Pharmaceutical Industry and FDA Preparing for ICH M7 Implementation

The pharmaceutical industry is concerned about the cost involved in setting up complex computational systems required for evaluating potential genotoxic impurities in drug substances and drug products under ICH M7. The guidance goes into effect in January 2016 for drugs in clinical development and in July 2017 for new and generic drug applications as well as drugs with post-approval changes.

The pharmaceutical industry and FDA are gearing up for implementing the International Conference on Harmonization’s M7 guideline on genotoxic impurities.

The guideline goes into effect on Jan. 1 for drugs in clinical development and in July 2017 for new and generic drugs and for post-approval changes.

Pharmaceutical industry officials said that one of the more challenging aspects of implementing the guideline is having two separate computational systems for assessing a drug’s genotoxicity.

ICH M7 was adopted in June 2014, and is slowly being phased in and implemented in the US and the EU. The focus of the guideline is on DNA reactive substances that have a potential to directly cause DNA damage when present at low levels, leading to mutations and therefore potentially causing cancer.

M7 offers guidance on how to use computational tools for predicting bacterial mutagenicity outcomes based on established knowledge.

The guidance also outlines an approach to identifying, assessing and controlling mutagenic impurities in drug substances and drug products.

The guideline introduces the concept of the staged Threshold of Toxicological Concern (TTC) for acceptable impurity levels, which can be used during development and commercial manufacture.

For assessing risk, ICH focuses on impurities that exceed a 1.5 microgram /day TTC, which corresponds to a theoretical one in 100,000 excess lifetime risk of cancer.

According to the ICH M7 business plan, the document was meant to fill a void and address in greater detail regulatory expectations for evaluating and controlling genotoxic impurities.

The current ICH guidelines on impurity evaluation under Q3A and Q3B provide guidance on how to identify genotoxic impurities but give no guidance on acceptable levels.

Genotoxic impurities can arise in unexpected situations. In 2008, genotoxicity concerns prompted a recall and suspension of authority for Roche to market an HIV treatment, Viracept (nelfinavir mesylate) in the EU.

The problems came to light after contaminated batches reached the market in March 2007 and patients began reporting that the tablets smelled strange. Upon investigation, Roche determined that in the last few months of 2006 and in early 2007, some batches of the active substance had become contaminated with a known genotoxic substance, ethyl mesylate, also knowns as ethyl methanesulfonate (Also see "EMEA Clarifies Ambiguities in Genotoxicity Guideline" - Pink Sheet, 1 Jan, 2008.).

The contamination occurred after a holding tank was cleaned using ethanol. Before the tank was dry it was filled with methane sulfonic acid, a chemical used to make Viracept. The remaining ethanol reacted in the holding tank with the methane sulfonic acid to form very high levels of ethyl mesylate, which wound up in some finished batches.

ICH drills down to known carcinogens

To help manufacturers decide where to focus their efforts at impurity control, the ICH in June 2015 drafted an addendum to M7 that lists the 15 chemicals commonly used in pharmaceutical manufacture that are considered to be carcinogens and mutagens and should be used with caution.

The draft addendum was released for public consultation on June 9.

For each chemical, ICH lists the acceptable intakes or the permissible daily exposures, the potential for human exposure, and information on its mutagenicity/genotoxicity and carcinogenicity, the regulatory and published limits and how the acceptable limits were calculated.

Some of the listed chemicals are acrylonitrile, benzyl chloride, ethyl chloride, glycidol, hydrazine and methyl chloride.

Benefits of ICH M7

Andrew Teasdale, principal scientist and chair of the genotoxic advisory group at AstraZeneca, said one of the main benefits of ICH M7 is the risk-based approach that allows manufacturers to design a control strategy to accommodate their processes and products.

The guidance said that the control strategy should be consistent with the risk management principles in ICH Q9.

A robust control strategy, said the guidance, should minimize the need for end-product testing by shifting controls further upstream.

Teasdale said that “historically control has been focused upon developing complex [analytical] methods even though in many cases there is compelling evidence to conclude an impurity would be removed in the downstream. ICH M7 provides greater flexibility through a series of control options to evaluate such risk based on the physical properties of a mutagenic impurity relating to process conditions.”

At the Generic Pharmaceutical Association’s fall technical conference on Nov. 4, Stephen Miller, FDA’s CMC-Lead with the Office of New Drug Products in the agency’s Office of Pharmaceutical Quality, outlined the four options for manufacturers for controlling impurities:

• Option 1 is to monitor the impurity in the drug substance if the acceptance level is at or below the TTC;
• Option 2 is to monitor the impurity in the starting material, intermediate or in-process control if the acceptance criteria is at or below the acceptance limit;
• Option 3 is to monitor the impurity in the starting material, intermediate or in-process control if the acceptance criteria is above the TTC, with demonstrated understanding of fate and purge and associated process controls; and
• Option 4 is to design robust process controls to reduce the risk of impurity level above the TTC to negligible.

Computational testing cited as challenge

Yet pharmaceutical industry officials said that a major challenge will be having the computational systems in place and having the necessary expertise to ensure the accuracy of the results.

The guidance expects manufacturers to use QSAR (quantitative structure activity relationship), or a computational toxicology assessment, to predict the outcome of a bacterial mutagenicity assay known as the Ames test.

