Re‐Evaluating Acceptable Intake: A Comparative Study of N‐Nitrosomorpholine and N‐Nitroso Reboxetine Potency

ABSTRACT

Establishing regulatory limits for Drug Substance‐Related Impurities (NDSRIs) is challenging due to the limited genotoxicity and carcinogenicity data available for many of these impurities, often leading to conservative approaches. In this study, we evaluated the genotoxic potential of two structurally related nitrosamines: N‐nitrosomorpholine (NMOR) and N‐nitroso reboxetine. Compared to the well‐studied NMOR, there is little toxicological information available for N‐nitroso reboxetine. Currently, both compounds have an acceptable intake value of 127 ng/day, based on a read‐across using the available carcinogenicity data of NMOR. While both compounds tested positive in a series of in vitro and in vivo assays, we found that the mutagenic potential of N‐nitroso reboxetine was significantly lower than that of NMOR. The benchmark dose (BMD) analysis of in vivo mutagenicity data supports an acceptable intake of 24,000 ng/day for N‐nitroso reboxetine. Computational studies, carried out using the quantum‐mechanical CADRE program, were consistent with in vitro and in vivo outcomes, suggesting an acceptable intake at or above 1500 ng/day for N‐nitroso reboxetine. In comparison to NMOR, this prediction is supported by lower computed reactivity in the hydroxylation step, greater steric hindrance of the alpha carbons, and more facile proton transfer in the heterolysis toward the aldehyde metabolite. The data presented in this work can be used to refine and improve the Carcinogenic Potency Categorization Approach (CPCA). It also underscores the importance of collaboration between regulatory authorities, the pharmaceutical industry, and scientific researchers to address potential risks while avoiding overestimation of the acceptable intake limits for certain NDSRIs.

1. Introduction

Genotoxicity tests for impurities in drug products are crucial for identifying and characterizing the genotoxic potential of these compounds to ensure that any associated risks are effectively managed (Bercu et al. 2009; Eichenbaum et al. 2009; Fioravanzo et al. 2012; Galloway 2017). These tests are conducted in accordance with the International Council for Harmonization (ICH) M7 guidelines (ICH 2023). Hazard assessments of impurities in a drug product typically begin with database and literature searches to gather carcinogenicity and bacterial mutagenicity data. The Threshold of Toxicological Concern (TTC) concept was developed to set a safe intake limit for humans to chemicals with limited toxicological data, thereby mitigating potential risk (Hartung 2017; Snodin 2018). A TTC value of 1.5 μg/day has been identified as appropriate for lifetime intake of genotoxic impurities, with an associated excess cancer risk of less than 1 in 100,000 (Munro et al. 2008; Snodin and McCrossen 2012). However, the TTC principle does not extend to N‐nitrosamines (Thresher et al. 2020; Snodin 2023).

N‐nitrosamines are a large class of chemicals characterized by a common functional group “N—N=O,” attached to other simple or more complex chemical structures (Beard and Swager 2021). During the process of metabolic activation, cytochrome P450 (CYP) enzymes mediate the α‐hydroxylation of nitrosamines, forming unstable α‐hydroxy nitrosamines. These intermediates further decompose into highly reactive alkyl diazonium ions, capable of forming DNA adducts (Abu‐Bakar et al. 2022; Dobo et al. 2022; Fahrer and Christmann 2023). If not properly repaired, these DNA adducts can cause mutations, potentially leading to carcinogenesis (Snodin et al. 2024). Some N‐nitrosamines are classified as probable (group 2A) or possible (group 2B) human carcinogens by the International Agency for Research on Cancer (IARC) based on results from animal studies (Li et al. 2021; Vikram et al. 2024).

Human exposure to nitrosamines can result from various sources such as food and beverages, tobacco smoke, rubber, or personal care products (Gushgari and Halden 2018). Since 2018, the detection of various nitrosamines in drug products has led to multiple recalls of medications from markets (Parr and Joseph 2019; Tuesuwan and Vongsutilers 2021). To address the need for a framework to assess and mitigate the risk of nitrosamines, regulatory authorities have been actively working on guidance for nitrosamines in drug products with updates periodically. In August 2023, the FDA issued a final guidance that recommends acceptable intake (AI) limits for Nitrosamine Drug Substance‐Related Impurities (NDSRIs) based on the Carcinogenic Potency Categorization Approach (CPCA) (FDA 2023), a method that was also adopted by other health authorities (HAs) (Canada 2023; EMA 2023).

The CPCA uses structure–activity relationships (SAR) concepts to evaluate the molecular structure of nitrosamines, focusing on identifying structural features that either increase or decrease carcinogenic potency (Goller et al. 2024; Kruhlak et al. 2024). Each nitrosamine is assigned a potency score based on its structural features. Depending on the potency score, nitrosamines are categorized into different potency classes with a predetermined AI limit, ranging from 18 or 26.5 to 1500 ng/day (Canada 2023; EMA 2023; FDA 2023).

The CPCA represents a major advancement in nitrosamine risk assessment, enabling a quick determination of AI limits for NDSRIs while promoting a consistent approach across various jurisdictions. However, this approach is sometimes overly conservative, as it might not consider a broad spectrum of factors that may affect the potential genotoxicity and carcinogenicity of NDSRIs (Bercu et al. 2024; Chakravarti et al. 2024; Jolly et al. 2024; Powley et al. 2024). Therefore, it is essential to regularly validate the CPCA against real‐world data and case studies to refine and improve the methodology. This, along with judicious benchmarking to higher‐level in silico modeling, such as quantum‐mechanical calculations, will over time improve CPCA's reliability and accuracy. The current efforts within the HESI‐FDA GTTC Mechanism‐based Genotoxicity Risk Assessment (MGRA) Working Group (https://hesiglobal.org/gttc‐nitrosamines/) reflect the community's shared commitment to this task.

Genotoxicity testing data can be used as an alternative method to justify a higher AI limit than that derived from the CPCA. The bacterial reverse mutation assay, also known as the Ames test, is widely accepted by regulatory agencies globally as a standard method for in vitro mutagenicity testing (Ames et al. 1975; Maron and Ames 1983; Trejo‐Martin et al. 2022; ICH 2023). Typically, negative results from a properly conducted Ames test in a standard conditions would classify an impurity as non‐mutagenic; however, this does not apply when evaluating nitrosamines due to concerns about the test's sensitivity for this class of compounds (Mori et al. 1985; Lijinsky 1987; Li et al. 2023). To address these concerns, a modified version of the traditional test, referred to as the Enhanced Ames Test (EAT) has been recommended, which utilizes specific bacterial strains, the pre‐incubation (30‐min) platform, recommends solvents and solvent volumes to minimize potential interference with Cytochrome P450s, the use of both rat and hamster liver S9 at 30%, and two nitrosamine‐specific positive controls. Recent studies have confirmed that the modified conditions enhanced the sensitivities of Ames tests in detecting the mutagenicity of multiple nitrosamines (Heflich et al. 2024; Thomas et al. 2024). Currently, the EMA may permit a control limit of 1.5 μg/day for nitrosamines that test negative in an EAT‐compliant Ames. However, the FDA may not accept negative Ames results as sole support for the absence of mutagenic properties and could request additional in vitro mammalian cell mutation assays and in vitro metabolism data to support the 1.5 μg/day AI (EMA 2023; FDA 2023).

Additional in vivo mutagenicity data may be required to support AI limits exceeding 1.5 μg/day.

