Managing nitrosamine toxicological risk in late-stage development

NDSRIs are especially challenging during late development because CMC processes, specifications, clinical supply and regulatory commitments may already be established. New issues can arise from the synthetic route, formulation or drug-product interactions, or else the stability programmes showing impurity exposure above an acceptable intake (AI). Toxicological evaluations must therefore define a scientifically appropriate safety limit while supporting practical development decisions.

Commonly unique to the active pharmaceutical ingredient, NDSRIs therefore lack impurity-specific carcinogenicity data. The US Food and Drug Administration (FDA), in partnership with the European Medicines Agency (EMA) and other international health authorities, developed and adopted the carcinogenic potency categorisation approach (CPCA) to address this gap. Assigning an NDSRI to one of five predicted potency categories – based on the number and distribution of α-hydrogens and other activating or deactivating structural features surrounding the nitrosamine-forming centre – the approach focuses on the α-hydroxylation pathway responsible for the metabolic activation of many nitrosamines.^1,2

CPCA provides a rapid, defensible strategy for defining conservative AIs for NDSRIs”

CPCA provides a rapid, defensible strategy for defining conservative AIs for NDSRIs. However, late-stage programmes may be unable to achieve the corresponding control limits, which range from 18 to 1,500 ng/day across categories one to five. Difficulty achieving a CPCA-derived AI does not establish that the value is toxicologically inappropriate. Where more informative evidence is available, compound-specific mutagenicity or toxicology data, or scientifically justified read-across, may support an alternative AI.

Navigating CPCA for nitrosamine impurity control

CPCA assignment can be supported by expert rule-based in silico tools, but expert review remains essential to verify the prediction and evaluate out-of-domain structural features. FDA publishes CPCA-derived AIs separately from limits supported by compound-specific data or read-across, reinforcing that CPCA is a scientifically grounded starting point rather than the final product-quality control limit.²

Parallel toxicology and CMC decisions are needed to determine the most appropriate evaluation strategy. Where chemically and mechanistically similar NDSRIs can be identified from public sources, including compounds with FDA-established AIs, read-across may provide a practical data gap-filling approach when the analogue has robust carcinogenicity or relevant mutagenicity data.² Not every NDSRI requires testing. Before generating data, teams should establish how the information will influence AI derivation, product-quality decisions or the impurity-control strategy.

Read-across

A defensible read-across should integrate structural similarity, metabolism, mechanistic plausibility and toxicological relevance, rather than rely solely on descriptor-based similarity metrics”

Surrogate selection is highly dependent on local chemical similarity around the N-nitroso environment. Relevant attributes include α-carbon substitution, α-hydrogen availability, steric and electronic influences, metabolic activation and the resulting DNA-reactive intermediate. Priority should be given to analogues with robust mutagenicity or carcinogenicity data, similar N-nitroso environments and comparable DNA-reactive intermediates.³ A defensible read-across should integrate structural similarity, metabolism, mechanistic plausibility and toxicological relevance, rather than rely solely on descriptor-based similarity metrics.

Mutagenicity testing

Relevant datasets for the target NDSRI or a read-across analogue may include the enhanced Ames test (EAT), complementary mammalian-cell mutagenicity assays and metabolism studies. In vitro studies using liver microsomes or human hepatocytes may help evaluate reactive-intermediate formation. Mammalian systems can address uncertainties in bacterial results, evaluate mutagenic or clastogenic effects and strengthen the overall weight of evidence.⁴

Setting acceptable intakes for NDSRIs

During late development, full two-year rat bioassays or six-month transgenic mouse studies are rarely pragmatic due to marketing authorisation timelines and the substantial quantities of NDSRI required. Nonetheless, a compound-specific AI may be derived using ICH M7 principles. A tumourigenic dose rate 50 percent (TD₅₀)-based approach provides an established basis for linear extrapolation. Benchmark-dose modelling instead uses dose-response data to estimate a dose associated with a predetermined biological response, such as a 10 percent increase in tumour incidence, together with a confidence bound. Regardless of the approach, the point-of-departure study must be reliable, use relevant endpoints and appropriate dose selection, and provide an interpretable dose-response relationship supporting a biologically credible assessment of carcinogenic risk.⁵

Transgenic rodent (TGR) gene-mutation assays conducted under OECD Test Guideline 488 have emerged as a pragmatic source of compound-specific potency information when lifetime carcinogenicity studies are not feasible. ICH M7 provides the foundation for deriving compound-specific AIs using linear-extrapolation principles.⁶ TGR assays have shown a strong correlation (r² = 0.95) between mutagenic potency (BMDL₅₀) and carcinogenic potency (TD₅₀),⁷ and complementary evidence has been reported for an NDSRI evaluated using Ames and in vivo mutation testing.⁸ A chemical-specific AI protective for a 10⁻⁵ lifetime cancer risk may therefore be derived by applying ICH M7 linear-extrapolation principles to a suitable TGR-generated BMDL₅₀. The BMDL₅₀ is the lower 95 percent confidence limit of the dose associated with a 50 percent increase in mutation frequency above background in the most sensitive relevant tissue.

