chapter 11
Conclusions
We studied in this book the mechanism of several physical processes in reacting complex molecules, in particular in biomolecules. This way we hope to contribute to the understanding of the dynamics and functioning of biological macromolecules. In particular this refers to enzymes, which are the basic molecular machines working in living systems. Since biological macromolecules operate on many thousands of degrees of freedom we concentrated on the study of simple model systems, as e.g. the dynamics of clusters consisting of a smaller number of atomic units, the dynamics of conformations and of transitions between conformations. In this context we analysed physical mechanisms as the transitions beween two potential wells, the nonlinear coupling between oscillatory modes, Fermi resonance, excitation of solitons in chains of nonlinear springs and the effects of coloured noise. The analysis of these complex processes was based on methods of nonlinear dynamics, stochastics and molecular dynamics. We did not intend to present a complete pattern of specific enzymatic reactions or to figure out the ways of their effective control. Instead we concentrated on the most important aspects, as e.g.the role of nonlinear excitations and of stochastic effects. We tried to give a survey of the state of art in the field and to present several original results obtained in the last years by our groups. Let us underline that this book poses problems of the molecular dynamics of biomolecules rather than providing their comprehensive solutions. Theoretical considerations precede experimental studies. The ideal experiments must use individual molecules, whereas, the conventional spectroscopic techniques study the molecular ensembles and the corresponding results imply averaging over a large numberof molecules in slightly different conformational states. Note that "ideal" experiments may use synthetic homogeneous polymers with incorporated fragments exhibiting special optical properties. As for the selected degrees of freedom, the studies in this field can be continued with the use of the computer experiments. We must proceed here from the simple "cluster" models to more elaborate models exhibiting complicated motions in hundreds degrees of freedom. Such an approach necessitates application of advanced software and strong computational power. 311
312
Conclusions
Let us consider now several more specific results obtained in this book: In Chapter 2 we considered several specific features of 2-d test particle transitions between two potential wells. Our approach goes beyond transition state theory and Kramers theory. In Chapter 3 we studied deviations from the Arrhenius behaviour mainly by molecular dynamics simulations. The main results obtained may be summarized as follows: The reactive transitions are not overdamped, they show strong deviations from a strict Arrhenius relation between transition time and temperature, the noise acting on a reacting site is not white but has the character of a coloured noise. In Chapter 5 we concentrated on the role of hard (soliton-like) excitations. Based on the model of Toda rings we have shown that hard excitations may lead to energy spots at the active site which lead to strong enhancements of reactive transitions. Further we studied on a simple model the influence of entropic effects. In Chapter 6 (as well as in Chapter 1) we studied the effect on Fermi resonances on transitions, showing that this resonance is another candidate for the enhancement of reactions. In Chapter 6 we considered also the diffusion limitation of the enzymatic reaction rate. Both substrate and product were represented by rather simple molecules. However, the reaction of peptide bond breaking may involve rather long peptides or their fragments. In Chapter 1 we explained the interaction of chymotrypsin with a polypeptide chain. CT active site effectively binds aromatic amino acids (e.g. phenylalanine) whereas small residues easily get through the AS pocket. What is the mechanism of this binding? We may suggest a mechanical analogy of a thread drawn through a needle eye? This problem can be considered based on the PDB data and the theory developed recently for estimating the rate of a polymer thread drawing through a hole in a membrane. This is related, in particular, to drawing of polymer substrates through the pores of the artificial vesicles containing CT molecules. In several Chapters of the book and in particular in Chapter 10, we considered the transfer of macromolecules from one conformational state into another. It is expedient to extend the consideration to the problem of ion transport through membranes of living cells taking into account the molecular machines incorporated into these membranes. ATPase represents a remarkable example of a molecular machine transforming electric energy into mechanic or chemical (and vice versa). Recent experimental investigations shed light on the physical principles of operation of this "engine". A comprehensive physical model of this molecule could be a significant step forward in interpreting the principles of operation of the molecular machines in general. We are sure that the understanding of regularities of the nonlinear Brownian motion in complex nonstationary force fields can substantially contribute to the development of such models. In the near future, the natural and artificial molecular machines will find new applications in biotechnology. We expect e.g. the advent of the "photosynthetic batteries", further molecular machines will be used in computers. But even at
313
present, the understanding of the fundamentals of molecular machines can help in technologies employing enzymes, in particular, aimed at developing new medicals. To conclude, this book conveys some of the spirit of the physical approach to the enormeous complexity of the functioning of reacting biomolecules. We hope that this approach is useful for the understanding of real biomolecules and quote finally a sentence taken from Eigen's foreword to Volkensteins remarkable book on the physical approach to biological evolution.