The Transgenic Rodent (TGR) mutation assay is regarded as the industry standard for in vivo mutagenicity testing of nitrosamines because it accounts for the metabolic activation and detoxification processes that occur in vivo, which are essential for accurately assessing the mutagenic potential of nitrosamines (Lambert et al. 2005; Wills et al. 2017; Powley et al. 2024). However, producing and maintaining transgenic rodents is costly, contributing to the high expense of conducting TGR assays. The demand for these animals often exceeds the supply, particularly in nitrosamine‐related testing, causing significant lead time to obtain the necessary transgenic animals and delaying the testing process. The development of error‐corrected next‐generation sequencing (ecNGS) methods, which enable direct measurement of mutations from wild‐type animals, offers a promising solution to this challenge and provides further mechanistic insights into the mutagenic process induced by nitrosamines (Valentine 3rd et al. 2020; Marchetti et al. 2023). In our previous studies, we showed that the in vivo comet assay was also highly sensitive in the detection of DNA damage induced by N‐nitrosodiethylamine (NDEA) in rodents (Bercu et al. 2023; Zhang et al. 2024).

In the current study, we evaluated the genotoxicity profiles of two structurally related nitrosamines: NMOR and N‐nitroso reboxetine (Figure 1) with a series of in vitro, in vivo, and in silico testing. NMOR is a known animal carcinogen demonstrated to induce various types of tumors in animal models. Some of the primary tumors associated with NMOR exposure include liver, lung, larynx, trachea, nasal cavity, and esophagus (Ketkar et al. 1983; Lijinsky et al. 1984, 1991; Hecht et al. 1989; Klein et al. 1990). The α‐hydrogen score of 1, with counts of α‐hydrogens (2,2), combined with a deactivating feature of +1 (N‐nitroso group in a morpholine ring), would categorize both NMOR and N‐nitroso reboxetine into CPCA potency Category 2, with a predetermined lifetime acceptable intake of 100 ng/day. The established acceptable intake (AI) value for NMOR is 127 ng/day, based on a TD50 value of 0.127 mg/kg/day in the most sensitive tissue, as determined by a robust study from the Carcinogenic Potency Database (CPDB) (Kruhlak et al. 2024).

FIGURE 1.

Structures of NMOR (116.1 g/mol) (a) and N‐nitroso reboxetine (342.4 g/mol) (b).

Compared to NMOR, the mutagenic and carcinogenic potential of N‐nitroso reboxetine is less well understood, despite both molecules featuring a morpholine ring with a nitroso group attached to the nitrogen atom. The same AI of 127 ng/day was assigned to N‐nitroso reboxetine using a read‐across approach, with the TD50 of NMOR as the point of departure. Although both compounds tested positive in our in vitro and in vivo genotoxicity studies, N‐nitroso reboxetine exhibits significantly lower mutagenic potency compared to NMOR. The benchmark dose lower confidence limit (BMDL) for N‐nitroso reboxetine, determined from the most sensitive in vivo mutation data (Duplex Sequencing), was about 190 times higher than that of NMOR, indicating that the CPCA/read‐across approach may have overestimated the potency of N‐nitroso reboxetine. The validated CADRE model, which assesses key events in the metabolism of N‐nitrosamines from first principles (Kostal and Voutchkova‐Kostal 2023) was consistent with the in vivo and in vitro assays, indicating that N‐nitroso reboxetine is less carcinogenic than NMOR by at least an order of magnitude in terms of predicted potency.

2. Methodology

2.1. Bacterial Reverse Mutation Test (Ames)

NMOR (ID: ArZ‐NT002350; Cas No. 59‐89‐2) was synthesized by ArZa Bioscience Limited, Bolton, United Kingdom, and the conduct of the Ames Assay was performed at Pfizer Research and Development, Groton, CT, U.S.A.

N‐nitroso reboxetine (Catalog No. SZ‐R059002) was synthesized by SynZeal Research Pvt. Ltd., Gujarat, India, for use in all testing, and the conduct of the Ames Assay was performed at Labcorp Early Development Laboratories Ltd., North Yorkshire, England.

Four strains of Salmonella typhimurium bacteria (TA98, TA100, TA1535, and TA1537) and one strain of Escherichia coli bacteria (WP2 uvrA pKM101) were used in Ames tests. The mammalian liver post‐mitochondrial fractions (S‐9) used for metabolic activation were obtained from Molecular Toxicology Incorporated, USA, where they were prepared from male Sprague Dawley rats or male Syrian hamsters induced with 5,6 Benzoflavone/Phenobarbital. Treatments were carried out both in the absence and presence of S‐9 by addition of either buffer solution or S‐9 mix, respectively. The S9 mixture was prepared immediately before starting an assay. Complete S9 Mixture contained the following components:

Ingredient	Final content per mL in:
	10% or 30% S‐9 mix
Sodium phosphate buffer pH 7.4 (SPB)	100 μmoles
Glucose‐6‐phosphate (disodium) (G‐6‐P)	5 μmoles
β‐Nicotinamide adenine dinucleotide phosphate (NADP) (disodium)	4 μmoles
Magnesium chloride (MgCl2)	8 μmoles
Potassium chloride (KCl) 33 μmoles	33 μmoles
Water	To volume
S‐9	100 or 300 μL

The Ames test was conducted following the recommendations of the OECD 471 test guideline (OECD 2020). S. typhimurium (TA98, TA100, TA1535 and TA1537) and E. coli (WP2 uvrA pKM101) were checked for strain characteristics (histidine or tryptophan dependence, rfa character and uvrB character if applicable, and resistance to ampicillin) prior to assay conduct. Triplicate plates without and with S‐9 (rat and hamster) for test article, vehicle (or in some instances six replicates) and positive controls were run.

Quantities of test article (0.1 mL), vehicle control solution (0.1 mL), or positive control (0.05 to 0.1 mL), bacteria (0.1 mL), and S‐9 mix or buffer (0.5 mL), or in tests with N‐nitroso reboxetine, an additional 0.5 mL of 100 mM sodium phosphate buffer (pH 7.4), were mixed together and incubated while shaking at 37°C for 30 min before combining with 2 mL supplemented agar at ~45°C, followed by rapid mixing and pouring onto Vogel‐Bonner minimal agar plates. When set, the plates were inverted and incubated for ~3 days at 37°C. Following incubation, these plates were examined for evidence of precipitate and cytotoxicity, and revertant colonies were counted either electronically or manually.

N‐nitroso reboxetine was tested with all five strains in the absence and presence of 10% rat and hamster liver S9. NMOR was tested using water as the solvent in all five strains in the absence and presence of both 10% and 30% rat and hamster liver S9. Additionally, NMOR was tested using DMSO as the solvent with TA100 and TA1535 in the absence and presence of both 10% and 30% rat and hamster liver S9.

Test Articles were considered mutagenic (positive) if the increase in mean revertants at the. The peak of the concentration response was equal to or greater than 3× (for strains TA 1535 and TA 1537) or 2× (for strains TA 98, TA 100, and WP2 uvrA pKM101) the mean vehicle control value and above the corresponding acceptable vehicle control range (95% CL).

2.2. Comet Assay

2.2.1. Characterization and Preparation of Test and Control Articles

N‐nitroso reboxetine was purchased from SynZeal (Lot # SRL‐1012‐119). For the N‐nitroso reboxetine dosing formulations, Microcrystalline Cellulose PH 102 was added to the API at a 1:1 ratio, then mixed with a vehicle consisting of 0.5% (w/v) methylcellulose A4M in purified water. A vehicle containing 75 mg/mL Microcrystalline Cellulose PH 102 in 0.5% (w/v) methylcellulose A4M in purified water was used to dose the vehicle control animals. A 30 mg/mL solution of EMS was prepared in purified water and used to dose the positive control animals approximately 4 h prior to tissue collection on Day 9.

NMOR was purchased from Tokyo Chemical Industry (Lot # 7BBWG). For the NMOR dosing formulations, Microcrystalline Cellulose PH 102 was added to the API at a 1:1 ratio, then mixed with a vehicle consisting of 0.5% (w/v) methylcellulose A4M in purified water. A vehicle containing 1.5 mg/mL Microcrystalline Cellulose PH 102 in 0.5% methylcellulose A4M in purified water was used to dose the vehicle control animals. A 30 mg/mL solution of EMS was prepared in purified water and used to dose the positive control animals approximately 3 h prior to tissue collection on Day 10.

