Structure based drug design and LIE models for GPCRs Peter Kolb kolb@docking.org Shoichet Lab ACS 237 th National Meeting, March 24, 2009 p.1/26
[Acknowledgements] Brian Shoichet John Irwin Mike Keiser Brian Kobilka Daniel Rosenbaum Juan José Fung Jérôme Hert Jonathan Mason NIH Anne Marie Jørgensen ACS 237 th National Meeting, March 24, 2009 p.2/26
[utline] The β 2 -adrenergic receptor LIECE Docking against β 2 AR Ligands Ligand Efficacy The approach Earlier examples LIECE for β-adrenergic receptors Conclusions ACS 237 th National Meeting, March 24, 2009 p.3/26
[ββ 2 AR xtal structure] T4-lysozyme (for stability) Cerezov et al., Science 318:1258-1265 Rosenbaum et al., Science 318:1266-1273 ACS 237 th National Meeting, March 24, 2009 p.4/26
[ββ 2 AR xtal structure] binding site T4-lysozyme (for stability) Cerezov et al., Science 318:1258-1265 Rosenbaum et al., Science 318:1266-1273 ACS 237 th National Meeting, March 24, 2009 p.4/26
[ββ 2 AR xtal structure] binding site T4-lysozyme (for stability) Cerezov et al., Science 318:1258-1265 Rosenbaum et al., Science 318:1266-1273 ACS 237 th National Meeting, March 24, 2009 p.4/26
[AA beautiful binding site] ACS 237 th National Meeting, March 24, 2009 p.5/26
[AA beautiful binding site (2)] ACS 237 th National Meeting, March 24, 2009 p.6/26
[AA beautiful binding site (2)] F193 5.32 W109 3.28 S203 5.42 H H 2 N N312 7.39 D113 3.32 V114 3.33 ACS 237 th National Meeting, March 24, 2009 p.6/26
[Docking] Calculations were done with DCK 3.5.54. Docking spheres where distributed based on the positions of the heavy atoms of carazolol and slightly altered to achieve more uniform coverage. The T4-lysozyme insertion was removed, as it carried a net charge of +9. We docked the lead-like subset of the ZINC database, which contains about 1M molecules. DCKING ACS 237 th National Meeting, March 24, 2009 p.7/26
[What we predicted] Four classes of substrates have been predicted: classic compounds with all canonical pharmacophoric features of antagonists. bridge compounds connecting the two walls of the binding pocket (Thr195 to Tyr308). sulfonamide compounds. others. We managed to acquire 12, 9, 2 and 2 compounds of each class, respectively. ACS 237 th National Meeting, March 24, 2009 p.8/26
[Results] Classic compounds: 4 hits. Bridge compounds: 0 hits. Sulfonamides: 0 hits. thers: 2 hits. 6 hits out of 25 tested 24% hit rate ACS 237 th National Meeting, March 24, 2009 p.9/26
[The six hits] Me Et Me H Me 1 H 2 N + Me Me 0.009 µm rank 6 N S N + Me 5 1.1 µm 6 3.2 µm rank 97 rank 156 H N + H N + Me N + H H 2 H 2 2 2.0 µm 3 0.47 µm 4 0.75 µm rank 37 rank 42 rank 104 Me N + H 2 Et Kolb et al. PNAS, accepted. ACS 237 th National Meeting, March 24, 2009 p.10/26
[The six hits] Me Et Me H Me 1 H 2 N + Me Me 0.009 µm rank 6 N S N + Me 5 1.1 µm 6 3.2 µm rank 97 rank 156 H N + H 2 N + H 2 Me 2.0 µm 3 0.47 µm rank 37 rank 42 N + H 2 Me H 4 N + H 2 0.75 µm rank 104 Et Kolb et al. PNAS, accepted. ACS 237 th National Meeting, March 24, 2009 p.10/26
[For comparison: five β 2 antagonists] H H 2 N+ HN H N + H 2 HN N + H 2 H ICI 118551 Carvedilol Carazolol F 3 C N N CH 3 H N + H 2 H N H 2 H + N H2 CGP 20712A Alprenolol ACS 237 th National Meeting, March 24, 2009 p.11/26
[The hits are all inverse agonists] ACS 237 th National Meeting, March 24, 2009 p.12/26
[The search for hay in a haystack] LGICs β-lactamase β 2 AR proteases kinases GPCRs 15 10 5 0 1500 1000 500 0 % of library with significant assignment number of molecules with significant assignment ACS 237 th National Meeting, March 24, 2009 p.13/26
LIECE ACS 237 th National Meeting, March 24, 2009 p.14/26
[Ranking compounds: LIECE] Linear Interaction Energy using Continuum Electrostatics (Huang and Caflisch, J Med Chem 47:5791-5797). G bind α E vdw + β E elec + γ E E elec = E complex (E protein + E ligand ) by CHARMM. = E coul + G solv G solv is calculated with the PBEQ solver in CHARMM. Known inhibitors are needed to calibrate α,β and γ. Minimisation to a native like conformation Energy evaluation with solvation ACS 237 th National Meeting, March 24, 2009 p.15/26
