Transcript In silico methods: ADMET vs receptor affinity

In silico ADME/Tox in drug
design
“Bioinformatics IV”
(Computational Drug Discovery)
Wednesday 7 June 2006
CMBI, University of Nijmegen
Lars Ridder, Organon
What makes a good drug ?
• Good activity/selectivity on the right
target
BUT ALSO !!!
• Absorption
• Distribution
• Metabolism
• Excretion
• Toxicity
ADME/Tox
Reasons for drug failure in
Clinical Development (>80%)
ADME/Tox
Role of in silico ADME/Tox
Research
Development
$300m
$500m
4-5yrs (30%)
8-10yrs (70%)
Does the
compound work in
man?
Identify ADME/Tox problems
Failure rate over 80-90%
earlier in the process
(safety, efficacy)
More emphasis on ADME/Tox
properties in lead optimization
Market
In-house design cycle
Guide optimisation
based on in silico
models
Screening, hit-optimization, lead selection, lead optimization, SOPP, development
Validate/refine models
based on new
pharmacological data
Absorption/Distribution/Metabolism
Pharmacokinetic parameters
•
•
•
•
Oral bioavailability =
fraction of dose that enters
blood circulation (after 1st
pass metabolism in the liver)
Absorption = fraction of
dose that passes the gut wall
Clearance (CL) = amount of
blood cleared per time unit
Volume of distribution (Vd)
= (I.V.) Dose / Initial plasma
concentration
Absorption
MW < 500, non-polar
Most common route of drug absorption
Membrane permeation
Water
C1
Membrane
C2
Penetration rate = P x A x (C1-C2)
Depends on physicochemical
properties of drug, e.g. lipophilicity,
MW, hydrogen bonding, etc.
P = partition into membrane
A = effective surface area of membrane
C1-C2 = concentration gradient
Hydrogen bond donors and acceptors
H
O
O
H
H
H
H
O
O
H
R
H
O
H
O
H
Absorption requires desolvation, which becomes more
difficult with an increasing number of hydrogen bonds
The octanol/water model
water
O
O
OH
octanol
OH
log P  log
[ AH ]oct
[ AH ]wat
The octanol/water model
water
O
octanol
O
OH
OH
log P  log
[ AH ]oct
[ AH ]wat
[ AH ]oct  [ A ]oct
log D  log
[ AH ]wat  [ A ]wat
O
O
O
O
The octanol/water model
water
O
octanol
O
OH
OH
log P  log
[ AH ]oct
[ AH ]wat
[ AH ]oct  [ A ]oct
log D  log
[ AH ]wat  [ A ]wat
O
O
O
O
logD = logP - log(1 + 10pH-pKa)
logD = logP + log(1 - fionized)
LogD depends
on pH !
pH-range in GI tract
pH
(fed)
pH
(fasted)
3-7
1.4-2.1
5-6.5 6.5
6.5-8 6.5-8
5-8
ClogP
Calculating logP from structure:
• Fragmentation of solute molecule by identifying Isolating
Carbons (IC = not doubly or triply bonded to a hetero atom)
• Remaining fragments are characterized by topology and
“environment” (i.e. the type of IC’s bound to it)
• ClogP is a sum of (tabulated or estimated) contributions of all
fragments + isolating carbons + ”corrections”
• Where “corrections” are made for intramolecular polar, dipolar
and hydrogen bond interactions as well as electronic (aromatic)
interactions (modified Hammett approach)
ClogP - examples
O
OH
HO
Fragment
Value
6 x IC (arom) 0.78
Carboxy
-0.03
Hydroxy
-0.44
4 x Hydrogen 0.91
Electronic int. 0.34
ClogP
1.56
Exp. logP
1.58
ClogP - examples
O
OH
OH
Fragment
Value
6 x IC (arom) 0.78
Carboxy
-0.03
Hydroxy
-0.44
4 x Hydrogen 0.91
Electronic int. 0.34
H-bonding
0.63
ClogP
2.19
Exp. logP
2.26
ClogP vs. Caco-2
Caco-2 = in vitro assay to measure absorption rate
10
y = -1.4107x + 7.3791
R2 = 0.4115
9
8
clogP
7
6
5
4
3
2
1
0
0
0.5
1
1.5
log(caco-2)
2
2.5
3
Lipinski’s Rule-of-5
• Lipinski (1997) selected 2245 orally
active drugs from the World Drug Index
(WDI)
• Distribution analysis suggested that
poor absorption is more likely when:
– Mol. Weight > 500
– ClogP > 5
– Nr. of H-bond donors > 5
– Nr. of H-bond acceptors > 10
Correlation to in vivo (rat) absorption
In-house rules
based on:
ClogP
MW
H-bond donors
H-bond acceptors
But also:
•
•
Polar surface area
Nr. of rotatable bonds
These simple physicochemical properties largely
determine bioavailaility !