2.2.2. Test System

Male C57BL/6 mice, approximately 8–10 weeks old at the initiation of dosing, were used in the comet assay study. Animals were housed in a controlled environment at 68°F–79°F and a relative humidity of 30%–70% with a 12‐h light/dark cycle. Animals were fed Certified Irradiated Rodent Diet 2916C (Envigo Teklad Global Diet). Municipal drinking water, further purified by reverse osmosis, was provided ad libitum.

2.2.3. Experimental Design

All animals were dosed by oral gavage once daily for 9 (N‐nitroso reboxetine) or 10 (NMOR) consecutive days beginning on Day 1 using a dose volume of 10 mL/kg. Individual dose volumes were calculated based on the animals' most recently recorded body weights. Five mice per dose group were tested. All animals in each positive control group were dosed by oral gavage once on the last study Day, ~3–4 h prior to necropsy.

2.2.4. Liver Comet Sample Processing and Evaluation

Liver samples from surviving animals were collected ~3–4 h post dose on Day 9 (for study of N‐nitroso reboxetine) or Day 10 (for study of NMOR) and processed for comet analysis. The samples were minced to prepare a single‐cell suspension. Four slides were prepared for each animal by mixing samples in low‐melting agarose, layered onto microscope slides precoated with high‐melting agarose. The slides were incubated overnight in lysis solution and. electrophoresed in electrophoresis buffer (pH > 13) for 30–40 min the next day. The slides were stained with SYBR GoldTM stain and blind coated. Overall, 150 cells (75 cells from 2 slides) per animal were scored using the KometImage Analysis System Version 7.0.1.43.

2.2.5. Statistical Analysis for Comet Assay

The comet parameters were calculated using the slide as the unit of measurement (Lovell and Omori 2008). The measurements were log transformed, and a summary statistic was calculated for each slide. Prior to taking the log, a small constant (0.001) was added to each measured value to avoid taking the log of zero. The summary statistic for %Tail DNA (PCT_TDNA) is the 75th percentile of the log transformed values.

Descriptive statistics were generated for each group. To test for an increasing trend across dose groups and for an overall test effect, a one‐way analysis of variance (ANOVA) on all groups was conducted, with two‐sided trend tests on vehicle control and test article‐treated groups and two‐sided pairwise comparisons of each group to the vehicle control group. The trend tests were performed sequentially using linear contrasts, and the pairwise comparisons were done using Dunnett's test.

For the sequential trend tests, if the initial test with all groups included in the trend analysis was significant (trend p‐value ≤ 0.05), it was concluded that the high dose group was different from the control. The trend test was continued with the high dose group dropped, and the trend test was performed on the remaining groups. This process was repeated until a nonsignificant p‐value was found, or the test was performed on only the low dose group and control.

2.3. Big Blue/Duplex Sequencing

2.3.1. Characterization and Preparation of Test and Control Articles

NMOR was purchased from Tokyo Chemical Industry (Lot # 7BBWG). Microcrystalline Cellulose PH 102 was added to the API at a 1:1 ratio, then mixed with a vehicle consisting of 0.5% (w/v) methylcellulose A4M in purified water. A vehicle containing 3 mg/mL Microcrystalline Cellulose PH 102 in 0.5% (w/v) methylcellulose A4M in purified water was used to dose the vehicle control animals. A 4 mg/mL solution of ENU (Sigma‐Aldrich, lot# MKCT1926) was prepared in PBS (calcium and magnesium free; pH 6.0–6.1) daily on Days 1–3 and was used to dose the positive control animals.

N‐nitroso reboxetine was purchased from SynZeal (Batch# SRL‐1012‐119). Microcrystalline Cellulose PH 102 was added to the API at a 1:1 ratio, then mixed with a vehicle consisting of 0.5% (w/v) methylcellulose A4M in purified water. A vehicle containing 30 mg/mL microcrystalline cellulose (Avicel PH102) in 0.5% (w/v) methylcellulose A4M in purified water was used to dose the vehicle control animals. A 4 mg/mL solution of ENU (Sigma‐Aldrich, lot# MKCR7896) was prepared in PBS (calcium and magnesium free; pH 6.0–6.1) daily on Days 1–3 and was used to dose the ENU positive control animals.

2.3.2. Test System

Big Blue C57BL/6 transgenic mice were ordered from Taconic Laboratories for Gentronix. At the time of dosing initiation, the animals were 8–10 weeks of age, and the body weights of the animals ranged from 18.9 to 26.3 g. Animals were housed in a controlled environment at 68°F–79°F and a relative humidity of 30%–70% with a 12‐h light/dark cycle. Animals were fed Certified Irradiated Rodent Diet 2916C (Envigo Teklad Global Diet). Municipal drinking water, further purified by reverse osmosis, was provided ad libitum.

All procedures performed on animals were in accordance with regulations and established guidelines and were reviewed and approved by an Institutional Animal Care and Use Committee or through an ethical review process.

2.3.3. Experimental Design

All animals (except those in the positive control group) were dosed by oral gavage once daily for 28 consecutive days using a dose volume of 10 mL/kg/day. Individual dose volumes were calculated based on the animal's most recently recorded body weight. Each experimental group (vehicle control, test article and positive control) includes six animals. DNA isolated from mice exposed to N‐ethyl‐N‐nitrosourea (ENU) by oral gavage at 40 mg/kg/day once daily on Study Days 1, 2, and 3 and necropsied on Study Day 31 was used as positive controls.

2.3.4. Tissue Collection

All animals were euthanized by gas anesthesia (isoflurane) followed by exsanguination on Day 31 and necropsied. For N‐nitroso reboxetine, the liver, bone marrow, glandular stomach, duodenum, and blood were collected, while for NMOR, the liver, bone marrow, kidney, and duodenum were collected. Tissues were flash frozen in liquid nitrogen and stored in a freezer set to −80°C. Tissues were maintained at −80°C until processing for cII mutant analysis and Duplex sequencing.

2.3.5. cII Mutations—Extraction, Processing of Genomic DNA, and Analysis

Genomic DNA was isolated from liver and bone marrow following RecoverEase Methods (Agilent, Santa Clara, CA). The lambda phage shuttle vectors were recovered from the genomic DNA and were packaged into empty lambda bacteriophage capsids, creating infectious phage using Transpack (Agilent, Santa Clara, CA). The phage was then used to infect E. coli and was grown in temperature‐selective conditions overnight for titer plates and approximately 40–48 h for mutant plates. Plaques were scored for both titer plates and mutant plates, and the mutant frequencies for each tissue were calculated. At least 125,000 phage particles were evaluated from at least two packagings.

Three titer plates per tissue per animal were scored for packaging efficiency. Five mutant plates were scored per tissue per animal. The cII mutant frequency was calculated by the ratio between the number of mutant phages and the total phages screened for each tissue analyzed from each animal.

2.3.6. Duplex Sequencing—Extraction, Processing of Genomic DNA, and Analysis

Genomic DNA was extracted from liver and/or kidney (NMOR only) of at least four (based on numerical order) viable Big Blue mice per group with DNeasy Blood & Tissue Kits (QIAGEN). Approximately 500 ng of extracted DNA from each sample was fragmented to a median size of about 200–300 base pairs with the TwinStrand DuplexSeq Kit‐Enzymatic Fragmentation Module. Duplex libraries were constructed from enzyme‐digested DNA using the DuplexSeq Mouse Mutagenesis kit (TwinStrand Biosciences Inc., Seattle, WA). Briefly, library preparation included end‐repair and A‐tailing, followed by the ligation of DuplexSeq adapters containing unique molecular identifiers (UMI). Target regions of DNA were enriched by hybrid capture using the Mouse Mutagenesis panel, followed by PCR amplification using the included kit reagents. Final libraries were quantified with a Qubit fluorometer and Agilent TapeStation system. Pooled samples were sequenced on the Illumina NovaSeq 6000 system using an S4 flow cell with a 2 × 150 bp read length.

