ICH M7 – Worked Examples using the new Nexus workflow

ICH M7 – Worked Examples using the new
Nexus workflow
43rd ICGM – New Orleans
Scott McDonald
Director of Member Services
[email protected]
Agenda
• The new ICH M7 Classification workflow in Nexus
• The workflow in practise
• Expert review under ICH M7
• Worked examples
ICH M7 Classification Workflow
• New feature in Nexus assists users undertaking expert
review by presenting relevant information for ICH M7
• Inputting an API with its impurities allows:
• Derek and Sarah Predictions
• Return of mutagenicity and carcinogenicity data
• Calculated classification for ICH M7
• User can interact with and modify classification
• Reports can be generated which include a classification
card for each impurity
What is an ICH M7 Class?
What’s Changed?
What’s Changed?
What’s Changed?
ICH M7 Classifications in Nexus
Active
Overall
Ames
Inactive
Other
Active
Class 1
Other
Class 2
Inactive
Class 5
Overall
Carc.
Overall
Carc.
Overall
Carc.
Active
Inconclusive
Inactive/
Other
Class 5
Inactive
Class 5
The user is
presented with
classifications
which require
expert review
No
Class 3
Yes
Non-mutagenic
Class 4
Active/Other
Derek
Alert
No
Alert
Alert in
API?
Sarah
Yes
Mutagenic
Inconclusive
Positive
Class 3
Negative
Class 5
OoD/Equiv.
Inconclusive
Overall carcinogenicity and Ames data
• Carcinogenicity
 Carcinogenicity Potency Database

http://toxnet.nlm.nih.gov/cpdb/
 In-house Vitic database if applicable
• Mutagenicity
 Lhasa curated Ames data