"Many biologists do not believe that their subject lends itself to the scrutinity of physical theory. They certainly admit that that one can simulate biological phenomena by models that can be expressed in a mathematical form. However they do not believe that biology can be given a theoretical foundation that is defined within the general framework of physics. Rather, they insist on a holistic approach, banning any reduction to fundamental principles subject to physical theory. This is a misconception, not of biology, but of physical theory. The aim of theory is not to describe reality in every detail, but rather to understand the principles that shape reality". References M.V. Volkenstein (1994):"Physical Approaches to Biological Evolution. With a Foreword by Manfred Eigen", Springer, Berlin-Heidelberg
This page is intentionally left blank
List of authors
Alexander P. Chetverikov, Faculty of Nonlinear Dynamics, Saratov State University, Astrahanskaya 83, 410026 Saratov, Russia; e-mail: [email protected] Olga A. Chichigina, Faculty of Physics, M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] Andrey Yu. Chikishev, International Laser Center M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] Werner Ebeling, Humboldt University at Berlin, Institute of Physics, Invalidenstr. 110 and Charite, Hessische Str. 2, D-10115 Berlin, Germany; e-mail: [email protected] Boris A. Grishanin, Faculty of Physics, M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] Martin Jenssen, Institute of Forest Research, D-16225 Eberswalde, Germany e-mail: [email protected] Stanislav V. Kroo, Faculty of Physics, M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] Alexey V. Netrebko, Institute of Mechanics, M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] 315
316
List of authors
Nina V. Netrebko, Faculty of Physics, M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] Viktor Yu. Podlipchuk, Institute of Physics, Humboldt-University at Berlin, Invalidenstr. 110, D-10115 Berlin, Germany; e-mail: [email protected] Yury M. Romanovsky, Faculty of Physics, M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] Mikhail G. Sapeshinsky, Faculty of Fundamental Science, Bauman Moscow State Technical University, 2-ja Baumanskaya, 107005, Moscow, Russia; e-mail: [email protected] Lutz Schimansky-Geier, Humboldt University at Berlin, Institute of Physics, Invalidenstr. 110, D-10115 Berlin, Germany; e-mail: [email protected] Konstantin V. Shaitan Faculty of Biology, M.V.Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail: [email protected] Ekaterina V. Shuvalova, Faculty of Physics, M.V. Lomonosov Moscow State University, Vorobyevy gory, 119992, Moscow, Russia; e-mail:[email protected] Peter Talkner, Institute of Physics, University of Basel, Klingenbergstr. 82, CH4056 Basel, Switzerland, e-mail: [email protected] Alexander A. Valuev, Department of General Physics, Moscow Institute of Physics and Technology, Pervomayskaya 9, 141700 Dolgoprudny, Moscow Region; e-mail: [email protected]
Index
E Einstein relation 13, 41, 61, 149 electrostatic field 210 energy accumulation 153ff energy beating 25 energy diffusion 2, 4, 88, 122 energy localization 148 entropy effects 41, 175ff enzymatic reactions 166 enzyme model 11, 26, 166 enzyme substrate complex 8, 182 enzyme substrate interaction XII ester bond 8, 219, 229 ethan 11, 266 evolution operator 72, 247
