Introduction to:

Modelling Molecular
Interactions and Dynamics

Bioinformatics II

M. Meuwly
Department of Chemistry
University of Basel
1 Introduction

Experimental Theoretical
techniques methods
Xray, NMR Development of
(dynamics and structure) mathematical
models v(r)

Light/Xray/neutron scattering
(dynamics and structure) Development of
Development of new methods to
theories and models explore models
to rationalize and
Imaging/CryoEM predict experimental
(dynamics and structure) observations
Exploration of model
phenomenology and
Calorimetry, pKas, properties
thermodynamics,
physical measurements

Understanding biomolecular
structure, dynamics and function

1
2 Force field

A force field is an empirical approximation for expressing structure-

energy relationships in biopolymers
It is a compromise between speed and accuracy
Common form (CHARMM):

X1
E(r1 , r2 , . . . , rN ) = kib (di d0i )2
2
bonds
X 1
+ ki (i i0 )2
2
angles
X
+ ki [1 + cos(ni i i )]
torsions
!12 !6
1 X min dmin
ij dmin
ij + qi qj
+ ij 2
2 dij dij dij
nonbond

2
2.1 Energy terms

From specctroscopy, IR, NMR

vbond = 12 kib(di d0i )2

Energy
Bonds

Bond

vangle = 12 ki (i i0 )2

Energy

Angles

Angle

vdihedral = ki[1 + cos(3i)] +

Energy
Dihedrals

ki [1 cos(i )]
Dihedral angle

From quantum chemistry, thermodynamics

 12  6 
dmin dmin
Energy

Van der Waals vvdW = min

ij
ij
dij 2 ij
dij

Distance
Energy

qi qj
Electrostatics vCoulomb = dij
+
Distance
3
2.2 Parametrization

4
6 Molecular dynamics
6.1 Basics

Atomic positions (coordinate file)

Covalent structure (topology file)
Effective temperature
Potential energy function (parameter file)
(through kinetic energy)
Additional atoms (solvent, counterions)
Forces on each atom
Special features (PBC, constant T and/or P)
Atomic velocities

13
6.2 Equations

Fi =mi ai Fi = gradi E E = Ebonding + Enonbonding

1. solve for ai at t dE
dri = Fi = mi ai (t)

2. update vi at t + t/2 vi (t + t/2) = vi (t t/2) + ai (t)t

3. update ri at t + t ri (t + t) = ri (t) + vi (t + t/2)t
4. go to 1.

Timestep controls accuracy of numerical solution.

Fundamental timestep is determined by high frequency vibrations (co-
valent bonds t = 1015 sec).
Highest frequency motions, i.e., hydrogen atom vibrations, can be re-
moved with holonomic constraints.

14
6.3 Thermodynamic variables T and P

Statistical ensembles connect microscopic to macroscopic

Microcanonical (NVE, entropy)

Canonical (NVT, Helmhotz free-energy)

P
2
T = m v /(3kb )

Isothermal-isobaric (NPT, Gibbs free-energy)

P = kinetic + virial contributions

Thermostats, barostats allow to choose the appropriate ensemble.

Andersen, Nose, Hoover.

15
4 Sampling techniques
4.1 Energy minimization

Potential Energy

Reaction coordinate

Minimization follows gradient

Reaches the nearest local minimum
Steepest descent, conjugate gradient
8
 4.2 Metropolis Monte Carlo (Boltzmann statistics)

Metropolis Monte Carlo yields an ensemble (Boltzmann statistics).

Ergodicity: every accessible point in configuration space should be reached in a finite
number of Monte Carlo steps from any other point.
Kinetics are usually not meaningful.

9
4.3 Simulated annealing (good for sampling but no ensemble)

High Temperature

Cooling
Potential Energy

Reaction coordinate

10
 Cycle
300 330 360 390
4.4 Parallel tempering (equilibrium Monte
0 1 2 3 I
Carlo scheme)
M non-interacting copies of the system at different Tm 1 II
0 2 3

A state is defined by
1 2 0 3 III
T1 , T2 ,..., TM
z
 }| {
1 M
X = xm(1) , . . . , xm(M ) , xim (q i , pi )m 1 2 3 0 IV

1 3 2 0 V

In order to converge toward equilibrium the detailed

balance should be satisfied. Therefore: T(K)

300

320

340

360

380

400

420

440

460

480
1, 0,

0

w(X X ) =
exp(), > 0.

200


where [n m ] E(xim ) E(xjn ) .

