Transcript Presentation

Investigation of CDK2 Inhibitor Potency using
Electrostatic Potential Complementarity and the
Fragment Molecular Orbital Method
Creating high-value drug discovery
innovation alliances
Evotec AG, 5th Joint Sheffield Conference on Chemoinformatics, July 2010
Overview
 Molecular Shape and Electrostatic Considerations in Ligand Binding
 Case Study: Cyclin-Dependent Kinase 2 (CDK2)
 Understanding Complex Interactions During H2L/F2L/LO
 The Fragment Molecular Orbital (FMO) Method
 Application of FMO Calculations
PAGE 2
Classical Lock and Key Problem
“Everything should be made as simple as possible, but not simpler.” - Einstein
 What is required to effectively describe protein::ligand interactions?
 Ligand and receptor features to consider:
 Shape
 Charge and electrostatic potential
 Dynamics
Ligand “Keys”
PAGE 3
Receptor “Lock”
Scoring Ligand Shape and Electrostatic Potential
Tanimoto Coefficient
How similar are these?
A
B
 Tanimoto coefficient is widely used to compare chemical
similarities
 Gaussian Tanimoto compares ligand shapes in 3D
IA
OAB
Tanimoto = 1 = A and B are identical
 Electrostatic Tanimoto (TES) is calculated in the similar manner
as for Gaussian Tanimoto but an electrostatic field overlap is
used instead of volume overlap1
 Implemented in MOE2 and is high throughput (10,000s cmpds)
PAGE 4
IB
1) Jennings and Tennant, J. Chem. Inf. Model. 47, 1829-1838, 2007
2) MOE (The Molecular Operating Environment) http://www.chemcomp.com
Ligand-Based Shape and Electrostatic Potential
Calculations
 Gaussian Tanimoto is a fast shape comparison application, based on the idea that molecules have
similar shape if their volumes overlap well and any volume mismatch is a measure of dissimilarity
 Used as a virtual screening tool which can rapidly identify potentially active compounds with a
similar shape to a known hit or lead compound
 TES score is sensitive to subtle changes in ligand electrostatics
 Semi-empirical atomic charges using AM1-BCC is recommended 1,2
AM1-BCC is parameterized for good correlation with HF 6-31G* charges 3
Gauss = 1.00
TES = 1.00
PAGE 5
Gauss = 0.97
TES = 0.40
Gauss = 0.99
TES = 0.64
1)
Tsai et al., Bioorg. Med. Chem. Lett., 18, 3509-3512, 2008
2)
Jennings and Tennant, J. Chem. Inf. Model. 47, 1829-1838, 2007
3)
Balyl, et al., J. Comput. Chem., 132-146, 132, 2000
Gauss = 0.97
TES = 0.54
Case Study: CDK2
Astex AT7519 (CDK2 inhibitor) as an example
 Can ligands be effectively represented and compared using measures of shape and
electrostatics?
 Case example taken from the literature.1,2 CDK2 fragment-based screened identified a number of
hits. 28 ligands taken from this work were examined.
 Pharmacological inhibitors of cyclin-dependent kinases (CDKs) are currently being evaluated for
therapeutic use against cancer and neurodegenerative disorders amongst many other diseases.
1) P. G. Wyatt et al., J. Med. Chem. 51, 4986-4999, 2008
PAGE 6
2) M. Congreve et al., J. Med. Chem. 51, 3661-3680, 2008
Gaussian Tanimoto
