R.A. Jolly et al.

Regulatory Toxicology and Pharmacology 152 (2024) 105672

small, low molecular weight nitrosamines, with AI limits of 18 or 26.5
ng per day from the EMA and FDA respectively. While a reasonable
starting point in the absence of other information, the use of read across
methods based only on 2D structure and local similarity or SAR rules
based on steric hindrance and/or mitigating structural features local to
the nitrosamine generally results in overly conservative estimates of AI.
Read across techniques using the most similar 2D structural analogs
have been employed for decades, and there have always been challenges
with its application (Patlewicz et al., 2014). While chemical structure is
associated with potency, small changes in structure can have a large
impact on potency.
Traditionally, read-across has involved four key areas of comparison
including 1) 2D structural similarity 2) intrinsic physicochemical
properties, 3) mechanistic properties and 4) potential metabolism. If
there is additional information other than structure and local SAR, it
should be used to inform and bolster a read across analysis.
Read across considering physicochemical properties combined with
global structural similarity is more likely to be consistent with empirical
data (Lester et al., 2023). Use of intrinsic properties of molecules is
considered a key part of read across. Evaluation of predicted physicochemical properties such as LogP and solubility for NFLX, NDLX and
NATX show that NNK is not the best comparator. For example, a high
LogP indicates a high lipophilicity which, in turn, supports the lower
predicted solubility of the NDSRIs. In our dataset, all three NDSRIs have
high predicted LogP values, consistent with the APIs but not consistent
with NNK (Table 2). Such properties are designed into API at the outset
(Meanwell, 2011). Large differences in important physicochemical
properties such as the Log P between NNK and the NDSRIs suggest that,
if present in drug product, the NDSRIs are less likely to be soluble and
therefore less likely to be as bioavailable when compared with NNK.
Such intrinsic properties support a lower overall potency of NFLX, NDLX
and NATX when compared with NNK.
Given that the adverse outcome pathway (AOP) of nitrosamine
activation and reactivity is well understood, it is logical to use mechanistic data to interrogate potency of nitrosamines. Modeling and evaluating quantum mechanical (QM) parameters of the nitrosamine
activation pathway provides a useful way to assess mechanism. QM
modeling is one area that can provide a deeper mechanistic understanding and even be used to predict nitrosamine reactivity (Wenzel
et al., 2022). It can model several steps in nitrosamine activation
including interaction with P450 and reactive intermediate formation
and reactivity. QM assessments (De et al., 2024) for several classes of
nitrosamines have shown that QM modeling both supports the principles
of the CPCA framework but also highlights gaps and inconsistencies in
that framework. Importantly, QM modeling may provide a way to predict potency where the CPCA cannot differentiate the chemistry. Kostal
(Kostal and Voutchkova-Kostal, 2023) has published data showing that
QM parameters can allow for a binning approach to nitrosamine potency, aligned with potency levels proposed by Bercu et al. (2023a,b) e.
g. unknown nitrosamines can be separated into high (AI 26.5 ng/day),
medium (AI 150 ng/day) and low (AI > 1500 ng/day) potency groups
based on mechanism. In the work described here, the QM modeling
showed that the change in stretching vibration frequency reflects the
long-range electronic effect on the reactivity. The relative electronic
effect of the oxygen atoms at the beta or gamma position, a strong
electron-withdrawing group like CF3, and aromatic group substitutions
on the potency of the compounds are reflected in Table 2. A local model

