Adding Detail to Models used in DrugDesign

1
Adding Detail to Models used in Drug-Design

• Christopher Woods

2
Structure-Based Drug Design

• Knowledge of receptor structure
• Experimental X-ray or NMR
• Theoretical homology model
• Optimise ligands for binding and specificity
  based on structure

3
Contents

• Two complementary pieces of research
• Undertaken at the University of Southampton,
  under the supervision of Dr Jonathan Essex
• Flexible-receptor docking - Dr Richard Taylor
• Role of tight binding waters - Caterina Barillari

4
Part 1 Docking

• Docking - predict how a ligand might bind to a
  protein active site
• Multiple protein-ligand configurations are
  generated, with each configuration being given a
  score
• Hopefully(!) the configuration with the best
  score corresponds to the experimental binding
  geometry
• For different ligands, score used to rank ligands

5
Existing Small-Molecule Docking

• There are many docking algorithms.
• 127 references in this 2002 review
• Taylor, R.D. et al., J. Comput. Aided Mol. Des.
  16, 151-166 (2002).
• Most algorithms use the rigid receptor
  hypothesis.
• Protein is mostly rigid, e.g. in GOLD only the
  polar hydrogens are allowed to move.

6
Protein Flexibility in Drug Design

• Multiple conformations of a few residues
• Acetylcholinesterase
• Phe330 flexible
• Acts as a swinging gate
• Teague, S.J., Nature Reviews, 2, 527-541 (2003)

7
Protein Flexibility in Drug Design

• Acetylcholinesterase
• Ligand binding causes significant changes in
  sidechain configurations
• Table 1 in the Teague paper lists some 30
  pharmaceutically relevant flexible targets
• Protein flexibility is important in ligand design

8
Flexible Protein Docking

• Two broad types of flexible docking methods
• Ensemble Docking
• Ensemble docking involves docking to a collection
  of protein conformations
• Efficient, but difficult to ensure that the
  ensemble contains all important conformations
• Induced Fit
• Induced Fit methods allow the protein
  conformation to change during the actual docking
  process
• Carlson, H.A., Curr. Opin. Chem. Biol., 6,
  447-452 (2002)

9
Flexible Soft Docking

• We developed our own flexible docking method
• Taylor, R. et al., J. Comput. Chem., 24,
  1637-1656 (2003)
• Use Monte Carlo with a rotamer library for large
  side-chain moves
• Effect of water included via GB/SA continuum
  solvation model
• Use a soft-core potential function, which is
  annealed to improve sampling

10
Results
11
Testing the Model

• Model optimised on Arabinose Binding Protein
• 15 complexes were studied
• Flexible docking found the X-ray structures, but
  could not uniquely identify them
• Need to refine the scoring function?
• Further validation is required

12
Importance of Water

• Water is amazing
• Docking work represented water using an implicit
  (fuzzy) solvent
• This models bulk water well, but does not do so
  well for structural waters in the active site
• Docking what waters to leave in?
• SBDD which waters should we aim to displace?

13
Displacing Water Is Good
Neuraminidase DANA (1f8b) Ki 4 ?M
Zanamivir (1nnc) Ki 1.3 nM
14
Is Displacing Water Always Good?

• Are all waters worth displacing?
• The cost of displacing a very tightly bound water
  may not be recovered on ligand binding
• Very useful to know which waters are loosely
  bound, and which are tightly bound

15
Free Energy Calculations
16
Test Systems

• Binding free energies of 51 water molecules were
  calculated
• Dataset included six proteins Neuraminidase,
  HIV-1 Protease, FXa, Scytalone Dehydratase, OppA
  and Trypsin
• Dataset included 5-6 ligands for each protein
  (Res. lt 2.5 Å)

17
Application to Neuraminidase

• Positive free energies of water A show that it is
  easy to displace
• Large, negative free energies of water B show
  that it is tightly bound

18
Classification of Waters

• Conserved water molecules are more tightly bound
  than those displaced by ligands.
• Calculations confirm what is intuitively
  expected.
• We are confident that we can use the absolute
  binding free energy of the water to the protein
  to classify whether or not the water is
  conserved, or can be readily displaced.

19
Advanced Statistical Models

• Free energy calculations are expensive
• A statistical model is needed to predict
  tight-binding waters from only the structure
• Advanced statistical methods are being used to
  find correlations between the calculated water
  binding free energy and different molecular
  descriptors

20
Advanced Statistical Models

• Started from 30 descriptors and 50 datapoints.
• Used linear and non-linear modelling.
• PCA, PLS, GFA, GAM
• Keep finding the same key descriptors
• Apolarity, polarity or H-Bonds, SuperStar
  probability

21
Non-linear Models
a cubic splines b GAM
5 term model
22
Conclusion

• We have developed methods that allow for protein
  side-chain flexibility during docking
• We are validating these methods and extending
  them to include limited protein backbone motion
• We have developed methods to calculate the
  strength of binding of water molecules
• We are developing statistical models to predict
  which water molecules are tightly bound

23
Acknowledgements

• Docking
• Dr. Richard Taylor (Rich.Taylor_at_ucb-group.com)
• Dr. P. Jewsbury, Astra Zeneca
• Donna Goreham, Sebastien Foucher
• Prediction of tight binding waters
• Caterina Barillari (C.Barillari_at_soton.ac.uk)
• Dr. R. Viner and the Drug Design group at
  Syngenta
• University of Southampton
• Dr. Jonathan Essex (J.W.Essex_at_soton.ac.uk)
• University of Southampton ISS
• Funding
• Astra Zeneca, Syngenta, EPSRC