number of compounds
•
•
•
•
Bioavailability
120
100
80
60
< 30%
> 30%
40
20
0
Good
Moderate
Bad
Monika classification
Good =
no properties out of range
Medium = 1 property out of range
Bad =
> 1 property out of range
Pharmacokinetic modeling
•
•
•
•
PK-sim
Cloe
PKexpress
Gastroplus
Advanced Drug Delivery Reviews 50 (2001) S41–S67
Distribution
Most important organ:
The brain
• Drugs acting on the central
nervous system (CNS) must
cross the blood-brain barrier
(BBB)
• Peripheral drugs may be
required not to pass the BBB
to avoid CNS side effects
• Physicochemical properties
are important (again)
• Efflux by P-gp mediated active
transport
Metabolism/Excretion
Metabolic enzymes
Lipophylic metabolites
Cytochrome P450
e.g. Flavin monooxygenases
Dehydrogenases
Hydroxylation, dealkylation,
N-oxidation, epoxidation,
dehydrogenation, etc.
Phase I: (mostly) oxidation
Polar metabolites
+glutathione
+H2O
+glucuronate
+sulphate +acetate
Phase II: conjugation
Hydrophylic metabolites
+methyl
Contributions of Phase I and Phase II
enzymes to drug metabolism
ADH, alcohol dehydrogenase; ALDH, aldehyde dehydrogenase; CYP, cytochrome P450; DPD,
dihydropyrimidine dehydrogenase; NQO1, NADPH:quinone oxidoreductase or DT diaphorase;
COMT, catechol O-methyltransferase; GST, glutathione S-transferase; HMT, histamine methyltransferase;
NAT, N-acetyltransferase; STs, sulfotransferases; TPMT, thiopurine methyltransferase;
UGTs, uridine 59-triphosphate glucuronosyltransferases. [Evans (1999) Science 286: 487]
Cellular localisation of metabolic enzymes
• Endoplasmitic
reticulum (ER) of
intestinal- and liver
cells contain P450
• Cytosol contains
Phase II metabolic
enzymes
Xray structures of P450
• CYP 2C5 from rabbit was 1st
mammalian P450 to be
crystallized in 2000 *
• the substrate access channel is
likely to be buried in the
membrane
• Structures of most important
human CYPs (2C9, 3A4 and
2D6)
* [Williams et al. (2000) Mol. Cell 5:121]
Structure of P450
Substrate access
Heme = catalytic centre
Cytochrome P450 (CYP)
Reactive iron-oxo
intermediate:
“Compound 1”
In-house data: Compounds tend to
be very stable or very unstable
Lipophilicity is an important
factor in microsomal stability
Unstable Stable
10
20
30
40
50
60 70 80
T1/2 (mins)
90 100 110 120
70
70
60
60
50
50
Stable
Stable
40
40
Unstable
Unstable
30
30
20
20
10
10
99
77
55
44
66
88
ClogP
ClogD
33
22
11
00
-1
-1
0
0
-2
-2
10
10
-3
-3
compounds
compounds
(ClogD discriminated better
between stable and unstable
than ClogP)
160
140
120
100
80
60
40
20
0
0
-4
-4
In vitro measurement of metabolic
stability in microsomes = ER
membrane fraction of liver cells
Ncompounds
Phase 1 metabolism vs. lipophilicity
The Cytochrome P450 family
CYP3A4
Family
Subfamily
Individual
protein
Isoenzyme specificity
• Various isoenzymes
have different but
overlapping substrate
specificities
• (CR indicates flatness
of molecule)
[Lewis (2002) Drug Disc. Today 7:918]
Individual variation in P450 activity
• Genetic polymorphism
– Defective gene: “poor metabolizer” (e.g. CYP 2C19: >20% in
Asians)
– Gene multiplication: “extensive metabolizer” (e.g. CYP 2D6)
• Enzyme induction
-> Increased protein synthesis
• Enzyme inhibition
• Enzyme activation
(CYP 3A4)
Drug-drug
interactions !
