Control Strategy for the Manufacture of Brivanib Alaninate, a Novel Pyrrolotriazine VEGFR/FGFR Inhibitor

Article abstract of DOI:10.1021/op500126u

This manuscript describes the control strategy for the commercial process to manufacture brivanib alaninate. The active pharmaceutical ingredient is a prodrug which is susceptible to hydrolysis. In addition to controlling hydrolysis, a robust strategy was

Full text of DOI:10.1021/op500126u

Article
pubs.acs.org/OPRD
Control Strategy for the Manufacture of Brivanib Alaninate, a Novel
Pyrrolotriazine VEGFR/FGFR Inhibitor
Paul C. Lobben,* Evan Barlow, James S. Bergum, Alan Braem, Shih-Ying Chang, Frank Gibson,
Nathaniel Kopp, Chiajen Lai, Thomas L. LaPorte, David K. Leahy, Jale Muslehiddinoglu,
̆
̈
Fernando Quiroz, Dimitri Skliar, Lori Spangler, Sushil Srivastava, Daniel Wasser, John Wasylyk,
Robert Wethman, and Zhongmin Xu
Chemical Development, Bristol-Myers Squibb Company, One Squibb Drive, New Brunswick, New Jersey 08903, United States
S
* Supporting Information
ABSTRACT: This manuscript describes the control strategy for the commercial process to manufacture brivanib alaninate. The
active pharmaceutical ingredient is a prodrug which is susceptible to hydrolysis. In addition to controlling hydrolysis, a robust
strategy was required in order to control input and process-related impurities. Three significant aspects of control include
understanding of the reaction parameters in order to minimize the regioisomer during the alkylation with (R)-propylene oxide,
development of a design space through statistical models to control impurity formation, and the use of in situ FT-IR to monitor
the hydrogenolysis of the Cbz protecting group.
INTRODUCTION
• Alkylation with RPO to incorporate the stereocenter of
the parent drug 2 and maximize regioselectivity.
• Development of a neutral and mild process that removes
the carbobenzyloxy (Cbz) protecting group while
minimizing hydrolysis of the resulting prodrug.
• Application of PAT for the hydrogenolysis to allow real-
time control and ensure product quality and improve
process robustness.
■
Brivanib alaninate (1) is a new oncology therapy with potential
applications against a wide variety of tumor types and several
stages of disease progression.1,2 This alanine ester prodrug is
orally active and a dual inhibitor of the vascular endothelial
growth factor receptor-2 (VEGFR-2) and fibroblast growth
factor receptor (FGFR) tyrosine kinases, both of which
modulate angiogenesis. Angiogenesis is characterized by the
formation of a vascular network of blood vessels that serve
growing tissue.3 The compound is expected to possess broad-
spectrum antitumor activity for use in the treatment of
hepatocellular carcinoma, colorectal cancer, and fibroblast
growth factor driven tumors. Brivanib alaninate, at a maximum
daily dose of 800 mg, was evaluated in phase 3 clinical trials.
The building blocks for the synthesis of brivanib alaninate
were obtained from simple starting materials and our previously
disclosed “quality gatekeeper” intermediate 3 (Scheme 1).4 The
overarching development goal was to assemble these building
blocks in a safe, robust, and efficient manner with optimum
control over the critical quality attributes (CQAs) of the drug
substance. The CQAs included identity, assay/potency,
impurities (including stereoisomers), residual solvents, and
particle size. (R)-Propylene oxide (RPO) and Cbz-L-alanine
were used to install the side chain and the alanine prodrug
moiety, respectively.5 The Cbz protecting group was chosen
such that it could be removed under mild and neutral
conditions in order to minimize hydrolysis of the ester. We
expected that these two chemical processes would require in-
depth understanding of the process parameters as a means to
control process-related impurities and develop a robust
manufacturing process. The following discussion details the
development challenges and the aspects of the control strategy
required to meet the CQA of the drug substance and ensure
consistent process performance and product quality.
RESULT AND DISCUSSION
■
The last two steps to manufacture brivanib alaninate include
two consecutive telescoped sequences, each composed of two
chemical reactions. The first requires saponification of the
pivalate protecting group of 3 with an alkoxide base (Scheme
1). Without isolation, the resulting hydroxypyrrole 6 is then
alkylated with (R)-propylene oxide. The resulting parent drug 2
(brivanib) directly crystallizes (direct-drop) from the reaction
mixture. A subsequent recrystallization from acetone and water
is required to further minimize process-related impurities,
including residual (R)-propylene oxide.6 Installation of the
prodrug begins with the acylation of the secondary alcohol with
carbobenzyloxy-L-alanine (Cbz-L-ala). The unisolated inter-
mediate7 is telescoped directly into the hydrogenolysis, using
Pd/C and hydrogen in THF. After a distillation to effect a
solvent exchange to ethyl acetate (EtOAc), the active
pharmaceutical ingredient (API) is crystallized with the
addition of heptanes in an overall yield of 61% with 99% purity.
Special Issue: Application of ICH Q11 Principles to Process
Development
Received: April 15, 2014
© XXXX American Chemical Society
A
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development
Article
Scheme 1. Proposed Commercial Route for the Synthesis of Brivanib Alaninate (1)
Figure 1. Key impurities resulting from alkylation with (R)-propylene oxide.
PROCESS DEVELOPMENT
■
Penultimate Step: Deprotection and Alkylation. The
penultimate process required 1 equiv of NaOMe8 for complete
deprotection of the pivalate ester 3. Excess reagent afforded
methanolysis of the product via ipso substitution at the indole-
aryl ether functionality. A wide variety of solvents were
identified to be compatible with the deprotection reaction. We
realized that the choice of solvent would play a critical role in
both the overall efficiency of the process and resulting product
quality. Ultimately, acetonitrile (CH3CN) was chosen based on
an acceptable rate of conversion and a favorable impurity
profile, vide infra. An added benefit of using CH3CN was that
the intermediate sodium salt could be telescoped directly into
the alkylation. If isolated as solid or stored as a solution, the salt
of intermediate 6 slowly degraded. Although the degradants,
mainly methanolysis/hydrolysis byproducts, were easily purged
in the crystallization, an impact on yield was observed, and
therefore, it was valuable to minimize these side reactions.