400
frame
High T replicas jump from basin to basin (inter-basin)

600
Low T replicas explore a single valley (intra-basin)

800
Rao and Caflisch, J. Chem. Phys. 119, 4035, 2003 1000

11
8 Free-energy barriers and timescales

barrier crossing

Free Energy

Reaction coordinate

To cross a free-energy barrier = 0 exp(G /kB T ) with 0 1012 s:

1 kcal/mol : ps, 5 kcal/mol : ns, 10 kcal/mol : s or longer

Sampling should exceed timescales of interest by 10-fold.

System size and complexity increase required timescales (equilibration
of ions, complex landscapes, multiple minima)
17
9 Approximations in molecular dynamics

Approximations inherent to the force field (E)

Systematic error:
Calculations of free energy differences is still very difficult.
Time scale and sampling problem
Statistical error:
Conformational transitions that require > 0.1 1s cannot be
simulated (yet) by conventional molecular dynamics techniques.

Other simulation approaches:

MD with implicit solvent (approximate)
Brownian dynamics
Monte Carlo (move definitions are difficult for macromolecules)

18
Differences between Force Fields
Differences between Force Fields
Differences between Force Fields
Practical Considerations for Calculating Energies

Force Field Ab initio

System Size Several 10000 atoms 20 heavy atoms (correl.)
1000 atoms (HF, [DFT])

Application Structures Structures

Conformational Search Energetics
Non-covalent interactions Reactions

Limitations Bond-breaking Very time consuming

Fixed atomic charges Dynamics often impossible
[Quantitative Information]
 Free energy: classical definition

The free energy is the energy left for once you paid the tax to entropy:

!G = !H " T!S
Enthalpic Entropic

! Hydrogen bonds ! Loss of degrees of freedom

! Polar interactions ! Gain of vibrational modes
! Van der Waals interactions ! Loss of solvent/protein structure
! ... ! ...
Theoretical Predictions: ! Approximate: empirical formula for all contributions
! Exact: using statistical physics definition of G

G = -KBT ln(Z)
 Free energy: statistical mechanics definition

G = !k BT ln(Z ) where Z = # i e! " Ei

is the partition function

Free energy differences between 2 states (bound/unbound, )

are, therefore, ratios of partition functions

# ZA &
!G = GA " GB = "k BT ln % (
$Z ' B

Free energy simulations aim at computing this ratio using various

techniques.
 Relation with chemical equilibrium

A+B # AB KA = Kb =
[ A'B']
" [ A ] [ B]
[ A ] [ B]
AB #
" A+B KD = Ki =
[ A'B']
Kb : binding constant, Kd : dissociation constant, Ki : inhibition constant

!G binding = "RTln K A = RTln K D = !H " T!S

"Gbinding (kcal/mol) -2 -4 -6 -8 -10 -12 -14 -16

Weak asso. Strong asso.

KD (mol/l) 10 -3 10 -6 10 -9 10 -12
Connection micro/macroscopic:
thermodynamics and kinetics
Absolute binding
free energies: !G
" KA
e - RT!G = KA
Relative binding
Free Energy Association Constant free energies: !!G
" KA / KA

Binding free energy

profiles: !G(#)
" KA, Kon, Koff
Microscopic Structure Biological function
The free energy is the main function behind all process
KA =
[ AB]
A) Chemical equilibrium [ A ] [ B]
+
!G binding = RTln K A A B KD = 1 / KA AB

B) Conformational changes

PConf 1
!G conf = k B Tln R = kB N A
PConf 2

C) Ligand binding

PUnbound
!G binding = k B Tln
PBound

D)
 Free energy: computational approaches

# ZA &
!G = GA " GB = "k BT ln % (
$Z ' B

Free energy simulations techniques aim at computing ratios of

partition functions using various techniques.

Z = # i e! " Ei

Sampling of important Computation of energy

microstates of the system of each microstate
(MD, MC, GA, ) (force fields, QM, CP)
 Connection micro/macroscopic: intuitive view

E1, P1 ~ e-$E1

Expectation value E2, P2 ~ e-$E2

1
O = #
Z i
Oie! "Ei E3, P3 ~ e-$E3

Where Z = e # ! "Ei
i
is the partition function E4, P4 ~ e-$E4

E5, P5 ~ e-$E5
 Central Role of the Partition Function

Z = # i e! " Ei

1
O = #
Z i
Oi e ! " Ei ...

Expectation Value

! " ! ln(Z ) %
ln(Z) = U p = k BT $
# !V '& N ,T
E = G = -kBT ln(Z)
!"
Internal Energy Pressure Gibbs free energy
 Binding free energy decomposition: MM-PBSA, MM-GBSA

Averaged over an MD simulation trajectory

!E gaz of the complex (and isolated parts)
Gaz Lig + Prot Lig:Prot
lig prot comp
!Gbind = !E gaz + !Gdesolv " T !S
!G solv !G solv !G solv

