Kinetic understanding using NMR reaction profiling

Article abstract of DOI:10.1021/op200202k

The combination of kinetic understanding and reaction modeling has been successfully applied to the development of processes from laboratory to manufacturing plant. Although extensively used in bulk chemistry, polymers, and the oil industry [ Bayer Technology Services, http://www.bayertechnology.cn/ uploads/media/0707-e-300dpi.pdf, July 2011; Lawrence Livermore National Laboratory, http://www1.eere.energy.gov/vehiclesandfuels/pdfs/merit-review-2011/ fuel-technologies/ft010-pitz-2 011-o.pdf,July 2011; Shin, S. B.; Han, S. P.; Lee, W. J.; Chae, J. H.; Lee, D. I.; Lee, W. H.; Urban, Z.Hydrocarbon Process. 2007, April) 83; Baumer, C.; Urban, Z.Hydrocarbon Process. 2007, June) 71 ], it has not been exploited to its full potential in the pharmaceutical industry. We present a fast and efficient methodology for kinetic modeling of chemical reactions using ¹H NMR reaction monitoring that can be used for the process understanding and development of active pharmaceutical ingredients. The parameters that are important for the development of a good, reliable model for the prediction and optimization of reaction conditions are discussed. The hydrolysis of acetic anhydride was chosen to illustrate the methodology because it is mechanistically and kinetically well established.

Full text of DOI:10.1021/op200202k

ARTICLE
pubs.acs.org/OPRD
Kinetic Understanding Using NMR Reaction Profiling
Flavien Susanne,* David S. Smith, and Anna Codina*
Chemical Research and Development and Analytical Research and Development, Pfizer World Wide Research and Development,
Sandwich Laboratories, Ramsgate Road, Sandwich, Kent CT13 9NJ, United Kingdom
ABSTRACT: The combination of kinetic understanding and reaction modeling has been successfully applied to the development
of processes from laboratory to manufacturing plant. Although extensively used in bulk chemistry, polymers, and the oil industry
[Bayer Technology Services, http://www.bayertechnology.cn/uploads/media/0707_e_300dpi.pdf, July 2011; Lawrence Liver-
more National Laboratory, http://www1.eere.energy.gov/vehiclesandfuels/pdfs/merit_review_2011/fuel_technologies/ft010_pitz_
2 011_o.pdf, July 2011; Shin, S. B.; Han, S. P.; Lee, W. J.; Chae, J. H.; Lee, D. I.; Lee, W. H.; Urban, Z. Hydrocarbon Process.
2007, April, 83; Baumer, C.; Urban, Z. Hydrocarbon Process. 2007, June, 71], it has not been exploited to its full potential in the
pharmaceutical industry. We present a fast and efficient methodology for kinetic modeling of chemical reactions using 1H NMR
reaction monitoring that can be used for the process understanding and development of active pharmaceutical ingredients. The
parameters that are important for the development of a good, reliable model for the prediction and optimization of reaction
conditions are discussed. The hydrolysis of acetic anhydride was chosen to illustrate the methodology because it is mechanistically
and kinetically well established.
’ INTRODUCTION
function of the concentration of both starting material
([Ac2O]) and reagent ([H2O]), eq II. The Arrhenius eq III
gives the dependence of the rate constant (k) of hydrolysis
on the temperature (in Kelvin) and activation energy (Ea), A
being the pre-exponential Arrhenius factor and R the gas
constant.
In the conditions of a high excess of water, the hydrolysis of
acetic anhydride can be considered to be pseudo-first-order,
simplifying the kinetic parameters calculation. The integrated
rate law can be rearranged such as the eq IV where [Ac2O]0
is the initial concentration of acetic anhydride. Plotting
ln[Ac2O]/[ Ac2O]0 against time creates a straight line with
slope ꢀk. This can be repeated for different temperatures.
From the Arrhenius eq V, a plot ln k versus 1/T produces a
straight line where Ea and ln A can be determined from the
slope and the intercept.
Reaction modeling using NMR data has been the subject of
multiple research papers.5ꢀ7 NMR together with in silico meth-
ods is a powerful tool for the study of reaction kinetics. NMR
spectroscopy provides structurally rich data, from which it is
possible to confirm or elucidate the structure of starting materi-
als, intermediates, and products.8 It is intrinsically quantitative,
overcoming the need to determine response factors required by
other techniques such as HPLC. It is not intrusive, and it is
potentially able to detect every compound involved in the
reaction provided NMR active nuclei are present and there is
sufficient signal dispersion. Despite its power, NMR is not the
technique of choice in pharmaceutical process development.9
Even in the early stages of development, reaction monitoring by
NMR tends to be only sporadically used by NMR and/or
chemical engineering specialists. Nevertheless, the pharmaceu-
tical industry is giving increased levels of attention to reaction
understanding as a way to ensure the quality of the active
pharmaceutical ingredient (API) and manufacturing process
robustness throughout its entire lifecycle.
ln½Ac2Oꢁ=½Ac2Oꢁ0 ¼ ꢀ kt
ðIVÞ
The mechanism of the hydrolysis of acetic anhydride has been
extensively studied during the last century.10 The intermediate
scheme proceeds via three irreversible steps: addition of water,
elimination, and proton transfer (Scheme 1), the addition being
the rate-limiting step.11
ln k ¼ ꢀ Ea=RT þ ln A
ðVÞ
First-order reactions are rare in the pharmaceutical industry.
Therefore, in this study we treated the hydrolysis of acetic
anhydride as a second-order type of reaction to demonstrate
our methodology for more general application.
Ac2O þ H2O f 2AcOH
ðIÞ
In this contribution, the kinetic parameters A and Ea were
calculated using Dynochem12 from a series of experiments includ-
ing variation of (1) initial concentrations of reagents and (2) the
ꢀ d½Ac2Oꢁ=dt ¼ k½Ac2Oꢁ ½H2Oꢁ
3
1
¼ A eꢀE =RT ½Ac2Oꢁ ½H2Oꢁ
ðIIÞ
a
reaction temperatures. H NMR was used to monitor the
3
3
3
evolution of starting material and product concentration as a
k ¼ A eꢀE =RT
ðIIIÞ
function of time.
a
3
The overall reaction can be represented in a simplified eq I.
The rate of hydrolysis, ꢀd[Ac2O]/dt, can be expressed as a
Received: August 1, 2011
Published: October 26, 2011
r
2011 American Chemical Society
61
dx.doi.org/10.1021/op200202k Org. Process Res. Dev. 2012, 16, 61–64
|

