Artifacts Generated during Azoalkane Peroxy Radical Oxidative Stress Testing of Pharmaceuticals Containing Primary and Secondary Amines

Article abstract of DOI:10.1002/jps.24667

We report artifactual degradation of pharmaceutical compounds containing primary and secondary amines during peroxy radical-mediated oxidative stress carried out using azoalkane initiators. Two degradation products were detected when model drug compounds

Full text of DOI:10.1002/jps.24667

RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
Artifacts Generated During Azoalkane Peroxy Radical Oxidative
Stress Testing of Pharmaceuticals Containing Primary and
Secondary Amines
MARCELA NEFLIU,1 TODD ZELESKY,2 PATRICK JANSEN,3 GREGORY W. SLUGGETT,2 CHRISTOPHER FOTI,2
STEVEN W. BAERTSCHI,3,4 PAUL A. HARMON1
1Merck Research Laboratories, Merck & Co., Inc., West Point, Pennsylvania 19486
2Pfizer Inc., Analytical Research & Development, Eastern Point Road, Groton, Connecticut 06340
3Eli Lilly and Company, Lilly Research Laboratories, Indianapolis, Indiana 46285
4Baertschi Consulting, LLC, Carmel, Indiana 46033
Received 24 May 2015; revised 1 September 2015; accepted 4 September 2015
Published online 13 November 2015 in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/jps.24667
ABSTRACT: We report artifactual degradation of pharmaceutical compounds containing primary and secondary amines during peroxy
radical-mediated oxidative stress carried out using azoalkane initiators. Two degradation products were detected when model drug
compounds dissolved in methanol/water were heated to 40°C with radical initiators such as 2,2ꢀ-azobis(2-methylpropionitrile) (AIBN). The
primary artifact was identified as an ␣-aminonitrile generated from the reaction of the amine group of the model drug with formaldehyde
and hydrogen cyanide, generated as byproducts of the stress reaction. A minor artifact was generated from the reaction between the
amine group and isocyanic acid, also a byproduct of the stress reaction. We report the effects of pH, initiator/drug molar ratio, and type
of azoalkane initiator on the formation of these artifacts. Mass spectrometry and nuclear magnetic resonance were used for structure
elucidation, whereas mechanistic studies, including stable isotope labeling experiments, cyanide analysis, and experiments exploring the
ꢁC
effects of butylated hydroxyanisole addition, were employed to support the degradation pathways. 2015 Wiley Periodicals, Inc. and the
American Pharmacists Association J Pharm Sci 104:4287–4298, 2015
Keywords: artifacts; stress testing; forced degradation; oxidation; autoxidation; free radicals; chemical stability; strecker reaction;
degradation products; mass spectrometry; stability
INTRODUCTION
is designed to accelerate oxidative kinetics so as to induce in
a short period of time an observable response for molecules
susceptible to autoxidation induced by peroxy radical attack.7
Because of the complexity of the oxidative system, the test has
historically suffered from some selectivity issues in creating
the peroxy radical activity, allowing artifactual degradation to
be observed.8 The current study represents another step in re-
fining the test conditions and understanding its limitations.
Peroxy radicals, known to selectively abstract weakly-bound
H atoms (bond dissociation enthalpy < ꢀ90 kcal/mol),4,9 can
be generated for the purpose of the stress test through ther-
mal decomposition of radical initiators such as 2,2ꢀ-azobis(2-
methylpropionitrile) (AIBN)10,11 or 4,4ꢀ-azobis(4-cyanovaleric
acid) (ACVA).7,11 The subsequent, rapid reaction with dissolved
oxygen generates the desired peroxy radicals as shown for
AIBN in Scheme 1. Usually, the experiment is carried out
in dilute drug solutions at 20 mol% or greater of the radi-
cal initiator.12 The long-lived azoalkane-derived peroxy radi-
cals may therefore encounter drug molecules from which they
may abstract H atoms, but may also readily undergo recom-
bination/decomposition as shown in Scheme 2. Characteristic
to tertiary peroxy radicals, the recombination/decomposition of
the azoalkane peroxy radicals leads to the undesired formation
of alkoxy radicals, estimated to react 104–105 times faster with
organic substrates, than the corresponding peroxy radicals.13
As a result, the selectivity of the test for weakly bound H atoms
is lost, as the drug may preferentially react with alkoxy rad-
icals at a variety of sites within the drug molecule. For drug
A key component of pharmaceutical stress testing is the execu-
tion of simple solution-based experiments to reveal intrinsic hy-
drolytic and oxidative susceptibilities of new drug compounds.1
Stress testing is used to support various aspects of drug devel-
opment including development and optimization of stability-
indicating methods and relevant control strategies. Identifica-
tion of potential drug degradation products from solution-based
experiments may also facilitate the risk assessment process
for the evaluation of degradation products for potential mu-
tagenicity (ICH M7).2 It is therefore desirable that solution
stress testing conditions trigger degradation pathways that are
relevant not only to the drug substance, but also to the as-
sociated drug product dosage forms. Primary hydrolytic drug
degradation products are often predicted via solution-phase
chemistry,3 however, evaluation of propensity of drugs to ox-
idize is more complex.4–6 Peroxy radical mediated oxidation,
a type of autoxidation, is induced by trace impurities (e.g., in
various excipients) and is considered the most common route of
drug oxidation.6 The solution stress test involves peroxy radi-
cal formation via thermal decomposition of azo compounds and
Correspondence to: Christopher Foti (Telephone: +650-378-2135; Fax: +650-
522-5472; E-mail: chris.foti@gilead.com)
This article contains supplementary material available from the authors upon
request or via the Internet at http://wileylibrary.com.
Journal of Pharmaceutical Sciences, Vol. 104, 4287–4298 (2015)
ꢁ
C
2015 Wiley Periodicals, Inc. and the American Pharmacists Association
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015
4287