of IR frequency as a surrogate for bond dissociation energy was correlated for nitrosamines similar to NDSRIs under evaluation. This analysis
showed a high correlation of these two factors (r = 0.9) and allowed for
local modeling and prediction of AIs for the NDSRIs. While assuming
similar metabolism for several of the metabolic steps for nitrosamine
activation, including alpha carbon hydroxylation, this analysis supports
that these NDSRIs would have lower potency when compared to NNK.
Traditional assessments of carcinogenic risk have been based on
tumor data from lifetime bioassays in rodents. These assessments
employed linear (non-threshold) low-dose extrapolation of the tumor
bioassay data to derive a TD50 which, in turn, was used to estimate a
dose associated with a theoretical excess cancer incidence of 10− 5 (ICH
M7 (R2) 2023). This animal- and time-intensive approach is not practical or in keeping with 3Rs principles (Hubrecht and Carter, 2019)
given the sheer number of impurities that may require assessment. More
streamlined studies, such as studies of in vivo mutagenicity (proximal to
tumor formation via the pathway of nitrosamine carcinogenicity) or the
use of other methodology such as in vivo duplex sequencing, can provide
a mechanistically sound and more efficient risk assessment of nitrosamine impurities. Bercu, Valentine and others have also shown that
duplex sequencing is highly consistent with TGR mutation results,
providing a way to assess mutation without the use of transgenic animals
(Bercu et al. (2023a,b); Valentine et al., 2020).
There is considerable literature data defining the metabolism of the
APIs from which these NDSRIs are derived (Mandrioli et al., 2006; Lantz
et al., 2003; Yu et al., 2016). These API molecules generally do not
undergo alpha carbon hydroxylation as a major metabolic pathway
which is associated with formation of reactive intermediates and more
predominant for LMW nitrosamines such as NDEA, NDMA or NNK. The
principal metabolic pathway via CYP2D6 for these APIs have been
described in the literature. By extension, LMW nitrosamines, which are
half the molecular size of the NDSRIs, are more likely to undergo such
metabolism than NDSRIs. This, higher molecular weight is one factor
supporting the lesser mutagenic potency of the NDSRIs when compared
with the LMW nitrosamines.
The science of carcinogenicity risk assessment has evolved to
incorporate a better understanding of biologic processes. Carcinogenesis
is a multifactorial process, and the adverse outcomes pathway (AOP) is
not linear (Kobets and Williams 2019; Cho et al., 2022). Not all mutations result in formation of a tumor due, for example, to DNA repair
mechanisms and immune surveillance with consequent killing of
DNA-damaged cells which supports a threshold-based approach.
Furthermore, the mutation frequency in the current TGR studies show a
clear threshold of effect.
Regulatory guidance such as the ICH M7 (R2) (2023) and EMA’s
Assessment Report (2020) were developed prior to the use of mutation
as a bonafide surrogate endpoint for carcinogenicity (e.g., Johnson et al.,
2021). A mechanism-based risk assessment paradigm using genotoxicity
AOPs is a scientifically justified and warranted alternative to the linear,
low-dose extrapolation approach based on rodent carcinogenicity data.
For nitrosamines, mutation is the relevant and sensitive endpoint for the
assessment of carcinogenic risk. Mutation is acknowledged as the key
precursor event to nitrosamine-induced carcinogenicity for which the
mode of action is well-established; this has been repeatedly demonstrated in numerous scientific studies (e.g., Li and Hecht 2022a,b). The
transgenic rodent model is a robust, well-validated model to assess
mutagenicity in vivo (OECD 488 2022) and is informed by scientific
study over many years. The OECD 488-compliant TGR model is the
accepted standard to assess mutagenicity in vivo. Being an vivo study,
metabolic aspects are accounted for and mutational events in relevant
tissues, e.g., liver for nitrosamines, can be assessed directly.
The AI derived using the NOEL or BMDL from the in vivo mutagenicity data (versus a TD50 value) is an appropriately conservative estimate of risk because the mutagenicity endpoint, as a required precursor
to carcinogenicity for nitrosamines, is more conservative than a tumor
endpoint. This was demonstrated for NDMA and NDEA, where the

2

FDA 2023, EMA 2023, Bercu 2023.
Calculated using the lower of the NOEL or BMDL value from the Big Blue
transgenic data in alignment with the ICH M7(R1) addendum approach. The
lowest value listed in mg/kg was divided by 50,000 and multiplied by 50 kg to
obtain an AI (e.g. for NATX, 4.4 mg/kg/50,000 *50 kg*1000000 ng/mg =
4400 ng per day). These AI calculations are consistent with those described in
M7R2. Instead of a TD50, the NOEL or BMDL are employed.
3

9

31 / 34 ページ