Attention then focused on the development of the alkylation
reaction, and specifically the control of the process-related
impurities. Two impurities that required strict control were the
regioisomer 4, which results from alkylation at the central
carbon of (R)-propylene oxide and alkylation of the product at
the indole-nitrogen to afford impurity 5 (Figure 1). The latter
impurity was difficult to reject in the crystallization and
subsequent recrystallization. The regioisomer impurity 4 ranges
between 5 and 10% prior to crystallization, and it effectively
purges in the crystallization. In addition it has a high tolerance
in the downstream processing.9 Intermediate hydroxypyrrole 6
was controlled through optimization of process parameters, vide
infra.
In neat acetonitrile, 50% conversion with 20% byproducts
occurred. Conversely, the use of water as the primary solvent
provided complete conversion, albeit at the expense of poor
regioselectivity. 67−80% water in acetonitrile afforded com-
plete conversion, an acceptable impurity profile, and allowed
for the direct crystallization of the parent drug 2 (Figure 2).
While a lower ratio of water increased the regioselectivity, it
also significantly slowed the reaction rate, thereby increasing
the amount of methanolysis and hydrolysis byproducts. Further
Figure 2. Impact of the solvent ratio on regioselectivity and
conversion.
optimization of the alkylation was realized through a design of
experiments (DoEs). A recrystallization step from acetone/
water was required for additional control of impurities 4 and 6
and RPO.
Control Strategy: (R) Propylene Oxide Alkylation.
Once the optimal solvent ratio of 2:1 v/v water-to-acetonitrile
was established, a design of experiments (DoE) protocol was
implemented to delineate the impact of the key parameters on
the impurity profile. The advantage of a DoE as compared to an
empirical univariate approach is the ability to efficiently define
the interactions between parameters. Results from the DoE
indicated that the equivalents of (R)-propylene oxide, water/
acetonitrile ratio, and batch temperature required tight control
in order to deliver product of a consistent quality. The
alkylation of the salt of hydroxypyrrole 6 is a relatively slow
reaction reaching completion in 12−60 h at 20−40 °C. The
reaction time is particularly sensitive to temperature. For
example a difference of three degrees in the batch temperature
impacts the reaction time by up to 5 h. Although temperature
increases the rate of the reaction, this parameter was also
observed to increase side reactions and impurities.10 The two
key impurities were generally found to be orthogonally formed
at different points in the design space, but both increased at
elevated temperatures. The worst conditions for the N-
alkylation impurity 5 were found to be low water/acetonitrile
B
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development
Article
Figure 3. 95% Prediction intervals for the model-predicted design space impurities 4 (a, left) and 5 (b, right). The color axis to the right of the graph
represents the levels of impurity.
ratio and high equivalents of (R)-propylene oxide. Conversely,
the worst conditions for regioisomer 4 were found to be high
ratio water/acetonitrile and low equivalents of propylene oxide.
A tight but workable range for each of the three parameters was
defined. The next step was to understand the purgability of the
key impurities in downstream processing to define a robust
multivariate design space.
Spiking studies determined that the API process could
tolerate up to 1.9% of the regioisomer impurity 4 and 0.34% of
the N-alkylation impurity 5. Accordingly, based on the process
capability for purging each impurity, the alkylation design space
was established such that a maximum of 5.9% of the
regioisomer impurity 4 and a limit of 0.34% for 5 would be
present in the first-drop direct crystallization of parent drug 2
prior to the required recrystallization step. Impurity 5 does not
purge in the crystallization or recrystallization.11 The
regioisomer 4 purges roughly by 50% in the first-drop
crystallization, affording an internal acceptance criteria [upper
limit] of ≤5.9%. A further 67% reduction was observed in the
recrystallization step to achieve the tolerance of 1.9%.
To increase the understanding of the multivariate inter-
actions and guide the design space development, a statistical
model was developed to predict the level of the two critical
impurities across the range of parameters.6 The data collected
from laboratory experiments and pilot plant batches was used
to develop a multivariate regression for each impurity as a
function of the three key process parameters. These equations
were further refined to evaluate the upper 95% prediction
intervals of the model in order to establish the edge of failure.
The equations for the 95% prediction interval were obtained by
fitting the estimates provided by SAS software for each of the
calibration data points used to generate the model.12 Using
these equations as guides to define the design space provided a
conservative set of reaction conditions to ensure quality.
Figure 3 shows the multidimensional graphs generated from
the statistical models. The left graph highlights the range of 4
(4.3−5.9 area %) observed when operating within the design
space. The right graph similarly displays the level of impurity 5
within the design space, ranging from 0.17 to 0.34 area %. The
design space is quite sensitive to the specification limit of 1.9%
for 4 previously described for the parent drug 2. For example,
tightening the limit of impurity 4 by 0.1% to 1.8% reduces the
space by 33%.
constraints were considered in order to maximize the process
robustness:
• Align ranges with the capabilities of manufacturing
equipment
•
•
10% variation on (R)-propylene oxide charge
5 °C temperature operating window.
Five pilot plants batches at 100 kg scale and 11 laboratory
experiments on 20 g scale were completed in order to verify the
predicted results. The results agreed with the predicted values.
The final design space for the alkylation is as follows:
equivalents of (R)-propylene oxide of 4.4−5.5, internal batch
temperature of 21−31 °C, and a ratio of water to acetonitrile of
1.8−2.3:1.
Final Step: Acylation with Cbz-L-alanine. With a robust
process for the synthesis of parent drug 2 developed, attention
focused on the acylation of the chiral secondary alcohol with
Cbz-L-alanine to produce the noncrystalline intermediate 7
(Figure 5). Although several protecting groups for the primary
amine were considered, the carbobenzyloxy (Cbz) group was
preferred for the ability to remove the protecting group under
neutral conditions, and its commercial availability as a starting
material. The final step can be divided into four processes: (1)
the acylation with Cbz-L-alanine, (2) hydrogenolysis to remove
the protecting group, (3) crystallization of the API, and (4)
wet-milling to afford the specified particle size. In consideration
of the impact that unit operations can have on the CQAs of a
drug substance, including purity and particle size, the
development of each step will be examined separately.
