Protein Dynamics, Allostery and Function
Lecture 1. Molecular Forces
Xiaolin Cheng UT/ORNL Center for Molecular Biophysics
SJTU Summer School 2017 1 Lecture 1, SJTU Summer School 2017 2 Lecture 1, SJTU Summer School 2017 3 Ribosome: an machine to produce proteins.
Mitochondrion: a factory to produce energy.
Lecture 1, SJTU Summer School 2017 4 Structures: protein
Lecture 1, SJTU Summer School 2017 5 Protein Structure
Lecture 1, SJTU Summer School 2017 6 Lecture 1, SJTU Summer School 2017 7 Protein Folding
Anfinsen’s dogma: The na ve structure corresponds to the state with the lowest free energy of the protein-solvent system.
But how? Lecture 1, SJTU Summer School 2017 8 Levinthal Paradox
Suppose (i) each amino acid has 3 conforma ons (ii) a protein consists of 100 amino acid residues, with a total conforma on number 3100 ≈ 5 x 1047 (iii) 100 psec (10-10 sec) are required for a conforma onal change
Then A random search of all conforma ons would require around 1030 years. Nevertheless, folding of a real protein takes place in msec to sec order.
So Protein folding cannot be via a random search.
Lecture 1, SJTU Summer School 2017 9 Funnel Theory
For a mul dimensional smooth landscape like (a), to find its minimum is simple. A real landscape is much more complex, with mul ple local minima (folding traps) like (c).
Lecture 1, SJTU Summer School 2017 10 Funnel Theory
enthalpy-entropy compensa on Gain of binding energy pulls pep de down; reduc on of entropy keeps pep de up.
Lecture 1, SJTU Summer School 2017 11 Model for Protein Folding
Lecture 1, SJTU Summer School 2017 12 Forces on Proteins
Lecture 1, SJTU Summer School 2017 13 Force Scales
room ATP covalent glucose temperature hydrolysis bond oxidation
secondary bond
vdw hydrophobic
HB
salt bridge covalent
Lecture 1, SJTU Summer School 2017 14 Bonded Forces
Lecture 1, SJTU Summer School 2017 15 Non-Bonded Forces
A repulsive component resul ng from the Pauli exclusion principle that prevents the collapse of molecules. A rac ve or repulsive electrosta c interac ons between permanent charges (in the case of molecular ions), dipoles (in the case of molecules without inversion center), quadrupoles (all molecules with symmetry lower than cubic), and in general between permanent mul poles. Induc on (also known as polariza on), which is the a rac ve interac on between a permanent mul pole on one molecule with an induced mul pole on another. This interac on is some mes called Debye force. Dispersion (usually named a er Fritz London), which is the a rac ve interac on between any pair of molecules, including non-polar atoms, arising from the interac ons of instantaneous mul poles.
Lecture 1, SJTU Summer School 2017 16 Force Field Functions
Lecture 1, SJTU Summer School 2017 17 Coulomb Interaction charge-charge interac ons
N2 complexity – requires fast algorithms, e.g., PME, FMM, Mul -grid
Entropic effect is significant
The dielectric constant of water is strongly temperature dependent, and decreases by 0.46% per Kelvin near room temperature.
TS = -1.38G
Lecture 1, SJTU Summer School 2017 18 Coulomb Interaction
Charge-dipole interac ons Induced dipoles
Dipole-dipole interac ons Induced dipoles
Lecture 1, SJTU Summer School 2017 19 Cation–π Interactions
Lecture 1, SJTU Summer School 2017 20 Cation–π Interactions
+ + + + + + + M Li Na K NH4 Rb NMe4 -ΔG [kcal/ 38 27 19 19 16 9 mol] ri o n [Å] 0.76 1.02 1.38 1.43 1.52 2.45
Lecture 1, SJTU Summer School 2017 21 Cation–π Interactions
Lecture 1, SJTU Summer School 2017 22 Cation–π Interactions
Lecture 1, SJTU Summer School 2017 23 Cation–π Interactions
Lecture 1, SJTU Summer School 2017 24 π Stacking
The benzene dimer is experimentally bound by 8–12 kJ/mol (2–3 kcal/mol) in the gas phase with a separation of 4.96 Å between the centers of mass for the T-shaped dimer.
Lecture 1, SJTU Summer School 2017 25 π Stacking
intermolecular overlapping of p-orbitals in π-conjugated systems, so they become stronger as the number of π- electrons increases.
