Mirabilis: Towards the prediction of purge factors A tool for predicting

Mirabilis: Towards the prediction of purge
factors
A tool for predicting the purging of
(potentially) mutagenic impurities during
synthesis
43rd ICGM, March 2016
New Orleans
Chris Barber
[email protected]
Benefits of a purge argument under ICH M7
Risk of potential
mutagenic
impurity
Ames
Ames
or 2 in
silicoTest
models
-ve
Treat as a nonmutagen
Present a purge
argument for
absence…
+ve
Control & monitor
in final API


Risk mitigated
A recap on the control options in M7
Option
Details
1
• Test impurity in API and show below threshold level
• Periodic testing may be possible if can show sufficient
consistency otherwise routinely measure
2
• Test in precursor (or reagent..) to show below threshold level
3
• Test in precursor (or reagent..) with acceptance ABOVE threshold
level when also supported by evidence that final impurity levels
are below threshold following subsequent process steps
4
• Demonstrate understanding of process and consequent purge
sufficiently to not need any analytical testing
A cross-industry consortium…
• 11 companies sponsoring development of s/ware and
science
• Product release scheduled for end of 2016
• 2016 plans
• Agree standard approach to calculating purge factors
• Identify gaps in knowledge
• Prioritise work for collecting specific data
• Provide data to support a prediction
• Test / use the software prototypes
• Engage with regulators
Introducing Mirabilis
• Software to support estimating & expert-review of purge values
• Web-based to support collaborative (internal) use
• Support a consistent industry-standardised approach
• Provide supporting commentary and data for expert review
• Generates reports suitable for regulatory submission
A theoretical approach to estimating purge factors
2013, 17, 221
Physicochemical Parameters
Reactivity
Solubility
Volatility
Ionisability
Physical Processes – chromatography
Purge Factor
Highly Reactive = 100
Moderately reactive = 10
Low Reactivity / un-reactive = 1
Freely Soluble = 10
Moderately soluble = 3
Sparingly Soluble = 1
Boiling point >200C below that of the reaction/ process solvent = 10
Boiling point +/− 100C that of the reaction/ process solvent = 3
Boiling point >200C above that of the reaction/ process solvent = 1
Ionisation potential of GI significantly different to that of the desired product
Chromatography – GI elutes prior to desired product = 100
Chromatography – GI elutes after desired product = 10
Others evaluated on an individual basis
Scope of science needed
chromatography
Scavenger resins
Physical
processes
Recrystallisation
Reactivity
1
Mirabilis
Solubility
Precipitation
Liquid-liquid
extractions
Solid-liquid
extractions
Evaporation
Reaction grid - origins
• Internal tool from AZ to assign a reactivity purge factor
• Summarised expert knowledge of impurity reactivity under
common reaction conditions
8
Reaction grid - matching reaction + impurity class
Impurity Class
Reaction types
10
• Values agreed by expert elicitation
• High level factors – 1, 10, 100
• Conservative by design
Mirabilis – supporting expert assessment of purge
2015
2016
•
•
•
Expert elicitation of purge factors
Standardised ‘best practise’
Scientifically accepted approach
•
Data to support an expert decision
Knowledge alert
Machine-learnt
support
Supporting
kinetic studies
Proposed PMI / MI Purge Factor Decision Tree
(Roland Brown, Vinny Antonucci, Mike Urquhart)
Impurity requires management as PMI or MI
Establish PMI / MI strategy based upon comparison of Predicted purge factor
(Mirabilis) vs Required purge factor calculated from TTC or PDE requirements
> 1000x
ICH M7 Option 4
Data collection not
required
Include predicted
purge factors in
submission
> 100x
ICH M7 Option 4
> 10x
Potential M7 Option 4
Collect experimental data on purge properties
(solubility, reactivity, etc.) to support scientific
rationale.
Measure purge factors, including
trace analyses as required, to
support scientific rationale.
Include predicted purge factors in submission for
developmental API route(s). Additionally, include
supporting experimental data on purge properties
in submission for commercial API route.
Include predicted and measured
purge factors in submission.
Typically, more detailed datasets are
expected for commercial vs.
developmental API routes
ICH M7 Option 1,2,3
Analytical testing and/or specification(s)
required at SMs, Intermediates, or API,
including trace analyses (as required).
If measured purge factor
is insufficient, then ICH M7
Option 4 is not justified
What supporting information is needed?
• The amount of supporting information needed is
dependent upon the size of the overall purge
• Two key users
• Submitting (process) chemist
• Regulatory reviewer
What supporting information is needed?
• For the process chemist…
• Why that purge value?
• Mechanistic explanation, literature support, examples
• Will it hold up in my specific case?
• Are there exceptions?
• What is the impact of my conditions?
• Where is there uncertainty (risk)?
• What information could I provide to reduce this?
• Has the approach been validated; is it conservative?
• How should I present this information?
• What experimental details are needed?
Knowledge Alert
• Summary
• Brief synopsis
• Overall purge value
• Range / confidence
• Additional information to support / provide confidence
• Mechanistic rationale & expert assessment
• Impact of key parameters
•
•
•
•
•
•
•
Specific substrate
Specific reagents
Impact of solvent
Time
No. of equivalents
Temperature
Potential competing / alternative reactions
Machine mined approach to reaction purge
Apply filters:
Reaction database
1) Transformation class
2) Functional group
Dataset of similar reactions
• Mining a library of ~1 million patent reactions
• Still in research
• Aim to present supporting examples where the impurity is
• stable to these conditions
• reactive to these conditions
• Identify relevant examples to underpin an expert decision
• Support expert commentary
• Sensitivity of purge to changes
Using kinetic studies to support a purge value
2015, 19, 1517
• Measure rates of loss of ‘impurity’ whilst changing temp
• Can derive rate constant (k) and activation energy (Ea)
Undertaking experiments to fill knowledge gaps
17
Objectively assessing the approach
• Starting to undertake ‘benchmarking’ within the consortium
• “How well do the purge values match experimental results?”
• Identification of gaps (in process or knowledge)
• A peer-reviewed publication to help adoption of the approach
• Defining what detail is needed to support a submission
• Produce some guidance as to what to include
• cf Establishing best practise in the application of expert review
of mutagenicity under ICH M7. Reg Tox and Pharm 2015, 73, 367
• …..not about software – but about the science!
Software
•
•
Impurities highlighted
“Tramlines” track impurities
through synthetic route
Next steps
• Mirabilis interim releases being evaluated by consortium
• Product release scheduled for end 2016
• Collaborative publications
• To ensure peer-reviewed and accepted approach
• Initial scientific focus on reaction-based purge
• Other mechanisms for purge under initial investigation
• Solid/Liquid - liquid extraction, solubility, volatility…
Questions
• Acknowledgements
• Lhasa consortia members
• Dr Andy Teasdale (AZ)
• Lhasa staff
•
•
•
•
Liz Covey-Crump
Martin Ott
Susanne Stalford
Sam Webb
A theoretical approach to estimating purge factors
Teasdale..Risk assessment of genotoxic impurities in new chemical entities: Strategies to demonstrate control.
Organic Process Research and Development, 2013, 17(2), 221
ICH M7 guidelines allow for in silico purge arguments
What supporting information is needed?
• For the reviewer…
• Why that purge value?
• Can I be confident in the knowledge behind this?
• Access to an explanation, literature support, examples
• Will it hold up in this specific case?
• Are there exceptions?
• What is the impact of varying the conditions?
• What is the risk in accepting this prediction?
• Has the approach been validated; is it conservative?
• Is there anything that could reduce this risk?
• Has the approach been applied consistently?
• Software should ‘standardise’ the application