Endogenous mutations were measured with a 48 kb panel targeting 20 regions spread throughout the mouse genome. The regions are balanced to provide an unbiased sampling of representative sequence contexts throughout the genome (GC content, genic/non‐genic, coding/non‐coding, etc).

The FASTQ files generated from sequencing were analyzed with the TwinStrand DuplexSeq Mutagenesis App on DNAnexus. The consensus calling and postprocessing, variant calling, and interpretation were performed as previously described (Valentine 3rd et al. 2020). The DS mutation frequency (MF) was calculated by the ratio between mutant duplex bases and total on‐target duplex bases sequenced.

Sequencing data were deposited to the Sequence Read Archive (SRA) with BioProject ID PRJNA1200773.

2.3.7. Statistical Analysis for Big Blue/Duplex Sequencing

Statistical analysis of mutant/mutation frequency data was performed using SAS software. A square root transformation was applied to the mutant/mutation frequency data. The transformed data were analyzed using a one‐way analysis of variance on all groups (vehicle control, positive control, and test article treated groups), with two‐sided trend tests on the vehicle control and test article treated groups and one‐sided pairwise comparisons of each group (including positive control) to the vehicle control group. The trend tests were performed sequentially using linear contrasts, and the pairwise comparisons were done using Dunnett's test.

For the sequential trend tests, if the initial test with all groups included in the trend analysis was significant (trend p‐value < 0.05), then it was concluded that the “highest” dose group was different from the “lowest” dose group, and a subsequent trend test was performed on all groups except the “highest” dose group. Testing continued in this manner until a nonsignificant result was obtained or the test was performed on only the two “lowest” dose groups. All subsequent trend tests were one‐sided in the direction suggested by the data in the initial test.

2.4. Benchmark Dose Modeling

BMD modeling was performed with PROAST version 70.1 (released on 19‐10‐2020 and accessed via https://proastweb.rivm.nl/). The BMD response was based on mutant/mutation frequency for both the cII and DS mutational analyses. For the comet assay, the response was based on the percent tail. The benchmark response (BMR), that is, critical effect size (CES), was set at 50% for all genotoxicity endpoints. The BMD was model averaged, with 200 bootstrap runs and an AI criterion of 2.

2.5. CADRE Computational Modeling

2.5.1. Quantum–Mechanical (QM) Calculations

The tiered structure of the CADRE N‐nitrosamine model was reported previously (Kostal and Voutchkova‐Kostal 2023) and has been used for NDSRI evaluations across industry in support of AI‐limit assessment for over 2 years. In contrast to the CPCA or read‐across, CADRE relies on the electronic structures of impurities, where density functional theory (DFT) is used to evaluate key transformations in the metabolic activation pathway from first principles. The model assesses ionization and tautomer states of the compound prior to DFT calculations and uses aqueous Monte Carlo (MC) simulations in conjunction with mixed quantum and classical mechanics calculations (QM/MM) to compute physicochemical properties and solute‐solvent energetics (i.e., Coulomb and van der Waals interactions). This enables CADRE to define a biologically relevant conformational landscape and to gauge the bioavailability of N‐nitrosamines. To capture reactivity, CADRE relies on the well‐accepted QM‐FMOT (Frontier Molecular Orbital Theory) approach (Kostal 2018), which uses global and atom‐based steric factors and electronic indices calculated at the mPW1PW91/MIDIX+ level of theory (Kostal and Voutchkova‐Kostal 2023). The mPW1PW91/MIDIX+ method was developed to calculate accurate energies of reaction, barrier heights, and electron affinities of large molecules at a reasonable cost (Lynch and Truhlar 2004). The method was shown to yield more accurate hydrogen‐abstraction energetics (the presumed rate‐determining step in N‐nitrosamine activation to the diazonium) than the benchmark second‐order Møller‐Plesset perturbation theory at ca. 20% of the computational cost (Lynch and Truhlar 2004).

2.5.2. Statistical Modeling

CADRE employs the R language (R 2009) and environment for statistical computing (version 4.1.2), which is used for data analysis, linear regressions, and linear discriminant analyses (LDAs) to predict N‐nitrosamine potency. Multivariate normality (mvn) of descriptors is determined using functions in the mvn library; the original descriptor selection was carried out with a genetic algorithm, as implemented in the library genalg, using 100 iterations with a mutation probability of 0.05 and the Bayesian Information Criterion (BIC) to avoid overfitting. Internal performance was estimated using the leave‐one‐out (LOO) cross‐validation method, and final model selection was based on performance in external validation (Kostal and Voutchkova‐Kostal 2023). The dataset used for model training was obtained from the Lhasa Carcinogenicity Database (LCDB, carcdb.lhasalimited.org). The LDA models used in this study were trained to (i) classify N‐nitrosamine contaminants into three potency categories: potent COCs (Cat 1, TD₅₀ ≤ 0.15 mg/kg), COC compounds (Cat 2, 0.15 < TD₅₀ ≤ 1.5 mg/kg), and non‐COCs (Cat 3, TD₅₀ > 1.5 mg/kg), and (ii) distinguish carcinogens from true non‐carcinogens. The overall accuracy of the CADRE tool was reported to be 77% in external testing for the LDA models and > 80% for the MLR models; CADRE extrapolation to the NDSRI space was explored and found to be reliable (Kostal and Voutchkova‐Kostal 2023), owing to the physics‐led approach to the model, that is, its reliance on the underlying chemistry (vs. chemicals in the training set).

3. Results

3.1. Bacterial Reverse Mutation Test (Ames)

NMOR was evaluated at concentrations ranging from 15 to 5000 μg/plate for the potential induction of reverse mutations (revertants) in S. typhimurium tester strains TA 1535, TA 1537, TA 98, and TA 100, and E. coli strain WP2 uvrA pKM101 using the in vitro pre‐incubation platform. When water was used as the solvent, there was a higher number of mean revertants per plate in the presence of rat liver S9 with TA 100 (max 2.9× with 10% and 4.5× with 30%), TA 1535 (max 21.5× with 10% and 38.5× with 30%), and E. coli (max 1.9× with 10% and 2.2× with 30%). Furthermore, there was a higher number of mean revertants per plate in the presence of hamster liver S9 with TA 98 (max 1.6× with 10% and max 2.3× with 30%), TA 100 (max 10.2× with 10% and 10.6× with 30%), TA 1535 (max 152.0× with 10%, max 112.2× with 30%), and with E. coli (max 3.4× with 10%, max 2.8× with 30%) (Figure 2).

FIGURE 2.

Ames assay dose‐response data for NMOR (water) using test strains TA98, TA100, TA1535, TA1537, and WP2 uvrA pKM101 in the absence of S9 and in the presence of 10% and 30% rat and hamster S9.

In a confirmatory assay (Figure 2 Test 2), Salmonella strains TA 98 and TA 1537 in the presence of hamster liver S9 metabolic activation (10%) and E. coli strain WP2 uvrA pKM101 in the presence of hamster liver S9 metabolic activation (10%) and rat liver S9 metabolic activation (10% and 30%) were exposed in triplicate to NMOR (water as solvent) over a narrow concentration range of 1000–5000 μg/plate to address weak increases observed in original tests. The results in the repeat tests were similar to the original tests, demonstrating reproducibility of the assay.