[riginal study: proteases] protein α β β-secretase 0.2737 0.1795 standard deviation ±0.0395 ±0.0461 HIV1-PR 0.1690 0.0168 standard deviation ±0.0196 ±0.0199 Two proteases, yet clearly different coefficients parameters are not transferable, right? ACS 237 th National Meeting, March 24, 2009 p.16/26
[The human kinome] ACS 237 th National Meeting, March 24, 2009 p.17/26
[LIECE for kinases] protein α β CDK2 0.2866 0.0520 standard deviation ±0.0171 ±0.0183 Lck 0.2735 0.0046 standard deviation ±0.0182 ±0.0237 P38 0.2699 0.0264 standard deviation ±0.0210 ±0.0170 For kinases, parameters seem to be transferable. Consequently, we tried to derive a universal model. ACS 237 th National Meeting, March 24, 2009 p.18/26
[LIECE for kinases (2)] -5 LIECE model for kinases 165 compounds; 3 different X-ray structures -5 Prediction for EphB4 37 compounds; RMSE=1.30 kcal/mol -5 Prediction for EGFR 128 compounds; RMSE=1.42 kcal/mol -6-7 -8 73 cpds in 1ke5 (CDK2) 51 cpds in 1qpe (lck) 41 cpds in 1m7q (p38) -6-7 -8-6 -7-8 LIECE prediction -9-10 -11-12 -9-10 -11-12 -9-10 -11-12 -13-13 -13-14 -14-14 -15-15 -15-16 -16-15 -14-13 -12-11 -10-9 -8-7 -6-5 exp G -16-16 -15-14 -13-12 -11-10 -9-8 -7-6 -5 exp G -16-16 -15-14 -13-12 -11-10 -9-8 -7-6 -5 exp G G calc = 0.2898 E vdw + 0.0442 ( E coul + G solv ) Kolb et al. J Med Chem 51:1179-1188 ACS 237 th National Meeting, March 24, 2009 p.19/26
enzymatic assays cell-based assay [Kinase hits] compound structure MW IC 50 IC 50 IC 50 CH predicted G [Da] [µm] [µm] [µm] % [kcal mol 1 ] H 2N S NH NH 1 349 76 n.d. n.d. n.d. -8.8 2-8 HN N N N N R R= 2 (CH 2 ) 4 H 353 1.4, 1.8 6.8 5.6, 7.9 neg. -11.0 3 (CH 2 ) 3 CH 3 337 1.5, 1.5 8.3 n.d. neg. -10.7 4 (CH 2 ) 2 CH 3 323 1.9 n.d. n.d. neg. 5 ortho-methoxy-phenyl 387 27 n.d. n.d. 14% @ 20 µm 6 (CH 2 ) 3 CH 3 353 50 n.d. n.d. 28% @ 20 µm 7 (CH 2 ) 2 Ph 385 30% @ 14 µm e n.d. n.d. 34% @ 20 µm 8 (CH 2 ) 2 -meta-methylphenyl 399 29% @ 5 µm e n.d. n.d. 14% @ 20 µm ACS 237 th National Meeting, March 24, 2009 p.20/26
LIECE for GPCRs ACS 237 th National Meeting, March 24, 2009 p.21/26
[What we used for β 2 AR] 7 ligands, all enantiopure, from Dallanoce et al., BMC 14, 4393. 20 ligands that belong to the well-known set of antagonists (all S) and are described in Baker et al. Issues on the ligand side: R chirality conformational flexibility N * * NHt Bu different efficacies H ACS 237 th National Meeting, March 24, 2009 p.22/26
[LIECE for β 2 AR] -6-8 fitting compounds G pred = 0.4584 E vdw calculated G -10 + 0.1780 E coul + 0.2047 G solv -12-14 RMSE q 2 = 0.6639 = 1.23 kcal/mol -14-12 -10-8 -6 experimental G ACS 237 th National Meeting, March 24, 2009 p.23/26
[Looks nice, but... ] -6 3 2-8 fitting compounds 6 hits 4 G pred = 0.4584 E vdw calculated G -10 5 6 + 0.1780 E coul + 0.2047 G solv -12-14 1 RMSE q 2 = 0.6639 = 2.24 kcal/mol -14-12 -10-8 -6 experimental G ACS 237 th National Meeting, March 24, 2009 p.24/26
[AA combined model for β 1 AR and β 2 AR] -6 3 2 calculated G -8-10 -12-14 fitting compounds 5 6 1 4 6 hits G pred = 0.3806 E vdw + 0.1424 E coul + 0.1576 G solv q 2 = 0.5542 RMSE = 1.40 kcal/mol RMSE = 1.75 kcal/mol -14-12 -10-8 -6 experimental G ACS 237 th National Meeting, March 24, 2009 p.25/26
[Some thoughts at the end] Docking to the β 2 -adrenergic receptor worked nicely...... we get a high hit rate (24 %) with one potent binder.... we see two previously unexplored scaffolds.... some of it is due to the bias in present-day libraries. LIECE models for GPCRs are difficult because...... the receptors are rather flexible and there are at least two distinct states (activated/inactivated) rigid receptor.... most molecules are chiral and flexible. We get reasonably predictive models, however. VdW is still the most important contribution, but electrostatics is similar. ACS 237 th National Meeting, March 24, 2009 p.26/26