1. The Many Roles of Computational Science in Drug Design and Analysis Mala L. Radhakrishnan Department of Chemistry, Wellesley College June 17, 2008 DOE CSGF Fellows Conference, Washington, D.C.

2. Amprenavir bound to HIV-1 protease (HIV) Kim et al, J. Am. Chem. Soc. 117:1181, 1995 Shafer, Clinical Microbiology Rev., 15:247, 2002 …target rapidly mutates!

3. Understanding binding specificity and promiscuity Developing methods for optimal drug cocktail design Designing broadly-binding HIV-1 protease inhibitors HIV−1 Protease

4. Ligand (drug) Receptor (target) Goal: create a high-affinity interaction  low  G

5. Ligand (drug) Receptor (target) “Reality”: Quantum mechanics Statistical Mechanics

6. Model: Atoms as spheres Point charges (Rigid Binding) Ligand (drug) Receptor (target)

7. Model: Atoms as spheres Point charges (Rigid Binding) + + - - Ligand (drug) Receptor (target) + -

8. + + - - Drug Target + - + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − Interactions with solvent

9. + + - - + - Loss of solvation  FAVORS LOW CHARGE MAGNITUDES Drug-target interaction  FAVORS HIGH CHARGE MAGNITUDES + + + + − − + + + + − − + + + + − − + + + + − − + + + + − − ++ ++ −− ++ ++ −− + + + + − −

10. + + - - + - Our Model “solvent continuum”: high dielectric + mobile ions

11. Designing Toward Multiple Targets + - + - - + + - + + - ? Physical Properties  Tailored Recognition

12. Theoretical Framework + LR C Solvent continuum Kangas and Tidor, J Chem Phys, 109:7522-7545, 1998  G bind = vdW + SASA + q L T Lq L + q R T Rq R + q R T Cq L GG charge value The binding profile looks like a paraboloid. + … + - - ++ -- - + ++ -- + + - -

13. Theoretical Framework A promiscuous ligand A specific ligand Radhakrishnan and Tidor, J. Phys. Chem. B, 111:13419-13435, 2007

14. Model Systems Radhakrishnan and Tidor, J. Phys. Chem. B, 111:13419-13435, 2007

15. Smaller ligands are more promiscuous than larger ligands Theory and Model Systems Agree: Drug charge distribution  Numerical Experiment Theory GG qRqR Binding profile for a big ligand Binding profile for a small ligand black:small red: big

16. Some Other Insights Smaller ligands are more promiscuous Hydrophobic ligands are more promiscuous because they are near the center of biological charge space Hydrophobic ligands are more promiscuous because they are not as sensitive to shape differences Flexibility makes polar and charged ligands more specific, but allows for greater overall binding affinity to multiple partners. Asymmetric groups can lead to increased promiscuity Radhakrishnan and Tidor, J. Phys. Chem. B, 111:13419-13435, 2007

17. HIV−1 Protease Understanding binding promiscuity Developing methods for optimal drug cocktail design Designing broadly-binding HIV-1 protease inhibitors

18. Rational Cocktail Design Drug 2 Drug 1 target variants decoy

19. Cocktail Design as an Optimization Problem Can also combinatorially design individual molecular members simultaneously Minimize # drugs in cocktail All targets covered by at least one drug Minimize sum of binding energies Size of cocktail is optimal E drugs targets+decoys affinity thresholds A drugs targets A’ 1 0 0 0 1 1... pre-process out decoys (mxn)

20. Optimally covering model ensembles -- - - 0 0.5 0.5 -0.5 0 -0.5 0 Radhakrishnan and Tidor, J. Chem. Inf. Model. 48:1055-1073, 2008

21. Tiling the Mutation Space -0.5 0.5 0.5 -0.5 0 -0.5 0 0.5 0.5 -0.5 -0.5 0.5 q(target,1) q(target,2) q(target,1) q(target,2) 0

22. HIV−1 Protease Understanding binding promiscuity Developing methods for optimal drug cocktail design Designing broadly-binding HIV-1 protease inhibitors

23. Designing into Multiple Targets: HIV-protease Wild Type V82A I84V D30N L90M I50V L63P, V82T, I84V

24. Methods: Molecular “scaffolds” F OH Functional groups F Michael Altman et al. J. Amer. Chem Soc., 130: 6099-6113, 2008  G, fast energy function  G,  High resolution energy function Dead-End Elimination, A*: List of tight- binding compounds

25. Design of Broadly − Binding HIV − 1 Protease Inhibitors Wild Type V82A I84V D30N L90M I50V L63P, V82T, I84V Physical Properties 07 0 7 Energy threshold (kcal/mol)Size of optimal cocktail Michael Altman et al. J. Amer. Chem Soc., 130: 6099-6113, 2008

26. Summary Physical Framework for Binding Promiscuity and Specificity Methods For Designing Toward Target Ensembles HIV-1 Protease as a Case Study

27. The Many Roles of Computation Computation can… …map out key interactions in a binding site. …design drugs. …predict properties that make drugs specific or “promiscuous.” …design drug cocktails. …model cellular events. …design novel experimental systems. …etc. ! …analyze clinical data

28. Acknowledgements MIT Wellesley -Bruce Tidor- Bilin Zhuang and Andrea Johnston -Michael Altman -Dave Czerwinski -Celia Schiffer, Tariq Rana, Michael Gilson and groups. Funding: DOE CSGF, NIH, Wellesley College