On January 5, the National Medical Products Administration (NMPA) issued an announcement on the application of the "M7(R2): Assessment and Control of DNA-Reactive (Mutagenic) Impurities in Drugs to Limit Potential Carcinogenic Risks" (hereinafter referred to as the M7(R2) Guidelines).
The M7(R2) guidelines focus on DNA-reactive substances that can directly cause DNA damage at low levels, leading to DNA mutations and therefore cancer. Bacterial remutation of a substance can be assessed based on a review of the available literature and/or a computerized toxicology assessment.
In this article, we will introduce the relevant considerations for the application of the M7(R2) guidelines from the perspective of the computer toxicology method (i.e., the (quantitative) structure-response relationship ((Q)SAR)).
Common toxicity databases
Generally, we can obtain impurity carcinogenicity and mutagenicity data through database and literature search, and the common toxicity database is shown in Table 1.
Table 1 Common toxicity database
However, in many cases, the impurities are emerging substances, the basic information of the substances is unknown, and the relevant toxicological data are difficult to find in databases or literature, so it is necessary to use the QSAR method to predict and analyze the toxicity of impurities.
Two complementary QSAR prediction methods
The M7(R2) guideline states that the QSAR methodology is used for toxicity assessment to predict the outcome of bacterial mutagenicity assays. Two complementary QSAR forecasting methods should be used for forecasting. One method should be based on expert knowledge rules, and the other should be based on statistics.
These forecasting methods used in the QSAR model should follow the general validation principles developed by the Organisation for Economic Co-operation and Development (OECD).
✍️ Expert knowledge rules approach
Use established rules to link the structure of the substance to known toxicity. Most of the predictions used in the expert knowledge-based rules approach are derived from published literature, regulatory bodies, or specific sources, and then these rules are artificially imported into the software.
Based on these rules, the method of warning structure identification is used to identify potentially toxic structural fragments in the structure of matter. Some software can also indicate the likelihood of toxicity by alerting the structure to mechanistic information.
✍️ Statistical methods
Mathematical models are used to determine the properties inherent in the structure of a substance and to make probabilistic predictions about the likelihood of toxic effects through comparison with the structure of other substances. These models are often built on large data sets, and when people enter the structure of a substance, the software automatically compares it with the data set and gives a probability value of the probability of toxicity occurring. For statistical methods, the methods used to predict toxicity endpoints are different in different models.
Whether it is an expert rule method or a statistical method, the QSAR model used needs to comply with the validation guidelines set by the OECD for the construction and use of QSAR models:
- clear toxicity endpoints;
- explicit algorithms;
- identified model application domains;
- have appropriate goodness-of-fit, robustness, and predictive ability;
- Explain the mechanism.
According to the requirements of the M7(R2) guidelines, the following needs to be considered in the QSAR model prediction:
Click image for full size
In order to facilitate the verification and use of the model by other personnel or regulatory agencies, model developers usually create a QSAR Model Reporting Format (QMRF) according to a unified template, which is used to summarize and report the key information of the QSAR model, including the relevant model toxicity endpoints, algorithms, application domains, and validity required by the QSAR model validation principles formulated by the OECD.
The European Union Reference Laboratory for Alternative Testing for Animal Testing (EURL ECVAM) of the European Union Joint Research Centre (JRC) has compiled the QMRF files they collected into a QSAR model database, including 154 models of toxicity endpoints such as physicochemical, toxicology, ecotoxicology, and environmental behavior, which can be considered as valid models that can be validated for use for regulatory purposes.
Common QSAR software
At present, there are many models that can be used to predict bacterial revert mutagenicity, and if applicants use the models in the literature or developed by themselves to make predictions, they need to submit relevant materials such as QMRF reports to the regulatory authorities for model validation. Of course, in most cases, applicants will choose the developed QSAR software on the market to make predictions directly, and the common QSAR software is shown in Table 2.
Table 2 Common QSAR software
These software are very user-friendly both for applicants and for regulatory review. Users only need to enter the structure of the substance, and they can obtain information such as the prediction results of the relevant toxicity endpoints, whether the target substance is in the application domain, or whether the prediction results are reliable. The QMRF files of QSAR models in many software can also be queried from the software documentation file or the JRC QMRF database.
Some software will also set up special prediction modules and report templates according to regulatory requirements, such as Leadscope, which provides users with the prediction option of ICH M7, which includes two models: expert knowledge rules and statistical rules, and the generated reports also include impurity classification based on the prediction results.
Although there are many types of QSAR software and the operation page is becoming more and more friendly, not every software model can be applied to all substances, and everyone needs to pay attention to choosing the right and appropriate prediction software in the process of use.
Many people simply think that "commercial software" is better than "free software", and from a scientific point of view, whether it is commercial software or free software, the model is verified to be valid according to the OECD guidelines. When actually forecasting, it is not possible to judge which is better by "price", but needs to be comprehensively analyzed in combination with the specific target substance and the forecast results. For example, whether the toxicity endpoint of the model is bacterial mutagenesis, whether the prediction model covers the two methods of expert knowledge rules and statistical rules, whether the target substance is within the application domain of the model, and whether the information contained in the prediction results can meet the requirements of supervision and risk assessment.
Sometimes different software will produce inconsistent or even opposite results, so we need to consider more information synthesis such as QSAR software, toxic mode of action, training set, and similars.
It should be noted that screenshots of prediction results or software reports generated by the software cannot be directly used for drug registration, and more detailed QSAR reports need to be submitted for review by regulatory agencies, including:
- material information (name, structure of matter);
- software information (name, version, database);
- toxicity endpoints;
- test data;
- QSAR prediction results;
- application domain information;
- ICH M7 classification;
- Relevant expert notes and supporting documents, etc.
summary
The QSAR prediction toxicity endpoint under the M7(R2) guideline is clear, and the method requirements are relatively clear, but due to the inconsistency of the results between the two methods in the actual prediction, it is often necessary to combine more information and expert knowledge to analyze the final results.
In addition, the results generated by the software cannot be directly submitted for the QSAR report, so the relevant information needs to be summarized and displayed in more detail and clarity for the regulator to evaluate.
Computational toxicology training recommended
In order to promote the development of the discipline of environmental computational toxicology in mainland China and cultivate professionals in the field of computational toxicology, Dalian Key Laboratory of Chemical Risk Prevention and Control and Pollution Prevention and Control Technology (Dalian University of Technology) and Hangzhou Yile Standard Technology Co., Ltd. jointly held the "4th Environmental Computational Toxicology Technical Workshop" in Yangzhou, Jiangsu Province, on April 27-28, 2024.
- Chemicals Registration QSAR Forecasting;
- Pharmaceutical ICH M7 QSAR analysis;
- Pesticide registration QSAR;
- QSAR modeling;
- QSAR software.
This article is translated and organized by Yireh, please indicate the source for reprinting!
Eel will keep you updated with more compliance information!
If you have any questions about the analysis of pharmaceutical ICH M7 QSAR, please feel free to discuss them in the comment area