A wide range of commercially available reagents and
conditions used to activate the carboxylic acid toward
esterification with 2 was evaluated. Activation of a carboxylic
acid via the acid chloride was not practical due to epimerization
of the chiral center. The resulting diastereomer from
epimerization of Cbz-L-alanine was not purged in the
crystallization, and thus, its formation had to be controlled. 1-
(3-(Dimethylamino)propyl)-3-ethylcarbodiimide hydrochlor-
ide (EDAC-HCl) as the acylating reagent proved to be robust
with minimal epimerization and facile work-up.
Acetone, THF, EtOAc, and CH3CN each offered >90%
conversion after only 5 h, with a de greater than 95%. A solvent
screen revealed that the rate of conversion was diminished in
highly polar aprotic solvents; NMP and DMF afforded only
70% conversion with 95% de after 10 h at ambient temperature.
THF was the preferred solvent due to an increase in the
solubility of parent drug 2 and the ability to telescope directly
into the hydrogenolysis process without a solvent exchange.
Finally, experimental verification for the proposed design
space was focused on the regions predicted to form the highest
impurity levels. In building this new design space, the following
C
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development
Article
Figure 4. Hydrogenolysis process-related impurities.
Figure 5. Implementation of FT-IR as part of the hydrogenolysis control strategy.
However, DMF as a cosolvent was necessary, particularly at
lower temperatures, in order to increase the solubility of the
EDAC hydrochloride and the resulting urea byproduct.13
Due to the modest nucleophilicity of parent drug 2, DMAP
was required to catalyze the acylation.14 However, Cbz-L-
alanine was observed to racemize under reaction conditions,
presumably due to the keto−enol tautomerization of the
polarized acylpyridinium intermediate.15 Optimum control of
the diastereomer impurity was realized by charging the Cbz-L-
alanine last to the reaction mixture at 0 °C. The unisolated
intermediate 7 (Figure 5) was stable to epimerization under
reaction conditions even when subjected to prolonged reaction
times.
The Cbz-alanine byproducts and alanine-related oligomers
were found to impact the efficiency of the final crystallization.
Immediate quench of the reaction was required to minimize
oligomerization of Cbz-ala. When telescoped through the
hydrogenolysis, the resulting di-, tri-, and tetra-alanine
oligomers negatively impacted the purging of impurity 8.16
Furthermore, an aqueous work-up removed additional Cbz-L-
alanine-related byproducts. The work-up consisted of sequen-
tial aqueous acidic, basic, and neutral washes. Following the
aqueous work-up and an azeotropic distillation to dry the
solution, the THF solution of intermediate 7 was telescoped
into the hydrogenolysis. The control strategy for the acylation
of parent drug 2 focused on (1) low temperature at −5−5 °C,
(2) charging the Cbz-L-alanine last, and (3) immediate quench
of the reaction upon reaching the end point.
142 (BASF) afforded a balanced rate of reaction and a favorable
impurity profile. As the API molecule possesses sensitive
functional groups such as a primary amine, an ester, and a labile
indole−aryl ether bond, the choice of solvent was limited to
polar aprotic solvents.17 Both EtOAc and THF proved effective
solvents for the hydrogenolysis. THF was selected because of
improved catalyst dispersion of the water wet-catalyst and a
resulting lower catalyst loading. Ethyl acetate has a liability of
N-acylation of the primary amine catalyzed by the carbon
dioxide byproduct.18
Control Strategy: Hydrogenolysis PAT. To develop a
robust strategy, three robust in-line FT-IR methods were
developed to monitor both the concentration of starting
material 2 and the concentration of carbon dioxide byproduct
in the reaction mixture.19 Full conversion was required due to
poor purging of the intermediate 7 in the final crystallization.
The reaction end point was set at >99.9% conversion by in-
line FT-IR due to the limitation of off-line HPLC. A fast
reaction, that is, an overcharge of catalyst (or too long of a
reaction), was to be avoided in order to minimize over-
reduction and to afford impurity 8. As with other in-process
related impurities, the rate of formation of impurity 8 depends
on the rate of the desired reaction. If the reaction was
proceeding as expected, it was important to stop the
hydrogenolysis when the end point was reached by immediately
purging hydrogen and CO2. However, the FT-IR method20 was
found not to be sensitive to less than 2.0 wt % of intermediate
7. As a result, a chemometric model using partial least-squares
combined with the reaction kinetics to accurately predict the
reaction end point. The chemometric model alone was not
enough. Once the level of intermediate 7 reached 5%, rate data
was collected until a level of 3% was reached. At this point, the
rate constant was obtained and the time to reach 99.9%
conversion was estimated based on the first order approx-
imation.21 The initial reaction kinetics was used to enabled the
detection of catalyst poisoning which may result in a slow
reaction and eventual hydrolysis of the prodrug.22 Upon
Final Step: Hydrogenolysis. Hydrogenolysis of inter-
mediate 7 proved to be complicated due to the lability of 1.
