Memories and the future:

1
Memories and the future From experimental to in
silico physical chemistry
Han van de Waterbeemd AstraZeneca, DMPK Alderley
Park, Macclesfield, UK

2
Overview

• Why physchem data?
• Wet screening (in vitro)
• Web screening (in silico)
• Future developments

3
Medchem evolution

• lt1980 target affinity/binding using intuition and
  experience
• gt1980 structure-based design
• gt1995 drug/lead filters such as rule of five
• gt2000 property-based design
• gt2005 in silico/in vitro (in combo) approaches

protein crystallography
attrition analyses
physchem/DMPK considerations
HT property screening
4
Key ADME questions

• Drugability
• Attrition
• Appropriate PK
• Target affinity
• vs
• ADME

Carlson and Segall, Curr.Drug Disc. 34-36 (2002)
5
ADMET screening strategy

• Biopharmaceutical (physchem) profiling
• Pharmacokinetics
• Metabolism
• Early toxicology
• In vitro wet screening
• In silico web screening
• In combo
• In cerebro

6
Wet screening (in vitro measurement)
7
Han very early days

• Leiden (PhD)
• log P vs log k
• Are rate constants of partitioning useful in
  QSAR?

8
Han early days

• Lausanne (post-doc with Bernard Testa)
• pKa - Apple III, IBM PC
• log kHPLC - first attempts to HT
• log P aV L
• hydrophobicity polarity
• size hydrogen bonding

9
Han early days

• Roche (Molecular Properties Group)
• pKa (GLpKa101, John Comer, Colin Peake)
• log kHPLC
• log Papp (artificial membranes pre-PAMPA,
• Gian Camenisch)
• PAMPA (Manfred Kansy)
• PSA polar surface area
• Van de Waterbeemd and Kansy, Chimia 46 (1992)
  299-303

10
Han more recent days

• Pfizer (automated ADME screening)
• log D - 96 well plates
• log S
• PAMPA
• Pfizer (in silico ADME)

11
Lessons learned

• Calculation goes faster
• Computed data often good enough
• No need to measure too much
• In silico for virtual compounds
• But, good quality experimental data are needed to
  build robust models

12
Kinetic vs equilibrium
Caco-2 PAMPA (cm/s)
log P log D
Water Membrane Water
log k (w/o) a log P b log (bP1)
c Kubinyi, 1978 Van de Waterbeemd et al, 1981
13
Permeability lipophilicity scale
Caco-2
Permeability?
PAMPA
Absorption
log Ddodecane
log Doct
Lipophilicity (log P/D)
In reality sigmoidal relationships
14
Web screening(in silico prediction)
15
Why in silico ?

• Lots of compounds (libraries, parallel
  synthesis)
• Lots of data (in vitro ADME/physchem screening)
• Screening is expensive
• In vitro models not always predictive for in
  vivo
• (e.g. Caco-2, PAMPA)
• In silico models to complement and/or replace
• in vitro/in vivo
• Only option for virtual compounds
• Guide in decision-making

16
In silico

• Sound QSAR and molecular modeling methods/tools
  are available
• Commercial and in-house solutions for physchem
  and ADME screening data
• Modeling and simulation for human PK
• Confidence is growing

17
In silico solubility ?

• Artificial GI fluid and buffered water are
  models
• for solubility in human GI
• In silico models of these surrogate conditions
  are
• therefore a model of a model
• What is predictive power of such solubility
  models?
• We dont take solid state properties into
  account!

Human GI Artificial GI Aqueous buffer
r2 0.7 r2 0.7 r20.5
18
In silico PAMPA and Caco-2 ?

• Caco-2 and PAMPA are models for oral absorption
• In silico models of Caco-2 and PAMPA are
• therefore a model of a model
• What is predictive power of such models?

in vivo in vitro
in silico Human A Caco-2/PAMPA Caco-2/P
AMPA models
r2 0.7 r2 0.7 r20.5
model x model random
19
C. Lupfert, A. Reichel, Chem.Biodivers. 2
(2005) 1462-1486
good
uncertain
poor
20
Unravelling the processes
ADME
Bioavailability Liver first-pass
metabolism Absorption Transporters Gut-wall
metabolism Permeability Lipophilicity Mole
cular size Molecular shape Flexibility H
ydrogen bonding Solubility
In vitro and in silico screens?
21
Prediction of A
Design
Clinical Candidate
Lead Optimization
Development
Lead Profiling
ACAT PBPK ppb pKa logD Caco-2 PAMPA Peff Vmax,
Km Solubility 78
Single Descriptors MWlt500 0ltClogPlt4 0ltlogDlt3 PSAlt
140A2 80-90
QSAR Structural Descriptors 75
R-o-5 MWlt500 ClogPlt5 HBAlt10 HBDlt5 gt60
Population 78 10
A human measured 76 15 !!
22
Towards prediction paradise?
Solubility A F
CL Vd
Dose
T1/2
log D
ADME Activity Toxicity
IC50
Tox
Van de Waterbeemd and Gifford, Nature Revs. Drug
Disc. 2 (2003) 192-204
23
Future developments

• Property-based design is best practise
• In combo approach established in drug discovery
• Further progress in silico QSAR technology
• New ADME/T world
• Pharma industry fully adapts in silico approach
  to design, screening, and optimisation

24
In vitro in silico in combo

• Integration of experimental and computational
• technologies
• Reducing cost of screening
• Maximising data information

Yu and Adedoyin, Drug Disc.Today 8, 852-861
(2003) Dickins and Van de Waterbeemd, DDT
Biosilico, 2, 38-45 (2004)
25
ADME technologies - autoQSAR

• Automated model building and updating

in combo
Data
Build in silico model
Update in silico model
in vitro priorities
J.Cartmell et al, J.Comp.-Aid.Mol.Des. 19 (2005)
821-833
26
In vitro logP conferences

• Great series of meetings,
• Excellent Proceedings
• Lausanne 1995, 2000
• Zurich 2004, 2009

27
In silico EuroQSAR conferences
QSAR has its attraction
28
References
Volume 5 ADME-Tox Approaches (B. Testa and H.
van de Waterbeemd), Elsevier, November 2006
29
Thanks
et bon appetit
30
(No Transcript)