Based on CDK2-Bound Alignment
PAGE 7
Fragment Hit
Clinical Candidate
2VU3
32
31
30
2VTT
2VTQ
27
26
25
24
2VTP
21
2VTO
20
19
2VTN
17
16
2VTL
13
2VTI
2VTS
2VTR
2VTJ
9
2VTA
Hit
Gaussian
Tanimoto
Coefficient
2VTM
3nM
2VU3
32
31
30
2VTT
2VTQ
27
26
25
24
2VTP
2VTO
21
20
19
2VTN
17
16
2VTL
2VTI
13
2VTS
2VTR
2VTJ
9
2VTM
2VTH
2VTA
2VTH
AT7519
Absolute Difference in pIC50
AT7519
0.0
3nM
1.5
4.0
Absolute
Difference
In pIC50
Hit
PAGE 8
Fragment Hit
Clinical Candidate
Electrostatic Tanimoto - TES
Based on CDK2-Bound Alignment
PAGE 9
Fragment Hit
Clinical Candidate
2VU3
32
31
30
2VTT
2VTQ
27
26
25
24
2VTP
21
2VTO
20
19
2VTN
17
16
2VTL
13
2VTI
2VTS
2VTR
2VTJ
9
2VTA
Hit
Electrostatic
Tanimoto
Coefficient
2VTM
3nM
2VU3
32
31
30
2VTT
2VTQ
27
26
25
24
2VTP
2VTO
21
20
19
2VTN
17
16
2VTL
2VTI
13
2VTS
2VTR
2VTJ
9
2VTM
2VTH
2VTA
2VTH
AT7519
Case Study: CDK2
Optimisation of Shape and Electrostatics
2VU3 AT7519, 47nM
Which interactions are the most important?
What happens when you have a complicated interaction that requires better understanding?
PAGE 10
1)
Chau, P-L., and Dean, P.M., J. Comput.-Aided Mol. Design., 8, 513-525, 1994
Understanding Complex Interactions during
H2L/F2L/LO
Which interactions are the most important?
What happens when you have a complicated interaction that requires better understanding?
Multiple equivalent
binding modes
Interactions not represented in
docking/MM forcefields
“Defragmentation” of large ligands
to determine group efficiency
More complex methods required – e.g., free energy and/or quantum mechanical calculations
PAGE 11
Fragment Molecular Orbital (FMO) Method
Method and throughput
Fragmentation of peptide
PIE (Pair Interaction Energy)
 Full quantum computation of protein::ligand
complexes has been practically impossible
until recently due to extremely large resources
required for computing
 The fragment molecular orbital method1 (FMO)
was proposed by K. Kitaura and co-workers
– Highly suitable for calculation of large (biological)
systems in parallel computing environment2,3
– Implemented in GAMESS QM suite
Calculations for systems with 200300 atoms are routinely ran at Evotec
(~10/day ) using MP2 / 6-31G* ,
6-31G(3df,3pd) for Cl and S
1) Kitara et al., Chem. Phys. Lett., 313, 701-706, 1999
PAGE 12
2) Komeiji et al., Comput. Biol. Chem., 28, 155-161, 2004
3) Fedorov et al., J. Comput. Chem., 25, 872-880, 2004
– PIEDA4,5 (Pair interaction energy decomposition
analysis) provides detailed ligand/protein
interaction information
4)
5)
Fedorov, D. G., and Kitaura, K., J. Comput. Chem., 28, 222-237, 2007
Nakano et al., Chem. Phys. Lett., 351, 475-480, 2002
The Cl-p Interaction in a Protein::Ligand Complex
 Calculated interaction energy is 2-3 kcal/mol
depending on the chloro species
HF/6-311G++(3df,2pd)
MP-2/6-311G++(3df,2pd)
MP-2/cc-PVTZ
3
Energy (kcal/mol)
 Cl-p interaction is an attractive interaction,
where the major source of attraction is the
dispersion force
0
 Optimal distance is ca. 3.6 Å
 HF interaction is repulsive
 Electron correlation method, such as MP2,