In tests in which DMSO was used as the solvent, there was a higher number of mean revertants per plate in the presence of rat liver S9 with TA 100 (max 1.7× with 10% and 1.5× with 30%) and TA 1535 (max 22.1× with 10% and 11.3× with 30%), as well as with hamster liver S9 with TA 100 (max 8.9× with 10% and 11.3× with 30%) and TA 1535 (max 154.9× with 10% and 142.9× with 30%) (Figure 3).

FIGURE 3.

Ames assay dose‐response data for NMOR (DMSO) using test strains TA100 and TA1535 in the absence of S9 and in the presence of 10% and 30% rat and hamster S9.

In a confirmatory test with TA 100 in the presence of rat liver S9 (10% and 30%) was exposed in triplicate to NMOR (DMSO as solvent) over a titrated concentration range (313–5000 μg/plate) to address the weak increases observed in the initial test. The response seen with 10% rat liver S9 was similar to that seen in the initial test (1.8×) while the response seen with 30% rat liver S9 was higher (max 2.3×). It was concluded that NMOR was mutagenic in the Exploratory Bacterial Mutagenicity Assay.

N‐nitroso reboxetine was evaluated at concentrations ranging from 5–5000 μg/plate with each of the five bacterial strains in the absence and presence of S‐9. In the presence of both rat and hamster S9 (10%), increases (max 6.3× and max 6.4×, respectively) in the mean revertant value were observed in strain TA1535 at concentrations of 1600 and 5000 μg/plate (Figure 4), providing evidence for the mutagenic activity of N‐nitroso reboxetine in this assay system. Additionally, there was evidence of an insoluble compound on the plates at concentrations ≥ 500 μg/plate with all of the strains in the absence and presence of S9. In the absence of metabolic activation, there was evidence of cytotoxicity at concentrations ≥ 500 μg/plate in TA 98 and E. coli, and at the highest concentration tested (5000 μg/plate) with TA1535 and TA1537.

FIGURE 4.

Ames assay dose response data for N‐nitroso reboxetine (DMSO) using test strains TA98, TA100, TA1535, TA1537, and WP2 uvrA pKM101 in the absence of S9 and in the presence of 10% rat and hamster S9. Precipitate was observed at concentrations ≥ 500 μg/plate with all strains. Cytotoxicity was observed in the absence of S9 at concentrations ≥ 500 μg/plate with E. coli and TA98 and at the highest concentration tested (5000 μg/plate) with TA1535 and TA1537.

3.2. Mortality, Clinical Signs, and Body Weights

In the 10‐day in vivo comet assay, one animal administered NMOR at 1 mg/kg/day was found dead on Day 1. All remaining animals survived to scheduled study termination.

Clinical signs observed in the study included eye closure in 1 animal on Day 10 at 15 mg/kg/day. There was no statistically significant decrease in body weight observed at any of the doses throughout the 10‐day study.

In the 9‐day in vivo comet assay, one animal administered N‐nitroso reboxetine at 250 mg/kg/day was found dead on day 3 of the dosing phase. All other animals survived to their scheduled termination. One animal administered N‐nitroso reboxetine at 10 mg/kg/day had decreased activity, rough haircoat, tiptoe walking, malocclusion, and thin appearance. There was no statistically significant change in the mean body weight of animals administered N‐nitroso reboxetine over the course of the study.

In the 31‐day Big Blue assay, animals administered NMOR at 30 mg/kg/day were euthanized on Day 1 and Day 3. Test‐article related clinical signs were limited to the 30 mg/kg/day group, which included decreased activity and eye closure. There were no test article‐related differences from controls in body weight parameters. In the 31‐day Big Blue assay for N‐Nitroso Reboxetine, there were no test article‐related mortality, clinical signs, or changes in mean body weight over the course of the study.

3.3. Comet Assay

In livers of C57BL/6 mice administered NMOR, there was test article‐related higher %Tail DNA at doses of 1, 3, 5, 10, and 15 mg/kg/day. These increases were statistically significant (pairwise p < 0.001; trend p < 0.001). No increase in the % tail DNA was observed in the lower dose groups, including 0.1 and 0.3 mg/kg/day (Figure 5A).

FIGURE 5.

Assessment of DNA damage induced by NMOR and N‐nitroso reboxetine in dose range‐finding studies. Data based on 150 comet cells scored per animal. **Statistically significant at the 0.01 level.

In livers of C57BL/6 mice administered N‐nitroso reboxetine, there was test article‐related higher %Tail DNA at doses of 10, 30, 90, 250, and 750 mg/kg/day. These increases were statistically significant (pairwise p < 0.001; trend p < 0.001) (Figure 5B).

3.4. Duplex Sequencing in DRF Studies

Mutation frequencies were measured by Duplex Sequencing at 20 target regions in the livers and kidneys of C57BL/6 mice administered NMOR and in the livers of C57BL/6 mice administered N‐nitroso reboxetine.

Administration of NMOR for 10 days induced statistically higher MFs at doses of 1, 3, 5, 10, and 15 mg/kg/day in the livers of C57BL/6 mice. Statistically higher MFs were observed in the kidneys at doses of 3, 5, 10, and 15 mg/kg/day (Figure 6A).

FIGURE 6.

Assessment of mutations induced by NMOR (a) and N‐nitroso reboxetine (b) with Duplex sequencing in dose range‐finding studies. The mutation frequency is determined by the ratio between mutant duplex bases and total duplex bases. *Statistically significant at the 0.05 level; **Statistically significant at the 0.01 level.

We tested MFs in the livers of C57BL/6 mice administered N‐nitroso reboxetine at the highest dose (750 mg/kg/day) and the lowest dose (10 mg/kg/day). Administration of N‐nitroso reboxetine for 9 days did not induce statistically higher MFs at the dose of 10 mg/kg/day, while a 2.4‐fold increase in MFs was observed in the livers of animals administered N‐nitroso reboxetine at the dose of 750 mg/kg/day, and the increase was statistically significant (pairwise p < 0.001) (Figure 6B).

3.5. cII Mutant Frequency

Administration of NMOR at doses of 3 and 10 mg/kg/day for 28 days induced statistically higher mutant frequencies relative to the vehicle control (pairwise p < 0.001, trend p < 0.001) at the cII gene in the liver of Big Blue male mice (Figure 7A). The mean mutant frequencies of the 3 and 10 mg/kg/day groups were above the 95% historical background control limit for group means in the liver of male Big Blue mice (Table S1).

FIGURE 7.

Assessment of cii mutations induced by NMOR and N‐nitroso reboxetine in Big Blue studies. Animals were administered vehicles alone or test articles daily by oral gavage for 28 days followed by 3‐day expression period. Animals from the positive control group were administered ENU by oral gavage on days 1, 2, and 3 and necropsied on day 31. **Statistically significant at the 0.01 level.

Administration of NMOR at doses up to 10 mg/kg/day for 28 days did not cause a statistically higher mutant frequency relative to the vehicle control at the cII gene in the bone marrow of Big Blue male mice (Figure 7A). Mean mutant frequencies of all treatment groups were within the historical 95% control limit in the bone marrow of Big Blue male mice (Table S2).

Administration of N‐nitroso reboxetine at doses up to 300 mg/kg/day did not cause statistically elevated MF at the cII gene in the liver or bone marrow of Big Blue male mice (Figure 7B). The ENU treatment resulted in a statistically significant increase in mutant frequencies for both tissues tested, demonstrating the utility of the test system to detect and quantify induced mutants, following exposure to a known direct‐acting mutagen (Figure 7A,B).

3.6. DS Mutation Frequency

Administration of NMOR at doses of 0.03 or 0.1 mg/kg/day for 28 days did not cause a statistically (pairwise comparison) higher mutation frequency relative to the vehicle. Control at the 48 kb regions in the liver of male Big Blue mice. Administration of NMOR at doses of 0.3, 1, 3, or 10 mg/kg/day induced statistically higher mutation frequencies relative to the vehicle control in the liver (Figure 8A). Mean mutation frequencies of these four treatment groups were above the historical 95% control limit in the liver of Big Blue male mice (Table S3).