Three chemical pathways required control: (1) formation of
parent drug 2, (2) over-reduction to afford impurity 8, and (3)
transamidation to afford dialanine impurity 9 and parent drug
2. Catalysts with high activity would typically afford nearly
quantitative indole cleavage impurity 8 (Figure 4). A slow
reaction routinely resulted in >5% hydrolysis of the prodrug 1
to form parent 2 as well as formation of the dialanine 9. ESCAT
D
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development
Article
reaction completion, over-reduction impurity 8 continues to
form in the presence of the catalyst, while dissolved carbon
dioxide catalyzes the hydrolysis of the product. Therefore, it
was important to remove carbon dioxide and filter the catalyst
immediately. During the purge of hydrogen, the FT-IR method
also monitored the level of carbon dioxide to an end point of
600 ppm or less. This was then followed by filtration. The
dissolved carbon dioxide catalyzes the hydrolysis of the prodrug
at a rate of 2% per week at <10 °C. The real-time monitoring of
the reaction, steps 1−3, was important to ensure quality (Figure
5). The immediate filtration of the catalyst, step 4, was also
important to minimize the formation of impurity 8. The control
strategy for the successful hydrogenolysis of intermediate 7
hinged on four points (Figure 5):
pressure, catalyst loading, headspace, and kLa and at high values
for concentration and water content (KF). A low hydrogen
mass transfer condition will lead to slower reaction kinetics and
the possibility of incomplete reaction and unacceptable
impurity profile.25 These parameters afforded a reaction time
of 9.5 h with an acceptable in-process impurity profile. All 12
experiments passed the minimum initial kinetic target of 50%
conversion over 20 min and in-process impurity profile. None
required a kicker charge of catalyst. The proven acceptable
range (PAR) for each parameter is shown in Table 1.
Table 1. Proven Acceptable Range for Hydrogenolysis of
Intermediate 7
Parameter
Low
Target
High
(1) measuring the initial reaction kinetics
(2) estimate the reaction end point
(3) purging the carbon dioxide byproduct
(4) filtering off the spent catalyst
Reactor Headspace (vol %)
Pressure (psig)
40
45
60
28
35
45
Reaction temperature (°C)
23
26
35
Catalyst loading, dry basis (wt % relative to
parent drug 2)
7.75
8.14
9.30
Control Strategy: Hydrogenolysis Proven Acceptable
Ranges. Early development indicated that seven parameters
impacted the robustness and quality of the process. Utilizing a
10 wt % loading of a 5 wt % Pd/C catalyst at a concentration of
5−10 L/kg was found to balance throughput while maintaining
quality and process robustness. Equipment design prevented
the hydrogen pressure from exceeding 45 psig; pressures below
30 psig and/or temperatures below 23 °C led to a slower
reaction with a risk of an incomplete conversion and significant
ester hydrolysis. In contrast, reaction temperatures greater than
35 °C led to faster reaction kinetics and over-reduction to
afford the indole-cleavage impurity 8.23
Conc (L/kg relative to parent drug 2 input)
9.32
0.06
0.1
7.77
0.08
6.99
0.3
Mass transfer coefficient (s−1
)
Water content (KF wt %)
0.5
Final Step: Crystallization. The last step in the process
was the crystallization and isolation of the drug substance. THF
was evaluated as the crystallization solvent but afforded a low
bulk density material with an unpredictable morphology, both
of which complicated drug product formulation. Although the
primary amine group is susceptible to acylation, ethyl acetate
was found to afford a robust crystallization with regard to
effectively controlling the input-related impurities. With
heptanes as the preferred antisolvent, the crystallization of
the API afforded rosette-like agglomerates (Figure 7). In the
manufacturing process, in-line Raman spectroscopy replaced
the original GC and HPLC methods for (1) initial batch
quantitation, (2) solvent exchange end-point (vol % THF), and
(3) in-process concentration (wt % 1) prior to antisolvent
addition. Raman proved26 an effective in-process control
instrument and was advantageous for the following reasons:
A minimum headspace of 40% was required in order to
minimize the level of dissolved CO2 byproduct (Figure 6).24 In
• Continuous monitoring vs periodic sampling
• Cycle time reduction positively impacted impurity profile
• Robustness around samplingno risk of product
precipitation prior to analysis.
In order to develop a robust design space, the crystallization
process was developed and optimized via single and multi-
variate experiments. For the design space, an 8-variable 2-level
factorial DoE was chosen in order to be able to resolve the
main effects. The key parameters evaluated included the initial
concentration in EtOAc, temperature, the ratio of antisolvent,
and the rate of addition. The measured responses included
losses to the mother liquor, yield, and the resulting impurity
profile. Oiling of the product was avoided by charging the
antisolvent in two portions with an intermediate seeding step.
Although dilute conditions only impacted yield, high
concentration caused premature nucleation and increased the
risk of an unacceptable impurity profile. Similarly, less volume
of heptanes impacted yield and increased risk of not achieving
nucleation upon seeding, whereas greater volumes increased
the risk of an unacceptable impurity profile and oiling of the
product prior to seeding. A high batch temperature increased
the levels of N-acylation, whereas a lower batch temperature led
to premature nucleation and a risk to the impurity profile.
Figure 6. Impact of headspace on the hydrogenolysis kinetics.
order to ensure quality and robustness, the reaction was
developed to provide kinetics such that the reaction was
complete within 10 h across the range of parameters. The initial
reaction kinetics over the first 20 min were monitored by FT-
IR.22 The data from laboratory experiments and multiple DoEs
were combined and analyzed in order to generate a statistical
model to control each of the two key impurities: parent drug 2
and the input intermediate 7.6
To verify the proposed ranges, a multivariate Placket−
Burman design was completed with 12 experiments for the
seven parameters. Included in the design was the predicted
worst case scenario centered at low values for temperature,
E
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development
Article
Figure 7. Rosette-like primary particles (left) and after wet-milling (right), taken by SEM.
Wet-milling of the crystallized slurry was required in order to
meet the particle size specification (CQA) necessary for
successful drug product formulation. Wet-milling is preferred
to dry grinding because of the reduction in the potential for
dust explosion, occupational exposure, handling, and cycle time.
The operating ranges for the main parameters (e.g., rotator
stators, mill power, and flow rate) were established for the
specific equipment.27 After isolation, the cake was then
deliquored using a mixture of EtOAc in heptanes and
subsequently washed with heptanes. A constant filtration rate
with a minimal resistance indicated the wet cake was
noncompressible. The heptanes wet cake typically dried in
less than 8 h under full vacuum at 50 °C. Neither
agglomeration nor attrition occurred using intermittent or
continuous drying cycles in a filter dryer.