4288
RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
acetate, were obtained from Sigma–Aldrich (St. Louis, Mis-
souri). Diphenhydramine hydrochloride and ethyl ether
(anhydrous) were obtained from Fisher Scientific (Pittsburgh,
Pennsylvania), methanol and acetic acid were obtained from
J.T. Baker (Center Valley, Pennsylvania), and potassium
phosphate monobasic, acetonitrile, formic acid, and trifluo-
roacetic acid, from Fluka (St. Louis, Missouri). [13C,15N]-AIBN
(2,2ꢀ-azobis(2-methylpropio[13C,15N]nitrile)) was obtained from
The Chemistry Research Solution (Bristol, Pennsylvania).
Sodium cyanide was obtained from MP Biomedicals (Santa
Ana, California), taurine was obtained from TCI America
(Portland, Oregon), and naphthalene dialdehyde was obtained
from Invitrogen (Carlsbad, California).
Scheme 1. Formation of peroxy radicals from AIBN.
compounds with little reactivity toward peroxy radicals, the
reaction with alkoxy radicals therefore produces false positive
test results.
The role of AIBN peroxy radicals in generating artifactual
degradation was first indicated by Nelson et al.8 and recently
demonstrated by Watkins et al.14 The authors showed that ad-
dition of a small amount of methanol to the solvent system ef-
fectively prevented the alkoxy radicals from encountering and
reacting with the dilute drug substance, by providing a sacri-
ficial H atom donor. The use of methanol has been generally
adopted in the recent years12 and has greatly improved the cor-
relation of the azoalkane stress test results with the actual,
long term, drug product stability results.
It is within this framework that we report significant non-
oxidative, or artifactual degradation even with the use of
methanol solvent, for several amine-containing model drug
molecules oxidatively stressed using azoalkane initiators. By
briefly exploring the effects of several experimental parameters
on the artifact yield, including pH, initiator/drug molar ratio,
and the type of the azoalkane initiator, we attempted to learn
the significance of these findings to the pharmaceutical scien-
tist conducting oxidative stress testing for amine-containing
drug molecules. A variety of experiments were also conducted
to investigate the degradation mechanisms, including isotopic
labeling studies using 18O2, methanol-d3, and [13C,15N]-AIBN,
cyanide analysis, and experiments that probed the impact of
butylated hydroxyanisole (BHA), a peroxy radical scavenger,
on artifact formation. Chemical structures proposed based on
tandem and high-resolution MS data were subsequently con-
firmed using retention time, mass, and fragmentation pat-
tern correlations with compounds that were synthesized for
this purpose and characterized by nuclear magnetic resonance
(NMR).
EXPERIMENT AND ANALYSIS
Azoalkane-Based Oxidative Stress Testing Conditions
Several oxidative stress experiments were carried out for the
purpose of this paper. In general, a methanol/water solution
containing the model drug compound and the azoalkane ini-
tiator was thermally stressed at 40°C using AIBN or AAPH,
and the artifact yield was monitored over a given period of
time. Methanol solvent was used to quench the formation of
the alkoxy radicals.
In Experiment 1, all seven model drug molecules (fluvox-
amine maleate, aminodiphenylethanol, baclofen, norfloxacin,
carvedilol, propranolol HCl, or diphenhydramine HCl) shown in
Figure 1 were oxidatively stressed using either AIBN or AAPH
as the azoalkane initiators. Solutions consisting of 0.2 mg/mL
of each compound and 5 mM AIBN or AAPH in 55/45 (v/v)
methanol/water or 55/45 (v/v) methanol/0.1 M KCl/HCl buffer
(pH 2.0), acetate buffer (pH 4.0), or phosphate buffer (pH 6.0 or
8.0) were transferred to standard HPLC vials and placed on the
autosampler tray. The experiment was conveniently performed
by heating the sample vials directly on the autosampler tray,
which was maintained at 40°C. At the preset time points of 4,
8 (or 12), 24, 48, and 72 h, each sample solution was analyzed
by UPLC or HPLC, according to the procedure described below.
Experiment 2 was focused on determining the effect of the
initiator excess relative to the model drug. Initiator/drug mo-
lar ratios between 2.5 and 50 mol/mol were examined using
carvedilol solutions of 0.1, 0.25, 0.5, 1.0, and 2.0 mM, whereas
AIBN was maintained at 5 mM. Measurements were conducted
Materials
All chemicals were used as received. 2,2ꢀ-azobis(2-meth- as described above, at a pH of 6.0.
ylpropionitrile) (AIBN), 2,2ꢀ-azobis(2-methylpropionamidine)
In Experiment 3, 25 mM of BHA were added to solutions con-
dihydrochloride (AAPH), ( ) propranolol hydrochloride, taining 0.2 mg/mL norfloxacin and 5 mM AIBN, to investigate
(1R,2S)-(−)-2-amino-1,2-diphenylethanol (aminodiphenyleth- the effect of the peroxy radical inhibitor on the artifact yield. So-
anol), fluvoxamine maleate, ( ) baclofen, norfloxacin, lutions were prepared in ACN/MeOH/water 8:1:1 (v/v/v) rather
carvedilol, potassium cyanate, iodoacetonitrile, BHA, than MeOH/water, to minimize BHA inactivation through H-
methanol-d3, 18O2, and the following buffer materials: bonding with the solvent.15
ammonium hydroxide, sodium hydroxide (pellets), hydrochlo-
Stress test experiments using isotopically labeled reagents
ric acid (concentrated), potassium chloride, and sodium are described separately below.
Scheme 2. Recombination/decomposition of AIBN peroxy radicals.
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015
DOI 10.1002/jps.24667

RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
4289
Figure 1. Model drug compounds.
HPLC and Liquid Chromatography–Mass Spectrometry Analysis
MS Analysis
High-resolution and tandem mass spectrometric experiments
for structure characterization were carried out in the posi-
tive ion mode using an LTQ Orbitrap XL mass spectrometer
(Thermo Electron North America LLC) coupled with an elec-
trospray ionization source. A spray voltage of 4.5 kV, sheath
gas flow rate of 45 (in arbitrary units), capillary voltage of 30 V,
and capillary temperature of 275°C were used. High-resolution
data were acquired using a resolving power of 30,000. Tandem
MS experiments were performed using collision-induced dis-
sociation mode with structure-dependent normalized collision
energy setting of 20–30 (in arbitrary units). MS fragmentation
results can be found in the Supporting Information.
Reversed-phase chromatography using C18 columns and gra-
dient elution was carried out using either an Acquity UPLCꢁR
(Waters Corporation, Milford, Massachusetts) coupled to a sin-
gle quadrupole mass detector or an Agilent 1200 HPLC (Agilent
Technologies, Santa Clara, California) coupled to an LTQ Orbi-
trap XL mass spectrometer (Thermo Electron North America
LLC, West Palm Beach, Florida). Both systems were equipped
with binary pumps, heated autosampler trays and column com-
partments, and photodiode detectors. Chromatographic condi-
tions were optimized for each drug and radical initiator and
are compiled in Table S1 of the Supporting Information. The
autosampler tray was held at 40°C as required for stress test-
ing and the detection wavelength was the extracted 8max for
each compound (Table S1, Supporting Information). Artifact
yields were calculated as peak area % relative to the sum of
API-related peaks by UV, at the characteristic detection wave-
lengths of the model drug compounds. The mass of the reaction
products was monitored by liquid chromatography–mass spec-
trometry (LC–MS), at unit resolution.
Isotopic Labeling Experiments
Two oxidative stress experiments where either methanol or
AIBN were replaced by their isotopically labeled counterparts
methanol-d3 and [13C,15N]-AIBN, respectively, were conducted
using aminodiphenylethanol following the procedure described
earlier for Experiment 1. In a third experiment, the stress test
DOI 10.1002/jps.24667
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015

4290
RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
was carried out using aminodiphenylethanol in an atmosphere
enriched with 18O2. The experiment was performed following
the general procedure described earlier in Experiment 1 with
the following deviations. Deionized water was used in the place
of the buffer component of the sample solution. Deionized wa-
ter, AIBN stock solution, and aminodiphenylethanol stock so-
lution were individually purged with nitrogen, combined and
placed into a Parr pressure vessel (Moline, Illinois) resulting
in sample concentrations and solvent ratios described in Ex-
periment 1. The Parr vessel was pressurized to approximately
100 psi with nitrogen gas and vented to atmospheric pressure
and repeated five times. The Parr vessel was pressurized to ap-
proximately 22 psi using 18O2 gas and placed into a 40°C oven
for 72 h. All samples were analyzed by LC–MS. High-resolution
and tandem mass spectrometry measurements were conducted
to confirm the incorporation of the isotope labels into the degra-
dation products. The characterization of the labeled products is
presented in the Supporting Information.
Scheme 3. Non-oxidative degradation of primary and secondary
amines during peroxy radical-mediated stress using AIBN or ACVA.
spectrum and the 1D carbon spectra were referenced using the
TMS signal, set to 0.00 ppm. NMR spectra can be found in the
Supporting Information.
RESULTS AND DISCUSSION
We found that drug molecules containing primary or secondary
amine functionalities undergo significant non-oxidative or ar-
tifactual degradation under peroxy radical-mediated oxida-
tive stress using azoalkane initiators, even in the presence of
methanol. Several commercially available compounds were se-
lected as model drugs for the purpose of this paper; however, the
phenomena described were originally observed on proprietary
molecules containing secondary amine functionalities. In these
experiments, two degradation products or artifacts were de-
tected, both displaying UV absorbance spectra similar to those
of the respective APIs. The major degradant showed a char-
acteristic mass gain of 39 amu, whereas the minor degradant
showed a 43 amu gain, resulting from addition of C2HN and
CHON, respectively, as calculated based on high-resolution MS.
Moreover, fragmentation patterns indicated that the amine
groups of the original drug substances had been structurally
modified. Given these results and our evolving understanding
of the chemistry (vida infra), we proposed the structures shown
in Scheme 3, corresponding to an "-aminonitrile (AN; the ma-
jor artifact) and a urea derivative (UD; the minor artifact).
For the purpose of structural characterization, both compounds
were synthesized from representative primary and secondary
amine parents and characterized by LC–MS and NMR. Simi-
larities in the retention times, masses, and fragmentation pat-
terns confirmed that the products synthesized, and shown in
Scheme 3, are those formed in the azoalkane experiments. As
the mechanism of the artifactual degradation observed during
the oxidative stress test was not readily apparent, additional
experiments were carried out to understand both the scope of
the problem (i.e., false positive reactivity) and the underlying
mechanisms.
Cyanide Analysis
Cyanide levels were measured in thermally stressed samples
of AIBN using a derivatization procedure similar to that de-
scribed by Sano,16 followed by UPLCꢁR analysis of the resulting
2-cyano-1-ethanesulfonic acid benzoisoindole with UV detec-
tion. The amount of AIBN remaining was determined as peak
area % relative to peak area at time zero. Details are provided
in the Supporting Information.
Preparation of the Oxidative Artifacts for Structure Confirmation:
Synthesis and Preparative SFC
Aminodiphenylethanol and propranolol HCl were reacted with
iodoacetonitrile and potassium cyanate to produce the corre-
sponding "-aminonitrile (AN) and urea derivative (UD), re-
spectively. Product isolation for structure identification was
performed using a Berger Minigram SFC system (Waters Cor-
poration) equipped with a variable wavelength detector. Sol-
vent from collected fractions was removed by centrifugal evapo-
ration using a Genevac EZ-2 Plus solvent evaporator (Genevac,
Ipswich, UK). All methods used a CO2 mobile phase with
methanol modifier and gradient elution. Detection wavelength
was 215 nm except for the UD of aminodiphenylethanol which
was detected at 220 nm. Injection volumes were between 400
and 500 :L. Experimental details on the syntheses of these ar-
tifacts and the SFC methods for their individual isolation can
be found in the Supporting Information.
Nuclear Magnetic Resonance
All 1D and 2D data were collected at 298 K using a Bruker-
Biospin 5 mm BBFO probe on a AVANCE III NMR spec-
trometer (Bruker-Biospin, Billerica, Massachusetts) operating
Product Yields Under Routine Oxidative Stress Conditions and the
Effect of Experimental Parameters
at 500 MHz for milligram scale samples or using a Bruker- Seven model compounds (Fig. 1) containing primary, sec-
Biospin 5 mm TCI cryoprobe on a AVANCE III NMR spec- ondary, and tertiary amine functionalities were chosen to in-
trometer operating at 600 MHz for a submiligram sample. vestigate the novel degradation chemistry. Except for amin-
The following data were collected: 1D proton, 1D carbon, 1H- odiphenylethanol, all of the model compounds are known active
1H gradient COSY (correlation spectroscopy), 1H-13C multiplic- pharmaceutical ingredients. In order to determine the general-
ity edited HSQC (heteronuclear single quantum coherence), ity of the artifactual reaction for amines, the compounds were
and 1H-13C HMBC (heteronuclear multiple bond correlation), subjected to AIBN stress under conditions similar to those used
1H-1H NOESY (nuclear overhauser effect spectroscopy), or in the original experiments (described under Experiment 1).
1H-1H ROESY (rotating-frame overhauser enhancement spec- Oxidatively stressed samples were analyzed by LC–MS to ob-
troscopy). A typical sample was dissolved in 0.6 mL of 99.9% serve the appearance of the degradation products with mass
deuterated dimethylsulfoxide (DMSO-d6). Both the 1D proton gains of 39 and 43 amu, corresponding to the AN and UD,
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015
DOI 10.1002/jps.24667

RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
4291
Figure 3. AN and UD yields (as area% relative to the sum of related
peaks, by UV) after 72 h, at 40°C in MeOH/buffer (pH 6.0) 55:45 (v/v),
at 0.1, 0.25, 0.5, 1, and 2 mM carvedilol and 5 mM AIBN.
solutions of 0.2 mg/mL, fall between 5 and 10 mol/mol (Table 1).
As shown in Figure 3, the artifact yields (especially for the ma-
jor degradant) decreased significantly when the AIBN/drug mo-
lar ratio was less than 10, explaining to some extent the spread
in reaction rates observed in Figure 2. Overall however, the ar-
tifact yields remained high enough to be problematic over the
range of AIBN/drug molar ratios routinely used for the stress
test, which is about 3–10 mol/mol.
The effect of varying the pH over a range of 2–8 was inves-
tigated by replacing the water component of the diluent, with
0.1 M buffers, as described in Experiment 1. The final pH of
the buffered solution containing the model drug and AIBN was
measured for each preparation and was used to generate the
data plots displayed in Figure 4. The pH of the MeOH/water
preparation was also measured and plotted in Figure 4. The
complete data set can be found in Tables S4 and S5 of the
Supporting Information. The results demonstrate that artifact
formation was generally favored by increasing pH, with AN
yields reaching maximum levels at pH 7–8. Both degradants
remained undetected below a pH of 2 (AN) or 4 (UD). Over-
all, the pH trend is consistent with the amine groups playing
roles of nucleophiles in the degradation pathway, the lower AN
yields at higher pHs being attributed to the instability of the
intermediate species in the proposed pathways (vida infra).
Figure 2. AN (a) and UD (b) yields (as area% relative to the sum of
related peaks, by UV) at 40°C in MeOH/water 55:45 (v/v), at 0.2 mg/mL
drug and 5 mM AIBN.
respectively. Yields were then calculated as peak area% rela-
tive to the sum of the API-related peaks, by UV. Results indicate
that artifact formation appears to be a general phenomenon for
primary and secondary amines under the conditions specified,
but not for tertiary amines. Although reaction rates vary sig-
nificantly, AN remained by far the major degradant with yields
ranging from 1% to 13% after 3 days (Fig. 2a), whereas UD was
measured at 0.2%–1.2% (Fig. 2b). In some cases, additional low
intensity peaks were observed, but none were investigated.
The effect of the AIBN/drug molar ratio was examined for
carvedilol over a range of 2.5–50 mol/mol, at a fixed AIBN
Stable Isotope Labeling Studies
concentration of 5 mM (see Experiment 2 for details). For In order to understand the source of the new C, N, and O atoms
comparison, the molar ratios calculated for the model drug in the AN and UD degradants, azoalkane oxidative stress tests
Table 1. Summary of Molar Concentration, Native pH in the MeOH/Water Diluent, Molar Ratio AIBN/Drug, Yield of "-Aminonitrile Artifact as
Area% Relative to the Sum of Drug-Related Peaks by UV, and Molar Concentration of the "-Aminonitrile Artifact at pH 6, for All Model Drugs
Native pHa
pKa
AIBN/Drug (mol:mol)
% AN after 72 hc
AN (mM)c
b
Model Compound
Model Drug (mM)
Aminodiphenylethanol
Fluvoxamine maleate
Baclofen
Norfloxacin
Carvedilol
0.94
0.63
0.93
0.63
0.49
0.68
8.1
6.3
6.6
7.3
7.9
5.1
8.9
5.3
8.0
5.4
8.0
10.2
7.4
2.3
3.7
1.4
14.8
9.6
0.3
0.022
0.023
0.013
0.093
0.047
0.002
9.4
10.3
8.7
8.2
9.5
Propranolol HCl
aMeasured in the MeOH:water (55:45, v/v) reaction mixture.
bCalculated using ACD/ChemSketch, ver 11.02.
cMeasured in the MeOH:buffer (pH 6) (55:45, v/v) reaction mixture.
DOI 10.1002/jps.24667
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015

