R.A. Jolly et al.

Regulatory Toxicology and Pharmacology 152 (2024) 105672

activation, e.g. hydroxylation and reactive intermediate formation and
reactivity. Such computational approaches including QM calculations
are shown in this manuscript for three NDSRIs. Understandably, gaining
regulatory acceptance of such novel approaches in read across remains a
challenge and uptake has been slow, although the recent structure activity relationship (SAR) work (Cross and Ponting, 2021) has been
instrumental in forming the basis of and improving upon the CPCA.
NDSRIs formed during the synthesis of drug substance can be more
readily qualified through carcinogenicity assessment of the API which
contains the NDSRI impurity than NDSRIs that form subsequently in
drug product due to the presence of nitrosating agents in excipients or
from other sources. Many NDSRIs are qualified in vitro using the Ames
assay described in OECD Guideline 471 (Heflich et al., 2020; OECD
guidelines 471, 2020). However, that Ames assay has been called into
question by Health Authorities (HAs) as not sensitive enough for nitrosamines, despite data indicating it is as or more predictive for assessing
nitrosamine mutagenicity than for other mutagenic compounds
(Trejo-Martin et al., 2022). Accommodations to the assay to address
perceived deficiencies, such as including increased amount of S9, the use
of hamster S9, and a longer preincubation for enhanced bioactivation,
have been implemented (FDA, 2023; EMA, 2023). A negative result from
this updated version of the Ames test, termed the Enhanced Ames Test
(EAT) in some regulatory guidelines, has allowed for increased limits for
NDSRIs (FDA, 2023; EMA, 2023) but only to the threshold of toxicological concern and not to the level of an ordinary impurity per ICH
Q3A/B.
An inability to sufficiently de-risk NDSRIs with the Ames assay has
led pharmaceutical companies to test NDSRIs in transgenic rodent
models, such as the Big Blue® or Mutamouse® to assess the potential for
mutation (Schmezer et al., 1998; Jacobson-Kram et al., 2004). These in
vivo mutagenicity assays, which are validated and conducted according
to regulatory guidelines (OECD 488 2022) and Good Laboratory Practices (GLPs), are considered adequate to qualify the mutagenic potential
of impurities (ICH M7 (R2) 2023) although, despite the high demand,
only a limited number of laboratories are currently capable of conducting such studies. Although studies such as the Ames assay and
transgenic rodent studies have been employed for qualitative hazard
assessment, their use for quantitative risk assessment use was discouraged in the ICH M7 (R2) Guidance Q&A (ICH, 2023) and has not been
commonly employed to date. Historically, the literature indicates that in
vitro potency in the Ames assay is not strongly correlated with carcinogenic potency (Purchase, 1985; McCann et al., 1988), the opposite
appears to be true for in vivo transgenic mutation data (Aoki 2017). One
path forward for evaluation of NDSRIs impurities is to use a relative
potency approach based on AIs generated using in vivo rodent mutagenicity studies. At the least, such an approach could be used qualitatively
to better categorize NDSRIs with respect to potency, e.g., to inform
whether they are in the ICH M7 (R2)-designated Cohort of Concern
(ICH, 2023). The data can additionally be employed in a more quantitative manner, e.g., to derive more meaningful AI limits. The current
paper describes an evaluation of NDSRIs that can potentially form in
Prozac® (fluoxetine), Cymbalta® (duloxetine), and Strattera® (atomoxetine) - three highly structurally similar examples of the oxetine drug
family and which have been commercially available for decades. These
drugs are selective reuptake inhibitors of the neurotransmitters: serotonin (SSRI, fluoxetine), serotonin and norepinephrine (SSRI/SNRI,
duloxetine), and norephinephrine (SNRI, atomoxetine) and are prescribed for the treatment several neurological disorders (FDA 2022,
2023a, 2023b). Herein, we describe the derivation of AIs for the nitrosated forms of each drug using computational assessments, read across
methodology, and in vivo mutation data for these potential impurities.
Furthermore, these analyses demonstrate that in vivo mutation data can
be used to derive a conservative and quantitative AI for NDSRI
molecules.

2. Materials and methods
The structures and Lilly serial numbers (LSNs) for fluoxetine,
duloxetine, and atomoxetine as well as the NDSRIs for each API are
shown in Fig. 1. LSNs of the NDSRIs are LSN3868255 for N-nitroso
fluoxetine (NFLX), LSN3868254 for N-nitroso duloxetine (NDLX) and
LSN3868306 for N-nitroso atomoxetine (NATX) respectively and all
LSNs were synthesized at Wuxi labs (China) with a purity greater than
99%.
2.1. Computational assessments based on structure and physicochemical
properties
Computational assessments, including structure and substructure
similarity and physicochemical property calculations on the NDSRIs,
were conducted using both Leadscope Model v3.0.2–4 (Columbus, OH)
and QSARflex software v1.6 (Mayfield, OH) programs.
2.2. Quantum mechanical modeling
For the quantum mechanical (QM) modeling, TD501 values from
compounds with structural similarity to the NDSRIs were collected from
the Lhasa carcinogenicity database (Lhasa Carcinogenicity Database
(lhasalimited.org). In particular, the compounds bearing similar structural and electronic features of these NDSRIs, such as those compounds
with or without an oxygen substitution at the β and γ carbon atoms as
well as compounds with an electron withdrawing group (EWG; e.g. CF3),
were included in this analysis. All computational calculations were
performed using the Gaussian 16 suite of programs (Frisch, 2016).
Geometrical structures of molecules were optimized using density
functional theory i.e., at the M062X/6-31+G (d,p) level of theory (Zhao
and Truhlar, 2008). Previous reports indicate that M062X function

Fig. 1. Structures of APIs and NDSRIs.

1
The TD50 is defined as the dose required to halve the probability of a
subject (animal) remaining without tumors throughout a lifetime of exposure.

2

24 / 34 ページ