needed to probe the interaction accurately
-3
3.0
4.0
Distance (Å)
5.0
 For example – serine protease inhibitor series1
PAGE 13
1)
Shi, Y., et al., J. Med. Chem. 51, 7541-7551, 2008
2)
Imai et al., Protein Science, 16, 1229, 2008
Application of FMO Calculations
PIE and PIEDA (Facio)1,2 and PIO (Pair Interacting Orbitals)3,4
PIEDA diagram
Exchange
Electrostatic
CT & Mixed
Dispersion
PDB: 1WCC
IC50 = 350mM
-48.40kcal/mol
PIO analysis
Glu81
Phe80
Phe82
His84
PAGE 14
Leu134
1)
2)
Suenaga, M., J. Comput. Chem. Jpn., 4 (1), 25-32, 2005
Suenaga, M., J. Comput. Chem. Jpn., 7 (1), 33-53, 2008
3)
4)
Fujimoto, H.; Koga, N.; Fukui, K. J. Am. Chem. Soc. 1981, 103, 7452.
Fujimoto, H.; Yamasaki, T.; Mizutani, H.; Koga, N. J. Am. Chem. Soc. 1985,
107, 6157.
Application of FMO to FBDD
Astex AT7519 (CDK2 inhibitor) as an example
PDB: 2VTA
IC50 = 185mM
LE = 0.57
PDB: 1WCC
PDB: 2VTN
PDB: 2VTO
IC50 = 0.85mM
IC50 = 0.14mM
LE = 0.44
LE = 0.39
Development
discontinued
AT7519
IC50 = 0.047mM
IC50 = 350mM
LE = 0.40
LE < 0.51
LE = -RT ln(IC50)/heavy atom cout
P. G. Wyatt et al., J. Med. Chem. 2008, 51, 4986-4999
PAGE 15
M. Congreve et al., J. Med. Chem. 2008, 51, 3661-3680
PDB: 2VTP
IC50 = 0.003mM
LE = 0.45
Application of FMO to FBDD
PIEDA and PIO (Pair Interacting Orbitals)
PIEDA diagram
Exchange
PIO analysis
Electrostatic
CT & Mixed
Dispersion
PDB: 1WCC
IC50 = 350mM
-48.40 kcal/mol
Phe80
Phe82
 PIEDA identifies the nature of
ligand/protein interactions
– H-bond, VDW, p-p etc
 PIO analysis used to visualize and provide
3D information on the interactions
– Interacting orbitals, direction of charge transfer
(vacant-occupied MO interaction)
His84
PAGE 16
Leu134
= Direction of CT
Application of FMO to FBDD
1WCC core modifications: FMO virtual SAR
1
6
2
3
7
8
4
5
9
10
Removal of the
chlorine detrimental to
the fragment binding
12
 Medium throughput (up to few 100s input)
FMO analysis can be rapidly carried out to
answer SAR questions
 The technique is highly effective for prioritizing
the initial fragment expansion directions or
optimization for larger ligands
PAGE 17
13
DE
11
IC50 = 350mM
IC50 = 7mM
DE = Sum PIE – Sum PIE (1WCC fragment)
Application of FMO to FBDD
Tracking the PDB: 2VTA development path by FMO analysis
repulsive
attractive
PDB: 2VTA
IC50 = 185mM
-41.71 kcal/mol
Glu81
PDB: 2VTN
PDB: 2VTO
IC50 = 0.85mM
IC50 = 0.14mM
-61.24 kcal/mol
-64.81kcal/mol
Phe80 Lys33-Asp145
Salt bridge
Phe82
Leu83
His84
PAGE 18
Leu134
Val18
Val18
R1
O
H
N
O
FMO Heatmap Analysis
R2
N
H
N
N
H
Sum of the PIE
R1
IC50 (uM)
R2
Cl
0.063
NH
*
*
Energy
Kcal/mol
32
-96.55
31
-100.76
30
-94.99
27
-81.81
26
-79.53
25
-71.99
24
-70.77
20
-60.92
F
OMe
0.052
NH
*
*
F
F
F
NH
0.910
*
*
F
NH2
F
0.038
*
*
F
OH
F
0.019
*
*
F
F
0.012
*
*
F
F
0.025
*
*
F
1.600
PAGE 19
F
*
*
Binding Enthalpy Comparison to MM Methods
Known Binding Modes from X-ray Structures
FMO
Affinity dG
PAGE 20
GB IV
Alpha HB
London dG
ASE
Method
R2
FMO
0.68
GB IV
0.10
London dG
0.47
Affinity dG
0.18
Alpha HB
0.42
ASE
0.67
Summary