Lecture 1, SJTU Summer School 2017 26 Hydrogen Bond
a rac ve interac on of a hydrogen atom with an electronega ve atom
occur between molecules (intermolecularly), or within different parts of a single molecule (intramolecularly)
5 to 30 kJ/mol - stronger than a van der Waals interac on, but weaker than covalent or ionic bonds. electrosta c dipole-dipole interac on, but has some features of covalent bonding: direc onal and strong, produces interatomic distances shorter than sum of van der Waals radii, and usually involves a limited number of interac on partners. Lecture 1, SJTU Summer School 2017 27 Hydrogen Bond
Lecture 1, SJTU Summer School 2017 28 Van der Waals Force
London dispersion forces - a rac ons between atoms, molecules, and surfaces.
caused by correla ons in the fluctua ng polariza ons of nearby par cles (a consequence of quantum dynamics).
The strength of London dispersion forces is propor onal to the polarizability of the molecule, which in turn depends on the total number of electrons and the area over which they are spread.
Lecture 1, SJTU Summer School 2017 29 Van der Waals Force
Whereas the func onal form of the a rac ve term has a clear physical jus fica on, the repulsive term has no theore cal jus fica on. It is used because it approximates the Pauli repulsion well, and is more convenient due to the rela ve computa onal efficiency of calcula ng r12 as the square of r6. Lecture 1, SJTU Summer School 2017 30 Van der Waals Force
a rela vely good approxima on and due to its simplicity is o en used to describe the proper es of gases, and to model dispersion and overlap interac ons in molecular models.
Lecture 1, SJTU Summer School 2017 31 Non-Bonded Forces
Other noncovalent interac ons include hydrogen bonds, van der Waals forces, charge-transfer interac ons, and dipole-dipole interac ons. These forces exist between molecules when they are sufficiently close to each other. The forces consist of four types: Dipole–dipole forces, ion–dipole forces, dipole-induced dipole force or Debye forces, instantaneous dipole-induced dipole forces or London dispersion forces.
Bond type Dissocia on energy (kcal) Covalent 400 Hydrogen bonds 12-16 Dipole–dipole 0.5 - 2 London (van der Waals) Forces <1
Lecture 1, SJTU Summer School 2017 32 Hydrophobic Effect
a phenomenological (entropic) force resul ng from the en re system's sta s cal tendency to increase its entropy, rather than from a par cular underlying microscopic force.
The hydrophobic effect represents the tendency of water to exclude non-polar molecules.
The effect originates from the disrup on of highly dynamic hydrogen bonds between molecules of liquid water by the nonpolar solute.
The hydrogen bonds are par ally reconstructed by building a water "cage" around the hexane molecule. The water molecules that form the "cage" (or solva on shell) have substan ally restricted mobility.
Lecture 1, SJTU Summer School 2017 33 Hydrophobic Effect
The hydrophobic effect is responsible for effects related to biology, including: cell membranes and vesicles forma on, protein folding, inser on of membrane proteins into the nonpolar lipid environment and protein-small molecule associa ons.
Lecture 1, SJTU Summer School 2017 34 Hydrophobic Effect
can be quan fied by measuring the par on coefficients between water and non-polar solvents: ΔG = ΔH - TΔS. These components are experimentally determined by calorimetry.
As a result of an entropy-enthalpy compensa on, the hydrophobic effect (as measured by the free energy of transfer) is only weakly temperature-dependent and became smaller at the lower temperature, which leads to "cold denatura on" of proteins.
Lecture 1, SJTU Summer School 2017 35 Electrostatic Interaction
Electrosta c or Coulomb poten al describes the interac ons between pairs of par al charges.
This formula indicates that the electrosta c interac on of two charges of 1e separated by the distance of 7 Å is about 50 kcal/mol.
Because VEL decays as 1/r, the electrosta c interac ons are considered as long-ranged. It is important to devise methods reducing their N2 scaling.
Lecture 1, SJTU Summer School 2017 36 Time Scale Challenge
V K (b b )2 K ( )2 = ∑ b − 0 + ∑ θ θ −θ0 bond angle K (1 cos(n )) + ∑ φ + φ −φ0 1 µs = 1 billion steps dihe 1 week of wall clock time q ⋅q + ∑∑ i j = 604, 800 seconds i j>i rij à 0.6048 millisecond per step 12 6 ⎡⎛σ ⎞ ⎛σ ⎞ ⎤ 4 ⎢⎜ ij ⎟ ⎜ ij ⎟ ⎥ + ∑∑ ε ij − i j i ⎢⎜ r ⎟ ⎜ r ⎟ ⎥ > ⎣⎝ ij ⎠ ⎝ ij ⎠ ⎦
1. N-body problem O(N2)àO(NlogN), O(N) e.g. PME, FMM Faster force evalua on 2. Parallelization
Enhanced sampling techniques increase the barrier crossing rate
Lecture 1, SJTU Summer School 2017 37 Particle Mesh Ewald
1 N ' q q E(r ,...,r ) i j 1 N = ∑∑ 2 i, j=1 n | rj − ri + nL |
PME reduces the complexity in treating the Coulomb interaction to NlogN.