(R)-1-(4-(4-Fluoro-2-methyl-1H-indol-5-yloxy)-5-
methylpyrrolo[1,2-f ][1,2,4]triazin-6-yloxy)propan-2-ol (2). A
reactor was charged with acetonitrile (183.1 kg), followed
successively by pivalate ester 3 (169.9 kg, 428.6 mol) and
additional acetonitrile (100.2 kg) as a rinse. The solution was
cooled to 12 °C and then was treated with 25 wt % sodium
methoxide in MeOH (92.7 kg, 429.0 mol) at <18 °C. The
alkoxide charge was followed by a rinse with acetonitrile (27.6
kg). The reaction mixture was maintained at 10−20 °C under
constant agitation. After 1 h, HPLC analysis indicated the
deprotection reaction was complete at >99.0% conversion. The
reaction mixture was diluted with water (913.8 kg) at <30 °C.
The batch temperature was adjusted to 29 °C prior to the
charge of (R)-propylene oxide (129.5 kg, 2230 mol) and
acetonitrile (21.5 kg) as a rinse. After the vent of the reaction
vessel was closed, the resulting reaction mixture was stirred at
20−30 °C for 1 d. Sampling the slurry after 20 h indicated
>98% HPLC conversion. The crystallized product was isolated
by filtration and was washed with 20% acetone (aq) (1119.7
kg). The isolated product was then recrystallized from aqueous
acetone. The cake was dissolved in 90% acetone (aq) (1035.4
kg) at 50−60 °C. The resulting clear solution was treated with
water (1020.3 kg) over 45 min at 50−60 °C. The batch
temperature was cooled to 20 °C over 3 h. The resulting slurry
was aged for 4 h and filtered. The cake was washed with 20%
acetone (aq) (119.7 kg) to afford 112.0 kg (71% yield) of
parent drug 2 as an off-white crystalline solid in 98% purity. 1H
NMR (300 MHz, DMSO-d6) 11.34 (1 H, br s), 7.90 (2 H, d, J
= 6.6 Hz), 7.14 (1 H, d, J = 8.8 Hz), 6.90 (1 H, t, J = 7.1 Hz),
6.24 (1 H, s), 4.90 (1 H, d, J = 4.8 Hz), 4.08−3.95 (1 H, m),
3.92 (2 H, d, J = 6.6 Hz), 2.50 (3 H, s) 2.41 (3 H, s), 1.18 (3 H,
d, J = 6.6 Hz).
(S)-((R)-1-(4-(4-Fluoro-2-methyl-1H-indol-5-yloxy)-5-
methylpyrrolo[1,2-f ][1,2,4]triazin-6-yloxy)propan-2-yl)-2-
(benzyloxycarbonylamino)propanonate (7). THF (525.2 kg)
and DMF (70.2 kg) were charged to a reactor, followed
successively by parent drug 2 (73.8 kg, 200 mol), EDAC-HCl
(58.0 kg, 300 mol), and DMAP (0.740 kg, 5.99 mol). The solid
charges were followed with a rinse of THF (65.9 kg). The
reaction mixture was cooled to −1 °C prior to charging Cbz-L-
alanine (54.3 kg, 244 mol) and was subsequently stirred for 3 h.
HPLC analysis indicated <0.2 RAP of secondary alcohol 2. The
reaction mixture was quenched with 0.5 M H3PO4 in 15 wt %
brine (643.8 kg) while maintaining the batch temperature <20
°C. The biphasic reaction mixture was warmed to 20 °C with
constant agitation for 30 min and then allowed to settle for 30
CONCLUSION
■
The development of a control strategy for the manufacture of
brivanib alaninate utilized DoE, generated statistical models,
visualized the design spaces, applied FT-IR/PAT, and
developed methods. In the penultimate process to afford
parent drug 2, it was found that impurities 4 and 5 both
increased at high temperatures but otherwise were at odds with
each other, which required balancing of the equivalents of (R)-
propylene oxide and the ratio of water/acetonitrile. The
resulting design space was constrained to the tolerance of the
impurities to downstream processing. Hydrogenolysis of
intermediate 7 was complicated by the relative instability of
the prodrug 1 (brivanib alaninate) under the reaction
conditions. Finalization of the process required the inves-
tigation of seven parameters using DoE. FT-IR was
instrumental in being able to monitor the deprotection in
real-time, detect operational/equipment issues, and stop the
hydrogenolysis at the end point so as to minimize over-
reduction of the biaryl ether. In-line Raman spectroscopy
provided additional control and robustness to the crystallization
process. The API was isolated by filtration as a white to off-
white crystalline powder in 76% overall yield and 99% purity.
EXPERIMENTAL SECTION
■
General. All solvents, reagents, and starting materials were
supplied by commercial vendors and were used without further
purification. Unless otherwise stated, all reactions were carried
out in glass-lined reactors and individual unit operations were
carried out under an inert atmosphere of nitrogen. Reported
yields are uncorrected for potency.
F
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development
Article
min. The rich organic layer was isolated and washed with 4 wt
% K2CO3 in 15 wt % brine (858.4 kg) and was agitated for 30
min and then was allowed to settle for 60 min. The rich organic
phase was split and subsequently washed with 15 wt % brine
(905.0 kg). After 30 min under agitation the biphasic mixture
was allowed to settle for 60 min. The rich organic layer was
isolated and subsequently azeotropically distilled and concen-
trated to 400 L and a final KF of <0.5 wt %. The solution was
filtered to remove precipitated salts to afford Cbz-L-alanine
ester 7 as a 27.3 wt % solution in THF (420.2 kg) in near
quantitative yield. A sample could be concentrated and dried to
the mother liquor and filter dryer. The dry cake was packaged
in double polyethylene bags, with desiccant between each bag;
and each bag was sealed with a twist-tie. The bags were placed
in a HDPE outer container with a secure lid, fitted with a
compressible rubber seal. 1H NMR (400 MHz, CDCl3) 8.31 (1
H, s), 7.83 (1 H, s), 7.25 (1 H, s), 7.00 (1 H, d, J = 8.6 Hz),
6.95 (1 H, dd, J = 15.4, 8.6 Hz), 6.28 (1 H, s), 5.36−5.30 (1 H,
m), 4.08−4.00 (2 H, m), 3.57 (1 H, dd, J = 14.0, 6.9 Hz), 2.47
(3 H, s), 2.40 (3 H, s), 1.66 (3 H, s), 1.38 (3 H, d, J = 6.4 Hz),
1.35 (3 H, d, J = 7.1 Hz).
