Using waterswap to predict protein-ligand binding affinities
1. Using waterswap to predict and
understand binding affinities
Christopher Woods
2. Introduction
• Developer of software and algorithms to
predict protein-ligand binding free energies
• Binding free energy measures binding
affinity, can be directly related to Ki
• Developed “waterswap”. First-principles,
calculation of absolute binding free energies
20. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
21. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
22. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
23. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
λ=0.0
24. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
λ=0.0
100%
0%
25. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
λ=0.2
80%
20%
26. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
λ=0.5
50%
50%
27. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
λ=0.8
20%
80%
28. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
λ=1.0
0%
100%
29. Perform Thermodynamic Integration (TI) along the
Waterswap λ coordinate.This results, directly,
in the absolute binding free energy
!20$
!18$
!16$
!14$
!12$
!10$
!8$
!6$
!4$
!2$
0$
0.0$ 0.2$ 0.4$ 0.6$ 0.8$ 1.0$
Free$Energy$/$kcal$mol01$
λ$
30. Perform Thermodynamic Integration (TI) along the
Waterswap λ coordinate.This results, directly,
in the absolute binding free energy
!20$
!18$
!16$
!14$
!12$
!10$
!8$
!6$
!4$
!2$
0$
0.0$ 0.2$ 0.4$ 0.6$ 0.8$ 1.0$
Free$Energy$/$kcal$mol01$
λ$
31. Perform Thermodynamic Integration (TI) along the
Waterswap λ coordinate.This results, directly,
in the absolute binding free energy
!20$
!18$
!16$
!14$
!12$
!10$
!8$
!6$
!4$
!2$
0$
0.0$ 0.2$ 0.4$ 0.6$ 0.8$ 1.0$
Free$Energy$/$kcal$mol01$
λ$
ΔGbind
52. …but,
• waterswap is easy to use…
• …but setting up a protein-ligand complex
for simulation requires expert knowledge
and is not trivial
• waterswap results depend on the quality of
the input model
65. !
Simulation should not try to
compete with experiment.
!
The job of simulation is to
provide inspiration and insight
66. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
67. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
68. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
69. Free Energy Decomposition
• As we integrate the total waterswap
binding free energy...
• ...we also integrate free energy changes in
the “protein” box and the “water” box
• Result are free energies that tell you if a
ligand’s binding strength comes from a
natural affinity for the protein, or an
aversion to water
70. -6.5
-6.0
-5.5
-5.0
-4.5
-4.0
-3.5
-10 -8 -6 -4 -2
Experiment/kcalmol-1
Simulation / kcal mol-1
R2=0.14
1
2
3
4
8 10
6
7
5
9
-6.5
-6.0
-5.5
-5.0
-4.5
-4.0
-3.5
-26 -24 -22 -20 -18 -16 -14 -12
Experiment/kcalmol-1
Simulation / kcal mol-1
R2=0.84
1
23
4
5
6
108
79
Specificity driven by “water” box, i.e.
the more hydrophobic the ligand, the less it
wants to be in the water box, and the more it
wants to be in the protein box.
!
This shows that a “better” ligand is only better
because it is more hydrophobic
Protein Box Water Box
71. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
72. Waterswap uses a λ-coordinate to swap a ligand and a
water cluster between a protein box and a water box
Protein Box Water Box
E = (1 )[Eprotein:cluster + Ewater:ligand]+
( )[Eprotein:ligand + Ewater:cluster]
Eresidue:cluster
Eresidue:ligand
73. Free Energy Decomposition
• As we integrate the total waterswap
binding free energy...
• ...we also integrate the individual
contributions from all of the binding site
residues
• Result is a “free energy” that indicates
whether the residue:ligand or
residue:water complex is more stable
77. ure S1. Experimentally measured binding affinities for the 10 ligands studied in thi
k (m-chlorobenzyl) and for the ten benzamidine analogs. Binding affinities are tak
m Muley et al., doi: 10.1021/jm9016416 (reference 32 in our paper).
8
10
-9
-8
-7
-6
-5
-4
-3
0 1 2 3 4 5 6 7 8 9 10
BindingAffinity/kcalmol-1
Ligand
m-chlorobenzyl
benzamidine
Replacing m-chlorobenzyl group with benzamidine
group systematically improves binding of the ligands
78. Conclusion
• Waterswap enables direct, first-principles
calculation of absolute binding free energies
• (but results depend on quality of model!)
• Free energies can be decomposed to per-
residue and per-water components
• Aim is to provide inspiration and insight
79. Appendix
• waterswap is just one of our tools…
• Also have ligandswap, which calculates
relative binding free energies by swapping
one ligand with another
• Also have waterview, that lets you quickly
visualise water dynamics in a binding site,
e.g.
80.
81.
82. Acknowledgements
• Organisers for inviting me and allowing me to talk
• You for your attention
• Dr. Maturos Malaisree (doing most of the work!)
• Dr. Julien Michel (discussions and providing thrombin test system)
• Prof.Adrian Mulholland, Simon McIntosh-Smith, Ben Long
• EPSRC and now BBSRC for funding
• eInfraStructureSouth for GPU compute
• ACRC (Bristol) for CPU compute
• Get the software at http://siremol.org
• Get in touch via Christopher.Woods@bristol.ac.uk
83. Identity Constraint
• How do we “identify” the cluster of water to be
swapped with the drug?
• We developed the identity constraint.This is a
new way of labelling water molecules in a
simulation that is based on where the molecule
is in space, rather than where it is located in the
input coordinate file.
• Allows definition of water clusters without
using restraints or external perturbations
96. Reflection Sphere
• Only waters
whose centers
are inside the
sphere can move
• Any move that
takes the center
of a water outside
the sphere is
reflected back
into the sphere
• This prevents
waters from
leaving
97. Grid Electrostatics
• Interactions inside
reflection sphere
calculated normally
• Interactions between
reflection sphere atoms
and atoms within buffer
(dotted sphere)
calculated normally
• Coulomb interactions
between reflection
sphere and fixed atoms
outside the buffer are
calculated using a pre-
computed cubic grid
98. Grid Electrostatics
• Use of pre-computed
grid means that there is
no penalty to using a
long-range electrostatic
cutoff
• Compatible with
advanced boundary
conditions, such as
reaction field or force-
shifted cutoff
• Fine grid (0.5 Å) and
tri-linear interpolation
give high accuracy
compared to direct
calculation