Lecture 1, SJTU Summer School 2017 38 Particle Mesh Ewald
43,222 atom system using a 128 × 128 × 128
Jaguar XT5 benchmark, 2010 Fitch BG et al. SC2006
For biomolecular simula ons, the messages to be sent and their associated latency dominate the FFT computa on me.
Lecture 1, SJTU Summer School 2017 39 Particle Mesh Ewald
number of nodes 1 M Atom Virus on Titan GPU Lecture 1, SJTU Summer School 2017 40 Fast Multipole Method
FMM reduces the complexity in treating the Coulomb interaction to N.
FMM vs. PME (par cle-mesh Ewald) - Linear me complexity - Allows non-periodic or periodic boundaries - Avoids 3D FFTs for be er parallel scaling - Spa al separa on allows use of mul ple me stepping - Can be extended to other types of pairwise interac ons (e.g., vdW)
Lecture 1, SJTU Summer School 2017 41 Fast Multipole Method
Yokota et al., Computer physics communications, 2011,1272
Strong scaling performance of an FMM based electrostatic model.
Nagasaki Advanced Computing Center: 128 nodes, each with 4 GPUs
20 million atoms and one billion unknowns, required only one minute on 512 GPUs Lecture 1, SJTU Summer School 2017 42 Fast Multipole Method
“achieved 4.4 PFLOPS in a 520 million-atom simulation with FMM”
Lecture 1, SJTU Summer School 2017 43 Implicit Solvation Models
Explicit Solvent Average Density Implicit Solvent http://feig.bch.msu.edu/main-research-methodology.html
It is possible to construct an implicit solvent model by approximating the medium outside the water-excluded volume as a continuum with electrostatic, entropic, and viscous properties that match water.
Lecture 1, SJTU Summer School 2017 44 Implicit Solvation Models
The solvation free energy of the solute in conformation X:
ΔG(X ) = ΔGnp (X ) + ΔGelec (X ) where ΔGnp is the change of solvation free energy in going from nothing to the non- polar solute and ΔGelec is the change of free energy in going from the non-polar (uncharged) form of the solute to the polar (charged) form of the solute.
Lecture 1, SJTU Summer School 2017 45 Polar Solvation
Consider the free energy of placing a charge q at the center of a spherical cavity in a solvent.
1 ⎛ 1 ⎞ q2 Consider the solvent to be a uniform ΔGelec = − ⎜1− ⎟ 2 R dielectric medium with dielectric constant . -⎝ ε ⎠ ion ε - - The dielectric is polarized by the charge at the center of the spherical cavity and will - +q - produce a reaction field that opposes the - - electric field produced by the solute charge. -
1 ⎛ 1 ⎞ q2 ΔGelec = − ⎜1− ⎟ 2 ⎝ ε ⎠ Rion
This expression is known as Born model, it gives the charging free energy (hydra on energy) of a single ion in a spherical cavity.
Lecture 1, SJTU Summer School 2017 46 Polar Solvation For the general case of a solute of arbitrary shape with several par al charge sites, the electrosta c free energy is given by, 1 G q rf Δ elec = − ∑ iφi 2 i rf aq vac φi = φi −φi
φ sa sfies the Poisson equa on ∇⋅[ε(r)∇φ(r)] = −4πρ(r) - Analy cal solu on available for spherical, - - ε0 cylindrical, or planar symmetry - a - - +q - -
Lecture 1, SJTU Summer School 2017 47 Poisson-Boltzmann Theory
The electrostatic potential related to charge density is given by Poisson’s law
∇⋅ε(r)∇φ(r) = −4πρ(r) Mobile ions and the Poisson-Boltzmann equation _ (r) z exp( z (r)) + ρm = ∑ iρi,bulk − iβφ Expand at low-salt concentration + ∇⋅ε(r)∇φ(r) −κ 2φ(r) = −4πρ(r) _ where, κ 2 ~ βI + + in: ε1 _ The Tanford and Kirkwood model for protein _ N q + 2 (r!) i (r! r!) ∇ φ1 = −∑ δ − i i=1 ε1 2 ! 2 ! + ∇ φ2 (r) −κ φ2 (r) = 0 out: ε2
Lecture 1, SJTU Summer School 2017 48 Numerical Solution of PB
Numerical solution (FD, BE, FEM) The finite difference formulation: spatial derivatives are approximated using neighboring points. A successive overrelaxation method used to get rapid convergence in solving the linear systems obtained from the finite difference discretization; The boundary element method: utilizes analytical solutions obtained in terms of Green’s functions and discretization on the domain surface (molecular surface); The finite element method: an adaptive multilevel approach based on tetrahedral elements to create a dense mesh to capture the dielectric discontinuity across the molecular surface.
