Fill and Sign the Fedex Authorization Letter for Pickup Form

Rapid Calculation of Polar Molecular Surface Area and its Application to the Prediction of Transport Phenomena David E. Clark Computer-Aided Drug Design Rhone-Poulenc Rorer, UK Background/Rationale • Andersen Consulting report* - 3 NCEs/year • Reduce fall-out rate in development • Nature of compounds, not just number of compounds is important • “Fail fast” is the watchword • In vitro ADME screening - synthesized compounds • Computational screens - virtual compounds * Banerjee, P. “Re-inventing Drug Discovery” (July 1997). Overview • Intestinal Absorption – Lipinski’s rule-of-5 – Polar Surface Area (PSA) • Blood-Brain Barrier (BBB) Penetration – Introduction – Application of PSA calculations • Applications in drug design Note: passive absorption only! Intestinal Absorption - Rule of 5 Compound flagged as having possible absorption problems if any two of the following rules apply to it: • • • • Molecular weight > 500.0 ClogP > 5.0 (MlogP > 4.15) Number of H-bond acceptors (any N,O) > 10 Number of H-bond donors (any N-H, O-H) > 5 Reference: • Lipinski, C.A. et al. Experimental and Computational Approaches to Estimate Solubility and Permeability in Drug Discovery and Development Settings. Adv. Drug Deliv. Rev. 1997, 23, 3-25. Intestinal Absorption - Rule of 5 Advantages: • Widely adopted by industry • Quick to calculate • Should give few false negatives Disadvantages: • A crude prediction method • Lots of false positives • Not discriminating enough Intestinal Absorption - PSA Background: • Polar surface: N, O; H attached to N or O • Palm et al. correlated “dynamic” PSA with fractional absorption in humans for 20 carefully selected drugs • Sigmoidal fit: r2=0.94, RMSE=9.2% (but slow!) • Using single conformation: • Sigmoidal fit: r2=0.94, RMSE=9.1% Reference: • Palm, K. et al. Polar Molecular Surface Properties Predict the Intestinal Absorption of Drugs in Humans. Pharm. Res. 1997, 14, 568-571. PSA Calculation Method 1) SMILES representation of compound 2) CONCORD generates approximate 3-D structure 3) Structure optimized using Maximin2 in SYBYL 4) Van der Waals surface calculated by MOLVOL 5) Inhouse program determines PSA CPU time: ~10 secs/structure (SGI R10000) Fully automated: intranet interface/batch scripts Intestinal Absorption - PSA PSA > 140 A^2 => possible absorption problems Fractional Absorption/% 100 75 50 25 0 0 50 100 150 200 Polar Surface Area (Concord)/A^2 250 Intestinal Absorption - PSA Validation Study 1: • 74 compounds taken from Wessel et al. Human FA (%) 100 75 50 25 0 0 50 100 150 200 250 300 Polar Surface Area (Angstroms2) Intestinal Absorption - PSA Seven apparent “false negatives” (shown as crosses) • 6/7 shown to be actively transported: – methotrexate: folate carrier, AZT: rCNT – lisinopril, amino-beta-lactams: dipeptide carrier • Etoposide - erratic bioavailability, active transport? • Remaining 67 compounds are classified correctly by PSA >140 criterion • Rule-of-5 only generates warnings for 2 out of the 5 poorly absorbed compounds (%FA < 10) Reference: • Wessel, M.D. et al. Prediction of Human Intestinal Absorption of Drug Compounds from Molecular Structure. J. Chem. Inf. Comput. Sci. 1998, 38, 726-735. Intestinal Absorption - PSA Validation Study 2: • Kansy set - 24 compounds with % absorption data Classified experimentally (human fractional absorption): • High (abs. > 70%): 19 • Moderate (< 30% abs. < 70%): 2 • Low (abs. < 30%): 3 Reference: • Kansy, M. et al. Physicochemical High Throughput Screening: Parallel Artificial Membrane Permeation Assay in the Description of Passive Absorption Processes. J. Med. Chem. 1998, 41, 1007-1010. Intestinal Absorption - PSA Kansy experimental PAMPA classification: • High: 13. (Correct) • Moderate: 8. (6 errors) • Low: 3. (Correct) PSA classification: • High (PSA < 83): 16. (Correct) • Moderate (83 < PSA < 113): 4. (2 errors) • Low (PSA > 113): 4. (1 error) Applying PSA > 140 cutoff predicts low absorption compounds correctly Intestinal Absorption - PSA Conclusions: • Slower to calculate than rule-of-5 • But more accurate at prediction • PSA >140 criterion for poor absorption seems robust • Why does PSA seem to work? • False positives indicate other criteria are necessary for more accurate prediction of absorption, e.g. pKa, logD O O + N NMe2 BBB Penetration - QSARs Introduction - QSARs for logBB (BB=Cbrain/Cblood): • Kansy/van de Waterbeemd: – Chimia 1992, 46, 299-303. – PSA, molecular volume • n=20, r=0.84, s=0.45, F=19.5 – training set too small, not general enough • Abraham et al. (AB): – J. Pharm. Sci. 1994, 83, 1257-1268. – 5 empirical solute descriptors • n=57, r=0.95, s=0.20, F=99.2 – not yet automated, missing fragments BBB Penetration - QSARs • Lombardo et al. (LO): – J. Med. Chem. 1996, 39, 4750-4755. – Free-energy of solvation (AMSOL) • n=55, r=0.82, s=0.41, F=108.3 – Computationally expensive • Norinder et al. (NO): – J. Pharm. Sci. 1998, 87, 952-959. – MolSurf parameters, PLS analysis (3 comps.) • n=56, r=0.91, s=0.31, F=87.0 – Computationally expensive BBB Penetration - QSARs • Luco (LU): – JCICS 1999, 39, 396-404. – Topological descriptors, PLS model (3 comps.) • n=58, r=0.92, s=0.32, F=102.0 – Interpretability? • In-house equation (RPR): logBB = -0.0148 PSA + 0.152 ClogP + 0.139 n=55, r=0.89, s=0.35, F=95.8 – Large training set, automatic, fast – Can also use MlogP with good results BBB Penetration - Test Sets Prediction set 1: Compound Y-G14 Y-G15 Y-G16 Y-G19 Y-G20 SKF89124 SKF101468 Mean abs. error: logBB values Expt AB -0.30 -0.31 -0.06 -0.01 -0.42 -0.41 -1.30 -0.14 -1.40 -0.57 -0.43 -0.44 0.25 0.24 0.30 LU -0.14 0.02 -0.51 -0.04 -0.29 -0.26 0.17 RPR -0.30 -0.09 -0.59 -0.24 -0.53 -0.56 -0.11 0.42 0.37 Ref: Abraham, M.H. et al. Drug Des. Discov. 1995, 13, 123-131. BBB Penetration - Test Sets Prediction set 2: Cmpd 31 32 33 34 35 Expt 0.00 -0.34 -0.30 -1.34 -1.82 Mean abs. error: logBB values LO NO -0.14 -0.58 -0.28 -1.11 -0.46 -0.75 -0.64 -0.99 -0.82 -1.35 LU -0.01 0.01 -0.45 -0.93 -1.31 RPR -0.25 -0.75 -0.70 -1.26 -1.77 0.41 0.29 0.24 0.52 Refs: Lombardo, F. et al. J. Med. Chem. 1996, 39, 4750-4755. Norinder, U. et al. J. Pharm. Sci. 1998, 87, 952-959. Luco, J.M. J. Chem. Inf. Comput. Sci. 1999, 39, in press. BBB Penetration - Test Sets Prediction set 3 • Most stringent test • 25 diverse drug compounds from Luco (1999) • Luco results: – mean abs. error = 0.43 – RMSE = 0.54 • RPR results: – mean abs. error = 0.50 – RMSE = 0.59 BBB Penetration Conclusions: • Predictive capability of RPR equation comparable to other methods • Fully automated, properties are quick to calculate • Easy to interpret • Physically sensible: logBB = -0.0148 PSA + 0.152 ClogP + 0.139 • PSA reflects H-bonding capacity (desolvation) • ClogP indicates lipophilicity • Brain is more lipophilic than blood Applications in Drug Design • (Virtual) compound assessment – Intranet interface for medicinal chemists – Batch processing for large sets • Combinatorial library design – Product-based reagent selection • Database for analysis – Graphical viewing of trends/patterns – Validation against inhouse experimental data – Example: Caco-2 %abs vs. PSA Conclusions • Intestinal Absorption – Lipinski’s rule-of-5: fast but crude – Polar Surface Area: slower but more accurate. PSA > 140 for poor absorption • Blood-Brain Barrier (BBB) Penetration – Fast and relatively accurate predictions using PSA/ClogP-based QSAR • Widely applied to inhouse drug discovery programs Acknowledgments RPR (UK): • A.J. Ratcliffe, I. McLay (Medicinal Chemistry) • S. Pickett, P. Bamborough, R. Lewis (CADD) • C. Brealey, N. Wilsher, K. Page (DDD) • P. Delpy, R. Stewart (Research Computing) Others: • K. Luthman (Tromso University, Norway) • L. Dodd (Polytechnic University, New York)