E gaz = E elec + E vdw + !E int ra

Sol Lig + Prot Lig:Prot comp
!Gdesolv = !Gsolv " (!Gsolv
lig prot
+ !Gsolv )
!Gbind
!T"S = !T(S comp ! (S prot + S lig ))
S = Strans + Srot + Svib B. Tidor and M. Karplus, J. Mol. Biol., 1994, 238, 405

!Gsolv = !Gsolv,elec + !Gsolv,np

comp
!Gdesolv = !Gsolv,elec " (!Gsolv,elec
lig prot
+ !Gsolv,elec (
) + # SASA comp " (SASA lig + SASA prot ) )
Depending on the way !Gsolv,elec is calculated:
Molecular mechanics Poisson-Boltzmann Surface Area (MM- PBSA)
J. Srinivasan, P.A. Kollmann et al., J. Am. Chem. Soc., 1998, 120, 9401

Molecular mechanics Generalized Born Surface Area (MM- GBSA)

H. Gohlke, C. Kiel and D.A. Case, J. Mol. Biol., 2003, 330, 891
Summary

Structural Optimization
Energy minimization
Normal Mode

Force Field Molecular Dynamics

Macroscopic Properties

Monte Carlo
Example from Research: CO motion in myoglobin
Experiments

F. Schotte et al., Science 300, 1944 (2003)

Ligand Dynamics in Mb
Simulation techniques:

Classical Molecular Dynamics

Conventional Force Field for Protein (CHARMM)

Use realistic model for CO (CO is neutral, has small dipole

but large quadrupole moment)

Use explicit solvation with water

Extended simulation times (several ns)

Analysis via Fourier Transformation of dipole autocorrelation function

Ligand Dynamics in Mb
IR spectrum for dissociated CO in native Mb

Experiment Simulations

M. Lim et al.,
J. Chem. Phys., 102 4355 (1995)
Ligand Dynamics in Mb Protocol: 50 ps of MD simulations
CO treated at B3LYP/6-31G**
Stochastic boundary conditions
Simulations

D. R. Nutt and M. Meuwly,

PNAS , 101, 5998 (2004)
Ligand Dynamics in Mb
IR spectrum for dissociated CO in native Mb

Experiment Simulations

M. Lim et al.,
J. Chem. Phys., 102 4355 (1995)
Advanced Topic: Ligand Rebinding in Mb

Possible Reaction Pathways in MbCO

A, B, Xe4
Ostermann et al., Nature, 404, 205 (2000)

A, B, Xe4, Xe1, B
Frauenfelder et al., PNAS, 98, 2370 (2001)
Nienhaus et al., Biochem., 42, 9647 (2003)

A, B, Xe4, Xe3, Xe1

Bossa et al., Biophys. J., 87, 1537 (2004)

A, B, Xe1
Srajer et al., Biochem, 40, 13802 (2001)

A, B, C competing with A, B, S
Scott+Gibson, Biochem., 36, 11909 (1997)
Advanced Topic: Ligand Rebinding in Mb
Simulation techniques:

Classical Molecular Dynamics

Conventional Force Fields for initial and final state

Detect Crossing via energy criterion

Carry out crossing using a mixing algorithm

Use explicit solvation with water

Extended simulation times (several ns)

Ligand Rebinding in Mb

Explicit rebinding dynamics in MbNO

B Method: A
B
Energy

Propagate on surface B
Locating of crossing
(EA = EB)
Backpropagation
A Surface crossing to
surface A A
R(Fe-X)
Ligand Rebinding in Mb
Explicit rebinding dynamics in MbNO

Extended recrossing region Rebinding is nonexponential in time

Rebinding into secondary Time constants: 1 = 3.8 ps (5.5 to 28 ps)
minimum 2 = 18.0 ps (50 to 280 ps)
Ligand Rebinding in Mb

Rebinding dynamics in MbCO on free energy curves

Precalculate the free energy curve for bound and

unbound CO motion using umbrella sampling.
Solve the Smoluchowski equation for G(q).

Data derived from experiment

J. S. Olson and
G. N. Phillips
JBC, 271 17593 (1996)
Ligand Rebinding in Mb

Rebinding dynamics in MbCO on free energy curves

unbound

Energy
bound
R(Fe-X)
G(kcal/mol)

B Xe4

R(Fe-CM)
Ligand Rebinding in Mb

Rebinding dynamics in MbCO on free energy curves

From Xe4

(kcal/mol) ns
4.0 100
5.0 280
7.5 1770

For native MbCO: inner barrier 4.3 kcal/mol vs 4.5 kcal/mol from experiment
For L29F mutant: rebinding time 10 ps vs rapid escape to Xe4 pocket