1) The document summarizes an evaluation of methods for predicting hydrogen bond strength, including Peter Kenny's accurate but slow quantum mechanics (QM) approach and the faster but less accurate Cresset molecular modeling software.
2) Applying Kenny's QM approach to 9 kinase ligand cores, excellent correlations with potency were found except for two outliers, while Cresset captured trends but not substituent effects.
3) A combination of Kenny's QM predictions with Cresset screening was proposed to effectively optimize ligand hydrogen bonding for potency, with each method's strengths mitigating the other's weaknesses.
2. Background
• Motivation: answer recurrent chemist question to predict
basicity of HB acceptor/acidity of HB donor, in order to correlate
with affinity
• Quick overview of literature highlighted the methods of Peter
Kenny (then at AstraZeneca) as simple and accurate for both
HB acceptors and donors strength assessment
method rigourously tested and sufficiently described to quickly reproduce
2
3. Peter Kenny’s method
• Uses projected electrostatic field rather than two-body
calculation
avoids need to correct for Basis Set Superposition Error
• From HF/6-31G* minimized geometries
HB acceptors (J. Chem. Soc. 1994, 2, 199-202)
calculate electrostatic potential minimum along acceptor lone pair axis
HF/6-31G* calculation in GAUSSIAN
HB donors (J. Chem. Inf. Model. 2009, 49, 1234-1244)
calculate electrostatic potential value 0.55 Å away from donor H in D-H
direction
B3LYP/6-31+G** is most predictive (GAUSSIAN)
3
5. Implementing Kenny’s approach
From HF/6-31G* minimized geometries (PC GAMESS, aka Firefly)
• HB acceptors
calculate electrostatic potential grid around acceptor atom
±2 Å around acceptor, 0.05 Å grid spacing (80^3=512 000 grid points!)
HF/6-31G* calculation in PC GAMESS
• HB donors
calculate electrostatic potential value 0.55 Å away from donor H in D-H
direction
B3LYP/6-31+G** calculation in PC GAMESS
5
6. Experimental data
• HB strength quantified by measuring association constants
for donor-acceptor complexes in nonpolar solvent
• All experimental data taken from Abraham et al, J. Chem.
Soc. Perkin Trans. 2 1989, 10, 1355-1375
Acceptors (UV) Donors (IR)
logK logK
HO NO2 (CH3CCl3) N
(CH3CCl3)
O
6
7. HB donors: B3LYP/6-31+G** V(0.55) results
logK exp vs. V (0.55)
3.5
3
• Identical results to Kenny’s
2.5
only QM softwares differed
2
1.5
in-house B3LYP/6-31+G**
1
Linear (in-house B3LYP/6-31+G**)
0.5
R² = 0.9311
0
0.3 0.31 0.32 0.33 0.34 0.35 0.36 0.37 0.38
7
9. In-house HF/6-31G* GAMESS grid protocol
Pearson r -0.97
Spearman rho -0.96
Vmin_HF(kJ/mol)
Marginally worse results than published by Kenny
9
10. HF/6-31G* from RM1 geometry
Pearson r -0.96
Spearman rho -0.94
Vmin_HF(kJ/mol)1
Much faster calculation, marginally worse correlations
10
11. Cresset on Kenny JChemSoc94 dataset
Pearson r -0.53
Spearman rho -0.64
Cresset_fieldsize
Weak correlation of field size with logK
Cresset seems unable to predict substituent effects
11
13. New Cresset force field effect
Force field FF2 FF3
Pearson r -0.53 -0.55
Cresset FF3
Spearman rho -0.64 -0.64 Cresset FF2
Cresset_fieldsize
Systematic shift to larger Field Sizes observed with FF3
Substituent effect remains unaccounted for
13
14. Example: kinase ligand scaffold hopping
• 9 topologically similar scaffolds synthesized in kinase project so far
• X-ray structure of ligand-kinase complex GK+1
available for main compound series GK+2
GK
mainly 1 point interaction with hinge GK+3 NH
is potency related to HB acceptor strength? GK+3 Ar
GK+4
Core
II
• Rather favorable case in Cresset FieldStere GK+5
but FieldStere only returns overall similarity score
no field point near donor nitrogen in some cases
14
15. Example scaffold hopping
QM results
• Modified Kenny method applied on all 9 core acceptor N, with no sidechain present on cores
excellent pIC50-Vmin correlation except for 2 cores in THP subseries
in valinol subseries, core III is the only outlier
IC50 for core V in THP subseries might be wrong? Maybe a substituent effect?
literature search indicates core III is basic enough to protonate at assay pH (pKa=6.9)!
Core
Core II Core
I I
Core
IX Core
pIC50
Core Core II
Core
III VII VI Core
III Core
IV
O X
N
X Core X
N
X
Core
X
Core
X Ar V O X
Core
X Ar
V
N X N X
THP valinol
Vmin (kJ/mol) Vmin (kJ/mol)
15
16. Example scaffold hopping
Cresset results
• No exact correlation of field size with pIC50 observed
• Cresset still partially picks up main trend and ranking
Core Core Core
I II I
Core
pIC50
Core Core
Core IX
VII Core II
III
III Core
IV
Core Core
VI Core
N N V
O X
Core
X
V X
Core
X
X X Ar O X X Ar
N X N X
Field Size Field Size
16
17. Summary
• Peter Kenny’s highly predictive QM HB strength prediction method identified and
implemented for both HB acceptors and donors
slow calculation for large systems
I/O preparation currently labor intensive low throuput
• Direct comparison of QM results with Cresset Field Sizes possible for HB acceptors
Cresset captures general trend but fails to incorporate substituent effects
could future XED force field releases take such effects into account?
• Combination of Cresset and the QM approach could prove effective for ligand HB
profile optimization
QM HB assessment method surprisingly predictive of ligand potency
Cresset best alternative for high throughput searches, particularly when substituents are
maintained (e.g. scaffold hopping in FieldStere)
17
18. Acknowledgements
• Cresset
Mark Mackey
Martin Slater
Tim Cheeseright
Andy Vinter
• Galapagos computational chemistry group
Pieter Stouten
Cornel Catana
Nicolas Triballeau
Miriam Lopez-Ramos
18