4292
RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
were carried out using isotopically labeled AIBN, methanol, or
O2, whereas the molecular weights of the products were moni-
tored by LC–MS. Experimental details can be found in Exper-
iment and Analysis section, whereas the data are summarized
in Scheme 4. By replacing AIBN with [13C,15N]-AIBN, a mass
gain of 2 amu was observed for both the AN and UD artifacts,
indicating that the labeled nitrile group was incorporated into
the products. The source of the new O atom in the UD degradant
was probed by saturating the sample solution with 18O2. The
observed mass increase of 2 amu indicated that the new O
atom derived from the molecular oxygen dissolved in solution,
rather than from water or methanol. The proposed pathway for
this incorporation begins with the addition of molecular oxy-
gen to the C-centered AIBN radical generated after nitrogen
expulsion, as shown in Scheme 1. Lastly, a mass gain of 2 amu
corresponding to two deuterium atoms, was observed for the
AN degradant after replacing the methanol with methanol-d3.
Incorporation of two deuterium labels from the –CD3 group
conclusively indicates that the methanol solvent is the origin of
the last unknown atoms in the AN degradant, the methylene
group.
Formation of AN in AIBN-Stressed Samples
The proposed mechanism for the formation of the AN
degradants is essentially an “accidental” Strecker reaction.17–19
The Strecker reaction is a well-known process in which "-
aminonitriles are synthesized from ammonia or amines and
carbonyl compounds in aqueous solutions containing alkaline
cyanide. During AIBN stress testing of pharmaceutical com-
pounds, the Strecker reagents (cyanide and formaldehyde)
Figure 4. AN (a) and UD (b) yields (as area% relative to the sum of
related peaks, by UV) after 72 h, at 40°C in MeOH/buffer 55:45 (v/v),
at 0.2 mg/mL substrate and 5 mM AIBN, as a function of pH.
Scheme 4. Isotope labeling results using aminodiphenylethanol.
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015
DOI 10.1002/jps.24667

RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
4293
Scheme 5. Formation of AN from reaction of drug molecules with formaldehyde and cyanide (lower) generated in situ from reaction of 2-cyano-
2-propyl peroxy radicals with methanol.
Table 2. Summary of the Cyanide Levels Produced in Thermally
Stressed AIBN Solution
are generated in situ as byproducts of the stress reaction
(Scheme 5). We found that methanol, used to quench the AIBN
alkoxy radicals, plays a key role in this process. By abstracting
H-atoms from methanol, alkoxy radicals form acetone cyanohy-
drin, which further dissociates into HCN and acetone,20 the
equilibrium being quickly established in the presence of ba-
sic compounds.21 The resultant methanol radicals quickly oxi-
dize to peroxy radicals, which can disproportionate to generate
formaldehyde. The drug amine group reacts with formalde-
hyde to form the imine intermediate, which further reacts
with cyanide (Scheme 5), explaining the incorporation of the
nitrile and methylene groups observed during the labeling
experiments. Previous reports showing formation of cyanide
and acetone cyanohydrin during AIBN-initiated oxidation of
benzene,22 as well as formation of formaldehyde during the ox-
idative stress test carried out in the presence of methanol,8 sup-
port the proposed mechanism. Table 2 shows our own measure-
ments of cyanide levels generated during the thermal stress
of AIBN in the absence of the test drug. Data show that the
cyanide concentration increases in time and is proportional
to the extent of AIBN decomposition. Interestingly, the high-
est AN concentrations attained during our experiments (nor-
floxacin, pH 6, Table 1) were equimolar with respect to cyanide,
indicating that cyanide is the limiting reagent in the Strecker
reaction. The decrease in the AN yields at high pH values
(Fig. 4) is also consistent with the proposed mechanism: cyanide
and hydroxide ions react competitively with the imine interme-
diate in Scheme 5. Above pH 9, the hydroxide concentration
becomes comparable with that of the cyanide (Table 2), causing
a decrease in the yield of the AN product.
AIBN
Degraded AIBN
Time
AIBN
AIBN Degraded
CN
(mM)
Converted to
point (h) (% Initial) (mM)
(mM)
CNa (%)
0
100
96.4
92.6
88.0
5.24
5.05
4.85
4.61
–
0.015
0.042
0.068
0.093
–
24
48
72
0.19
0.38
0.63
9.9
9.0
7.5
All values are average of n = 3.
aNote that each molecule of AIBN can generate two CN ions.
Formation of the UD in AIBN-Stressed Samples
The detailed mechanism for the formation of the UD
degradants is more difficult to support. Although the origin
of the new C, N, and O atoms in the molecule was determined
through isotope labeling studies (Scheme 4), critical informa-
tion about the reaction pathway was revealed through BHA
experiments (see Experiment 3 for details). By conducting the
oxidative stress test in the presence of BHA, the AIBN per-
oxy radicals are expected to be effectively quenched and con-
verted to AIBN hydroperoxide (ROOH). Therefore, any changes
in the degradant yields caused by addition of BHA should in-
dicate a direct correlation with the AIBN peroxy radicals, or
any of the species deriving from them (Scheme 1). Note that in
order to avoid inactivation of BHA through H-bonding with
solvent molecules, these experiments were carried out in a
DOI 10.1002/jps.24667
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015

