The Prediction of Chromosomal Effects and Their ... - Lhasa Limited

The Prediction of Chromosomal Effects
and Their Mechanism of Induction in
Derek for Windows
RV Williams 1 & M Hayashi 2
1 Lhasa Limited, Leeds, UK
2 National Institute of Health Sciences, Tokyo, Japan

Overview
• In silico prediction of chromosomal effects
• Development of a Derek for Windows model
• Common mechanisms
• Evaluation of predictive performance
• Conclusions

Chromosomal effects
• In vitro tests for chromosomal effects are part of
the established genotoxicity test battery
• Such tests are often low-throughput and
relatively expensive in comparison to the Ames
test
• Computer systems offer an alternative approach
for the identification of chemicals that may
induce chromosomal effects

N
H 2
13.0
Approaches to in silico toxicology
OH
26.5
8.2
36.1
S
O
O
58.4
Statistical model
O
O
14.7
Algorithm
QSAR
Multicase (MC4PC)
ADMEworks
Structure
Reactivity
Activity
Mechanism
Expert system rules
Metabolic data
Human expert
SAR
Derek for Windows

Derek for Windows
• Derek for Windows is a knowledge-based expert system
that predicts the toxicity of a compound from its
chemical structure
• Predictions are qualitative and based upon structural
alerts, rules and examples
• A broad range of toxicological endpoints are covered
� Carcinogenicity
� Genotoxicity
� Hepatotoxicity
� HERG channel inhibition
� Reproductive toxicity
� Skin sensitisation

Derek for Windows

Genotoxicity Coverage
• Genotoxicity coverage in Derek for Windows has
historically focussed on the prediction of Ames
test mutagenicity
• A project was therefore initiated to improve the
prediction of structural and numerical
chromosomal effects (“chromosome damage”)

Mechanism
related to existing
mutagenicity alert
Extend existing
mutagenicity alert
Project strategy
Chemical class causing
chromosome damage
Mechanism not
related to existing
mutagenicity alert
Develop new
chromosome
damage alert

Developing new alerts
In vitro chromosomal aberration data sets
Analyse
Common toxicophores
Prioritise
Ranked common toxicophores
Gather mechanistic evidence
and additional examples
Mechanism-based structural alerts

Chromosomal aberration data sets
• Sofuni data book
• CGX data set (Kirkland et al)
• Ishidate data set
• Vitic database
• Data set of marketed pharmaceuticals (Snyder
and Green, Snyder et al)

Toxicophore identification
• Toxicophores were identified primarily by
analysis of the Sofuni data book
• Two techniques were used during this process
� Visual analysis
� Computer analysis using ChemTK lite

Example case for toxicophore
identification
5-Substituted uracils

Toxicophore identification
• Visual analysis highlighted that the data sets
contained four 5-substituted uracils
• All four compounds gave positive results in the
in vitro chromosomal aberration test
• These were developed into a toxicophore,
comprised of the common structural features
allowing for some generalisation

Toxicophore identification
F
O N
HN
O
O
HN
O
O
O
N
O
N
F
1,3-Bis(2-tetrahydrofuranyl)
-5-fluorouracil
1-(2-Tetrahydrofuranyl)
-5-fluorouracil
HO
O
O
HN
O
O
O
N
N
H
F
F
OH
5-Fluorodeoxyuridine
5-Fluorouracil

Known active
O
HN
O
N
H
F
Toxicophore identification
O
N
O
N
R
Where R =
any heteroatom
Unknown activity
O
HN
O
N
H
Cl

Toxicophore identification - results
• 103 toxicophores have been identified using this
approach

Toxicophore development
• The next step was to prioritise the toxicophores,
using a weight of evidence approach
• Prioritised toxicophores were then further
developed using information in the published
literature including
� Mechanistic evidence
� Additional example compounds

Example case for toxicophore
development
5-Substituted uracils

Initial
toxicophore:
O
N
O
N
Toxicophore development
R
Where R =
any heteroatom
Additional
examples included:
O
N
O
HN
N
OH
O
O
F
OH
Capecitabine, a clastogenic
5-fluorocytidine
Underlying
mechanism:
5-Fluorouracils and 5fluorocytidines
inhibit
thymidylate synthetase

O
Alert features
Mechanism-based
structural alert for
5-fluoropyrimidines
N
NH 2
N
F
Based on the mechanism of action, the alert is restricted to 5-fluoropyrimidines
O
N
O
N
F

O
N
O
HN
N
OH
O
O
F
OH
O
Alert features
Mechanism-based
structural alert for
5-fluoropyrimidines
N
NH 2
N
F
The alert is also restricted to 5-fluoropyrimidines that can form or mimic nucleotides
O
N
O
N
F
O
O
N
O
O
N
F

