is the clear recognition that proximity to the final API is a key factor in determining risk (i.e. purging capacity of the downstream process), that early stages will, in general, represent a lower risk of carryover than later stages. This point is returned to later in the guideline within the control options section and is again a key, new, aspect in the guideline.
In the context of synthetic impurities, mutagenic impurities can arise via three sources:
1 Mutagenic reagents used deliberately in the synthesis. Many of the common reagents used in the synthesis of the DS are mutagenic. The use of such reagents, i.e. methyl iodide, epichlorohydrin, etc., is effectively unavoidable; it is simply impractical to construct C⏤C and C⏤N bonds without the use of such reagents [21].
2 Mutagenic intermediates – often the use of a deliberately formed, highly reactive, intermediate is required – examples include tosylates, hydrazides, and epoxides; such an intermediate being deliberately utilized to effect an efficient synthesis.
3 Side reactions. Perhaps the most difficult to assess, those impurities formed as a result of predictable side reactions. Wherever possible this should be based on existing scientific knowledge. This has been drawn into sharp focus by events surrounding N‐nitrosamines. The risk of MIs arising from side reactions is the focus of Chapter 11.
2.2.6.2 Degradation Products
Unlike earlier guidelines, ICH M7 specifically addresses the issue of mutagenic degradants. Similar to DS impurities, the guideline makes clear the primacy of ICH Q3B [8] and the identification thresholds for the product. It also makes clear again the need to focus on degradants likely to be present in the final DP. While clearly helpful, this nevertheless does not define how the risk posed by degradants should be evaluated. The guideline provides some advice defining degradants in terms of actual degradants and probable degradants. Actual degradants are those observed to form over prolonged storage at ambient temperature, i.e. ICH long‐term accelerated storage conditions. Probable degradants are defined in terms of those observed under accelerated conditions, e.g. 40°/75%. Reference is made within the guideline to the use of knowledge derived from stress studies. The evaluation of mutagenic degradants is examined in detail by Baertschi et al. [22]. The use of stress studies to identify major degradative pathways and their associated primary degradants is underpinned by the established relationship between degradants formed under stress conditions to those seen under ambient conditions. Baertschi [23] demonstrated that degradants formed under ambient conditions were a contained “sub‐set” of those observed to form under stressed conditions, this is illustrated in Figure 2.2, where the relationship between idealized and realistic degradation knowledge landscapes is considered.
Figure 2.2 Interrelationship between degradant classes.
Hypothetical degradants arise from in silico and in cerebro assessments; potential degradants are observed as major degradation products in stress testing and accelerated stability testing; actual degradation products are those that arise under ICH long‐term (real‐time) storage conditions.
Another important consideration in the assessment of degradants is what is an appropriate identification threshold within the context of a stress study. The identification threshold defined within ICH Q3B [7] for a DP with a dose of >10 mg – 2 g is set at 0.2% or 2 mg. This is relative to a typical maximal level of degradation of typically 2% total degradation in DP. In the context of a stress study where degradation levels may well exceed 10% total, an adjustment factor of five to the identification threshold would seem pragmatic, raising this to a value of 1.0%. Using such a structured approach will ensure that impurities identified during a stress study are the primary degradants and hence commensurate with the focus of the guideline on degradants likely to be present in the final DP.
Overall in terms of the impurity assessment the impurities for consideration are reflected in Figure 2.3.
Figure 2.3 Potential sources of mutagenic impurities.
This topic is examined in detail in Chapter 14.
2.2.7 Hazard Assessment
The emphasis of the guideline (Section 6) now shifts and focuses on an assessment of the mutagenic potential of impurities identified in the preceding risk assessment. Such an assessment is typically made through the use of in silico SAR systems. The guideline defines the need to apply two (Q)SAR methodologies. One methodology should be expert rule based, and the other methodology should be statistical based; however, the guidance does not define which software packages are preferable; this decision is left to the end user. Importantly, it also highlights the need for an expert evaluation of the results.
The use of two methodologies throws up a number of different permutations; these include not only situations where predictions are conflictory in nature but also out of domain predictions. These arise where the molecule in question, or at least a significant proportion of the molecule, is not recognized by the training set of the in silico tool, and hence it cannot accurately predict its mutagenic potential. In such circumstances, expert evaluation is required to make an overall consensus prediction. Barber et al. [24] examine this in detail. They describe the various scenarios potentially encountered (Figure 2.4), examining how to address conflictory predictions as well as out of domain predictions. Barber et al. provide advice on how to challenge predictions made by both rule based and (Q)SAR systems as well as providing a series of examples that serve to provide effective practical illustration of the key points made within the paper.
Figure 2.4 Decision matrix when evaluating two in silico predictions.
In a related study, Green et al. [25] examined the relative predictive performances of popular commercial in silico systems. Using a data set of some 801 chemicals and pharmaceutical intermediates, they showed the overall accuracy of each of the systems was generally comparable, ranging from 68 to 73%; however, their studies showed significant differences in sensitivity of each system (i.e. how many Ames positive compounds are correctly identified) results varying between 48 and 68%. The studies did not, however, identify any stand out system or specific combination of rule based/(Q)SAR systems. Perhaps the most significant finding of the studies was the number of contradictory predictions observed when two different methodologies were applied, i.e. those where one system predicted positive and the other did not or the statistical models were not able to make a prediction. Over one‐third of all the compounds in this 801 compound data set were seen to give a conflictory prediction. The authors concluded there is clearly a need for expert opinion to be applied to determine the appropriate classification.
Ultimately, the outcome of any such assessment is then classified using the system defined by Mueller et al. [26]. This is shown below in Table 2.2.