Avoid drugs being metabolized via a single route !
Occurrence of major polymorphisms
Ingelman-Sundberg et al. (1999) Trends in Pharm. Sciences20:342
Impact of P450 polymorphism
Ingelman-Sundberg et al. (1999) Trends in Pharm. Sciences20:342
Metabolite identification
• It is often important to identify the
metabolites formed by P450’s:
– Identification of toxic metabolites
– Knowledge about the site of metabolism
can be used to design metabolically more
stable compounds (e.g. by
modifying/blocking the labile site in a
molecule)
P450 metabolism
• Which metabolites are formed by
P450’s depends on:
– If and how (i.e. in what orientation) a
compound is bound to the active sites of
individual CYP’s
– The chemical reactivity of various sites of a
molecule towards CYP catalyzed
mechanisms
In silico methods
• Binding in CYP active site
– Docking
– Pharmacophore
• Reactivity of ligand sites
– QM methods
• Metabolism rules
– Expert knowledge
– Empirical scoring
Modeling Ligand binding to CYP2C19
by homology modeling and docking
Drug
Reaction
Amitriptyline
Diazepam
Flunitrazepam
Ifosfamide
Imipramine
Indomethacin
Lansoprazole
Methoxychlor
Moclobemide
Omeprazole
Pantoprazole
Phenobarbital
Phenytoin
Rabeprazole
R-mephobarbital
R-warfarin
Sertraline
Testosterone
Venlafaxine
Nirvanol
Clomipramine
Nordazepam
Nortriptyline
Trimipramine
Citalopram
Carisoprodol
S-mephenytoin
Fluoxetine
Hexobarbital
Tolbutamide
Cyclophosphamide
Desogestrel
Progesterone
N-deMe
N-deMe
N-deMe
C-hydrox, alifatic
N-deMe
O-deMe
C-hydrox, aromatic
O-deMe
C-hydrox, alifatic
O-deMe
O-deMe
C-hydrox, aromatic
C-hydrox, aromatic
O-deMe
C-hydrox, aromatic
C-hydrox, aromatic
N-deMe
ox, -OH to =O
O-deMe
C-hydrox, aromatic
N-deMe
C-hydrox, alifatic
N-deMe
N-deMe
N-deMe
N-deMe
C-hydrox, aromatic
N-deMe
C-hydrox, aromatic
C-hydrox, alifatic
C-hydrox, alifatic
C-hydrox, alifatic
C-hydrox, alifatic
Reactive
conformation
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
top ranked
found
found
found
found
found
found
found
not found
not found
not found
docking failed
docking failed
docking failed
Assessing chemical susceptibility towards
CYP metabolism based on QM calculations
Many CYP reactions begin with abstraction of aliphatic H•
sites of oxidation
Primary/benzylic
Benzylic/CI-methyl benzylic
Benzylic/O-demethylation
Benzylic/O-demethylation
Benzylic/secondary
1/2 Hexanol
1/2 Octanol
Benzylic/substituted
Benzylic/substituted
Benzylic/substituted
Gact
-4.39
-0.74
0.4
0.72
1.28
-1.85
-2.14
0.51
0
1.81
5
R2 = 0.86
4
calculated (AM1)
substrate
Ethylbenzene
a-Chloro-p-xylene
2-Methylanisole
4-Methylanisole
1,3-Diphenylpropane
Hexane
Octane
1-Phenyl-3-(4-F-phenyl)propane
1-Phenyl-3-(4-Me-phenyl)propane
1-Phenyl-3-(4-F3Me-phenyl)propane
3
O
2
1
0
 site2 
 RT ln 
 x1  H rad  x2  IPrad

 site1 
0
1
2
3
-1
experimental (kcal/mol)
Works for small molecules – for larger drug molecules a
combination of high level modeling and QM calculations will
ultimately result in more accurate predictions
4
5
Derivation of metabolic rules
• Example: rule for N-acetylation
R
NH2
H
N
R
O
[NH2:1] >> [N:1]C(=O)C
• Apply on training set of 7307 reactions
metabolites generated in total
metabolites match experimental product
probability
1223
122
122/1223 = 0.10
Refined rules for N-acetylation