Organic Process Research & Development
ARTICLE
Scheme 1. Mechanism of acetic anhydride hydrolysis by water
Table 1. Summary of experimental reaction conditions and
NMR acquisition parametersa
temperature (°C)
D2O
acetic anhydride
acquisition
spectro-
mass
(mg)
[ ]
pad total
N
t0
bath
meter
V (μL)
(M) (min) (h)
1
2
3
∼50 24.8
∼50 14.8
25
15
5
750
750
750
750
750
78.0
77.9
77.7
38.4
19.2
0.91
0.91
0.91
0.47
0.24
1
2
1
3
50 500
5.0
10
1
10
2
4 401500 24.5
5 401900 24.9
25
25
2
4
a N = experiment number, t0 = time between initiation of the reaction
and starting the acquisition, V = volume, [] = concentration, pad = pre-
acquisition delay.
’ EXPERIMENTAL SECTION
The reaction was performed at different concentrations and
temperatures (Table 1). Each reaction was prepared by the
addition of an appropriate amount of acetic anhydride and 750
μL of deuterium oxide (D2O) into a test tube placed in an oil
bath set at the required temperature. They were then mixed in
the test tube, transferred to a 5 mm NMR tube, also at the same
temperature, and immediately placed in the NMR spectrometer
(Varian Inova 500 MHz equipped with a 5 mm triple-resonance
(1H/15N/13C) probe), set at the same temperature as that of the
bath. Once in the NMR spectrometer, the homogeneity of the
magnetic field was adjusted by gradient shimming on the z-axis,
Figure 1. 1H NMR spectra (aliphatic region expansion) of the acetic
anhydride (0.47 M, 25 °C) hydrolysis vs number of scan (s). The peaks
correspond to the methyl protons of the acetic anhydride (left) and the
acetic acid (right).
Dynochem was chosen to build a kinetic model based on the
rate eq II, which describes the hydrolysis of the acetic anhydride
assuming first-order kinetics for both acetic acid and water
concentration. Ea and A are adjustable kinetic parameters.
The time-resolved experiments (time vs concentration) for
different conditions were imported into the program and were
used to solve the rate equation and calculate the optimum value
for Ea and A. The LevenbergꢀMarquardt fitting algorithm was
used, which calculates different theoretical reaction profiles by
iterating Ea and A, and looks for the minimal deviation between
calculated and experimental data. This is expressed as the sum of
the squares deviation (SSQ) in eq VI where j is the experiment
number, M the number of experiments, i the point number and N
the number of points, (yij ꢀ yjmod) is the difference between
experimental and calculated values and wj a weighing factor for
each of the different experiments.
1
and an array of H experiments was acquired. The probe had
previously been tuned and matched with a sample of similar
composition. The delay between initiation of the reaction and
starting acquisition was accurately measured for most of the
reaction conditions (Table 1). The duration of the pulse program
was also taken into account to obtain an accurate time value. The
course of the reaction was followed by 1H NMR. Key parameters
are contained in Table 1.
Data were processed and analyzed using MestReNova, 6.1.1,
build 6384 (MestreLab Research). The regions of interest were
integrated, and the values obtained were transferred to Dyno-
chem (3.2.1.0) in order to build a kinetic model.
’ RESULTS AND DISCUSSION
M
N
NMR experiments were acquired as described in the Experi-
mental Section. The experiments were designed to cover a wide
range of concentration of substrate and of temperature to enable
a good understanding across a large design space. Two peaks were
observed in the spectra (Figure 1): one depleting and the other
increasing with time. They correspond to the methyl protons of the
acetic anhydride (2.19ꢀ2.03 ppm) and acetic acid (2.01ꢀ1.86
ppm), respectively. These two regions were integrated, and the
values of the integrals were converted to moles.
2
SSQ ¼
wj
∑ ∑
ðy ꢀ yijmodÞ
ðVIÞ
ij
j¼1
i¼1
The program reports the best values for Ea and A after
convergence or after a predefined maximum number of itera-
tions. Ea and A are reported with an associated confidence
interval, expressed as a percentage of the parameter value.
In the case of the acetic anhydride hydrolysis, values of ln
A = 8.05 ( 1.02 sꢀ1 and Ea = 11.4 ( 0.4 kcal/mol were obtained
62
dx.doi.org/10.1021/op200202k |Org. Process Res. Dev. 2012, 16, 61–64