Lecture 1, SJTU Summer School 2017 49 Numerical Solution of PB
Advantages Disadvantages
Finite difference and • Fast solvers • Non-adaptive uniform mesh methods • Low memory overhead • Poor solution resolution • Cartesian mesh • Previous parallel methods complicated and inefficient
Boundary element methods • Smaller numerical systems • Less efficient solvers • Easier interaction • Only applicable to linear evaluation problem
Finite element methods • Highly adaptive • Previous solver and • Relatively fast solvers adaptive methods inadequate • Previous parallel methods complicated and inefficient adapted from Nathan A. Baker’s slides, North Dakota State University, 2003
Lecture 1, SJTU Summer School 2017 50 AMFPB
300
250 direct calculation 6-digit accuracy 3-digit accuracy 200
150
100
50 Time (seconds) Time
0 0 5000 10000 15000 20000 25000 30000 30S ribosome Number of Nodes 500 subunit 3-digit accuracy 400 6-digit accuracy 88,431 atoms 300 343,028 elements and 171,288 nodes
200 21 min on our desktop machine (Intel(R) 100 Xeon(TM) CPU 3.00 GHz, 4GB memory), with memory (megabytes) memory 2.6GB memory 0 0 5000 10000 15000 20000 25000 30000 Number of Nodes <1 min on our 64 core Dell workstation
Lecture 1, SJTU Summer School 2017 51 Reaction Field Method
An approximate local algorithm for electrostatic calculation.
ε0
ε1 qi
rc
Tironi et al. J. Chem. Phys. 1995, 102, 1
Lecture 1, SJTU Summer School 2017 52 Reaction Field Method
Schulz et al., J. Chem. Theory Comput., 2009, 2798
28 ns/day for a 5.4 million atom system running on 30k cores; >300TF for pure water systems on 150K cores
Lecture 1, SJTU Summer School 2017 53 Tree-code GB Generalized Born electrostatics 1 1 q q G GB (1 ) i j Δ pol = − − ∑ 2 ε ij fGB 2 2 rij 1/ 2 fGB = [rij + αiα j exp(− )] 4αiα j
The particle-cluster interaction energy
Xu et al. J Chem Phys, 2011, 064107
Lecture 1, SJTU Summer School 2017 54 Image Charge Method
The reaction potential can be approximated by M +1 image point charges plus a few correction terms,
Xu et al, SIAM J. Appl. Math. 2013
Lecture 1, SJTU Summer School 2017 55 56 Model for protein folding: (i) Framework model describes a step-wise mechanism to greatly narrow the conformational search. This involves a hierarchical assembly whereby local elements of secondary structure are formed according to the primary sequence, but independent from tertiary structure. These elements then diffuse until they collide, whereupon they coalesce to form the tertiary structure. (ii) Nucleation model suggests that tertiary structure forms as an immediate consequence of the formation of secondary structure. Nucleation occurs through the formation of native secondary structure by only a few residues (e.g. a beta-turn, or the first turn of an alpha-helix), and structure propagates out from this nucleus. (iii) Hydrophobic collapse model hypothesises that the native protein conformation forms by rearrangement of a compact collapsed structure. Hydrophobic collapse to form a molten globule therefore constitutes an early step in the folding pathway. (iv) Nucleation-condensation model suggests that a diffuse folding nucleus is formed in (ii) and consolidated through the transition state, concomitant with tertiary structure formation. (i), (ii), and (iv) suggest the formation of kinetic intermediates, whereas (iii) does not. (A. R. Fersht. Curr. Opin. Struc. Biol., 1997, 7, 1, 3-9.)
Lecture 1, SJTU Summer School 2017 57 Dynamics of Protein Structures
Lecture 1, SJTU Summer School 2017 58