Three more specific rules for N-acetylation
Arom
NH2
H
N
Arom
79 / 357 = 0.22
R
NH2
H
N
R
122/1223 = 0.10
Aliph
NH2
O
O
H
N
Aliph
33 / 417 = 0.08
Hetero
NH2
O
H
N
Hetero
10 / 88 = 0.11
O
Refined rules for N-demethylation
H
N
NH2
CH3
0.8
10/13 = 0.77
0.6
H
N
N
CH3
CH3
11/20=0.55
Probability
CH3
0.4
0.2
H
N
aliph
CH3
aliph
NH2
102/266 = 0.38
0
20
CH3
CH3
aliph
CH3
109/434 = 0.25
CH3
R
N
182/1052 = 0.17
H
N
N
R
22
23
24
 H (hydrogen abstraction)
H
N
N
aliph
21
R
R
CH3
NH
2/87 = 0.02
25
Current rule base at Organon
19
• 148 rules
• phase I and phase II
metabolism
• Probabilities range from
0.006 (glycination of aliphatic
carboxyls)
to 0.77
anilines)
(demethylation of methyl-
19
12
24
22
14
4
13
dealkylation
carbon oxidation
reductions
condensation
glucuronidation
8
13
hydroxylation
hetero oxidation
hydrolysis
other phase I
other phase II
Evaluation: Sulfadimidine
1
H
N
S
O
N
O
O
N
H
H
N
N
O
S
O
2
O
H2N
S
S
O
N
H
N
H
N
O
N
N
H
N
O
O
Sulfadimidine
H2 N
5
N
N
OH
N
3
H2N
O
S
O
4
H2N
N
H
O
N
Prediction(rank)
N
O
S
OH
N
N
H
OH
N
O
OH
Application: metabolic stability
Predicted
Rank 1
O
N
N
N
N
N
O
N
HO
N
N
N
N
O
O
-> Confirm experimentally
by mass spectroscopy
Med Chem optimisation:
increased metabolic stability
O
N
F
N
N
N
N
O
Toxicity
Systemic Toxicity
Organ Specific Toxicity
•
•
•
•
•
•
• Blood/Cardiovascular
Toxicity
• Hepatotoxicity
• Immunotoxicity
• Reproductive Toxicity
• Respiratory Toxicity
• Nephrotoxicity
• Neurotoxicity
• Dermal/Ocular Toxicity
Acute Toxicity
Subchronic Toxicity
Chronic Toxicity
Many endpoints
Genetic Toxicity
Many mechanisms
Carcinogenicity
Developmental
Toxicity
-> Tough
problem
• Photo toxicity
Prediction of toxicity
Biology
Activity
(Toxicity)
Rules/ToxQSAR
icophores
Statistics
Analytical
methods
Chemistry
Structure
Reaction mechanisms
• Expert or rule-based systems
• QSAR or “correlative” methods
Example expert system:
Derek
• 303 knowledge
based alerts or
toxicophores
• 35 tox. endpoints
• refs to literature
included
• Works well e.g. for
mutagenicity
Example expert system:
Derek output
LHASA PREDICTIONS
Carcinogenicity
Alert overview: 107 Aromatic amide
R1
N
O
R2 R2
R1 = C (aryl)
R2 = C, H
Cellular metabolism required for activity. The best evidence indicates that hydroxylamino compounds are proximate carcinogenic forms. The above
functional group can be converted to hydroxylamine by hydrolases, oxidases, or reductases endogenous to most tissues.
References:
General principles for evaluating the safety of compounds used in food-producing animals.
Title:
Author: Food and Drug Administration (FDA).
Source: Food and Drug Administration Report, 1986, III-7-III-17, July 1994 revision available at
"http://www.fda.gov/cvm/guidance/guideline3toc.html".
Locations:
Paracetamol !
Example QSAR method:
APA Acute toxicity model
• 37400 IP-mouse LD50 data
• Classification
– Knowledge from literature
– Properties identified form
decision trees
• QSAR based on fragments
• Overall R=0.8 for test-set
•
•
•
•
•
Absorption
Distribution
Metabolism
Excretion
Toxicity
Research
Guide optimisation
based on in silico
models
Screening, hit-optimization, lead selection, lead optimization, SOPP, development
Validate/refine models
based on new
pharmacological data
Development
Decrease failure rate !
Market