Acceptable intake challenges

A lifetime cancer-risk-based AI is not equivalent to a temporary or interim limit. The EMA states that less-than-lifetime (LTL) adjustment should not be used routinely to calculate nitrosamine limits but may be considered, after consultation with competent authorities, as a temporary measure while reductions are implemented.⁵ FDA separately publishes interim AIs for specified circumstances and describes them as temporary measures rather than permanent specification limits.²

FDA previously did not recommend applying the ICH M7(R2) LTL framework to nitrosamine impurities owing to uncertainties concerning cumulative-dose relationships, DNA-repair capacity and combined exposure through polypharmacy. In September 2026, FDA revised this position based on new scientific evidence and stated that ICH M7-style adjustments may be scientifically supported for individual nitrosamine impurities in drug products intended for limited treatment durations.^9,10

FDA identifies adjustment factors of 80 for treatment of one month or less, 13.3 for more than one month through 12 months, and 6.67 for more than one year through 10 years. The factor must account for the worst-case treatment duration and total exposure to multiple nitrosamines must remain within a 10⁻⁵ overall cancer-risk framework. FDA evaluates proposals case by case and continues to expect root-cause investigation and impurity mitigation. Interim limits are not automatically interchangeable with LTL-adjusted AIs or appropriate as permanent specifications.⁹

Read-across also has limitations when metabolism, mechanism or dose-response information is insufficient for quantitative extrapolation. Differences in methodology, analogue data gaps or reliance on qualitative similarity can prevent establishment of a defensible quantitative AI. Robust in vitro mutagenicity data may therefore provide important complementary evidence.

ICH Q3A/B thresholds

In addition to, or in the absence of, read-across, an EAT with the NDSRI is generally expected as a minimum dataset. EAT evaluates bacterial mutagenicity under conditions designed to improve nitrosamine sensitivity. Under the EMA framework, a negative GLP-compliant EAT can support control at 1.5 µg/day, numerically corresponding to the ICH M7 lifetime threshold of toxicological concern, whereas FDA may expect complementary in vitro mammalian and metabolism data to support this level.^2,11 Alternative AIs above 1,500 ng/day generally require additional compound-specific evidence, including in vivo mutagenicity data.

The key distinction is whether additional data support a higher AI while the NDSRI remains controlled as a mutagenic impurity or support its reclassification as non-mutagenic. EMA, but not FDA, permits ICH Q3A/B qualification and control thresholds when in vivo and in vitro evidence supports treatment as a non-mutagenic impurity.^2,5,11 FDA remains cautious because the relationship between negative in vivo mutagenicity findings and carcinogenic potential through other mechanisms is not fully elucidated.²

Toxicology and mitigation in parallel

Once exposure exceeds the applicable AI, toxicological refinement and impurity mitigation should proceed together. Deferring mitigation while awaiting a favourable toxicology result can consume limited late-stage flexibility, while manufacturing difficulty should not lower the scientific standard for AI derivation.

A pragmatic approach should include three features. First, determine whether the NDSRI presents a meaningful toxicological concern and whether exposure can be controlled before initiating additional studies. Second, require every study or read-across assessment to have a clear decision-making purpose and outcome plan. Third, plan for either result: additional evidence may refine the AI or confirm that the existing conservative framework remains appropriate.

Organisations should establish decision frameworks defining when CPCA is sufficient, when additional evidence may influence development decisions and how toxicology, CMC, quality and regulatory functions will respond when exposure exceeds an established limit. This enables consistent responses to evolving science and regulatory expectations while maintaining patient safety, product quality and development timelines.

Ultimately, read-across, mutagenicity testing and compound-specific AI derivation should be pursued only when they can meaningfully reduce uncertainty or alter development strategy. Integrating toxicology, CMC, quality and regulatory expertise into a unified workstream supports patient safety while reducing the risk of clinical holds, submission delays or programme failure.

References

1. US Food and Drug Administration. Determining recommended acceptable intake limits for N-nitrosamine impurities in pharmaceuticals: development and application of the carcinogenic potency categorization approach. Silver Spring (MD): US Food and Drug Administration; 2025.

2. US Food and Drug Administration. CDER nitrosamine impurity acceptable intake limits. Silver Spring (MD): US Food and Drug Administration; updated 2026 Aug 6.

3. Ponting DJ, de Oliveira Silveira G, Rowell P, et al. Establishing best practice for N-nitrosamine read-across and surrogate selection. Regul Toxicol Pharmacol. 2026;166:106009.

4. Atrakchi A, Puglisi R, Bercu J, et al. Nitrosamine Ames data review and method development: proceedings of a US FDA/HESI workshop. Mutagenesis. 2026;41(4):225-236. doi:10.1093/mutage/geag011.

5. European Medicines Agency. Questions and answers for marketing authorisation holders/applicants on nitrosamine impurities in human medicinal products. EMA/409815/2020 Rev.23. Amsterdam: European Medicines Agency; 2025.

6. International Council for Harmonisation of Technical Requirements for Pharmaceuticals for Human Use. ICH M7(R2) addendum: application of the principles of the ICH M7 guideline to calculation of compound-specific acceptable intakes. Geneva: ICH; 2023.

7. Jolly RA, Trejo-Martin A, Bercu JP, et al. Ames concordance with the in vivo transgenic rodent gene mutation assay for NDSRIs and relative in vivo TGR potency with nitrosamines with robust dose-response carcinogenicity data. Regul Toxicol Pharmacol. 2026;167:106051. doi:10.1016/j.yrtph.2026.106051.

8. Glowienke S, Onken U, Elhajouji A, et al. Genotoxicity evaluation of a valsartan-related complex N-nitroso-impurity. Regul Toxicol Pharmacol. 2022;134:105245. doi:10.1016/j.yrtph.2022.105245.

9. US Food and Drug Administration. Application of less-than-lifetime adjustment to nitrosamine impurities. Silver Spring, MD: FDA; 2026 Sep 24.

10. Felter SP, Bercu JP, Jolly RA, et al. Applicability of the ICH M7 guidelines for assessing less-than-lifetime exposure to nitrosamine impurities in drug products. Regul Toxicol Pharmacol. 2026;106245. doi:10.1016/j.yrtph.2026.106245.

11. European Medicines Agency. Appendix 3: Enhanced Ames test conditions for N-nitrosamines. EMA/120337/2024. Amsterdam: European Medicines Agency; 2024.