Organic Process Research & Development
ARTICLE
Figure 3. Reaction profile for experiment 2 conducted with 0.91 M
acetic anhydride at 15 °C. The depletion of acetic anhydride is
represented with squares, and the formation of acetic acid, with circles.
The solid and dotted lines represent the prediction calculated by
Dynochem.
experimental error within this study are the failure to accurately
control the temperature throughout the reaction, which is
especially important in nonadiabatic reactions, and the failure to
accurately measure the total reaction volume, which is not
necessarily the sum of reagent volumes. The accurate measure-
ment of the mass of reagents is also important. This is especially
difficult when working on a small scale.
Figure 2 shows the reaction profiles for experiments 3 and 5
with good agreement between experimental and calculated
values. The agreement is not as good in experiment 2 (Figure 3).
In this last case the delay between initiation of the reaction and
starting the NMR acquisition was not as accurately measured as
in the experiments 3, 4, and 5, and this is the root cause of the
poorer fit of the experimental and calculated data.
Figure 2. Reaction profile for experiments 3 and 5 conducted respec-
tively with 0.91 M acetic anhydride at 5 °C (A) and 0.24 M at 25 °C (B).
The depletion of acetic anhydride is represented with squares, and the
formation of acetic acid, with circles. The solid and dotted lines represent
the prediction calculated by Dynochem.
The methodology described in this contribution has some
limitations and would not be the method of choice for the direct
study of biphasic systems or of very fast or exothermic reactions
because of issues of nonhomogeneity of the sample, the need to
mix all reagents and solvents outside of the NMR magnet, and
the lack of good temperature control throughout the whole
duration of the experiments, respectively. However, it has the
potential to be widely applied to pharmaceutical process chem-
istry projects because it is inherently quantitative, simple, fast,
and easy to use. It also provides structural information which can
assist the identification of intermediates. The authors believe that
a model does not necessarily have to be built with experimental
data derived from large-scale reaction conditions or from mi-
micking large-scale conditions (e.g., kilo lab and pilot plant). It
can be built during the small-scale experimental design stage for
use in reaction understanding and optimization. The chemical
model so developed will then be in place and ready to combine
with the reactor modelling data for use in supporting process
scale-up.
Table 2. Experimental activation energies and ln A values for
the acetic anhydride hydrolysis
activation energy
(kcal/g mol)
In A
(L/(g mol s))
3
3
this work
Asprey et al.11
Glasser and Williams (1971)13
Cleland and Wilhelms (1956)14
Eldridge and Piret (1950)15
11.4
10.9
10.8
10.6
10.3
8.05
7.66
7.95
7.80
7.53
after 400 iterations. The confidence interval for both was <10%,
which is considered to be very good. The quality of the fit is
illustrated in Figure 2, which show a high degree of overlap between
calculated and experimental reaction profiles for the reaction at 25 °C
and acetic anhydride initial concentrations of 0.91 and 0.24 M.
The value of the kinetic parameters Ea and ln A obtained for
the acetic anhydride hydrolysis using the methodology reported
in this contribution are also in good agreement with the values
reported in the literature16,17 for the same reaction, but using
different techniques (Table 2).
’ CONCLUSIONS
A kinetic model was derived from NMR experiments for the
hydrolysis of acetic anhydride with water. The results demon-
strate that 1H NMR data obtained by monitoring reactions in an
NMR tube can be sufficient to build a good, reliable kinetic
model, provided the experiments are conducted in a controlled
Kinetic studies require accurate measurement and control
of reaction conditions. Factors that could contribute to the
63
dx.doi.org/10.1021/op200202k |Org. Process Res. Dev. 2012, 16, 61–64