Dataset of 10,695 compounds
 In-house Vitic database if applicable
Addendum to ICH M7
• M7(R1) Addendum: Assessment and Control of DNA
Reactive (Mutagenic) Impurities in Pharmaceuticals to
Limit Potential Carcinogenic Risk
 http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/
Guidelines/Multidisciplinary/M7/M7_Addendum_Step_2.pdf
• Acceptable limits have been derived for a set of common
chemicals considered to be mutagens and carcinogens
• AIs automatically reported to users for exact matches
Addendum to ICH M7
• M7(R1) Addendum: Assessment and Control of DNA
Reactive (Mutagenic) Impurities in Pharmaceuticals to
Limit Potential Carcinogenic Risk
 http://www.ich.org/fileadmin/Public_Web_Site/ICH_Products/
Guidelines/Multidisciplinary/M7/M7_Addendum_Step_2.pdf
• Acceptable limits have been derived for a set of common
chemicals considered to be mutagens and carcinogens
• AIs automatically reported to users for exact matches
ICH M7 Workflow – SDF Input
ICH M7 Workflow – API Selection
Select the API from the inputted
file and map whether it is a known
mutagen or carcinogen
ICH M7 – Other data sources
The option to link to an in-house
Vitic database or import other
QSAR results is available
ICH M7 Workflow Results Screen
ICH M7 Workflow – DX and SX Predictions
ICH M7 Workflow – DX and SX Predictions
ICH M7 Class is calculated from data inputs
Each impurity is classified
according to the data available
in addition to predictions
Users can also input
experimental results for
mutagenicity or carcinogenicity
Source of prepopulated data
Carcinogenicity data comes from
the CPDB which gives activity in
single-cell or multi-cell assays
Source of prepopulated data
Carcinogenicity data comes from
the CPDB which gives activity in
single-cell or multi-cell assays
Lhasa Ames data comes from
the Lhasa curated data
ICH M7 Workflow – User Inputs
Details of the Derek and Sarah
Predictions for each impurity
can be found here
Users can specify any additional
Ames, Carcinogenicity or QSAR
results and provide comments
for reporting
ICH M7 – User defined classification
Users can change the ICH M7 class manually according to their expert
assessment. This is indicated by an asterisk in the main results screen
ICH M7 Reports
ICH M7 Reports
ICH M7 Classification Workflow Summary
• Assists expert review of impurities
• Supplements Derek and Sarah predictions with public
and in-house data
• Calculates ICH M7 classification based on predictions,
data and user inputs
• Expert review comments can be added and classification
cards can be generated
Expert review
Regulatory Toxicology and Pharmacology
2015, 73, 367-377
ICH M7 Workflow – Worked Example
O
API
S
Cl
O
N
N
N
N
N
O
Impurity 1
O
S
O
N
Cl
O
N
N
Impurity 2
O
O
Impurity 3
Impurity 4
Cl
B
S
O
Cl
Impurity 5
F
O
O
Impurity 6
Impurity 7
O
F
–
F
Impurity 8
Disclaimer:
This is not intended to be a realistic situation. These compounds have
been chosen to demonstrate the ICH M7 workflow functionality in Nexus
and to illustrate some interesting cases for expert review.
All expert review is the opinion of Lhasa Limited only
ICH M7 Workflow
ICH M7 Workflow
ICH M7 Workflow
• Derek predicts plausible
• Alert 354 fired for aromatic amines
• Clear (mechanistic explanation) and close examples support
the positive prediction.
• Sarah predicts positive
• Four positive hypotheses
• Supported by close, relevant examples
• Conclusion
• Positive
• Class 3
Derek: Inactive
Sarah: 4 Hypotheses (all negative) with closely related examples
• Derek predicts inactive
• No alerts fired
• No features of concern
• Sarah predicts negative
• Four negative hypotheses
• Phenyl amine hypothesis has closely related relevant examples
• Conclusion
• Negative
• Class 5
Derek:
1 Alert (Plausible)
Sarah: 1 Hypothesis (Negative) with no related examples
• Derek predicts plausible
• Fires alert 842 (Triazoline)
• Plausible mechanism described
• Sarah Predicts negative
• Single hypothesis doesn’t cover the concerning feature
• In domain
• Conclusion
• Derek fires an alert with closely related examples. Sarah
has triazoline in training set but cannot support a hypothesis.
Sarah prediction is not strong enough to overrule Derek.
• Positive
• Class 3
Derek:
Inactive
Sarah: 2 Hypotheses (11% Positive). Amine unlikely to be mutagenic
• Derek predicts negative
• Query structure matches exclusion pattern for alert 354
• Sarah predicts positive but with a low confidence
• Related examples are negative or could operate through a
different mechanism
• Conclusion
• Weak positive result can be overturned on the basis that
there is no mechanistic reason for the molecule to be
mutagenic
• Negative
• Class 5
Derek:
Inactive
Sarah: No Hypotheses (Equivocal). Even range of activities of examples
• Derek predicts inactive
• Dialkyl sulphates fire alert 27 but it specifically excludes
monoalkyl sulphates
• Sarah predicts equivocal
• Supporting examples show that dialkyl sulphates are positive
but monoalkyl sulphates are negative
• Conclusion
• Overrule Sarah as most relevant examples in Sarah are
negative, Derek alert comments provide support and
mechanistically monoalkyl sulphates are not nucleophilic
• Negative
• Class 5
Derek:
Alert 027 Fires
But – alert commentary suggests that longer
chain alkyl chlorides may not be Ames positive:
Sarah: Two Hypotheses (Negative). Closely related examples are negative
• Derek predicts plausible
• Alert advises that alkyl chlorides are weakly positive and this
is lost with longer chain analogies
• Specifically note that n-butyl chloride is negative
• Sarah predicts negative
• Close analogues are all negative
• Conclusion
• There is some uncertainty here
• There is a good basis for making a negative call
• The closest example in Sarah is a good analogue for read across.
However would want to dig more deeply into the data for it to ensure the
study was 5 strain and done correctly (Vitic provides this)
• May have to test
Sarah: Two Hypotheses (Negative). Closely related examples are negative
Derek:
Alert 315 Fires (Equivocal)
But there may be a dependency on the solvent:
Do carboxylic/sulfonic acid halides really present a mutagenic and carcinogenic risk as impurities
in final drug products?
Amberg, A., (2015). Organic Process Research & Development, 2015,
doi:10.1021/acs.oprd.5b00106
• Derek predicts equivocal
• Alert 315 fires. Comments note that the outcome is
dependent on the solvent used.
• Sarah predicts negative
• 12 examples contain SO2Cl
• One positive example is an aromatic nitro compound
• Conclusion:
• Solvent dependent. However fires same alert as API
• Negative
• Class 4
Derek:
Alert 746 Fires
The original query molecule is excluded from the alert due to the
deactivation of 2,6 disubstitution:
Sarah:
Out of Domain for both compounds.
The alert written in Derek is based on proprietary data which cannot
be displayed in Sarah without revealing the structure.
Publication of arylboronic acids shows similar SAR
• Derek predicts negative with unclassified features
• Trifluoroborate is known to be active
• But is deactivated by di-ortho substituents (alert 746)
• This is covered by the alert but the data is proprietary so cannot
be used with respect to unclassified features
• Sarah gives an out of domain prediction
• No relevant training compounds are provided
• Conclusion
• Proprietary compounds in Derek fill the gap and provide the basis of an
argument
• However the out of domain in Sarah and the unclassified feature in Derek
create a very high bar which means regulators are likely to want to see
multiple disubstituted examples with negative Ames data
• May have to test
Application
ICH M7 Classification Report
Summary
• Automatic and transparent ICH M7 workflow implemented
•
•
•
•
•
Expert rule-based predictions (Derek Nexus)
Statistical-based predictions (Sarah Nexus)
CPDB data and Lhasa curated Ames data
In-house Vitic data (where available)
M7(R1) Addendum
• Allows modification of results during expert review
• In-house data
• Other (Q)SAR data
• User comments
Questions?
THANK YOU
ADDITIONAL EXPLANATION
SLIDES
Composition of Lhasa Curated Ames data
• Public data:
• Feng, Vitic NTP, Helma, ISSSTY, FDA CFSAN, Kirkland,
Benchmark, MPDB, Bursi, Vitic Ames Summary, DX
Mutagenicity Examples, Acid Halide dataset
• Proprietary data:
• 3 companies
• Structures curated and standardised. Duplicates
removed
• 10,695 compounds
• 4,690 +ve, 4,883 –ve, 418 inconclusive, 28 equivocal, 676
conflicted
How is data combined?