FIGURE 8.

Assessment of mutations induced by NMOR and N‐nitroso reboxetine by Duplex Sequencing with tissues collected from Big Blue studies. The mutation frequency is determined by the ratio between mutant duplex bases and total duplex bases. *Statistically significant at the 0.05 level; **Statistically significant at the 0.01 level.

Administration of NMOR at doses of 0.03, 0.1, 0.3, or 1 mg/kg/day for 28 days did not cause a statistically (pairwise comparison) higher mutation frequency relative to the vehicle control at the 48 kb regions in the kidney of male Big Blue mice. Administration of NMOR at doses of 3 or 10 mg/kg/day induced statistically higher mutation frequencies relative to the vehicle control in the kidney (Figure 8A). Mean mutation frequencies of these three treatment groups were above the historical 95% control limit in the kidney of Big Blue male mice (Table S4).

In the liver of Big Blue male mice, treatment with N‐nitroso reboxetine at doses of 10, 30, 100, and 300 mg/kg/day induced dose‐related and statistically significant increases in mutation frequencies (pairwise p < 0.001, trend p < 0.001) relative to the vehicle control (Figure 8B). The mutation frequencies from all treatment groups fall outside the 95% control limits of the distribution of the historical vehicle controls (Table S3).

3.7. DS Simple Base Substitution Spectra

To assess whether NMOR treatment alters the mutational spectra in the Big Blue mouse genome, we examined the simple base substitution spectra in the liver and kidney of Big Blue mice administered varying doses of NMOR. The administration of NMOR led to increases in the mutation frequency (MF) across all six mutation subtypes in the mouse liver (Figure 9A). The primary mutation types observed were T>G transversions, followed by T>A transversions and T>C transitions. The proportion of each mutation subtype is illustrated in Figure 9B. Despite dose‐related increases in the total counts of T>C transitions, their proportion showed a decreasing trend due to more significant increases in other mutation subtypes (i.e., T>A, T>C, and T>G). In contrast, the proportion of C>T transitions was less changed and continued to be the predominant mutation type at all NMOR dose levels in the kidney (Figure 9C,D). The predominant mutation type in the liver of animals administered N‐nitroso reboxetine was C>T transitions at all dose levels determined by both mutation frequency (Figure 9E) and their proportion (Figure 9F).

FIGURE 9.

Simple base substitution spectra for NMOR and N‐nitroso reboxetine in Duplex sequencing studies. Dose levels and mutation subtypes are color‐coded. Top panels (a and b): Livers collected from animals administered NMOR. Middle panels (c and d): Kidneys collected from animals administered NMOR. Bottom panels (e and f) livers collected from animals administered N‐nitroso reboxetine. Panels on the left (a, c, e): Mutation frequencies were calculated as the specific subtype single‐nucleotide mutation count divided by the total duplex bases. Panels on the right (b, d, f): Proportion of individual mutational subtypes.

3.8. DS Trinucleotide Spectra

We next analyzed the mutation patterns induced by NMOR and N‐nitroso reboxetine treatments within a trinucleotide context. In mouse liver, vehicle control samples predominantly exhibited C>T transitions, commonly found in the ACG, CCG, GCG, or TCG context. A notable shift in the trinucleotide spectra was observed in the liver of animals treated with NMOR at a dose of 10 mg/kg/day (Figure 10, top two panels). Compared to vehicle controls, there were overall increases in T>A, T>C, and T>G mutations in NMOR‐treated liver samples. Additionally, there were significant increases in C>T mutations in the kidneys of animals treated with NMOR at the same dose. Both liver and kidney shared some mutational hotspots, such as T>A transversions at CTA and CTG contexts, T>C transitions at GTA and GTC contexts, and T>G transversions at GTA and GTC contexts (arrows in Figure 10).

FIGURE 10.

Trinucleotide spectra for NMOR and N‐nitroso reboxetine in liver and kidney. Arrows highlight trinucleotide contexts that exhibit greater prominence of C>T, T>A, T>C or T>G mutations. The predominant C>T mutations were frequently found in ACG, CCG, GCG or TCG trinucleotide context in vehicle control groups. Trinucleotide spectra of N‐nitroso reboxetine are different from that of N‐nitroso reboxetine in liver of Big Blue mice.

N‐nitroso reboxetine preferentially induces C>T transitions in the liver, a mutation subtype frequently observed from spontaneous deamination. However, N‐nitroso reboxetine also increased the mutation frequency across various trinucleotide contexts for C>T transitions and is not limited to high‐frequency contexts of ACG, CCG, GCG, or TCG seen in vehicle controls.

3.9. Benchmark Dose

The BMDs calculated from Duplex sequencing data in Big Blue mouse livers are shown in Table 1 for NMOR and N‐nitroso reboxetine. NMOR has a BMD range of 0.024–0.07 mg/kg/day. N‐nitroso reboxetine has a significantly higher BMD confidence interval of 4.49–12.1, although both compounds are categorized in CPCA Category 2, suggesting that CPCA modeling may have overpredicted the potency of N‐nitroso reboxetine.

TABLE 1.

BMDs for in vivo duplex sequencing endpoints.

Treatment	BMDL50	BMDU50	AI (ng/d)
NMOR	0.024	0.07	127 a
N‐nitroso reboxetine	4.49	12.1	24,000 b

Note: BMDL is the BMD response (50% CES) at the lower 90th confidence interval. BMDU is the BMD response (50% CES) at the upper 90th% confidence interval.

^a AI published by Health Authorities.

^b Derived by dividing the BMDL₅₀ N‐nitroso reboxetine for mutation induction in the liver by the BMDL₅₀ for NMOR for the same endpoint and then multiplying the result by the published AI for NMOR.

3.10. CADRE Modeling

Consistent with the CPCA categorization, NMOR was predicted as a potent COC in CADRE (TD₅₀ ≤ 0.15 mg/kg), confirming an AI of less than 150 ng/day. In contrast, N‐nitroso reboxetine was predicted as a non‐COC (Cat 3, TD₅₀ > 1.5 mg/kg), suggesting a conservative AI of 1500 ng/day. A detailed discussion of these outcomes is offered in the next section.

4. Discussion

Nitrosamines in pharmaceuticals can be broadly categorized into small‐molecule nitrosamines and NDSRI. Initially, regulatory bodies concentrated on small‐molecule nitrosamines, such as NDMA and NDEA, which are not structurally related to the active pharmaceutical ingredient (API). These nitrosamines were observed in various medications and raised concerns due to their potential carcinogenicity (Charoo et al. 2019; Shephard and Nawarskas 2020; Shaik et al. 2022). However, the focus has since expanded to include NDSRIs, which are structurally related to the API and can form during the manufacturing process or storage of the drug (Holzgrabe 2023; Schlingemann et al. 2023). Many NDSRIs lack specific mutagenicity and carcinogenicity data; this is partly because NDSRIs are a relatively new area of focus, and their structural diversity makes it challenging to gather comprehensive data quickly, since each NDSRI may require individual testing.

The recommendation for the CPCA by regulatory authorities like the EMA, HC, and FDA stems from the need to better assess and manage the risks associated with NDSRIs. It offers a fast applied SAR‐based method to determine AI limits for NDSRIs without robust genotoxicity or carcinogenicity data available. However, this method does not consider other important factors, including physicochemical properties of NDSRIs such as molecular weight, solubility, potential metabolic pathways, and DNA adduct formation and stability, which may significantly impact their mutagenic and carcinogenic potency (Thomas et al. 2022; Chakravarti 2023). For example, it has been proposed that AIs derived through the read‐across method should be adjusted according to the molecular weight of the target compound (Snodin 2023; Bercu et al. 2024). This is based on the rationale that each nitroso group in a nitrosamine can produce only one diazonium ion, which can result in a single DNA alkylation and potentially one DNA mutation. Consequently, the number of nitroso groups per mass unit, which correlates with the molecular weight of the nitrosamine, directly influences its carcinogenic potency (Fine et al. 2023). Specifically, N‐nitroso reboxetine is anticipated to be less potent than NMOR due to its molecular weight being three times higher than that of NMOR (Bercu et al. 2024).