The guidance states that “two QSAR prediction methodologies that complement each other should be applied. One methodology should be expert rule-based and the second methodology should be statistical-based.”

The absence of structural alerts from these two methods will be sufficient to conclude that the impurity is of no mutagenic concern.

The guidance further recommends the use of experts to provide additional supporting evidence on the results or to reconcile conflicting outcomes.

High cost cited

Steve Baertschi, president of Baertschi Consulting, said that the guidance imposes high costs on manufacturers, not only in having the two systems, but in resolving different computational results.

He said that the industry pushed for the use of one system instead of two since two systems might yield too many results that are false positive and can incorrectly assume too many drugs are genotoxic when they are not.

Teasdale concurred that while a “vast majority of companies have prepared for this, the biggest challenge is using two systems with conflicting predictions.”

Teasdale cited one study showing that running the two programs yielded disparate results 25% of the time.

“How do you weigh the system so that you’re not getting a lot of false positives?

Teasdale said that to reconcile conflicting results, it is important to “understand the relative strengths and weaknesses of the differing predictions. In the case of the rule-based system, what rule was triggered and how well validated is the rule? In the case of the statistical prediction, how has the molecule been classified, are the compounds it has based the prediction on closely related? Or could it be that the compound set it has been grouped with is not closely related and that the feature identified is not relevant in the context of the molecule. This unfortunately needs expertise.”

Teasdale said that this new expectation has prompted “a lot of uncertainty as a result of this and you will see a lot more scrutiny of in silico systems.”

How to get the best result

Naomi Kruhlak, the lead of the chemical informatics program at FDA who spoke at the GPhA fall technical conference on Nov. 4, described in more detail FDA’s expectations for computational modeling under ICH M7.

FDA’s Chemical Informatics Program is an applied regulatory research group that creates chemical structure-linked toxicological and clinical effect databases and evaluates data mining and QSAR software.

The group also develops toxicological and clinical effect prediction models through collaborations with software companies. The team provides QSAR evaluations for drugs and degradants to FDA drug safety reviewers.

Kruhlak confirmed that FDA will be looking to see that new and generic drug applications as well as post-approval changes each have two computational models in place to assess genotoxicity: the expert rule based model and a statistical based model. Both should be able to predict Ames mutagenicity.

Kruhlak said that conducting an empirical Ames assay for every potential and known impurity is not feasible or justified because of the high cost involved in using the vitro test, especially for assessing the multitude of impurities that may be present in a particular drug substance.

The Ames test is a widely used method that relies on bacteria to test whether a given chemical can cause mutations in the DNA of the test organism. It is a biological assay to assess the mutagenic potential of chemical compounds.

She said QSAR or in silico models are better at handling large volumes of impurities. These models should instead be used to predict the results of Ames tests.

She said that QSAR modeling identifies correlations between chemical structural features and biological activity.

Kruhlak said that QSAR statistically derived models use partial least squares regression analysis and support vector machines. Expert-rule based models, however, capture knowledge from proprietary data.

Kruhlak said that FDA uses the following software packages in doing QSAR analyses for drug compounds:

• For its statistical models it uses CASE Utra/MC4PC developed by MultiCASE and Model Applier Statistical Models developed by Leadscope.
• For its expert-rules based models it uses Derek Nexus developed by Lhasa Limited.

She pointed out that FDA has a research collaboration agreement in place with these software vendors.

To ensure that manufacturers are getting the best and most accurate results and to avoid discrepancies between tests, manufacturers should do the following:

• Check that the impurity structure is correct;
• Check for experimental Ames data;
• Generate predictions for the impurity structure;
• Determine the credibility of the reasoning for the predictions; and
• Check for experimental data for chemicals with similar structures.

Agency training reviewers on QSAR

FDA reviewer David Green said that in anticipation of ICH M7 adoption, the agency is training reviewers on assessing impurities in drug master files (DMFs).

Green said that DMF reviewers will assess the mutagenicity of APIs in three steps.

First, the reviewer will calculate the TTC for an API based on the Maximum Daily Dose of the reference ANDA and the daily intake per duration of treatment. The reviewer will be adding a TTC limit to the table for the referenced ANDA. A specified limit of 1.5 micrograms a day daily intake is used for a drug product that has a lifetime duration of use per M7.

Next the reviewer will be looking at the firm’s synthesis to determine if there are any chemical compounds that contain structural alerts. Green said that FDA is training staff on QSAR assessments to ascertain whether the drug tests positive for genotoxicity or is negative.

To assist reviewers in identifying compounds that are carcinogenic, FDA will be using a chemical Carcinogenesis Research information System (CCRIS) database which is a publicly accessible database maintained by the National Cancer Institute.

The databank is organized by chemical record, and contains carcinogenicity, tumor promotion, tumor inhibition and mutagenicity test results derived from the scanning of primary journals.

“All that matters is if study data is available. We do not read or interpret any of the studies. If data is available a deficiency is triggered.”

The last step is that reviewers will determine the appropriate response once structural alerts have been identified.

Green said that FDA expects industry will be doing these QSAR assessments once ICH M7 goes into effect.

He said that “it has been an incredible experience. Great collaborations have come out of this. It has been a great training experience for our reviewers. But in reality we want to get to the point where we are not doing this. We want to get back to our traditional system of reviewing.”