Prediction for 5-fluorouracil

Results of toxicophore development
• To date, 50 of the 103 toxicophores have been
investigated
• Currently the Derek for Windows knowledge
base contains 63 alerts for chromosome damage
• From this work, several common mechanisms
for the induction of chromosomal effects that do
not involve direct damage to DNA have become
apparent

Why is mechanism important?
• Contributes to the definition of alert scope
• Aids determination of in vivo significance
• Provides transparency to predictions
• Is included as one of the OECD Principles for
(Q)SAR Validation designed to facilitate
regulatory acceptance of in silico predictions

Common mechanisms leading to
chromosomal effects

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation
Xanthines:
Inhibition of cell cycle checkpoint function
O
N
O
N
N
N
(Caffeine)
4-Hydroxystilbenes:
Inhibition of ribonucleotide reductase
HO
OH
OH
(Resveratrol)

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation
N
H 2
Thymine and derivatives:
Imbalance of the nucleotide pool
HN
O
N
HO
HO
N
N
O
O
N
H
NH
O
O
O
P
O
O
Na +
Na +
(Thymine)
Guanine and derivatives:
Imbalance of the nucleotide pool
(Sodium 5’-guanylate)

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation
HO
Vinca alkaloids:
Numerical chromosomal aberrations
N
H
O
O
O
N
N
OH
H
N
OH
H
O O
O
(Vinblastine)
Di- or tri-phenylethylenes:
Numerical chromosomal aberrations
OH
O
(Diethylstilbestrol)

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation
N +
N +
e-
O 2
Bipyridinium compounds:
N +
-
O 2
N +
Cl Cl
(Paraquat)
N N +

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation
Polynitrophenols:
Uncoupling of oxidative phosphorylation
O 2 N
OH
NO 2
(2,4-Dinitrophenol)

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation
Isothiazolinones:
Covalent interaction with proteins
O
NH
S
(Benzisothiazolinone)
N-Polyhaloalkylthio compounds:
Covalent interaction with proteins
O
O
N S
Cl
Cl
Cl
(Captan)

Chromosomal effect mechanisms
Disruption or inhibition of DNA synthesis or repair
Spindle function disruption
Generation of reactive oxygen species
Energy depletion
Thiol reactivity
Intercalation
Psoralens:
Disruption of DNA helix
O
O O O
Hydroxyanthraquinones:
Disruption of DNA helix
OH O OH
O
(5-Methoxypsoralen)
(Danthron)

Evaluation of predictive performance

How computer systems are evaluated
• Computer systems offer an alternative approach
for the prediction of chromosomal effects
• The performance of computer systems can be
evaluated using data sets of chemicals with
known activities
• Key measurements include sensitivity and
specificity

Evaluation of three computer systems
• Three computer systems have been developed
and evaluated by the NIHS
� Derek for Windows, ADMEworks and Multicase
• Chromosomal aberration data from the
Japanese database of existing chemicals was
used for the evaluation (GLP tests)
Positive 98
Negative 121
Total 219

Evaluation results
• The best results were obtained when the
computer systems were combined
• In this approach a majority verdict was used to
give an overall result
Derek for Windows
Multicase
Admeworks
Overall result
+
+
+
-
+
+
Positive
- -
- +
- -
Negative

Combined systems - detailed results
Chrom ab. test
Chrom ab. test
+
-
In silico
+ -
63 26
18 81
Applicability: 85.8% (188/219)

Combined systems - detailed results
Chrom ab. test
Chrom ab. test
+
-
In silico
+ -
63 26
18 81
The activity of 63/89 positive
compounds was correctly predicted

Combined systems - detailed results
Chrom ab. test
Chrom ab. test
+
-
In silico
+ -
63 26
18 81
The inactivity of 81/99 negative
compounds was correctly predicted

Combined systems - detailed results
Chrom ab. test
Chrom ab. test
+
-
In silico
+ -
63 26
18 81
Sensitivity
70.8 %
Specificity
81.8 %
Concordance
76.6 %
Applicability: 85.8% (188/219)

Conclusion
• This work has demonstrated that it is feasible to define
alerts for the prediction of chromosomal effects
• Derek for Windows now contains 63 alerts for the
chromosome damage endpoint and work is on-going
• The approach allows mechanistic insight and
demonstrates promising predictive performance
• When computer systems are used in combination,
predictive performance can be improved

• Lhasa Limited
� Kate Langton
� Carol Marchant
� Russ Naven
Acknowledgements
• National Institute of Health Sciences
� Akihiko Hirose
� Eiichi Kamata