Structure-based maximal affinity model predicts small-molecule druggability

Transcription

1 Structure-based maximal affinity model predicts small-molecule druggability Alan Cheng IMA Workshop (Jan 17, 2008) Druggability prediction Introduction Affinity model Some results 1

2 Why estimate druggability? 60% of programs fail in HTS and Hit-to-lead Brown & Superti-Furga Drug Discovery Today (2003) Traditional way: Sequence homology Certain gene families tend to be druggable e.g., Kinases and GPCRs Used to estimate druggable genome Hopkins & Groom Nature Rev. Drug Disc. (2002) Unprecedented targets and gene families Not all members of a gene family are equally druggable 2

3 HTS way: Screening a diverse library NMR screening hit-rate* Diverse compound collection screening hit-rate Reagent, screening investment * Hajduk et al. J Med Chem Biophysically-inspired way: Structure-based Druggable Undruggable Qualitative, intuitive: Can we make this quantitative? 3

4 Concept of maximal affinity Maximal affinity of ligands ~1.5 kcal/mol/atom Kuntz et al. PNAS (1999) Extend to binding sites? Restrict to drug-like ligands Oral drugs tend to have drug-like properties 20% 15% 10% 5% 0% weight Molecular Weight (Da) 30% 25% 20% 15% 10% 5% 0% Polar surface tpsa area (tpsa, A 2 ) >90% of oral drugs fall within physiochemical ranges Marketed oral tablets in MDDR v.2001 Similar to Lipinski et al. 2001, Palm et al

5 Translating to the protein binding pocket 550 MW ~ 300A 2 Maximal affinity predicted, ΔG MAP-POD Non-polar surface area for binding site 300A 2 surface area ~ 550 MW ΔG MAP-POD ~ γ(r) A NP 300 A total One fitted parameter Curvature-dependent HPO desolvation term γ(r) (kcal/mol) = 45 kcal/mol/a 2 p = 1.4A Sharp et al. Science (1991) Dill et al. J Phys Chem B (2003) De Young & Dill, J Phys Chem (1990) Curvature r (Å) 5

6 Implementation Algorithms Precisely defining pocket for surface area calculation Liang, et al. (1998) Protein Sci. Generate tetrahedra representation Calculate protein core Define pocket tetrahedra Koehl, POCKET Curvature: New sphere fitting approach using geometric inversion Brannan, Esplen, Gray (1999) Geometry Appolonius (200BC) Coleman, Burr, Souvaine, Cheng (2005) Proteins 6

7 All models are wrong, some are useful. George Box Validation on 27 targets Druggable Undruggable 7

8 Scholarship finds outliers are prodrugs Druggable Undruggable Prediction of druggable and difficult targets 8

9 Prediction and validation with two novel targets Unprecedented targets/ unprecedented gene families Predictions made before targets entered portfolio Screened 11k chemical space diverse compound set Predicted Druggability Raw hits 20 um, 60% cutoff Confirmed hits IC50<5uM IC50<1uM Fungal HSD raw hits 2 confirmed hits 0 confirmed hits H-PGDS raw hits 33 confirmed hits 11 confirmed hits Cheng et al. (2007) Nature Biotechnology Do experimental maximal affinities correlate? Experimental affinities for orally bioavailable compounds. Literature mining; values are approximate (combination of Kd s, Ki s, IC50 s) Correlation is very encouraging Cheng et al. (2007) Nature Biotechnology 9

10 Druggability in practice: Caveats Binding site structures are treated explicitly Predictions are for oral, passively absorbed, noncovalent drugs Large conformational changes (especially loops) Unspecified binding sites Metal chelation Covalent adducts Active transport Prodrug strategy Alternate delivery/approaches Take a measured risk for compelling biology These are predictive risk assessment tools Significant conformational change Druggability prediction Druggable Target space Model based on nonpolar desolvation Correlation with HTS and Phase II outcomes Disease modifying 10

11 Expanding druggable space Structure-based drug design Allosteric sites Structure-based drug design Shape VDW, hydrophobic Can be optimized by eye with reasonable success. Charge Hydrogen bonds, Ionic pairs More difficult to optimize b/c affinity is not as intuitive (not just interaction, also desolvation) prot-ligand ligand protein ΔG electrostatic = ΔG interaction + ΔG desolvation + ΔG desolvation 11

12 Tidor Lab charge optimization Ligand desolvation (Q 2 ) Free Energy Net Electrostatic Energy Protein desolvation Ligand charge P-L interaction Coulombic (Q) Tidor et al. Protein Sci. (1998) Charge optimization in lead progression Applied to available series of six co-crystal structures for neuraminidase (antiviral target) Goal, retrospectively study utility in lead progression 12

13 Neuraminidase case study Focus on R-groups Increasingly potent compounds-- generally R-groups closer to optimal charge distribution Lead optimization results in charge optimization Armstrong, Tidor, Cheng. J Med Chem (2006) Crystallographic water for Oseltamivir binding Optimal charge distribution provides an explanation for crystallographic water 13

14 Towards Identifying Druggable Allosteric Sites Druggable Functionally relevant Protein surface Computational bioinformatics approach Statistical coupling analysis 1. Large sequence alignment 2. Identify coupled residues 3. Map to structure Lockless & Ranganathan, Science (1999) 14

15 Local version of druggability equation Potential allosteric sites in p38a/kinases Top site identical to small molecule allosteric inhibitor site recently identified in cabl (Nature Chem. Biol. 2006) Other predicted site: Inhibitor recently found for Jnk1 (Abbott Pharmaceuticals, Oct 2007, Manuscript in preparation) Coleman, Salzberg, Cheng, J Chem Inf Model (2006) 15

16 Summary and References Druggability Nonpolar desolvation drives maximal drug-like affinity. This is quantitatively useful. Proteins (2005) 61, Nature Biotechnology (2007) 25, Expanding druggable space Charge optimization is helpful for SBDD in polar binding sites Finding allosteric sites by combining functional residue prediction and druggability predictions. J Med Chem (2006) 49, J Chem Inf Model (2006) 46, Acknowledgements Computational geometry Ryan Coleman (Pfizer, Tufts Univ.) Diane Souvaine (Tufts Univ.) Structure-based druggability Kate Smyth and Patricia Soulard (Pfizer Biology) Qing Cao, Daniel Caffrey, Anna Salzberg, Enoch Huang, RTC MI colleagues Advice from Eric Fauman, Ken Dill (UCSF), Pfizer Cambridge and Pfizer Global R&D colleagues Charge optimization Kathryn Armstrong (MIT, Pfizer) Bruce Tidor (MIT) Allosteric sites Anna Salzberg (Brandeis, Pfizer) alan.cheng@amgen.com 16