While the CPCA framework is valuable, it relies on computational predictions and surrogate data, which can sometimes lead to overestimations of applicable intake limits. Experimental or higher‐grade (e.g., QM) computational data reflecting real biological responses can validate or refute these predictions, helping to refine and improve the accuracy of models like the CPCA over time. In this study, we compared the genotoxicity profiles of two compounds using a series of in vitro, in vivo, and physics‐led in silico tests. Our goal was to establish a realistic acceptable intake (AI) for N‐nitroso reboxetine based on its true genotoxic potency, thereby enabling more effective control of the risks posed by this NDSRI.

Because nitrosamines need metabolic activation to produce reactive intermediates that interact with DNA and have the potential to cause mutations, regulators are concerned that the standard Ames test may not always be sensitive enough to evaluate the mutagenic potential of nitrosamines. This concern is heightened by historical instances where in vivo carcinogens have shown negative results in the Ames test (Rao et al. 1979; Andrews and Lijinsky 1980; Lijinsky 1987). However, recent studies have shown that modified study designs, which include a 30‐min pre‐incubation method, 30% rat/hamster liver S9, and specific tester strains, could enhance the sensitivity of the Ames test for detecting the mutagenicity of N‐nitrosamines (Heflich et al. 2024; Kajavadara et al. 2024; Thomas et al. 2024). In our study, we initially conducted the standard OECD‐compliant Ames test for N‐nitroso reboxetine before health authorities recommended the Enhanced Ames Test for nitrosamine impurities in drug substances. When using 10% rat or hamster liver S9 and DMSO as a solvent, we observed increases in revertant numbers in TA1535, with a maximum of 6.3× for 10% rat S9 and 6.4× for 10% hamster S9. In a more recent Ames test for NMOR, we used both rat and hamster liver S9 at two different concentrations (10% and 30%) to compare their effects on mutagenic potency. We found that hamster liver S9 was more effective in the metabolic activation of NMOR, even at lower concentrations; specifically, 10% hamster liver S9 induced a stronger mutagenic response than 30% rat liver S9. When testing NMOR under similar conditions to N‐nitroso reboxetine (i.e., 10% rat or hamster liver S9, DMSO as solvent, and 30‐min pre‐incubation), we observed increases in revertant numbers in TA1535, with a maximum of 22.1× for 10% rat S9 and 154.9× for 10% hamster S9. Additionally, there were a higher mean number of revertants per plate in TA100, with a maximum of 1.8× for 10% rat liver S9 and 8.9× for 10% hamster S9. Therefore, NMOR exhibited greater mutagenic potency than N‐nitroso reboxetine in the in vitro tests, as evidenced by both the number of tester strains that tested positive and the magnitude of the mutagenic responses. Concerns have been raised about DMSO potentially inhibiting the metabolic activation of cytochrome P450 (Hakura et al. 2010; Li et al. 2010; Valicherla et al. 2019). Under the conditions of our study, both water and DMSO appear to show comparable results in the mutagenicity tests of NMOR in the selected tester strains.

The in vivo comet assay is a widely recognized method for measuring single or double DNA strand breaks (Azqueta et al. 2019; Cordelli et al. 2021; Vodicka et al. 2023). Our previous studies have shown that the in vivo comet assay is highly sensitive in detecting DNA damage induced by NDEA treatment in both mice and rats (Bercu et al. 2023; Zhang et al. 2024). Additionally, it demonstrates great specificity in testing non‐genotoxic NDSRIs such as N‐nitroso ramipril and N‐nitroso quinapril (Cheung et al. 2024) In the current study, both NMOR and N‐nitroso reboxetine induced dose‐related DNA damage, as measured by % Tail DNA. However, N‐nitroso reboxetine required much higher concentrations than NMOR to achieve comparable fold increases relative to vehicle controls. This significant difference was also reflected in their mutagenic potential in the dose range‐finding studies, as measured by Duplex sequencing. For instance, administering NMOR at 10 mg/kg/day for 10 days resulted in over an 8‐fold increase in mutation frequencies in the liver of C57BL/6 mice, whereas administering N‐nitroso reboxetine at the same dosage for 9 days did not cause any significant increase in mutation frequencies compared to the vehicle control.

The TGR assay, often regarded as the industry standard for in vivo mutagenicity testing, clearly shows a marked difference in the mutagenic potential of the two compounds after 28‐day treatment. NMOR caused a statistically significant increase in mutant frequency at a dose as low as 3 mg/kg/day in the liver of Big Blue mice. In contrast, N‐nitroso reboxetine did not induce a significant increase in mutant frequency in either the liver or bone marrow of treated animals at a dose of up to 300 mg/kg/day. Using the highly sensitive Duplex sequencing, both compounds tested positive for mutation induction in the liver. With the Duplex sequencing data derived from the liver of transgenic mice in the 28‐day Big Blue study, BMD analysis was performed to compare the mutagenic potency of NMOR and N‐nitroso reboxetine. NMOR has a BMDL₅₀ of 0.024 mg/kg/day, but the BMDL₅₀ of N‐nitroso reboxetine is about 190 times higher than that of NMOR, underscoring that N‐nitroso reboxetine is a much less potent mutagen compared to NMOR.

The analysis of mutational spectra reveals that NMOR has a wide‐ranging mutagenic impact, leading to various DNA base substitutions, notably high frequencies of C>T, T>A, T>C, and T>G mutations. These mutations may result from a diverse array of DNA adducts formed during NMOR metabolism (Li and Hecht 2022). The distribution and metabolism of N‐nitroso[¹⁴C]morpholine ([¹⁴C]NMOR) were examined using autoradiographic and in vitro techniques in Sprague–Dawley rats (Lofberg and Tjalve 1985). Overall, 8 h post‐injection of [¹⁴C] NMOR, higher levels of radioactivity, measured in dpm/mg wet tissue, were observed in the liver (260.5 ± 45.9) compared to the kidney (77.3 ± 11.4), which may contribute to the slight difference in NMOR's mutational spectra between these organs. The mechanisms of metabolic activation and DNA adduct formation by N‐nitroso reboxetine are less understood. However, the predominance of C>T transitions in its mutational spectra suggests that N‐nitroso reboxetine, despite sharing the morpholine ring with NMOR, might undergo a different mutagenic process. Further detailed studies comparing the in vivo metabolisms and DNA adducts formed by NMOR and N‐nitroso reboxetine are necessary to elucidate the distinct mutational spectra and significant differences in the mutagenic potency of these compounds.