4294
RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
solvent consisting primarily of ACN (80%).15 The fraction of
MeOH was reduced to 10%, which has been previously shown
to remain sufficiently effective in quenching alkoxy radicals.14
Results show that addition of 25 mM of BHA effectively in-
hibits the formation of the AN artifact, whereas the UD levels
remain practically unchanged (Fig. 5). These observations in-
dicate that species deriving from the AIBN peroxy radicals are
key to the formation of the AN artifact, consistent with our pro-
posed mechanism. In contrast, the lack of effect on UD levels
indicates that the AIBN peroxy radicals are not key to the for-
mation of UD. Recall that the isotope labeling studies showed
incorporation of the AIBN nitrile group and molecular oxygen
into the UD molecule (Scheme 4), suggesting the participation
of AIBN oxidation products to the formation of this artifact. We
therefore propose that an alternate oxidation pathway of AIBN
accounts for this incorporation, involving the formation of an
AIBN peroxide within the AIBN “solvent cage,” as shown in the
top portion of Scheme 6. Formation of AIBN peroxy radicals
(Scheme 1) follows the expulsion of molecular nitrogen and the
rapid oxygenation of the carbon-centered radicals at diffusion-
controlled rate in oxygen-saturated solutions.23,24 Geminate re-
combination between the peroxy radical and the remaining C-
centered radical may, however, occur within the solvent cage,25
producing an AIBN peroxide. We propose that the UD products
originate in the AIBN peroxide, which further breaks down to
ultimately form isocyanic acid (Scheme 6), known to generate
UDs in reaction with amines.26 This pathway was not affected
by the presence of BHA in the bulk solution, consistent with the
proposal of rapid reaction within the solvent cage, explaining
the lack of correlation between the yield of the UD product and
the addition of BHA.
Figure 5. AN (a) and UD (b) yields for norfloxacin (0.2 mg/mL) in the
absence of BHA [MeOH/water 55:45 (v/v)] and in the presence of 25
mM BHA [ACN/MeOH/water 8:1:1 (v/v/v)], at 40°C and 5 mM AIBN.
Choice of Azoalkane Initiator to Minimize AN and UD Artifacts
We have shown that primary and secondary amines undergo
artifactual degradation reactions that may reach significant
Scheme 6. Proposed formation of UD through AIBN oxidation in the solvent cage to form the AIBN peroxide, followed by conversion to isocyanic
acid and reaction with drug molecule.
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015
DOI 10.1002/jps.24667

RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
4295
Scheme 7. Oxidative decomposition products of AIBN, ACVA, and AAPH.
yields under the conditions routinely used for oxidative stress completion, the individual products were purified by SFC
testing using AIBN and ACVA. Observing false positive re- and characterized by NMR and LC–MS. The retention times,
activity for drug compounds is undesirable, as it may impact masses, and fragmentation patterns of the synthesized com-
resources and development timelines due to efforts required pounds were matched with those of the corresponding artifacts
to address chemical stability issues. Therefore, taking steps to detected in the stressed samples.
avoid such unrepresentative reactivity may need to be consid-
Accurate mass measurements of the synthesized artifacts
ered, unless mass measurements are routinely carried out for (Table 3) show molecular ions [M+H]+ that correlate to the
degradant profiling (which would allow a quick way to differ- protonated empirical formulas consistent with the structures
entiate these artifacts). As discussed in this paper, decreasing shown. Fragmentation patterns, illustrated in Figures S2, S4,
AIBN/drug molar ratio or solution pH may reduce or even in- S6, and S8 of the Supporting Information section, are also con-
hibit artifact formation. Other options, including the choice of sistent with the proposed structures.
solvent or azoalkane initiator may also be available to the phar-
maceutical scientist.
One dimensional and extensive 2D NMR experiments were
performed for the assignments of the proton and carbon spec-
Based on the mechanism described in Scheme 5, azoalkane tra for the AN and UD artifacts of aminodiphenylethanol and
initiators lacking the nitrile group should not produce cyanide propranolol (Fig. 6). The data are consistent with the struc-
during thermal stress, and therefore would not produce AN ar- tures shown in Table 3 and the proton and carbon assign-
tifacts in reaction with amines. For example, the oxidative de- ments for each compound. For instance, the carbon spectrum
composition of three commonly used initiators, AIBN, ACVA, of the AN artifact of aminodiphenylethanol shows a quater-
and AAPH27 is summarized in Scheme 7. As expected, cyanide nary signal at 118.5 ppm, indicative of a cyano group car-
is generated from AIBN and ACVA, through decomposition of bon. In the HMBC spectrum, this carbon resonance shows
the corresponding cyanohydrins, but not from AAPH. The in- correlations to the methylene protons at 3.57 and 3.09 ppm,
ability of AAPH to induce AN formation was confirmed for both and a three-bond correlation to the amino group proton at
primary and secondary amines at pH values ranging between 2.64 ppm. Together with the COSY and HSQC data, the re-
2 and 8, as described in Experiment 1. In the case of the UD sults indicate that the methylene group of the AN is attached
artifacts, ACVA is expected to parallel AIBN and generate iso- to the nitrogen atom of aminodiphenylethanol. The UD product
cyanic acid as described in Scheme 6; however, the mechanism shows characteristic urea carbonyl signal at 157.7 ppm as well
cannot be directly translated to AAPH. Although formation of as the unique amino protons singlet at 5.52 ppm. The signal
the UD artifacts was observed in this case, the reaction path- of the alkyl urea amino proton appears as a doublet at 6.53
way was not investigated.
ppm that is coupled to a proton at 4.79 ppm by COSY data.
This proton signal, assigned as the nitrogenated methine on
the basis of the attached carbon resonating at 58.8 ppm by
HSQC experiment, shows a key three-bond HMBC correlation
to the urea carbonyl signal, establishing the connectivity be-
tween the amino group of the starting material and the urea
carbonyl. The other artifacts of propranolol were similarly char-
Structure Confirmation—Synthesis of Artifacts, LC–MS, and NMR
For structure confirmation purposes and peak tracking stud-
ies, the AN and UD artifacts of representative primary and
secondary amine model drug compounds were synthesized as
detailed in Experiment and Analysis section. For example,
aminodiphenylethanol and propranolol were reacted with
iodoacetonitrile and potassium cyanate to produce the corre-
sponding AN and UD artifacts, respectively. After reaction
1
acterized. The H NMR and 13C NMR spectra of the artifacts
can be found in Figures S10–S17 of the Supporting Information
section.
DOI 10.1002/jps.24667
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015