1
ASSOCIATED CONTENT
* Supporting Information
afford an off-white amorphous solid. H NMR (400 MHz,
■
CDCl3) 8.17 (1 H, br s), 7.84 (1 H, s), 7.41 (1 H, s), 7.35−7.28
(5 H, m), 7.03 (1 H, d, J = 8.6 Hz), 6.95 (1 H, t, J = 7.7 Hz),
6.30 (1 H, s), 5.36−5.32 (2 H, m), 5.11 (2 H, br s), 4.43−4.40
(1 H, m), 4.02−3.99 (2 H, m), 2.46 (3 H, s), 2.41 (3 H, s), 1.44
(3 H, d, J = 7.2 Hz), 1.38 (3 H, d, J = 7.2 Hz).
S
A description of the procedures used to handle (R)-propylene
oxide (RPO) during isolation and recrystallization; the
statistical models that were generated for impurities 4 and 5
and the data visualization tools; a discussion of the design space
for the manufacture of parent drug 2; and a discussion of the
development of the PAT FT-IR methods used to monitor the
hydrogenolysis of intermediate 7. This material is available free
of charge via the Internet at http://pubs.acs.org.
(S)-((R)-1-(4-(4-Fluoro-2-methyl-1H-indol-5-yloxy)-5-
methylpyrrolo[1,2-f ][1,2,4]triazin-6-yloxy)propan-2-yl)-2-
aminopropanonate (1). Cbz-L-alanine ester 7 (420.7 kg) as a
27.3 wt % solution in THF was charged to a stainless steel
reactor. The batch was diluted with additional THF to a target
concentration of 5.0 L/kg to achieve a nominal headspace of
45%. After three successive nitrogen purges up to 30 psig, a 5%
Pd/C, 50% water wet, catalyst (12.0 kg) was charged to the
reactor. Following the requisite pressure test, the reactor was
again purged, without agitation, three times with nitrogen up to
30 psig, and three times with hydrogen up to 30 psig. The batch
temperature was established at 26 °C. The reaction mixture was
only then agitated at 30 psig hydrogen for 6 h, at which time
the reaction was deemed complete by the FT-IR. The
chemometric model predicted >99.9% conversion. The initial
reaction kinetics after 20 min indicated >80% conversion and
that a kicker charge of catalyst was not necessary. The reaction
end point was confirmed by HPLC analysis at >99.7%
conversion. The vessel was depressurized and the reaction
mixture was purged continuously with nitrogen until the
concentration of carbon dioxide was below 600 ppm as
measured by FT-IR. The reaction mixture was filtered and the
reaction vessel and filter train were rinsed with THF (100.0 kg)
to afford in-process prodrug 1 in THF (609.4 kg).
An in-process solution of prodrug 1 in THF (609.1 kg) was
transferred to a reactor. The stream was assayed by in-line
Raman spectroscopy at 14.0 wt %, affording a calculated in-
process yield of 96.9% (85.3 kg) from parent drug 2. The THF
solvent was exchanged by distillation under reduced pressure
for ethyl acetate. The distillation and solvent exchange were
both monitored by in-line Raman spectroscopy to achieve an
end point of 6.0 vol % THF and a concentration of 19.9 wt %.
The solution was heated to 55 °C, and one volume of heptanes
(263.1 kg) was added over 40 min at >50 °C. Seeds of 1 (250
g) were charged to the solution in order to induce nucleation.
The resulting thin slurry was aged for 20 min prior to the
charge of the second volume of heptanes (263.1 kg), also over
40 min at >50 °C. The resulting slurry was cooled to 20 °C
over 2 h and aged overnight before recirculating through a wet-
mill to afford a D90 particle size of 51 μm. The slurry was
isolated by filtration, and the mother liquor was displaced with
20% EtOAc in heptanes (206.2 kg), followed by two successive
displacement washes with heptanes (272.6 kg). The cake was
dried under full vacuum at 50 °C, which upon discharge
afforded 53.4 kg (61% yield) of 1 as a white to off-white
crystalline powder. Product losses of ∼15% were attributed to
AUTHOR INFORMATION
Corresponding Author
*Phone: (732) 227.7191. E-mail: paul.lobben@bms.com.
■
Notes
The authors declare no competing financial interest.
ACKNOWLEDGMENTS
■
The authors greatly acknowledge the following individuals for
their direct contribution to a successful technology transfer to
manufacturing: Jose E. Tabora (modeling), Michael Hrystak
(PAT), and San Kiang (early FT-IR method). Additional
appreciation is given to Michael Randazzo and Jaan Pesti for
careful reading of the drafts of the manuscript.
REFERENCES
■
(1) (a) Crispino, G. A.; Hamedi, M.; Laporte, T. L.; Thornton, J. E.;
Pesti, J. A.; Xu, Z.; Lobben, P. C.; Leahy, D. K.; Muslehiddinoglu, J.;
Lai, C.; Spangler, L. A.; Discordia, R. P. Process for the Preparation of
[(1R), 2S]-2-Aminopropionic acid 2-[4-(4-fluoro-2-methyl-1H-indol-
5-yloxy)-5-methylpyrrolo[2,1-f][1,2,4]triazin-6-yloxy]-1-methylethyl
ester. PCT Int. Appl. WO/2007/124332. (b) Cai, Z.-w.; Zhang, Y.;
Borzilleri, R. M.; Qian, L.; Barbosa, S.; Wei, D.; Zheng, X.; Wu, L.;
Fan, J.; Wautlet, B. S.; Mortillo, S.; Jeyaseelan, R.; Kukral, D. W.;
Kamath, A.; Marathe, P.; D’Arienzo, C.; Derbin, G.; Barrish, J. C.;
Robl, J. A.; Hunt, J. T.; Lombardo, L. J.; Fargnoli, J.; Bhide, R. S. J.