The CADRE modeling outcomes are consistent with the above in vitro and in vivo assays, placing N‐nitroso reboxetine in the non‐COC potency category (TD₅₀ > 1.5 mg/kg, suggesting a conservative AI of 1500 ng/day). Here, it is important to understand the difference in the prediction for NMOR and N‐nitroso reboxetine from first principles. In CADRE, global electronic properties are derived from the Frontier Molecular Orbital Theory (FMOT), while local properties are expressed via the Fukui function, $f\left(r\right)={\left[\frac{\partial \rho \left(r\right)}{\partial N}\right]}_{\nu \left(r\right)}$, which measures propensity of an atom (or molecule) to accept or donate electron density (Torrent‐Sucarrat et al. 2010). Here, we evaluated susceptibility to undergo radical chemistry at the α‐carbon positions, $f^0\left(C_{\alpha }\right)=\left[{\rho }_{N+1}\left(C_{\alpha }\right)+{\rho }_{N-1}\left(C_{\alpha }\right)\right]/2$, which reflects the initial rate‐determining C‐H alkylation step. Maxima in the computed $f^0\left(C_{\alpha }\right)$ index correspond to greater propensity for this process. From Table 2, in comparing NMOR and N‐Nitroso Reboxetine, NMOR ($f_{\mathit{avg}}^0\left(C_{\alpha }\right)$ = 0.031) was nearly three times more susceptible to radical chemistry than N‐nitroso reboxetine ($f_{\mathit{avg}}^0\left(C_{\alpha }\right)$ = 0.011). In gauging local electrophilicity of the α‐C, $f^{+}\left(C_{\alpha }\right)=\left[{\rho }_{N+1}\left(C_{\alpha }\right)-{\rho }_N\left(C_{\alpha }\right)\right]$, likewise, NMOR showed to be more susceptible to a nucleophilic attack ($f_{\mathit{avg}}^{+}\left(C_{\alpha }\right)$ = 0.033) than N‐nitroso reboxetine ($f_{\mathit{avg}}^{+}\left(C_{\alpha }\right)$ = 0.013). These results are further supported by computed bond dissociation energies (BDEs), which reflect the vertical enthalpy change to form the radical intermediate in the hydroxylation step. Here, NMOR was ca. 1.1 kcal/mol more reactive than N‐nitroso reboxetine, and, crucially, the weakest α‐C—H bond, identified by calculating the Wiberg bond index (Bridgeman et al. 2001), was shown to be considerably less sterically hindered in NMOR than in N‐nitroso reboxetine. Measured via the solvent‐accessible surface area (SASA), NMOR was more accessible to the cytochrome P450 Fe = O⁴⁺ by over 3 Ų compared to N‐nitroso reboxetine, owing to the lack of β‐C substitution.

TABLE 2.

Computed electronic and steric parameters that underscore the difference in NMOR and N‐nitroso reboxetine CADRE predictions of carcinogenic potency.

Compound	CADRE LDA model prediction	SASA, weakest α‐C‐H bond (Å2)	Caco‐2 (nm/s)	Average radical susceptibility (α‐carbon)	Average electrophilic susceptibility (α‐carbon)	Vertical bond dissociation energy (weakest α‐C‐H bond, kcal/mol)
NMOR	Potent COC (TD50 ≤ 0.15 mg/kg)	22.6	505.4	0.031	0.033	112.9
N‐nitroso reboxetine	Non‐COC (TD50 > 1.5 mg/kg)	19.2	290.1	0.011	0.013	114.0

To understand this preference, it is important to consider the conformational landscape of both NDSRIs. While NMOR is only hindered by the “bent” N‐nitroso moiety, which is limited in rotation due to partial resonance, N‐nitroso reboxetine has a tertiary‐carbon substitution on the β—C, where two phenyl rings effectively block one of the α—Cs. To what extent these moieties play a role in hindering the α—C is difficult to estimate from a single structure; thus, CADRE uses QM/MM/MC simulations to compute SASA (and other properties) as ensemble averages (Kostal and Voutchkova‐Kostal 2023), which, in the case of N‐Nitroso Reboxetine, involved sampling of 15 × 10⁶ configurations in explicit aqueous solution. Figure 11 depicts the lowest‐energy structures of both N‐nitroso reboxetine and NMOR, where the N‐nitroso group is rotated away from the aromatic rings in the former, presumably due to greater solvation of this conformer, which also avoids the repulsion between the N‐nitroso and the π‐electron density of the benzene ring (when N‐nitroso oxygen faces toward the aromatic ring). From Figure 11, we can assume the benzene ring favors a syn orientation with the α—C due to a favorable H–π interaction, as is well‐known for aromatic molecules, and is worth ca. 1–1.5 kcal/mol for alkyl–benzene pairings (and more for polarized H interactions) (Ribas et al. 2002). This interaction stabilizes N‐nitroso reboxetine, thus making it less reactive, and accounts for the steric hindrance on the α‐C, further limiting hydroxylation at this position.

FIGURE 11.

Lowest‐energy states of N‐nitroso reboxetine (green) and NMOR (blue), as identified from QM/MM/MC simulations in explicit aqueous medium and post‐optimized at the mPW1PW91/MIDIX+ level of theory. Key distances discussed in the text are shown using dashed lines.

Beyond the hydroxylation step, an argument can be made that NMOR is more reactive than N‐nitroso reboxetine in the heterolysis step of the N‐nitrosamine activation to the diazonium, where a proton is transferred from the α‐C‐OH to the N‐nitroso to cleave the aldehyde. We have shown in our previous reports (Kostal and Voutchkova‐Kostal 2023; Kostal 2024) that as the average distance between the hydrogen on the α‐C and the oxygen of the N‐nitroso decreases, the potency of cyclic N‐nitrosamines increases, owing to the implied lowering of the donor‐acceptor energy barrier and the increased likelihood of quantum tunneling in proton transfer (Krishtalik 2000). To that end, there is a strong correlation between the calculated TD₅₀ values and this metric across cyclic alkyl N‐nitrosamines in the LCDB database (R ²  ~ 0.8 for structures with reliable data) (Kostal 2024). Here, N‐nitroso reboxetine showed a greater average distance (2.8 Å) than NMOR (2.6 Å), which is consistent with the rest of our computational analysis that strongly indicates N‐nitroso reboxetine to be less potent than NMOR.

Lastly, in terms of bioavailability, we predicted Caco‐2 (i.e., apparent Caco‐2 cell permeability in nm/s), which is a model for non‐active transport across the gut‐blood barrier that was shown to be relevant to N‐nitrosamine potency in our previous report (Kostal and Voutchkova‐Kostal 2023). From Table 2, N‐nitroso reboxetine had a lower permeability rate than NMOR (290 vs. 505 nm/s). Though both rates are considered to be high (> 100 nm/s), their relative difference further supports our prediction of considerably lower potency for N‐nitroso reboxetine over NMOR.

5. Conclusion

NMOR is a well‐studied mutagen that has been demonstrated to be a carcinogen in rodent studies. It tested positive in our in vitro Ames assay, in vivo comet assay, Big Blue assay, and Duplex sequencing. N‐nitroso reboxetine, although mutagenic both in vitro and in vivo, has a significantly higher BMDL₅₀ than that of NMOR (4.49 vs. 0.024 derived from Duplex sequencing data), which supports an AI of 24,000 ng/day when comparing the potency of NMOR. Consistent with these results, the CADRE computational analysis, which addresses the metabolism of N‐nitrosamines from first principles to estimate carcinogenic potency, showed that N‐nitroso reboxetine is less potent than NMOR by at least an order of magnitude. These outcomes were reasoned using computed reactivity in the hydroxylation step, steric hindrance of the alpha‐carbons, and estimated barrier for the heterolysis to generate the aldehyde metabolite. This work provides a more accurate risk estimate for N‐nitroso reboxetine using horizontally integrated experimental and physics‐led in silico data, helping to refine the framework for predicting the potency of other novel nitrosamines.

Author Contributions

Shaofei Zhang: manuscript drafting, experiment planning, data analyses. Jennifer Cheung: manuscript drafting, experiment planning, data analyses. Jakub Kostal: manuscript drafting, data analyses. Adelina Voutchkova‐Kostal: manuscript drafting, data analyses. Maik Schuler: manuscript drafting, experiment planning, data analyses.

Disclosure

Shaofei Zhang, Jennifer Cheung, and Maik Schuler are employees and shareholders of a pharmaceutical company.

Supporting information

Data S1.

Data Availability Statement

The data that support the findings of this study are openly available in SRA at https://www.ncbi.nlm.nih.gov/sra, reference number PRJNA1200773.