Organic Process Research & Development
ARTICLE
manner with high attention to detail. It is important to accurately
measure the mass of the starting materials and solvents together
with the time delay between the initiation of the reaction and the
start of the NMR acquisition. It is also critical to control the
temperature during the preparation of the sample and through-
out the reaction in order to minimize experimental errors.
Despite NMR being a very powerful and widely available
technique, we believe that it is underutilized in the chemical
process development arena. It has great potential value for kinetic
modelling in combination with orthogonal data from other
techniques (such as LC, GC, or IR).
’ AUTHOR INFORMATION
Corresponding Author
Flavien_susanne@yahoo.fr; acviladoms@gmail.com.
’ ACKNOWLEDGMENT
We thank Santiago Dominguez and Carlos Cobas (Mestrelab
Research) for software support and development and Joe
Hannon (Dynochem), Christopher Pilkington and Pfizer collea-
gues (Wilfried Hoffmann, Julian Smith, Don Richards) for useful
discussions.
’ REFERENCES
(1) Bayer Technology Services, http://www.bayertechnology.cn/
uploads/media/0707_e_300dpi.pdf, accessed in July 2011
(2) Lawrence Livermore National Laboratory, http://www1.eere.
energy.gov/vehiclesandfuels/pdfs/merit_review_2011/fuel_technologies/
ft010_pitz_2011_o.pdf, accessed in July 2011
(3) Shin, S. B.; Han, S. P.; Lee, W. J.; Chae, J. H.; Lee, D. I.; Lee,
W. H.; Urban, Z. Hydrocarbon Process. 2007, No. April, 83.
(4) Baumer, C.; Urban, Z.; Matzopoulos, M. Hydrocarbon Process.
2007, No. June, 71.
(5) Garrido, M.; Larrechi, M. S.; Rius, F. X.; Mercado, L. A.; Galiꢀa,
M. Anal. Chim. Acta 2007, 583, 392.
(6) Khajeh, M.; Botana, A.; Bernstein, M. A.; Nilsson, M.; Morris,
G. A. Anal. Chem. 2010, 82, 2102.
(7) Zeng, H.; Lee, Y.; Hilty, C. Anal. Chem. 2010, 82, 8897.
(8) Ciobanu, L.; Sweedler, J. V.; Webb, A. G. NMR Imaging in
Chemical Engineering; Wiley-VCH Verlag GmbH & Co. KGaA: Weinheim,
2006; p 123.
(9) Bernstein, M. A. Spectrosc. Eur. 2007, 19, 11.
(10) Visentin, F.; Gianoli, S. I.; Zogg, A.; Kut, O. M.; Hungerb€uhler,
K. Org. Process Res. Dev. 2004, 8, 725.
(11) Asprey, S. P.; Wojciechowski, B. W.; Rice, N. M.; Dorcas, A.
Chem. Eng. Sci. 1996, 51, 4681.
(12) Dynochem 2011, Scale-up Systems, www.scale-up.com, 2011
(13) Glasser, D.; Williams, D. F. Ind. Eng. Chem. Fundam. 1971,
10 (3), 516.
(14) Cleland, F. A.; Wilhelm, R. H. AIChE J. 1956, 2, 489.
(15) Eldridge, J. W.; Piret, E. L. Chem. Eng. Prog. 1950, 47, 363.
(16) King, R. P.; Glasser, D. S. Afr. Ind. Chem. 1965, 19, 12.
(17) Shatynski, J. J.; Hanesian, D. Ind. Eng. Chem. Res. 1993, 32, 594.
64
dx.doi.org/10.1021/op200202k |Org. Process Res. Dev. 2012, 16, 61–64