Med. Chem. 2008, 51, 1976−1980. (c) Bhide, R. S.; Cai, Z.-W.; Zhang,
Y.-Z.; Qian, L.; Wei, D.; Barbosa, S.; Lombardo, L. J.; Borzilleri, R. M.;
Zheng, X.; Wu, L. I.; Barrish, J. C.; Kim, S.-H.; Leavitt, K.; Mathur, A.;
Leith, L.; Chao, S.; Wautlet, B.; Mortillo, S.; Jeyaseelan, R.; Kukral, D.;
Hunt, J. T.; Kamath, A.; Fura, A.; Vyas, V.; Marathe, P.; D’Arienzo, C.;
Derbin, G.; Fargnoli, J. J. Med. Chem. 2006, 49, 2143−2146.
(d) Borzilleri, R. M.; Zheng, X.; Qian, L.; Ellis, C.; Cai, Z.-w.;
Wautlet, B. S.; Mortillo, S.; Jeyaseelan, R.; Kukral, D. W.; Fura, A.;
Kamath, A.; Vyas, V.; Tokarski, J. S.; Barrish, J. C.; Hunt, J. T.;
Lombardo, L. J.; Fargnoli, J.; Bhide, R. S. J. Med. Chem. 2005, 48,
3991−4008. (e) Hunt, J. T.; Mitt, T.; Borzilleri, R.; Gullo-Brown, J.;
Fargnoli, J.; Fink, B.; Han, W.-C.; Mortillo, S.; Vite, G.; Wautlet, B.;
Wong, T.; Yu, C.; Zheng, X.; Bhide, R. J. Med. Chem. 2004, 47, 4054−
4059.
(2) (a) Padilla, I.; Siu, L. L. Expert Opin. Invest. Drugs 2011, 20, 577−
586. (b) Ayers, M.; Fargnoli, J.; Lewin, A.; Wu, Q.; Platero, J. S. Cancer
Res. 2007, 67, 6899−6906.
G
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX

Organic Process Research & Development
Article
(3) (a) Saharinen, P.; Eklund, L.; Pulkki, K.; Bono, P.; Alitalo, K.
Trends Mol. Med. 2011, 17, 347−362. (b) Ivy, S. P.; Wick, J. Y.;
Kaufman, B. M. Nature Rev., Clin. Oncol. 2009, 6, 569−579.
(c) Kiselyov, A.; Balakin, K. V.; Tkachenko, S. E. Expert Opin. Invest.
Drugs 2007, 16, 83−107.
(4) Pesti, J. A.; LaPorte, T.; Thornton, J. E.; Spangler, L.; Buono, F.;
Crispino, G.; Gibson, F.; Lobben, P.; Papaioannou, C. G. Org. Process
Res. Dev. 2014, 18, 89−102. Please see the referenced paper for a
detailed discussion on the development and manufacture of “quality
gatekeeper” intermediate 3. The processes to prepare intermediate 3
proved to be robust and scalable, and they allowed optimum control
over input and process-related impurities.
(5) A prodrug of 2 was required in order to increase bioavailability.
(6) For further discussion, see the Supporting Information.
(7) Despite a thorough screen, Cbz-alanine intermediate 7 was an
amorphous noncrystalline solid.
(8) Deprotection of intermediate 3 is achieved with a wide variety of
sodium and potassium alkoxides. Sterically hindered alkoxides such as
sodium isopropoxide and potassium tert-butoxide resulted in a slower
rate of deprotection. The use of sodium hydroxide led to cleavage of
the indole-aryl ether bond. Reduction with LiAlH4 afforded an
unidentified impurity, >10 area %. Sodium methoxide was chosen due
to quality and cost considerations.
(9) The moderate regioselectivity is likely due to polarity and/or H-
bonding, slightly facilitating one transition state. However, screening
20 Lewis acids to soften the oxirane, according to Pearson’s hard−soft
acid−base (HSAB) theory, eroded the regioselectivity in all cases.
(a) Ho, T.-L. Chem. Rev. 1975, 75, 1−20. (b) Hanson, R. M. Chem.
Rev. 1991, 91, 437−475. (c) Maheswara, M.; Subba, K.; Rao, V. K.;
Do, J. Y. Tetrahedron Lett. 2008, 49, 1795−1800. (d) Halimehjan, A.
Z.; Gholami, H.; Saidi, M. Green Chem. Lett. Rev. 2012, 5, 1−5.
(10) An excess of (R)-propylene oxide was required due to the
competing hydrolysis, as approximately 20% of the starting material is
hydrolyzed to (R)-1,2-propanediol under the reaction conditions.
(11) The N-alkylation impurity 5 was found to be relatively less
soluble than the desired parent drug. A screen of crystallization
solvents ranging from ethers, acetates, and hydrocarbon all failed to
purge this impurity. A variety of absorption techniques, including
carbon pads, also failed to reduce the level of this impurity.
(12) The analysis was performed with SAS software (Statistical
Analysis Systems, SAS Institute Inc., SAS/STAT(R) 9.2 User’s Guide,
2nd ed, 2009.).
(13) The addition of DMF as a cosolvent was found to greatly
enhance the rate of the reaction in THF and EtOAc by increasing the
solubility of the EDAC and reducing precipitation of the urea
byproducts. 5−15 vol % DMF gave enhanced reaction rates without
dramatically compromising de; the use of 20% DMF does begin to
erode the diastereoselectivity, and without DMF the reaction mixture
is difficult to stir, especially at lower temperatures.
(14) (a) Hassner, A.; Krepski, L. R.; Alexanian, V. Tetrahedron 1978,
34, 2069−2076. (b) Myers, A. G.; Glatthar, G.; Hammond, M.;
Harrington, P. M.; Kuo, E. Y.; Liang, J.; Schaus, S. E.; Wu, Y.; Xiang, J.-
N. J. Am. Chem. Soc. 2002, 124, 5380−5401. (c) Kamijo, T.;
Yamamoto, R.; Harada, H.; Iizuka, K. Chem. Pharm. Bull. 1983, 31,
3724. (d) Basel, Y.; Hassner, A. J. Org. Chem. 2000, 65, 6368−6380.