A Arrhenius law ix, 1, 5, 81, 122 acetylcholin esterase (ACE) xi, 8, 26 acetylcholin xi, 8, 209 activation processes ix, 122, 129, 131, 136 active site (AS) 28, 154, 181 autocorrelation function 49, 135 B barrier crossing 16 bistable potential X, 2, 3, 23, 78, 127 Brownian motion 2, 37ff Brownian particle 2, 37ff C canonical distribution, 38ff chymotrypsin (CT) 28, 29 conformational energy levels, 287ff conformation of proteins 285ff correlation function 139 conformation 26 cluster 26 cluster models 26ff cluster dynamics X, 26ff cnoidal waves 168 collision force 129, 133 collision spectrum 133 conformation of proteins xiii coulored noise X, 54, 124
F Fermi resonance XI, 17ff, 25, 181 flicker noise 155 fluctuations 38ff Fokker Planck equation, VIII, 80 force distribution 151 friction constant 5, 263, 280 free energy 2, 11, 45, 287 H hydrogen bond (H-bond) 8, 15 hard particles X, 130 harmonic noise 56, 58, 124, 137 heat bath 2, 25, 121 heat, specific 145, 149, 151 hydrogen bond 28, 255 hydrolysis 10
D damping of oscillation XII, dipol moment 209, 216 dissociation of molecules 130 dynamic structure factor 169, 170
K kinetics of enzymatic reactions 230 317
318 Kramers rate 4, 85ff Kramers theory VIII, 85ff, 122 Kramers-Moyal moments 75ff L Langevin equation 4, 11, 18, 65ff, 143 Lennard-Jones potential 13, 14, 127, 128 linear response 5 Iff M Markovian process, 37ff mean first passage time 95ff mean square fluctuation 153 Michaels-Menten equation 230 molecular dynamics 12, 126, 130, 269 molecular machine XIII, 26, 209 Morse potential XI, 14, 128 Morse springs 176 molecular scissors 8 N Nyquist Theorem 55ff non-Kramers kinetics 136 O oscillating potential landscape 120 P peptide bond 8, 9, 17 Perrins pendulum 43 Pippard oscillator 21 Pippard potential 20, 23 polypeptide 29 potential landscape 103 protein 11, v26 protein dynamics 11 protein machine 26 proton transfer 29, 247, 255 Q Q-factor 267 quantum tunneling 2, 5, 247 R Raman effect 17 Raman spectra 264 rate enhancement 172, 178 recombination 131 reaction rate 1, 10, 81ff red noise 53 resonances 18ff
Index S Schrodinger equation XII, 247, 260 Sinai billard X, 114 soft particles X, 130, 274 soliton 146ff solitonic exitation 141ff spectral density 124, 133, 135, 152 Stokes-Lamb theory 263ff subglobules 29, 273 subglobul oscillation 263 substrate 29 substrate inhibition 241 stochastic transitions 2, 85ff, 122 structure factor 157ff, 162 synapse 209 T test particles 17 three-minima potential 104, 120 Toda lattices XI, 141ff, 150 Toda method 106, 193 Toda rings 141ff Toda spring 146ff Toda potential 14, 143 transition rates 16, 89, 122 transition time 5, 25, 123 W wave function 247 white noise 4, 48
Stochastic Dynamics of
Reacting Biomolecules This is a book about the physical processes in reacting complex molecules, particularly biomolecules. In the past decade scientists from different fields such as medicine, biology, chemistry and physics have collected a huge amount of data about the structure, dynamics and functioning of biomolecules. Great progress has been achieved in exploring the structure of complex molecules. However, there is still a lack of understanding of the dynamics and functioning of biological macromolecules. In particular this refers to enzymes, which are the basic molecular machines working in living systems. This book contributes to the exploration of the physical mechanisms of these processes, focusing on critical aspects such as the role of nonlinear excitations and of stochastic effects. An extensive range of original results has been obtained in the last few years by the authors, and these results are presented together with a comprehensive survey of the state of the art in the field.
ISBN 981-238-162-7
World Scientific www. worldscientific.com 5099 he