4296
RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
Table 3. Accurate Mass Measurements for Synthetically Derived "-Aminonitrile (AN) and Urea Derivative (UD) Artifacts of
Aminodiphenylethanol and Propranolol
Theoretical
Mass Mea-
surement
(m/z)
Protonated
Empirical
Formula
Protonated
Monoisotopic
Mass (Da)
Mass
Deviation
Artifact
AN
Artifact Structure
253.13338
C16H17ON2
253.13354
−0.6
−0.9
UD
257.12823
C15H17O2N2
257.12845
AN
299.17509
303.17044
C18H23O2N2
299.17540
303.17032
−1.0
UD
C17H23O3N2
0.4
CONCLUSIONS
drug molecules, significant yields may be observed for these
artifacts. These findings should help simplify data interpreta-
tion during the peroxy radical-mediated oxidative stress testing
by allowing the pharmaceutical scientist to readily understand
that degradation products with a mass gain of either 39 amu
(AN) or 43 amu (UD) over the amine parent, are most likely ar-
tifacts and therefore should be disregarded. Such degradants
are not characteristic to the peroxy radical-oxidation chemistry.
The most effective strategies to minimize or prevent artifact for-
mation include lowering the reaction pH and switching to al-
ternative azo initiators that do not produce AN artifacts, such
as AAPH. Decreasing the initiator/drug ratio was also shown to
reduce artifact yields; however, an excess of initiator should be
maintained as recommended elsewhere.12,14 Alternatively, LC–
MS can be used to simply identify and exclude the artifacts,
following mass identification.
We show that primary and secondary amines undergo arti-
factual degradation during azoalkane-based oxidative stress
testing even when carried out in the presence of signifi-
cant amounts of methanol. These observations are notable, as
methanol is routinely utilized as co-solvent to reduce the oc-
currence of “false positive” degradants, and improve the qual-
ity of the oxidative stress test as a probe for the propen-
sity of molecules to autoxidize via peroxy radical initiated
processes.12 The artifacts, an AN and UD (Scheme 3) are
not products of autoxidation, but arise from the reaction be-
tween the amine substrate and byproducts of the stress re-
action. Mechanistic studies revealed that AN is produced via
a Strecker reaction between the amine group of the drug
molecule and formaldehyde and HCN generated during the
oxidative decomposition of nitrile-containing azoalkane ini-
tiators. UD species are believed to arise via a more com-
plex mechanism, from the rearrangement and decomposition
of an AIBN peroxide intermediate to yield isocyanic acid,
which can easily react with amines to produce urea deriva-
tive. We show that under the conditions routinely used in
the pharmaceutical industry for oxidative stress testing of
ACKNOWLEDGMENTS
The authors would like to thank Dr. Fangming Kong, Ron
Morris, and Victor Soliman for conducting NMR and MS ex-
periments to aid in the structure elucidation of the artifacts.
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015
DOI 10.1002/jps.24667

RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
4297
Figure 6. NMR chemical shif assignment data for AN and UD artifacts of aminodiphenylethanol and propranolol.
REFERENCES
12. Baertschi SW, Jansen PJ, Alsante KM. 2011. Stress testing: A pre-
dictive tool. In Pharmaceutical stress testing: Predicting drug degra-
dation; Baertschi SW, Alsante KM, Reed RA, Ed. 2nd ed. New York:
Informa Health Care, pp 10–48.
13. Hendry DG, Mill T, Piszkiewicz L, Howard JA, Eingenmann HK.
1974. A critical review of H atom transfer in the liquid phase: Chlorine
atom, alkyl, trichloromethyl, alkoxy, and alkylperoxy radicals. J Phys
Chem Ref Data 3:937–978.
14. Watkins MA, Pitzenberger S, Harmon PA. 2013. Direct evidence of
2-cyano-2-propoxy radical activity during AIBN-based oxidative stress
testing in acetonitrile-water solvent systems. J Pharm Sci 102:1554–
1568.
15. Foti MC. 2007. Antioxidant properties of phenols. J Pharm Phar-
macol 59:1673–1685.
16. Sano A, Takimoto N, Takitani S. 1992. High-performance liquid-
chromatographic determination of cyanide in human red-blood-cells
by precolumn fluorescence derivatization. J Chromatogr-Biomed Appl
582:131–135.
17. Strecker A. 1850. Ueber die Ku¨nstliche Bildung der Milchsa¨ure
und einen neuen, dem Glycocoll Homologen Ko¨rper. Ann Chem Pharm
75:27–45.
18. Galletti P, Pori M, Giacomini D. 2011. Catalyst-free strecker reac-
tion in water: A simple and efficient protocol using acetone cyanohydrin
as cyanide source. Eur J Org Chem 3896–3903.
19. Wang J, Liu XH, Feng XM. 2011. Asymmetric strecker reactions.
Chem Rev 111:6947–6983.
1. Baertschi SW, Jansen PJ, Alsante KM. 2011. Stress testing: A pre-
dictive tool. In Pharmaceutical stress testing: Predicting drug degra-
dation; Baertschi SW, Alsante KM, Reed RA, Ed. 2nd ed. New York:
Informa Health Care, pp 10–48.
2. ICH. 2014. Assessment and control of DNA reactive (mutagenic)
impurities in pharmaceuticals to limit potential carcinogenic risks M7.
3. Waterman KC, Adami RC, Alsante KM, Antipas AS, Arenson DR,
Carrier R, Hong JY, Landis MS, Lombardo F, Shah JC, Shalaev E,
Smith SW, Wang H. 2002. Hydrolysis in pharmaceutical formulations.
Pharm Dev Technol 7:113–146.
4. Hovorka SW, Schoneich C. 2001. Oxidative degradation of pharma-
ceuticals: Theory, mechanisms and inhibition. J Pharm Sci 90:253–269.
5. Waterman KC, Adami RC, Alsante KM, Hong JY, Landis MS,
Lombardo F, Roberts CJ. 2002. Stabilization of pharmaceuticals to
oxidative degradation. Pharm Dev Technol 7:1–32.
6. Johnson DM, Gu LC. 1988. Autoxidation and autoxidants. In Ency-
clopedia of pharmaceutical technology; Swarbrick J, Boylan JC, Ed. 1st
ed. New York: Wiley, pp 415–450.
7. Harmon PA, Boccardi G. 2011. Oxidative susceptibility testing. In
Pharmaceutical stress testing: Predicting drug degradation; Baertschi
SW, Alsante KM, Reed RA, Ed. 2nd ed. New York: Informa Health
Care, pp 168–191.
8. Nelson ED, Thompson GM, Yao Y, Flanagan HM, Harmon PA. 2009.
Solvent effects on the AIBN forced degradation of cumene: Implications
for forced degradation practices. J Pharm Sci 98:959–969.
9. Lienard P, Gavartin J, Boccardi G, Meunier M. 2015. Predicting drug
substances autoxidation. Pharm Res 32:300–310.
10. Boccardi G, Deleuze C, Gachon M, Palmisano G, Vergnaud JP. 1992.
Autoxidation of tetrazepam in tablets—Prediction of degradation impu-
rities from the oxidative behavior in solution. J Pharm Sci 81:183–185.
11. Boccardi G. 1994. Autoxidation of drugs—Prediction of degrada-
tion impurities from results of reaction with radical-chain initiators.
Farmaco 49:431–435.
20. Li C-H, Stewart TD. 1937. Comparison of the basicity of aliphatic
amines in different solvents. J Am Chem Soc 59:2596–2599.
21. Stewart TD, Fontana BJ. 1940. The effect of solvation upon the
dissociation of acetone cyanohydrin. J Am Chem Soc 62:3281–3285.
22. Talat-Erben M, Onol N. 1960. The reaction of 2-cyano-2-propyl free
radicals with oxygen. Can J Chem 38:1154–1157.
23. Maillard B, Ingold KU, Scaiano JC. 1983. Rate constants for the
reactions of free-radicals with oxygen in solution. J Am Chem Soc
105:5095–5099.
DOI 10.1002/jps.24667
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015

4298
RESEARCH ARTICLE – Pharmaceutics, Drug Delivery and Pharmaceutical Technology
24. Neta P, Huie RE, Ross AB. 1990. Rate constants for reactions of 26. Stark GR. 1965. Reactions of cyanate with functional groups of
peroxyl radicals in fluid solutions. J Phys Chem Ref Data 19:413– proteins. III. Reactions with amino and carboxyl groups. Biochemistry
513.
4:1030–1036.
25. Bevington JC. 1956. Reaction between diphenylpicryl- 27. Werber J, Wang YJ, Milligan M, Li XH, Ji JA. 2011. Analysis of 2,2 ‘-
hydrazyl and 2-cyano-2-propyl radicals.
1132.
J
Chem Soc 1127– azobis (2-amidinopropane) dihydrochloride degradation and hydrolysis
in aqueous solutions. J Pharm Sci 100:3307–3315.
Nefliu et al., JOURNAL OF PHARMACEUTICAL SCIENCES 104:4287–4298, 2015
DOI 10.1002/jps.24667