(15) Inorganic and trialkylamine bases were completely ineffective.
Pyridines substituted in the two-position were expected to induce A1,3
strain in keto−enol tautomerization, thus minimizing epimerization. In
reality, <10% conversion was observed with 2-susbtituted pyridine
(methyl, methoxy, bromo, dimethylamine) catalysts. Although 4-
methoxy-, 4-methyl-, and 4-pyrrolopyridines were effective in
promting the acylation, these catalyst gave slightly inferior de,
compared with DMAP, as did 1-methylimidazole.
into the crystallization, this mixture of byproducts was shown to
negatively impact the tolerance of in-process related impurities during
the final crystallization.
(17) THF, isopropyl acetate, n-butyl acetate, 2-methyltetrahydrofur-
an, and ethyl acetate solvents were acceptable for the hydrogenolysis.
Concern with acetone was Schiff-base formation, and alcohols had the
possibility to hydrolyze the ester. Dichloromethane was discounted
due to the risk of forming bridged methylene dimers with primary
amines,
̆
(18) (a) Muslehiddinoglu, J.; Lobben, P.; Leung, S.; Spangler, L.;
̈
Kiang, S. Catal. Today 2007, 123, 164−170. (b) For a somewhat
related account detailing the impact of carbon dioxide on ruthenium
catalyzed reductive amination, please see: Strotman, N. A.; Baxter, C.
A.; Brands, K. M. J.; Cleator, E.; Kraska, S. W.; Reamer, R. A.; Wallace,
D. J.; Wright, T. J. J. Am. Chem. Soc. 2011, 133, 8362−8371.
(19) Stirred tank reactors under hydrogen pressure may need to be
purged with nitrogen prior to sampling. PAT offers the benefit of not
disturbing the gas composition until the end point.
(20) Three distinct peaks for intermediate 7 were utilized. The most
distinct peak due to a CO stretching band is unique for the input
material, and is located at 1746−1670 cm−1, and the bending regions
at 1216−1194 cm−1 and 1136−1109 cm−1. For carbon dioxide the
asymmetric stretch for the molecule is easily distinguished as it appears
at a higher wavenumber than the bending modes and at a lower
wavenumber than the stretching modes of organic compounds. The
peak appears as a singlet at 2339−1 not a doublet, indicating it is
dissolved in solution.
(21) By knowing the mass of intermediate 7 charged to the reaction
vessel and accurately measuring the concentration by FT-IR, the
proper catalyst loading was assured. FT-IR is used to develop a safe
process and monitor sensitive reactions. For recent examples see:
(a) Dunetz, J. R.; Berliner, M. A.; Xiang, Y.; Houck, T. L.; Salingue, F.
H.; Chao, W.; Yuandong, C.; Shenghua, W.; Huang, Y.; Farrand, D.;
Boucher, S. J.; Damon, D. B.; Makowski, T. W.; Barrila, M. T.; Chen,
R.; Martinez, I. Org. Process Res. Dev. 2012, 16, 1635−1645. (b) Sosa,
A. C. B.; Conway, R.; Williamson, R. T.; Suchy, J. P.; Edwards, W.;
Cleary, T. Org. Process Res. Dev. 2011, 15, 1458−1463. (c) Pesti, J.;
Chen, C.-K.; Spangler, L.; DelMonte, A. J.; Benoit, S.; Berglund, D.;
Bien, J.; Brodfuehrer, P.; Chan, Y.; Corbett, E.; Costello, C.; DeMena,
P.; Discordia, R. P.; Doubleday, W.; Gao, Z.; Gringras, S.; Grosso, J.;
̆
Haas, O.; Kacsur, D.; Lai, C.; Leung, S.; Miller, M.; Muslehiddinoglu,
̈
J.; Nguyen, N.; Qiu, J.; Olzog, M.; Reiff, E.; Thoraval, D.; Totleben,
M.; Vanyo, D.; Vemishetti, P.; Wasylak, J.; Wei, C. Org. Process Res.
Dev. 2009, 13, 716−728.
(22) The availability of 5 wt % kicker charge of catalyst was a
contingency in case of an unplanned deviation in process
parameter(s), including slow reaction kinetics recorded by FT-IR,
which may result in hydrolysis of the prodrug.
(23) For example, the rate of formation of over-reduction impurity 8
increased five times with a 10 °C increase in batch temperature.
(24) On lab scale in EtOAc and with <10% headspace, periodic or
continuous venting of the headspace was required to reach full
conversion. Conversely in THF, stalling occurring with continuous
venting was due to the preferential removal of hydrogen.
(25) The mass transfer coefficient (kLa) is a function of the H2
addition mode, agitation rate, reactor fill level (% headspace), the
reactor, and configuration (propeller type, baffles).
(26) The THF vol % was calculated using data from calibration
model prediction and the linearity. Standards of 1 in THF/EtOAc
solutions were prepared and the THF adjusted from 2 to 15 vol %
while the temperature ranged from 35 to 45 °C.
(27) The robustness of the wet-milling process was demonstrated
using an IKA Turrax 81-M-006 (IKA DISPAX-REACTOR DR 2000/
10, staged with two medium-grade rotor stators and a pumping stage.
Particle size was measured using dry method analysis on a Malvern
Mastersizer 2000. Laser light scattering (LLS) analysis (Lasentec)
proved effective in monitoring the wet-milling process in real-time.
(16) Mixing together the acylating reagents in the absence of
secondary alcohol 2, the alanine-related byproducts were identified by
HPLC and LCMS to be composed of alanine anhydride, the presumed
rearranged Cbz-L-alanine N-acylurea, MW 378 (EDAC + Cbz-
alanine), and di-, tri-, and tetra-Cbz-alanine oligomers with a molecular
weight of 223 + (223−18)n, where n = 1, 2 or 3, respectively. Spiked
H
dx.doi.org/10.1021/op500126u | Org. Process Res. Dev. XXXX, XXX, XXX−XXX