Gaussian Tanimoto can be used to assess similarly shaped compounds to actives

TES can be used to assess which docking pose is the best during VS

TES used to identify suboptimal interactions for further development

FMO can be used to identify which binding pose from a VS has the optimal
interactions with a receptor

FMO can be used to indentify subtle changes required to improve binding enthalpy

Molecular interactions reflected in the binding enthalpy are critical variables in lead
optimisation
PAGE 21
Current and Future Work

PAGE 22
Currently assessing protein::ligand complementarity methods
Current and Future Work

PBSA treatment of free energy of solvation can be used to rationalize overestimated
enthalpic terms in FMO

Free energy of binding QSAR models are highly predictive

Need for improved treatment of
PAGE 23

Solvation

Entropy

Salts and Metals
Acknowledgements
Evotec CADD Group
Richard Law
Osamu Ichihara
Alex Heifetz
Chemical Computing Group
MOE svl Scripts
Andrew Henry
Simon Grimshaw
Guido Kirsten
Kristina Grabowski
FMO Developers
Dmitri Fedorov
PAGE 24
Your contact:
Dr. Mike Mazanetz
Senior Scientist, Computational Chemistry
+44 (0) 1235 44 1342
[email protected]
Appendices
PAGE 26
Estimation of Binding Free Energies
Entropy – Enthalpy Compensation
Aim of free energy calculation in a VS campaign is to rank-order molecules
such that if a selection of high-ranking compounds is obtained and analysed,
it is likely that some will show activity.
However, compound activity is likely to span about 5 log orders in
magnitude, which equates to free energy range of around
5.5 kcal/mol at 37°C.
PAGE 27
IC50 µM
DG
0.01
-11.34
0.1
-9.92
1
-8.51
10
-7.09
100
-5.67
R: universal gas constant ≈ 1.986 cal/Kmol
T: temperature 310 K
1)
Williams, D., et al., Angew. Chem. Int. Ed., 43, 6596-6916, 2004
Estimation of Binding Free Energies
Basic equations and two thermodynamic terms
 Relationship between to Ki (IC50) and the free enegy of binding
DG = -RT lnKD
 Free energy of ligand binding consists of two thermodynamic terms
DG = DH – TDS
•
•
Binding enthalpy
Notoriously difficult to optimize due to strict three dimensional requirements
Enthalpic improvement is often not reflected in better affinity, because of the associated
entropy-loss (desolvation)
Binding entropy
Dependent primarily on the hydrophobic effect and conformational entropy
Easier to optimize and less affected by compensating enthalpy changes
PAGE 28
Key SBDD Concepts

Entropy-enthalpy compensation phenomenon

Desolvation penalty (4-8 kcal/mol per polar group)

Origin of hydrophobic interaction (entropy-driven effect, re-organization of surface
water network)

Two terms contribute to the entropy of binding
Desolvation entropy (always favourable, about 25 cal/mol Å2 for a carbon atom)
Conformational entropy
•
Overcoming enthalpy/entropy compensation
Well placed H-bond can make a favourable enthalpic contribution of the order of -4 to -5 kcal/mol (1000 –
5000 fold increase in affinity)
Hydrogen bonds should be aimed at already structured regions of the protein
Try achieving multiple H-bonds for flexible residues – positive cooperativity
Be aware of the forced solvent exposure of hydrophobic groups
PAGE 29
Free Energy of Binding Thermodynamic Cycle
+
ΔGSolv, Ligand
ΔGBind, Solv
ΔGSolv, Receptor
+
ΔGSolv, Complex
ΔGBind, Vac
ΔGBind, Solv = ΔGBind, Vac + ΔGSolv,Complex  ( ΔGSolv,Ligand + ΔGSolv,Receptor )
ΔGSolv = ΔGElec, ε=80  ΔGElec, ε=1 + ΔGHydro
ΔGVac = ΔEMM  T·ΔS
PAGE 30
Estimation of Binding Free Energies
All 28 CDK2 ligands
ΔGBind, Solv = ΔGBind, Vac + ΔGSolv,Complex  ( ΔGSolv,Ligand + ΔGSolv,Receptor )
Actual pIC50
ΔGSolv = ΔGElec, ε=80  ΔGElec, ε=1 + ΔGHydro
ΔGVac = ΔEMM  T·ΔS
PBSA using a single
energy-minmized
structure
Number of
rotatable bonds
r2 = 0.81
q2 = 0.76
Predicted pIC50
FMO sum PIE
PAGE 31
1)
Rastelli G., et al., J. Comput. Chem., 31(4), 797-810, 2009